Thomas Banks

Quantum Mechanics: An Introduction

Thomas Banks

Quantum Mechanics: An Introduction

Cover Credit: Cover image “Dead or Alive?” (oil on wood panel, 18 × 24 inches) provided by Ricardo Mulero
(www.ricardomulerodesign.com), used with permission.

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2019 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper
International Standard Book Number-13: 978-1-4822-5506-5 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the
validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the
copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to
publish in this form has not been obtained. If any copyright material has not been acknowledged, please write and
let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or
utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including pho-
tocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission
from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com
(http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers,
MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of
users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has
been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only
for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Names: Banks, Tom, 1949- author.
Title: Quantum mechanics: an introduction / Thomas Banks.
Description: Boca Raton: CRC Press, Taylor & Francis Group, 2018.
Identifiers: LCCN 2018025041 | ISBN 9781482255065
Subjects: LCSH: Quantum theory—Textbooks.
Classification: LCC QC174.12 .B35545 2018 | DDC 530.12—dc23
LC record available at https://lccn.loc.gov/2018025041

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents

PREFACE xv
AUTHOR xix

Chapter 1  Introduction 1

1.1 WHAT YOU WILL LEARN IN THIS BOOK 1

1.2 WHAT YOU SHOULD KNOW ABOUT LINEAR ALGEBRA 5
1.3 THE ESSENTIALS OF PROBABILITY THEORY 9
1.4 PHILOSOPHICAL INTRODUCTION 14
1.4.1 The Essentials of Quantum Mechanics 15
1.4.2 Unhappening 19
1.4.3 Quantum Mechanics of a Single Bit 20
1.4.4 What is Probability? 20
1.5 A NOTE ON MATHEMATICS 22
1.6 FEYNMAN’S LECTURES 23
1.7 ACKNOWLEDGMENTS 23
1.8 EXERCISES 24

Chapter 2  Two State Systems: The Ammonia Molecule 27

2.1 INTRODUCTION AND MOTIVATION 27

2.2 THE AMMONIA MOLECULE AS A TWO STATE SYSTEM 28
2.3 PHYSICAL QUANTITIES AS MATRICES 30
2.4 OPERATIONS AS MATRICES 34
2.5 THE EIGENVALUES OF A MATRIX 36

v
vi  Contents

2.6 THE BORN RULE AND UNCERTAINTY 38

2.6.1 Born’s Rule in Pictures 40
2.7 WHICH OPERATORS ARE DIAGONALIZABLE? 42
2.7.1 Vectors, Functions, and Diagonal Matrices 44
2.8 QUANTUM DYNAMICS OF THE AMMONIA MOLECULE 45
2.9 SUMMARY 52
2.10 SKIP THIS UNLESS YOU ARE INTERESTED IN PHILOSOPHY 53
2.11 FOUR STATE SYSTEMS 56
2.12 FURTHER EXERCISES 60

Chapter 3  Quantum Mechanics of Single Particle in One-Dimensional

Space I 63

3.1 INTRODUCTION 63
3.2 TRANSLATION INVARIANCE AND FREE PARTICLE MOTION 64
3.2.1 No Experimental Definition of Infinity 64
3.2.2 Free Particles on an Infinite Continuous Line 68
3.2.3 Primer on the Fourier Transform and Fourier Series 77
3.3 LAGRANGIAN AND HAMILTONIAN MECHANICS: A REMINDER OR
A PRIMER 79
3.4 FURTHER EXERCISES 83

Chapter 4  Quantum Mechanics of a Single Particle in One-Dimensional

Space II 89

4.1 INTRODUCTION 89
4.2 A MORE MATHEMATICAL DESCRIPTION OF ONE-DIMENSIONAL MOTION 89
4.3 CONTINUUM OR DELTA FUNCTION NORMALIZATION 93
4.4 TIME EVOLUTION OF A FREE PARTICLE WAVE FUNCTION 96
4.5 THE DOUBLE SLIT EXPERIMENT 99
4.6 A WORKED EXERCISE 102
4.7 FURTHER EXERCISES 107
 Contents  vii

Chapter 5  The Harmonic Oscillator 113

5.1 INTRODUCTION 113

5.2 QUANTIZING THE SIMPLE HARMONIC OSCILLATOR 113
5.3 QUANTIZATION OF FIELDS AND WAVE–PARTICLE DUALITY 121
5.4 FERMI–DIRAC STATISTICS 127
5.5 EXERCISES 129

Chapter 6  Review of Linear Algebra and Dirac Notation 133

6.1 INTRODUCTION 133

6.2 SEPARABLE HILBERT SPACES 134
6.3 UNITARY TRANSFORMATIONS, UNITARY MATRICES,
AND CHANGES OF BASIS 135
6.4 NORMAL OPERATORS ARE DIAGONALIZABLE OPERATORS 138
6.5 CONTINUUM EIGENVALUES AND UNBOUNDED OPERATORS 139
6.6 SUMMARY 141
6.7 DIRECT SUMS AND TENSOR PRODUCTS 142
6.8 THE GENERALIZED UNCERTAINTY RELATION 143
6.9 SYMMETRIES AND CONSERVATION LAWS 144
6.9.1 Dynamical Symmetry Groups 147
6.9.2 Projective Representations of Symmetry Groups 149
6.9.3 Examples 152
6.10 EXERCISES 160

Chapter 7  Rotation Invariance and the Hydrogen Atom 165

7.1 INTRODUCTION 165

7.2 UNITS AND SYMMETRIES 166
7.3 IRREDUCIBLE REPRESENTATIONS OF THE COMMUTATION RELATIONS 168
7.4 ADDITION OF ANGULAR MOMENTA 175
7.5 THE HAMILTONIAN OF SPHERICALLY SYMMETRIC POTENTIALS 176
viii  Contents

7.6 THE PERIODIC TABLE 181

7.7 THE SPIN OF THE ELECTRON 186
7.8 SPIN PRECESSION AND SPIN RESONANCE 190
7.9 STERN–GERLACH EXPERIMENTS 191
7.10 EXERCISES 192

Chapter 8  Scattering Electrons on a Nucleus 197

8.1 INTRODUCTION 197

8.2 POSITIVE ENERGY EIGENFUNCTIONS AND SCATTERING AMPLITUDES 197
8.3 ANALYSIS OF THE COULOMB SCATTERING AMPLITUDES 200
8.4 EXERCISES 204

Chapter 9  Charged Particles in a Magnetic Field 207

9.1 INTRODUCTION 207

9.2 THE LORENTZ FORCE AND LANDAU’S SOLUTION 207
9.3 EXERCISES 212

Chapter 10  The Meaning of Quantum Measurement and the Emergence

of a Classical Reality 215

10.1 INTRODUCTION 215

10.2 COUNTING STATES AND COLLECTIVE COORDINATES 216
10.3 COLLECTIVE COORDINATES 217
10.4 INTERFERENCE OF CLASSICAL HISTORIES 220
10.5 SCHRÖDINGER’S CAT AND SCHRÖDINGER’S BOMB 222
10.6 THE EINSTEIN–PODOLSKY–ROSEN (EPR) PARADOX, BELL’S
THEOREM, AND OTHER ODDITIES 228
10.7 THE RELATION BETWEEN THEORY AND EXPERIMENT IN QM 234
10.8 EXERCISES 236
 Contents  ix

Chapter 11  Sketch of Atomic, Molecular, and Condensed Matter Physics 239

11.1 INTRODUCTION 239

11.2 THE BORN–OPPENHEIMER APPROXIMATION 240
11.3 COLLECTIVE COORDINATES, ROTATIONAL LEVELS, AND PHONONS 242
11.3.1 Water 243
11.3.2 Phonons 244
11.4 THE HYDROGEN MOLECULAR ION 245
11.5 DENSITY FUNCTIONAL THEORY 247
11.6 ELEMENTS OF THE THEORY OF CRYSTALLINE SOLIDS 250
11.7 BAND STRUCTURE 252
11.7.1 A Simple Model 255
11.8 THE FERMI LIQUID THEORY OF CONDUCTORS 256
11.9 EXERCISES 259

Chapter 12  Quantum Statistical Mechanics 261

12.1 INTRODUCTION 261

12.2 QUANTUM FIELD THEORY OF FERMIONS 263
12.3 STATISTICAL MECHANICS OF BOSONS AND FERMIONS 264
12.4 THE PARTITION FUNCTION 265
12.5 THE LOW TEMPERATURE LIMIT 271
12.6 STATISTICAL MECHANICS OF A FERMI LIQUID 275
12.7 PLANCK’S RADIATION LAW 277
12.8 EXERCISES 278

Chapter 13  Perturbation Theory: Time Independent 281

13.1 INTRODUCTION 281

13.2 BRILLOUIN–WIGNER PERTURBATION THEORY FOR
NONDEGENERATE LEVELS 281
13.2.1 Relation to Rayleigh–Schrödinger Perturbation Theory 285
x  Contents

13.3 DEGENERATE PERTURBATION THEORY 285

13.4 THE FEYNMAN–HELLMANN THEOREM 287
13.5 EXAMPLES 287
13.5.1 The Stark Effect 287
13.5.2 The Zeeman Effect 289
13.5.3 Fine Structure and the Weak-Field Zeeman Effect 290
13.5.4 Coulomb Expectation Values of Powers of R 292
13.5.5 A Three-Dimensional Example 294
13.5.6 Degenerate Perturbation Theory in a Macroscopic System 295
13.6 EXERCISES 298

Chapter 14  Perturbation Theory: Time Dependent 301

14.1 INTRODUCTION 301

14.2 DYSON’S FORMULA 301
14.3 THE DIRAC PICTURE 303
14.4 TRANSITION AMPLITUDES 303
14.5 ELECTROMAGNETIC RADIATION FROM ATOMS 304
14.6 INCOHERENT PERTURBATIONS AND RADIATIVE DECAY 306
14.7 SELECTION RULES 309
14.8 EXERCISES 310

Chapter 15  The Adiabatic Approximation, Aharonov--Bohm, and Berry Phases 315

15.1 INTRODUCTION 315

15.2 ADIABATIC ENERGY EIGENSTATES 315
15.3 THE BERRY PHASE 318
15.4 THE AHARONOV–BOHM EFFECT 320
15.5 ANYONS, FERMIONS, AND THE SPIN STATISTICS THEOREM 322
15.6 EXERCISES 326
 Contents  xi

Chapter 16  Scattering Theory 329

16.1 INTRODUCTION 329

16.2 GENERAL FORMALISM 330
16.3 POTENTIAL SCATTERING 331
16.4 THE BORN APPROXIMATION 332
16.5 PHASE SHIFT ANALYSIS 333
16.5.1 The Effective Range Approximation 334
16.6 RESONANCES 335
16.7 A PARTIALLY WORKED EXERCISE: THE δ SHELL POTENTIAL 336
16.8 EXERCISES 339

Chapter 17  The JWKB Approximation 341

17.1 INTRODUCTION 341

17.2 THE JWKB EXPANSION 341
17.3 THE JWKB APPROXIMATION IN ONE DIMENSION 343
17.4 COMMENTS ON THE SOLUTIONS 344
17.5 THE JWKB APPROXIMATION FOR THE PROPAGATOR 345
17.6 THE JWKB APPROXIMATION FOR ENERGY LEVELS 346
17.7 THE JWKB APPROXIMATION TO THE WAVE FUNCTIONS OF
EIGENSTATES 348
17.7.1 Examples 350
17.8 THE DECAY OF METASTABLE STATES 352
17.9 MORE EXAMPLES 354
17.10 THE JWKB APPROXIMATION FOR PHASE SHIFTS 356
17.11 EXERCISES 357

Chapter 18  The Variational Principle 361

18.1 INTRODUCTION 361

18.2 GENERAL PROPERTIES 361
xii  Contents

18.3 EXAMPLES 362

18.4 THE HARTREE AND HARTREE–FOCK APPROXIMATIONS 365
18.5 THE LANCZOS METHOD 370
18.6 EXERCISES 372

Chapter 19  The Feynman Path Integral 375

19.1 INTRODUCTION 375

19.2 TWO DERIVATIONS OF THE PATH INTEGRAL FORMULA 375
19.3 THE PATH INTEGRAL FOR A HARMONIC OSCILLATOR 380
19.4 MORE GENERAL POTENTIALS 385
19.5 PATH INTEGRALS AT FINITE TEMPERATURE 386
19.6 PATH INTEGRALS AND THE JWKB APPROXIMATION 388
19.7 PATH INTEGRALS FOR SPIN AND OTHER DISCRETE VARIABLES 389
19.8 FERMIONS AND GRASSMANN INTEGRATION 391
19.9 FURTHER EXERCISES 396

Chapter 20  Quantum Computation? 399

20.1 INTRODUCTION 399

20.2 QUANTUM INFORMATION 400
20.3 PAGE’S THEOREM REDUX, MONOGAMY, AND CLONING 402
20.4 QUANTUM KEY DISTRIBUTION 404
20.5 QUANTUM TELEPORTATION 406
20.6 GATES FOR QUANTUM COMPUTERS 408
20.7 COMPUTATIONAL COMPLEXITY 410
20.8 CAN WE BUILD A QUANTUM COMPUTER? 413
20.9 EXERCISES 414

Chapter 21  L’Envoi: Relativistic Quantum Field Theory 417

 Contents  xiii

Appendix A  Interpretations of Quantum Mechanics 423

Appendix B  The Dirac Delta Function 447

Appendix C  Noether’s Theorem 449

Appendix D  Group Theory 453

Appendix E  Laguerre Polynomials 461

Appendix F  Summary of Dirac Notation and Linear Algebra 463

Appendix G  Answers to Selected Problems 469

Bibliography 523

Index 535
Preface

Quantum mechanics is at the basis of most physical science and modern electronic technology,
and has increasing relevance to the biological sciences. It’s also the most confusing subject
in the world, because it seems to deny the very foundations of logic. Logic is a precise
distillation of our intuitive grasp of how things work. It seems to have nothing to do with
particular physical situations where we’ve become used to the fact that our intuition is only
an approximation. For example, Galileo appreciated that Aristotle’s intuitive notion that
rest is the natural state of bodies was wrong, and invented the relativity of frames moving at
uniform velocity. Einstein realized that Galileo’s laws relating the kinematics of two frames
moving at uniform velocity were only approximately correct, valid when the velocity was
much smaller than that of light. All of this is a bit confusing when you first encounter it,
but it’s not actually that mind boggling, and once you understand how the correct formulas
reduce to the non-relativistic ones (which are straightforward and make intuitive sense) when
the velocity is small compared to that of light, it actually can be pretty easy to come to terms
with relativity, and even develop an intuition for it. The goal of this text is to help the reader
develop a similar understanding of Quantum Mechanics (QM).
While there are many available textbooks on quantum mechanics, they almost uniformly
present certain aspects of the subject in a manner that reflects the confusions encountered
by inventors of quantum mechanics. These confusions include the meaning of wave–particle
duality and the correct interpretation of measurements. Quantum Mechanics: An Introduction
presents the subject from a modern perspective. It includes an elementary discussion of field
quantization, the only proper way to understand wave particle duality, at a very early stage.
The essentials of field quantization are not difficult, because fields are just collections of simple
harmonic oscillators (the standard example used in elementary texts). The interpretation of
particles as excitations of quantized fields is the only way to understand the identity of
particles and the peculiar statistical properties of multi-particle states observed in the real
world. All extant textbooks either give incorrect explanations of the origin of Bose and Fermi
statistics, or introduce these laws as an additional postulate. On the other hand, all working
quantum theorists know that the statistics of identical particles is a consequence of quantum
field theory.

xv
xvi  Preface

Field theory is introduced in Chapter 5, and used to simplify the discussion of quantum
statistical mechanics in Chapter 12. It’s invoked in Chapter 11 as well, in our brief discussion
of density functional theory.
The book also explains the interpretation of measurements in terms of decoherence,
including the correct explanation of how the classical world we experience emerges from the
underlying quantum formalism (e.g., order of magnitude estimates of deviations from classical
behavior) in a new way, to provide a more accurate and rounded picture for the reader. While
detailed derivations of the principles of decoherence are difficult, the description of the results
of those derivations is straightforward to understand at an elementary level.
The third major innovation in this book is the decision to include a brief discussion in
Chapter 11 of the principal approximation methods used in many body physics. Students
going on to careers in areas that use quantum mechanics will learn about these in more
advanced courses, but all students need a glimpse of the way that quantum mechanics explains
the world around us.
Throughout, I’ve attempted to emphasize the key principle that quantum mechanics is a
probability theory, in which not all quantities that appear in the equations of motion of the
fundamental variables can take definite values at the same time. As a consequence, histories
cannot be predicted definitely, even if we have a precise account of a complete set of initial
data. Moreover, the quantum formalism is mathematically inevitable, since given any list of
data, we can introduce matrices, which change one data point into another. Once we do this,
quantum probabilities are defined, even in systems where we’re able to ignore them because
the equations of time evolution only require the values of quantities that are simultaneously
definite.
Those of a mathematically rigorous turn of mind will find my discussion of continuous
spectra and unbounded operators lacking in precision. Von Neumann’s famous book cleared
up most of the issues, and there are many fine books on the mathematics of quantum mechan-
ics to which one can turn for the details. Physics students, for the most part, are impatient
about such things, and it would distract from their absorption of the already difficult con-
ceptual and computational issues of quantum mechanics.

ORGANIZATION OF THE BOOK

I’ve chosen to begin with a careful explanation of the differences between classical and quan-
tum probability. This takes place in terms of the simplest quantum system: a two state
system or q-bit, where the algebra involved is elementary. I also describe why the two lowest
energy states of the ammonia molecule form a good testing ground for comparing the two
theories, and how quantum mechanics wins that test in a decisive manner.
The beginning of the text also introduces a key theme: the relation between symmetries
and conservation laws (energy conservation is a consequence of time translation symmetry).
 Preface  xvii

In classical mechanics, this is a sophisticated theorem, proven by Emmy Noether, but in

quantum mechanics it follows from the very definition of symmetry. Finally, we’ll understand
at this very early stage, the relation between discrete energy levels and the spectrum of light
emitted by matter.
My treatment of the free particle is based on symmetry principles: invariance under
spatial translations and Galilean boosts. The harmonic oscillator is treated by algebraic
methods, which enable us to obtain both the energy eigen-values and the eigen-functions with
a minimum of computation. We do this through the introduction of coherent states, a simple
topic that is usually reserved for more advanced courses. Apart from computational simplicity,
the introduction of coherent states enables us to expose the real connection between particles
and physical waves, and to dispel the false notion that the Schrödinger wave function is
a physical wave, rather than a device for computing probabilities. Finally, this discussion
enables us to introduce photon creation and annihilation operators at an early stage. This is
conceptually useful in discussions of transitions among energy levels.
An important choice that must be made in any quantum mechanics text is where and
when to go over the mathematics of Hilbert space. I do this by introducing it informally in
Chapters 2, 3, and 4, providing a formal introduction in Chapter 6, and summarizing the
important rules in a brief appendix. Similarly, group theory, which is not discussed in detail,
is split between Chapter 6 and an appendix.
Chapter 7 covers the hydrogen atom. We first solve for the spherical harmonics using the
algebraic techniques of angular momentum theory. Then we solve the radial problem two
ways, first using the traditional power series solution, and secondly (in a problem set) with
an algebraic method, which explains the accidental degeneracy of the hydrogen spectrum.
Chapters 8 and 16 are devoted to scattering theory, the first for the exactly soluble
Coulomb potential, while the second is a more general discussion. The book introduces
scattering theory in the spherically symmetric context, rather than using artificial one-
dimensional examples. One-dimensional scattering is treated through an extensive worked
problem set in Chapter 4, on the square well and barrier.
Chapter 9 is about Landau levels. This subject is often omitted from textbooks at this
level, but it’s the basis for an enormous amount of modern activity, so it’s important to
include.
Chapter 10 finally deals with the thorny problem of the proper interpretation of quantum
mechanics, and with measurement theory. The point of view emphasized here is that, while
QM always gives us a mathematical definition of probabilities for histories of any given com-
plete commuting set of quantities, these probabilities do not satisfy the “sum over histories
rule” for total probability, which leads to Bayes’ notion of conditional probability. The inter-
pretation of quantum predictions in terms of actual experiments depends on the existence
of systems with large numbers of variables for which that history sum rule is satisfied with
xviii  Preface

accuracy exponential in the number of atoms in the subsystem on which those variables oper-
ate. This is the phenomenon of decoherence, and we briefly review the order of magnitude
estimates necessary to demonstrate its plausibility.
Chapters 13, 14, 17, and 18 sketch the main approximation methods that have been
used to solve problems in QM. These chapters follow fairly standard lines. Chapter 15 is on
the adiabatic approximation, Berry phases, and the Aharonov–Bohm effect. It also discusses
anyon statistics in two spatial dimensions and the idea of changing statistics by flux attach-
ment. We learn that fermions in any dimension can be thought of as bosons coupled to a Z2
gauge field. In a one-semester course, I usually include the discussion of the Aharonov–Bohm
effect along with Landau levels.
Chapter 19 discusses Feynman’s path integral formulation of quantum mechanics in some-
what more detail than is found in most textbooks. Chapter 20 is a brief introduction to
quantum information and quantum computer science. While far from complete, its aim is to
enable the reader to get a head start on a fascinating, rapidly developing field. Lastly, the
first appendix discusses a variety of attempts to interpret quantum mechanics in a “realist”
fashion, while the others are devoted to technical and mathematical details.

FOR INSTRUCTORS
This textbook is intended for an advanced (junior or senior level) undergraduate quantum
mechanics course, or a first year graduate course, for physics and math majors, depending on
the level of preparation of the students. The whole book is intended for a full year course. A
single semester course may be constructed using Chapters 1–7, 9 and 12, in addition to
an optional lecture briefly presenting the main idea in Chapter 10. It may also be used in
a physical chemistry or materials science course, as long as the students have had linear
algebra, and will be useful to computer scientists who are interested in studying the subject
from a physics standpoint. Readers should be comfortable with basic notions of linear algebra,
and are encouraged to review the matrix representation of a linear operator, with particular
emphasis on the fact that it depends on a choice of basis. The book uses operator algebra as its
primary computational tool, rather than differential equations, because these methods involve
much less mathematical manipulation, and are of much greater general utility. The differential
equation form of the Schrödinger equation is actually only really useful for artificial problems
involving a single particle.

INSTRUCTOR RESOURCES
Solutions to problems are available to course instructors upon request. Please visit the book’s
page at http://www.crcpress.com/9781482255065.
Author

Thomas Banks is a Distinguished Professor of Physics at Rutgers University and Emeritus

Professor at University of California at Santa Cruz. He was born and brought up in New York
City and got his undergraduate education at Reed College in Portland, Oregon, majoring in
Physics and Mathematics, and his Ph.D. in physics at M.I.T. in 1973. He then joined the
faculty at Tel Aviv University, rising to the rank of full Professor before he left in 1985. In
1986, he became a Professor at U.C., Santa Cruz, and, in 1989, moved to become a founding
member of the New High Energy Theory Center at Rutgers University. From 2000 to 2015
Banks split his time between Santa Cruz and Rutgers, and returned to Rutgers full time
in 2015. In 2014, Banks married Anne Elizabeth Barnes, a judge on the Georgia Court of
Appeals. He has two children from a previous marriage. Banks has held visiting professorships
at Stanford, the Stanford Linear Accelerator Center, the Weizmann Institute of Science, and
the Institute for Advanced Study in Princeton, New Jersey. He has been a Guggenheim
Fellow, a Fellow of the American Physical Society, and in 2011, was elected to the American
Academy of Arts and Sciences. His research has mostly been in the fields of particle physics
and cosmology.

xix
 CHAPTER 1

Introduction

1.1 WHAT YOU WILL LEARN IN THIS BOOK

The formalism of quantum mechanics (QM) was developed some 90 years ago. Since then,
tens of textbooks on the subject have appeared. The reader deserves to know why she/he
should choose to learn QM from this book, in preference to all the others. There is no
better explanation than a list of the things you should be able to learn from reading it.
Comparing it to one of the older texts, you will find some differences in emphasis and some
differences in the actual explanations of the physics. The latter were inserted to correct what
this author believes are errors, either conceptual or pedagogical, in traditional presentations
of the subject. The following list includes both topics where our presentation differs from the
traditional one and topics that parallel tradition.

• Any mathematical description of a physical system consists of a list of the possible

states the system is in. Physical quantities characterizing the system can be thought of
as functions on the space of states. A completely equivalent mathematical description
is to view the set of states as the basis of a vector space and the physical quantities as
matrices A, diagonal in that basis. This description is convenient because operations
on the system then become a certain kind of nondiagonal matrix, U . If one does not
know what state the system is in, one can introduce a probability distribution ρ whose
diagonal matrix elements pi give you the probability of being in the i-th state. The
expected or average value of the quantity given by the matrix A is just given by the
formula
hhAii = tr (ρA). (1.1)
The trace of a matrix means the sum of its diagonal matrix elements. The mathematical
result that leads to QM is that we can extend this formula to

hhM ii = tr (ρM ). (1.2)

1
2  Quantum Mechanics

for any matrix M which is diagonal in a basis related to the original basis by a transfor-
mation that preserves the lengths of (complex) vectors. Such operators1 are given the
name normal. The resulting formula gives a probability for the quantity M to take on
one of its eigenvalues. What is remarkable is that if we make this interpretation of the
formula, then the quantity M has uncertain values even when we know exactly which of
the original states the system is in. Conversely, if we accept the eigenvectors of M , for
which M takes on a definite value, as allowed physical states of the system, then all of
our original physical quantities are uncertain in that state. The essence of QM is that
we accept all normal operators as possible physical quantities characterizing the sys-
tem, so that the theory has an intrinsic uncertainty built into it, not related to our lack
of knowledge or failure to measure details. This new kind of probability theory violates
some of our intuitions about what a probability theory should do. We will explore this
idea in more detail in the second part of this introduction and in Chapter 10.
Another, completely equivalent way of describing this new probability theory is illus-
trated in Figure 1.1: We view the answer to a Yes/No question as the vertical or
horizontal position of a switch, along a pair of axes with positive orientation. If we
draw another unit vector in the plane, it defines a pair of positive quantities, p1,2 , that
sum to one: the squares of the dot products of that vector with the original axes. In
QM, we consider each of these unit vectors as defining a possible state of the system,
with pi interpreted as the probability that the system will be found to be in each of the
Yes/No states. p1 is also the probability that, assuming the answer to our question is
No, the system is in the state defined by the unit vector in the picture, and similarly
for p2 if the answer is Yes. Operators diagonal in different bases are just physical quan-
tities that have definite values in the states described by those bases. Since the process

a2 + b2 = 1

Figure 1.1 Pythagoras’ theorem defines a new kind of probability.

1
The same linear operation or operator has different matrices if we use different orthonormal bases of the
vector space to describe vectors as n-tuples of complex numbers.
 Introduction  3

of diagonalizing matrices involves the solution of algebraic equations, we are forced to

consider complex vector spaces, instead of the real space shown in the figure, in order
to find the most general allowed state of our system.

• Equations of motion in physics relate the time rate of change of physical quantities to
the present values of other physical quantities. Classical mechanics is the special case
of QM in which all the physical quantities appearing in the equations of motion have
definite values in the same state. Our current state of knowledge suggests that the
equations that describe our world at the microscopic level do not have this property.

• QM’s unification of physical quantities characterizing the state of a system and opera-
tions on that system leads to a transparent proof of Noether’s theorem, which relates
symmetry operations on a system to conserved physical quantities. In particular, time
translation invariance leads to a conservation law called energy conservation. The cor-
responding matrix/operator is called the Hamiltonian H and leads to Heisenberg’s
form of the equations of motion of QM, for any physical quantity (normal operator) A,
whose definition has no explicit time dependence.

i~∂t A = HA − AH ≡ [H, A]. (1.3)

• The eigenvalues of the Hamiltonian operator are the allowed energies of the system in
states that have definite energy. In states that correspond to classical motions which
extend only into finite regions of space, these eigen-energies are discrete, leading to the
famous quantization laws that give the subject its name. For unbounded motions, the
energy eigenvalues are continuous.

• A particular example, with only bounded motions, is the harmonic oscillator, a good
first approximation to almost any system perturbed a little bit from its lowest energy
state. Harmonic oscillators are particularly important because fields obeying linear
field equations like Maxwell’s equations are just collections of oscillators. We will see
that oscillators have two natural kinds of excitations of their lowest energy state: small
quantized oscillations with definite energy and large coherent excitations with indefinite
energy. The coherent excitations are parameterized by classical solutions of the system,
and if the parameters defining the coherent states are large, then the corresponding
quantum states have small uncertainties in the classical values of the coordinates, at
all times. When promoted to the field theory context, these two kinds of excitations
correspond, respectively, to particles, which are automatically identical and obey what is
called Bose-Einstein statistics for multiparticle states and classical fields. This duality
between two kinds of states of a quantized field theory is the proper interpretation
of the phrase wave-particle duality. Historically, and in many current textbooks, that
4  Quantum Mechanics

phrase is applied to the description of particle states by Schrödinger “wave functions.”

You will learn why that use of the phrase is misleading by the fifth chapter of this book.
• Many of the particles in the real world, although identical, do not obey Bose–Einstein
statistics, but rather another rule, called Fermi-Dirac (FD) statistics, which was moti-
vated by Pauli’s Exclusion Principle, one of the keys to understanding the Periodic
Table of chemical elements. The explanation of FD statistics is quite a bit more com-
plicated, and we will have to learn about the Aharonov–Bohm effect in Chapter 15
before we understand it completely. Leave this as a mysterious teaser for now.
• The application of these ideas to motion of a single particle in a spherically symmetric
potential leads to the theory of angular momentum. By Noether’s theorem, this is equiv-
alent to studying how quantum states transform under rotations of coordinates. Since
such motions are compact, they lead to quantization rules: in this case, the correct form
of Bohr’s rules for quantizing angular momentum. We will then move on to the study
of radial motion, which will lead us to Bohr’s formula for the energy levels of hydrogen.
• To study most interesting physical systems, we have to understand the dynamics of
multiple particles, and we must abandon exact solutions of the equations. Before mov-
ing on to that, we pause to explain the relation between the mathematics of QM and
our everyday experience of an apparently classical world. The key to understanding
that is the notion of a collective coordinate of a macroscopic system. A macroscopic
system is one containing of order 1020 atoms or more, which is in an energy range where
a huge number of different, closely spaced, multiatomic energy levels are excited. A col-
lective coordinate of such a system is an average over all of the atoms, like their center
of mass. We will argue that the QM uncertainties in collective coordinates are of order
10−10 or smaller, and that the violations of the “intuitive” rules of probability theory
20
by the quantum predictions for these quantities are smaller than e−10 . This means
that the latter violations are unobservable, even in principle. The world we are used
to corresponds to observations only of such collective coordinates. A proper under-
standing of this fact removes much of the mystery of QM, associated with phrases
like “Schrödinger’s Cat,” “Collapse of the Wave Function,” “Spooky Action at a Dis-
tance,” etc.
Quantitatively detailed treatments of multiparticle QM require the use of large com-
puters and a variety of approximations. The two most important approximations are
the Born-Oppenheimer approximation, which exploits the fact that nuclei move much
more slowly than electrons, and some form of Hartree’s self-consistent field approxima-
tion. The modern form of the latter is called Density Functional Theory, and we will
be able to give a quick sketch of it, but in very little detail. It will be enough to give
us a rough explanation of the periodic table and the gross properties of solids.
 Introduction  5

• The quantum treatment of the statistical mechanics of multiparticle systems, which

leads to the resolution of Gibbs paradox, the Planck black body spectrum, and the
phenomena of Fermi surfaces and Bose–Einstein condensation, will be dealt with in
Chapter 12.

• Much of the rest of the book is devoted to discussions of various analytic approxima-
tion schemes for quantum problems. Some of this material is presented from a fresh
perspective, but for the most part, it follows roughly traditional lines.

• Chapter 15 on Berry phases and the Aharonov–Bohm effect is somewhat novel, in that
it presents an explanation of Fermi statistics in terms of the A–B effect and a simple
presentation of the properties of anyons, particles that can exist only in two space
dimensions, and which obey statistical rules different from either bosons or fermions.

• Chapter 19 on Feynman’s Path Integral formulation of QM presents a topic often omit-

ted from textbooks at the undergraduate level. It also covers Schwinger’s alternative
functional differential equation derivation of the path integral as well as path integrals
for fermionic variables and other systems with only a finite number of quantum states.

• Chapter 20 is a quick introduction to Quantum Computing. Its sole purpose is to enable

you to pick up one of the good texts on the subject and get into it rapidly. For the most
part, we have stuck to the notation and nomenclature of this book, rather than intro-
ducing a whole chapter written in the foreign language of quantum computer scientists.

1.2 WHAT YOU SHOULD KNOW ABOUT LINEAR ALGEBRA

The basic premise of the approach to QM used in this book is that the mathematics of
linear algebra defines a new kind of intrinsic probability theory, in which not all quantities
can take definite values at the same time. If the equations of motion relate quantities that
cannot be definite at the same time, then the concept of probabilities for histories, which
is central to the way that classical physicists and philosophers think about probability, can
only be an approximate one, valid for certain systems containing many fundamental degrees
of freedom, and only for certain average or collective properties of those systems. The key
mathematical notion that defines QM probability theory is the notion of change of basis
in a vector space over the complex numbers. It cannot be stressed often enough that the
mathematical surprise that leads to QM is that the generalization of Pythagoras’ theorem
to N -dimensional complex vector space can be interpreted as saying that every unit vector
defines a probability distribution over the set of all other unit vectors, the absolute square of
the projection of one vector on another. This leads to the idea that any diagonalizable linear
operator on the vector space is just as good a candidate for a quantity that can be measured,
6  Quantum Mechanics

as any other. All of the basic properties of quantum systems, and their violation of classical
logic, flow from this observation. It is therefore incumbent on any would-be cognoscenta of
QM to have a thorough knowledge of linear algebra.
You should know the definition of a complex vector space and understand that the rep-
resentation of a vector |vi2 in that space as a column of complex numbers:

v1
 
 .. 
|vi =  .  ,
vN

depends on a choice of orthonormal basis vectors |en i. The word orthonormal means that the
scalar products (Dirac notation again) of these vectors with each other satisfy

hei |ej i = δij .

The components vn are the coefficients in the expansion

N
X
|vi = vn |en i.
n=1

The notation hv| for a given (column vector) |vi means the corresponding transposed row
vector, but with its elements subjected to complex conjugation

hv| = (v1∗ . . . vN
∗
).

This is the representation for hv| in the transposed basis hei |, for row vectors, corresponding
to the basis |ei i for column vectors. The reader would do well to convince him/herself at this
point that the expression hv|wi is equal to the complex number
N
vi∗ wi .
X
hv|wi =
i=1

A key mathematical fact that leads to QM is that if both vectors have unit length, the
absolute square of this number obeys all the mathematical properties that one would need
to call it “the probability that, if one were in the state of a system represented by the unit
vector |vi, then a measurement designed to detect whether one was in the state represented
by |wi would give a positive answer.” This violates only one rule of classical logic: The
Law of the Excluded Middle. That law takes as the definition a state that one cannot be in
2
We are here using Dirac’s notation for vectors, which will be explained in Chapter 2, and more extensively
in Chapter 6.
 Introduction  7

two states simultaneously. Ultimately, like any other law in a scientific theory, the Law of
the Excluded Middle must be tested by experiment, and it fails decisively for experiments
performed on microscopic systems. Mathematically, this law is equivalent to claiming that
every state encountered in the world is an element of some particular orthonormal basis.
Orthonormal bases are not unique. In two- or three-dimensional space, we are familiar
with the fact that any choice of orthogonal axes is related to any other by a rotation. The
corresponding ambiguity in N -dimensional complex space is an N -dimensional unitary trans-
formation. The most important thing to remember is that the column vector notation for
|vi depends on the choice of basis, but the vector itself is independent of that choice. This
is like saying that the position of your house on the earth is independent of the longitude
latitude coordinates we generally designate it by. One can define another set of longitude
latitude coordinates by choosing the “North Pole” to be any other point on earth and draw-
ing the corresponding arcs. The ones we use are convenient because of their relation to the
rotation axis and magnetic field of the earth, but the South Pole would be just as good. An
observer looking at our solar system from the outside might have chosen the plane of the
ecliptic to define Earth’s equator, or the plane of galactic rotation, etc. These changes of the
definition of the coordinate system change your house’s designation by a pair of numbers,
but they do not change where your house is.
Linear operations on a vector space are mappings that take vectors into vectors and
satisfy
A(a|vi) + b|wi) = aA(|vi) + bA(|wi).
From now on, we will drop the round brackets around the argument of a linear operator and
write A(|vi) ≡ A|vi. Once we have chosen a basis, A is represented by a matrix, an array of
rows and columns of complex numbers. We will see that we can compute those numbers as

Aij = hei |A|ej i.

In words, the number in the i-th row and j-th column of the matrix is given by the scalar
product of the i-th orthonormal basis vector, with the action of A on the j-th orthonormal
basis vector. It is the component of A|ej i along the |ei i axis. The matrix looks different for
different choices of bases. In particular, for certain operators, called normal, there is a basis
where the matrix is diagonal. This is the basis of eigenvectors, |ai i of A satisfying

A|ai i = ai |ai i.

The process of diagonalizing A, given its matrix in some random orthonormal basis, is simply
the process of finding the orthonormal basis of eigenvectors.
To reiterate, the fundamental principle of QM is that Pythagoras’ theorem in N -
dimensional complex vector space can be thought of as a probability theory for unit vectors
8  Quantum Mechanics

in that space. Each unit vector |ei is a state of a physical system, and it defines a probability
distribution that tells you that if you are in the state |ei, the probability to be in any other
state |f i is the absolute square of the scalar product hf |ei (called a probability amplitude).
The classical approach to the same system would only allow states that are elements of a
particular orthonormal basis. In that case, the same rule tells us that the probability to be in
some allowed state, given that one is definitely in another, is exactly zero. This is sometimes
called The Law of the Excluded Middle, and it is not true if one allows any unit vector to be
an allowed state. At this point, students should go back to our discussion of defining expec-
tation values as traces of operators, and try to understand how that discussion of probability
relates to the current one. The connection involves the projection operator P (e) on a state |ei,
which is defined to equal 1 when applied to |ei and to give zero when acting on the subspace
of vectors orthogonal to |ei. It is better if you work out that connection for yourself, to learn
it more thoroughly.
The intuitive reason that we think that knowing the state of a system should determine
that there is no chance of being in any other state, is that we think of determining the state
by “doing all possible measurements of the properties of the system.” This is simply incorrect
in QM. The quantum rule is that one determines a state by doing all possible compatible
measurements. That is, the mathematics of the theory, and the way it assigns values to
quantities, is such that not all possible quantities that could be measured can be known
with precision at the same time. One can try to make this intuitively plausible by thought
experiments first described by Heisenberg. One determines the position of a particle by
“looking where it is.” Maxwell’s theory of light tells us that in order to do that with precision
∆x you have to scatter light off it with wavelength less than or equal to ∆x. Maxwell tells
us that light carries momentum, but in classical electrodynamics one can transfer arbitrarily
small momentum with any wavelength. However, the quantum theory of light (see Chapter 5)
~
says that the minimum momentum carried by photons of wavelength ∆x is ∆x , so position
and momentum cannot be known simultaneously with arbitrary precision.
If there was anything you did not understand about the mathematics in the previous
few paragraphs, you should probably learn it well before starting to try to learn QM from
this book.
To summarize: in QM, every vector |si of length one in the vector space representing the
system, is considered a possible state of the system. If that vector is an eigenstate of the
operator A, with eigenvalue a, then the theory predicts that a measurement of the quantity
represented by A will find the value a with probability one. If |si is not an eigenstate of A,
then the probability of finding the eigenvalue ai is |hs|ai i|2 .3 In Chapter 10, and elsewhere

3
This assumes all the eigenvalues are different. If some of them are the same, the formula is summed over
all eigenvectors with the same eigenvalue.
 Introduction  9

throughout the book, we will discuss at some length the meaning of the words probability
and measurement that we used in the preceding paragraphs.

1.3 THE ESSENTIALS OF PROBABILITY THEORY

Before moving on, let us review the essentials of the classical theory of probability. We will do
this for a system that has only a finite number of states, labeled by an integer 1 ≤ i ≤ N . For
a simple coin flipping experiment, we have N = 2. The mathematical definition of infinity and
of continuous variables always involves a limit from finite systems, so that measurements of
any finite precision will never be able to distinguish a system described by an infinite number
of states from such a finite system.
A mathematical probability distribution for a finite system is simply a function p(i)
defined on the finite set of integers, such that p(i) ≥ 0 and i p(i) = 1. There are,
P

roughly speaking, two approaches to the physical interpretation of p(i), which we will call
the Bayesian and frequentist interpretations of probability. The Bayesian approach is tied
to psychology. p(i) represents the expectation that if one tries to determine the state of the
system, one will find the state i. In the frequentist approach, one defines the probability in
terms of repeated trials. One flips a coin K times, determines the fraction of times it comes
up heads fK (1) and defines fK (2) = 1 − fK (1), and then takes the limit K → ∞ of fK (i) to
be p(i). One can think of the Bayesian interpretation as the theoretical model of the proba-
bility distribution and the frequentist definition as the experimental method for testing the
theory.
The problem is that we can never really take K to infinity. So no actual experiment can
carry out the rigorous frequentist definition of probability. If one has a theory of p(i), like
“the coin is not weighted, so p(1) = p(2),” and one finds 20 million heads in a row, one
cannot say that the theory is wrong, because the theory predicts that there is a probability
2−20,000,000 that the first 20 million tosses will come up heads. All you can say is that “the
probability that the unweighted theory is wrong is very close to one.” From a strictly logical
point of view, this means that the frequentist definition of probability is circular for any finite
number of experiments. On the other hand, there is a clear sense in which, for K this large,
one is close to the required limit, and one should simply say that the “equally weighted”
theory is wrong.
Indeed, in most real systems, one tries to determine the state one is in by measuring
variables that nominally take on all possible real values. This means that the possible values
one can find by doing the measurement are distributed more densely than the precision
of the measuring apparatus. In such systems, there is an unavoidable measurement error in
determining what the state of the system is. Thus, experimental results are quoted with “error
bars.” This means that the results of any experiment are themselves given by a probability
distribution. Experimental physicists work hard to eliminate or estimate “systematic errors,”
10  Quantum Mechanics

which are caused by defects in the apparatus or biased changes in the environment which skew
the results in a particular direction. They then assume that the environmental factors over
which they do not have systematic control are given by a Gaussian probability distribution
2
P (x) ∝ e−a(x−x0 ) , where a is estimated from the results of multiple trials of the experiment.
In the exercises below, you will see one example of how Gaussian probabilities arise in a very
general way when one is dealing with a large number of random events.
If one has only a probabilistic theory of what the results should be, this probability
distribution has to be combined with the experimental probability distribution of what the
results actually were. One displays the comparison of theory to experiment as a graph with
various bands surrounding a line of theoretical predictions for the functional relationship
between two measurable quantities. The bands represent confidence intervals, which take
into account both the theoretical and experimental uncertainties in the problem.
As an example, the Large Hadron Collider at the CERN laboratory in Geneva, Switzer-
land, announced in 2012 that “the Higgs boson had been discovered at 5 standard deviation
level with a mass (times the square of the velocity of light) of about 125 × 109 eV,” the level
particle physicists consider a significant discovery of a new particle. Five standard deviations
correspond to probability of about 1 in 3.5 million. This is the probability that if the Higgs
particle does not exist at a mass close to that, the data that CERN scientists collected in
Geneva, Switzerland, would be at least as far from the predictions of a theory where the
Higgs particle had a significantly different mass as this data is, because of a random sta-
tistical fluctuation. The theoretical predictions here depend on QM, which is intrinsically
probabilistic and so that random statistical fluctuation could be either a reflection of that
theoretical probability or of some uncontrolled random event in the experimental apparatus.
Readers should note the extreme care and almost legalistic precision with which one must
phrase scientific conclusions, if one wants to be completely accurate. Usually, scientists use
shorthand phrases like “5 sigma” to refer to such a discovery. The assumption is that anyone
professional will understand the small print and ultimately probabilistic guesses that go into
any statement about a discovery.
In classical probability theory, the distribution p(i) represents uncertainty about some of
the factors that determine the state of the system at a given time. One is given a box into
which 6 red, 5 blue, and 17 black balls have been inserted, and blindly reaches in and pulls
5 6
out a ball. It could be any color, but the probabilities of blue, red, and black are 28 , 28 , and
17
28 , respectively. Here, the uncertainty is encoded in the word blindly. If the person reaching
in were able to feel colors with their hand, then this would not be a problem in probability
theory.
A more amusing example is recounted in the book The Eudaemonic Pie [4]. Some of the
inventors of chaos theory had discovered experimentally that roulette wheels obeyed a form
of “low-dimensional chaos.” That is simply the statement that the apparently random fall of
the ball into different numbered slots was actually fit to high accuracy, by an equation with
 Introduction  11

a small number of parameters, which one could determine by observing a reasonably small
number of spins of the wheel. The odds in the casinos in Las Vegas and Reno were set by
assuming that the probability distribution for the fall of the ball was completely random. The
principle of odds making is to set them so that the customer wins sometime, but assuming
the distribution is random, the net flow of money goes to the casino. One does not want to
skew them so badly that no one will play the game. The chaos theorists were able to beat the
odds because they could come up with a better ansatz for the distribution, which overcame
the slight edge the casino owners had built in to the odds.
So, in classical probability theory, the initial distributions themselves must be determined
experimentally, or guessed on the basis of some theory of completely random events. There
is a variety of such general theoretical models, appropriate to different assumptions about
the randomness.
Given a probability distribution for the initial state of some system, one makes predictions
in physics by writing down equations of motion. These take the form
i(t + 1) = g(i(t)). (1.4)
Here, the 1 refers to some unit of time. Our insistence that there are only a finite number of
states means that we can only contemplate discrete time evolution, but the time unit could be
as small as we like. One usually studies evolution equations that are invertible. This means
that one can follow the evolution backwards or forwards, starting from any time t. Note
that this is a much weaker requirement than time inversion symmetry, which says that the
evolution in the backward direction is the same operation as that in the forward direction.
Invertibility is the same as saying that g(i) = S(i), where S is some permutation which
exchanges each state label i with exactly one other label.
Classical physicists assumed that in such a situation, the only detectable properties that
the system has were simply functions f (i) which take values on each state. The evolution
law for such a function is just
f (i, t + 1) = f (i(t + 1)) = f (g(i)). (1.5)
The probability distribution p(i) looks just like another function, satisfying some constraint
equations, but its evolution law is different.
p(i, t − 1) ≡ p(g(i)). (1.6)
That is, the time evolution for the probability distribution goes backwards, relative to the
evolution of the detectable quantities. To see why, note that the average or expectation value
of any function at time t is calculated in terms of the value of the probability distribution at
the initial time, t = 0 via the equation
p(g −1 (j))f (j) =
X X X
hf (i, t + 1)i = p(i)f (g(i)) = p(j, t + 1)f (j), (1.7)
i j j
12  Quantum Mechanics

where we have used the fact that the evolution is invertible to redefine the summation variable
by j = g(i). This says that we can view the time evolution of the system either in terms
of the time dependence of physical quantities with a fixed probability for initial conditions,
or as a time-dependent probability distribution to be in any given state with the value of
a given quantity f (j) completely fixed at the initial time. In QM, as we will see, these two
different ways of thinking about time evolution are called the Schrödinger and Heisenberg
pictures.
The second point of view allows us to think about a more general situation in which some
of the parameters that determine the function g(j) are themselves uncertain. Most generally,
we can make these transformations depend on time, which we denote by gt (j) and let the
time-dependent transformation be random, with some prescribed probability distribution.
One can show that if the probability distributions for those variables at different times are
independent of each other, then one gets a similar evolution equation for the time dependence
of the probability distribution. Similar means that the equation still relates the probability
distribution at time t + 1 to that at time t, and that it is linear as a function of p(j, t). The
linearity in p is the crucial property that allows us to define probabilities for histories and
formulate Bayes’ rule for conditional probabilities.
A history i(t) is simply some particular sequence of states. The probability of such a
history, stretching from t = 0 to t = k is simply
k
Y
P [i(t)](k) = p(i(t), t). (1.8)
t=0

The square brackets in P [i(t)] indicate that this probability depends on all the intermediate
points. We will now show that one can write p(i, t) as a sum over all possible histories,
weighted by the probabilities for individual histories.
The space of all complex valued functions f (i) is a vector space.4 A basis in this space is
the set of functions
ej (i) = δij .
We can write the probability distribution as
N
X
p(i, t) = pj (t)ej (i). (1.9)
j=1

The mapping i → g(i) acts on the basis by

X
ej (g(i)) = Sjl el (i), (1.10)
l

4
For the present section, we need to invoke only real-valued functions, but in QM, we will need the more
general space.
 Introduction  13

where Sjl is a permutation matrix. Its nonzero elements are a 1 in the j-th row and l-th
column, where j = g(l). There is exactly one nonzero element in each row, and they are all
in different columns.
Now let us rewrite our time evolution equation

p(i, t + 1) = p(g −1 (i), t) = pj (t)(S −1 )lj el (i).

X
(1.11)
j,l

In terms of the coefficients in the expansion of p(i, t) in terms of basis functions

X
pj (t + 1) = (S)lj pl (t). (1.12)
l

We can now iterate this equation to get the full solution in terms of initial conditions
l(t) l(t−1) l(1)
X
pj (t) = Sj Sl(t) . . . Sl(2) pl(1) (0). (1.13)
l(1)...l(t)

You will verify in Exercise 1.8 that the multiple sum over indices in this formula is precisely
a sum over probabilities of histories, assuming that the histories are related to the initial
condition by the equation of motion i(t+1) = i(g(t)). In Exercise 1.9, you will show that if the
matrices S at different times are not necessarily the same, but are picked independently from
a random ensemble of permutations whose probability distribution is chosen independently
at each time, then the same sum over histories formulation is correct as long as we introduce
the same randomness into the evolution law of the histories.
The sum over histories solution to the evolution law for probability leads directly to Bayes’
law of conditional probability. Suppose we consider some intermediate time 0 < ti < t. Then
we can divide all histories into those whose state at ti lies in some particular subset Σ of the
set of all states, and those which do not. We can then say that if we make an observation
at ti , which verifies that the state at that time lies in Σ, then we can throw out all the
histories that do not satisfy that condition, and define a conditional probability distribution
over the subset Σ. That distribution is the sum over restricted histories, divided by a factor
that accounts for the fact that the total probability of the restricted histories is less than one
in the original distribution. Bayes’ conditional probability rule is the instruction to construct
such a distribution, based on observation.
Readers should be alert to the fact that the discussion above contains the seeds of QM.
It shows that we can reformulate all of conventional classical physics in terms of linear
transformations, the matrices S, on a vector space. In Exercise 1.10, you will show that as a
consequence of the special properties of permutation matrices, all functions of the probability
distribution satisfy the same equations we have derived above. QM replaces the permutation
matrices by more general unitary matrices acting on a complex vector space. Functions of
14  Quantum Mechanics

the original distribution no longer satisfy the sum over histories rule and the functions that
do satisfy the rule are complex functions called probability amplitudes and cannot be thought
of directly as probability distributions. There is a unique sensible definition of a probability
distribution that one can construct from these complex functions. This definition is called
Born’s Probability Rule, and it says that a probability is the absolute square of a probability
amplitude. Since the absolute square of a sum of complex numbers is not, in general, the
sum of their absolute squares, probabilities in QM will not satisfy the sum over histories rule,
which allows us to define conditional probabilities. The fact that Born’s rule has the form
of an absolute square is just a consequence of the fact that Pythagoras’ theorem in complex
inner product spaces defines, for every unit vector, |ei, an infinite number of functions, one
for each orthonormal basis in the space. These functions are just the absolute squares of
the projections hei |ei of the vector |ei on the basis vectors |ei i. They are all nonnegative,
and for each basis they sum up to one. That is, they have the mathematical properties
that we would assign to the phrase “|hei |ei|2 is the probability to find that the system is
in the state represented by |ei i, assuming that we have determined it to be in the state
represented by |ei”. QM follows from assuming that this sentence applies to systems in the
real world. It violates the assumption of traditional logic, that we define different states of
the system by saying that we have determined that if we are in one state, then we cannot be
in another. This assumption is based on an incorrect extrapolation of macroscopic experience,
namely that we can always measure all of the properties that might determine the state of a
system, simultaneously. The task of “understanding QM” really amounts to demonstrating
that certain systems obeying the rules of QM can behave like the idealized systems of classical
logic, to a sufficient degree of accuracy to account for our missing the correct rules. We will
see that the key to demonstrating this is that typical macroscopic objects are composed
20
of > 1020 atoms. Such a system has c10 quantum states with c ≥ 2, and under normal
conditions (temperatures far removed from absolute zero), the system explores a double
exponentially large number of those states. Collective variables, averages over all the atoms
like the center of mass position, then obey the sum over histories rule of classical probability
theory, with double exponential accuracy.

1.4 PHILOSOPHICAL INTRODUCTION

The rest of this introduction is important, but it is not important to read it before you
start the meat of the book. If the philosophizing makes you impatient, skip to the next
chapter and come back to this at your leisure. Its main message can be summarized in
a couple of sentences. QM is an intrinsically probabilistic theory. The randomness of the
world at the microscopic level cannot be attributed to our inability to measure everything
with sufficient accuracy, but stems from the mathematical definition of the theory. The
theory of QM identifies certain properties of macroscopic objects, made of large numbers
 Introduction  15

of constituents, which obey the laws of classical probability theory (where all probability is
attributed to ignorance/measurement error) with incredible accuracy. That accounts for the
apparent classical nature of the macroworld we live in, which shapes all of our intuitions.
Now, go on or skip, at your discretion (but please first read the two Feynman chapters, which
are mentioned at the end of this chapter).

1.4.1 The Essentials of Quantum Mechanics

QM is the most confusing subject in the world, because it seems to deny the very foundations
of logic. Logic is a precise distillation of our intuitive grasp of how things work. It seems
to have nothing to do with particular physical situations where we have become used to
the fact that our intuition is only an approximation. For example, Galileo appreciated that
Aristotle’s intuitive notion that rest is the natural state of bodies was wrong, and invented the
relativity of frames moving at uniform velocity. Einstein realized that Galileo’s laws relating
the kinematics of two frames moving at uniform velocity were only approximately correct,
valid when the velocity was much smaller than that of light. All of this is a bit confusing
when you first encounter it, but it is not mind boggling, and once you understand how the
correct formulae reduce to the nonrelativistic ones, which make “intuitive sense,” when the
velocity is small compared to that of light, it is pretty easy to come to terms with relativity,
and even develop an intuition for it.
But how could a similar situation hold for LOGIC? How could logic be “a little bit
wrong”? The key to answering these questions involves the notions of probability and uncer-
tainty. We are used to assuming that at any one time, measurable quantities have definite
values, and that given values of enough quantities, one can predict what the values of anything
else will be in the future. The essence of QM is that this is not true. The fact that a quantity,
which has been measured, had a definite value at the time it was measured, is tautological.
However, our experience tells us that in order to predict the future value of something, we
need to know not only its current value, but also something else, its “rate of change.” For
example, Newton’s equations predict the future motion of a particle, given the present values
of both its position and velocity. It turns out that in QM you cannot determine the precise
value of both of these quantities at the same time. Measuring one with precision introduces
an UNAVOIDABLE uncertainty into the other. This uncertainty has nothing to do with
the clumsiness of our measuring apparatus. It is built into the fundamental mathematical
structure of QM and the definition of what velocity is in that structure.
At the time QM was invented, probability was a concept familiar to classical physicists,
gamblers, and insurance brokers. The world seems to be full of events that seem unpredictable,
and we are naturally interested in assessing the risk that some disaster or piece of good luck
will occur. Classical physicists always assumed that the reason for this unpredictability was
our inability in practice to measure all of the variables necessary in order to make predictions.
16  Quantum Mechanics

This was particularly plausible given the picture of ordinary objects as large collections of
microscopic constituents, whose miniscule size and vast numbers made the measurement of
their properties problematic. So, by the 19th century, there was a well-developed theory
of probability, based on the assumption that everything was precisely measurable and pre-
dictable, in principle, but rarely in practice. One of the problems of this theory is that the
probability distributions cannot usually be known in advance. One must make guesses about
what they are, and refine those guesses based on data collected about the system under
observation. However, it turns out that fairly general mathematical assumptions about prob-
ability distributions enable one to make successful statistical predictions about uncertain
events, and to estimate the likelihood of those predictions being right. In the limit of a large
number of independent trials, most distributions are well approximated by a normal or Gaus-
sian distribution. A great popular account of this theory can be found in [1] and a classic
treatise on the mathematics is the two volume work of Feller [2].
One of the fundamental laws of this classical probability theory is Bayes’ law of conditional
probability. Without going into the details, Bayes’ law says, essentially, that once one has
measured a definite value of some quantity, one can define new probability distributions in
which that quantity has no uncertainty and proceed to calculate the probabilities of other
things. Philosophically, this fits perfectly with the assumption that everything could have
been known in principle, and that it is only our laziness and incompetence, which prevent
us from making completely accurate predictions. A classic example of Bayesian conditional
probabilities is the probability that a given hurricane will hit a particular city. Early on
in the evolution of hurricane Katrina in 2005, there was some probability, according to the
equations used by the National Weather Service of the United States, that it would hit
Galveston, Texas instead of New Orleans, Louisiana. These are two cities about 400 miles
apart on the Gulf Coast of the United States. Once the hurricane hit New Orleans, one could
define a new probability distribution in which the probability of hitting Galveston was zero,
and that of hitting New Orleans was one, and make future predictions for the evolution of
the storm given that new piece of data.
There is a mathematical fact, associated with the use of Bayes’ law, which will be useful in
understanding QM. If we write the equations of probability theory as differential equations,
then they have the kind of locality we are used to from the equations for fields in classical
physics. The time derivative of the probability distribution at a point depends only on the
behavior of the distribution at very nearby points, so that the influence of a change at a point
takes time to propagate to distant points. However, when we use Bayes’ law, an event at one
point (observation of a hurricane hitting New Orleans) immediately changes the value of the
probability distribution at a distant point (Galveston), changing the value there to zero in a
way that is not prescribed by the equation. There is of course nothing funny going on here.
The probability distribution is not a physical field, it is an expression of the uncertainty in the
position of the storm. Changing it to zero at Galveston is just adding another piece of data,
 Introduction  17

to our algorithm for predicting what will happen in the future in this particular sequence of
events. The theory does not predict what will happen in that particular sequence, but only
the frequency of occurrence of different outcomes in the limit of an infinite sequence of reruns
of the same initial data. The philosophy behind the use of Bayes’ rule in this instance is that
in principle we could have predicted that the hurricane would hit New Orleans. The reason
that we did not is that we did not know all the initial conditions and so our probabilistic
equations included initial conditions that would have led the hurricane to hit Galveston.
Mathematically, the reason that we are able to use Bayes’ law on solutions of the National
Weather Service equations is that these are linear equations for the probability that the
hurricane will be in a certain place at a certain time. If we think about all possible histories
of the system, the linearity of the equations leads to a rule

• The probability P (x, t) for the hurricane to be at a particular place at a particular time
is equal to the sum of the probabilities of all histories x(s) (for 0 ≤ s ≤ t) that have
initial conditions whose probability is nonzero in the original distribution P (x, 0), and
that have x(t) = x.

Probabilities in QM do not obey the laws of classical probability theory, and in particular,
they do not in general obey Bayes’ rule or the history sum rule above. Given any system,
there is a complete set of compatible detectable quantities,5 which could, in principle, be
measured with absolute precision at the same time. There are actually a continuous infinity
of different compatible sets, but when one set is in a state where its members have definite
values, the others all have uncertain probability distributions. There are two striking things
about these probability distributions, when compared to those of classical probability theory.
The first is that they are all quite definite mathematical functions, completely determined
by the values of the measured detectables. In classical probability theory, the distributions
must be discovered from experiment and only take on a priori functional forms in certain
limits. More striking and much more peculiar is the fact that QM probability distributions do
not, in general, satisfy the Bayesian conditional probability rule, which allows us to replace
uncertainty by certainty when a measurement has been made.
We will see that the last feature has profound philosophical consequences, which are best
summarized in a phrase invented by Rutgers professor, Scott Thomas, “Objective Reality is
an Emergent Phenomenon.” Emergent concepts are basic objects in an approximate theory
of nature, which have no counterpart in a more exact theory which underlies it. The notion of
water as a continuous fluid is a prime example. Real water is made up of discrete molecules.
The nature of the emergence of reality in QM is still a matter of some debate, and there
are many researchers who hope to find at least an interpretation of QM which allows for
5
The traditional word to use here is observable, rather than detectable. Detectable is preferable because it
does not hint that the “observation” must be performed by a conscious observer.
18  Quantum Mechanics

some kind of underlying reality. We will defer most of our discussion of these abstruse issues
to Appendix A. However, it is important to stress that all popular discussions of nonlocal
action of things on each other in QM has to do with such interpretations, rather than with
the use of QM as a probability theory to make predictions about the results of experiments.
Let us be a little more precise here about the meaning of the phrase Objective Reality
in Thomas’ aphorism. In classical probability theory, there are probabilities for histories
of any system. If zi represent all possible detectable quantities characterizing the system,
then a history is a time-dependent set zi (t). Even if we divide the system’s characteristics
up into visible quantities va (t) and hidden ones hA (t), there will still be probabilities for
histories of the va (t). Those probabilities will satisfy the “obvious” rule that if one makes
an observation at time t0 , one can divide up the histories according to whether they agree
with that observation or not and base future predictions on the result of that observation,
throwing out those histories which did not agree with that observation. This statement is
true in QM as well, as long as we actually make the observation. The italicized phrase is
the source of much of the confusion about the meaning of QM, because it seems to imply
some connection between human intervention and the basic laws of physics. We will learn
later that a more proper form of the phrase is as long as the “observation” is a quantum
entanglement between the microscopic property being observed and the collective coordinate
of some macroscopic object. The point of this long winded phrase is that the microscopic
rules of QM probability theory do not have the property that probabilities of histories can be
manipulated in this way. The probability of finding the values va (0) and va (T ) is not the sum
of the probabilities of all histories va (t) with those initial and final values. However, there
are special systems, called macroscopic objects, and certain variables, like the center of mass
position of such an object, for which the violations of this probability sum rule are incredibly
20
small (< 10−10 ). “Realist” interpreters of QM believe that a notion of probabilities for
histories, obeying the usual rules, is essential to a notion of Objective Reality. This is clearly
untrue if we take the commonplace meaning of objective reality, which is merely a record
of the things that actually happen in the universe, with no predictive power. It is only if
we want to make a predictive theory of what will happen given the maximum information
about what has happened, that we might want to use the notion of probabilities for histories
obeying the classical sum rule. Experiment shows that those predictions are wrong. QM is a
mathematically beautiful (and somewhat inevitable) alternative theory of prediction, which
does agree with experiment.
To those who agree with Prof. Thomas’ epigram, objective reality in the sense described
above, is not an exact property of any quantum variable. However, objects that we call
macroscopic are made of huge numbers of atoms N > 1020 . Such macroscopic objects have
collective coordinates, like the center of mass, which are defined in terms of weighted averages
over all the atoms in the object. One can show that the uncertainties in the values of the
collective coordinates are of order N −1/2 or smaller. Furthermore, the violations of the rules
 Introduction  19

of classical probability theory for the distributions of these collective variables, are, under
normal circumstances6 , of order e−N . So, collective variables behave a lot like classical objects
were supposed to behave in classical physics. Their values are not very uncertain, and the
uncertainty that there is can, with extraordinary accuracy, be mistaken for uncertainty due
to measurement error. According to this view, our brains and our bodies live in this fictitious
macroworld of certainty and it was not until we became sophisticated enough to probe the
atomic constituents of the matter around us, that we were forced to recognize the correct,
quantum mechanical, rules, which govern the world.

1.4.2 Unhappening
One of the most disturbing features of the emergent nature of the concept of “happening”
in QM is what can only be called “unhappening.” In classical probability theory, when we
use Bayes’ rule to throw away part of the probability distribution, and renormalize the part
we keep, we are doing the correct thing physically. That is, if we are only using probability
because of our ignorance, then every new piece of data about the world, which tells us “what
really happened,” should be used to reduce the uncertainty in our probability distribution. In
QM, this is not true. In QM, when we use Bayes’ rule, upon a single observation of a particular
value for some collective coordinate, to throw out the part of the probability distribution
that predicted another value, this only gives the correct answer for the probabilities of later
observations, as long as the macroscopic object (or some other macroscopic object whose
collective coordinates were determined by it) continues to exist. If the object disintegrates
into elementary particles, we must go back to the initial probability distribution, before its
truncation by the use of Bayes’ rule, to get correct predictions for future observations. We
will give a particularly poignant example of this in Chapter 10.
It is important to note that classical probabilities are just a special case of the probabilities
defined by QM, and classical mechanics is a special case of QM, defined by the requirement
that all quantities appearing in the equations of motion can be definite at the same time.
This means that in classical mechanics, we can define probabilities for histories of the system,
which allows us to say that some things definitely occurred in the past, by using Bayes’ law
to consider only probabilities conditioned on the results of past experiments. No real system
exactly obeys these classical rules, but the collective coordinates of macroscopic systems,
under “normal” conditions, obey them with such fantastic accuracy that experiments to
detect the deviations would take far longer than the current age of the universe (see Chapter
10) and require impossible amounts of isolation of the system and accuracy of the measuring
device. As noted in the previous paragraph, these statements only remain valid as long as
the macrosystems in question do not disintegrate into elementary particles.

6
We will be more precise about what we mean by normal circumstances in Chapter 10.
20  Quantum Mechanics

1.4.3 Quantum Mechanics of a Single Bit

We will begin our discussion of QM with the simplest possible system, one which has only
two states, corresponding classically to a single Yes/No question. As we will see, the molecule
of ammonia, NH3 can be approximated by such a system, in a certain energy regime. We
will see that all the machinery of QM, the mathematics of linear algebra, can be introduced
in a purely classical discussion of this system, as if we were computer scientists, discussing a
single bit. Quantities which a classical physicist/logician would think of as measurable can
be modeled as diagonal matrices, while classical operations which change the state of the
system are off diagonal matrices. A general classical probability distribution is a diagonal
matrix, and the usual formula for the expectation value of a quantity is written as a trace of
the product of the matrix for the quantity with that for the distribution. This trace formula
immediately generalizes to all matrices, even if they are not diagonal. For those matrices
which are diagonal in some orthonormal basis, the trace can be interpreted as a probability
distribution for that matrix to take on one of its eigenvalues.
This discussion will show that QM is, in a certain sense, inevitable. That is, for any
system, even one that we think of classically, we can introduce quantum variables, which
have uncertainty even when we have completely fixed the values of the classical variables.
The choice of which variables are definite corresponds, for a two state system, to a choice of
basis in a two-dimensional vector space. From the point of view of linear algebra, this choice
seems arbitrary. The difference between classical and quantum mechanics is in the nature of
their equations of motion. Classical mechanics has initial conditions which can all be definite
at the same time, while in more general quantum mechanical systems, only some of the initial
variables are compatible with each other. The collective coordinates of macroscopic quantum
systems do not obey classical mechanics, but they do, up to fantastically small corrections of
20
size < e−10 , obey a classical stochastic theory, in which the uncertainties are very small (of
order 10−10 or smaller). These mathematical facts about the quantum theory are sufficient
to explain why our intuitions about the logic of the world are incorrect.
The present author is among those who believe that we will never find a consistent
interpretation of the facts of the quantum world, which admits the concept of an underlying
reality with exact probabilities for histories. This is not a settled question, and we will try
to avoid injecting our prejudices into most of this text.

1.4.4 What is Probability?

QM adds confusion to what one may worry is already a complicated issue in classical probabil-
ity theory, namely how we should think about probabilities. The inventors of the theory, par-
ticularly those whose primary interest was in gambling or other financial transactions, clearly
thought of it as a sophisticated way of guessing an unpredictable future. This interpretation
 Introduction  21

is clearly tied up with human psychology. A discussion of probability from this point of view,
which attempted to make very precise rules about how to guess and how to use additional
data to assess and improve the quality of one’s guesses, was given by Bayes [3] in the 18th
century. In modern times, this view of probability is given the label Bayesian interpretation.
If you are a gambler or a financier, this is certainly the way you think of probability. Exper-
imental physicists and theorists who follow their work closely also use Bayesian reasoning
quite frequently. Looking at experimental lectures you will often see plots, which include lines
indicating the predictions of a theory, and colored stripes following the lines, which indicate
things like “the 95% confidence interval.” Translated into full English sentences, this means
the region of the graph where, with 95% probability the data actually lie, given all the possi-
ble random and systematic errors. In graphs referring to the behavior of microscopic systems,
these errors include the fact that the quantum theory does not make definite predictions for
the number of events of a certain type, but only predicts (see below) the ratio of the number
of events to the number of runs of the experiment in the limit that the number of runs goes
to infinity. There may also be different bands telling you that the theoretical calculation has
“uncertainty,” but this is a completely different sort of error and stems from the fact that we
can usually solve the equations of QM only approximately. The use of the word confidence
interval in this context is the reflection of the Bayesian outlook on the meaning of probability.
As probability theory became more and more important in science, scientists searched for
a more “objective” way of thinking about it, which removed the human psyche and words like
confidence from the unbiased description of nature by the combination of mathematics and
observation. This led to what is called the frequentist interpretation of probability. Accord-
ing to this paradigm, you test a prediction which gives probabilistic answers by repeating
your experiment N times, and recording the fraction of times the experiment produces each
possible result. As N → ∞, these fractions converge to the predicted probabilities if the
theory in question is correct. This is indeed an objective definition of probabilities, but it
is problematic, because it is impossible to take N to infinity, even if the universe lasts for
an infinite amount of time. To illustrate the problem, flip a coin 2000 times and observe
that it always comes up heads. The probability for that, assuming the coin is unbiased is
2−2000 ∼ 10−500 , a pretty small number, but this does not prove that someone has weighted
the coin. If you think it does, would you bet your life on it? Would you bet the lives of all your
loved ones? Would you bet the lives of the entire human race? Obviously, these questions
all have subjective answers, which depend on who you are and what your mood is. This is
the reason that experimental physicists, who test probabilistic predictions (or even definite
predictions that are tested with imprecise machinery) by applying the frequentist rule with
a finite number of trials, cite their results in terms of a Bayesian confidence interval. We can
try very hard to be completely objective about the data, but no finite amount of effort can
completely eliminate the need for “leaps of faith.”
22  Quantum Mechanics

These interpretational problems have nothing to do with QM. They would be there for a
completely deterministic theory about which we were ignorant of some of the initial data, and
since some of the initial data have to do with the performance of the measuring apparatus
itself, or external influences interfering with the machinery (cosmic rays, sound waves, the
electromagnetic field generated by a radio 4 km away, etc.), we are always ignorant of some
of the data. We continue to do experimental and theoretical physics despite these obstacles,
because we believe that we can control these sources of error well enough that we are happy
with the small size of the required leap of faith.
The interpretation of QM as a new kind of probability theory is certainly correct, and
is the only interpretation that has been tested by experiment. If it is the final word on how
to interpret the mathematics, then we will just have to live with the intrinsically indefinite
nature of probabilistic predictions. We will explore alternative explanations in Appendix A.

1.5 A NOTE ON MATHEMATICS

This subsection is addressed to two different groups of people. First to teachers of a more
conventional course in QM: When the Schrödinger equation for a nonrelativistic QM is writ-
ten in position representation, it is a partial differential equation. If we are discussing the
QM of a single particle, the Schrödinger equation is similar to Maxwell’s equations: it is
a differential equation in time and space. It is tempting, and some would argue pedagogi-
cally preferable, to utilize the students’ familiarity with electrodynamics as a crutch. Indeed,
some students have definitely expressed a preference for this approach. In the long run, it is
a mistake. The methods of partial differential equations (PDEs) are practically useless for
understanding complicated QM problems involving many particles or an indefinite number
of particles, and these are the vast majority of systems of interest in particle physics, nuclear
physics, and condensed matter physics. The solution of problems via operator algebra, which
is the approach taken in most of this book, is simpler (but more abstract and less familiar)
and introduces methods, which are more useful in real applications of QM.
Perhaps equally important is the fact that concentration on the single particle Schrödinger
equation and its mathematical analogy to Maxwell’s wave equation, misleads students com-
pletely about the nature of what is called wave–particle duality. Let me say it clearly once
and for all. The Schrödinger wave function is NOT a classical wave, but instead defines a
probability distribution.7 For N particle systems, the Schrödinger equation is a differential
equation in 6N + 1 variables, and is not a wave in space. Wave particle duality has to do with
the fact that multiparticle systems can be described by quantum fields: operators which obey

7
More precisely, the density matrix, which contains the entire physical content of the wave function, defines
a probability distribution for all normal operators in the Hilbert space of wave functions.
 Introduction  23

wave equations in space and time. These quantum fields have states that behave approximately
like classical fields, and other states that behave approximately like particles.
Secondly, for those with a mathematically rigorous turn of mind: I have known a lot of
brilliant mathematicians who had a hard time reading QM texts because of the nonrigorous
treatment of operators and Hilbert spaces. This book will be no exception. Physicists find
that excessive attention to mathematical rigor slows us down, and is difficult for most of
our students, who are more interested in the use of physics to understand the world. In my
opinion, the correct approach is that of von Neumann: if you are bothered by a statement
in a physics book, work out the correct explanation yourself. Much of the necessary rigor is
supplied in von Neumann’s famous book [5] of which a beautiful new translation has appeared
recently [6].
Finally, it should be emphasized that linear algebra is a prerequisite for this course. We
will review it, using the Dirac notation beloved by most physicists and rather less popular
among mathematicians, in Chapter 6, but you will be expected to know enough to follow
the first five chapters. You can of course skip to Chapter 6 to brush up on things, or in
extremis, to learn linear algebra from scratch, but that is not the best way to profit from this
book. Another part of mathematical physics that would be helpful to know is the subject
called analytical mechanics, which is to say, everything about Lagrangians, Hamiltonians,
Poisson brackets, and the Hamilton-Jacobi equation. We will not use much of this in the
book, but there is a short summary of it in Chapter 4, and the nomenclature will be helpful
in our discussion of particles in magnetic fields and in the chapters on Path Integrals and
the JWKB approximation.

1.6 FEYNMAN’S LECTURES

Before you go on any further, you should read the first two chapters of Volume 3 of the
Feynman lectures on physics. You can find them here: www.feynmanlectures.caltech.edu.
They illustrate the puzzling nature of QM in a beautifully simple way.

1.7 ACKNOWLEDGMENTS
In writing this book, I have benefited from a lot of help from colleagues and students. I would
like to thank Eliezer Rabinovici, Patrick Hayden, Daniel Harlow, Subir Sachdev, Sriram
Shastry, Michael Dine, and Jacob Barandes for important comments and advice on portions
of the manuscript. I would especially like to thank Satish Ramakrishna for providing very
careful solutions to many of the problems. Pouya Asadi and Conan Huang are responsible
for the figures. Ricardo Mulero provided me with the beautiful cover art. Finally, I would
like to thank my beautiful wife Anne for the patience she has shown as I worked on this book
and all the love and support she gives me every single day.
24  Quantum Mechanics

1.8 EXERCISES
1.1 The Let us Make a Deal Problem: On this famous quiz show a contestant is shown three
doors and told that behind one of them there is a fabulous prize. After the contestant
has chosen a door, one of the other doors is opened, and shown to have no prize (or
a booby prize) behind it. The contestant is then given the option to change the door
he/she has chosen. What is the best strategy for getting the prize?

1.2 If a couple has three children, what is the probability, given no further information,
that two of them are girls? Suppose you know that one of the children is a girl named
Florida. What is the probability that two of the children are girls? Does this depend
on the choice of name? FYI, Florida is a girl’s name that used to be popular, but has
gone out of fashion.

1.3 Drop a needle on a plane surface. Take the origin to be one end of the needle and draw
radial lines at angles 2πk
N , starting in a random direction. What is the probability that
the direction of the needle is between the k and k +pth radial line? How does the answer
change if one takes the origin to be at another point?

1.4 Here is a problem that is easy to state, but hard to analyze. Given a fairly weighted
coin, what is the probability that in 100 flips there will be a run of at least K heads in
a row? What if we ask for exactly K heads in a row? The way to set this up is to define
a quantity P (i) which is 1 if the ith throw is a head and zero if it is a tail. Each run
is characterized by a set of values for these 100 variables, and there are 2100 possible
values. The question we are trying to answer is, out of all of these possible “states of
the system,” how many of them have P (i)P (i + 1) . . . P (i + K − 1) = 1 for some value
of i. The trick is that we have to worry about double counting. E.M. Purcell, the Nobel
Prize winning Harvard physicist and author of a classic text on electrodynamics, solved
the much harder problem of determining the probability given, “dumb luck” of a player
having an n game hitting streak in baseball. At the time he did the calculation in the
1980s, the only streak in the history of baseball, which was not, “what we could expect
from random probabilities for hits and outs” was Joe DiMaggio’s streak in 1941.8

1.5 The probability of getting K heads in N throws of a fair coin is given by

N!
2N K!(N − K)!
8
This calculation is reported in a lovely essay by S.J. Gould [7]. You will also be touched by Gould’s essay
[7] on estimating his own chance of beating the odds after his cancer diagnosis. Note that the Wikipedia
article on Dimaggio cites some work disagreeing with Purcell’s calculations.
 Introduction  25

This is maximized at K = N/2 if N is even. Show that as N → ∞ the probability

of getting N2 (1 − x) heads, with x kept constant as N goes to infinity, is a Gaussian
distribution of the form
2
P (x) = Ae−bx .
Calculate the value of b. What happens if N is odd?

1.6 Suppose the probability of finding some quantity Q to have the value x is

P (x) = e−f (x) ,

where f (x) > 0. Given N independent copies of the system, show that as N → ∞,
the fluctuation of Q away from its expectation value goes to zero. For the system with
multiple copies, QAV is defined to be the average of the Q of the individual copies.
That is X xi
QAV = .
N
Show that fluctuations of Q away from that expectation value have, generically, a
Gaussian distribution with a width that scales to zero like N −1/2 . What characterizes
the nongeneric exceptions? This result is called the Central Limit Theorem.

1.7 Political polls often quote a “margin of error.” This is defined in the following way.
Suppose, in a very large population N , a fraction p of the population prefers candidate
Thomas Jefferson and the rest prefer his opponent. Imagine you take a random sample
of n  N voters and find a fraction pn prefer Jefferson. There are n!(NN−n)!!
different
random samples, which give a distribution of pn . A sample of n people voting for
Jefferson’s opponent would give pn = 0, whereas a sample containing only Jefferson
voters would give pn = 1. Show that for N  n  1, the distribution of pn is Gaussian
and find its center and width. The margin of error usually quoted is one half of the
95% confidence interval for the true percentage p. The margins of error for different
confidence intervals are simply related by the value of the inverse error function at
those confidence levels.

1.8 Use the fact that the only nonzero matrix element of a permutation matrix is a 1 in each
row, with a different column for each row, to show that the formula for the time evolved
probability distribution can be thought of as a sum over histories. The contribution from
a fixed value of each of the summed matrix indices l(k) is the contribution of a single
history.

1.9 Show that the sum over histories formulation still works, if at each time the permutation
matrix S is replaced by a time-dependent matrix S(t), chosen from some probability
distribution of with no correlation between different times. The dynamics of the histories
26  Quantum Mechanics

is simply replaced by i(t) = g(i(t − 1), t), where the permutation g is chosen from the
same random distribution.

1.10 Use the fact that the only nonzero matrix element of a permutation matrix is a 1 in
each row, with a different column for each row, to show that p2 (j, t), or indeed any
function f (p(j, t)) satisfies the same sum over histories formula as p(j, t) itself.
 CHAPTER 2

Two State Systems: The

Ammonia Molecule

2.1 INTRODUCTION AND MOTIVATION

In this chapter, we will use a simple system with two states to motivate the claim that the
mathematical structure of Hilbert space (a vector space equipped with a positive definite
scalar product), the basis of quantum mechanics (QM), is implicit in even the classical view
of such a system. We will see that both the classical notion of a detectable quantity, whose
values differentiate among the states of a physical system, and the classical notion of an
operation, which changes the physical state of the system, are special cases of linear operators
on the Hilbert space of states of the system. Both detectable quantities and operations come
equipped with a natural notion of product and the algebra one gets by taking all possible
linear combinations of products of detectables and operations is the algebra of all linear
operators. Among those, the so-called normal operators are those whose matrix in some
orthonormal basis is diagonal. The original detectable quantities are simply those which are
diagonal in a particular basis. Any normal operator can be said to take on values if we allow
for the possibility that any normalized vector in the Hilbert space is a possible state of the
system.
Even more remarkably, any such normalized vector defines a probability distribution for
every normal operator to take on each of its allowed values. This separates the notion of
a probability theory from the classical context in which it was invented, i.e., as a way of
codifying our lack of knowledge about the state of all relevant variables, which determine the
time evolution of the system in question. In QM, probability and uncertainty are intrinsic to
the mathematical form of the theory, rather than concepts we introduce to parameterize our
ignorance.

27
28  Quantum Mechanics

|+i 3 = |−i 3 =

Figure 2.1 Model of the ammonia atom.

Classical physics is then seen as the special case of QM where the evolution of the state
of the system with time simply permutes the elements of a particular basis, so that the
subalgebra of the algebra of all linear operators, consisting of operators diagonal in that
basis, always takes on definite values.

2.2 THE AMMONIA MOLECULE AS A TWO STATE SYSTEM

In any chemistry department, you can find little models of molecules made out of colored balls
and sticks. The balls represent atoms, the sticks the chemical bonds between them. These
models, as we will learn, are only crude visualizations of the real properties of molecules,
but the model of ammonia, a pyramid with a nitrogen atom at its apex and three hydrogen
atoms arrayed in an equilateral triangle beneath it (Figure 2.1), will allow us to set up the
QM of ammonia in a certain energy regime. We imagine a situation where the molecule is
isolated from the rest of the world. Excitations of the molecule around its lowest energy state
could be classified according to its states of motion. First of all, we could imagine excitations
of the individual electrons in its constituent atoms.1 It turns out that such excitations have a
characteristic energy scale ranging from 1 to 103 electron-volts (eV). Next come vibrational
excitations, in which individual atoms oscillate around their equilibrium positions in the
molecule (which are the positions assigned to the atoms in the Chem Lab Model), and
overall rotations of the molecule, as if it were a rigid body. We can also move the center of
mass of the molecule, but in the absence of external forces, its motion will have constant
velocity, and we can always work in an inertial reference frame where the center of mass is at
rest. The excitation energies of rotational and vibrational motions have characteristic energy
scales which are smaller than atomic excitations by powers of the electron mass divided by
the masses of the nuclei.
In classical physics, one could get arbitrarily low energy of rotation or vibration by letting
the atoms oscillate slowly, or continuously lowering the angular momentum. We will see later
that QM quantizes these energy levels so that any transitions between rotational, vibrational,
or electronic excitation energies are quantized multiples of the characteristic scales. However,

1
At a deeper level, we could imagine excitations of the nuclei and of the individual protons and neutrons in
the nuclei. These may be neglected for reasons similar to those we will discuss in Chapter 11 for electronic
motions.
 Two State Systems: The Ammonia Molecule  29

if we assume that the physics of an isolated ammonia molecule is invariant under space
reflection x → −x, then classical reasoning leads us to the conclusion that there are, for
each state of atomic, rotational, or vibrational motion, actually two states of the molecule,
degenerate in energy, corresponding to configurations related by reflecting the nitrogen atom
through the plane of the hydrogens (Figures 2.2 and 2.3).
Let us denote these two states by the symbols |±i, where we suppress the information
about the particular state of vibration, rotation, and atomic excitation. |+i is the state where
the nitrogen atom is in the positive 3 direction. This direction is defined to be perpendicular
to the plane of the hydrogens, and having an orientation defined by the right-hand rule. |−i
is the mathematical representation of the state with the opposite orientation. This funny
notation for a state of a system was invented by Dirac. It is an incredibly useful way of
encoding the notion of what a state is, in QM, but it is an equally valid tool in classical
mechanics. The reason it is so useful is that, mathematically, states are really lines in a huge
vector space, and Dirac’s notation is a very quick way to mechanize the operations you can
do in vector spaces. But we do not yet know any of that, so let us just accept the notation
as the professor’s quirk, for the moment. Alternatively, you can consult the last appendix
of the book, where you can find a quick summary of Dirac Notation and linear algebra, or
jump to Chapter 6, where you will find an extensive exposition of these subjects.
According to classical reasoning, these two states would have definite energies, and the two
energies would be equal, by reflection invariance. A detectable quantity which distinguishes
between the two states is the electric dipole moment, D. Since it is odd under reflection,
D has the values ±d in the two states.

|+i 3 =

Figure 2.2 The up state of ammonia.

| − i3 =

Figure 2.3 The down state of ammonia.

30  Quantum Mechanics

2.3 PHYSICAL QUANTITIES AS MATRICES

A crucial step on the way to QM is the observation that we can think of the two states as
vectors in a two-dimensional vector space
!
1
|+i = , (2.1)
0
!
0
|−i = , (2.2)
1
and the measurable quantities as matrices

E = e1 D = dσ3 ,

where !
1 0
1= , (2.3)
0 1
!
1 0
σ3 = . (2.4)
0 −1
The rule is that the value of a given quantity in a given state is gotten by acting with the
matrix representing that quantity on the vector representing the state, obtaining the value of
the quantity, times the original vector. That is, the states are eigenstates of the energy and
dipole moment matrices, with eigenvalues equal to the values of the corresponding quantities.
Our notation |±i for states is known as Dirac Notation. Dirac invented it as a symbol
for column vectors, and called such vectors kets. We will eventually be discussing systems
with more than two states, so we might as well introduce the notation in that more general
context. For any n-dimensional column vector
v1
 
 .. 
|vi ≡  .  ,
vn
there is a corresponding row vector
 
hv| ≡ v1∗ . . . vn∗ ,

and the scalar product between two vectors is denoted

n
vk∗ wk = hw|vi∗ .
X
hv|wi ≡
k=1
 Two State Systems: The Ammonia Molecule  31

Dirac called row vectors bras, so that the scalar product is a bra-ket or bracket. It is one
of those jokes that only sound funny to physicists. The real virtue of Dirac notation is that
it allows us to construct a self-explanatory notation for the linear operator Dvw called the
outer product or dyadic formed from two vectors. Dirac’s notation is

Dvw = |vihw|.

In words, what this operator does to a vector |ui is to take its scalar product with |wi
and multiply that number times the vector |vi. In particular, if |ei is a vector with norm 1
(he|ei = 1) then
|eihe|
is the projection operator, which, for any other vector |vi gives the projection of |vi along
|ei.
Let us take a moment to recall the definition of the term linear operator, which was used
in the previous paragraph. In any vector space over the complex numbers, we can construct
linear combinations of a pair of vectors |vi and |wi according to the formula

a|vi + b|wi.

Thinking about vectors as n-tuples of complex numbers, you know exactly what this means.
A linear operator is a mapping that takes any vector |vi into another vector T |vi. Linearity
means that the map satisfies the rule

T [a|vi + b|wi] = aT |vi + bT |wi. (2.5)

The representation of vectors as n-tuples of complex numbers depends on a choice of what is

called a basis in the space of all vectors. We can consider the vectors |ek i whose representation
as a column of numbers has a 1 in the k-th row and zeroes everywhere else. Then the complex
column representing |vi can be thought of as the numbers that appear in the expansion
X
|vi = vk |ek i.
k

With our definition of the scalar product, the vectors |en i satisfy hem |en i = δmn , which is
the definition of a basis that is orthonormal, shorthand for “orthogonal and normalized to 1.”
The concept of basis should be familiar to you from two- and three-dimensional geometry.
Unit vectors in the x, y, and z directions form an orthonormal basis of three-dimensional
space. From these examples, you are familiar with the fact that orthonormal bases are not
unique. We can rotate them in any direction in three-dimensional space. The same is true
of the n-dimensional complex vectors that are of interest in QM. We call a linear operator
U , a unitary transformation or unitary operator, if the vectors U |en i ≡ |fn i also form an
32  Quantum Mechanics

orthonormal basis. We will explore the conditions for a transformation to be unitary in the
following chapters.
Given an orthonormal basis and a linear transformation T , we can construct the square
array of numbers
Tmn = hem |T |en i. (2.6)
This array is called the matrix of the transformation (or the operator) T in the |en i basis.
You should convince yourself that you can think of Tmn as the coefficient of |em i, in the
expansion of the vector T |en i in the |en i basis.
Returning to our two state ammonia molecule, we say that we have identified the energy
and electric dipole moment as linear operators in a space with orthonormal basis |+i and
|−i, and that the matrices of these operators are diagonal in that basis. We can also think
of a probability distribution for the different states as a diagonal matrix ρ, with eigenvalues
p± ≥ 0, p+ + p− = 1. For any probability distribution, the expectation value of a quantity,
which takes the values Ai in the i-th state of a system, is just
X
hAi ≡ Ai p i .
i

Another term for this is the average value of the quantity A, but in QM, we always use the
term expectation value. Exercise (2.5) asks you to show that the expectation value, hP i, of
any polynomial in the quantities E and D is just

hP (E, D)i = Tr P (E, D)ρ.

Recall that the trace of a matrix is the sum of its diagonal elements. It is important to
note that, despite the appearance of matrices, all of these formulae are just a compact
mathematical way of discussing the properties of a bit in classical logic. A bit in classical
computer science (Turing) is the same as an irreducible Yes/No question. This is a question,
which completely characterizes the state of a system. By definition, if you can characterize a
system by the answer to a single Yes/No question, then that system has two states. So, our
low-energy ammonia molecule provides a mathematical model applicable to any two state
system, in physics, logic, or computer science.
P (E, D) is just a general function of the two physical variables, energy and dipole
moment, one of which takes the same value in both states of the system and the other
of which takes opposite values. Any quantity, which takes definite values in the two states
is a function of these two (prove this in Exercise 2.15). The eigenvalues of ρ are just the
probabilities that the system is in each of its possible states, and the trace formula is just the
usual statement that the expectation value is the sum of the possible values of a quantity,
weighted by the probability that it takes on each of these values. At this point, you should
do Exercises 2.1 and 2.2, which allow you to refresh your memory of matrix multiplication.
 Two State Systems: The Ammonia Molecule  33

Exercise 2.1 The action of a matrix on a vector is denoted by the algebraic formula
v i → j Mji v j . Use this formula to evaluate the action of the matrix
P

!
0 a
(2.7)
b 0

on the vector !
c
. (2.8)
d

Exercise 2.2 The product of two matrices M and N is denoted M N and is defined by
letting M act on each of the column vectors making up N , according to the rule in Exercise
2.1. In symbols,
X
(M N )ij = Mki Njk . (2.9)
k

Evaluate the product of two general 2×2 matrices and show that M N 6= N M . The difference
M N − N M ≡ [M, N ] is called the commutator of the two matrices. The inverse of a matrix
is defined by the equation M M −1 = 1, where the matrix 1 is the unit matrix, which has
ones on its diagonal and zero everyplace else. Find the inverse for a general 2 × 2 matrix, if
it exists, and describe the criterion for the inverse to exist. Show that the commutator of a
matrix and its inverse vanishes.
Answer to Exercise 2.2: The equation for the inverse is
! ! !
−1 a b e f 1 0
MM = = . (2.10)
c d g h 0 1

The product of the two matrices is

!
ae + bg af + bh
. (2.11)
ce + dg cf + dh

The equations for the inverse are thus

ae + bg = 1 = cf + dh, af + bh = 0 = ce + dg. (2.12)

So, g = −(c/d)e, h = −(a/b)f , and 1 = (a − bc ad

d )e = (c − b )f . The last two equations
have solutions if and only if ac − bd 6= 0. This combination of matrix elements is called the
determinant of the matrix M and denoted det M . It measures whether the rows and columns
of the matrix are linearly independent vectors. The inverse matrix is
34  Quantum Mechanics
!
−1 −1 d −b
M = (ad − bc) , (2.13)
−c a

and it is easy to verify that M −1 M = 1.

In classical probability theory, a state is called pure if it corresponds to probability one
for one of the elementary states, and zero for all of the others. If a state is not pure, it is
called mixed. In matrix language, pure states are characterized by the matrix equation
ρ2 = ρ.
Although we have introduced it in the context of a two state system, this description in terms
of matrices is completely general, for any finite number of elementary pure states. N states
corresponds to a theory of N × N matrices. In classical logic, all matrices representing data
about the system are simultaneously diagonal, as are all probability distributions.

2.4 OPERATIONS AS MATRICES

Things become a little more interesting when we recognize that the classical operation of
reflecting the nitrogen through the plane of the hydrogens is also represented by a matrix.
The matrix is !
0 1
σ1 = , (2.14)
1 0
and Exercise 2.3 asks you to verify that it indeed does what it was claimed to do.

Exercise 2.3 Use the rules for acting with matrices on vectors to show that

σ1 |±i = |∓i.
From this equation, it follows that σ12 = 1. Verify that this is true using matrix multiplication.
This algebraic equation for σ1 expresses the fact that if we do two successive reflections, we
return the system to its original state.
The operation of evaluating the dipole moment does not commute with this reflection
σ1 σ3 = −σ3 σ1 ≡ −iσ2 ,
from which it also follows that σ22 = 1. Calculation (Exercise 2.4) shows that
!
0 −i
σ2 = . (2.15)
i 0

We can also write the general product relation

σa σb = δab + iabc σc ,
 Two State Systems: The Ammonia Molecule  35

where the sum on c from 1 to 3 is left implicit (this is called the Einstein summation conven-
tion). The symbol abc is totally antisymmetric in its three indices and has the value 123 = 1.
It is the object we use to define the cross product of two vectors,

(A × B)a = abc Ab Bc .

Again we are summing over repeated indices, and from now on we will use the Einstein
summation convention without comment. Do Exercise 2.6 if you want some practice with the
epsilon symbol.
The fact that these matrices do not commute is obvious, from the point of view of
classical logic, if we think of them both as operations on the system. One operation evaluates
the dipole moment in a particular state, while the other changes which state the molecule is
in. Of course you get a different result if you evaluate the dipole moment before or after you
change the state!
It is because we can think of both the diagonal matrices, which act on a state by mul-
tiplying it by a number, and matrices like σ1 , as operations on the system, that we use the
words matrix and operator almost interchangeably. A matrix is just a way of characterizing
what a linear operator does to a particular orthonormal basis.2
Now let us consider something peculiar. Evaluate

Tr ρσ1 = 0. (2.16)

This is true no matter what the probabilities are to be in |±i3 , and in particular it is true
when we know that we are certainly in one of those states. On the other hand, since σ12 = 1
it is natural to think that it might “take on the values ±1,” just like σ3 . Indeed, we can find
vectors of length 1
1
|±i1 = √ (|+i3 ± |−i3 ), (2.17)
2
such that
σ1 |±i1 = ±|±i1 . (2.18)
For diagonal matrices, the trace with the density matrix gave the expectation value for the
operator σ3 , in a probability distribution for it to take on one of its possible values. We can
interpret the formula in the same way for σ1 , and come to the conclusion that σ1 is uncertain
even when we know the precise state we are in. That is, we could accept the proposition that
the normalized vectors |±i1 defined possible states that the system could be in, in which
2
One can also associate matrices with the action of an operator on a basis which is not orthonormal. We
will have little or no use for these in the finite dimensional case, although we will discuss matrix elements
of operators in continuous, overcomplete bases in Chapters 5 and 19.
36  Quantum Mechanics

σ1 had definite values. The density matrix saying we are definitely in the state |+i1 would
be the projection operator on this vector.

(1 + σ1 )
P1+ = . (2.19)
2
It is easy to see that
σ3 |±i1 = |∓i1 , (2.20)

so that
Tr (P1+ σ3 ) = 0. (2.21)

That is, in the state |+i1 , σ1 has a definite value, while σ3 is maximally uncertain.
This observation and conjecture lead one to ask which normalized vectors can be thought
of as states of the system, and which operators correspond to quantities that can take on
definite values. A quick review of the theory of diagonalization of matrices, will lead us to the
conclusion that any normalized vector is acceptable, and any operator whose eigenvectors
form a complete orthonormal basis is a potentially observable quantity.

2.5 THE EIGENVALUES OF A MATRIX

The fact that matrices satisfy algebraic equations like σ12 = 1, is not a surprise. In fact, there
is a general theorem, for any square matrix in any number of dimensions. Recall that the
determinant of a square n × n matrix det(M ) is the sum of all possible products of n distinct
elements, with a sign that you probably learned in terms of an algorithm where you go down
the left-hand column of the matrix, changing sign at every step, and evaluate the determinant
in terms of products of the matrix element in the k-th row of the left-hand column times
determinants of cofactor matrices obtained by omitting the k-th row and first column. For
a diagonal matrix, this algorithm just gives the product of diagonal matrix elements. Given
any matrix M , its characteristic polynomial is

PM (x) = det(xI − M ),

where I is the n × n unit matrix. If the matrix is diagonal, with diagonal matrix elements
mi , this just gives
Y
PM (x) = (x − mi ).
i

We can easily see from this that if we make the same polynomial of the matrix, then

PM (M ) = 0.
 Two State Systems: The Ammonia Molecule  37

This is (the simplest case of) the Cayley-Hamilton theorem. Now recall a simpler theorem,
namely that for any two square matrices

det(AB) = det(BA),

so that
det(SBS −1 ) = det(B),
for any invertible matrix S. This generalizes Cayley–Hamilton to a large class of nondiagonal
matrices, namely all those that are diagonalizable by a similarity transformation S. If M =
SDS −1 , where D is diagonal, then

det (xI − M ) = det (S[xI − D]S −1 ) = det (xI − D), (2.22)

so M and D have the same characteristic polynomial. It is also true for any polynomial q(x)
that q(M ) = Sq(D)S −1 , so that PM (M ) = SPD (D)S −1 = 0. In fact, the Cayley–Hamilton
theorem is true (but we will not use it) for any matrix.
At this point let us recall that linear algebra is a prerequisite for this course, stop treating
you like a babe in the matrix woods, and freely use concepts from linear algebra. If you
are uncomfortable with your understanding of linear algebra, you should skip to Chapter 6
at this point. This will have the added advantage that you will review linear algebra using
the Dirac notation that we will employ throughout this book. A shorter summary of Dirac
notation and linear algebra can be found in Appendix F.
The matrices σa are all examples of matrices that can be diagonalized by conjugation by
a unitary transformation. That is, there is a unitary transformation, Ui (i.e., Ui† Ui = 1) such
that Ui† σi Ui is diagonal. Equivalently there is an orthonormal basis of eigenvectors such that

σi |±ii = ±|±ii .

For example,
1
|±i1 = √ (|+i3 ± |−i3 ).
2
In the basis |±ii , σi is diagonal, while the other two have off diagonal matrices. In fact, any of
the matrices σ(n) ≡ na σa where na is a real three-dimensional vector, can be diagonalized.
P

In this new notation, the states we called |±i are now |±i3 . You can easily verify that for
the vectors |±i1 ,
σ1 |±i1 = ±|±i1 .
The factor of √12 is inserted so that these vectors are orthonormal. These words refer to the
scalar product on our vector space. We will discuss this important concept in more detail
below, but you should start trying to remember things about scalar products from your linear
algebra class.
38  Quantum Mechanics

The reason that it is interesting to look at orthonormal bases is that they are the quantum
version of a complete set of independent alternatives in classical probability theory. The
projection operators Pn ≡ |en ihen | have the property of density matrices. Their eigenvalues
are positive and sum to 1. Furthermore,

Tr (Pm Pn ) = δmn . (2.23)

This equation is interpreted as meaning that if one is in the state specified by Pm , then the
probability of being in the state Pn is zero unless m = n. The normalization condition on
states,
1 = hs|si = Tr(|sihs|), (2.24)
is the statement that probabilities sum to one in any state. What distinguishes QM from
classical mechanics is that we allow ourselves to contemplate the quantities

Tr (|en ihen ||fm ihfm |) (2.25)

for two different orthonormal bases and interpret this number as the probability that the
system is in one of the f states, given that we know that it is in one of the e states (or vice
versa: the formula is symmetric under interchange of the two bases). If the time evolution
law for states does not preserve a particular orthonormal basis, we are forced to work with
these more general probabilities.
To say this all in another way, in addition to the original matrices, which we identified
with quantities detectable or measurable for the up/down bit of the ammonia molecule,
there are other matrices, either operations on the up/down states, or algebraic combinations
of operations and measurable quantities, which satisfy algebraic equations in terms of matrix
multiplication and addition. For some of them we can, using the rules of linear algebra, find
linear combinations of the vectors we originally identified with states of the system, in which
the different solutions of those algebraic equations are “realized.” Linear algebra treats all of
these realizable operators (which are called normal operators) in a democratic fashion. QM
results from exploring the consequences of this democracy for physics, that is, of viewing
every normal operator as something that can take on a definite value in some state of the
world.

2.6 THE BORN RULE AND UNCERTAINTY

To reiterate: the essence of QM lies, in the fact that we can generalize the trace formula for
expectation values, to a large class of nondiagonal matrices A. The formula

hAi = Tr (Aρ)
 Two State Systems: The Ammonia Molecule  39

may be evaluated as
X
hAi = ak pr |hak |ri|2 ,
r,k

for any matrices A and ρ, which are diagonalizable by a unitary transformation (Exercise 2.7).
The vectors labeled ak and r are the orthonormal eigenvectors of A and ρ, respectively. The
ak are the two eigenvalues of A. If the eigenvalues pr satisfies the probability (in)equalities
above, then so does the probability distribution
X
P (ak ) ≡ pr |hk|ri|2 ,
r

as a consequence of completeness and orthonormality of the two bases (Exercise 2.8). Note in
particular that even if pr = 1 for some r, which would, according to classical logic, imply that
all quantities were known with absolute certainty, P (ak ) < 1 unless the two bases k and r are
identical (up to permutation). Note also that, while we have introduced all of this notation
for a simple bit, with only two states and a single nontrivial question, our remarks about
matrices are completely general, and apply to a system with any number of states. Finally,
note that if pr = 1 (and thus all other probabilities zero) then the probability of finding A
with eigenvalue ak is
|hak |ri|2 , (2.26)

the square of a complex number called the probability amplitude that A takes on the value ak .
This equation is known as Born’s Rule for QM probabilities. It is usually introduced
as an ad hoc postulate, or following a discussion like Feynman’s lecture on the double slit
experiment. Here, we have derived it by generalizing classical probability formulae, written
in matrix form, to matrices that are diagonalizable, but not diagonal in the same basis as
the (probability) density matrix ρ. Born’s formula for probability is one of the two most
important equations in QM, the other one being
X
1= |en ihen |, (2.27)
n

which is called the resolution of the identity, and is true for any choice of the orthonormal
basis |en i. The individual terms, |en ihen |, in the sum are projection operators onto the one-
dimensional spaces spanned by a single member of the orthonormal basis (we have said this
before, but prove this to yourself, if it is not obvious to you). What this equation says, in
words, is that the unit operator is the sum of the projection operators onto the elements of
any orthonormal basis.
40  Quantum Mechanics

2.6.1 Born’s Rule in Pictures

Let us consider Born’s Rule when the state of the system is pure, and the operator A is the
projection operator onto some pure state |f i. Then

A = |f ihf |, (2.28)
ρ = |eihe|, (2.29)
Aρ = (hf |ei)|f ihe|, (2.30)
ρA = (he|f i)|eihf |. (2.31)

In the example where !

1
|ei = and
0
!
1 eiα
|f i = √ ,
2 1
you should verify that these formulae give
−iα
!
e√
eiα 0
ρA = √ 2 , (2.32)
2 1 0

e√iα
!
e−iα 1
Aρ = √ 2 . (2.33)
2 0 0
Note that the trace of these matrices is the same, a consequence of the general theorem that
Tr AB = Tr BA, and is equal to |hf |ei|2 , which is equal to 1/2 for our example.
This leads to a geometrical interpretation of Born’s rule, illustrated in Figure 2.4.
If |ei and |f i are two unit vectors, then the quantity he|f i is the projection of the |ei
vector on the |f i vector, which is the name for the component of the |ei vector along |f i, in
a basis, one of whose elements is |f i. If the vectors were real, this would be cos(θ), the cosine
of the angle between them. Since they are complex, the projection is a complex number. Its
complex conjugate hf |ei is the projection of |f i along |ei.
The projections fi of any unit vector along the elements of any orthonormal basis |ei i
satisfy Pythagoras’ theorem X
|fi |2 = 1.
Born’s rule tells us that we should interpret this geometrical formula as defining a probability
distribution. Each unit vector is a possible physical state of the quantum mechanical system.
|fi |2 is interpreted as the probability that, if the system is in the physical state represented by
 Two State Systems: The Ammonia Molecule  41

a2 + b2 = 1

Figure 2.4 Geometric interpretation of Born’s rule.

|f i, that a measurement which attempts to find out whether it is in one of the physical states
|ej i, will find that j = i. In order to truly understand the implications of this sentence, we
will have to make a quantum model of what we mean by the word measurement. We will
postpone this difficult question until Chapter 10.
The picture also helps us to understand the meaning of the name projection operator.
Given the unit vector |f i, the projection operator |f ihf | operates on any vector and gives us
the projection of that vector along |f i. This is illustrated in Figure 2.5.
To summarize: one can formulate all of the classical information about the two apparently
degenerate states of the ammonia molecule, and operations which change from one state to
the other, in terms of the algebra of two by two matrices, acting in a vector space whose basis
vectors represent the two states. The formulae for expectation values of classical measurable
quantities generalize to arbitrary matrices diagonalizable by a unitary transformation and
give us a probability theory in which, in any state of the system, not every quantity can be
known definitely.
Another way to think about quantum probability theory is that it uses the mathematical
similarity between probability formulae and Pythagoras’ theorem for the projections of one
unit vector on an orthonormal basis of others, to define a notion of probability in which there
is uncertainty for something, even in states of the system where a complete set of things is
completely determined.

|vi

P |vi

Figure 2.5 A projection operator.

42  Quantum Mechanics

Probability theory was invented to deal with situations in which things were definite,
and evolved in a definite manner, but where we lacked knowledge, whether through sloth
or inability to measure, about all the relevant initial conditions. We think of it as a way
of guessing intelligently, about the things we have not measured. The initial probability
distributions in a classical probabilistic theory are guessed, and then revised on the basis of
experiment, to fit the data. In contrast, the intrinsic quantum probability theory has nothing
to do with guessing. Once we have specified the state completely, by measuring the maximal
set of compatible quantities, the probability distributions for all other quantities are exactly
computable. One might say that QM takes the guesswork out of probability. At this point,
the reader should do Exercise 2.9, and find the probabilities, according to Born’s rule, for
the operator na σa , where na is a real three vector, to take on its two possible values ±|n|,
assuming that the system is in one of the states |±i3 .
The essence of QM is that we accept Born’s rule, which defines, for every normalized
vector in our space (our two-dimensional space in the present example) and every matrix
A which is diagonalizable by a unitary transformation, a mathematical function, Pψ (ak ) ≡
|hak |ψi|2 , which has the properties of a probability distribution for the quantity A to take
on one of its eigenvalues, ak when the system is in the state ψ. QM is the statement that
this mathematical function should be considered a real probability distribution for physical
quantities.

2.7 WHICH OPERATORS ARE DIAGONALIZABLE?

Now let us take up the question of which linear operators A, are candidates to be measurable
properties of our system. Our fundamental probability assumption is that the quantity
apk Pψ (ak ),
X
hh(Ap )ii = Tr (Ap |ψihψ|) = hψ|Ap |ψi = (2.34)
k
is the expectation value of the value of Ap when we make repeated measurements of it, after
having prepared the system in the state represented by the unit norm vector |ψi. We will have
to explain what we mean by measurement and state preparation, in order to make contact
between this mathematical probability theory, and things that happen in the real world, but
we will put off that discussion until Chapter 10, when we have a lot more experience with
the mathematics of QM.
In order for this formula to have the properties of a probability distribution, there must
be an orthonormal basis |ak i such that
A|ak i = ak |ak i. (2.35)
Now consider the matrix of A in any other orthonormal basis
X
Amn ≡ hem |A|en i = hem |ak ihak |A|al ihal |en i, (2.36)
kl
 Two State Systems: The Ammonia Molecule  43

where we have used the resolution of the identity for the basis |ak i twice.
We now use the eigenvalue equation and the orthonormality relation hak |al i = δkl , to
write this as
∗
Amn = Umk ak Unk , (2.37)
where Umk = hem |ak i. As a matrix equation, this is A = U DU † , where the Hermitian
conjugate U † is the matrix obtained by both complex conjugating and transposing (flipping
row and column indices) the matrix U . D is the diagonal matrix of A in the |ak i basis. The
resolutions of the identity and orthonormality relations for the two bases |ak i and |em i are
(prove it!) equivalent to the matrix equation
U U † = U † U = 1. (2.38)
As we have said, such matrices are called unitary matrices. Their rows and columns are the
two sets of orthonormal basis vectors. They are the analog of ordinary rotations in the vector
space representing three-dimensional Euclidean space.
Now note that
AA† = U |ak |2 U † = A† A. (2.39)
Any operator that commutes with its Hermitian conjugate is called normal. The fact
that [A, A† ] = 0 is obvious in the original basis where the matrix of A is diagonal. On
the other hand, if we have a polynomial involving some collection of matrices Mi , then
P (U † Mi U ) = U † P (Mi )U, for any unitary matrix. One can prove this identity by noting
that all the “intermediate” factors of U † U can be replaced by 1 due to unitarity. (Informal
Exercise: Prove this statement.)
The converse statement, that any normal operator can be written as A = U DU † , is called
the spectral theorem, and the set of ak is called the spectrum of the operator. To prove it,
note that det(A − a) is an N -th order polynomial in the complex parameter a and so has N
roots ak . This means that A − ak is not an invertible operator, and so must satisfy
(Bk )p |ak i = 0, (2.40)
where Bk = A − ak , for some value of p and some state |ak i. If A is normal, this means that
hak |[Bk† Bk ]p |ak i = 0. (2.41)

If p is even, this is the norm of the state [Bk† Bk ]p/2 |ak i, while if p is odd, it is the norm of
Bk [Bk† Bk ](p−1)/2 |ak i. The only state with vanishing norm is the zero state, so we find that
Bkq |ak i = 0, where q is p/2 or (p − 1)/2. We can continue reducing the value of p in this way,
until we conclude that Bk |ak i = 0, which is the statement that |ak i is an eigenvector of A
with eigenvalue ak . The eigenvectors with different eigenvalues are orthogonal because
hak |A|am i = hak |am iam = hak |am iak , (2.42)
44  Quantum Mechanics

if we act with either A to the right or A† to the left. Note that hak |A is the complex conjugate
row vector corresponding to A† |ak i = a∗k |ak i. For a degenerate subspace with the same
eigenvalue, we can always choose an orthonormal basis. So, a general normal operator is
diagonalizable by a unitary matrix.
In most books on QM, one is told that operators corresponding to measureable quantities
must be Hermitian: H = H † . Hermitian operators are obviously normal. Furthermore, for
any operator
1
A = (A + A† )/2 + i (A − A† )/2 ≡ H1 + iH2 , (2.43)
i
where Hi = Hi† . This is analogous to breaking a complex number into its real and imaginary
parts. If A is normal, then [H1 , H2 ] = 0. So, a normal operator is equivalent to a pair of
commuting Hermitian operators. There is no reason not to allow any normal operator as a
“physically measurable” quantity in QM, and it is convenient because it allows us to solve
any algebraic equation involving physical quantities, and get another physical quantity.

2.7.1 Vectors, Functions, and Diagonal Matrices

Later on in this book, we are going to encounter vector spaces, in which the vectors are
functions of continuous variables. A lot of students get confused transferring what they know
about linear algebra into the context of function spaces, so it is worth explaining that we can
always think of linear algebra in terms of functions. If vn are the components of a vector |vi
in some basis, then we can associate with them a complex valued function, defined on the
finite set of integers from 1 to N , by the rule

fv (n) = vn . (2.44)

The space of all such functions is a vector space, since we can add them and multiply them
by complex numbers, and the scalar product takes the form
N
fw∗ (n)fv (n).
X
hw|vi = (2.45)
n=1

The basis vectors are functions which vanish on all integers between 1 and N except for one.
Linear operations become operations on functions. A particularly interesting one is the finite
difference operator
[∆f ](n) = f (n) − f (n − 1). (2.46)
If we take a limit where we think of N going to infinity and define x = Nn , then x becomes
a continuous variable in the limit and N ∆ approaches the derivative operator, when it acts
on differentiable functions of x. You should write out the matrix corresponding to the differ-
ence operator ∆ to make sure you understand this. Note that this correspondence between
 Two State Systems: The Ammonia Molecule  45

vectors, and functions defined on the discrete set 1 . . . N , depended on a choice of basis.
Different choices of basis will give different functions for the same vector |vi. When we get to
continuous variables, we will see that the famous Fourier Transform is just a relation between
the functions that represent a vector in two different bases.
Another confusion that arises when thinking about functions as elements of the space
is that functions also act as multiplication operators on the space of functions. This also
has a finite dimensional analog. Given a vector vn , or the associated function fv (n) one can
construct a diagonal matrix, whose n-th diagonal element is vn . When one acts with that
matrix on a vector whose components are wn , then one gets the vector whose components are
vn wn = fv (n)wn . The function corresponding to this new vector is the product fv (n)fw (n).
These remarks may seem sort of silly in the finite dimensional context, but they are the
bridge that allows you to make the transition to thinking about function spaces.

2.8 QUANTUM DYNAMICS OF THE AMMONIA MOLECULE

The reader is entitled to be confused at this point, since we started out with a seemingly
classical description of the two low-energy states of the ammonia molecule and have already
introduced quantum mechanical variables, which cannot take all definite values at the same
time. How is it possible then, to have a classical description of ammonia, even in principle?
Physics is all about dynamics. It is a set of equations of motion, which tell us how to
predict the state of a system at a later time, given its state now. In Newtonian mechanics,
this prescription consists of a set of differential equations3

ż i = f i (z).

Time flows continuously. The variables z i are differentiable functions of t. The values of z(t0 )
enable us to predict/retrodict all past or future values of z(t). The key point in these equations
is that, we assume that all of the different z i can take on definite values simultaneously. QM
emerges upon realizing that this assumption is not necessary, if we are willing to live with an
intrinsically probabilistic description of the world, as opposed to one in which probability is
introduced merely as a device to compensate for our lack of ability or desire to measure all of
the z i (t0 ) precisely. The only way to see whether the special case of C(lassical) or C(ertainty)
mechanics describes the real world, is to do experiments.
For our simple low-energy ammonia molecule, a classical theory must describe the evolu-
tion of the system in discrete jumps. We started out by assuming that there is no continuous
way to go between two states that differ by a reflection of the nitrogen atom through the plane
3
We can always reduce ordinary differential equations with higher derivatives to first order equations by
simply calling the derivatives names and adding them to the list of variables. An N -th order equation for
a single function is equivalent to N first-order equations for that function and its derivatives up to the
N − 1 order.
46  Quantum Mechanics

of the hydrogens, without having much more energy than we have available. For simplicity,
let us imagine that the jumps happen every fixed time interval τ . Introduce the frequency
f = 1/τ . Then the state of the system at time t is

|ψ(t)i = (σ1 )[f t] |ψ(0)i.

The square bracket notation in the exponent stands for “greatest integer in.” In this classical
version of the theory, the energy of both states is the same. It is definite and conserved during
the motion. σ3 , which determines the value of the dipole moment, switches sign every time
an interval τ passes. A classical physicist would not want us to talk about measuring the
variables σ1,2 (though the classicist would let us talk about σ1 as an operation on the system),
but we can use the trace formula to evaluate their expectation values. The initial density
matrix is the projection matrix onto whichever of the two states |±i3 is the initial state, and
it switches to the projector on the other state every time interval τ . The expectation values
of σ1,2 are zero in either of these states, so we conclude that these two variables are always
maximally uncertain. The most general diagonalizable 2 × 2 matrix has the form

A + Bna σa ,

where na is a real unit three vector (prove this in Exercise 2.10), and A, B are complex
numbers. With the classical evolution law we have given, the components of this matrix
proportional to 1 and σ3 remain definite for all time, while σ1,2 are maximally uncertain.
That is, if we interpret the formula

TrρA

in the probabilistic manner suggested above, and choose ρ to be the projection operator on
either of the vectors |±i3 , then we find the expectation values of σ1,2 are both zero. Given
the possible values, ±1, of those two operators, and the fact that the system only has two
states, we learn that the probability to be in each of the eigenstates of ta σa is 12 , if ta is
perpendicular to the 3 axis.
One thing that is a little annoying about this classical system is that the time evolution
has jumps, whenever t passes through an integer times τ . There is nothing wrong with this.
In particular, if our experimental time resolution is much larger than τ , we would never see
the discrete jumps. However, it is amusing to note the purely mathematical result that if
ω ≡ π2 f , then
σ1N = e−iωt(σ1 −1) ,
when t = N τ . So, we can view the discrete classical evolution as a sequence of snapshots of
a continuous evolution of the state

|ψ(t)i = e−iωt(σ1 −1) |ψ(0)i ≡ U (t)|ψ(0)i,

 Two State Systems: The Ammonia Molecule  47

if we contemplate expanding our notion of what states the system can be in, to include all
the two-dimensional complex vectors that are swept out by this evolution. Mathematically
this extension of the concept of state is simple, and in a certain sense obvious. Conceptually
it is a profound deviation from the logic of classical physics. The idea of allowing the system
to be in any normalized state is responsible for all of the uncertainty principles of QM.
The evolution operator is unitary (Exercise 2.11), and so the two vectors |±i3 are trans-
formed into two other orthonormal vectors at any given time. That means that, at each time,
there is some matrix, which is diagonal in that instantaneous orthonormal basis, but it is not
the original matrix σ3 , which we identified with the detectable electric dipole moment of the
molecule.
Let us step back and review the formula e−iωt(σ1 ) that we just used. What does it mean?
Matrix multiplication allows us to define polynomials of matrices, so in principle, up to
questions of convergence, it also allows us to define any function of a matrix that is given by
a power series. So,
∞
(−iωtσ1 )n
e−iωt(σ1 ) =
X
.
k=0
n!
Using σ12 = 1, the matrix appearing in the even terms in this sum is always the unit matrix,
while the odd terms are all proportional to the matrix σ1 . The even and odd terms in an
exponential of a complex number just sum up to the cosine and i times the sine, so
e−iωt(σ1 ) = cos(iωt) − i sin(iωt)σ1 .
Another way to think about functions of operators, if they are diagonalizable, is to think
in the basis where the operator A is diagonal, and just define a function f (A) to be the
diagonal matrix whose matrix elements are f (ai ). We can then express this function in any
other orthonormal basis by using the unitary operator UA , which rotates the basis where A
is diagonal, into that other basis. In other words, we define functions of operators so that
f (U † AU ) = U † f (A)U . This equation is obviously correct for any function defined as a power
series, and we simply take it to be the definition of f (A) for any diagonalizable A.
The density matrix of the system evolves according to
ρ(t) = U (t)ρ(0)U † (t). (2.47)
This is “obvious” for pure states and follows for general mixed states by linearity of the time
development operation:
bi Ai U † (t) = bi U (t)Ai U † (t).
X X
U (t)

Note that the overall phase eiωt disappears from the formula for ρ(t). We can study the time
development of expectation values in the system by noting that (see Exercise 2.7), by the
cyclicity of the trace (Tr AB = Tr BA)
48  Quantum Mechanics

Tr A(t)ρ = Tr Aρ(t), (2.48)

where the time evolution of the Heisenberg picture operator A(t) is defined by

A(t) ≡ U † (t)A(0)U (t). (2.49)

That is, we can either evaluate expectation values of fixed operators in a time-dependent
state (the Schrödinger picture) or time-dependent operators in a fixed state (the Heisenberg
picture). Whichever way we do it, we find that, even if we start in a state where σ3 has a
definite value, it will not have a definite value at a generic time. Indeed

σ3 (t) = (cos(ωt) + i sin(ωt)σ1 )σ3 (cos(ωt) − i sin(ωt)σ1 )

= cos(2ωt)σ3 − sin(2ωt)σ2 .
Note, for future (Chapter 7) reference, that this looks like a rotation by an angle −2ωt in
the 2 − 3 plane, if we think of the three Pauli matrices as a three-dimensional vector. Since
the expectation value of σ2 in the initial state vanishes, we have

hσ3 (t)i = cos(2ωt) = cos2 (ωt) − sin2 (ωt).

That is, it appears that the system has a time-dependent probability distribution, with prob-
ability p(t) = cos2 (ωt) to have a positive value of the dipole moment, and 1 − p(t) to have a
negative value.
We can now try to compare the predictions of the classical and quantum models for the
evolution of the system with data. In principle, in order to do this, we have to understand
what it means to measure a microscopic property of a molecule. In classical physics, this is a
conceptually straightforward, if practically difficult operation. In QM, we will have to discuss
the concept of measurement more carefully, but we will wait till we understand the subject a
bit better before doing that (Chapter 10). For the moment, let us just imagine that we know
how to do the measurement.
The classical prediction is that the molecule jumps from positive to negative electric
dipole moment (EDM), or vice versa, every ωπ seconds. It will always be found to have the
same magnitude and only its sign will flip. At intermediate times, it will remain in whatever
state it got to after the last flip. The QM prediction is the same4 at precisely these discrete
times, but the prediction for intermediate times is different. The quantum prediction is that
at an intermediate time, the EDM will be either ±, the same classical magnitude, but that
the choice of sign is random. There is no way to predict precisely what will happen at
intermediate times. However, if we repeat the experiment over and over again, with exactly
4
Actually there is a subtle difference. See below.
 Two State Systems: The Ammonia Molecule  49

the same conditions, we will find that the frequency of plus versus minus signs follows the
cos2 (ωt) law. Needless to say, we would not be talking about this if it were not for the fact
that experiments show the QM prediction to be the correct one for ammonia molecules.
An important subtlety appears when we think carefully about the quantum versus clas-
sical predictions for what will happen at the discrete times. Since the quantum prediction
of anything is always probabilistic, we really have to do the experiment many times if we
want to test the quantum prediction, even when QM predicts certainty. After all, a single
measurement at an intermediate time does not prove the quantum law is correct, since it is
only a probabilistic law. The prediction of probability one for the discrete times can only be
checked by showing that the dipole never has the wrong orientation at those times. In the
classical theory, by contrast, we take the stance that, since we have completely measured
the initial conditions, the theory makes a definite prediction, whose truth can be determined
by a single experiment. It is in this subtlety that we see the real distinction between a theory
which considers probability to be a mere consequence of failure to measure things precisely,
and one in which probability is intrinsic.
QM can explain an even more striking experimental fact if we make a hypothesis about
how to identify energy in the QM formalism. In order to do this, we have to remember the
connection between energy conservation and time translation invariance. There is a very
general theorem about this connection, called Noether’s theorem, which is proved in the
Lagrangian formulation of classical mechanics (see Appendix C). For simplicity, we can just
think about Newton’s law for a classical particle moving in one dimension, under the influence
of a potential:
d2 x dV
m 2 =− .
dt dx
If either the mass or the potential depend explicitly on time, then energy is not conserved.
Time translation invariance is necessary for conservation of energy and Noether showed that
it is also sufficient.
Noether’s theorem, the connection between symmetries and conservation laws, is one of
the most profound results in all of theoretical physics. One of the attractive features of QM is
that, as a consequence of its unification of detectable quantities for a system and operations
on that system, Noether’s theorem is almost a definition of a symmetry. With the notable
exception of time reflection symmetry, which we will discuss in Chapter 6, both detectable
quantities and symmetry transformations are normal operators. Symmetries, however, obey
two restrictions, which are not required of general detectable operators. Given any pair of
states, the quantity |hs1 |s2 i|2 is the probability, assuming the system is in one of those states,
that one finds the projection operator on the second, equal to 1. This should still be true
after subjecting each state to a symmetry operation |si i → (|W (si )i). That is,

|hs1 |s2 i|2 = |hW (s1 )|W (s2 )i|2 . (2.50)

50  Quantum Mechanics

One general way to satisfy this is for W to be implemented by a linear unitary transformation

|W (s)i = W |si; W W † = 1. (2.51)

The other possibility exploits the fact that the probability formula is invariant under complex
conjugation of hs1 |s2 i. We will explore it in Chapter 6.
The other requirement for the quantum mechanical version of a symmetry transformation
is analogous to the requirement in classical mechanics. If we transform the initial state of
some quantum system, it should lead to an equivalent time evolution. This means that the
symmetry operator U has to commute with the time translation operator. In equations, the
Heisenberg operator
W (t) ≡ U † (t, t0 )W (t0 )U (t, t0 ) = W (t0 ). (2.52)
is a conserved quantity! This is the quantum mechanical version of Noether’s theorem. If you
know the classical derivation of the theorem, or read the appendix, you will conclude that
the quantum version is simpler.
Our classical and quantum evolution laws for the ammonia molecule are both time trans-
lation invariant (in the classical case, by discrete multiples of τ ). A more general set of laws
would allow the variation of the state at each time to be arbitrary. In the classical case, this
would just mean that at each time step we could choose to flip the EDM, or not. Quantum
mechanically, we could have
|ψ(t)i = U (t, t0 )|ψ(t0 )i,
with an arbitrary time-dependent unitary matrix. At any time, we could diagonalize

U (t, t0 ) = diag(e−iω1 (t)t , e−iω2 (t) ).

The two frequencies ωi (t), would be time dependent. So, it is tempting to identify the time-
independent parameter ω of the time translation invariant system with the conserved energy.
Frequency and energy have different units, so we must introduce a constant ~, with dimen-
sions of energy times time, or action and define

E = ~ω.

It turns out that this conversion factor is universal for all systems.5 The constant ~ is called
Planck’s constant (it is actually the constant Planck originally defined, divided by 2π because
of the conversion between cycles per second and radians per second). In centimeter-gram-
second units, it has the value 1.05457266(63) × 10−27 erg-s. Theoretical physicists often use
natural units in which ~ = 1 and energy and frequency are simply identified.
5
This is a consequence of energy conservation. It turns out that the laws of QM say that frequency is always
conserved, so if different systems had different conversion factors, energy would not be conserved.
 Two State Systems: The Ammonia Molecule  51

Energy conservation is more than just the fact that ω is constant in time. In the time-
independent case, U (t) is diagonal in the basis where σ1 is diagonal, for all time. Thus, the
Hamiltonian operator
H ≡ ~ω(σ1 − 1),
is time independent, H(t) = H(0). This is the full statement of energy conservation. The
Hamiltonian, H, as it is called, is the quantum mechanical version of energy, and like every-
thing else in QM, it is a matrix/operator. What is more interesting is that this matrix has
two different eigenvalues 0 and −2~ω. The corresponding eigenstates are
1
|±i1 = √ (|+i3 ± |−i3 ).
2
Notice that under the space reflection operation, which changes the sign of the dipole moment,
both these states go into themselves, up to a phase ±1. Note also that we are perfectly free
to define the reflection operation on the states by

|±i3 → −|∓i3 ,

since the extra minus sign does not affect the transformation property of the expectation
values of any operator.
In the classical theory, where the only allowed states were |±i3 , we were forced to conclude
that reflection invariance forced the energies of the two states to be exactly equal to each
other. Here, we see that these two eigenstates of the Hamiltonian can have different energies,
in a way that does not violate reflection symmetry. So, we have a new prediction: there should
be a small splitting, 2~ω between the classically degenerate energy levels of the ammonia
molecule, and this splitting should be related to the frequency of oscillation of the probability
distribution for the electric dipole moment.
We can test this prediction by exposing the molecule to light. One of the most charac-
teristic and puzzling features of atomic systems from a classical point of view was the fact
that they absorb and emit light at discrete characteristic frequencies. The historical origin of
quantum theory was Planck’s work on the blackbody spectrum. His resolution of the classical
paradox was that light of a given frequency carries discrete amounts of energy, given by the
formula E = ~ω. This relation was used by Einstein to understand the photoelectric effect,
and by Compton to understand the scattering of light by electrons. And we indeed find that
ammonia molecules absorb and emit light of a fixed (microwave) frequency, much lower than
most other characteristic frequencies of atoms and molecules. Later, we will understand the
small size of this energy difference in terms of the concept of quantum mechanical tunneling.
We said earlier that it was tempting to identify (~×) the frequencies with which the
eigenstates of the time-independent evolution operator oscillate, as energies. In fact, even if
we would never heard of the concept of energy, the general discussion of symmetries in QM
52  Quantum Mechanics

tell us that these frequencies are indeed constants of the motion and we would have been
led to define a conserved quantity, the frequency operator H/~ for any time-independent
quantum system.

2.9 SUMMARY
Let us summarize all this by saying it in a different way. The quantifiable data about any
system that we think of classically, can be organized by making a list of all possible states of
n of the system and then listing all functions f (n) on that space of states, which could con-
ceivably correspond to measurable properties of the system. This is mathematically identical
to inventing a vector space, whose basis is labeled by the different states, with the functions
realized as diagonal matrices. A probability distribution is also a diagonal matrix ρ, with
nonnegative matrix elements summing up to one. Operations that change the state of the
system are nondiagonal matrices. One can then make the mathematical observation that the
formula
Tr f ρ,

for computing the expectation value of a function in a given probability distribution, can
be generalized by replacing f by any diagonalizable matrix, even if it is not diagonal in the
original basis. If the matrix is diagonalizable by a unitary transformation (which is equivalent
to the requirement that [A, A† ] = 0), then the above formula defines a probability distribution
p(a) for A to take on one of its eigenvalues. Even if the original probability density ρ specified
a definite value for each of the functions f , the probabilities for diagonalizable matrices which
do not commute with the original functions do not give definite values for these variables.
The question of whether this intrinsically uncertain probability theory is relevant to a
particular physical system depends on its dynamics. If time development always takes ele-
ments of the original basis into other elements of the basis (i.e., if it just permutes the
states), then we can just declare that the only measurable operators are those diagonal in
that basis and ignore the fact that many other quantities are uncertain. This is the analog
of classical mechanics for a system with a finite number of independent projection oper-
ators. However, the most general probability conserving operation on states is given by a
unitary operator U (t, t0 ), which can continually change the basis. Future predictions of the
history of a general quantity will be probabilistic, because the time derivative of some variable
which is definitely known cannot be known with certainty. This is QM. Experiment shows
that it is the relevant form of dynamics for systems of atomic scale and smaller. We will
learn later that our illusion that macroscopic objects do not suffer from such uncertainties
is a consequence of the fact that they are made of a large number > 1020 of atomic con-
stituents, and that we only observe coarse grained average properties of the states of these
atoms.
 Two State Systems: The Ammonia Molecule  53

2.10 SKIP THIS UNLESS YOU ARE INTERESTED IN PHILOSOPHY

The probability theory defined by QM differs in two important ways from the probability
familiar to classical physicists and gamblers. In the latter, we are forced to use probability
because we cannot (or at least have not) measured all of the variables needed to make
predictions about the future of our system. As a consequence, the probability distributions
are not known a priori, but must be established by experiments, or by making assumptions
like “the coin is not biased, the dice are not loaded,” or that the distributions follow certain
laws that have been established by observation of large classes of systems. In contrast, once
one has established a definite initial state in QM, the probability distributions for all quantities
are calculable with no recourse to observation. So, in a certain sense, QM takes the uncertainty
out of probability theory, but introduces intrinsic uncertainty into our predictions about the
future, even for initial conditions measured with infinite precision.
The second difference is the violation of Bayes’ rule of conditional probability. Consider
two mutually exclusive alternative states A and B of the system at some time, t0 . This means
that if A is true, then B is false and vice versa. Then in either a classical or quantum theory
of probability, we have

P (x = A or x = B) = P (x = A) + P (x = B),

at time t0 . Bayes’ rule assumes that this linearity will persist in time. In a classical theory,
this is reasonable, since ẋ is a function of x, which is simply to say that it can be definite
when x is. So,
∂P
Ṗ = ẋ + P (x)V̇ (x),
∂x
and similarly for higher time derivatives. Here, V is the volume element in the space of x
variables. The second term comes from the fact that the time evolution might not preserve
the volume element in the space of x variables, and so the probability density must change
in order to keep the total probability of being in a fixed region constant. These equations
are linear in P . Even if the x variables are in constant interaction with a random medium,
we always make the approximation that the medium’s effect comes through modifications
of the laws of motion of x by friction, random forces, etc., and we obtain a linear evolution
equation for P . As a consequence, we can say that if we measure x = A at t0 , then we can,
without making a mistake, ignore the fact that the theory predicted that x could have been
B, and model the future evolution of the system by saying that x was definitely equal to A.
In QM, this rule is simply invalid.6 The evolution equation is linear, but it is not an equation
6
In the Bohmian approach to QM, described in the appendix, one rewrites quantum mechanics in a manner
that looks like a probability theory for particle trajectories. However, the evolution equation for the
trajectories depends on the probability distribution itself, in a bizarre manner. Bayes’ rule is not satisfied.
54  Quantum Mechanics

for the probability but for the “Schrödinger wave function,” or probability amplitude. As we
have seen, the rules for probabilities that we sketched above imply that probability is the
absolute square of the complex probability amplitude (this is Born’s rule). So, it does not
evolve linearly and does not in general satisfy the Bayes rule of conditional probabilities.
We will see later that the essence of the measurement process in QM is to correlate the
microscopic state of an atomic system with a macroscopic variable (a pointer), for which the
quantum rules of probability obey Bayes’ rule with exquisite accuracy. All of the “paradoxes”
associated with “collapse of the wave function” are really just a consequence of the application
of Bayes’ rule to probabilities for macroscopic objects. The quantum violations of Bayes’ rule
for macrovariables are exponentially small and cannot be measured, even given a time much
longer than the history of the universe.
Note that because the Schrödinger equation
∂
i~ |ψ(t)i = H|ψ(t)i
∂t
is linear, we can apply a sort of quantum Bayes’ rule to the Schrödinger equation directly.
This is what is called collapse of the wave function upon measurement. We will see that it
only makes sense when the wave function is a linear combination of states with different
histories of macroscopic collective coordinates, in which case it implements Bayes’ rule for
probabilities, and defines what it means for something to “definitely happen” to a macroscopic
variable. The mathematically precise meaning of that phrase is that QM in general predicts
only probabilities,7 but it can define conditional probabilities for macrovariables (a shorthand
phrase we will use from now on to denote “the collective coordinates of macroscopic systems,
which are defined so that interference terms between different histories are too small to
be measured, even in principle”) for which one can define probabilities for conventional
histories by using Bayes’ rule. The approximate histories in this emergent classical probability
theory are called decoherent histories, because the phase coherence between parts of the wave
function describing different histories is wiped out. However, this use of Bayes’ rule is both
approximate and evanescent, in the sense that once the last collective variable involved in
a decoherent history has gone out of causal contact with a part of the system that has
disintegrated into elementary particles, then the correct predictions for future measurements
done on those particles must use the original wave function derived from the Schrödinger
equation, rather than the one gotten by applying Bayes’ rule after observing individual
histories of the collective coordinates. That is, we use the approximate Bayes’ rule obeyed by
collective coordinates to split the wave function into parts that correspond to fixed histories
of the collective coordinates. This procedure only makes sense if we can monitor the collective
7
These probabilities can be used in two ways: the Bayesian approach, which uses them to inform human
judgment about the validity of theories, and the frequentist approach, which requires us to do the same
experiment over and over again, to eliminate as much as possible the reliance on subjective criteria.
 Two State Systems: The Ammonia Molecule  55

coordinates in question. If they are inaccessible, then we must go back to the probabilistic
predictions of the original wave function. This is all perfectly reasonable if we think of the
wave function only as a device for computing probabilities. It is troublesome if we think of it as
a physical object. In Chapter 10, we will describe a thought experiment called Schrödinger’s
Bomb, which illustrates the question in a quite dramatic fashion.
The foregoing was a lot to swallow, and you should not be surprised if you do not
understand it at this point. The examples in Chapter 10 will clarify these difficult ideas.
We conclude this chapter by reiterating the important lessons we have learned:

• When we think about vectors in ordinary space, we know that we can refer them
to different orthogonal coordinate systems, and that the same vector has different
representations as a collection of numbers in different coordinate systems. The same
is true of vectors |vi in a general N -dimensional complex vector space. If |en i is an
orthonormal basis, then the coefficients in |vi = n vn |en i are the column vector
P

representation of the vector in that basis. Similarly, operators have different matrix
forms in different bases. The set of operators diagonal in a given basis |an i have the
form
X
A= f (an )|an ihan |,
n

since |an ihan | is the projection operator on |an i.

• Given a vector |si of length 1, the positive numbers P (an ) = |han |si|2 have the mathe-
matical properties associated with a probability distribution for the operator A to take
on the eigenvalue an . We call vectors of length 1, states of the system, and describe
these numbers as “the probability of being in the state |an i when the system is in the
state |si.” QM is essentially the statement that we take these probability distributions
as making predictions about actual physical systems. The key feature of these distribu-
tions is that not all variables can be certain at the same time, because not all operators
are diagonal in the same basis. When the equations of motion relate operators with
mutual uncertainty, then we are doing QM.

• We can study QM with either time-dependent states and time-independent opera-

tors (Schrödinger picture) or time-dependent operators and time-independent states
(Heisenberg picture). The physical predictions are the time dependence of expectation
values of operators.

The way this math applies to physics can be understood from the following table, which
contrasts the classical and quantum treatments of a system with N states
56  Quantum Mechanics

Classical Quantum

Observables Operators Diagonal in Operators Diagonal in

Basis |ci i ANY Orthonormal Basis
Pure States Projection Operators on |ci iProjection Operators on
ANY Unit Vector
Mixed States Diagonal Matrices Tr (ρ) = 1 Matrices Tr (ρ) = 1 w/
w/ nonnegative eigenvalues nonnegative eigenvalues

Expectation Values Tr (ρA) Tr (ρA)

Time Evolution Sequence of Permutations of Discrete or Continuous Set

Basis |ci i of Unitary Operators U (t)

In both frameworks, we can allow the time dependence to be carried by the states or
the observables. It is often easiest to think in the Schrödinger picture, where the vectors
corresponding to pure states are simply evolved into |ci i → S(tn )|ci i or |v(0)i → U (t)|v(0)i.
Both frameworks contain observables whose value cannot be certain at all times. These are
diagonalizable operators whose eigenstates are not (at any given time) the unit vectors |ci (tn )i
or |v(t)i. However, in the classical case, we can simply ignore these operators and declare
they are not observable.

2.11 FOUR STATE SYSTEMS

In order to understand the formalism, we will now study the case of a system with four
states. This is quite complicated, but simple enough to actually allow for an exact analytic
solution. Finding the eigenvalues of a matrix requires us to solve the equation

det (M − e) = 0,

which is a polynomial equation of order N , where N is the dimension of the Hilbert space. It
is only possible to give analytic formulae for the solutions of a polynomial when N ≤ 4. The
four-dimensional case is also the first case in which dimension is not a prime number. In all
such composite dimensions one can construct a description of the space as what is called a
tensor product of two smaller spaces whose dimensions involve fewer prime factors. We will
discuss this concept in more detail in Chapter 6, and use the four state system to understand
the nonintuitive nature of quantum probability theory in Chapter 10.
 Two State Systems: The Ammonia Molecule  57

To have a physical picture of a four state system, consider two different8 molecules, each
of which has, like ammonia, two low lying energy states. The detectable quantities of the two
individual molecules are two commuting copies of the Pauli matrix algebra.

σa σb = δab + iabc σc . (2.53)

τa τb = δab + iabc τc . (2.54)

σ a τb = τb σ a (2.55)
Again, we are using the summation convention over the three valued indices we denote by
a, b, c, d . . .
Now consider the following 4 × 4 matrices, which act in the combined Hilbert space of the
two molecule system. If the interaction energies between the two molecules are of the same
size as the splittings between their two lowest lying states, and thus much smaller than the
energy gap to higher molecular states, we can approximate the two molecule system by this
four state system. The matrices we consider are 1, σa , τa , σa τb , and there are 1+3+3+9 = 16
of them. 1 is the four by four unit matrix. We think of it as a tensor product of two 2 × 2 unit
matrices, 14 = 12 ⊗ 12 defined in the following way: for each matrix element of the matrix on
the right of the tensor product symbol, write the two by two matrix on the left of the tensor
product symbol, multiplied by that matrix element. In symbols,
 
! ! ae af be bf
a b e f ag ah bg bh 
⊗ = . (2.56)
 
c d g h  ce cf de df 
cg ch dg dh

Exercise 2.15 Show that

A⊗B =B⊗A
for any pair of 2 × 2 matrices. Note that if you write tensor product matrices in block form,
so that !
A11 B A12 B
A⊗B = ,
A21 B A22 B
then the two different orders use a different labeling for the basis of the tensor product space,
so the two orders are only conjugate by a permutation matrix. Avoid this by considering the
action on the basis |i, Ii.
All 16 of the matrices we wrote above should be thought of in this way. That is σa4×4 =
σa ⊗ 12 , etc. Using this fact you should be able to do
8
We make them different in order to avoid the constraints of identical particle statistics, which we will
discuss in Chapter 5.
58  Quantum Mechanics

Exercise 2.16 Show that the 16 matrices formed by tensor products of 2 × 2 matrices are
linearly independent. Hint: Consider all of the matrices in the basis where τ3 is diagonal.
Then the matrices (12 orσa ) ⊗ (τ± , or(1 ± τ3 ) each have a different 2 × 2 block of their 4 × 4
matrix not equal to zero. The result then reduces to the 2 × 2 case.

This result means that we can write any 4 × 4 matrix as a linear combination of these 16
matrices. Note that all of the matrices are Hermitian, so any Hermitian matrix, which could
be the Hamiltonian for the interacting molecules, is a real linear combination of these 16.
If the system is time translation invariant, the coefficients will be independent of time. The
unit matrix term simply shifts the overall zero of energy, so it has no physical consequences.
We will save the case of the most general Hamiltonian for the last exercise, and concentrate
on easily soluble examples of the form

H = ma σa ⊗ 12 + 12 ⊗ na τa + ma nb /Eσa ⊗ τb . (2.57)

Exercise 2.17 Write the Hamiltonian above explicitly, as a 4 × 4 matrix.

To solve for the eigenvalues of this Hamiltonian, we note that the operators Q1 ≡ ma σa ⊗ 12
and Q2 ≡ 12 ⊗na τa , both commute with H, as well as with each other. If several diagonalizable
operators commute with each other, then they are all diagonal in the same basis. Equivalently,
the eigenvectors of H can all be chosen to be eigenvectors of both Qi . In fact, the eigenvalue
of H is completely determined by the simultaneous eigenvalues of Q1,2 . That is

h = q1 + q2 + q1 q2 /E. (2.58)

The eigenvalues qi have dimensions of energy, and the energy scale E is put in so that the
same is true for the eigenvalues of the Hamiltonian. We can label the states by the eigenvalues
of Q1,2 . Those eigenvalues are determined by the equations

Q21 = ma ma 14 , (2.59)

Q22 = na na 14 . (2.60)
√
This tells us that√each of these operators has two degenerate states with eigenvalues q1 ± m2
for Q1 and q2 ± n2 for Q2 . The eigenstates can be computed using your knowledge of the
two state problem. You can call them

|q1 , q2 i ≡ |qi i.

Notice the labeling of the states by a pair of two valued indices, rather than a single four
valued index. This is a very general property of tensor product Hilbert spaces. A choice of
basis in each of the two spaces involved in the product determines a basis in the full space.
 Two State Systems: The Ammonia Molecule  59

Exercise 2.18 Find the eigenvectors |±, ±i as four-dimensional column vectors, in the basis
where 12 ⊗ τ3 and σ3 ⊗ 12 are diagonal.
The physical significance of the Hamiltonian eigenvalues is two fold. First of all, they give
us the four possible values of the energy for the two molecule system. In the real world, the
molecules can make transitions between these states by emitting light quanta (photons) and
the energies (and thus frequencies/colors) of the emitted photons will be differences between
these values. The second role of the eigenenergies is to determine the time evolution of a
general initial state X
|s(0)i = v(qi )|qi i. (2.61)
qi

It evolves to qi +q2 +q1 q2 /E

v(qi )e−i
X
t
|s(t)i = ~ |qi i. (2.62)
qi

The time-dependent density matrix is

q1 +q2 +q1 q2 /E−p1 −p2 −p1 p2 /E
e−i
X
t
ρ(t) ≡ |s(t)ihs(t)| = ~ |q1 q2 ihp1 , p2 |. (2.63)
qi ,pj
√
In this equation, q1 and p1 are independently summed over the values ± na na , while q2 and
√
p2 are independently summed over ± ma ma .

Exercise 2.19 Write out the density operator ρ(t) explicitly as a 4 × 4 matrix in the basis
where Q1,2 are diagonal. Now write the same density operator in the basis where τ3 and σ3
are diagonal, using your knowledge of the 2 × 2 case to find the form of the eigenvectors |qi i
in this basis.

Given the time-dependent density operator ρ(t), we can compute the time-dependent
expectation value of any normal operator A as Tr(Aρ(t)). This gives the answer to any
prediction possible for this time-dependent system in QM. We could, for example, start in an
eigenstate of A, a state where A has a definite value a1 . Then the expectation value formula
tells us that the probability to find one of the eigenvalues ai when we measure A at time t is
Tr (|ai ihai |ρ(t)).
The traditional classical view of time dependence is that a detectable quantity has some
value A(t) at all times. The analog of this in QM is to write ρ(t) = U (t)ρ(0)U † (t), where
H
U (t) = e−i ~ t and note that (by cyclicity of the trace: Tr (AB) = Tr(BA))

Tr(Aρ(t)) = Tr (A(t)ρ(0)), (2.64)

where the Heisenberg operator

A(t) ≡ U † (t)AU (t). (2.65)
60  Quantum Mechanics

Thus, the time dependence of the detectable’s value can equally well be computed as the
expectation value of a time-dependent operator in an initial probability distribution. Note
that the time dependence of a product of operators, according to the Heisenberg equation, is
AB(t) = A(t)B(t), (2.66)
which means that we only have to solve the Heisenberg equations for a set of operators,
which generate the full operator algebra by matrix addition and multiplication.
Exercise 2.20 For our four state system, define operators ψ1,2 by insisting that
√
ψ1 | − n2 , q2 i = 0,
√ √
ψ1 | n2 , q2 i = | − n2 , q2 i.
ψ2 performs the analogous lowering operation on the q2 eigenvalue. Prove that ψ1,2 together
with their Hermitian conjugates, generate the operator algebra. Find the explicit matrix
forms of these operators in the basis where Qi are diagonal.

Exercise 2.21 Write and solve the Heisenberg equations of motion, which express time
derivatives of the ψ1 in terms of functions of themselves and their conjugates.

2.12 FURTHER EXERCISES

2.4 Evaluate the matrix σ2 defined by the matrix equation σ1 σ3 = −iσ2 using explicit
matrix multiplication. Show that σ22 = 1 follows both from the algebraic definition and
the explicit matrix form.
2.5 Using standard probability theory, evaluate the expectation value of a general polyno-
mial P (E, D) of the energy and dipole moment of the ammonia molecule. Show that
this is equivalent to the formula TrP (E, D)ρ.
2.6 The three-dimensional Levi-Civita symbol, or  symbol is denoted abc . 123 ≡ 1 and
for any other permutation of the three indices it is ±1 according to the evenness or
oddness of the permutation of 123. Show that
abc cde = δad δbe − δae δbd .
Use this equation to evaluate ∇ × (∇ × V), for a vector function V(x).
2.7 Show that the trace of a matrix is independent of the basis used to compute the trace.
First show that the diagonal matrix element of A in some orthonormal basis is equal to
hei |A|ei i, and then use the fact that 1 = i |ei ihei | for any choice of basis. Also, prove
P

that Tr AB = Tr BA for any two matrices.

 Two State Systems: The Ammonia Molecule  61

2.8 Consider a matrix A in an n-dimensional complex vector space and a density matrix
i pi |ei ihei |, where hei |ej i = δij , pi ≥ 0 and
P P
ρ = pi = 1. Assume that A has
an orthonormal basis of eigenstates A|ak i = ak |ak i, hal |ak i = δkl . Evaluate Tr (Aρ) =
Tr (ρA) = k hak |ρA|ak i ≡ k ak P (ak ). Show that the numbers P (ak ) are nonnegative
P P

and sum to one. That is, they satisfy the rules for a probability distribution for A to
take on one of its eigenvalues.

2.9 According to Born’s rule, if the ammonia molecule is in one of the states |±i3 , the prob-
√
abilities for the operator na σa to take on the values ± na na are given by the absolute
squares of the scalar products of these states with the corresponding normalized eigen-
vectors of the operator. Find these normalized eigenvectors as linear combinations of
|±i3 and compute the probabilities.

2.10 Prove that the most general 2 × 2 matrix diagonalizable by a unitary transformation
has the form A + Bna σa , where A and B are complex numbers and na is a real three
vector. First prove that any matrix M that is diagonalizable by a unitary must satisfy
[M, M † ] = 0. In proving this you will have to argue that every matrix can be written
as a linear combination of the unit matrix and the three Pauli matrices. Then use the
multiplication law of Pauli matrices to show that only the indicated form satisfies this
commutator equation. To show that any matrix of this form can indeed be diagonalized,
find the eigenstates of na σa and show that they are orthonormal. The basis formed
by these eigenstates is thus related to the basis where σ3 is diagonal by a unitary
transformation (we will see this quite generally in Chapter 6, but you can try to prove
it in the two-dimensional case if you would like). Finally show that A+Bna σa is diagonal
in this same basis.

2.11 Prove that every operator of the form eiα eina σa , where na is a real 3-vector, and α is
real, is unitary. One way to do this is to evaluate the exponential matrix explicitly using
the algebra of the Pauli matrices. You should find a diagonalizable matrix of the form
studied in Exercise 2.10. Show that the eigenvalues are pure phases and argue that this
implies that matrix is unitary.

2.12 Generalize 2.10 to n dimensions, in the following way. Argue that the commutator of
two Hermitian matrices is antihermitian, which means that it is i times a Hermitian
matrix. If λa is a basis for the space of all hermitian n × n matrices (show that there
are n2 independent elements in this basis), the previous sentence means that

[λa , λb ] = ifabc λc .

It is obvious that fabc is antisymmetric in its first two indices. Multiply the equation by
λd and take the trace to show that it is in fact totally antisymmetric. Use this to prove
62  Quantum Mechanics

a statement analogous to the first step in Exercise 2.9, which involves two complex
constants and a real vector of dimension n2 − 1.

2.13 Show in general dimension that eiH , where H is a hermitian matrix, is unitary. Assume
that H is diagonalizable by a unitary transformation (sketch of the proof in Chapter 6).

2.14 Show that for any matrix M diagonalizable by a unitary transformation

det M = etrlnM .

∗
2.22 Consider a general 4 × 4 Hermitian matrix Hij = Hji , in the basis where τ3 and σ3 are
diagonal. Use Mathematica, Maple or Sage to find the eigenvalues and eigenvectors of
the matrix. Write an explicit form for the time-dependent density matrix correspond-
ing to this Hamiltonian, starting from a density matrix diagonal in the original basis.
Evaluate the time-dependent expectation values of the projection operators on each of
the common τ3 and σ3 eigenstates.

2.23 Apply the general solution of Exercise 2.22 to the special Hamiltonian studied in the
text. At what point in the analysis does one see evidence for the conservation laws of
Q1,2 ?
 CHAPTER 3

Quantum Mechanics of a
Single Particle in
One-Dimensional Space I

3.1 INTRODUCTION
In this chapter, we will derive the quantum mechanics (QM) of a free particle moving on
an infinite line, from invariance properties. This will be another example of the quantum
version of Noether’s theorem. We will show that translation invariance implies the existence
of an operator P , which commutes with the Hamiltonian, and makes infinitesimal changes
in the value of the coordinate x of the particle on the line. We will derive the spectrum
of P and show that the Hamiltonian must simply be a function of it. Since the system is
also invariant under space reflection (x → −x) the expansion of H(P ) in powers of P/m
P2
must begin as 2m . That formula suggests that we identify the conserved quantity P with the
classical quantity called momentum. It leads to a system invariant under Galilean boosts to
frames moving with constant velocity. The simplest description of a particle moving under
the influence of external forces is to add a term V (X) to the Hamiltonian, which breaks
translation invariance.
We will also show that these considerations lead to the fundamental Heisenberg commu-
tation relation
[X, P ] = i~, (3.1)

which tells us that position and momentum cannot simultaneously have definite values. This
implies that there are no probabilities for histories of particle motion. We will briefly discuss
an alternative Hilbert space treatment of free particle motion, which does allow for the
notion of particle histories. In that “quantum theory of classical mechanics,” momentum

63
64  Quantum Mechanics

and position commute and can take on definite values simultaneously. Only experiment can
decide which of these two theories to use for a given system. Experiment confirms that the
quantum, rather than the classical description of particles, is the one relevant to real atoms
and molecules, as well as more elementary particles.

3.2 TRANSLATION INVARIANCE AND FREE PARTICLE MOTION

3.2.1 No Experimental Definition of Infinity
The real line is infinitely long and contains infinitesimally small intervals. Both of these
introduce mathematical complications. However, as physicists, we know that we will never
be able to know for sure if space is infinite, or infinitesimally divisible. Mathematicians know
that the definition of the infinite real line can be thought of by taking two kinds of limits of a
clock with a finite number of “minutes.” We can first take the limit where the minutes become
seconds, microseconds, picoseconds . . . and then the limit where the circumference of the clock
goes from 1 to 10 m to 100 million km . . . If you now ask an experimental physicist to test
the mathematician’s notion of infinity, she will tell you that she cannot. The mathematician
defines an infinite continuous line by studying the behavior of a finite discrete set of points,
and taking an imaginary limit in which the number of points gets larger and larger and
the distances between them smaller and smaller. The physicist, with finite resources and a
finite amount of time to do measurements can only verify the mathematical properties of the
infinite line with limited precision. In other words, she cannot tell the difference between an
infinite continuous line and a finite and/or discrete one, by discovering that the line is not
really infinite or continuous. For any given experiment, there will be a finite discrete model
of the line, which fits the data as well as the infinite continuous one.
It is useful to think of the infinite continuous line as the limit of a clock, because it
illustrates some very general basic features of QM. Look at Figure 3.1, which illustrates a
sequence of clocks of radius R, divided up into N equal “minutes,” for a range of N . We are
discussing motion on the rim of the clock. One of the detectable quantities associated with
motion on the rim of the clock is its position. We can describe this as a complex number
2πin
zn = Re N where n ranges from zero to N − 1. A more abstract way of thinking of the
different values of position is that they correspond to functions of n defined by em (n) = δmn .

...

Figure 3.1 Clocks with N minutes.

 Quantum Mechanics of Single Particle in 1D Space I  65

You should think of this equation as defining the n-th component of a vector |em i. The set
of all complex valued functions of n is a complex vector space, and the functions em (n) are
a basis for that space. That is,
X
f (n) = f (m)em (n). (3.2)
m

There is a position operator, U , in this vector space defined by

RU f (n) = zn f (n), (3.3)

and you should convince yourself that it is a linear operator. You will study all the properties
of this operator, and the translation operator we are about to introduce, in the Exercises.
For the moment, we are just using them to get some intuitions. We are thinking about the
position operator as defining a point on a circle of radius R, centered at the origin of the
complex plane. The functions em (n) are eigenstates of this operator, with eigenvalue zm . As
N gets large, the dimension of the space increases, and the numbers zn start to fill up the
rim of the clock densely.
We now want to think about an operator V , which moves you from one point on the
clock to the one with next larger n. That is

V em (n) = em+1 (n) = δm+1,n . (3.4)

On a general function f (n), we have

X
V f (n) = f (m)V em (n) = f (n − 1). (3.5)

Obviously, if we apply V N times, we get back to where we started, so V N = 1. This means

2πik
that the eigenvalues of V are e N , k = 0 . . . N − 1, and they form a “dual clock.” The
eigenvectors corresponding to these eigenvalues satisfy
2πik
fk (n − 1) = e N f (n). (3.6)

The solution to this is

1 2πik
fk (n) = √ e− N . (3.7)
N
We have chosen the normalization factor so that
X
|fk (n)|2 = 1. (3.8)
n

The eigenstates of V are discrete analogs of standing waves on the circle and k is what we
would call the wave number of the wave. Note that as N gets large, the spectrum of values
66  Quantum Mechanics

of k gets infinitely large, just like the spectrum of values of n. However, if we want to think
of 2π Nn as a continuous variable, the values of k remain discrete. Thus, a continuous circle
corresponds to a discrete infinite lattice of wave numbers. For finite N , both the clock rim and
the space of wavenumbers are discrete finite sets. As N goes to infinity, they both become
infinite, but in different ways. The circle remains compact but becomes continuous, while
wave number space becomes an infinite lattice.
We can also contemplate the opposite limit, in which space becomes an infinite regular
lattice, while wave number space is a circle, which in this context is called the Brillouin zone.
This limit has applications to the study of solids which have a periodic crystal structure.
We will discuss this briefly in Chapter 11. See Figure 3.2 for an illustration of how the two
complementary limits of continuous circle and infinite lattice are related in position and wave
number space.
For finite N , it is easy to see that the set of all wave number eigenfunctions is a basis
for the space. The easiest way to do this is to compute the scalar product of two different
wave numbers. You should do this for yourself before reading the next few lines. The scalar
product is

1 X 2πi (k−l)n
fk∗ (n)fl (n) =
X
e N . (3.9)
n N n

For k = l this is 1. For k 6= l it is the sum of all of the N th roots of unity, which equals zero.
So we have N orthonormal vectors in an N -dimensional space, and these vectors form a basis.
As we go through the next two chapters you will realize that this little computation is the
basis for the proof of Fourier and Plancherel’s theorems about writing any square integrable

Position Wave Number

The relation between continuous/discrete circles in position and wave number.

Figure 3.2
Arrowheads indicate the line extends indefinitely.
 Quantum Mechanics of Single Particle in 1D Space I  67

function as a Fourier series or Fourier integral. It will also be the key to understanding what
momentum is in QM and to Heisenberg’s Uncertainty Relation for position and momentum.
Now we return to the continuous circle, which has circumference 2πR. We can parame-
terize it by a variable x which goes from −πR to πR. The geometry of the circle is invariant
under translations of x if we make the identification x ≡ x + 2πRm, where m is an integer.
The Hilbert space of the theory becomes the space of periodic functions of x satisfying
Z πR
|f |2 < ∞. (3.10)
−πR

The variable x has dimensions of length, so a function eikx will be periodic if k = n/R,
where n is an integer. These functions just get multiplied by a phase eika when we translate
x → x + a, so they are eigenfunctions of the translation operator V (a), which is defined by

V (a)f (x) ≡ f (x + a). (3.11)

This one parameter set of operators is the limit of the discrete set V n in the N → ∞ limit.
So far we have not said anything about dynamics. We can make some very interesting
general conclusions about free particle motion on discrete or continuous circles by making
two assumptions:

• We do not need any more variables to specify the states of our particle than we have
already described.

• The dynamics of a free particle is invariant under translations around the circle.

The second assumption has some immediate mathematical consequences. It says that the
Hamiltonian matrix
hen |H|em i (3.12)
is the same as the matrix in the translated basis V |en i. That is

hen |V † HV |em i = hen |H|em i. (3.13)

Since this is true for every matrix element, it implies the operator identity

V † HV = H, (3.14)

or
HV = V H. (3.15)
Since H and V commute, they are diagonalizable in the same basis. On the continuous circle,
the same arguments go through with the operator V replaced by V (a).
68  Quantum Mechanics

What this means is that in the basis where V (a) is an infinite diagonal matrix with matrix
an
elements ei R , with n running over all integers, H is a diagonal matrix with matrix elements
H(n). Remarkably, this means that in any quantum theory of free particle motion, based
on these two assumptions, the allowed energies of a free particle on a continuous circle are
discrete. This is the famous quantization of continuous classical variables, which gives QM
its name. We will see later that it occurs any time the particle is confined in a fixed region
of space, or even when we insist that the probability of finding the particle at infinity goes
to zero rapidly enough for the function representing its state to actually be normalizable.
If we take the model of an infinite discrete line instead of a continuous circle, we can see
that quantization of energy no longer occurs. Notice that this system is exactly the same
Hilbert space as that of a particle on a circle. We simply think of the circle as wave number
eigenstates and the discrete line as positions. It is clear then that the translation operator
from one point to the other on the line has a continuous spectrum, so the argument above
will say the same for the energy operator.
We learn a general lesson: quantization of energy has to do with motions that are bounded
in space. In QM, where we can only talk about probabilities, this is equivalent to saying that
the probability of finding the particle at any finite point in space, goes to zero sufficiently
fast that the total probability is one. Systems in infinite space can also have discrete energy
eigenstates if they have bound states. The states with continuous energy eigenvalues, generally
look like free particle states at infinity and one can investigate what happens when one throws
a free particle into finite regions, where it interacts. This is called a scattering problem and
the corresponding eigenstates are called scattering states. We will learn more about them in
Chapters 4, 8, and 16.
In the previous Chapter, we argued that a “classical” theory of the two states of the
ammonia molecule was a special case of the quantum theory, in which time was taken discrete
and the time translation operator was always the matrix σ1 . One might have expected the
same would be true for a particle moving on a circle, but it is not. The classical physics of
a particle moving on a circle has a continuous energy spectrum. The reason is that it does
not satisfy the first assumption above. In classical physics, the velocity or momentum is an
independent variable at fixed time, which determined the future history of the system. So
a state is characterized by two parameters x, p, rather than only x. We will see below that
there is a particular form of QM on the Hilbert space of functions of these two variables,
which reduces to the standard formulation of classical mechanics. We are now ready to start
a more formal treatment of the QM of particles in one dimension.

3.2.2 Free Particles on an Infinite Continuous Line

Despite the issues of principle that we have pointed out in connection with the operational
definition of the notion of infinity, the use of continuous and infinite sets is useful to physicists
 Quantum Mechanics of Single Particle in 1D Space I  69

because it simplifies mathematical calculations and throws away nonuniversal behaviors of

discrete finite systems, which might not show up in a given set of experiments. In this section,
we will treat some infinities cavalierly, referring to Chapter 4 for more careful definitions.
We start by thinking about a particle moving freely on the real line. The salient fact
about this system is that it is invariant under translations of the position x of the particle.
Among the detectable quantities in this system are functions, A(x) of its position. For any
state |ψi, the probability amplitude to find all of these functions evaluated at a particular
value x = y is a complex number ψ(y) called the wave function of the particle. The continuous
analog of probabilities summing to one is
Z ∞
dx|ψ(x)|2 = 1.
−∞

We can think of the meaning of ψ(x) in Dirac notation as ψ(x) = hx|ψi, where |xi is “the
state where the particle is definitely at the point x.” If the real line consisted of discrete points,
this would be a perfectly sensible normalized state, but because we have taken the imaginary
limit of a continuous line, it turns out that its squared norm hx|xi is infinite. Nonetheless,
the formal expression Z a+
P (a, ) = dx|xihx|, (3.16)
a−

defines a perfectly sensible operator in the Hilbert space of a free particle on a line. It is an
operator whose value is one if the particle is between the two points a ±  and zero otherwise.
Its expectation value in a general state is
Z a+
hψ|P (a, )|ψi = dx |ψ(x)|2 . (3.17)
a−

To see this, consider the discrete line for a moment. A particle hopping from point to point
has a space of states which is infinite dimensional. A particular orthonormal basis consists of
precisely those states |ni (−∞ ≤ n ≤ ∞) where the particle is definitely at the n-th point.
A general normalized state will be characterized by its projections hn|ψi on this basis. The
projections satisfy X
|hn|ψi|2 = 1. (3.18)
n

The operator
n=b+M
X
PM (b) = |nihn|, (3.19)
n=b−M

is a projection operator which gives zero on a state |mi unless b − M ≤ m ≤ b + M . Now

imagine taking the limit as the spacing s between the discrete points goes to zero, and that
70  Quantum Mechanics

in this limit we replace ns by a continuous variable x. Then, recalling your first year calculus
course, you will realize that sums get replaced by integrals. The projector
R∞
PM (b) converges
to P (a = bs,  = M s), and the normalization condition is just −∞ dx|ψ(x)|2 = 1. The
“states” |yi are not normalizable vectors in our Hilbert space, because their wave function is
the Dirac delta function δ(x − y). We will explain this in more detail in Chapters 4 and 6. A
quick introduction to these delta function normalizable states can be found in Appendices B
and F.
For a free particle in infinite space, translation invariance should be a property of the
energy operator H. That is, a translated state, with wave function ψ(x + a), should have
the same expectation values of any function of the energy operator, as the state with wave
function ψ(x). Furthermore, as we discussed in the previous chapter translation should be
represented by an operator V (a) on |ψi, which satisfies

V (a)V (b) = V (b)V (a) = V (a + b),

as well as V (0) = 1. V should also be unitary, in order to conserve the probability inter-
pretation of the translated wave function (see Chapter 6 for an extensive discussion of the
necessity for choosing symmetry operations to be unitary, as well as the single exception to
this rule).
So we must have V (a) = eiKa (prove this in Exercise 3.1), where K is an operator with
real spectrum (the spectrum of an operator is the set of its eigenvalues). K is called the wave
number operator. It is Hermitian and the translation operators are unitary.

Exercise 3.1 Prove that a one parameter family of unitary operators satisfying U (a)U (b) =
U (b)U (a) = U (a + b), as well as U (0) = 1, has the form eiKa , where K is an operator with
real spectrum. Hint: the operators U (a) are unitary and commute with each other, so can all
be diagonalized in the same basis.

The probability distribution, which says that the state of the system is definitely |ψi, is the
projection operator P = |ψihψ|. The translated projection operator is P (a) = V (a)P V † (a),
so the condition that the Hamiltonian have the same expectation value after translation is

Tr (HV (a)P V † (a)) = Tr (V † (a)HV (a)P ). (3.20)

This must be true for every a and every projection operator P , and that is only true if

[H, K] = 0. (3.21)

To see this, note that, for a projection operator on the state |ψi we have

Tr (V † (a)HV (a)P ) = hψ|V † (a)HV (a)|ψi. (3.22)

 Quantum Mechanics of Single Particle in 1D Space I  71

So the expectation value of the difference between H and the translated Hamiltonian vanishes
in every state. If we now apply this rule to a state which is a superposition |ψi = |α1 ψ1 +
α2 i|ψ2 i and insist that the expectation value be independent of a for every pair of states and
arbitrary coefficients, we conclude that

hψ1 |V † (a)HV (a)|ψ2 i = hψ1 |H|ψ2 i, (3.23)

which is the same as the operator identity

V † (a)HV (a) = H.

This can only be true for all a, if K commutes with H.

In the previous section, we established that the time evolution of operators is
i i
A(t) = e ~ Ht A(0)e− ~ Ht , (3.24)

from which it follows that

i
Ȧ(t) =
[H, A]. (3.25)
~
This is called the Heisenberg equation of motion and is obeyed by every operator in the
Heiseberg picture, whose corresponding Schrödinger picture operator is independent of time.
The Heisenberg equation of motion then implies

dK i
= [H, K] = 0. (3.26)
dt ~
In other words, K is a conserved quantity. This connection between symmetries and conser-
vation laws is one of the most important general principles in theoretical physics. In QM, it
follows naturally from the idea that operations on the system are also measurable quantities.
A symmetry is, by definition, an operation on the system, which does not change the Hamil-
tonian. In QM, operations are implemented by unitary transformations, so this is equivalent
to saying that there is a group of unitary transformations, which has the same1 multiplication
laws as the group of symmetries. Every element of the group commutes with the Hamiltonian,
which, by Heisenberg’s equations, tells us that it is a conservation law. Groups that depend
on continuous parameters have an infinite number of different elements. However, as we will
see, they are all gotten by concatenating infinitesimal symmetry transformations, so if there
are a finite number of independent parameters, we only get a finite number of conservation
laws.
1
Actually, there is freedom for what is known as a central extension of the group, stemming from the fact
that quantum states are only defined up to an overall phase. We will not talk about this in this book,
except when discussing particles of half-integral spin.
72  Quantum Mechanics

Let us stop and reiterate that we have just proven the quantum mechanical version of
Noether’s famous theorem[8] from classical Lagrangian mechanics (see Appendix C). Almost
nothing we have said so far distinguishes the group2 of one-dimensional spatial translations,
from any other group of symmetries that depends on a continuous parameter a. Every such
group is represented in QM by a one parameter group of unitary operators U (a) = eiaG .
G is a Hermitian operator called the infinitesimal generator of the symmetry group. Invari-
ance of the expectation value of the Hamiltonian in an arbitrary state, under the symmetry
transformations then implies that the Hermitian operator G commutes with the Hamiltonian.
Returning to the example of the translation group, we note that since H and K commute,
they are diagonal in the same basis. Since the spectrum of K is nondegenerate (all the different
real numbers), this means that H is a function of K. That is, H just has some number in
each of the diagonal places where K takes on the real value, k, so its values are labeled by
H(k). We are being somewhat cavalier about the idea of matrices with continuous indices at
this point, and we will be much more careful about this in Chapter 6.
We will use one more fact about the classical mechanics of free particles, namely that the
velocity
P
Ẋ = ,
m
is a translation invariant quantity. This follows simply from the fact that it is a limit of a
difference of positions, which does not depend on the choice of origin. We conclude that the
momentum P commutes with K and so is also a function of K, which commutes with H.
The last fact deserves to be written down, because it implies
dP i
= [H, P ] = 0, (3.27)
dt ~
which is Newton’s equation for free particle motion. This is pretty amazing. We have derived
Newton’s equation for free particles by just using symmetries. We are not quite done yet.
Now let us examine the action of K on wave functions. The state eiKa |ψi is supposed to
be translated by an amount a. That means that its amplitude to be at the point x is the
same as the amplitude of the state |ψi to be at x + a. If we take a to be infinitesimal, this
implies that K acts on ψ(x) like 1i ∂x (we use a partial derivative in this one-dimensional
space because we will also want to talk about the time derivative of wave functions). If
we think of ψ(x) as “the column vector corresponding to |ψi in the |xi basis”, then this
partial derivative representation of K is like a matrix representation of an operator in a
finite dimensional Hilbert space.
The operators V (a) have eigen-states which satisfy
V (a)|ki = eika |ki, (3.28)
2
If you know nothing about group theory at this point, please take a moment to read the appendix on that
subject before going on. We will not use much of the math, but you should get used to the words.
 Quantum Mechanics of Single Particle in 1D Space I  73

and are also eigenstates of K, H(K) and P (K), with eigenvalues given by k, H(k) and P (k).

Exercise 3.2 Prove that the wave functions of K eigenstates are

1
ψk (x) = √ eikx . (3.29)
2π
Solution of Exercise 3.2: Using the partial derivative representation of K we have

1
Khx|ki = ∂x hx|ki = hx|K|i = khx|ki. (3.30)
i
The exponential function is the unique solution to this homogeneous differential equation
and we have chosen a conventional normalization for the exponential. You might think that
we should have determined the normalization by insisting that the total probability be 1.
However, in a wave number eigenstate, the probability density for finding the particle at
position x is uniform over the real line. This implies that these states cannot be normalized.
We can understand this a little better, and relate it to the analogous inability to normalize
the states |xi by putting the system on a circle, rather than the real line. Then, the allowed
n n
values of k are R where n is an integer and R the radius of the circle. The states |k = R i are
normalizable because the range of integration in the formula for the squared norm is finite.
The analogy becomes more appealing when we recognize that on the discrete infinite line,
where the group of translations consists of discrete shifts by an integer number of lattice
spacings, the wave numbers, eiks , appearing in the eigenstates of the discrete translation
operator are continuous, but periodic under k → k + 2π s . If wave number states live on a
circle, position eigenstates are discrete and normalizable. If position states live on a circle,
wave numbers are discrete and wave number eigenstates are normalizable.
The relationship between the amplitudes to be in position or wave number eigenstates is
now given by
Z ∞ Z ∞
dx −ikx
hk|ψi = dx hk|xihx|ψi = √ e ψ(x). (3.31)
−∞ −∞ 2π
This means that the wave function in wave number space is the Fourier transform of the
wave function in position space. The next few paragraphs will assume you know all about
Fourier transforms. The next subsection contains a primer for those readers to whom that
assumption does not apply.
Note that mathematics textbooks give a variety of choices of the factors of 2π in the
definition of Fourier transform. We have chosen a particular one based on our desire to have
the Fourier transform be symmetric between wave number space k and position space x.
Indeed, this discussion shows that the Fourier transform is the continuous analog of a change
of basis in a finite dimensional vector space, so it should be implemented by a unitary
74  Quantum Mechanics

transformation. The inverse of the Fourier transform expresses position space wave functions
in terms of wave functions in wave number space
Z ∞ Z ∞ dk
hx|ψi = dk hx|kihk|ψi = √ eikx ψ(k). (3.32)
−∞ −∞ 2π

We can make wave functions more localized than the wave number eigenstates by taking
superpositions Z ∞
dk
ψ(x) = √ f (k)eikx , (3.33)
−∞ 2π
where
dk
Z
|f (k)|2 = 1.
2π
The last equality is called Plancherel’s theorem[9], which says that the squared norm of a
function and its Fourier transform are the same. We want dx|ψ(x)|2 = 1, because that is
R

the total probability for finding the particle anywhere on the line.
In calling these integrals the norm of a wave function, we are using the relation described
above between quantum states and their wave functions, and anticipating a definition of the
Hilbert space in terms of square integrable functions, which we will give in the next chapter.
At this point, the reader may benefit from reviewing the discussion in Chapter 2 of finite
dimensional vector spaces, as function spaces on the set of integers from 1 to N . And while
we are on the subject of mathematical theorems, Fourier’s theorem[10] tells us that we can
represent any function on the real line as a Fourier transform, including those corresponding
to localized lumps.
Using the general rules of QM, we can now evaluate the time evolution of this general
wave function, assuming that the equation above gives its form at t = 0. It is

e−iH(k)t f (k)eikx .
X
ψ(x, t) = (3.34)
k

For a localized wave function, localized near x0 , f (k) can be concentrated near some k0
but must have lots of other nonzero values. The general rule of thumb is that the more
localized the Fourier coefficients f (k) are in k space, the less localized the position space
wave function is.
For example, we could try a Gaussian falloff
2
f (k) = N e−(k−k0 ) .
∞
dx |ψ(x)|2 = 1, and can be easily calculated.
R
The normalization factors are determined by −∞
For any such function, concentrated around k0 , but containing enough weight for other values
 Quantum Mechanics of Single Particle in 1D Space I  75

that ψ(x) is localized at x0 , then, assuming that H(k) is a smooth function of k near k = k0 ,
we can approximate the time evolution by
H 0 (k0 ) H 0 (k ) H 0 (k0 )
Z
H(k0 ) H(k0 )
−k0 ~ 0 )t
ψ(x, t) ∼ dkf (k)e−i ~
t
e−i ~
(k−k0 )t+ikx
= e−i( ~ ψ(x − t). (3.35)
~
That is, the center of the wave “packet” moves through space according to the equation
H 0 (k0 )
x(t) = x0 − t, (3.36)
~
which is the equation of motion for a free particle if
H 0 (k0 ) P (k0 )
= . (3.37)
~ m
If we compare this to the Heisenberg equation of motion
P (K) i
= Ẋ = [H(K), X], (3.38)
m ~
we learn that
P (K)
i[H(K), X] = ~H 0 (K) = ~ . (3.39)
m
This is solved by
∂
X = i~ , (3.40)
∂k
in the basis where K is diagonal.
To fix the function H(K), we recall from our discussion of the ammonia molecule, the
principle that frequency is the same as energy divided by ~. Using the classical formula for
energy,
P2
E= ,
2m
and the formulae above, we derive the relations

P = ~K, (3.41)

~2 K 2
. H= (3.42)
2m
Alternatively, we could have derived the same formulae by insisting that the theory of a free
particle be invariant under Galilean transformations (see Exercise 3.2). Yet a third way of
deriving this formula is to just assume that we are interested in k much smaller than some
specified inverse length L−1
c . Then H(k) will have a power series expansion. The term linear in
76  Quantum Mechanics

k vanishes by invariance under reflections, while the term quadratic in k must have dimension
of inverse mass. The factor of two may be thought of as a conventional normalization of the
definition of mass. Higher order terms involve (by dimensional analysis) a new parameter
with dimensions of velocity, which Einstein discovered had to be the velocity of light in
vacuum.
We also record the fundamental commutation relation we have derived:

[X, P ] = i~. (3.43)

This says that position and momentum are not compatible observables. Measuring one of
them introduces uncertainty in the other. Since both are needed in order to predict the
history of particle motion, the histories will be uncertain. Note however that when m is very
P
large, the uncertainty in the velocity Ẋ = m is very small, so the history will not be that
uncertain. This is part of the explanation of why macroscopic objects behave classically. We
will explore this in more detail in Chapter 10.
When we discussed the ammonia molecule, we said that there was a choice to be made
between having a classical theory of its motion and a quantum theory. Only experiment
could tell us which was right. The same is true for free particle motion. In the classical
theory, X and P are simultaneously measurable, and you will show in Exercise 3.3 that the
Hamiltonian operator, in the basis where both X and P are diagonal, is

H = i~ (∂x E∂p − ∂p E∂x ), (3.44)

2
p
where E = 2m + V (x) is the classical energy. That is, classical mechanics is QM in a Hilbert
space whose vectors are functions of x and p, with Hamiltonian operator H. Note that in this
case the Hamiltonian operator is not the energy E(p, q), although it does commute with it,
so that both quantities are conserved. As you will show in the exercises, we can choose initial
states in which both x and p are known with precision, and these evolve into other eigenstates
of x and p. More general operators, involving the derivatives w.r.t. x or p are uncertain, but
if we agree not to measure them, then the theory makes only precise predictions.
For microscopic particles, experiment shows that our quantum form of the equations for
a free particle is the correct one. We will see later that the classical theory is approximately
valid for collective coordinates of macroscopic objects.
To summarize: translation invariance of free particle dynamics in one dimension leads,
via general principles of QM, to the conclusion that the system has a conserved quantity
K which takes on all possible real values. The momentum and energy must be functions of
K. Using the classical formula for energy as a function of momentum, or imposing Galilean
symmetry, we conclude that P = ~K, and that [X, P ] = i~.
 Quantum Mechanics of Single Particle in 1D Space I  77

3.2.3 Primer on the Fourier Transform and Fourier Series

Given an infinite one-dimensional lattice, we can define a Hilbert space with orthonormal
basis vectors |ni. A general vector is given by
∞
X
|ψi = ψ(na)|ni. (3.45)
n=−∞

∞
X
|ψ(na)|2 < ∞. (3.46)
−∞

You can think of ψ(na) in two ways: it is both an infinite column vector and a function of
the lattice points. These are two different names to describe the same mathematical object.
We can define a linear operator on this space by translation through a single lattice point:
∞
X
V |ψi = ψ((n + 1)a)|ni. (3.47)
n=−∞

It has a continuous set of eigenvalues/eigenvectors

V |αi = eiα |αi. (3.48)

They are given explicitly by

√ ∞
X
|αi = N einα |ni, (3.49)
n=−∞

where N is a normalization constant. Note that α lives on a circle

α ≡ α + 2πm. (3.50)

Note that the function of na corresponding to this “vector” is

√ α
ψα (na) = N ei a na . (3.51)

We put air quotes around the word vector because it is not normalizable
X
hα|αi = N = ∞. (3.52)
n

More generally: X
hα|βi = N ein(β−α) . (3.53)
n
78  Quantum Mechanics

Given a vector |ψi we can define a function of α by

∞ ∞
e−inα hn|ψi = e−inα |ψ(na)i.
X X
ψ(α) = hα|ψi = (3.54)
n=−∞ n=−∞

Now if we evaluate
Z 2π Z 2π
δmn = hm|ni = dα hm|αihα|ni = N dα ei(n−m)α = 2πN δmn , (3.55)
0 0

we see that we must choose N = √1 . We can also derive from this (do it!) that
2π

∞
X
δ(α − β) = ein(α−β) . (3.56)
−∞

You should now think about all of these formulae and realize that you have proved Fourier’s
theorem: It is all about a unitary transformation mapping the space of square integrable
periodic functions ψ(α) into the space of square summable infinite sequences, which can be
thought of as the space of functions ψ(na) on an infinite lattice. We use the same letter
to denote the function on the lattice and the function on the circle, even though they have
different functional forms. The different nature of the argument tells you which one you are
talking about.
The Fourier transform is the limiting form of this Fourier series story, when we take the
limit of zero lattice spacing, so that x = na becomes a continuous variable.

Now a → 0 x = na. (3.57)

In this limit, the periodicity of α goes away.

α
e−inα → e−ikx , k= . (3.58)
a
The range of k now becomes infinite, −∞ < k < ∞ and the periodicity is gone. Sums over
the discrete variable n are replaced by integrals over x in the usual way, but we now have the
freedom to rescale x and k keeping kx fixed. In QM we do this to make the Fourier transform
look completely symmetrical between x and k.
dx
Z
ψ(k) = √ e−ikx ψ(x). (3.59)
2π
The normalization is determined by Fourier’s theorem, which is equivalent to
dk ik(x−y)
Z
δ(x − y) = e . (3.60)
2π
 Quantum Mechanics of Single Particle in 1D Space I  79

You should understand how to prove Fourier’s theorem, as well as Plancherel’s theorem about
the equality of the norms of a function and its Fourier transform, using this equation. In the
exercises, you will learn how to do the Fourier transform of a Gaussian. You should also
2
practice by figuring out the Fourier transform of the function e−b |x| and of a function which
is equal to 1 on the interval [b, c] and zero everywhere else. These last two depend only on
knowing how to integrate an exponential function between finite limits.

3.3 LAGRANGIAN AND HAMILTONIAN MECHANICS: A REMINDER

OR A PRIMER
In the best of all possible worlds, students would take a course in Lagrangian and Hamiltonian
mechanics before taking an advanced undergraduate course in QM. In this book, we will use
the actual formalism of analytical mechanics only sparingly, but like notions from group
theory, we will use the language of that subject more extensively. What follows is a bare
bones introduction to the subject.
The solution to Newton’s equations for a free particle is a straight line
p
x(t) = x0 + t. (3.61)
m
The fact that straight lines minimize distance, and Fermat’s principle that light travels in
such a way as to minimize the time, led 18th century physicists to formulate this equation
as a minimum principle. Namely, the path minimizes the integrated kinetic energy
Z t m dx 2
0=δ ds ( ) . (3.62)
t0 2 ds

This can be generalized to get Newton’s equation in the presence of a force

d2 x
m = F, (3.63)
ds2
if the force is conservative
F = −∇V,
by writing Z t dx 2
0 = δS ≡ δ ds [m( ) − V ]. (3.64)
t0 ds
Both of these results can be derived from Lagrange’s more general formula
Z t Z t ∂L ∂L
δ ds L(q i ˙
(s), q i (s)) = ds [δq i (s) + δq i˙(s) i ]. (3.65)
∂q i ∂ q̇
t0 t0
80  Quantum Mechanics

If the infinitesimal variation of q i (s) vanishes at the endpoints of the integral, we can integrate
by parts to write
Z t Z t ∂L d ∂L
δ ˙
ds L(q i (s), q i (s)) =δ ds δq i (s)[ − ]. (3.66)
t0 t0 ∂q i ds ∂q˙ i

The variational principle δS = 0 says that the correct path is determined by insisting that
this formula vanish for every differentiable function δq i (s), which vanishes at the endpoints.
This leads to Lagrange’s equations

∂L d ∂L
0=[ − ], (3.67)
∂q i dt ∂q˙ i

because we can choose δq i (s) to vanish outside of a tiny interval centered on any point in
the interval.
Defining the canonical momentum pi ≡ ∂∂L q̇ i , these equations take the form

∂L dpi
= . (3.68)
∂q i dt

The definition of pi can be viewed as an equation that determines q̇ i in terms of q i and pj .

This is made more explicit by defining the Legendre transform of L, which replaces it with
a function H(p, q), via X
H(p, q) ≡ q̇ i pi − L. (3.69)

The statement that H depends on q̇ i only through its dependence on pi is what we have
taken as the definition of pi . On the other hand, the statement that L depends on pi only
via its dependence on q̇ i reads
∂H
q̇ i = . (3.70)
∂pi
∂H ∂L
The q dependence of L and H is not Legendre transformed, so ∂q i = − ∂q i , and we have

dpi ∂H
=− i. (3.71)
dt ∂q

These are Hamilton’s form of the equations of classical mechanics. The function L is called
the Lagrangian and its Legendre transform H, the (classical) Hamiltonian. The integral of L
is called the action. For a particle moving in a potential in one dimension, the Lagrangian is
1
L = mẋ2 − V (x), (3.72)
2
 Quantum Mechanics of Single Particle in 1D Space I  81

the canonical momentum is

p = mẋ, (3.73)
and the classical Hamiltonian is
p2
H= + V (x). (3.74)
2m
The Hamiltonian equations of motion have two very interesting structures associated with
them. The first is the Poisson bracket, defined for two functions of p and q by
∂F ∂G ∂G ∂F
[F, G]P B = − . (3.75)
∂q i ∂pi ∂q i ∂pi

For the variables pi and q j themselves, we find that the Poisson brackets with themselves
vanish, while
[q i , pj ]P B = δji . (3.76)
Furthermore, the Poisson bracket satisfies (Exercise 3.11):

[F, G]P B = −[G, F ]P B ,

[AB, C]P B = A[B, C]P B + [A, C]P B ,

[A, [B, C]P B ]P B + [C, [A, B]P B ]P B + [B, [C, A]P B ]P B = 0.
These are identical to the algebraic identities satisfied by commutators in QM, and this
coincidence led Dirac to propose a general method for “quantizing” a classical system written
in Hamiltonian form, by simply replacing the Poisson bracket by the commutator of the
canonical variables, by multiplying the Poisson bracket formula by i~. For nonlinear functions
of the canonical variables, there are ambiguities in how one orders the products of p and q
to form quantum detectables from classical functions.3 This is, in particular, true for the
Hamiltonian function, so the Dirac procedure assigns many different quantum theories to
the same classical system. This is, as it should be. We will see below that classical behavior
arises in certain limits of the rules of QM, and it makes sense that details are lost in taking
a limit.
The second general property of these equations is that they can be formulated as a single
nonlinear partial differential equation. Consider the classical action, for a classical pathRthat
ends at some point q i (t) at time t. It has the form S(q(t), t). From the definition S = tt0 L,
we have
dS ∂S ∂S
= i q̇ i + = L = pi q̇ i − H(pi , q j ). (3.77)
dt ∂q ∂t
3
The ambiguity is ameliorated, but not removed, by insisting that real functions be mapped into Hermitian
operators.
82  Quantum Mechanics

Thus,
∂S ∂S
= −H( i , q j ). (3.78)
∂t ∂q
This is called the Hamilton–Jacobi equation. A solution of the H–J equation is determined
∂S i
by giving the value of pi = ∂q i for all q , at time t0 . This is the same information as the

initial conditions for Hamilton’s equations.

p2
For a one-dimensional system, with Hamiltonian 2m + V (x) the H–J equation reads

S0 2
∂t S = −[ + V (x)]. (3.79)
2m
We solve this by making the ansatz S = −Et + S0 , which reduces it to

S00 2 = 2m(E − V ), (3.80)

whose solution is Z x q
S0 = dy 2m(E − V (y)). (3.81)
x0

In this solution, we view x0 as an initial condition, while x is the position to which the
classical trajectory with energy E has gotten in time t. That time is determined by the
equation Z x
dy
t − t0 = . (3.82)
x0 ẏ

The equation s
∂H p ∂S 2(E − V (y))
ẏ = = = = , (3.83)
∂p m m∂y m
completes the determination of the classical trajectory in terms of the solution of the H–J
equation. By a similar, but somewhat more abstract argument, one shows that a complete
solution of the H–J equation determines all the classical trajectories for a general problem in
analytical mechanics.
The significance of the H–J equation for QM is that it is a partial differential equation in
the variables t and q i , just like the Schrödinger equation. In Chapter 17, we will explore the
JWKB approximation, in which the H–J equation is derived as a first order approximation
to the Schrödinger equation, analogous to the description of light waves by ray tracing.
There is a philosophical comment about classical mechanics that is relevant to some of
the material in Appendix A about interpretations of QM. The H–J equation contains all of
classical mechanics in the sense that once we have found its general solution, a particular
motion of the system is just a solution of the equations

q̇i = ∂i S, (3.84)
 Quantum Mechanics of Single Particle in 1D Space I  83

which goes through some initial point q0 = q(0). So we have a “field on the multidimensional
configuration space, which guides the motion of particles in that space.” Of course, we never
think of classical physics this way, because for any given initial q0 , the values of the field
S(q, t) at points away from the unique trajectory are irrelevant. However, this is a consistent
interpretation of classical physics, because nothing singular happens if we insist that only
one particular trajectory occurs in the real world.
Schrödinger invented his equation by thinking of the H–J equation as the geometrical
optics approximation to a wave equation, much as the eikonal approximation to Maxwell’s
Equations is used to derive the Newton-Fermat picture of light in terms of corpuscles following
fixed trajectories. The equation for the phase of the complex Schrödinger wave function
looks like the H–J equation, with corrections proportional to powers of ~. However, in QM,
the correction terms become singular in regions where the magnitude of the wave function
vanishes, and of course the phase is ill-defined there. Since the situation of a particle traveling
only along a fixed trajectory corresponds to a wave function that vanishes almost everywhere
at fixed time, the QM extension of the H–J equation does not allow for such a situation.

3.4 FURTHER EXERCISES

P 2
3.3 Consider a system with energy E = 2m + V (X). We want to define a quantum theory
in which P and X are compatible variables (commute with each other), and satisfy
Newton’s equations. Show that this is achieved with the Hamiltonian

∂E ∂ ∂E ∂
Hcl = ±i~[ − ], (3.85)
∂p ∂x ∂x ∂p

for one choice of the overall sign. This Hamiltonian is written in the basis where X
and P are both diagonal. Show that if we agree to measure only expectation values of
functions f (X, P ), then these are computed in terms of the square of the wave function

ρ(X, P ) = ψ ∗ (X, P )ψ(X, P ),

which satisfies
∂ρ ∂E ∂E
= ±[ Ẋ + Ṗ ],
∂t ∂X ∂P
where X(t) and P (t) satisfy the usual classical equations of motion. It is your job to
find the right choice of sign in both of these ambiguous equations. You will also find
that the choice of sign in the equation for ρ is opposite to what you would expect for a
function f (X(t), P (t)) and you should also explain why this is the right choice of sign
for a probability distribution. This shows that, as we saw for the ammonia molecule,
there is a classical theory of particle motion, which fits into the framework of QM, and
84  Quantum Mechanics

one must distinguish between the “classical quantum theory” and the real quantum
theory experimentally.

a. In the classical quantum theory of Exercise 3.3, we had to agree not to measure the
∂
operators ΠP,X = ~i ∂X,P . Show that the Schrödinger equation for Hcl is invariant
under the “gauge transformation”

ψ(X, P ) → eiθ(X,P ) ψ(X, P ),

and that the only “gauge invariant operators” in the system are functions f (X, P ).
This observation is the mathematical excuse for a person who wanted to believe
in classical mechanics to disregard the quantum version of classical mechanics. If
we introduce the idea that the phase of the wave function is redundant (this is the
meaning of the phrase gauge invariance), then the quantum formulation of Classical
Mechanics is just adding redundant information. Below we will see that there is a
completely different way to think about gauge invariance in regular QM, which leads
to Maxwell’s electrodynamics. What is the analog of this gauge invariance for a two
state system?

3.4 Compute the commutator

[X, P n ],
in two ways. First, use the representation where X is diagonal and P = ~i ∂x and act
with this commutator on a function ψ(x) and use the definition of the commutator as
the difference between the action of operators in two different orders. You will need to
evaluate ∂xn [xψ(x)] to do this. Now evaluate it by using Leibniz’ rule

[A, BC] = [A, B]C + B[A, C],

repeatedly.

3.5 Find the expectation values of P , P 2 , and P 3 in the states described by the (unnor-
malized) wave functions eik0 x (x2 + a2 )−p . You can use Mathematica, Maple, or Sage to
do the integrals. Remember to divide by the norm of the wave function ( hψ|A|ψi
hψ|ψi is the
formula for the expectation value if you do not normalize your states). How large does
p have to be in order for the wave functions to be normalizable?

3.6 Show that if a wave function is real, then the expectation value of P 2n+1 vanishes for
all nonnegative n. Hint: Use integration by parts to write the expectation value as the
total derivative of a function that vanishes at infinity.
 Quantum Mechanics of Single Particle in 1D Space I  85

3.7 Show that the operator defined in the representation where X is diagonal by
C(ψ(x)) = ψ ∗ (x),
is antilinear
C(aψ + bφ) = a∗ C(ψ) + b∗ C(φ),
and satisfies C 2 = 1. We will see in Chapter 6 that C is an essential part of the time
reversal operation in QM.
3.8 a. Show that the scalar product of C(φ) with C(ψ) is the complex conjugate of the
scalar product of φ with ψ.
b. Show that
hφ|C(|ψi) = hψ|C(|φi),
for any two vectors.
c. Show that the operator AC = CAC, has the same matrix elements between C(ψ)
and C(φ) as A† has between φ and ψ. In order to do this, you will have to show that
AC is an ordinary linear operator, so that you know how to define its Hermitian
conjugate. Hint: Insert 1 = C 2 twice into the expression for the matrix element of
A.
d. Show that for the momentum operator P C = −P and use this to explain the result
of Exercise 3.6.
3.9 Use Fourier series to solve the Schrödinger equation for a free particle moving on a
circle. Assume that the initial wave function has Fourier coefficients f (n).
3.10 Galilean transformations transform momentum and particle position according to P →
P + mv and X → X + vt, where m is the mass and v the velocity of the Galilean boost
transformation. An operator which generates these changes in the sense of
δO = −i[N, O]δv,
for infinitesimal v = δv, is
m P
N= [X − t].
~ m
The commutator of N with the Hamiltonian gives the Heisenberg equations of motion
for the Heisenberg operator N (t), which coincides with the Schrödinger picture operator
N at t = 0. In this case, we must take into account the fact that the Schrödinger picture
operator N depends explicitly on time. The correct Heisenberg equation is then
dN (t) i ∂N
= [H, N ] + , (3.86)
dt ~ ∂t
86  Quantum Mechanics

where the partial derivative symbol means derivative w.r.t. the explicit time dependence
in the Schrödinger picture. Show that the requirement that the equations of motion of
X and P = mẊ take the same form in different Galilean reference frames4 implies that
dN (t)
dt = 0. Show that, assuming translation invariance, this gives the usual form of the
free particle equations and determines the Hamiltonian as a function of P .

3.11 Find the normalization constant N for a Gaussian wave function

(x−x0 )2
ψ(x) = N e− 4∆2 ,

such that Z
dx ψ ∗ (x)ψ(x) = 1.

In solving this problem you should encounter the integral

Z
du u−1/2 e−u ,

which is Euler’s Gamma function Γ(z) evaluated at z = 1/2. Γ(z) satisfies

π
Γ(z)Γ(1 − z) = ,
sin(πz)

which will allow you to evaluate this.

a. Do the Gaussian integral Z

2 +y 2 )
dxdy e−a(x ,

in two different ways. First, it is the product of two one-dimensional Gaussians of

the type you encountered in Exercise 3.11. However, we can also do it by going to
radial coordinates in the x, y plane. Show that the second route leads to an easy
√
derivation of the fact that Γ(1/2) = π.

3.12 Find the expectation value of the operator X n in the normalizable wave function of
Exercise 3.11. The expectation value can be written as a ratio of two integrals. Evaluate
the numerator integral by differentiating the denominator.

3.13 Repeat exercise 3.12 for the operators P n .

4
You will have to think about what this phrase means in QM.
 Quantum Mechanics of Single Particle in 1D Space I  87

3.14 In an n-dimensional Hilbert space, define two operators Un and Vn by the equations

Unn = Vnn = 1,
2πi
Un Vn = Vn Un e n .
2πik
In the basis where Un is diagonal, its eigenvalues are e n . In this basis, Un is called
the clock operator. Show that in this basis, Vn is the matrix that “shifts the clock by
one tick.” Write it explicitly and verify that it satisfies the remaining two equations.
3.15 Show that every n × n matrix M can be written as
n
X
M= akl Unk Vnl .
k,l=1

3.16 In the limit n → ∞, the spectrum of U∞ becomes dense on the circle θ ∈ [0, 2π]. Write
the eigenvalues of U∞ as eiθ and consider the set of functions eikθ , the eigenvalues of
U k . The fact that this is a complete set of complex square integrable functions on the
circle (see the next chapter and Chapter 6 for definitions and more discussion) is the
∂
content of a famous theorem by Fourier. The operator Pθ = ~i ∂θ has an integer valued
spectrum (in units of ~) and the exponentials are its normalizable eigenvalues. Define
Pθ
Vα = eiα ~ and evaluate U k Vα U −k V−α .
3.17 Now write θ = Rx 2π, where x and R have dimensions of length. Find the spectrum of
the operator PRθ and show that in the limit R → ∞ at fixed x, this gives us the Hilbert
space of a single particle on an infinite line.
3.18 Prove that the Poisson bracket satisfies the identities:

[F, G]P B = −[G, F ]P B ,

[AB, C]P B = A[B, C]P B + [A, C]P B ,

[A, [B, C]P B ]P B + [C, [A, B]P B ]P B + [B, [C, A]P B ]P B = 0.

3.19 Solve for the eigenstates of the free particle Hamiltonian on the interval [−a.a], with
the boundary condition ψ(±a) = 0. These are called Dirichlet boundary conditions.
3.20 Repeat the previous exercise, but with the boundary condition ∂x ψ(±a) = 0. These are
called Neumann boundary conditions.
3.21 Solve the free particle on a finite interval with one Neumann and one Dirichlet boundary
condition.
 CHAPTER 4

Quantum Mechanics of a
Single Particle in
One-Dimensional Space II

4.1 INTRODUCTION
This chapter adds more mathematical detail to the description of one-dimensional particle
motion. It also discusses the double slit experiment in some detail. The chapter ends with
an important worked exercise, in which the reader is guided through the solution of both the
bound state and scattering state equations for a particle in a square well or barrier potential.
It is crucially important for every reader to work through this exercise in exhaustive detail.
One should understand why the bound state spectrum is discrete, and what the scattering
matrix is, as well as the relation between bound states and the analytic continuation of
scattering matrix elements.

4.2 A MORE MATHEMATICAL DESCRIPTION OF ONE-DIMENSIONAL MOTION

Now let us add a little more mathematical detail to our description. As we will see in Chap-
ter 6, the proper description of the space of states of a particle moving on an infinite one-
dimensional line is the set of all complex valued functions ψ(x) satisfying the condition
Z
dx |ψ(x)|2 < ∞. (4.1)

This is a vector space, in an obvious way: we can make linear combinations of functions with
complex coefficients. The resulting functions are square integrable, because of the inequality

89
90  Quantum Mechanics
Z Z Z
| dx f ∗ (x)g(x)| ≤ ( |f |2 |g|2 )1/2 . (4.2)

The quantity on the left-hand side

Z
dx f ∗ (x)g(x) ≡ hf |gi, (4.3)

defines the scalar product on this space. The inequality above is then precisely the Cauchy–
Schwarz inequality for this scalar product (see Chapter 6). The definition of distance between
two vectors in a finite dimensional Hilbert space is D2 ≡ |vn − wn |2 , which is the length of
P

the vector |vi − |wi. The generalization of this to the distance between two functions is
Z
D2 (f, g) = dx |f (x) − g(x)|2 . (4.4)

2
This space of functions is infinite dimensional. For example, the functions P (x)e−x , where
P (x) is an arbitrary polynomial, all belong to it. Saying that a finite linear combination of
these vanishes (i.e., that there is a finite basis for the space) is equivalent to saying that
some polynomial vanishes everywhere on the real line. Of course, polynomials of order N
have at most N distinct real zeroes, so this means the polynomial is zero. The monomials
xk for different k are linearly independent (using the same argument) so we have an infinite
number of linearly independent square integrable functions.
On the other hand, there is a discrete or countable basis for this space. Again, the
argument goes through polynomials. Consider a continuous complex valued function f (x) on
an interval [a, b]. Divide up the interval into N , not necessarily equal segments, of size about
b−a
N , and think about the equations

N −1
(N −1) k
X
PN −1 (xm ) ≡ ak xm = f (xm ),
k=0

for xm the middle of each of these segments. This is a set of N linear equations for the N
coefficients of PN . It will generally have a solution as long as the determinant of the matrix
k
Mm ≡ xkm is nonzero. If it is not, we can shift the xm slightly off the midpoints to make
the determinant nonvanishing. Since both the polynomial and f are continuous, they cannot
deviate too much from each other at other points on the interval. In the limit N → ∞, they
cannot differ at all. This is a nonrigorous proof that every continuous function on an interval
can be uniformly approximated by a polynomial.
Now consider a continuous square integrable function g(x) and look at
Z ∞ 2
D2 = |g(x) − PN (x)e−x |2 .
−∞
 Quantum Mechanics of Single Particle in 1D Space II  91

Thinking of the space of square integrable functions as an infinite dimensional Hilbert space,
D2 is the squared length of the vector difference between the vector g(x) and its approx-
2
imation by polynomials times a Gaussian. Note that here we are approximating g(x)ex
uniformly by polynomials.
Now choose a large number Y . The contributions to D2 outside the interval [−Y, Y ]
obviously go to zero as Y → ∞. Inside the interval, we can find a polynomial of order N (Y ),
1
with large enough N (Y ) that the contribution to D2 from the interval is < N (Y ) , because we
2
can uniformly approximate the continuous function g(x)ex by polynomials. It follows that
we can find a polynomial PN , with large enough N that makes D2 < 1/N . As we have said,
D is the infinite dimensional analog of the distance between two points in Euclidean space,
according to the Pythagorean formula. When the distance D goes to zero we say that the
2
sequence of functions PN e−x “converges to g in the L2 norm.” Now not all square integrable
functions are continuous, but they are all limits of piece-wise continuous functions, and you
can convince yourself by drawing pictures that every piece-wise continuous function can be
approached arbitrarily closely (in the sense of D going to zero) by continuous functions. So,
we have shown that polynomials times Gaussians are a countable basis of the space of square
integral functions. There are many other discrete, complete bases. Mathematically, what we
have shown is that the Hilbert space of square integrable functions is a separable Hilbert
space, which is the mathematical name for a space with a countable basis.
If we have two different bases of the same Hilbert space, there is a mathematical isomor-
phism that maps the space into itself, by taking |ek i into |fk i. The best thing about Dirac
notation is that it provides a simple formula for this mapping (and for many other transfor-
mations on Hilbert space). Recall that a dyadic operator determined by two vectors |vi and
|wi is
Dvw |f i = |vihw|f i. (4.5)
In words, we take the scalar product of |wi with |f i and then multiply this number by the
†
vector |vi. We also define the operation of Hermitian conjugation of operators by Dvw = Dwv .
Given this definition, it is easy to write down the transformation that takes one basis into
another:
X
|fk i = U (f, e)|ek i ≡ |fl ihel |ek i. (4.6)
l

In other words, X
U (f, e) = |fl ihel |. (4.7)
l

We just have to use the orthonormality of the |ek i vectors to verify this. The Hermitian
conjugate of the operator U (f, e) is also its inverse, U (f, e)† = U (e, f ), as you will verify
in Exercise 4.1. This is usually written U (e, f )† U (e, f ) = 1, and we have agreed to call
92  Quantum Mechanics

operators satisfying this property unitary operators. We will go into all of this in more detail
in Chapter 6, which will give you a chance to brush up on your linear algebra, while learning
Dirac notation more thoroughly. Eventually, if you want to become a quantum physicist, you
would like to be able to do manipulations of Dirac notation in your sleep.

Exercise 4.1 Consider a unitary matrix Uij . Show that its rows and columns, both form
orthonormal bases. Interpret this fact in terms of the dyadic construction of Exercise 4.2.

Exercise 4.2 Prove that the operators k |ek ihfk |, are unitary, for any choice of the two
P

orthonormal bases |ek i and |fk i. Prove that any unitary transformation1 has this form. (Hint:
Consider the action of U on the elements of an orthonormal basis. In the course of this proof,
you will realize that the dyadic representation of the unitary operator is not unique.)

We now want to introduce a particular unitary transformation, which is extremely useful

for any quantum mechanics (QM) problem involving particles. It is actually a formula you
probably know from your math-physics class, and which we used at the end of the last
chapter, the Fourier transform. This is defined as a mapping from functions to functions:
Z ∞ dp ikx ˜
Z
dp dy ik(x−y)
f (x) = √ e f (p) ≡ F [f˜] = e f (y). (4.8)
−∞ 2π~ 2π~

Here, ~k ≡ p, and from now on all integrals without limits go from −∞ to ∞. Actually, we
have written both the definition of the Fourier transform and Fourier’s theorem, which says
that if you apply the Fourier transform twice, you get back to the function you started with,
but evaluated at a negative value of the argument. In other words, the Fourier operator F
satisfies F 4 = 1 and F 2 = R, where Rf (x) ≡ f (−x).
Physicists like to summarize the Fourier theorem in a pair of equations

dk ik(x−y)
Z
e = δ(x − y). (4.9)
2π
and
dx ix(k−k0 )
Z
e = δ(k − k 0 ). (4.10)
2π
In these equations, k is the wave number, related to the momentum by p = ~k. These serve
to introduce the Dirac delta function, which is defined by the property.
Z
dx δ(y − x)f (x) = f (y) (4.11)

1
The phrases unitary transformation and unitary operator mean the same thing.
 Quantum Mechanics of Single Particle in 1D Space II  93

for any continuous function f (x). We will reserve the discussion of the mathematical prop-
erties of this object, which is not really a function but a distribution, to Appendix B. Its
defining property is that Z
dy f (y)δ(x − y) = f (x). (4.12)

From these equations, it follows that

dx dp dq i(kp −kq )x ˜∗
Z Z Z
∗
dx f (x)g(x) = e f (p)g̃(q) = dpf˜∗ (p)g(p). (4.13)
2π~
This shows that the Fourier transform is a unitary operator. It preserves the scalar product on
the space of square integrable functions. In the previous chapter, we called this Plancherel’s
theorem. We will be using the Fourier transform extensively in this book. Start getting used
to it by doing Exercise 4.3.
2
Exercise 4.3 Compute the Fourier transform of a Gaussian ψ(x) = e−ax +bx , as well as
an exponential function ψ(x) = e−a|x| . Compute the Fourier transform of a step function
S(x) = 0 for x > a or x < b < a, and S(x) = s on the interval [b, a].

4.3 CONTINUUM OR DELTA FUNCTION NORMALIZATION

We have emphasized that we can expand every state in the Hilbert space of square integrable
functions in terms of a countable basis. In equations, this says
Z
dy ψn∗ (y)f (y),
X
f (x) = ψn (x) (4.14)

which we can now write formally as

ψn (x)ψn∗ (y).
X
δ(x − y) = (4.15)

This is called the completeness relation for a complete orthonormal set of functions. It is a
continuous version of the relation
X
δkl = hfk |en ihen |fl i, (4.16)

where |fn i and |en i are any two orthonormal bases.

In a finite dimensional Hilbert space, any normal operator, satisfying [A, A† ] = 0 has a
complete basis of orthonormal eigenfunctions. This is not true in infinite dimensions. If we
take a multiplication operator
Fg ψ(x) ≡ g(x)ψ(x), (4.17)
then this could only have a fixed numerical value if ψ(x) vanishes everywhere except at some
fixed x = x0 . Such a function could not be in the Hilbert space of square integrable functions,
94  Quantum Mechanics

since if its value at x0 is finite, its norm is zero and so it should be identified with the zero
vector. On the other hand, if we take ψ(x) = δ(x − x0 ), then this does satisfy the eigenvalue
equation for Fg , with eigenvalue g(x0 ). The delta function is not a square integrable function
but it does satisfy Z
dy δ(x − y)δ(y − z) = δ(x − z), (4.18)

which means that “eigenfunctions” corresponding to different eigenvalues of the multiplica-

tion operator are orthogonal. This also says that those “eigenfunctions” have infinite norm,
and are not actually in the Hilbert space. We say that such states are delta function normal-
ized, and introduce the notation |xi for the abstract state corresponding to the delta function
located at x. Then (p ≡ ~k),
dp ik(x−y)
Z
hx|yi = δ(x − y) = e . (4.19)
2π~
For any other state |ψi, we can interpret the function ψ(x) that represents this state as
the overlap ψ(x) = hx|ψi, with these delta function normalized eigenstates of the position
operator x. The proper mathematical treatment of these delta function normalized states
is complicated[11]. We will use the physicist’s approach to mathematical rigor. We ignore
it and treat the position eigenstates as if they were discrete, replacing sums by integrals in
equations like the decomposition of the operator Fg into projection operators on a complete
orthonormal basis: Z
Fg = dx g(x)|xihx|. (4.20)

The equation for a momentum eigenstate

~ ∂
hx|pi = phx|pi (4.21)
i ∂x
is solved by
1 px
hx|pi = √ ei ~ , (4.22)
2π~
where we have chosen the normalization so that
dx ix(kp −kq )
Z
< p|q >= δ(p − q) = e . (4.23)
2π~
The factors multiplying x in the exponent are the wave numbers corresponding to the indi-
cated momentum. So, momentum eigenstates also have delta function normalization on the
real line. In Exercise 4.4, you will see how this gets modified on the circle (where we have
seen the eigenvalues are discrete) and in Exercises 3.7–3.9, you saw how all of this mysterious
infinite stuff can be realized in terms of limits of finite matrices.
 Quantum Mechanics of Single Particle in 1D Space II  95

Exercise 4.4 The Hilbert space of functions on a circle consists of periodic square integrable
functions f (θ) = f (θ + 2π) on the interval [0, 2π]. The eigenfunctions of the wave number
∂
operator K = 1i ∂θ are
1
ψk = √ eikθ .
2π
What are the allowed values of k? The fact that these are a complete orthonormal set of
functions is the content of the famous theorem by Fourier:
X fk
f (θ) = √ eikθ .
k 2π
1 2π Z
fk = √ f (θ)e−ikθ .
2π 0
For a circle of radius R, simply make the replacement 2π → 2πR. The momentum operator
is ~K = P . Write an equation for the Dirac delta function on the circle, using these formulae.
Answer to Exercise 4.4: eikθ is periodic, if and only if k is an integer. The delta function is
defined by Z 2π
f (θ) = f (α)δ(θ − α).
0
We use Fourier’s theorem in the form
X 1 Z 2π
f (θ) = eikθ f (α)e−ikα
k
2π 0

to conclude that
X 1
δ(θ − α) = eik(θ−α) .
k
2π
While we are on the subject of “generalized functions” like the Dirac delta function, let us
introduce the Heaviside step function θ(x), which is defined to be zero for x < 0 and 1 for
x > 0. Its real definition (see Appendix B) is in terms of its action on functions. Namely,
Z ∞ Z ∞
θ(x)f (x) = f (x). (4.24)
−∞ 0

Sometimes it is useful to define θ(0) = 1/2. Note that

Z
df ∞ df Z
dx θ(x − a) = = −f (a), (4.25)
dx a dx
where f is any differentiable square integrable function. Square integrability implies that
f (±∞) = 0. Doing a formal integration by parts on the integral from −∞ to ∞ with the
Heaviside function, and recalling the definition of the delta function, we conclude that
dθ(x − a)
= δ(x − a). (4.26)
dx
96  Quantum Mechanics

4.4 TIME EVOLUTION OF A FREE PARTICLE WAVE FUNCTION

We are now ready to use all of this math to study the propagation of free particles through
space in QM. At time t = t0 , the position space wave function hx|ψi is the Fourier transform
of the momentum space wave function, hp|ψi. We also learn that

e−ikx
hp|xi = √ , (4.27)
2π~
R
from the identity hp|ψi = dxhp|xihx|ψi.
We will take p−p0 2
< p|ψ(t0 ) >= N e−( 2∆p ) .
The normalization constant N is determined by insisting that the integral over all momenta
is 1: Z p−p0 2
N −2 = dpe−2( 2∆p ) .
p−p0
Introducing z ≡ ∆p , we write this as
Z Z ∞
−2 − 12 z 2 1 2
N = ∆p dz e = 2∆p dze− 2 z .
0
√ 1
If we define 12 z 2 = u and note that dz = ( 2u)− 2 du, this is
√ −1 Z ∞ 1
N −2 = ∆p 2 2 duu− 2 e−u .
0

This integral is the Euler Gamma function Γ(s), at s = 21 . If you do not know the Euler
Gamma function, look it up on the Web. If you know, or look up, the identity
π
Γ(s)Γ(1 − s) = ,
sin(πs)
√
we see that this integral is just π.
We introduce the Gamma function here because we will run into it in a few other places,
√
but the result Γ(1/2) = π can be proven simply by the following argument.
Z Z Z ∞ Z ∞
− 12 x2 2 − 12 (x2 +y 2 ) − 12 r2
[ dx e ] = dxdy e = 2π dr re =π du e−u = π. (4.28)
0 0

So,
π 1 π 1
N =( 2
)4 = ( 2
)4 .
2(∆p) 2(~∆k)
 Quantum Mechanics of Single Particle in 1D Space II  97

To see the time development of the wave function in position space, we remember that for
the free particle, energy eigenstates are also momentum eigenstates, so we can write

eikx −i p2 (t−t0 ) −( p−p

Z
0 2
hx|ψ(t)i = N dp √ e 2m~ e 2∆p ) . (4.29)
2π~
We convert this into an integral over k, defining ~k0 = p0 and ~∆k = ∆p,

eikx
Z
1 ~k2 k−k0 2
< x|ψ(t) >= ~ N 2 dk √ e−i 2m (t−t0 ) e−( 2∆k ) . (4.30)
2π
This is the same as
1
Z
k 0 )2 1 1 2 2
−( 2∆k
e ~ N 2 dk √ e− 2 α k +ak , (4.31)
2π
where
1 ~∆t
α2 = 2
+i .
2(∆k) m
k0
a= + ix.
2(∆k)2
By completing the square, shifting, and rescaling the integration variable, we find
k0 2 a2 √ 1
< x|ψ(t) >= e−( 2∆k ) e 2α2 α−1 2π~ 2 N. (4.32)

This is a Gaussian in position space. To find its center and width, we need to compute the
a2
real part of 2α 2 , which is

−(x2 − 2x ~k2m )
0

,
2(∆x)2
with
1 4(∆k)4 ~2 (∆t)2
(∆x)2 = (1 + ).
2(∆k)2 m2
The wave packet is initially narrow and concentrated near the classical trajectory, if ∆k is
big, and wide if ∆k is small. It spreads with time, and the spread becomes important more
quickly if ∆k is large, and less quickly if m is large.
This spread can be understood intuitively in terms of the probabilities of momentum
and position and the fact that they cannot both be sharp at the same time. We tried to
localize the wave function in position space at the initial time. However, the more we try to
do so the larger we must take ∆k, the uncertainty in momentum. But the time derivative
of the position is the momentum divided by the mass, so large ∆k means a big uncertainty
in velocity, which of course leads to a larger uncertainty in position at a later time. The
98  Quantum Mechanics

velocity uncertainty is inversely proportional to the mass, for fixed ∆k, which explains the
slower spread of the wave packet as the mass increases. Roughly speaking, the same thing
happens for any localized initial condition. In Exercise 4.5, you will relate the rate of falloff
of a function in position space to the rate of falloff of its Fourier transform in momentum
space. These results are all a consequence of the Heisenberg uncertainty relations, which we
will study in Chapter 6.
Indeed, we can understand the spread of the wave packet that we have found in a much
more general way, by thinking in the Heisenberg picture, where the wave function ψ(x) is
independent of time, but the operator X(t) satisfies
P
X(t) = X + t. (4.33)
m
We can choose ψ(x) such that X has very little uncertainty. Such a tightly focused function
has to have large derivatives, since it must fall rapidly to zero outside of a small interval in
x. The uncertainty in P is the average of the absolute square of dψ/dx, so this will be large.
This means that at large t, the uncertainty in X(t) is large ∼ ∆P t/m, which is equivalent
to the spread of the wave packet, in the Schrödinger picture.
In Exercise 4.11, we will estimate the consequences of these results for a variety of macro-
scopic bodies, from baseballs to the moon, and see that the predicted uncertainties are very
small. This is a part, but not the most important part, of the explanation of how classical
physics emerges as an approximation to the more fundamental quantum rules, which are
obeyed by baseballs as much as they are by electrons.
There is one point that one always has to keep in mind when thinking about quantum
uncertainties for macroscopic objects. Einstein once said that “I cannot believe that the
Moon exists only because a mouse looks at it.” In saying this, Einstein was objecting to an
interpretation of QM in which we think of the quantum state of a system as describing what
really happens to a physical system in any one history. As we will emphasize over and over
again, no such interpretation is correct.2 QM only tells us about probabilities, and the only
way to test a probabilistic theory, even when it says that the probability for something is very
close to one, is by doing the experiment over and over again, and comparing the frequencies of
occurrence to the probabilities predicted by the theory. For systems where we cannot repeat
the experiment, we can only treat the probabilities in a psychological, Bayesian sense. They
tell us, sometimes, what we expect to see in a single run of an experiment, but only if the
probability for a certain history is very close to one. So, the uncertainty predicted in the
motion of the moon just tells us that after a long enough time we cannot predict what will
happen to the moon, given a measurement of its current position and velocity. Einstein never
2
Modern attempts to find a set of “hidden variables,” such that quantum probabilities follow from definite
histories for the hidden variables, give up the possibility of predicting histories for the quantities we
actually measure. We will discuss this in Appendix A on interpretations of QM.
 Quantum Mechanics of Single Particle in 1D Space II  99

objected to this statistical interpretation of QM. However, he believed, because of his deep
understanding of classical mechanics, that all uncertainty had to be explained in terms of
probabilities for definite histories of systems. The statistical interpretation of QM denies the
existence of such probabilities, unless the systems in question are the collective coordinates
of macroscopic objects.
The other thing we have to emphasize about this computation of the spread of the wave
packet for a free particle is that real particles are never exactly free. In particular, whenever
we try to measure the position of a particle with a detector, we are introducing complicated
interactions between the particle and the detector, which change the predictions for the
spread of the wave packet. We will see that when particles move under the influence of
forces, the spread of their wave packets can be dramatically slower than the computation
we have just done would indicate. For example, for simple harmonic motion, certain wave
packets show no spread whatsoever.

4.5 THE DOUBLE SLIT EXPERIMENT

We are now in a position to study the double slit experiment that you were asked to read
about in Feynman’s Lectures, before starting this book. The basic idea is illustrated in
Figure 4.1, but we will use Schrödinger wave function language to describe what is happening.
We have a particle moving in a two-dimensional plane. It is described by some incoming
normalizable wave packet from the left. It encounters a barrier in the vicinity of x = ±a.

Angle (arb. units)

Intensity (arb. units) Intensity (arb. units) Intensity (arb. units)

Figure 4.1 The Double Slit Experiment.

100  Quantum Mechanics

The barrier is a potential V that is infinitely high between x = ±a, except for two slits in y,
of width D, where the potential is zero. The slits are centered at y = ±b. For simplicity, we
are going to consider a Hamiltonian for the particle of the form

p2x p2y
+ + V,
2mx 2my
with mx  my . This means that we can consider a very narrow wave packet in the x direction,
localized in a distance  a, without worrying about rapid spreading in the x direction. An
adequate approximation to the problem is found by simply starting the system off at t = 0
at a time when the wave function is “just emerging from the barrier.” This phrase makes
sense because of the control we have over the width of the wave packet in the x direction.
If we want to be considering states for which the expectation value of the energy is finite,
then the wave function has to vanish identically everywhere that the potential is infinite.
This means that at t = 0, we have

ψ(x, y, 0) = ψ− (x, y, 0) + ψ+ (x, y, 0),

where ψ± vanishes everywhere but in the slit centered at y = ±b. At this time, we can say
for this quantum particle, as we could for a bullet, that the particle either went through the
plus slit or the minus slit. The total probability that the particle is somewhere is the sum
of a probability P+ that it went through the plus slit, and a probability P− that it went
through the minus slit, and these two probabilities add up to one. The probabilities P± are
calculated by integrating |ψ± |2 over the slits where they are nonzero. The total probability
is calculated by integrating the absolute square of ψ over all space, and is normalized to one.
∗
The interference terms ψ± ψ∓ vanish identically.
For bullets, we use “logic” to conclude that for predictions of what happens for t > 0
we continue to add together the probabilities we would obtain for the conditional events
that the bullet went through one slit or another. For a classical probability theory, we can
do this because the equations that determine the evolution of probability with time are
linear in the probability density. Another way of saying this is that we are using Bayes’ Law.
Linearity of the equations for the probability density in a classical theory means that there
are probabilities for histories of the variables. We can divide all histories up into those which
pass through one slit and those which pass through another, at t = 0. The total probability is
the sum of the probabilities of one or the other of these classes of histories, and the linearity
of the equation for probability density tells us that this additivity remains true for all times.
This is not true in QM. We can define “conditional wave functions” ψ± , and follow their
evolution separately, because the Schrödinger equation is linear. It is even true that
Z
dxdy (|ψ+ |2 + |ψ− |2 ) = 1,
 Quantum Mechanics of Single Particle in 1D Space II  101

at all times, if it is chosen so initially. Nonetheless, given an initial wave function ψ = ψ+ +ψ− ,
when we ask what the probability is, at some t > 0 to find the particle at some point (x, y),
∗ ∗
it is not the sum of P+ (x, y, t) + P− (x, y, t). The interference term ψ+ ψ− + ψ− ψ+ , integrates
to zero, but is nonzero point by point because of spreading of the wave packet.
This would follow from calculations we have already done, if the wave functions were
Gaussians, but these wave functions actually vanish outside of a finite interval. In fact, one
can show that this makes the spreading even worse. You will do that in Exercise 4.5, but let
us just try an example. Suppose we have a wave function
y 2c
ψc = e−( b ) .
For very large c, this approaches a function that vanishes for |y| > b. Its Fourier transform
is proportional to Z
y 2c
dy eiky e−( b ) .
The result is an even function of k, so we take k positive and large. For large k we can do the
integral by the method of steepest descents: we find the stationary point of the exponent.
This is
ikb2c 2c−1
1
y∗ = [ ] ].
2c
2c+1
At the stationary point, the real part of the exponent behaves like −dk 2c−1 , where d is a
positive constant of order 1 in the large c limit. Thus, the larger the value of c, the more
slowly the momentum space wave function falls off at large momentum. The uncertainty in
the momentum grows larger, and at large times, this leads to an uncertainty in position worse
than that of the Gaussian.
Many readers find that the most mystifying part of Feynman’s discussion of the double
slit experiment is the disappearance of the interference pattern when one tries to determine
experimentally the slit through which the particle goes. There are two reasons for this mys-
tery. The first is that, despite Feynman’s care in saying that he is talking about probability,
and that in any given run of the experiment the detector goes “clunk” at some particular
point on the screen, his pictures give the impression that the Schrödinger wave function is a
physical wave. In fact, the picturesque wave pattern is something we put together to sum-
marize the results of many experiments. It is not a physical wave in space, measured at a
particular time. If it were, it would indeed be mysterious that the value of this physical thing
near the plus slit could affect the detector near the minus slit. This is a lot less mysterious
when we say, correctly, that what we are doing is measuring the frequency of occurrence of
hits on different parts of the screen in two different situations, one with a detector sitting at
the minus slit and one without it.
Conditional probabilities often behave in a “nonlocal” manner. If the equations for the
weather predict on a certain day that a particular storm has a probability to hit one of two
102  Quantum Mechanics

cities a few days later, then once a few days have passed we know that one of those two
predictions was wrong. If we thought of the probability distribution as a physical thing, it
would seem insane to suddenly change it in Galveston, because we see a storm hitting New
Orleans. Bayes’ rule of conditional probability tells us this is exactly the right thing to do.
Bayes’ rule is not in general applicable in QM, because probabilities only add when inter-
ference terms are negligible. If we want to understand why the interference terms suddenly
become negligible when we put a detector near one of the slits, then we have to understand
more about the properties of the detector than Feynman’s discussion supplies. We will see in
Chapter 10 that the crucial property is that the detector is a complicated system, made of
many atoms, with a huge number of microscopic atomic states corresponding to each posi-
tion of the “needle on the detector’s dial.” As a consequence, the interference terms in the
probability distribution for seeing the detector’s dial either register a hit or not, are unimag-
inably small, and Bayes’ rule is approximately valid. This is a general lesson: the quantum
rules for the probabilities that refer to macroscopic objects differ from the rules of a classical
statistical theory, which satisfies Bayes’ rule, by amounts which are of order e−cN where N
is the number of atoms out of which the object is composed. These differences are so small
that they are impossible to measure, even in principle.
The fact that the rule for computing probabilities of future events depends on whether the
detector is there or not is not mysterious. Indeed, we are describing very different experiments
in the two cases. In the first, we ask for a prediction of the probability of getting a hit on the
screen. In the second, we are asking for probabilities of hits, conditioned on the detector’s
behaving a certain way at an intermediate time. As we will see, the fact that the violation of
Bayes’ rule for observations done on the macroscopic detector is vanishingly small, completely
accounts for the difference in the predictions.

4.6 A WORKED EXERCISE

Exercise 4.6 As we will see in the next chapter, the generalization of the Schrödinger equa-
tion to particles moving under the influence of a force is simply to add the potential energy
V (x) to the Hamiltonian operator. In the following exercises, we will consider potentials,
which are locally constant, so that we can use free particle solutions to solve the equation. To
begin, solve the Schrödinger equation for a step function potential V (x) = V0 θ(x+a)θ(a−x).
The real line divides into three intervals, in each of which the equation reduces to that of
a free particle. By taking the integral of the Schrödinger equation in small intervals around
the points of discontinuity, you can figure out how to join the solutions in the three regions
and impose the condition of normalizability. For one sign of V0 , you should find only delta
function normalizable eigenstates, whereas for the other, you will find both delta function
normalizable and truly normalizable eigenfunctions. Show that the spectrum of the latter
is discrete and describe how it depends on V0 and a. The normalizable wave functions are
 Quantum Mechanics of Single Particle in 1D Space II  103

called bound
√ states. Show that the bound state spectrum depends only on the parameter
a
s0 ≡ ~ 2mV0 . By examining the equation for bound states graphically, show that the num-
ber of such states is finite, but goes to infinity as s0 → ∞. For small s0 , show that there is
only a single bound state. Now find the eigenfunctions for E > 0. There are two linearly inde-
pendent solutions, which can be characterized by saying that the associated time-dependent
solution moves toward or away from the center, near +∞. Alternatively, we can characterize
the solutions as incoming or outgoing at −∞. There must be two linear relations between
these four solutions, since only two of them can be linearly independent. Compute the 2 × 2
Scattering matrix or S-matrix, which expresses the two outgoing solutions in terms of the
two incoming solutions.

Exercise 4.7 Consider the square well potential of Exercise 4.6. For the case of the nor-
malizable wave functions, show that they decrease exponentially for large |x|. Think about
classical mechanics with the same potential, and show that motion in the region where the
quantum wave function is exponentially decreasing is simply not allowed. The fact that there
is nonzero probability for the particle to be found in this region is called quantum tunneling
through a barrier. We will study it more extensively in Chapter 17 on the JWKB approxi-
mation. Show that the equation for bound states depends only on the parameter s0 .

Exercise 4.8 For the continuum (i.e., scattering) wave functions in Exercise 4.6, we have
two possible behaviors e±ikx for positive k in each of the asymptotic regions x → ±∞.
By considering the time-dependent Schrödinger equation show that the choices of the sign
of k correspond to left and right moving waves in the two regions. Argue that the two
different signs in a given asymptotic region, correspond to two linearly independent solutions
of the second order ordinary differential equation for fixed k 2 . Argue that there are exactly
two linearly independent solutions. Thus, there must be linear relations between the four
possible asymptotic behaviors. Using your exact solutions, find these linear relations. Your
tool for finding these relations is continuity of the wave function and its derivative at the
points where the potential, and thus the second derivative of ψ is discontinuous.

Answer to Exercises 4.6–4.8: The solution to the Schrödinger equation with energy E is
p 2m
ψE = e± ~2
(V (x)−E)x
, (4.34)

where V (x) = V0 θ(a−x)θ(x+a). Integrating the Schrödinger equation near the discontinuities
in V (x), we find that the wave function and its first derivative must be continuous there
Figure 4.2. If E > 0, we have delta function normalizable solutions, while if E < 0, we can
get normalizable bound state solutions by picking only the falling exponential as x → ±∞.
The problem is invariant under reflections, so for the bound states, we can choose to look
at even and odd solutions. We can impose the continuity conditions only at x = a and they
104  Quantum Mechanics

V0

−a a

−V0

Figure 4.2 The potential for Exercises 4.6–4.8.

will automatically be satisfied for negative x. In the region near the origin, the wave function
depends on the crucial quantity
p
2m(V0 − E)
ik0 = , (4.35)
~
which gives k0 imaginary. The imaginary part of ak0 is equal to the classical action in units
of ~ of a particle traveling from −a to a, if E − V0 > 0. When the opposite inequality holds,
k0 is imaginary. In either case, we define k0 with the positive (imaginary) square root. We
define
√
−2mE
ik = . (4.36)
~
In addition, define r ≡ kk0 . In the region |x| > a, the wave function grows or falls exponentially
if E < 0. We must choose the falling solution at both positive and negative values of x,
to obtain a normalizable wave function. The continuity conditions for even and odd wave
functions at x = a are:

A cosh(|k0 |a) = Be−|k|a , Ar sinh(|k0 |a) = −Be−|k|a , (4.37)

A sinh(|k0 |a) = Be−|k|a , Ar cosh(|k0 |a) = −Be−|k|a , (4.38)

when k0 is imaginary. The hyperbolic sine and cosine are both positive, and these equations
have no solution. Thus, we must have the intuitively obvious condition E − V0 > 0, which
implies a potential well rather than a barrier and asymptotic energy less than the depth of
the well, in order to have a bound state. For k0 real, the matching conditions become

A cos(|k0 |a) = Be−|k|a , Ar sin(|k0 |a) = Be−|k|a , (4.39)

A sin(|k0 |a) = Be−|k|a , Ar cos(|k0 |a) = −Be−|k|a . (4.40)

 Quantum Mechanics of Single Particle in 1D Space II  105

Motivated by the bound on E, we define E = yV0 , with 0 ≤ y ≤ 1. The conditions become

y
r p
= − tan(s0 1 − y). (4.41)
1−y
y
r p
= − cot(s0 1 − y). (4.42)
1−y
q
These depend only on a single parameter s0 ≡ a −2mV ~2 . When s0 is small, we can expand
0

the trigonometric functions. The equation for odd solutions becomes y = 1/s20 , which is
inconsistent with 0 ≤ y ≤ 1. The even solution is y = s20 and is consistent. There is a single
√
bound state. When s0 is large, we can satisfy the equation with s0 1 − y ∼ nπ 2 , where n
nπ
is odd for even solutions and vice versa, as long as 2s0 < 1. The point is that the tangent
and cotangent take on any large value in the vicinity of their poles, so that as long as y is
sufficiently close to 1 we can match. These explicit formulae are only valid for large n, but
there are bound states for every value of n satisfying the inequality. As s0 goes to infinity, we
get an infinite number of states and the explicit formula becomes more exact. In this limit,
after adding a constant to make all the energies positive, the spectrum approaches that of
the infinite square well. Note, however, that s0 can be large for a shallow, but very broad,
well also. In that situation, we would still have continuum eigenstates.
Turning now to E > 0, we define
ikx −ikx
ψ + = A+
out e + A+
in e , x > a, (4.43)

ψ + = A−
out e
−ikx
+ A−
in e
ikx
, x < −a. (4.44)
2 k2
t −i ~2m
The subscripts in and out refer to the fact that, when multiplied by e , the relevant
part of the solution becomes a traveling incoming or outgoing wave. Note that the sign of the
exponential changes when we compare an incoming (outgoing) wave on the left and right.
To write the continuity conditions at x = ±a compactly, it is convenient to define z0 = eik0 a ,
z = eika , and r± = 1 ± r. Then we have

A−
in z
−1
+ A− −1
out z = Az0 + Bz0 , (4.45)

A−
in z
−1
− A− −1
out z = r(Az0 − Bz0 ), (4.46)
A+
out z + A+
in z
−1
= Az0 + Bz0−1 , (4.47)
A+ + −1
out z − Ain z = r(Az0−1 − Bz0 ). (4.48)
These are equivalent to a pair of 2 × 2 matrix equations expressing the vectors
A−
!
in/out
Ain/out ≡
A+
in/out
106  Quantum Mechanics

in terms of the vector !

A
Am ≡
B
z r+
Ain = [ + r− z0 σ1 ]Am , (4.49)
2 z0
1 r−
Aout = [ + r+ z0 σ1 ]Am . (4.50)
2z z0
The Scattering matrix or S-matrix is the transformation that takes Ain into Aout , so we have
1 r− 2 r+
S= [ + r+ z0 σ1 ] [ + r− z0 σ1 ]−1 , (4.51)
2z z0 z z0
These equations reflect the fact that there are only two linearly independent solutions of the
Schrödinger equation. The in and out states represent two independent bases for the Hilbert
space. These equations tell us how to transform between them.
Note that the two factors in the formula for the S-matrix commute with each other.
−1 † †
Thus, since S = S+ S− , we have S † S = S+ S+ [S− S− ]−1 . Now compute
† 1 r− 1 2
S+ S+ = [r− z0 + r+ z0−1 σ1 ][ + r+ z0 σ1 ] = [r+ 2
+ r− + r+ r− (z02 + z0−2 )], (4.52)
4 z0 4
† 1 r+ r+ 1 2
S− S− = [ + r− z0 σ1 ][ + r− z0−1 σ1 ] = [r+ 2
+ r− + r+ r− (z02 + z0−2 )]. (4.53)
4 z0 z0 4
Therefore, S † S = 1. The S-matrix is a unitary transformation between two bases of the
Hilbert space. Of course, the states of fixed k are not normalizable states in the Hilbert space,
but this calculation guarantees that the mapping of normalizable incoming wave packet states
into normalizable outgoing wave packet states, defined as superpositions of the fixed k states,
is a genuine unitary operator in the Hilbert space.
We can write the eigenvalues of the unitary S-matrix as e2iδ± (k) where δ± are called the
phase shifts.
(r− z0−1 ± r+ z0 )
e2iδ± = z 2 . (4.54)
(r− z0−1 ± r− z0 )
They are analytic functions of k, which will have poles whenever the denominator has a linear
zero. The equation for the poles is
r+
= −z02 . (4.55)
r−
For real k, there are no solutions of this equation because the left-hand side is a real number
greater than 1 and the right-hand side is a phase. For imaginary k, the left-hand side is also
a phase. If k = il the equation for a pole is
r− = ±z02 r+ . (4.56)
 Quantum Mechanics of Single Particle in 1D Space II  107

Some simple algebra shows that this is identical to the equation for bound states, and has
solutions only when V0 < 0. The fact that the S-matrix is an analytic function of energy,
with poles at the bound state energies, is a very general result of immense importance.

4.7 FURTHER EXERCISES

4.4 Consider the Fourier transform of a smooth function f (x) (see Appendix B for a defi-
nition of smooth). Z
˜
f (k) = dx f (x)eikx ,

for large values of k. Show that smoothness implies that the Fourier transform falls off
faster than any power of k. Compute the Fourier transform of (x2 + a2 )−b for those
values of b for which the function is square integrable. How does the falloff in wave
number space depend on b?
4.5 Compute the Fourier transform of a step function, θ(a − x)θ(a + x), where θ(x) is 1
for x > 0 and 0 for negative x, and demonstrate the relation between the width of the
interval over which the function is nonzero and the falloff in wave number space. Use
this to calculate the time evolution of a wave function that is initially confined in a
fixed interval.
4.9 Consider the plane wave normalizable solutions for the potential of Exercises 4.6–4.8, in
the case where there is no left moving or incoming wave in the region of large positive x.
Thus, ψ = A+out e
ikx
in this region. For x < −a we have
ψ = A−
in e
ikx
+ A−
out e
−ikx
.
A−in is called the incident amplitude. It is a wave moving toward the barrier from the
left. A− +
out is called the reflected amplitude. Aout is the transmitted amplitude. The
transmission coefficient,
A+ 2
T = | out | ,
A−in
gives us the probability of transmission through the barrier. Calculate it, for all values
of V0 , E, and a. Also define and calculate the reflection coefficient R and show that
R + T = 1. Show that T oscillates as a function of E and that there are values of E for
which it is equal to 1. That is, there are discrete energies for which the barrier becomes
completely transparent! This is called the Ramsauer–Townsend effect, and has been
observed in the laboratory.
4.10 Consider the potential V = −a2 θ(x). Consider a particle of mass m and energy E
traveling to the right from large negative x. Assume that for x > 0 there is no left
108  Quantum Mechanics

moving wave. Show that there is a reflected wave and calculate the reflection coefficient.
A classical particle would not of course experience such a reflection. Is it possible that
R = 1, i.e., for some values of energy, a quantum particle “cannot fall off a cliff” ?

4.11 Consider an initial state ψ(x, 0) which is normalizable and even so that hXi = 0 and
hX 2 i = (∆X)2 is finite. Using the Heisenberg picture, find the minimal value of the
uncertainty in the position of the particle after time t. For a freely moving baseball
of mass 0.5 kg and an initial uncertainty ∆X = .001 m, how long will it be before
the uncertainty is of order a meter? Assume the initial velocity is 100 km/h. Treat the
baseball as a point particle. Do the same calculation for the moon, whose mass is about
7.35 × 1022 kg and whose orbital velocity is about 3, 683 km/h. Note that for motion
in a Newtonian potential, the spreading of the wave packet is even slower[12].

4.12 Show that Gaussian wave functions saturate the Heisenberg uncertainty bound on
∆X∆P . Prove that these are the only wave functions that do this.
∞
xn r(x) = mn is finite for every n ≥ 0. This
R
4.13 Let r(x) ≥ 0 be a function such that −∞
function defines a positive definite scalar product on the space of polynomials P (x). By
the Gram–Schmidt process, we can find an orthonormal basis of polynomials
Z
r(x)Pn (x)Pm (x) = δmn .

Show that  
m0 m1 m2 ... mn
 m m2 m3 ... mn+1 
 1 
Pn (x) = cn det  ... .
 
 
mn−1 mn mn+1 . . . m2n−1 
1 x x2 ... xn
Find the value of cn .

4.14 Find conditions on the weight function r and the function g such that the Rodrigues
formula
1 dn
Pn (x) = [rg n ],
r dxn
defines a set of orthogonal polynomials with respect to r(x).

4.15 Consider a differential operator of the form

d2 d
H = a(x) + b(x) + c(x).
dx2 dx
 Quantum Mechanics of Single Particle in 1D Space II  109

Find the condition that the operator H is Hermitian in the scalar product space defined
in the previous exercise. Assume that r(x) is differentiable. This should give you two
conditions on the functions a, b, c, for a fixed choice of r.

4.16 Consider a non-Hermitian operator a, which satisfies [a, a† ]+ ≡ aa† + a† a = 1 as well as

a2 = 0 and define Q = (P − iW (X))a, where X, P are the coordinate and momentum
of a one-dimensional particle. Show that

[Q, Q† ]+ = 2H = P 2 + W 2 + W 0 (aa† − a† a).

Show that the operator in parentheses has eigenvalues ±1, and that a can be represented
by a two-dimensional matrix. The full Hilbert space consists of two component wave
functions Ψ = ψi (x). Show that if a state exists, satisfying

QΨ = 0,

then it is the ground state of the system. Find the explicit solution to this equation,
and the criterion that this candidate ground state wave function is normalizable. Find
the lowest eigenvalue. This system is called supersymmetric quantum mechanics [13].
d
4.17 Show that dx θ(x) = δ(x) by integrating the derivative against a smooth integrable
function and using integration by parts.

4.18 Show that

∞ eixs
Z
ds = θ(x). (4.57)
0 2πi(s + i)
Here  is a small positive number, which is taken to zero after doing the integral. Use
Cauchy’s theorem from complex analysis.

4.19 Consider the infinite square well, Exercise 4.6 with V0 → −∞. Shift the energy by
an infinite amount so that the energy inside the well is defined to be zero, and that
outside to be positive infinity. Show that there are only bound state solutions, and that
they correspond to solving the free Schrödinger equation with boundary conditions at
x = ±a.

4.20 Show, for every solution of the time-dependent Schrödinger equation of the infinite
square well, that there is a time T such that ψ(x, 0) = ψ(x, T ). You can use the
expansion of a general solution in energy eigenstates. Now consider the classical motion
in this potential. A particle will return to the same state at some time Tcl . Show that
Tcl depends on the energy. See [14] for an explanation for this discrepancy between
quantum and classical recurrences.
110  Quantum Mechanics

4.21 Solve for the bound states in the potential

~2 k12
V (x) = − [θ(x + a + b)θ(a − b − x) + θ(a + b − x)θ(x − b + a)],
2m
where 0 < a < b. This has the form of a double square well, symmetric around the
origin. We can view it as a very primitive model of an electron attracted to two different
nuclei, considered as infinitely heavy.3 Set up the equations that solve for the ground
state energy, and argue that the ground state energy is lowered as b is made smaller with
a fixed (corresponding to moving the nuclei, without altering the potential each nucleus
exerts on the electron. The argument here should go back to the original Hamiltonian).
If we now restore the finite mass of the nuclei, the variation of the ground state energy
will correspond to a force on the nuclei. Is it attractive or repulsive? We will do more
realistic problems of this type when we study the Born–Oppenheimer approximation
in Chapter 11.
4.22 In the previous problem consider b  1, with a fixed and of order 1. You should be able
to solve the bound state equations by hand in this limit. There is another, very inter-
esting, way to approach the solution. Argue that a solution, where the wave function is
approximately the bound state wave function for a single well, is a good approximate
solution of the one well problem in this limit. There are two such solutions, which are
degenerate in energy, because of the reflection symmetry of the potential. Consider
arbitrary linear combinations of these solutions, and determine which linear combina-
tion minimizes the expectation value of the Hamiltonian. What is the orthogonal linear
combination? How do these two linear combinations transform under the reflection
symmetry? This discussion should remind you of the ammonia molecule. Explain why.
4.23 Consider a general potential V (x) which has the property that V (x) → 0 rapidly as
x → ±∞. We will discuss the question of how rapid the falloff has to be when we talk
about scattering in a Coulomb potential. At ±∞ we can use solutions
ψ± = A± eikx + B± e−ikx .
Since the Schrödinger equation only has two linearly independent solutions, there have
~2 k2 t
to be two linear relations. The time dependence of all of these solutions is e−i 2m , so
the coefficients B+ and A− represent incoming waves, while the other two coefficients
are outgoing waves. We write the two linear relations as expressions for the outgoing
waves in terms of the incoming waves.
! ! !
B− S11 S12 A−
= .
A+ S21 S22 B+
3
The lightest nucleus weighs 2, 000 times as much as an electron.
 Quantum Mechanics of Single Particle in 1D Space II  111

The 2×2 S-matrix or scattering matrix relates incoming waves from either side to outgo-
ing waves on either side. Write formulae relating reflection and transmission coefficients
to S-matrix elements. Argue that the S-matrix is unitary S † S = 1, as a consequence
of conservation of probability.

4.24 Consider the Hamiltonian

P2 c
H= + 2,
2 X
where we have used units such that ~2 = m, which you can call unnatural units. Show
that the operator XP + P X ≡ D satisfies

[H, D] = −4iH.

Use this fact to solve the Heisenberg equation of motion for D. You will want to recall
that the operator H is time independent. Also show that if H has an eigenstate ψ1 (x)
with eigenvalue E1 , which is normalizable, or delta function normalizable, then it has
another one for every other real value of E with the same sign as E1 . What is the
corresponding eigenstate?

4.25 Argue that for E = 0, the eigenfunctions H are also eigenfunctions of the rescaling
operator D, and are therefore power laws. Show that they are not even delta function
normalizable, so E = 0 is not in the spectrum of H.

4.26 Argue that for c > 0 the expectation value of H in any normalizable state ψ(x) is
positive, and that this means that all eigenvalues are positive.

4.27 Solve the Schrödinger equation for the Hamiltonian in Exercise 4.24 exactly in terms
of Bessel functions. Use Mathematica, Maple, SAGE, or any handbook of functions
to determine the behavior of the solutions at the origin and at infinity, as well as the
conditions for normalizability. Show that for c > 0, there are a continuum of delta
function normalizable solutions, but no normalizable ones.
 CHAPTER 5

The Harmonic Oscillator

5.1 INTRODUCTION
This is probably the most important chapter in this book. It introduces the harmonic oscil-
lator, the single most important soluble problem in physics. It also contains a very brief
introduction to the theory of quantized fields, since quantized fields satisfying linear field
equations are nothing but collections of quantum harmonic oscillators. The real importance
of quantized fields is that they are the correct description of multiparticle states of all the
particles we have observed in the world, and explain why those states are either totally sym-
metric (Bose–Einstein particle statistics) or totally antisymmetric (Fermi–Dirac statistics)
under interchange of particle labels. The theory of quantized fields also explains the correct
meaning of the phrase “wave–particle” duality, which pervades much of the old quantum
mechanics (QM) literature and many current textbooks. Bosonic quantized fields have two
different kinds of states: their energy eigenstates are interpreted as collections of noninter-
acting particles, while large amplitude coherent states behave like classical waves. We will
explain the concept of coherent state first in the context of a single harmonic oscillator.
Coherent states turn out to be the most efficient tool for solving the dynamical equations of
an oscillator.

5.2 QUANTIZING THE SIMPLE HARMONIC OSCILLATOR

Simple harmonic motion is the most useful textbook problem in the history of physics text-
books. Its utility comes from the fact that almost every problem in physics has a lowest
energy state or ground state, and small deviations in energy from the ground state can be
decomposed into simple harmonic motions, with some spectrum of frequencies. The harmonic
oscillator is not just an artificial exactly soluble problem, but forms the basis for investigation
of a very wide range of systems.

113
114  Quantum Mechanics

Force is defined in classical mechanics as the rate of change of momentum of a particle.

In QM,
i
Ṗ = [H, P ], (5.1)
~
so in order to have a force, we must add a term to the Hamiltonian that does not commute
with P . The simplest possibility is to add a function of X, which we call the potential and
denote by V (X). V has dimensions of energy. It leads to the Heisenberg equation of motion
i
Ṗ = [V (X), P ]. (5.2)
~
We can understand how to compute this by taking V = X n , and observing that, for any
three operators,

[AB, C] = ABC − CAB = (AC − CA)B + A(BC − CB) = [A, C]B + A[B, C]. (5.3)

This is called Leibniz’ rule for commutators. You might note the similarity to the action of
a derivative on a product, but you have to be careful of the ordering of the terms. Applying
Leibniz’ rule with A = X n−1 , B = X, and C = P , we get

[X n , P ] = [X n−1 , P ]X + X n−1 i~ = in~X n−1 , (5.4)

where the last step follows by induction. It follows that for any potential that has a power
series expansion around some point x0 ,
dV
[V (X), P ] = i~ , (5.5)
dX
so that
dV
Ṗ = − . (5.6)
dX
This is the quantum mechanical version of Newton’s second law.
If V has a minimum at some point (which without loss of generality we can take to be
x = 0), then near that point we can expand V ∼ V0 + 12 kX 2 . The harmonic oscillator problem
approximates a more general problem by just taking V to be given exactly by this formula.
Also, we usually set V0 = 0 since this just shifts the zero of energy, which is not observable.
The Heisenberg equations of the harmonic oscillator are
P
Ẋ = , (5.7)
m
Ṗ = −kX, (5.8)
 The Harmonic Oscillator  115

from which we conclude that

P ci mk
ciẊ + Ṗ = ci − kX = (P + i X), (5.9)
m m c
√
for any c. If we choose c = km, then the time derivative of this complex linear combination
is proportional to itself, so that
√ √
P (t) ± i kmX(t) = e±iωt (P (0) ± i kmX(0)), (5.10)

where s
k
ω= . (5.11)
m
Notice that since we never had to multiply operators together, these manipulations are equally
valid in classical and quantum mechanics.
We define the creation and annihilation1 operators by
1
a† ≡ √ (mωX − iP ), (5.12)
2~mω
1
a≡ √ (mωX + iP ). (5.13)
2~mω
The operators X and P are Hermitian, because they have real eigenvalues, so a and a† are
Hermitian conjugates of each other. Their commutation relations with each other are really
interesting. We have
1
a† a = (mωX − iP )(mωX + iP ), (5.14)
2~mω
1
= (m2 ω 2 X 2 + P 2 − imω[P, X]). (5.15)
2~mω
Also,
1
aa† = (mωX + iP )(mωX − iP ), (5.16)
2~mω
1
= (m2 ω 2 X 2 + P 2 + imω[P, X]). (5.17)
2~mω
If we use [X, P ] = −[P, X] = i~, then these two equations can be written

1 P2 1
H= ( + kX 2 ) = ~ω(a† a + ), (5.18)
2 m 2
1
These are also called creation and destruction, raising and lowering, and ladder operators.
116  Quantum Mechanics

and
[a, a† ] = 1. (5.19)
Using Leibniz’ rule again, we can derive from these that

[H, a] = −~ωa, (5.20)

and
[H, a† ] = ~ωa† , (5.21)
which allow us to solve Heisenberg’s equations as

a(t) = e−iωt a(0). (5.22)

We should have expected this because the creation and annihilation operators are just mul-
tiples of the complex linear combinations of X and P that we discussed above. Finally, note
that the commutator between a and a† is the same as that between the multiplication oper-
∂
ator w by a complex variable, and the partial derivative ∂w , with respect to that variable.
The commutators of the Hamiltonian with the creation and annihilation operators, when
they are applied to an eigenstate of the Hamiltonian, give

[H, a]|Ei = −~ωa|Ei. (5.23)

This can be rewritten

Ha|Ei = (E − ~ω)a|Ei, (5.24)
which says that a|Ei is proportional to an eigenstate of H with eigenvalue lowered by ~ω.
We say proportional to, rather than equal, because eigenstates are defined to have norm 1.
A similar calculation shows that a† |Ei is proportional to an eigenstate of the Hamiltonian
with energy E + ~ω.
The norm of a|Ei is

1 1 (E − 12 )
||a|Ei||2 = hE|a† a|Ei = hE| (H − )|Ei = . (5.25)
~ω 2 ~ω
This fact is responsible for the alternate name of raising and lowering operators for a† and a.
In a couple of paragraphs, we will see that the allowed values of E are En = (n + 21 )~ω, where
n is a nonnegative integer. Let us denote the eigenstate with eigenvalue En by |ni. Then we
can write
√
a|ni = n|n − 1i, (5.26)
√
a† |ni = n + 1|n + 1i, (5.27)
 The Harmonic Oscillator  117

and by induction,
(a† )n
|ni = √ |0i. (5.28)
n!
On the other hand, the Hamiltonian is a sum of squares of Hermitian operators, which means
(Exercise 5.1) that all of its eigenvalues are positive. We cannot keep lowering the energy.
That is, there must be a lowest energy, or ground state, with positive energy, such that

a|0i = 0. (5.29)

The formula for the Hamiltonian in terms of creation and annihilation operators tells us
that this lowest energy state has energy 21 ~ω. We can obtain other states, with energies
(n + 12 )~ω, with n a positive integer, by acting n times on the ground state with a† . Since
X is proportional to a + a† , the states that we generate by acting on the ground state with
powers of a† generate all products of a polynomial times the ground state wave function. In
a moment, we will show that the ground state wave function is a Gaussian, so that all of
the states |ni are normalizable, because of the sort of calculation that we did in the previous
chapter. Furthermore, those calculations show that they form a complete basis for the Hilbert
space. This means that we have found the entire spectrum of the Hamiltonian, from these
simple algebraic considerations. The energy spectrum is ~ω(n + 21 ), where n is a nonnegative
integer.
In fact, we can do a lot more than that, by introducing what are known as coherent
states. It is worth doing this because these states have many uses, and illuminate the relation
between classical waves and quantum particles, which we will call wave–particle duality.2
Coherent states are defined by the formula:
†
|zi = eza |0i, (5.30)
3
where |0i is the ground state of the oscillator, satisfying

a|0i = 0.
2
In much of the literature on QM, wave–particle duality is used for a different notion: the supposed duality
between the Schrödinger wave function and the particles for whose properties that wave function computes
probabilities. This use of the phrase is misleading. In the rest of physics and mathematics, the word duality
refers to two equivalent descriptions of the same system, in terms of different variables. The Schrödinger
wave function is not another description of particles. Particles are entities whose properties we measure
in single experiments in the real world. The wave function is a machine for computing probabilities that
those measurements will give certain results. It cannot be determined by any single experiment, but only
by tabulating the frequencies of repeated trials. Our use of the phrase wave–particle duality refers to an
actual equivalence between a particle and a wave description of the space of states of quantum fields. The
intent is to supplant the older, misleading, usage.
3
The 0 on the right-hand side of this equation means the zero vector, while |0i is a nonzero state whose
corresponding wave function is a normalized Gaussian.
118  Quantum Mechanics

Thus, the ground state is the coherent state with z = 0. We will see in a moment that the
vectors |zi all have finite norm, but that norm is not equal to one. Although we should really
reserve the term state for vectors of norm 1, we will follow convention and refer to |zi as
a coherent state, and the unit vector √|zi , as a normalized coherent state. The coherent
hz|zi
states are of obvious utility as a generating function for the energy eigenstates; the n-th term
in the power series expansion in z of the coherent state is √1n! |ni. This is only the first of
their remarkable properties.
Let us solve the Schrödinger equation with |zi as an initial condition.
∞
i zn 1
|z, ti = e− ~ Ht |zi = e−iωt(n+ 2 ) (a† )n |0i = e−iωt/2 |z(t)i,
X
(5.31)
n=0
n!

where z(t) = e−iωt z. That is, the quantum evolution of a coherent state is, up to a phase
which just measures the ground state energy, exactly give by the classical evolution of the
coherent state parameter z. Indeed, we will see in a moment that both the position and
momentum space wave functions of the coherent state are Gaussians of time-independent
width, centered around a classically evolving position and momentum. Although this result
is special to the harmonic oscillator, it shows that the spreading of wave packets in position
space, which we found for free particles, can be drastically modified by forces acting on those
particles. This observation that coherent states satisfy classical equations of motion is at the
basis of the connection between particles and classical fields in QM.
The scalar product of two coherent states hw|zi is easily computed as
∗ † X (w∗ )m z n ∗
hw|zi = h0|ew a eza |0i = √ hm|ni = ew z , (5.32)
m,n m!n!
∗
where we have used hm|ni = δmn . Thus, the norm of a coherent state is ez z . We also learn
that different coherent states are not orthogonal to each other.
The final important property of coherent states is that

a|zi ≡ z|zi. (5.33)

The annihilation operator is not a normal operator, since it does not commute with its
Hermitian conjugate, but this does not mean it cannot have eigenstates. What it does mean
is that the eigenstates are not orthonormal, as we have just seen. To prove this identity, we
use the operator equation
[a, (a† )n ] = n(a† )n−1 , (5.34)
which we prove by applying the Leibniz rule for commutators

[A, BC] = B[A, C] + [A, B]C, (5.35)

 The Harmonic Oscillator  119

repeatedly. Then

X (za† )n X (za† )n
a|zi = a |0i = [a, ]|0i = z|z > . (5.36)
n! n!
We can find the position (ψ(x, z)) or momentum (ψ(p, z)) space wave functions for coherent
states by solving the equation a|zi = z|zi in the appropriate basis:

1 ∂
√ (mωx + ~ )ψ(x, z) = zψ(x, z).
2~mω ∂x
1 ∂
√ (i~mω + ip)ψ(p, z) = zψ(p, z).
2~mω ∂p
It is easy to solve these first-order linear Ordinary Differential Equations (ODEs), and the
answers are both Gaussians. You will work out the details, as well as the proper normal-
izations of these functions, in Exercise 5.2. The correctly normalized position space wave
function is
mω 1/4 − mω (x−x0 )2 ik0 x
ψ(x, z) = ( ) e 2~ e , (5.37)
π~
where s
2~
x0 = Re z ,
mω
and r
2mω
k0 = Im z .
~
This is the wave function corresponding to the normalized coherent state

hx|zi
ψ(x, z) = p . (5.38)
hz|zi

We have seen that the coherent states are not orthogonal. The statement that they are over
complete is:
Z 2
d z −zz∗
1= e |zihz|. (5.39)
π
∗
To prove this, we use the power series expansions of eaz and its complex conjugate and find
that the right-hand side of this equation is

1 d2 z ∗ m n −z∗ z
Z
(a† )n |0ih0|am
X
(z ) z e . (5.40)
m,n n!m! π
120  Quantum Mechanics

If we write z = reiθ , then the angular integral of the terms with m 6= n vanish. Recalling the
normalization
(a† )n
|ni = √ |0i,
n!
we find that the coefficient of the projection operator |nihn| in this sum is
2
Z ∞ 2
cn = rdre−r r2n . (5.41)
n! 0

Introducing u ≡ r2 , we rewrite
1
Z ∞
cn = due−u un = 1. (5.42)
n! 0

Thus, the right-hand side is equal to n |nihn| = 1.

P

As a final bonus, we can compute the position space eigenfunctions of the harmonic
oscillator Hamiltonian. The power series expansion in z of the coherent state is
∞
zn
(a† )n |0i.
X
(5.43)
n=0
n!

From our discussion above, we recognize the term proportional to z n as being proportional
to the n-th eigenstate. We have
∞
X zn
|zi = √ |ni. (5.44)
n=0 n!
Thus,
∞
X zn
ψ(x, z) ≡ hx|zi = √ ψn (x). (5.45)
n=0 n!
Since ψ(x, z) is a Gaussian, with z appearing only in the term linear in x, each of these
functions has the form r
mω − mω x2
ψn (x) = Hn (x )e 2~ ,
2~
where Hn (x) is a polynomial. It is called the n-th Hermite polynomial. These eigenfunctions
were originally found by solving the second-order ODE

hx|H|ψi = Ehx|ψi,

subject to boundary conditions of normalizability.

Many older textbooks emphasize that the Schrödinger equation is a differential equation,
and exploit the fact that similar methods are used to solve Maxwell’s equations, with which
 The Harmonic Oscillator  121

students already have some familiarity. You can follow this route in Exercise 5.3. The truth
is, it has very little utility. The operator methods we have exploited are far simpler and, more
importantly, they generalize to a large class of many body systems. There is no calculation for
the harmonic oscillator, which is more transparent in the language of differential equations
than it is in that of coherent states. For the most part, the utility of differential equations
in QM is restricted to problems that can be reduced, exactly or approximately, to ordinary
differential equations. Although such problems abound in textbooks, they are not easily
found in applications of QM to important problems in modern physics. On the other hand,
the use of operator algebra, particularly the algebra of creation and annihilation operators,
is a standard tool in most real world applications of QM.

5.3 QUANTIZATION OF FIELDS AND WAVE–PARTICLE DUALITY

Wave–particle duality is a phrase, which was thrown around a lot in the early days of QM
and still appears in many contemporary textbooks. It was almost universally interpreted
as referring to the Schrödinger wave function in coordinate space, as if it were a wave in
physical space. One cannot emphasize too strongly that this interpretation is misleading.
The Schrödinger wave function for anything more than a single particle, obeys a “wave
equation” in the 3N + 1 dimensional configuration space of particle coordinates (plus time),
which is not physical space.
The desire to redefine the phrase wave–particle duality is based on many years of teach-
ing QM to juniors, seniors, and graduate students. The vast majority of students come into
class thinking of the Schrödinger wave function as a wave in three-dimensional space and
are shocked to learn the obvious fact that it is not, for multiparticle systems. The histori-
cal concept of wave–particle duality was based on a conflation of false thinking about the
Schrödinger wave function, with the obvious fact that the electromagnetic field manifested
both as a particle and a wave.
Indeed, as we have seen, the Schrödinger wave function is just a device for keeping track
of the probabilities of a whole collection of variables. Its entire physical content consists of
the formula
hAi = Tr (APψ ),

for all normal operators on Hilbert space. In this formula, the left-hand side refers to the
statistical expectation values of the quantities associated with A, some of which are subject
to measurement. Pψ is the projection operator in Hilbert space on the physical state whose
coordinate space representation is the Schrödinger wave function ψ(xi ). In coordinate space,
Pψ is an integral operator, with kernel

K(x, y) = ψ(x)ψ ∗ (y).

122  Quantum Mechanics

We test the predictions of the theory by preparing the system in the same state over and over
again, and measuring each of the variables. Then we compare the frequencies of occurrence
of each value of each variable, with the above formula for the expected value. That is the
only role of the wave function in the theory. We do not measure the wave function directly.4
We can extract knowledge of the wave function only by doing repeated experiments with the
same initial state.
By contrast, a classical field, like Maxwell’s electromagnetic field, is a variable that we
measure in single experiments. In this sense, it is like a particle coordinate or momentum and
it is certainly not a probability distribution. In order to understand the relation of such fields
to particles, we have to study the quantum mechanical treatment of fields. This is usually
considered a very advanced subject and is often discussed only in graduate courses. In fact,
the rudiments of the theory of quantized fields are just a simple extension of our discussion of
harmonic oscillators. Discussing them here will enable us to understand a concept that really
deserves the name of wave–particle duality, the Planck–Einstein–Compton (PEC) notion of
photons, and the peculiar notion of identical particle statistics, with a very small incremental
effort. Physicists should cease to use the term wave–particle duality, when referring to the
relation between the Schrödinger wave function and the observables of particles.
The electromagnetic field is a complicated object, with six components, and a host of
subtle properties related to gauge invariance. We will restrict our attention instead to a field
with one component, satisfying the same wave equation as each component of the electric
and magnetic fields of a light wave

∂t2 φ(x, t) − c2 ∇2 φ(x, t) = 0. (5.46)

Most of the physics discussed in this book is invariant under Galilean transformations, or
can be thought of as a set of equations one obtains by going to the center of momentum
reference frame by a Galilean boost. The above wave equation is not. It involves a particular
velocity c of propagation of waves, whereas Galilean transformations add a constant to the
velocity. There are two reasons for studying this equation here. The most important one is
that the physics of atoms, molecules, and condensed matter systems is crucially dependent
on the properties of photons, and in particular on the PEC relation between photon energy-
momentum and frequency-wave number. The Planck distribution (Chapter 12) also depends
on the properties of photons/electromagnetic waves and the duality between them. Finally,
we should note that exactly the same wave equation describes the properties of sound waves
and the quasiparticles called phonons, in condensed matter systems. In that context, the
“contradiction” between Lorentz invariant wave equations and Galilean invariant physics is
resolved because the phonon wave equation is valid only in the rest frame of the material. The
4
See Chapter 10 on Measurement Theory for a more precise and detailed discussion of the meaning of this
sentence.
 The Harmonic Oscillator  123

mathematical treatment of this wave equation at this point in the book is natural because,
as we are about to see, it follows simply by copying our treatment of the harmonic oscillator.
The Fourier transform
d3 k ik·x
Z
φ(x, t) = e φ(k, t), (5.47)
(2π)3
turns Equation 5.46 into a continuous infinity of harmonic oscillator equations
d3 k ik·x
Z
e [φ̈(k, t) + c2 k2 φ(k, t)] = 0. (5.48)
(2π)3
Remember that if the Fourier transform of a function vanishes, then the function itself
vanishes.
If a continuous infinity of oscillators makes you nervous, we can put the system in a box,
with boundary conditions on φ. For example, if we impose periodic boundary conditions on a
box of size L, the wave numbers will be restricted to kn = 2πn L , where n is a vector of integers.
Any sort of boundary condition will reduce the continuous infinity to a discrete set of kn . If
you are still nervous about an infinite number of oscillators, we can agree that our experiments
are never going to probe wavelengths shorter than some minimum size r0 . This puts an upper
cutoff on the wave |kn | < r10 . None of this detail distorts the main message, which is that
a field satisfying a linear, translation invariant, field equation is nothing but a collection of
decoupled harmonic oscillators, one for each wave number. For our particular wave equation,
which is called the D’Alembert equation, or the massless Klein–Gordon equation, the relation
between the oscillator frequency and the wave number is ω = c|k|, which is the same as that
for light waves.
We can now quantize all these oscillators, using the techniques of the previous section.
The energy eigenstates consist of the ground state, and states obtained by acting on it with
some number of creation operators. These eigenstates satisfy
n
(H − E0 )a† (k1 ) . . . a† (kn )|0i = [ ~ω(kj )]a† (k1 ) . . . a† (kn )|0i.
X
(5.49)
j=1

That is, the eigenstates of the quantized field look like the states of a collection of noninter-
acting “photons” obeying the PEC relation between energy, momentum and wave number.
We say this because the states are labelled by a finite set of wave numbers, the energy is a
sum of contributions from independent wave numbers, and the energy and wave number are
related by the PEC formula. Note however that these “particles” are automatically indistin-
guishable, and that only states symmetric under particle interchange are allowed, because
the different creation operators all commute with each other.
On the other hand, time-dependent coherent states of these oscillators are parameterized
by functions z(k, t) satisfying the Fourier transformed classical field equations. This is just
124  Quantum Mechanics

our result from the previous section that the coherent state parameters satisfy the classical
oscillator equations, combined with the Fourier transform. The function

d3 k ik·x
Z
z(x, t) = e z(k, t), (5.50)
(2π)3

satisfies the classical field equations and completely characterizes a combined coherent state
of the oscillators with fixed wave number.
This then is the concept that deserves the name wave–particle duality. A single system,
the quantized field, has two different kinds of states. Its energy eigenstates are multiphoton
states, and have a particle interpretation. Its coherent states are associated with classical
fields. The two kinds of states are incompatible with each other, in the sense that eigenstates
of the matrices σ1 and σ3 were incompatible with each other for a two state system. In
an energy (and particle number) eigenstate, the probability for finding a coherent state
characterized by a fixed classical field history is

|hz(k, t)|a† (k1 ) . . . a† (kn )|0i|2 . (5.51)

A fixed energy eigenstate does not correspond to a definite classical field. Conversely, if we say
we have a fixed classical field, then energy and particle number have statistical fluctuations.
You will explore these fluctuations in Exercises 5.8 and 5.9. It does not really make sense
to treat a coherent state as a classical field unless the fluctuations in its energy are small
compared to the expectation value of the energy. In the exercises, you will see that this is
equivalent to saying that the field is large. More precisely, it says that the number of particles
in each classical mode of the field is much larger than one. Coherent states that do not satisfy
this condition are better thought of as superpositions of states of a small number of particles.
It should be emphasized that the classical nature of the coherent state parameters is
not approximate. For fields satisfying linear field equations, the exact quantum equations
of motion lead to classical field equations for the coherent state parameters. For strongly
nonlinear field theories, both the existence of approximate classical fields and the particle
interpretation of low-energy excitations about the ground state have to be rethought, but for
linear field theories, the properties of the simple harmonic oscillator cement the connection
between waves and particles.
As we noted above, the particle interpretation of the multiphoton states leads to another
surprise. In classical particle mechanics, particles are distinguishable, even if their Hamil-
tonian has an exact permutation symmetry. Speaking colloquially, we can imagine that we
watch the particles move around, starting from their original positions, and keep track of
which was which. When this kind of reasoning is applied to classical statistical mechanics, it
leads to the “Gibbs Paradox” in the entropy of mixing of two identical volumes filled with
identical classical particles.
 The Harmonic Oscillator  125

If one takes a box filled with equal volumes of two different types of gas, separated by
a partition, and removes the partition, the entropy goes up when the box has returned to
equilibrium.
According to Boltzmann’s statistical mechanics, this is explained by the fact that each
individual gas molecule has twice as much volume to move around in. However, this argument
makes no reference to whether the gases are the same or different. Experiment shows that
there is no corresponding decrease in temperature when the two sides of the partition are
filled with identical particles.
Gibbs solved this problem by postulating that identical particles were indistinguishable.
The particle interpretation of the excitations of quantum fields automatically produces par-
ticles that are indistinguishable. If we think of the field as the basic entity, then obviously
exchanging two localized excitations of the field does not change the field configuration, and
leads to the same physical state. Permutations of identical particles are, in this case, not
merely symmetries of the system, but redundancies of a particle description of the under-
lying field degrees of freedom.5 The experimental fact that there is no increase in entropy
when we remove a partition between two volumes filled with the same gas is evidence that
this identical particle picture is correct. However, it is not by itself evidence for QM.6
With small modifications, we can perform a similar quantization of the Schrödinger equa-
tion, viewed as a classical field theory in space time. If we write the Fourier transform
d3 k ik·x
Z
Ψ(x, t) = e b(k, t), (5.52)
(2π)3
then the Schrödinger equation for Ψ, combined with its complex conjugate, is equivalent to
~k 2
∂t b(k, t) = −i b(k, t), (5.53)
2m
~k 2 †
∂t b† (k, t) = i
b (k, t). (5.54)
2m
Again, these look like the equations for the creation and annihilation operators of an infinite
set of oscillators (where we have used the † notation for Hermitian conjugation of operators,
5
In modern high-energy physics, such redundancies are called gauge symmetries, and are discrete analogs
of the gauge invariance of the description of the electromagnetic field in terms of potentials.
6
For many years, Einstein advocated models in which particles would appear as localized solutions of non-
linear classical field equations. In such models, particle exchange would also be a redundancy. Modern
quantum field theory [15] unifies these two points of view. Localized solutions to nonlinear field theories
indeed exist. When these models are quantized, one finds that both the localized solutions and small
fluctuations around a translation invariant background are particles. The analysis is valid in an approx-
imation analogous to the JWKB approximation we will study in Chapter 17, which is dependent on the
existence of a small dimensionless expansion parameter g. In the small g limit, particles that arise as small
fluctuations, have much lower energy than particles that are localized classical solutions.
126  Quantum Mechanics

rather than the ordinary complex conjugate, in anticipation of the operator interpretation of
the Fourier coefficients of the field). We quantize the classical Schrödinger field by imposing

[b(k), b† (k0 )] = δ(k, k0 ), (5.55)

where the right-hand side is a Kronecker delta. We have again put the field in a box with
boundary conditions that make the allowed set of wave numbers discrete. The solution of the
time-dependent Schrödinger equation,

d3 k ik·x −i ~k2
Z
Ψ(x, t) = e e 2m b(k), (5.56)
(2π)3

is now interpreted, as was the field φ(x, t) above, as a quantized field operator, in the Heisen-
berg picture, acting in the Hilbert space of multiparticle states, which are the eigenstates of
the energy operator
X ~2 k 2
H= b† (k)b(k). (5.57)
k
2m
The multiparticle states will all be symmetric under any permutation of the particles.
Let us end this section by explaining why the misleading notion that wave–particle duality
had to do with the Schrödinger wave function arose. To this end, consider states of single
particles. These have the form
~k2 t
e−i 2m a† (k)|0i, (5.58)
and represent a single particle with momentum ~k. The time-dependent factor comes from
H
acting with e−i ~ t on a† (k)|0i. We can consider superpositions of such states, defined by a
complex function of wave number f (k):

d3 k −i ~k2 t
Z
|f i ≡ e 2m f (k)a† (k)|0i. (5.59)
(2π)3

These are states with a single particle in some superposition of momentum eigenstates, which
can be localized in x. The traditional single particle Schrödinger wave function is just the
Fourier transform
d3 k −i ~k2 t
Z
ψ(x, t) = e 2m f (k)e−ik·x . (5.60)
(2π)3
It then follows that the Schrödinger wave function satisfies the same wave equation as the
field Ψ(x, t) (indeed, one can think of it as the matrix element of the Heisenberg picture field
between the ground state and the state with a single particle). Richard Feynman was a great
advocate of using the principle that “the same equations have the same solutions,” to relate
the solutions of one physics problem to those of an already solved problem. However, the fact
 The Harmonic Oscillator  127

that two different quantities satisfy the same equation does not imply that they have the
same meaning. In this case, the quantized Heisenberg field operator, the classical coherent
state parameters, and the single particle Schrödinger wave function all satisfy the same field
equation, but they all have different meanings. The operator is an observable in QM and
can, in principle, be measured at each time. If the system is in a particular coherent state
then, as you will show in Exercise 5.10, the expectation value of the field operator will be
equal to the coherent state parameter, at all times. The statistical fluctuations around this
expectation value may be either large or small compared to the expectation value itself. You
will find that when the classical field value is large, the fluctuations are small.
The Schrödinger wave function is something else entirely. It tells us what the probabil-
ities are, in the single particle state |f i to find the particle with either a given position or
a given momentum. It cannot be probed by a single measurement, but only by repeated
measurements of identically prepared states. The same equations have the same solutions,
but the different uses of those solutions can have meanings that are conceptually different.
Let us summarize the important things we have learned in this section. First and foremost,
we have derived the PEC relation between the energy of the “smallest quantum of light of a
given frequency” and that frequency. We used this in our discussion of the ammonia molecule,
but now we really understand it. We have also learned that photons are identical particles,
because they are excitations of a quantized field. Gibbs’ classical discussion of the entropy of
mixing suggests that all particles should arise in this way, and this turns out to be correct.
There is, however, one subtlety in the discussion. There turn out to be two different kinds of
identical particles in QM (and in the world), called bosons and fermions. Our discussion so
far applies only to bosons, we will introduce fermions in a moment.
Finally, we learned that classical electromagnetic fields are coherent states of bosons. We
first introduced coherent states for a single harmonic oscillator as a sort of technical trick
for doing calculations. Their real importance is in showing how classical particle physics
and classical field theory both emerge from the single concept of a quantized field, the real
meaning of wave–particle duality. Not bad for a little algebraic trick for solving a simple
harmonic oscillator!

5.4 FERMI–DIRAC STATISTICS

The formula for an n-particle state with each particle in a distinct single particle state:
|k1 . . . kn i = b† (k1 ) . . . b† (kn )|0i (5.61)
guarantees complete symmetry of the multiparticle states under interchange of the particles,
because the b† (k) operators commute with each other. It turns out that many familiar par-
ticles do not obey this symmetry rule, but rather have states that are totally antisymmetric
under interchange. Remember that any permutation is a product of transpositions, in which
128  Quantum Mechanics

only two particles are interchanged. The new rule is that if the number of transpositions is
odd, the state gets multiplied by a minus sign. Particles obeying the symmetric rule are called
Bose particles or bosons, and are said to obey Bose–Einstein statistics. Particles obeying the
rule with the minus sign are called fermions and obey Fermi–Dirac statistics. One can imple-
ment the rule easily by insisting that any pair of creation operators a† (k1,2 ) anticommute
with one another

a† (k1 )a† (k2 ) + a† (k2 )a† (k1 ) ≡ [a† (k1 ), a† (k2 )]+ = 0. (5.62)

Extending this, as in the Bose case, to k1 = k2 we get

a† 2 (k) = 0. (5.63)

Those of you who have had a class including elementary atomic physics will have heard of
the Pauli Exclusion Principle. The previous equation is the mathematical expression of that
principle. For bosons, we can make a state with an arbitrary number of particles, in the same
single particle state. Fermions “don’t like neighbors”: you cannot put more than one in any
single particle state. This property is the key to the stability of atoms and nuclei, as well as
the solidity of solids and many other facts about the real world.
In Exercise 5.11, you will show that in order to be consistent with the anticommutation
relations for creation operators, their adjoints have to satisfy anticommutation relations as
well:
[a(k1 ), a† (k2 )]+ = δ(k1 , k2 ), (5.64)
where the right-hand side is a Kronecker delta, because we have put the system in a box to
make the wave numbers discrete. In fact, for a fixed value of k1 = k2 , we have already run
across operators with these properties when we studied two state systems, as you will recall
in Exercise 5.12. In Exercise 5.13, you will learn a trick, called the Jordan Wigner transfor-
mation, for constructing the full algebra of Fermion creation and annihilation operators out
of a collection of Pauli spin operators.
Given a classical wave equation, like

(∂t2 − ∇2 )φ = 0, (5.65)

or
(i~∂t − ∇2 )Ψ = 0, (5.66)
you can expand it in Fourier components and the corresponding “creation and annihilation
operators” which evolve with negative and positive frequencies. One can then quantize the
system as one would a harmonic oscillator, obtaining a theory of Bose particles, or one can
 The Harmonic Oscillator  129

choose the alternative Fermi–Dirac quantization, obtaining Fermi particles. This raises three
questions:
• How do we choose which statistics to use for each type of particle?
• Is there anything weird about the minus signs for Fermions?
• Are there other possibilities besides the Bose–Fermi alternative?
It turns out that the combination of relativity and QM resolves the first question. We will
learn later that every elementary particle carries an internal angular momentum called spin.
Spin describes the way the states of a particle at rest respond to rotations. It turns out that
the spin state of a particle can change by a minus sign under a 2π rotation, because that will
not affect the density matrix, which computes all physical expectation values in that state.
Particles exhibiting that peculiar minus sign are said to have half integral spin, while those
whose states are invariant under 2π rotations have integral spin. It turns out that when we
combine QM with relativity, and with the principle that signals cannot be sent faster than
the speed of light, half integral spin particles must be a Fermions, and integral spin particles
Bosons. This is called the Spin Statistics Theorem [16].
The odd thing about the minus sign is the following: The creation operators for two
fermions in different states do not commute with each other. Suppose that the states are
localized millions of miles apart. If we can create single fermions locally, then an experiment
right here and now, could be affected by an experiment done simultaneously millions of miles
away. The resolution of this potential paradox is that the rules of the game, and in particular
the Spin Statistics theorem, only allow even numbers of fermions to be created locally. As
we will see, given the Spin-Statistics connection, this follows from conservation of angular
momentum. Thus, if we insist on angular momentum conservation, then it is impossible to
change the number of fermions from even to odd, or vice versa.
Finally to the question of other types of statistics, it turns out that if the world only
had two space dimensions, then other kinds of particles, called anyons, are possible, but
in higher dimension they are not allowed. Nonetheless, there are certain condensed matter
systems, confined to a plane, which exhibit anyonic excitations. This is all connected to a
wonderful purely quantum mechanical phenomenon, called the Aharonov–Bohm effect, and
we will explore it in Chapter 15.

5.5 EXERCISES
5.1 Prove that the eigenvalues of the square of a Hermitian operator are all nonnegative, by
proving the stronger theorem that H 2 has nonnegative expectation value in any state.
The latter theorem is true for any sum of squares of Hermitian operators, and proves
the statement about eigenvalues for such sums as well.
130  Quantum Mechanics

5.2 Solve the equations for a coherent state in position and momentum representations:

1 ∂
√ (mωx + ~ )ψ(x, z) = zψ(x, z).
2~mω ∂x

1 ∂
√ (i~mω + ip)ψ(p, z) = zψ(p, z).
2~mω ∂p
Write down the solutions that have the normalization hw|zi = ew∗z , used in the text.

5.3 Solve for the harmonic oscillator eigenstates by direct solution of the second-order
equations.

a. The equation is
−~2 d2 kx2
[ + ]ψ = Eψ,
2m dx2 2
2
q
k
with the boundary condition that |ψ|2 < ∞. Define x ≡ ( k~2 m )1/4 and E ≡ ~
R
m
and show that
d2
[− 2 + y 2 ]ψ = 2ψ.
dy
y2
b. Show that at large |y|, the two linearly independent solutions behave like e± 2 .
A solution is normalizable only if it is a falling Gaussian at both ±y → ∞.
c. Write the solution as
∞
y2
ak y k e− 2 ,
X
ψ=
k=0

and convert the differential equation into a recursion relation for the coefficients ak .
Show that unless the recursion terminates, and the prefactor is a polynomial, the
y2
solution will behave like e 2 . Do this by approximating the recursion relation for
large k. This polynomial condition is only achieved for quantized values of .

5.4 Calculate the expectation value and the fluctuations of the number operator a† a in the
coherent state |zi. Remember that the coherent state is not normalized to 1 so you have
to divide by the norm to calculate expectation values.

5.5 Calculate the overlaps of the ground states of two harmonic oscillators with different
frequencies.
 The Harmonic Oscillator  131

5.6 The Klein–Gordon equation in d spatial dimensions is

∂2 X ∂2 1
( 2 2
− 2 )φ + 2 φ,
c ∂t i
∂xi `

~
where l is the Compton wavelength l = mc . Calculate the frequencies of modes of
wavenumber k (a d-dimensional vector). Use the result of the previous problem to
calculate the ground state overlap of the Klein–Gordon field with two different values
of l. You will have to put the system in finite volume in order to get a nonzero answer.
In high enough dimension (how high?), you will find the overlap is zero even in finite
volume if there is no ultraviolet cutoff on large wave numbers.

5.7 Use the relations

a† |ni = (n + 1)1/2 |n + 1i
and
a|ni = n1/2 |n − 1i
to write recurrence relations for the Hermite polynomials and their derivatives.

5.8 Compute the uncertainty in the energy operator H = ~ω(a† a + 12 ) in a coherent state
|zi. Remember that you have to divide by the norm of the state. Note that the constant
ground state energy cancels out of this computation.
† P
5.9 Compute the uncertainties in particle number N = k a (k)a(k) and energy
P †
H = k ~ω(k)a (k)a(k) in a coherent state of the Klein–Gordon field φ(x, t).

5.10 Show that the expectation value of the operators P and X in a coherent state are
proportional to the imaginary and real parts of the coherent state parameter. Generalize
this computation to the Klein–Gordon field.

5.11 Show that the anticommutation relation

[a(k1 ), a† (k2 )]+ = δk1 k2

is required for anticommuting creation and annihilation operators.

5.12 Show that the anticommutation relations

a2 = 0, [a, a† ]+ = 1

are realized in a two-dimensional Hilbert space.

132  Quantum Mechanics

5.13 Given an infinite collection of Pauli matrices σa (i), −∞ < i < ∞ show that the
operators Y
ψ(i) = σ+ (i) σ3 (j)
j>i

1
σ+ (i) = √ (σ1 (i) + iσ2 (i))
2
satisfy
[ψ(i), ψ(j)]+ = 0,
[ψ(i), ψ † (j)]+ = δij .
This is called the Jordan–Wigner transformation.

5.14 A different way to solve for the time evolution of a coherent state is the following: the
time evolution of the system, starting at t = 0 in a coherent state is given by
H
|zi → e−i ~ t |zi. (5.67)

We act on the time evolved state with the Schrödinger picture operator a (this is the
Heisenberg picture operator, at t = 0).
H H H H
ae−i ~ t |zi = e−i ~ t (ei ~ t ae−i ~ t )|zi. (5.68)

The quantity in parentheses is just the Heisenberg picture operator a(t). Use the Heisen-
berg equations of motion to evaluate a(t) and use this to reproduce the formula in the
text for the time evolution of |zi.
 CHAPTER 6

Review of Linear Algebra and

Dirac Notation

6.1 INTRODUCTION
This chapter provides a more formal introduction to the mathematics of Hilbert space, using
Dirac notation. It also contains a general discussion of symmetries in quantum mechanics
(QM) and a derivation of the generalized uncertainty relations for any pair of operators that
do not commute. Remember that the key idea of QM is to view the complex N -dimensional
generalization of Pythagoras’ theorem for a vector of length 1:

X
1= |hv|en i|2 , (6.1)
n

as the definition of a probability distribution for each such unit vector. The individual terms in
the sum are interpreted as the probability that, given that the system is in the state represented
by |vi, a measurement to determine whether the system is in the state |en i will find that it is
in fact in that state. This probability distribution over all orthonormal bases is equivalent to
a probability distribution over all normal operators, defined as operators that are diagonal
in some basis. The equivalent formula for operators is that the expectation value of Ak in
the state |vi is given by
hAk i = Tr Ak Pv , (6.2)

where Pv ≡ |vihv| is the projection operator on |vi. The theory of Hilbert space generalizes
all of this to the limiting case of N → ∞. There are lots of subtleties, the most signifi-
cant of which is that operators can have continuous spectra and there are no normalizable
eigenvectors for those eigenvalues.

133
134  Quantum Mechanics

6.2 SEPARABLE HILBERT SPACES

It is about time to review linear algebra a bit more formally than we have up till now. We
will discuss linear algebra in separable Hilbert spaces. A separable Hilbert space is an infinite
dimensional vector space, with a countably infinite basis and a scalar product. We will include
the finite dimensional spaces you know and love from linear algebra in our definition of a
Hilbert space. We will use Dirac’s notation |en i for some particular basis, and replace e by
other lower case Latin letters to denote other bases. We will also mostly follow the convention
that the basis of eigenstates (eigenvectors normalized to one) of a normal operator labelled
by a capital Latin letter like A, will be denoted |ak i. The major exception to this is the
Hamiltonian operator H, whose eigenvalues are energies, and whose eigenstates are labeled
|Ek i.
Given a basis |en i, we can define a scalar product such that the basis vectors are orthonor-
mal
hen |em i = δmn . (6.3)
A general vector is given by X
|vi = vn |en i, (6.4)
where X
kvk2 ≡ |vn |2 < ∞.
The scalar product of two such vectors is

vn∗ wn = hw|vi∗ ,
X
hv|wi = (6.5)

which identifies the components of |wi as a column vector and the components of hv| as a
Hermitian conjugate (transpose followed by complex conjugate) row vector. Dirac’s termi-
nology for column and row vectors is kets and bras, because they combine together to make
a bra(c)ket.1 Mathematicians call column vectors simply vectors, and row vectors are called
covectors, and thought of as complex valued linear functions defined on the vector space. The
scalar product is an identification of the vector space with the vector space of linear func-
tions defined on it. Given a basis of vectors |en i, there is a canonical basis of linear functions
defined by Fn (|em i) = δmn . If we define the identification of vectors with covectors/linear
functions by associating each vector with the function that is equal to one on it, and zero on
all other basis vectors, then we have defined the scalar product so that this particular basis
is orthonormal, a portmanteau word that says that each vector in the basis is normalized to
one and orthogonal to all the other vectors. We will always work with orthonormal bases,
except when using coherent states, which form an overcomplete basis.
1
One supposes that the c in bra(c)ket stands for the vertical line in the notation for scalar product.
 Review of Linear Algebra and Dirac Notation  135

It is easy to prove (Exercise 6.1) the Schwarz inequality, which states that
q
|hv|wi| ≤ kvkkwk. (6.6)

This inequality shows that the scalar product is finite, for all vectors with finite norm. It is
the generalization of the intuitive statement from two-dimensional space that the projection
of one vector on the direction of another is no longer than the vector itself. It also follows
easily that
kvk2 = hv|vi.
The final axiom defining a separable Hilbert space is that it is complete. That is, the scalar
product defines a norm or distance, by Pythagoras’ formula2
q qX
kv − wk = hv − w|v − wi = |vn − wn |2 . (6.7)

A sequence of vectors whose norm converges to zero is said to converge to zero, and com-
pleteness means that the limit |vi, in this sense, of any sequence of vectors, |vi i → |vi, is in
the Hilbert space.
If we have two different separable Hilbert spaces, with bases |en i and |fn i, respectively,
then the map |en i → |fn i maps one into the other in a way that preserves the vector space
structure and the scalar product, so there is really only one infinite dimensional separable
Hilbert space, just as there is only one finite dimensional Hilbert space for each dimension.
This sounds pretty remarkable when you realize it means that the whole standard model of
particle physics is somehow contained in the space of states of a single nonrelativistic free
particle.
What this result is really telling us is that physics is really about the choice of particu-
lar operators, which are measurable in a particular system. We often talk about “measuring”
any operator in our Hilbert space, but this is an exaggeration. Each system has certain simple
operators, which we have a prescription for measuring, and those are really the only ones of
interest. We are generally going to ignore this subtlety, but you should remember it.

6.3 UNITARY TRANSFORMATIONS, UNITARY MATRICES, AND

CHANGES OF BASIS
The column and row vector representations of vectors |vi and covectors hw| depend on the
choice of orthonormal basis. If we have another orthonormal basis |fn i, we can expand the
|en i basis in terms of it
2
Notice that in this equation, we have made a convention we will follow from now on, which is that sums
without limits on them are taken from 1 to ∞. Similarly, integrals without limits will be taken from −∞
to ∞.
136  Quantum Mechanics
X
|en i = |fk ihfk |en i (6.8)
k

Then, X
|vi = vn hfk |en i|fk i. (6.9)
k,n

Thus, the column vector that represents |vi in the |fk i basis is
(f )
vk = hfk |en ivn(e) , (6.10)

where we have left the sum over n implicit, using the Einstein summation convention.
The matrix
Ukn ≡ hfk |en i (6.11)
is thus the matrix that transforms the coefficients of a vector in the e basis into its components
in the f basis. It satisfies the defining equation for a unitary matrix

U U † = U † U = 1, (6.12)

where the 1 in this matrix equation is the unit matrix. To prove this identity, we use the
most important formula in Dirac notation, called resolution of the identity
X
1= |en ihen |. (6.13)
n

It is true for any orthonormal basis. The resolution of the identity is a special case of Dirac’s
notation for dyadic operators D(v, w) = |vihw|. The dyadics |ek ihel | are operators whose
matrix in the |ek i basis has a 1 in the k-th row and l-th column and zeros everywhere
else. The operator |ek ihek | (no sum) is the projection operator on the vector |ek i, and the
resolution of the identity follows from this.
Now recall that the Hermitian conjugate operation † is the same as complex conjugation
followed by transposition (or the other way around). So,

(U † )kn = Unk
∗
= hfn |ek i∗ = hek |fn i. (6.14)

Thus,
Ukn (U † )np =
X X
hfk |en ihen |fp i = δnp . (6.15)
n n

The other direction of this multiplication is equally easy to prove (Exercise 6.2).
The defining equation for unitarity is the same as the statement that both the rows and
the columns of the unitary matrix form an orthonormal basis. We can think of these as the
form of the e basis vectors in the f basis, and vice versa. This shows that the most general
 Review of Linear Algebra and Dirac Notation  137

unitary matrix represents the transformation between two different orthonormal bases. Con-
versely, we can get any orthonormal basis by acting with all possible unitary transformations
on some fixed orthonormal basis. The relationship between a pair of bases and a unitary
operator is best understood with an example. Consider the 4 × 4 unitary matrix
 
U11 U12 U13 U14
U U22 U23 U24 
 21
(6.16)

U31 U32 U33 U34 
 

U41 U42 U43 U44 .

The matrix elements should be thought of as the form of an operator U in a particular basis:
Umn = hem |U |en i. The basis vectors in that basis have the usual form: a one in a single
row and zeroes elsewhere. Consequently, the columns of the unitary matrix are the form of
the new basis vectors |fn i, expressed in the |en i basis. Correspondingly, the columns of U † ,
which are the complex conjugates of the rows of U , are the forms of the |en i basis vectors in
the |fn i basis. You should make sure that you understand how this paragraph relates to the
formulae
X
U= |fk ihek |, (6.17)
k

U† =
X
|ek ihfk |. (6.18)
k

It is worth rewriting all of this using the notion of linear operators. A linear operator on a
vector space is simply a function A(|vi), from the vector space to itself, which satisfies

A(a|vi + b|wi) = aA(|vi) + bA(|wi). (6.19)

We denote the action of a linear operator by

A(|vi) ≡ A|vi, (6.20)

in order to distinguish it from any old function from the vector space to itself.
A linear operator is completely defined by its action on a basis |en i, because linearity then
fixes its action on any linear combination of basis vectors. Let us define such an operator by

U † |en i = |fn i. (6.21)

Then,
hek |U † |en i = hek |fn i = Unk
∗
. (6.22)
138  Quantum Mechanics

That is, the matrix U † is gotten by sandwiching the operator U † between a pair of basis
vectors of the e basis. We will call such a sandwich a matrix element of the operator U † . Now
let us compute the matrix elements of the same operator in the |f i basis.

hfk |U † |fn i = hfk |ep ihep |U † |er iher |fk i

X
(6.23)
p,r

† †
hfk |U † |fn i = Unk Ukl Ulm . (6.24)
Note that in writing this equation, we could have replaced U † by any other operator. Thus,
† (e)
A(f )
mn = Unk Akl Ulm . (6.25)

In words, this equation says that we get the matrix of an operator in the |fk i basis, by
conjugating its matrix in the |ek i basis, by the unitary matrix relating the components of
vectors in the two bases.

6.4 NORMAL OPERATORS ARE DIAGONALIZABLE OPERATORS

If an operator A has a matrix that is diagonal in some basis, equal to a complex diagonal
matrix D, then its matrix in any other basis is

A = U † DU, (6.26)

where U is the matrix that connects the two bases. This matrix satisfies

[A, A† ] ≡ AA† − A† A = 0. (6.27)

In words, it commutes with its adjoint, and is called a normal matrix. The same terminology
is used for operators, independent of the basis. For any operator,
1 1
A = [A + A† ] + i [(−i)(A − A† )]. (6.28)
2 2
The two operators in square brackets are each equal to their own Hermitian conjugate, and
are called Hermitian. For normal operators, these two Hermitian pieces, analogous to the real
and imaginary parts of a complex number, commute with each other, and can be diagonalized
simultaneously. This means that the study of normal operators reduces to that of Hermitian
operators. It is common to state that all detectables in QM are Hermitian operators. Such a
claim would not allow us, e.g., the freedom to take square roots of observable quantities. It
is an arbitrary and overly rigid rule.
We will generally abuse language and use the words operator and matrix interchangeably.
When we do, you will understand that we mean the matrix of an operator in some particular
 Review of Linear Algebra and Dirac Notation  139

basis, which may or may not have been specified. We will also occasionally use the word
representation to refer to a particular choice of orthonormal basis. For example, in speaking
of particles, we often refer to the position representation or the momentum representation.
We have also used this word in a rather special context, referring to the Schrödinger and
Heisenberg forms of the equations of motion of QM. This context is so special that we
usually use the term Schrödinger and Heisenberg picture, instead of representation. Later
we will also discuss the Dirac or interaction representation of dynamics. The general idea
is that abstract operators are the intrinsic dynamical variables/detectable quantities in QM
and that a particular choice of basis represents the abstract operator as a different matrix.
Every such choice is related to every other one by a unitary transformation.
The spectral theorem is the statement that every operator, which satisfies [A, A† ] = 0, is
conjugate to a diagonal matrix via unitary transformation, A = U † DU . For a finite dimen-
sional space, counting of parameters in the two equations would seem to indicate that this is
true (Exercise 6.3). A rigorous proof in finite dimensions goes along the following lines. The
spectrum of an operator A is the set of complex numbers λ such that A − λ is not invertible.
In finite dimensions, this is equivalent to the statement that there is an eigenstate |λi of A,
with eigenvalue λ. Let d(λ) be the degeneracy or number of independent eigenstates with
the same eigenvalue, and let P (λ) be the projection operator on the subspace of the Hilbert
space spanned by these eigenstates. In Dirac notation,
X
P (λ) = |λ (i) ihλ (i)|, (6.29)
i

where
A|λ (i) i = λ|λ (i) i. (6.30)
Then, A(1 − P (λ)) is a normal operator (prove it) on the Hilbert space perpendicular to the
subspace on which P (λ) = 1. We now repeat the above procedure, until we have written A
as a linear combination of a complete set of commuting orthogonal projection operators
X
A= λi Pi , (6.31)
X
Pi Pj = δij Pj Pi = 1. (6.32)
i

This is called the spectral theorem, and it is one of the most important results in all of
mathematics.

6.5 CONTINUUM EIGENVALUES AND UNBOUNDED OPERATORS

There are two subtleties that are encountered in extending this analysis to the case of infinite
dimensional separable Hilbert spaces. The first is the phenomenon of continuous spectra,
140  Quantum Mechanics

which we have encountered when discussing the single particle position operator x. That
is, there can be values of λ for which A − λ is not invertible, which do not correspond
to normalized eigenvectors of A. The eigenvectors are singular limits of square integrable
functions, like the Dirac delta function. The mathematically rigorous way of dealing with
this phenomenon is to note that an object defined formally by
Z λ+
P (, λ) = da |aiha| (6.33)
λ−

is a well-defined projection operator for any choice of λ in the spectrum of A, and any positive
number . Mathematicians speak of a projection valued measure dP (λ) and write a general
normal operator A as Z
A= dλ λdP (λ). (6.34)

The integral runs over the spectrum of A in the complex plane. Physicists generally ignore
this subtlety, and treat the continuum “eigenstates” as if they were normalized vectors in the
Hilbert space, taking care only that none of the expressions they actually use give infinite
answers.
The other subtlety has to do with the fact that not all of the operators physicists discuss
are actually well-defined operators on the entire Hilbert space. For example, differential oper-
ators only make sense when applied to differentiable functions, but most square integrable
functions are not differentiable. An even simpler example is the position operator x itself.
Acting on an L2 function of x it gives xf (x), which might not be square integrable. Square
1
integrability of f (x) requires only that it fall faster than x− 2 − for small positive , while
square integrability of xf (x) requires a faster falloff. Actually this is not an independent
example, because the Fourier transform takes multiplication into differentiation.
Operators that are well defined on each vector in some complete basis of the Hilbert space
but not on every vector in the Hilbert space are called unbounded operators. In all physical
cases, they arise when we have a one parameter set B(z), of bounded normal operators3
and we define the derivative of it with respect to the parameter A = dB dz . In this case, the
spectral theory of A is derived from the spectral theory of the family. Whenever we have
one parameter groups of symmetries, the infinitesimal generators of the symmetry group are
constructed in this way.
However, we are often given an explicit form for the operator A, and asked to derive the
family B(z) from it. For example, we are given the Hamiltonian, and are asked to construct
i
the time evolution operator e− ~ Ht . Then, in principle, we have to prove the existence of
3
Bounded operators have the property that kBψik ≤ kBkk|ψik for a fixed positive number kBk and any
vector in the Hilbert space. For normal operators, kBk is the largest absolute value of any point in the
spectrum of B. See Exercise 6.4
 Review of Linear Algebra and Dirac Notation  141

bounded functions of A, from the properties of A alone. Very occasionally, in physics, there
are subtleties that require us to be a little more careful about the math of unbounded
operators than usual. We will encounter one such subtlety for zero angular momentum wave
functions in spherical coordinates, in Chapter 7.

6.6 SUMMARY
Let us end this section by summarizing the properties of operators in separable Hilbert
spaces, using Dirac notation. Given a linear operator A and a basis |en i the matrix elements
of the operator in that basis are
Amn = hem |A|en i. (6.35)
|ek ihek |, we have
P
Using the resolution of the identity 1 = k
X
hem |AB|en i = hem |A|ek ihek |B|en i, (6.36)
k

which is the conventional rule for matrix multiplication. We can rewrite the operator in terms
of its matrix and the dyadic operators |ek ihel | as
X
A= Akl |ek ihel |. (6.37)
kl

In particular, if A is a normal operator ([A, A† ] = 0, see the next paragraph), its form in the
basis given by its eigenvectors is
X
A= ak |ak ihak |. (6.38)
k

The adjoint of an operator A, denoted A† is defined by

hem |A† |en i ≡ hen |A|em i∗ . (6.39)
This equation is true in any basis (Exercise 6.5). An operator is called normal if [A, A† ] = 0.
This implies (this is the spectral theorem) in particular that if there is a basis where the
matrix of A is diagonal, then A† is also diagonal in that basis. Particular examples of normal
operators are the unitary operators, for which U U † = U † U = 1. The matrix of a unitary
operator is a unitary matrix, which is the same as the statement that both the rows and
columns of the matrix are an orthonormal basis for the Hilbert space. In fact, the unitarity
equation implies that, given an orthonormal basis |en i, the vectors |fn i = U |en i are another
orthonormal basis for the space.
The spectral theorem states that any normal operator can be written as
Z
A= daµ(a)|aiha|. (6.40)
142  Quantum Mechanics

The spectral weight µ(a) is a measure, which means that it can have delta function contribu-
tions δ(a − an ), as well as a continuous piece. For the discrete spectrum (the delta function
contributions), the corresponding symbol |an i is a genuine normalized eigenstate of A. In the
continuous spectrum, we think of |ai as a linear function on states, defined by the value of
the wave function ha|ψi ≡ ψ(a) for each normalizable state |ψi. The “continuous eigenstates”
obey delta function normalization
1
ha|bi = δ(a − b).
µ(a)
It is conventional to use the fact that every normal operator can be written as the sum of
two commuting Hermitian operators and to write the above formula in terms of integrals
over a real parameter a. The spectral density µ(a) then has real and imaginary parts.
When A has continuous spectrum, and we express operators in terms of their action
“in the basis where A is diagonal,” then they become integral operators, also called integral
kernels. For simplicity, we assume that the discrete and continuous spectrum of A are disjoint,
and we work in the subspace of the Hilbert space orthogonal to the discrete eigenvectors.
We write states in this subspace in terms of their wave functions ψ(a) in the A basis. The
completeness relation takes the form
Z
1= |aiµ(a)da ha|,

and we have
ha|a0 i = δ(a − a0 ).
It is convenient to absorb the square root of the nonsingular spectral density µ(a) into the
definition of the wave function: q
ψ(a) ≡ µ(a)ha|ψi. (6.41)
A general linear operator takes the form
Z
ha|B|ψi = B(a, a0 )ψ(a0 ). (6.42)

Note that differential operators can also be written in this form, with kernels like
d
D(x, y) = δ(x − y).
dx

6.7 DIRECT SUMS AND TENSOR PRODUCTS

Given a pair of Hilbert spaces, He and Hf , we can define two new Hilbert spaces, called the
direct sum, He ⊕ Hf , and direct or tensor product, He ⊗ Hf . It is simplest to do this in
 Review of Linear Algebra and Dirac Notation  143

terms of a choice of bases |en i in He and |fN i in Hf . The direct sum is the set of complex
an |en i + bN |fN i. It has dimension De + Df . The tensor product
P P
linear combinations
space is spanned by vectors anN |en i ⊗ |fN i. It has dimension De Df . We often drop the ⊗
P

sign when talking about vectors in a tensor product. A straightforward argument, which is
left to the reader, shows that the resulting spaces are independent of the choice of bases.
The notion of the direct sum underlies the spectral theorem. A given normal operator
decomposes the Hilbert space into a direct sum of eigenspaces, in each of which it takes
on a fixed numerical value. In the case of continuous spectra, the direct sum is replaced by
something called a direct integral, which is the rigorous definition of some of the expressions
we have written in Dirac notation.
In physics, the tensor product arises when discussing the joint states of independent
systems. For example, a pair of ammonia molecules, labeled A and B, at low energy, has
states
|+i3A ⊗ |+i3B , |+i3A ⊗ |−i3B , |−i3A ⊗ |+i3B , |−i3A ⊗ |−i3B . (6.43)
There are two general theorems about operators that act separately on each factor of the
tensor product. Such operators are denoted 1 ⊗ A2 or A1 ⊗ 1, and we have
(1 ⊗ A2 )(A1 ⊗ 1) = (1 ⊗ A1 )(A2 ⊗ 1) = A1 ⊗ A2 = A2 ⊗ A1 . (6.44)
We also have
Tr [A1 ⊗ A2 ] = Tr1 [A1 ]Tr2 [A2 ], (6.45)
where the traces with subscripts are traces over the individual factors in the tensor product.

6.8 THE GENERALIZED UNCERTAINTY RELATION

We have seen that if two operators commute, then they are simultaneously diagonalizable,
which means that there are states of the system in which they are both definite at the same
time. In this section, we want to make a quantitative estimate of the mutual uncertainty
of two operators that do not commute. Given any probability distribution, any quantity A
will have an expectation value hAi. We define a quantity Ā ≡ A − hAi, which has zero
expectation value. If the values of A had a Gaussian, or normal distribution, then the width
of the Gaussian would be determined by hĀ2 i. A wider Gaussian, with more uncertainty,
would correspond to a larger value for this mean square average. We therefore define the
uncertainty of A, ∆A, for any probability distribution by
q
∆A ≡ hĀ2 i. (6.46)
Let us now apply this definition to the probability distributions of two noncommuting Her-
mitian operators, A and B, in a pure quantum state |ψi. We first observe that
hψ|[Ā, B̄]|ψi = iImhψ|ĀB̄|ψi. (6.47)
144  Quantum Mechanics

This means that

|hψ|[Ā, B̄]|ψi|2 ≤ |hψ|ĀB̄|ψi|2 ≤ |hψ|Ā2 |ψi||hψ|B̄ 2 |ψi|. (6.48)

In the last inequality, we have used the Schwarz inequality

|hχ|φi| ≤ k|χikk|φk (6.49)

applied to the vectors |φi = B|ψi and |χi = A|ψi. We end up with the generalized Heisenberg
uncertainty relation
∆A∆B ≥ |hψ|[A, B]|ψi|. (6.50)

If we apply this to the position and momentum of a particle, which satisfy [x, p] = i~, the
right-hand side is independent of the state and we get

∆x∆p ≥ ~. (6.51)

This is Heisenberg’s famous uncertainty relation between position and momentum. The care-
ful reader will note that we have implicitly used the fact that [Ā, B̄] = [A, B], which is true
because the barred quantities differ from the unbarred ones by an additive multiple of the
identity, and the identity commutes with everything.

6.9 SYMMETRIES AND CONSERVATION LAWS

In QM, a symmetry transformation must be an invertible operation S, which takes one vector
in the Hilbert space into another. It cannot change the probability of finding that something
is true, so S(|vi) must have a scalar product with S(|wi), which has the same magnitude as
hw|vi. Unitary operators have this property, because they preserve the entire scalar product,
including its phase. There is a class of nonlinear operators which also satisfy the weaker
condition of preserving probability. An operator is called conjugate linear if

C(a|vi + b|wi) = a∗ C(|vi) + b∗ C(|wi). (6.52)

A conjugate linear operator can also have the property that the scalar product of C(|vi) and
C(|wi) is equal to hv|wi, the complex conjugate of the original scalar product. Such operators
are called antiunitary. The product of two antiunitary operators is a linear unitary operator
(Exercise 6.6), so any two antiunitaries differ by multiplication with a unitary. We will see
below that the operation of time reversal must be taken to be antiunitary. If time reversal,
T , is a symmetry operation of a given system, then any other symmetry is either represented
by a unitary operator or by T U where U is unitary.
 Review of Linear Algebra and Dirac Notation  145

Symmetry operations must conserve probability, but that is not sufficient. In order to be
a symmetry of a given system, an operation must also preserve its equations of motion. We
can write the equations of motion of QM in either the Schrödinger or Heisenberg pictures

i~∂t |ψi = H|ψi, (6.53)

or
~∂t O = i[H, O]. (6.54)
If we take
|ψi → S|ψi, (6.55)
where S is a putative symmetry operation, these equations must be preserved. In the case of
the Schrödinger equation, this means S|ψi should satisfy the same equation as |ψi. Recall,
that in the Schrödinger picture, operators are time independent. Acting with S on both sides
of the Schrödinger equation we get

±i~∂t S|ψi = SHS −1 S|ψi, (6.56)

where the plus sign corresponds to unitary S, while the minus sign corresponds to antiunitary
S. The symmetry criterion, for symmetries that do not reflect t → −t, is

SHS −1 = ±H, (6.57)

while if time reflection is involved then

SHS −1 = ∓H. (6.58)

The upper sign is for a unitary symmetry operation, and the lower for an antiunitary one. If
we work in the basis where H is diagonal, we see that the cases with a minus sign are possible
only if the spectrum of energy eigenvalues is symmetric around zero. Very few Hamiltonians
have this property. In particular, any Hamiltonian for interacting particles, which has the
property that particles become free when they are far from each other, cannot have a spectrum
that is symmetric around zero. The free particle Hamiltonian is bounded from below and
its spectrum is continuous. All bound states of the particles, which have energies below the
bound of the free Hamiltonian, have a discrete spectrum.4
Thus, with Wigner, we argue that symmetries which involve time reflection are antiuni-
tary, while those which do not are unitary. Note that this does not require that time reflection
4
Up till now, our quantum formalism has shared with classical mechanics the principle that the choice of the
zero of energy does not effect physics. If we allowed antiunitary symmetries that preserve time orientation,
or unitary symmetries that reverse it, the zero of energy would acquire an absolute significance.
146  Quantum Mechanics

itself is a symmetry. Indeed, a variety of experiments on elementary particle physics suggest

that it is not. However, there is a rigorous theorem in relativistic quantum field theory,
which shows that T CP , the combined operation of time reflection (T), space reflection (P),
and the operation (C) which exchanges particles and their antiparticles, is a symmetry of
any model of local interactions, which is invariant under space-time translation and Lorentz
transformations.5
Apart from TCP, we can always think about symmetries, which are linear unitary oper-
ations that commute with the Hamiltonian. Choose, among all the symmetry operations of
a given model, a maximally commuting set, Ui . The fact that Ui commute with the Hamil-
tonian implies that we can diagonalize them all, and that the time evolution of the system
does not change their joint eigenvalues eiφi . That is to say, the phases eiφi are conserved
quantities, constants of the motion.
We can also see this connection between symmetries and conservation laws by looking at
the Heisenberg picture equations of motion. In the Heisenberg picture, states are time inde-
pendent, and operators whose definition is not explicitly time dependent, vary according to

~∂t O = i[H, O]. (6.59)

For the symmetry operators Ui , this implies that ∂t Ui = 0, which is the Heisenberg picture
form of a conservation law.
This connection between symmetries and conservation laws was first discovered in clas-
sical mechanics by E. Noether. Noether actually proved the connection only for symmetries
that depend on a continuous parameter a. Her proof is reviewed in Appendix C. A contin-
uous family of symmetries in QM, depending on a single parameter, is a family of unitary
transformations U (a).6 We can always choose another parameterization a → f (a), where f
is an invertible continuous function. However, if [U (a), U (b)] = 0 for any two different values
of the parameter, then there is a natural choice for the parameterization.
First note that U (a)U (b) is a symmetry, if the original unitaries are. It is unitary, and
commutes with the Hamiltonian. Let us choose the parameterization such that U (a)U (b) =
U (a + b). If that is the case, we can write U (a) = eiQa , where Q is a Hermitian operator,
if we are willing to make the extra assumption that the dependence on a is differentiable.
This assumption is valid for all known continuous symmetry groups that have been used in
QM. Q is known as the infinitesimal generator of the one parameter group of symmetries
U (a). Clearly, [Q, H] = 0 and Q is the quantum analog of Noether’s conservation law. The
Noether connection between symmetries and conservation laws is tighter in QM than in
5
The evolving cosmology in which we live, does not have these symmetries, but they are currently violated
by very small amounts, inversely proportional to the age of the universe, so there is a good approximate
theory of the world which does have these symmetries.
6
This discussion should remind the alert reader of our treatment of spatial translations in Chapter 3.
 Review of Linear Algebra and Dirac Notation  147

classical Lagrangian physics, because in QM, all operations are also potentially measurable
quantities. The idea that an operation is connected to a conserved quantity is natural in this
context.
The set of all unitary symmetries of a given quantum system forms a mathematical
structure known as a group. Every symmetry transformation has an inverse and the product
of two symmetries is always a symmetry because we can evaluate its commutator with the
Hamiltonian using the Leibniz rule. More precisely, in QM we have a representation of an
abstract group of symmetries by a group of unitary transformations in a Hilbert space.
The classification of all possible groups and their unitary representations is one of the great
problems of mathematics, and it has not been solved. However, mathematicians have found
out an enormous amount of information about this problem, much of it generated by examples
from physics. We cannot possibly do justice to this material at the level of this course.
Appendix D on symmetries summarizes a few of the most important facts and concepts.
Above we discussed a maximal set of commuting symmetry operators Ui for a given
system. There can be other symmetries, which do not commute with the Ui . The full set of
symmetry operations forms a group. If that group is nonabelian, i.e., contains elements that
do not commute with each other, then the minimal faithful7 representation of it has dimension
greater than 1. A representation of a group is called irreducible if the only operators which
commute with all the group representation matrices are multiples of the unit matrix. Since
the Hamiltonian commutes with all symmetry group elements, it must take on the same
value in every state of an irreducible representation. If the group is nonabelian, this leads
to degeneracies in the spectrum of the Hamiltonian. Without such a symmetry explanation,
the occurrence of such degeneracies would not be expected. A random Hermitian matrix will
not have any equal pair of eigenvalues. We will see examples of this below when we discuss
rotation invariance and the hydrogen atom.

6.9.1 Dynamical Symmetry Groups

The Hamiltonian itself is a symmetry generator. We introduced it to describe the invariance
of quantum dynamics under time translation. Our brief discussion of nonabelian symmetry
groups suggests that there could be an expanded concept of symmetry, in which the symmetry
group contains operators that do not commute with the Hamiltonian. Indeed, such a concept
of symmetry was introduced into physics by Galileo, when he argued that the equations of
physics (the language here is a bit anachronistic) looked the same in all inertial reference

7
A faithful representation is one in which the full group multiplication table is represented by nontrivial
matrices. An example of a nonfaithful representation is one in which every group element is represented
by the unit matrix. A slightly less trivial one is the representation of the group of permutations in which
every even permutation is represented by the unit matrix, and every odd permutation is represented by a
matrix with eigenvalues ±1.
148  Quantum Mechanics

frames. Inertial reference frames are systems of Cartesian space coordinates, which are moving
with respect to each other at constant velocity, x → x + vt. This transformation is called a
Galilean Boost. For a particle of mass m, the momentum changes under such a transformation
by an amount
p → p + mv, (6.60)
and so the particle’s kinetic energy changes. The Hamiltonian is not invariant. Nonetheless,
the equations of motion in the new frame are identical to those in the old, if there is no
external potential. If the system contains multiple particles and the potential depends only
on coordinate differences, then the equations of motion still look the same in both frames.
Furthermore, only the kinetic energy changes
1X 1
mi vi2 → M v2 + v · P, (6.61)
2 2
P P
where M = mi is the total mass and P = mi vi is the total momentum. In nonrelativistic
physics, both the mass and the momentum are conserved if the system is invariant under
spatial translation, so this law for the change of the Hamiltonian is universal and involves
only conserved quantum numbers. For very small v = δv, the change in the Hamiltonian is

δH = δv · P, (6.62)

while the change in momentum is

δP = M δv. (6.63)
If the action of Galilean boosts on a state is

|ψi → (1 + iNi δvi )|ψi, (6.64)

where Ni is a Hermitian operator, then the corresponding action on operators is

δO = −i[Ni δvi , O]. (6.65)

Thus,
Pi = −i[Ni , H] (6.66)
and
M δij = −i[Ki , Pj ]. (6.67)
Similarly
tδij = −i[Ki , Xj ], (6.68)
where X is the center of mass position coordinate, the canonical conjugate of the total
P2
momentum. We can find a solution of these relations by insisting that H = 2M + Hint , where
 Review of Linear Algebra and Dirac Notation  149

Hint depends only on relative coordinates and their conjugate momenta, as well as the spins
of the constituent particles, and
M Pi
Ni = Xi − t . (6.69)
~ ~

Note that Ni has explicit time dependence. In Exercise 6.7, you will verify that this makes it
time independent, despite the fact that it does not commute with the Hamiltonian. This is in
accord with Noether’s classical theorem, which relates the invariance of the classical action
under Galilean boosts, to a conservation law. The form we have written for the Galilean
boost operator corresponds precisely to the quantum version of Noether’s expression. Those
observant readers with long memories will also notice that we have completed the second
justification for the form we wrote for the free particle Hamiltonian.
Symmetry groups which include the Hamiltonian, but do not commute with it, are often
called Dynamical Symmetries, or Spectrum Generating Symmetries. Galilean boosts tell us
that center of mass motion decouples from the internal dynamics, and completely deter-
mine the energy momentum relation (also called the dispersion relation) for the center of
mass motion. The true symmetry of the world, to which Galilean boost invariance is only
a slow motion approximation, is the Lorentz group. In QM, this symmetry has profound
consequences for dynamics. It forbids the existence of interacting systems in which particle
number is conserved, and when combined with the notion of causality/locality (which says
that quantum interference should respect the restrictions imposed by a maximal propagation
velocity), it puts extremely strong constraints on the properties and interactions of elemen-
tary particles. We do not have time to study this in this first course in QM, but there are
a number of excellent textbooks on relativistic Quantum Field Theory [17], where you can
follow the story further.
There are a few other examples of dynamical symmetries in problems of physical interest.
Perhaps the most powerful is the Runge–Lenz–Pauli symmetry of the hydrogen atom, which
completely determines the spectrum in terms of group theory. There are also a variety of
models in one space and one time dimension with similar spectrum generating symmetries
[18].

6.9.2 Projective Representations of Symmetry Groups

Before leaving the topic of symmetry, we should touch on a peculiar aspect of symmetry
groups in QM, which has no classical analog. A group is determined by its multiplication law
g1 g2 = g12 . In QM, we have a unitary transformation U (g) for each group element and we
might have expected that

U (g1 )U (g2 ) = U (g1 g2 ), (6.70)

150  Quantum Mechanics

but this is too strong a requirement. Suppose instead that

U (g1 )U (g2 ) = U (g1 g2 )eiφ(g1 ,g2 ) , (6.71)

where φ is real. The fundamental predictions of QM are predictions for expectation values of
the form hψ|A|ψi, and these are all insensitive to the extra phase in the multiplication law.
So such projective representations of groups are allowed in QM.
One may ask the question, “Why do we care?” if the choice of phase factor in the pro-
jective representation does not affect physical predictions. The answer is that it does affect
physical predictions through the choice of allowed representations of the group. The most
important example of this is the group of three-dimensional rotations SO(3). The most obvi-
ous representation of this group is a three-dimensional vector V i , and one can make products
of these V i . . . V j which are called tensors of rank n, if there are n factors in the product.
On the other hand, we may recall the 2 × 2 Pauli matrices, which have a multiplication
table
σa σb = δab + iabc σc , (6.72)
which is covariant under rotations if σa transforms like a three-vector. Consider the unitary
transformation
θ
U (θ, e3 ) = ei 2 σ3 = cos(θ/2) + i sin(θ/2)σ3 , (6.73)
and compute
U † (θ, e3 )σa U (θ, e3 ). (6.74)
It is obvious that this does not change σ3 . Using the Pauli algebra, it gives

U † (θ, e3 )σi U (θ, e3 ) = cos2 (θ/2) − sin2 (θ/2)σi − i sin(θ/2) cos(θ/2)(σ3 σi − σi σ3 ), (6.75)

for i = 1, 2. This is
cos(θ)σi + sin(θ)ij σj , (6.76)
where the two-dimensional Levi-Civita symbol is defined in the obvious way.
We see that U (θ, e3 ) is a unitary transformation, which operates on the Pauli matrices
like a rotation around the 3 axis, with angle θ. In the Exercises, you will show that U (θ, ea ) =
θ
ei 2 ea σa , where ea is a three-dimensional unit vector, gives a rotation with angle θ around the
ea axis. So we have a two-dimensional unitary representation of the rotation group.
Well, not quite. Let θ = 2π. The corresponding rotation is equal to the unit matrix, and
indeed, the Pauli matrices are invariant under this transformation. However,
θ
U (θ, e3 ) = ei 2 σ3 = −1. (6.77)

So the multiplication equation R(π, e3 )2 = 1 of the rotation group is not satisfied in this
representation. Note, however, that this transformation U = −1 commutes with all other
 Review of Linear Algebra and Dirac Notation  151

elements of the group. In Exercise 6.8, you will prove that this implies that we have a
projective representation. In this case, the phases are just ±1.
This representation is actually an ordinary, nonprojective representation of another group,
the group of unitary 2 × 2 matrices with determinant 1, which is called SU (2). Writing a
general 2 × 2 matrix as a sum of the unit matrix and the Pauli matrices:

U = eiφ (n0 + ina σa ), (6.78)

where n0 is real, you will show in Exercise 6.10 that the unitarity condition is that na is real
and n20 + n2a = 1, while the determinant condition implies φ = 0. The unitarity condition is
solved by n0 = cos(α) and na = sin(α)ea , and we recognize both the relation to 3D rotations
and the fact that the SU (2) group has the geometry of a sphere in 4 Euclidean dimensions.8
The three-dimensional rotation group is gotten from SU (2) by identifying elements ±U with
each other.
A construction like this is quite general. If G is a group, the set of elements that commute
with everything else in the group is called Z, the center of the group. We can now define a
new group called G/Z (G mod Z) by identifying group elements related by g1 = Z12 g2 , where
Z12 is an element of the center. If G has continuous parameters, we can think of the group as
a space with topology. One of the interesting topological questions one can ask about a space
is whether there are closed curves, which cannot be contracted to a point. For example, on
a sphere, in any dimension, there are no such closed curves. On the other hand, on a two-
dimensional torus, the surface of a donut, there are two different kinds of noncontractible
closed loops, going around the donut in two orthogonal directions (Figure 6.1). If our group
G, like the group SU (2) has no noncontractible loops, the group G/Z will have contractible
loops. Namely, take an open curve in SU (2), which goes from a group element U to −U . In
SO(3) = SU (2)/Z2 , this is a closed loop, but it cannot be contracted because it connects
two different points in SU (2).

Figure 6.1 Noncontractible loops on a torus, but not a sphere.

8
One may suspect that there is a four-dimensional rotation group hiding somewhere here. Find it in Exercise
6.9.
152  Quantum Mechanics

There is an amusing little experiment you can do, to convince yourself that the rota-
tion group has a noncontractible loop in it. Fill a glass of water half full. Hold the bot-
tom in the palm of your hand, stick your arm out and try to rotate the glass around
a full circle, without spilling. You will find that your arm is twisted in a way that you
cannot get out of without either undoing the rotation or doing another 2π rotation. See
https://blogs.scientificamerican.com/roots-of-unity/a-few-of-my-favorite-spaces-so-3/ for an
illustration of this, and of a similar experiment with belts or long braids of hair. It is the
second option that is the surprise. This shows not only that there are objects for which 2π
rotations are nontrivial, but also that a 4π rotation is always trivial. We have discovered the
hidden SU (2) group in a classical experiment.
Objects that transform under rotations like the two by two matrix representation we have
just discussed are called spinors. They were first discovered by Pauli, and the mathematician
Cartan investigated the mathematical properties of spinors in any space dimension. Given
our little example, it might seem that no localized particle could transform like a spinor.
Your hand with its glass of water is sensitive to the 2π rotation because your arm is attached
to your body. For a point particle, one could imagine something similar happening if there
were some kind of infinite string emanating from it. The string would have to extend out
to infinity in order to avoid questions about what is attached to the other end. The string
could not carry any energy or momentum and other things should be able to go right through
it without noticing that it is there. Sounds like magic. Actually we will see in Chapter 15
that we can build mathematical models which behave exactly like particles attached to such
magical invisible strings. There are particles that transform like spinors and are sensitive to
2π rotations, and they all behave as if they have magical strings attached, a property which
manifests itself in what is called Fermi-Dirac statistics.

6.9.3 Examples
Let us go back to the example of the ammonia molecule, with which we began our exploration
of QM. We will learn in Chapter 11, that the meaning of picture (Figure 6.2) of this molecule
that we drew, is tied to an approximation in which we can think of the nuclei of the atoms
composing the molecule as stationary sources of electric field, sitting at points in space.

|+i 3 = |−i 3 =

Figure 6.2 Ammonia in the Born–Oppenheimer approximation.

 Review of Linear Algebra and Dirac Notation  153

We know that this is impossible for quantum particles, because of their position and
momentum cannot be simultaneously definite. However the commutator of the position and
velocity of a particle is
~
[X, Ẋ] = i , (6.79)
M
so if there is some sense in which M is very large, then this uncertainty is small. Thus, a very
massive particle can stay in place for a long time, without violating the uncertainty relation.
Of course, mass and time have dimensions, so we have to say “long (heavy) compared to
what?” In a molecule, the electron mass sets a mass scale and if we have fixed energy, it also
sets a time scale. The basic idea of the so-called Born–Oppenheimer approximation is that
one can freeze the nuclei at fixed positions over time scales that are long enough that the
electrons settle into their ground state in the external potential of the fixed nuclei. This then
creates a potential (the Born–Oppenheimer potential) for the nuclei since the ground state
energy will depend on their positions. The nuclei sit in harmonic oscillator ground state wave
functions localized within a distance that scales like an inverse power of M from the minima
of the Born–Oppenheimer potential.
In the ammonia molecule, the positions of the three identical hydrogen atoms form a
plane and the symmetry under exchanging these atoms suggest that the minima of the B-O
potential should sit at the vertices of an equilateral triangle. The most symmetric configura-
tion for the nitrogen is obviously at the center of the triangle, but there is no reason for it
to be in the same plane as the hydrogens, so it probably chooses to be above or below the
plane. The Hamiltonian for Coulomb interactions between electrons and nuclei is invariant
under spatial reflections, so there must be two equilibrium positions of equal energy.
The full symmetry group of the stationary molecule is thus a combination of cyclic permu-
tation of the three hydrogen atoms, and the reflection in the plane of the hydrogens. These
two types of transformation commute with each other. The Z3 permutation is a discrete
rotation in the plane, and rotations in the plane are invariant under reflections through it.
Call the transformation on the Hilbert space of states of the nuclei, which performs a 2π/3
rotation U3 and the reflection U2 . Both operators commute with the Hamiltonian, and with
each other. They satisfy
U33 = U22 = 1. (6.80)
As a consequence, we can consider eigenstates of H such that
2πik
U3 |k, pi = e 3 |k, pi (6.81)
U2 |k, pi = eπip |k, pi, (6.82)
where k = 0, 1, 2 and p = 0, 1.
These are the only symmetries of the Hamiltonian with nuclei in the Born–Oppenheimer
potential, but when we add the electrons back to the system, the U3 transformation is part
154  Quantum Mechanics

of the full rotational symmetry. The states with k 6= 0 have nonzero angular momentum and
it turns out that this gives them higher energy. You can understand this by remembering
the classical dependence of energy on angular momentum, and adding to that the fact that
in QM, bounded motions have discrete eigenenergies. We will compute the rough size of
rotational levels of molecules in Chapter 11. As a consequence, the lowest energy states are
labelled by the two valued quantum number p. In terms of the notation of Chapter 2, these
are the states
1
|±i1 = √ (|+i3 ± |−i3 ) (6.83)
2
and the 3 index can now be thought of as the direction perpendicular to the plane of the
hydrogen atoms in the ammonia molecule. The reflection symmetry operator is precisely the
Pauli operator σ1 , so these are eigenstates of σ1 .
The general theory of symmetry shows us that these two states are eigenstates of the
Hamiltonian, because symmetry transformations can always be diagonalized simultaneously
with the Hamiltonian and the distinct eigenstates of the symmetry operator σ1 , |±i1 , form a
complete basis for the space. This implies that the Hamiltonian is a function of the symmetry
operator
H = a + bσ1 . (6.84)
We have just rederived our theory of the low-energy states of the ammonia molecule, using
symmetry conditions alone.
Another simple example of the use of symmetry in QM is the theory of a particle moving
on a plane, under the influence of a potential that depends only on the radial distance from
the origin. The Hamiltonian is
1
H = (P12 + P22 ) + V (R), (6.85)
2
q
where R ≡ X12 + X22 . Introduce Z = √12 (X1 + iX2 ) ≡ Reiφ . The system is invariant under
the rotation Z → eiα Z, for any α. Potentials, which preserve this symmetry, are functions
only of Z † Z. We can also have potentials of the form

VN = f (Z N , Z † Z) + h.c., (6.86)

where h.c. refers to the Hermitian conjugate operator. Z N will be invariant under a rotation
by angle α if
eiαN = 1, (6.87)
which is solved by the N -th roots of unity
2πik
e N ,
 Review of Linear Algebra and Dirac Notation  155

with k = 0, . . . N − 1. Systems with such a potential are said to be invariant under the group
ZN of N -th roots of unity (Exercise 6.15 asks you to prove that this set of complex numbers
2πi
forms a group under multiplication). This group is generated by the single element e N , which
means that every element of the group is a power of that element. Such groups are called
cyclic groups. They are the building blocks for the set of all finite abelian groups.
Given a group of symmetries in QM, one always asks how vectors in the space of states
transform under the group. For the group ZN , this classification is easy to do. We have an
infinite dimensional space of invariant wave functions ψ(z N , z ∗ z). (Note that we are using
small letters here. When speaking about the potential, which is an operator, we wrote it as
a function of the operator Z. Here we are describing states in terms of their wave function
in the “basis” where Z is diagonal. z is the eigenvalue of Z.)
We can also have wave functions of the form ψk = z k ψ(z N , z ∗ z), which transform as
2πik
ψk → e N ψk ,

under the generating element of ZN . The action under the other transformations is just
obtaining by iterating this equation. In Exercise 6.16, you will use Fourier’s theorem for
periodic functions to prove that this is the most general transformation law allowed. That
is, any wave function can be written as a linear combination of wave functions of one of
the types listed above. This is an example of a general theorem about QM representations
of groups which are compact. That term refers to groups for which the set of parameters
necessary to specify the most general group element is a closed and bounded subset of a real
vector space of finite dimension. Examples of compact groups are groups with finite numbers
of elements and groups of rotations or unitary transformations in any number of dimensions.
The action of the group ZN is also useful for describing the motion of particles on the
surface of a cone. You can make a cone by cutting an angular wedge out of a sheet of paper,
and then gluing the edges together (Figure 6.3).
A circuit around at fixed distance from the singular apex will traverse an angle less
than 2π. When that angle is 2π/N , we can get the same figure by identifying points on the

Figure 6.3 Making a cone by gluing.

156  Quantum Mechanics

original sheet under the transformations of ZN . It follows that, for cones with these opening
angles, the physics of the motion is obtained from physics on the plane, by simply restricting
attention to quantities invariant under ZN . That is, the Hilbert space of motion on the cone,
is the subspace of the Hilbert space of motions on the plane consisting of those wave functions
that are invariant under ZN .
In this context, the group ZN is not really a symmetry group, because every state in
the physical Hilbert space is invariant under it. Rather the group is there to show us that
the description of motion on the cone in terms of motion on the plane is redundant. Points
on the plane differing by a ZN transformation are just copies of points in the fundamental
wedge with angles between 0 and 2π/N .
The redundant description is useful. If you’d been asked to construct a description of
motion on a cone, you might have started by trying to solve the motion on some smooth
curved two dimension surface and then taken a limit where the curvature went to zero
everywhere except at one point, where it became infinite. For the discrete set of opening
angles for which the trick works, it is much simpler to solve the problem on the plane,
and then project out the solutions that are not invariant under ZN . This kind of use of
a symmetry group goes by the name of gauge symmetry because it is reminiscent of the
redundancy familiar from Maxwell’s electrodynamics. In electrodynamics, we describe the
electric and magnetic fields by
E = Ȧ − ∇φ, (6.88)

B = ∇ × A. (6.89)

These physical fields are left invariant if we subject the vector potential A to a gauge trans-
formation9
A → A + ∇χ, (6.90)

with χ̇ = 0. It is convenient to use the vector potential to solve Maxwell’s equations, but all
physical quantities are invariant under the gauge transformation. Gauge Symmetries have
turned out to be the basis of the standard model of particle physics. We will encounter
them again in Chapter 15, where we will use a Z2 gauge theory to give a more profound
understanding of Fermi Statistics. Gauge symmetries are also crucial for understanding exotic
particles called anyons that can only propagate in a plane. Finally, note that if we describe
multiparticle states of identical particles in terms of many body wave functions, instead of
quantum fields, then we must treat the permutation group Sn , which exchanges particles,
as a gauge symmetry or redundancy. Every state must be invariant under the subgroup
An of even permutations. Fermion states are odd under the Z2 factor group Sn /Zn of odd

9
This is not the most general form of gauge transformation. We use the special case to illustrate the analogy.
 Review of Linear Algebra and Dirac Notation  157

permutations (every odd permutation can be written as some particular transposition times
an even permutation).
As our final example of symmetry groups, we note that many molecules contain a plane10
which is the form of a regular N -gon, and the Hamiltonian of the model is invariant under
the geometrical symmetries of the polygon. To understand what these are, we denote the
2πik
positions of the vertices as complex numbers zk = Re N . Then the operations
2πik
zi → e N zi = zi+k , for all i,

and
zi → zi∗
are symmetries of the polygon. The addition is modulo N arithmetic. Geometrically, com-
plex conjugation is a reflection around the horizontal axis in the plane (Figure 6.4). In
Exercise 6.17, you will show that these operations form a group with 2N elements, called the
dihedral group, DN . If we denote the multiplication by the k-th root of unit Sk and complex
conjugation by C, then the 2N elements are Sk and Rk = CSk . Any product of two of these
operations gives another one on the list, and each operation has an inverse.

z∗

Figure 6.4 Complex conjugation.

10
The idea that a molecule has a fixed shape, is an approximate statement in QM, related to the fact that
the masses of nuclei are at least two thousand times the mass of the electron. As a consequence, the
uncertainties in nuclear positions are much smaller than those of electrons, and low-energy excitations
of the molecule can be treated as small perturbations around a picture where the nuclei are classical
particles at rest. We will explore this Born–Oppenheimer approximation in more detail in Chapter 11.
158  Quantum Mechanics

In QM, each of these symmetry transformations should be represented by a unitary

transformation in a Hilbert space. Given any two such representations, we can form another
one by making block diagonal matrices with each block being one of the representations.
This is what we called the direct sum above. Obviously, the symmetry restrictions on energy
eigenvalues will apply separately to each block, so it is sufficient to study only irreducible
representations, which cannot be decomposed into such blocks. For the dihedral groups, the
simplest such representation can be obtained by the following geometrical trick.
If we think of the vertices of the polygon as points in a two-dimensional plane, then the
symmetries are represented as two-dimensional real matrices. Let us do the example of D3 ,
the dihedral group of the equilateral triangle. The vector representations of the vertices are
!
1
z0 = ,
0

!
cos(2π/3)
z0 = ,
sin(2π/3)
!
cos(2π/3)
z0 = ,
− sin(2π/3)

In Exercise 6.18, you will write the six elements of the dihedral group as two by two real
matrices. Since the matrices preserve the length of vectors, they satisfy M T M = 1. Now you
can let the same real matrices act on a complex two-dimensional Hilbert space. Since they
are real, they satisfy M † M = 1, which means that they are unitary transformations. So we
have constructed a two-dimensional unitary representation of the group.
To examine the possibility of more general unitary representations, we note that the
transformations Sk = S1k all commute, so we can diagonalize all of their unitary representa-
tives by diagonalizing U (S1 ). The multiplication law of the group tells us that U (S1 )3 = 1,
so its eigenvalues can only be third roots of unity. Now note that the group multiplication
law also implies
CS1 = S−1 C = S2 C = S12 C. (6.91)

Acting with both sides of this equation on an eigenstate |ωi, we get

ωC|ωi = S12 C|ωi, (6.92)

so C|ωi is an eigenstate of S12 with eigenvalue ω. This implies that it is an eigenstate of

S1 with eigenvalue ω ∗ . The whole Hilbert space is a direct sum of spaces with eigenvalues
2πi
1, e± 3 . The transformation U (C) takes the eigenspace with eigenvalue zi into that with
 Review of Linear Algebra and Dirac Notation  159

eigenvalue zi∗ . This means that it commutes with U (S1 ) in the z0 eigenspace and acts like
the block matrix !
0 1
, (6.93)
1 0
mapping the two other eigenspaces (which, we conclude, must be of equal dimension since
U (C) is unitary (Exercise 6.19) ) into each other. Thus, on the subspace of complex eigen-
values of U (S), we have !
0 1
U (C) = (6.94)
1 0
and !
e2πi/3 0
U (S) = −2πi/3 . (6.95)
0 e
These are block matrices, with any finite block size B. In Exercise 6.20, you will show that
one can rearrange rows and columns so that this is equivalent to a direct sum of B identical
copies of the representation
2πiσ3
U (C) = σ1 , U (S) = e 3 . (6.96)

Rearrangement of rows and columns is implemented by a permutation matrix, which is a

special example of a unitary matrix, so we say that every representation with complex eigen-
values is unitarily equivalent to a direct sum of copies of the two-dimensional representation.
For the subspace where U (S) = 1, we can find two solutions of C 2 = 1, so we have a num-
ber of copies of one-dimensional representations, in each of which U (S) = 1 and U (C) = ±1.
However, the representations in which U (C) = 1 are not faithful representations: two different
group elements are represented by the same unitary matrix. To make faithful representations,
we combine the one-dimensional representations into two-dimensional representations, where
U (S) is the two-dimensional unit matrix and U (C) = σ3 .
The symmetry constraints on the physics of a system invariant under the group D3 are
simply that the Hamiltonian commute with every symmetry generator. This is equivalent to
the statement that it is proportional to the unit matrix in every irreducible representation
of the group. In the representations with complex eigenvalues of U (S), where the nonabelian
character of the group is preserved, this implies a degeneracy of eigenvalues. Conversely, if
a system exhibits an exact degeneracy of eigenvalues, then the group of unitary transforma-
tions in the degenerate subspace commutes with the Hamiltonian, and is a symmetry of the
problem. The only thing to be wary about in using this general statement is the following.
For most systems, the Hamiltonian is constructed in terms of some simple operators, like
positions and momenta of particles. It is complicated to find its eigenvalues and eigenstates.
Thus, the unitary symmetry operations coming from a degeneracy may not have a simple
160  Quantum Mechanics

expression in terms of the fundamental variables. In this case, we would consider the degen-
eracy accidental, and guess that a small deformation of the Hamiltonian written in terms of
position and momentum variables would entirely remove the degeneracy.
A similar remark should be made about the group of symmetries defined, for any Hamilto-
nian, by unitary operators diagonal in the same basis as the Hamiltonian. For those students
who know advanced classical mechanics, these are the analog of the action and angle variables
of classical mechanics. We have not emphasized this in our discussion of classical mechanics,
but one can phrase the Hamilton–Jacobi equation as a technique for finding a complete set of
conservation laws of any Hamiltonian system. The angle variables are conjugate (in the Pois-
son bracket sense) to the conservation laws and basically have a harmonic time dependence.
For generic systems, the conservation laws are extremely complicated functions of the origi-
nal position and momentum coordinates, and finding them is tantamount to finding all the
complicated solutions of the equations of motion. Symmetry transformations/conservation
laws are only useful in either classical or quantum mechanics, if they are relatively simple
functions of the fundamental variables out of which the Hamiltonian is constructed.

6.10 EXERCISES
p
6.1 Prove the Schwarz inequality |hv|ui| ≤ |hu|uihv|vi|.

6.2 Prove that if the diagonal matrix elements hs|A|si of an operator vanish for every state
in the Hilbert space, then the operator is equal to 0.

6.3 If two operators do not commute, then eA+B 6= eA eB . The corrections to this formula
are written in terms of two complementary results, the Baker–Campbell–Hausdorff
formula and the Zassenhaus Lemma (the two differ in writing the left-hand side of the
equation as the right-hand side plus a series of corrections), or vice versa, and they
involve multiple commutators of A and B. Derive the first two terms in the formula
eA+B = eA eB eC1 eC2 . . . where Ck involves k fold commutators of A and B.

6.4 Let A(t) be a time-dependent operator. Write the Karplus–Schwinger formula express-
d A(t)
ing dt e in terms of dAdt . A convenient trick is to introduce a second parameter
W (s, t) = esA(t)
and write a differential equation for ∂W
∂s .

6.5 The Gram–Schmidt orthogonalization procedure. Start with a nonorthonormal basis

|ai i, with hai |aj i = Cij . We want to construct an orthonormal basis, with |e1 i =
|a1 i
ha1 |a1 i1/2
, one of the basis vectors. The basic idea is to choose a second basis vector via
the formula
|e2 i ∝ |a2 i − |e1 ihe1 |a2 i.
 Review of Linear Algebra and Dirac Notation  161

The proportionality constant is determined so that |e2 i has norm one. We continue
in this fashion, subtracting from |ak i all its projections on |ei i with i < k and then
normalizing the resulting vector. Express the coefficients in the expansion of |ei i in
terms of |ai i in terms of quantities derived from the matrix Cij .

6.6 In an N -dimensional Hilbert space, consider a unitary operator satisfying U N = 1,

2πik
with nondegenerate spectrum. Its eigenvalues are e N for k = 1 . . . N . Denote the
corresponding eigenstates by |ki. Let V be the cyclic permutation operator that takes
2πi
|ki → |k+1 (mod N )i. Show that V is unitary, and satisfies V N = 1 and U V = V U e N .
Show that any operator in the space can be written in the form

N
X −1
A= akl U k V l .
k,l=0

6.7 Take the limit N → ∞ in Exercise 6.6. To be more precise, consider states of the form
2πk
|f i = fk |vk i, where V |vk i = e N |vk i, and |fk |2 = 1 and use Fourier’s theorem on
P P

Fourier series to show that these converge to periodic functions on a circle z = eiθ with
0 ≤ θ ≤ 2π. U converges to the multiplication operator by eiθ and V k → eiαpθ , where
α = Nk and pθ = 1i ∂θ . Show that the spectrum of pθ consists of all the integers.

6.8 The circle in the previous exercise had radius 1. θ was a dimensionless angle variable.
If we want to understand how things depend on the radius of the circle, we should
introduce a variable with dimensions of length. We write θ = 2π X R . Let us choose the
range of θ to be −π ≤ θ ≤ π so that − R2 ≤ X ≤ R2 . Similarly, we write pθ = KR. K
makes infinitesimal shifts in the variable X. The eigenvalues of K are integers divided
by R. In the limit R → ∞, they become a continuum. Show that in this limit, we
recover the Hilbert space of a free particle propagating on an infinite line, with K and
X playing their usual role.

6.9 Consider the interval [−1, 1] and a nonnegative function on it, µ(x),
R1
which you can take
to be piecewise continuous. Define the moments of µ by Mn = −1 µ(x)xn . Show that
you can find polynomials Pk (x) for all nonnegative k, such that
Z
µ(x)Pk (x)Pl (x) = δkl .

Do this explicitly for k = 0, 1, 2, 3, and show that the coefficients are expressed in terms
of the moments. Argue that the same will be true for all k. This exercise is an infinite
dimensional analog of Gram–Schmidt orthogonalization.
162  Quantum Mechanics

x2a = 1.
P
6.10 Let xa , a = 0 . . . 3 be coordinates of a point on the sphere in four dimensions a
Define
U = x0 + ixa σa ,
where σi are the Pauli matrices. Show that U (xa ) is unitary and has determinant equal
to one. Conversely, show that every 2 × 2 matrix with these properties has this form
(recall that the unit matrix and the Pauli matrices form a basis in the space of all
complex two by two matrices).

6.11 Evaluate
U † (xa )σi U (xa ) = Rij (xa )σj .
That is, show that the unit operator does not appear on the right-hand side. Show that
the 3 × 3 matrix Rij is real and satisfies RT R = 1, so that it is a rotation matrix.
This is a mapping of the group SU (2) of 2 × 2 unitary matrices into the group SO(3)
of three-dimensional rotations. Show that both U = ±1 are mapped into the identity
rotation.

6.12 Prove that if a Hamiltonian H is bounded from below H ≥ k, then any diagonal matrix
element of the resolvent operator
1
Ds (z) = hs| |si,
z−H
has an integral representation

ρ(w)
Z
Ds (z) = ,
C 2πi(z − w)

with ρ ≥ 0. The contour C encircles the spectrum of H, which lies on the interval
[K, ∞], in a counterclockwise manner.

6.13 Schur’s Lemma: Consider a subalgebra A of operators on a finite dimensional Hilbert

space, which acts irreducibly. That is, there is no proper subspace of the Hilbert space,
apart from the zero vector, which is left invariant by the action of A. If C is an operator
such that C and C † commute with every element of A, Schur’s lemma states that C
is proportional to the unit operator. Alternatively, if the Hermitian conjugate of any
operator in A is in A, then we need only assume that C commutes with A.

6.14 For diagonal operators, the identity

ln A
det A = etr
 Review of Linear Algebra and Dirac Notation  163

is obvious. Prove that it is true for all normal operators, using properties of the deter-
minant and trace. Now let us make a small change A → A + δA. Show that

det A → det A(1 + tr [δAA−1 ]),

to first order in δA.

6.15 Prove that the N -th roots of unity form a group under multiplication, generated by the
element with the smallest phase. The axioms for a group are that we have a multiplica-
tion rule for any two elements, which gives a third element as their product, that there
is an identity element which satisfies eg = g for every group element g, and an inverse
such that gg −1 = 1 = g −1 g, for every element g.

6.16 Prove that every function on the plane is a linear combination of functions of the form
z k fk , where fk is invariant under the group ZN and k = 0 . . . N − 1.

6.17 Show that the combined operations of multiplying a complex number by an N -th root
of unity, and complex conjugation, form a group with 2N elements.

6.18 Write the six elements of the dihedral group of the equilateral triangle as 2×2 matrices.

6.19 Show that unitarity of the matrix U (C) in a representation of the group D3 with
complex eigenvalues for U (S) implies that the dimensions of the two eigenspaces with
complex conjugate eigenvalues are equal.

6.20 Show that one can exchange rows and columns of the block 2B × 2B matrices,
U (C), U (S), representing the group D3 and write them as a direct sum of B copies
of the 2 × 2 matrices σ1 and σ3 .

6.21 Explain how the previous comment is related to the equation

A⊗B =B⊗A

for the tensor product of any two matrices. You will have to remember that the first
tensor product is defined by multiplying each matrix element of the matrix A, by the
matrix B.

6.22 Find, up to unitary equivalence, all unitary irreducible representations of the group D4 .

6.23 Consider a particle of charge q moving on a circle of radius r. If the circle encloses a
solenoid of magnetic flux F , show that after setting a bunch of parameters equal to 1,
the Lagrangian is
1
L = θ̇2 + αθ̇.
2
164  Quantum Mechanics

Construct the Hamiltonian and find its eigenfunctions and eigenvalues. Show that time
reversal symmetry is valid for this problem for two different values of α. One of them
is obviously α = 0, what is the other? Show that for the nonzero value of α the time
reversal invariant problem has a degenerate ground state.

6.24 The symmetry group of the problem in Exercise 6.23, thought of classically, is U (1)×T .
Show that when α is nonzero, the U (1) and T operations do not commute. The faithful
representations of this nonabelian symmetry group have more than one dimension. Show
that the degenerate ground state is a two-dimensional representation of this group.

6.25 Now add a general periodic potential that is a function of 2θ to the Hamiltonian. The
problem is no longer exactly soluble. Show that the nonabelian symmetry is still present
at the nonzero, T invariant, value of α. This means that the ground state is at least
doubly degenerate.
 CHAPTER 7

Rotation Invariance and the

Hydrogen Atom

7.1 INTRODUCTION
This chapter is devoted to the exact solution of the hydrogen atom, which is the basis for
much of our understanding of atomic physics. We first discuss rotationally invariant problems
in general. This leads to another Noetherian triumph: angular momentum is the generator of
rotations (just as the Hamiltonian and momentum operators generate time and space trans-
lations), and the symmetry determines all of its properties in quantum mechanics (QM).
In particular, this insight leads to the exact form of Bohr’s approximate formula for the
quantization of angular momentum. It also shows us that different angular momentum com-
ponents cannot commute with each other. This has the consequence that energy eigenstates
in rotationally invariant theories always come in 2l + 1 dimensional degenerate multiplets.
Finally, rotational invariance completely determines the form of the Schrödinger wave func-
tions, as functions of the angles of a spherical coordinate system, and reduces the problem
to an infinite set of one-dimensional Schrödinger equations. For hydrogen, we can solve all of
these equations in terms of functions called Laguerre polynomials. The hydrogen spectrum
has a degeneracy beyond that implied by angular momentum. This is a consequence of the
conservation of the direction of the semimajor axis of elliptical orbits, a quantity called the
Laplace–Runge–Lenz vector.
Finally, we will outline the explanation of Mendeleev’s periodic table in terms of the
hydrogen atom and the Hartree approximation for multielectron atoms.

165
166  Quantum Mechanics

7.2 UNITS AND SYMMETRIES

The hydrogen atom is modeled by a Hamiltonian

P2N P2 e2 Z
H= + e − . (7.1)
2mN 2me 4π0 |Xe − XN |
This Hamiltonian actually describes a whole family of one electron ions, with nuclear charge
Z and nuclear mass mN . The hydrogen atom is the case Z = 1, and mN = mproton ∼
2, 000 me . The potential depends only on the relative coordinate R = Xe − XN , so the
system is invariant under translations and the total momentum PT = PN + Pe , commutes
with the Hamiltonian and is conserved. If we go to the rest frame of the ion, then PT = 0,
and we can rewrite the Hamiltonian in terms of the relative momentum P = PN − Pe , which
satisfies
[Ri , Pj ] = i~δij . (7.2)
The Hamiltonian is

P2 e2 Z
H= − . (7.3)
2mR 4π0 |R|
The reduced mass is given by
me mN
mR = = me + o(me /mN ). (7.4)
me + mN
The correction to mR = me is less than 1 part in 2, 000.
If we introduce the Bohr radius by
4π0 ~2
aB = ∼ 5.29177 × 10−11 m (7.5)
me e2
and the Rydberg energy by
~2
|E1 | = = 13.6057eV ≡ ERydberg (7.6)
2me a2B

(the hydrogen ground state energy, like all bound state energies, is conveniently taken to be
negative. It is −1 in Rydberg energy units), then we can write a completely dimensionless
Schrödinger equation by defining R = aB X. We get
2Z
[−∇2 ψ − ]ψ = Eψ, (7.7)
r
√
where E is the energy in Rydberg units and r = X2 , measured in Bohr radius units
 Rotation Invariance and the Hydrogen Atom  167

We will begin by analyzing a more general spherically symmetric problem − Zr → V (r),

and then specialize to the Coulomb problem, which is exactly soluble. As usual, we first use
the three-dimensional rotational symmetry, to identify conservation laws, which commute
with the Hamiltonian, and then solve the Schrödinger eigenvalue problem with fixed values
of the conserved quantities.
Any rotation is a rotation in some plane, by some angle θ. It also has a sense: either
clockwise or counterclockwise. The set of all counterclockwise rotations in the [12] plane
is a one parameter group of symmetries, represented in QM by unitary operators eiθK12 ,
where −K21 = K12 ≡ K3 is a Hermitian operator called the infinitesimal generator of the
symmetry. θ goes from 0 to 2π. In writing K12 = K3 , we are using the geometric fact that
every plane in three dimensions has a unique axis perpendicular to it. Note that K3 changes
sign if we change the sense of rotation, but not if we do a reflection of all three coordinates.
More generally we define
Kij = ijk Kk , (7.8)
as the generator of infinitesimal rotations that take the i-axis into the j-axis. It is obvious
from this formula that the three components Kk should transform as a pseudovector under
rotations: a vector that does not flip sign under reflection. The fact that K1,2 transform as
the components of a two-dimensional vector under the rotation generated by K3 is expressed
via the equation
e−iθK3 K1 eiθK3 = cos θK1 + sin θK2 . (7.9)
For small θ, this says that
−i[K3 , K1 ] = K2 . (7.10)
A generalization of this formula, which works for rotations around any axis is just
[Ki , Kj ] = iijk Kk , (7.11)
where we have used the summation convention for the repeated index k.
It is easy to see intuitively why rotations do not commute with each other. The rotation
around the three-axis changes the one-axis into the two-axis, so you do not get the same
thing after applying it as you do before applying it. The real surprise in QM is that, for a
system invariant under rotations, we must have
[Ki , H] = 0, (7.12)
which, by the Heisenberg equations of motion implies that the pseudovector K is a con-
served quantity for rotation invariant systems. From classical mechanics, it is familiar that
angular momentum is conserved for spherically symmetry systems. Angular momentum is a
pseudovector, so it must be proportional to Ki . By dimensional analysis,
Li = ~Ki .
168  Quantum Mechanics

In principle, there could be a numerically constant in this equation, but as we will see,
choosing the constant equal to one gives us precisely Bohr’s rule for quantization of angular
momentum. In addition, you will prove in Exercise 7.1 that the classical formula

L=X×P (7.13)

or
Li = ijk Xj Pk (7.14)
gives precisely those commutation relations for Li .

7.3 IRREDUCIBLE REPRESENTATIONS OF THE COMMUTATION RELATIONS

Since the Li commute with the Hamiltonian, we will simplify the problem of solving the
Schrödinger equation by diagonalizing as many of the angular momentum operators as we
can before writing the eigenvalue problem. Here we come to one of the weird features of QM.
Since the angular momentum generators do not commute, we can diagonalize at most one of
the generators in any given state. In terms of probabilities, if we are in a state in which L3
has a precise value, then L1,2 are uncertain.
The square of the angular momentum

L2 ≡ L21 + L22 + L23 (7.15)

is invariant under rotations, and so must commute with all of the Li . You will verify this,
using the commutation relations, in Exercise 7.2. This means that we can find a complete
set of states |n, l, mi of the system in which both L2 and L3 are fixed numbers.

L2 |n, l, mi = ~2 l(l + 1)|n, l, mi, (7.16)

L3 |n, l, mi = ~m|n, l, mi, (7.17)

The object of the quantum theory of angular momentum is to find the values of l and m that
are allowed by the commutation relations. We do this by acting with the operators K1,2 on
these eigenstates, much as we found the spectrum of the harmonic oscillator. The additional
label n refers to the action of operators that commute with the angular momentum. In the
problem of a single particle in a spherically symmetric potential, a complete commuting set
of such operators consists of functions of the Hamiltonian. So in this problem, n will label
the different energy eigenstates for fixed values of l and m. Rotations are a symmetry of a
much broader class of problems, involving any number of particles. In these more complicated
systems, n might refer to many other conserved quantities, besides the energy. Our analysis
of the possible values of l and m will be valid in all of these systems.
 Rotation Invariance and the Hydrogen Atom  169

It will be simpler to work in terms of the dimensionless operators Ki . We have already

seen that K1,2 transform as a two-dimensional vector under rotations generated by K3 . We
can think of the two-dimensional plane as the complex plane, with complex coordinates

K± = K1 ± iK2 , (7.18)

K+ = (K− )† , (7.19)
in which case rotations act by K± → e±iθ K± . Using either this geometric intuition, or directly
from the fundamental commutation relations, we see that

[K3 , K± ] = [K3 , K1 ] ± i[K3 , K2 ] = i312 K2 ∓ 321 K1 = ±K± . (7.20)

In other words, K± are raising and lowering operators for the eigenvalue of K3 , just as
creation and annihilation operators raise and lower the eigenvalue of the harmonic oscillator
Hamiltonian.
The parallel with the harmonic oscillator is not exact, because we can compute

K+ K− = K12 + K22 + i[K2 , K1 ] = K2 + K3 − K32 (7.21)

and
K− K+ = K12 + K22 − i[K2 , K1 ] = K2 − K3 − K32 . (7.22)
In either order, these are products of an operator with its own adjoint, so the expectation
value of both the left- and right-hand sides of both equations, in any state, are positive.
Applying this to simultaneous eigenstates of K2 and K3 , we get

l(l + 1) ≥ m2 ∓ m. (7.23)

We know that K± raise and lower the eigenvalue, m of K3 by one unit, without changing
l. The only way we can satisfy the inequalities for all states is if there are minimum and
maximum values of m such that

K+ |l, mmax i = 0 = K− |l, mmin i. (7.24)

We then have

0 = hl, mmin |K+ K− |l, mmin i = l(l + 1) − m2min + mmin , (7.25)

0 = hl, mmax |K− K+ |l, mmax i = l(l + 1) − m2max − mmax . (7.26)

Combining these two equations, we find that

l(l + 1) = mmax (mmax + 1) = mmin (mmin − 1). (7.27)

170  Quantum Mechanics

The solution mmax = mmin − 1 of the second equality is incompatible with mmax ≥ mmin ,
so we must have
l = mmax = −mmin . (7.28)
Starting from |l, mmin i and acting p times with K+ , we raise m to mmin + p. For consistency,
this must stop at mmax , so
2l = mmax − mmin = p, (7.29)
with p a positive integer. We conclude that l is quantized in half integral units and m runs
between −l and l. Angular momentum is thus quantized, as Bohr hypothesized, but he got
two things wrong about the quantization rule. First, we find that L2 = ~2 l(l + 1), rather
than l2 , so Bohr’s rule is only right in the limit of large l.
The half integer rather than integer quantization can be eliminated, at least for orbital
angular momentum, by insisting that a rotation by 2π return every state to itself. The oper-
ator e2πiK3 must be the unit operator, which means that K3 must have integer eigenvalues.
This argument is too glib, as we will see below. In nonrelativistic physics, where one can make
a clear separation between the intrinsic spin of particles and their orbital angular momen-
tum, we can assign the funny behavior under 2π rotations to an intrinsic angular momentum
carried by a single particle. This separation is incompatible with relativity. We will sort this
confusion out in a little while, when we talk about spin.
In order to find the wave functions of the hydrogen atom, we need to use the representation
of the angular momentum operators in coordinate basis. This is usually done in spherical
coordinates, and we will record the angular momentum operators in spherical coordinates
now. The simplest is L3 , which simply shifts φ. It is thus

~
L3 = ∂φ . (7.30)
i
It is convenient to use complex combinations of the coordinates in the 12 plane

z = x1 + ix2 = r sin(θ)eiφ (7.31)

and its complex conjugate

z ∗ = x1 − ix2 = r sin(θ)e−iφ . (7.32)
The generators L± have to be proportional to e±iφ to satisfy the commutator [L3 , L± ] =
±~L± . Thus, they have the form

L+ = x3 ∂z∗ − z∂3 , (7.33)

L− = x3 ∂z − z ∗ ∂3 . (7.34)
 Rotation Invariance and the Hydrogen Atom  171

It is now a simple change of variables to show that

L± = ±~e±iφ (∂θ ± i cot(θ)∂φ ). (7.35)

We can now compute

L2 = L+ L− + L23 − ~L3 . (7.36)
In principle, we can compute the angular wave functions, which are called spherical harmon-
ics, by solving the differential equations

L2 Ylm = ~2 l(l + 1)Ylm , (7.37)

L3 Ylm = ~mYlm . (7.38)

The second equation is solved by

Ylm = eimφ Plm (θ),

and this turns the first equation into

sin(θ)∂θ [sin(θ)∂θ Ylm ] + m2 Ylm = −l(l + 1) sin2 (θ)Ylm . (7.39)

You should verify this as an unofficial exercise. This is called the associated Legendre equa-
tion and its solutions the associated Legendre functions. The equation came up in the 18th
and 19th centuries and the properties of these functions were studied extensively. You can
find tables and numerous recursion relations, integral representations, etc., in many online
resources.
Rather than following this time honored route, we will get to the Legendre functions by
using another representation of the sphere. Instead of thinking about spherical coordinates,
we can think about functions on the sphere as functions of a three-dimensional unit vector.
It is then obvious how the functions
na1 . . . nak (7.40)
transform under rotations. In particular,

Ka nb = iabc nc , (7.41)

where we use the summation convention.

There are two ways to use the unit vector representation of the sphere to make Legendre
functions, and we will start with the simplest way of constructing all of the Ylm for inte-
ger l. The space of all square integrable functions on the sphere is acted on by a unitary
representation of the rotation group U (R)f (na ) = f (Rab nb ), where R is a 3 × 3 rotation
matrix. We have just shown by abstract operator algebra, that there are subsets of functions
172  Quantum Mechanics

Ylm (n) = hn|l, mi which transform into themselves. For fixed l, there are 2l + 1 functions
so we say that we have a 2l + 1 dimensional representation of the rotation group. Under
rotations, its obvious that the subset of functions T̃ a1 ...al ≡ na1 . . . nal 1 for fixed l, transform
into themselves under rotations. If one contracts two of the indices of T̃ with the Kronecker
delta, one gets the T̃ with two fewer indices. This transforms into itself separately. So, for
example

1
T ab = T̃ ab − δ ab , (7.42)
3
transforms into itself. Note that it has five independent components, as we would expect
for a spherical harmonic of l = 2. The components that have fixed value of L3 = ~m are
obtained by taking complex linear combinations of components, n± = n1 ± in2 . That is T ++
has m = 2, T −− has m = −2, T 3± has m = ±1, and T 33 has m = 0. Now, for any l define,
recursively
X
T a1 ...al = T̃ a1 ...al − Cl [δ a1 a2 T a3 ...al − combinations]. (7.43)

The combinations are added and the constant Cl chosen to make sure that δ ai aj T a1 ...al = 0
for any pair of indices. The T ’s are called traceless symmetric tensors, and it is easy to
verify that the l-th T has 2l + 1 independent components. Up to normalization, the traceless
symmetric tensors are the spherical harmonics Ylm , which is to say the “column vector”
representatives of the states |l, mi in the basis where na are diagonal. Since any function on
the sphere can be written as a (in general infinite convergent) linear combination of the T̃ ’s
and therefore of the Ylm , we have found a basis of the whole Hilbert space. Note that each l
value occurs exactly once.
An easy way to do the counting of independent Ylm in the traceless tensor representation
is to work in terms of complex coordinates, n3 and n± = n1 ± in2 = e±iφ sin(θ). Using the
fact that n+ n− = 1 − n23 , we see that there is a complete set of independent components of
na1 . . . nal of the form Yl,±m = nm
± flm (cos(θ)), for 0 ≤ m ≤ l. The polynomials flm can be
determined in various ways. One is the orthonormality constraint
Z Z 1
∗
sin(θ)dθdφ Ylm Yln = 2π dx(1 − x2 )m flm (x)fln (x) = δnm . (7.44)
−1

Different people prefer different methods. Use the traceless condition or the orthogonality
condition at your own discretion. Of course, nowadays the easiest way to find a formula for
spherical harmonics is to look them up on the web. These analytical methods can be reserved
for those times when you want to calculate a Ylm after an electromagnetic pulse or cyber
attack has disabled the internet.
1
We are raising the indices here to avoid double subscripts.
 Rotation Invariance and the Hydrogen Atom  173

Now we want to rederive the same results using operator methods. Most people will find
these more cumbersome, but they are useful as an introduction to techniques for general
symmetry groups, where there are no obvious analogs of the simple functions T̃ a1 ...al , which
jump started the discussion of the previous paragraphs. These operator methods are most
useful for low values of |m − l|.
We decompose Ka into K3 and K± as in our derivation of the spectrum. The analogous
decomposition of na 2 is into n3 and n± = n1 ± in2 , and n23 = 1 − n+ n− . n± are simply the
variables z and z∗ introduced above, with the factor of r removed. It is then simple algebra
to show that
K± n± = 0, K± n∓ = ±2n3 (7.45)
and
K± n3 = ∓n± . (7.46)
We can write all monomials in components of na in terms of either

Zpq ≡ np+ nq−

or
n3 Zpq ,
since even powers of n3 are just polynomials in n+ n− . The Ka operators are all first order
differential operators, so they satisfy Leibniz’ rule when applied to a product of two functions:
Ka (f g) = Ka (f )g + f Ka (g). We conclude that

K+ (Zpq ) = 2qZp(q−1) n3 , (7.47)

K− (Zpq ) = −2pZ(p−1)q n3 , (7.48)

K+ (Zpq n3 ) = 2qZp(q−1) (1 − n+ n− ) − Z(p+1)q , (7.49)
K− (Zpq n3 ) = −2pZ(p−1)q (1 − n+ n− ) + Zp,(q+1) . (7.50)
The action of K3 on any of these functions, just multiplies it by p − q.
We would like to find polynomials, linear combinations of Zpq and Zpq n3 , which have
fixed values of K2 and K3 . These functions are the spherical harmonics Ylm (na ) = Ylm (θ, φ).
They are the coordinate basis representatives of the abstract states, |lmi. Recall that

n3 = cos(θ),

n± = sin(θ)e±iφ .
2
We have switched back to lowered indices here because we no longer have to write expressions that would
generate double subscripts.
174  Quantum Mechanics

These are in fact the three components of Y1m , as you can verify by specializing the general
analysis below to l = 1.
For general l, we start by looking for Yll , which satisfies K+ Yll = 0, and K3 Yll = lYll . The
obvious solution is
Yll ∝ Zl0 , (7.51)
and it is unique. The proportionality constant can be calculated by integrating |Zl0 | over the
sphere, since Yll is normalized to 1:
Z Z π (2l)!
dΩ|Zl0 |2 = 2π sin2l+1 (θ)dθ = 2π 2 .
0 22l (l!)2
Thus,
(2l)!
Yll = [π 2 ]−1/2 sinl (θ)eilφ . (7.52)
22l−1 (l!)2
A similar analysis shows that

(2l)! −1/2 l
Yl−l = [π 2 ] sin (θ)e−ilφ . (7.53)
22l−1 (l!)2

Our general discussion of angular momentum now tells us that

l−m
Ylm ∝ K− Yll , (7.54)

and we can carry out the differentiation by using the formulae above for the action of K−
on Zpq × (1, n3 ). When l − m is small, this is quite quick and efficient. On the other hand,
when m is close to −l we can instead start from Yl−l ∝ Z0l , and apply powers of K+ . Either
method generates all of the Ylm , but it is less tedious to apply one at the top of the m
spectrum and the other at the bottom. Another useful observation that comes from this
analysis is that, for any l and m = 0, Zpp = (n+ n− )p = (1 − n23 )p , so that Yl0 is a polynomial
in n3 = cos(θ). These polynomials are proportional to the Legendre polynomials [19] and can
be computed simply by requiring that they form an orthonormal basis of polynomials on the
interval [−1, 1].
The proper normalization of all spherical harmonics can be obtained by computing
l−m l−m
hll|K+ K− |lli. (7.55)
l−m
This is the squared norm of the state |K− |lli. The trick for doing this computation is to
note that
l−m l−m l−m−1 l−m−1
hll|K+ K− |lli = hll|K+ (K2 − K32 + K3 )K− |lli. (7.56)
 Rotation Invariance and the Hydrogen Atom  175

l−m−1
The state K− |lli is an eigenstate of K2 with eigenvalue l(l +1) and of K3 with eigenvalue
m + 1, so this is equal to
l−m−1 l−m−1
(l(l + 1) − (m + 1)2 + m + 1)hll|K+ K− |lli. (7.57)

We can keep on using this trick until we eliminate all of the raising and lowering operators
and get back to the normalized state |lli.

7.4 ADDITION OF ANGULAR MOMENTA

Suppose we have two systems, which have angular momenta j1 and j2 . The theory of addition
of angular momenta tells us which representations of the rotation group there are in the
composite system made by combining them. This formula is useful for studying states of two
particles, but also for a single particle when we decompose its total angular momentum into
spin and orbital parts. The mathematical problem is that we have two commuting copies
(1) (2) (1) (2)
Ja and Ja of the angular momentum. We can form their sum Ja = Ja + Ja , and verify
that it also satisfies the angular momentum algebra. We want to know how the full Hilbert
space decomposes into irreducible subspaces of dimension 2j + 1, where Ja2 = ~2 j(j + 1).
The full Hilbert space consists of the states

|j1 m1 j2 m2 i; −j1 ≤ m1 ≤ j1 ; −j2 ≤ m2 ≤ j2 . (7.58)

It has dimension (2j1 + 1)(2j2 + 1). In the language of Chapter 6, it is the tensor product
of the two individual Hilbert spaces. The values of J3 for each of these states are simply
m1 + m2 .
The analysis begins by noting that there is a unique state with J3 = j1 + j2 . The theory
of angular momentum then tells us that there must be a whole multiplet of states with J3
between j1 + j2 and −(j1 + j2 ). Now look at states with J3 = j1 + j2 − 1. There are two
such states |j1 m1 j2 m2 − 1i and |j1 m1 − 1 j2 m2 i. One linear combination of them is the
state |j1 + j2 j1 + j2 − 1i from the highest angular momentum multiplet. The orthogonal
linear combination must be the top of a new multiplet with j = j1 + j2 − 1. Next we look at
J3 = j1 + j2 − 2. There are three such states

|j1 m1 − 2 j2 m2 i |j1 m1 j2 m2 − 2i |j1 m1 − 1 j2 m2 − 1i.

Two linear combinations are accounted for by states in the two multiplets we have already
found, while the third one indicates the presence of a new multiplet, with j = j1 + j2 − 2.
You should convince yourself that this pattern repeats: at each level J3 = j1 + j2 − k there is
always exactly one extra state, which is not contained in the multiplets uncovered at previous
levels. The process stops, however, when k = 2jmin , the smaller of the two angular momenta,
176  Quantum Mechanics

because past that point we cannot lower J3 by lowering mmin . Thus, the lowest angular
momentum we get to is jmin = |j1 − j2 |. Thus, the addition of angular momenta j1 and j2
leads to one copy of total angular momentum for each value of j between j1 + j2 and |j1 − j2 |.
You should convince yourself that the sum of the dimensions of these representations sum
up to the dimension of the tensor product space.
There are now two different bases for the tensor product space, |j1 m1 j2 m2 i and |j m i.
The scalar products between them define the matrix elements of the unitary transformation,
which takes one basis into another. They are called Clebsch-Gordon coefficients and you can
find tables of them on the World Wide Web.

C(j, m; j1 , m1 , j2 , m2 ) ≡ hj m|j1 m1 j2 m2 − 1i. (7.59)

7.5 THE HAMILTONIAN OF SPHERICALLY SYMMETRIC POTENTIALS

Now we want to rewrite the Hamiltonian of a spherically symmetric problem by exploiting
our knowledge of angular momentum and rotation invariance. The Hamiltonian commutes
with angular momentum, so it must be constructed out of r, ∂r , and L2 . The potential is
simply a function of r. The kinetic term −∇2 scales like r12 , so it must be a linear combination

B f [K2 ]
−∇2 = A∂r2 + ∂r + . (7.60)
r r2
Since it is a second-order differential operator, the function f [K2 ] must be linear, a simple
multiple of K2 . Any higher power of K2 would be a higher order differential operator, and
could not come from the Laplacian. We can evaluate A and B by letting the operator act on
a function that is constant on the sphere, for which K2 ψ = 0, ψ = ψ(r). Then
xi
∂i ψ(r) = ∂r ψ. (7.61)
r
It follows that
2
−∇2 ψ = −(∂r2 + ∂r )ψ, (7.62)
r
so B = −2 = 2A. To determine the coefficient of the angular momentum term, we apply
the Laplacian to the function xr3 = cos(θ). This is a vector under rotations (l = 1) and is
independent of r so
K2 2
2
cos(θ) = 2 cos(θ).
r r
On the other hand,
x3 1
∇2 = 2∇(x3 ) · ∇( ),
r r
 Rotation Invariance and the Hydrogen Atom  177

where we have used the fact that x3 ∇2 1r = 4πx3 δ 3 (x) = 0. This gives us
x3 1 2
∇2 = 2∂3 ( ) = − 2 cos(θ).
r r r
This tells us that the coefficient in the K2 term is 1.

2 K2
−∇2 = −(∂r2 + ∂r ) + 2 . (7.63)
r r
That is, we have derived the separation of variables of the Schrödinger equation for a spheri-
cally symmetric potential, and found an efficient method to construct the exact angular wave
functions Ylm . We also understand why these are simultaneous eigenfunctions of H, the third
component of angular momentum, and the square of the angular momentum. The angular
momentum eigenvalues are quantized, reproducing Bohr’s rule for large l, as we might have
expected for wave functions confined to a compact space.
The eigenvalue equation for fixed angular momentum is
2 l(l + 1)
[−∂r2 − ∂r + + V (r)]fl (r) = El fl (r). (7.64)
r r2
For single electron ions, the potential is just
2Z
V (r) = − .
r
Energies are expressed in Rydbergs and lengths in Bohr radii. We now exploit the fact
that before rescaling everything depended only on the combination Ze2 . So, we can do√our
computations for hydrogen, and get the answer for ions of higher charge by scaling e → Ze
in the answer.
The normalization condition for the wave function is
Z ∞
|fl (r)|2 r2 dr = 1. (7.65)
0

Near r = 0, if l 6= 0 the wave function behaves like ra , with

−a(a − 1) − 2a + l(l + 1) = 0. (7.66)

The solutions are a = l and a = −l − 1. Only a = l is normalizable. For l = 0, both solutions

are normalizable, but the expectation value of the Coulomb term in the Hamiltonian is
logarithmically divergent if we choose a = −1. This would mean that if such a behavior were
allowed in the Hilbert space the Hamiltonian would be unbounded from below.
This is a place where we have to be a little more careful about the definition of unbounded
operators. To prove Hermiticity of the Hamiltonian, we have to integrate the radial derivatives
178  Quantum Mechanics

by parts. However, if we choose the singular behavior at the origin for l = 0, the surface terms
do not vanish. In fact, they diverge. So, functions are in the part of the Hilbert space where
H is a Hermitian operator, only if they go to a constant at the origin for l = 0. One says
that normalizable wave functions with the r−1 behavior are not in the domain of definition
of the unbounded Hamiltonian, and so cannot be eigenstates.
If we look at r → ∞, then the equation for the logarithm, S of the wave function becomes
2
−∂r2 S − (∂r S)2 − ∂r S = E − V (r). (7.67)
r
The potential goes to zero at infinity, and √
we know E < 0 if we want a wave function that goes
to zero. The leading solutions are S = ± −Er, and only the negative sign gives something
normalizable. Now we know we are in trouble. The boundary condition at the origin fixes
the wave function, up to an over all constant, but this means that in general, there will be
a nonzero coefficient √
A(E)e −Er .
The only way to get a normalizable solution is to tune E to one of the zeroes of A(E). This
is the origin of energy quantization for bound states.
Notice
√ that for positive energy, the behavior is very different. The two solutions behave
±i Er
like e , which correspond to incoming and outgoing radial waves. These solutions
describe scattering of electrons off the nucleus. The electron comes in from infinity, is
attracted by the Coulomb potential, but has enough energy to escape to infinity. There
is no quantization condition.
Next we note that we can simplify our equation by defining fl ≡ ul /r. The point is that

∂r f = ∂r u/r − u/r2

and
∂r2 f = ∂r2 u/r − 2∂r u/r2 + 2u/r3 ,
so that
2
∂r2 f + ∂r f = ∂r2 u/r.
r
√
In terms of ul and the rescaled variable y = r −E, the Schrödinger equation is
l(l + 1) y0
∂y2 ul = [ − + 1]ul , (7.68)
y2 y

where y0 = √2Z . The normalization condition for ul is just

−E
Z ∞
dy|ul (y)|2 = 1. (7.69)
0
 Rotation Invariance and the Hydrogen Atom  179

For the bound states, our analysis of the behavior near 0 and ∞ suggests that we write
ul (y) = y l+1 e−y v(y), (7.70)
P∞ k
where v(y) = k=0 ck y has a power series expansion. The equation for v is
y∂y2 v + 2(l + 1 − y)∂y v + [y0 − 2(l + 1)]v = 0. (7.71)
When we examine the action of the various terms in this equation on a power y k , we see
that the first term lowers the power by 1, the second has a term that also lowers the power
by 1 and a term that leaves the power fixed, and the third term leaves the power fixed. The
zeroth order term in the series gives
2(l + 1)c1 = [2(l + 1) − y0 ]c0 .
More generally, we get
(2(k + l + 1) − y0 )
ck+1 = ck . (7.72)
(k + 1)(k + 2l + 2)
The series terminates for the discrete values y0 = 2(kmax + l + 1). If it does not terminate,
its large k behavior is
2
ck+1 = ,
k+1
k
or ck ∼ 2k! , which is the expansion of e2y , giving the bad asymptotic behavior at infinity. If we
define n = kmax + l + 1, then we find that the energy in Rydbergs is quantized as En = − n12
Rydbergs. Rydberg’s original experiments showed that the spectral lines of hydrogen fit the
formula ωmn = n12 − m12 when measured in units of Rydbergs/~.
A curious feature of this formula is that the spectrum has a degeneracy beyond that
expected from angular momentum analysis. For any l, we can get the same energy as long as
n ≥ l +1. The degeneracy of the nth level is thus n2 . The origin of this degeneracy is a hidden
conservation law of the hydrogen atom. It is well known that the bound classical orbits in a
Coulomb potential are ellipses, so that the semimajor axis of the ellipse is a conserved vector
throughout the motion. The energy of an orbit does not depend on the direction of the
semimajor axis, but it does depend on its magnitude, and is in fact completely determined
by it. This conservation law was pointed out by Laplace in the 18th century and Runge and
Lenz in the 19th century and is called the Runge–Lenz vector or the Laplace–Runge–Lenz
(LRL) vector. Pauli showed that the algebra obtained by combining the quantum mechanical
L–R–L vector with angular momentum, completely determines the spectrum of hydrogen.
We do not have space here to go into the details. See Exercise 7.4.
The wave functions of the hydrogen atom and one electron ions have the form
s
8Z 3 (n − 1 − l)! − y l 2l+1
ψnlm (r, θ, φ) = Ylm (θ, φ) e 2 y Ln−l−1 (y), (7.73)
2n4 [aB (n + l)!]3
180  Quantum Mechanics

with
2Zr
y= .
naB
The functions L2l+1
n−l−1 (y) are polynomials called the associated Laguerre polynomials. Our
normalization conditions for these polynomials differ from, e.g., the form you will find in the
Wikipedia article on Laguerre polynomials.
The recursion relation for the coefficients of these polynomials is

2(k + l + 1 − n)
ck+1 = ck . (7.74)
(k + 1)(k + 2l + 2)
Recalling that the Gamma function obeys

Γ(z + 1) = zΓ(z), (7.75)

we see that
2k Γ(k + l + 1 − n)
ck = Γ(k + 2l + 2). (7.76)
Γ(k + 1)
You will explore more properties of Laguerre polynomials in Exercises 7.7–7.9.
Some simple examples are the ground state and first excited state radial wave functions

Z 3/2 − aZr
f10 = 2( ) e B, (7.77)
aB
Z 3/2 Zr − 2aZr
f20 = ( ) (2 − )e B , (7.78)
2aB aB
Z 3/2 Zr − 2aZr
f21 = ( ) √ e B. (7.79)
2aB 3aB
Notable general features are that higher Z implies wave functions closer into the nucleus,
while higher l or n have the opposite effect. For angular momentum, this is just a consequence
of the repulsive angular momentum barrier, which dominates the Coulomb attraction at small
r, while for higher energy states, it is just a consequence of our scaling argument.
The energy differences between different levels can be measured by exciting the hydrogen
atom out of its ground state, by subjecting it to an external force. While our simplified
treatment finds all of the excited levels of hydrogen to be stable, this is no longer true
once we consider corrections to the Coulomb approximation for the effect of the interaction
between hydrogen and the quantized electromagnetic field. That interaction includes terms in
which the atom can make transitions to a lower level and emit a single photon whose energy,
according to the Einstein ~ω rule, corresponds to radiation of a given frequency. Multiphoton
emissions can also occur but the probability for them is suppressed by powers of the fine
 Rotation Invariance and the Hydrogen Atom  181

structure constant αem ∼ 10−2 . These processes are allowed by energy conservation, and
they are the main processes that occur when the hydrogen atom is in a relatively isolated
environment. In such circumstances, since the system of lower energy atom plus emitted
photons has more states3 than the excited atom, the higher excited states will decay. They
are only metastable (because the fine structure constant is small, so the probability of decay
is small).
The emission spectra from hydrogen and other atoms have an enormous number of both
practical and pure science applications. They are a primary tool for identifying materials, as
well as the primary tool for probing the distant universe. Indeed, it was Hubble’s discovery
of a systematic red shift in the spectral lines from distant objects, which led to our modern
picture of an expanding universe. The association of energy levels with spectra gives rise to
names for the energy levels associated with the visual signatures of the spectra they give rise
to. Thus, we have s(harp), p(rincipal) and d(iffuse) for the spectra associated with l = 0, 1, 2.
These are followed by f(undamental), g, and so on down the alphabet. This terminology for
angular momentum states has entered into the colloquial language of physics, in ways that
has nothing to do with spectra. Thus, you will often hear particle physicists talking about s or
p wave scattering amplitudes (see the next chapter), or condensed matter physicists talking
about d wave superconductors. The latter terminology is particularly amusing, because it
refers to the behavior of the superconducting condensate, an approximately classical field, in
the ground state of certain types of superconducting materials, rather than to the behavior
of a quantum wave function.

7.6 THE PERIODIC TABLE

We now want to sketch the QM of multielectron atoms. The key idea is to treat them in
an approximation where each electron can be thought of as propagating in a self-consistent
field generated by all of the others. This idea goes back to Hartree [20] who proposed to
calculate the self-consistent field as the sum of the nuclear Coulomb field and the Coulomb
field generated by the electrons, which he wrote as

1 1
Z
ψ ∗ (y)ψi (y).
X
VHartree = d3 y (7.80)
4π0 |(x − y)| i i

3
In statistical mechanics language, this is just the statement that the final states of atomic ground state
plus far away photon have higher entropy than the initial state of an excited atom with no photon. In few
body systems, we usually say that “the phase space of the final state is larger.” The point is that while
transitions from excited atom to less excited atom plus photon, and atom plus photon to excited atom,
occur with equal probability when the photon is near the atom, the first of these always occurs, while the
latter only occurs if an incoming photon gets close enough to the atom.
182  Quantum Mechanics

The wave functions are the individual electron wave functions, which we take to be the
~2
lowest eigenstates of the self-consistent Hamiltonian − 2m e
∇2 + VHartree , consistent with the
constraints of Fermi statistics, which we will discuss below.
It turns out that, in the limit that the nuclear mass is  me , the idea of a self-consistent
field is exactly valid,4 but Hartree’s approximation to it is not. Much work in the physics of
atoms and ions goes into calculating the correct form of the electron density profile, which
determines the self-consistent field and the ground state energy. What is remarkable is that
the self-consistent field method is valid, even though the idea of a multielectron wave function,
which is simply a(n antisymmetrized) product of single electron wave functions is not. The
true multielectron wave function for an atom has complicated multiparticle correlations.
Nonetheless, a picture of the atom in terms of single electrons in a self-consistent potential
is sufficient for understanding the gross structure of the periodic table, if not the numerical
details of atomic ground state energies. It is what we will use. Within such a picture, the
two key concepts are screening of the nuclear charge by electrons and Fermi–Dirac statistics.
Let us begin by explaining the latter. Imagine that there were no correlation at all between
the wave functions of individual electrons. Then we could put each electron into the ground
state of the self-consistent potential. We would expect ionization energies to be at least N
times the ionization energy for hydrogen for an N electron atom, and apart from that, all
atoms would behave in exactly the same way. There would be no explanation for chemistry
as we know it. Pauli hypothesized his exclusion principle to avoid this theoretical disaster.
If no two electrons are allowed to be in the same state, then there can be only two electrons
in the single particle ground state, once we include the fact that the electrons have two spin
states, and that the self-consistent potential is spin independent. Similarly, there can be two
electrons in each of the ψ20 and ψ21 states. Note that we can retain the hydrogenic labeling
for the states, even though the self-consistent potential is not the Coulomb potential, as long
as it is a spherically symmetrical potential. The label n labels the number of nodes of the
radial wave function (Exercise 7.15). We should no longer expect the ψ20 and ψ21 states to
be degenerate, since the potential no longer has the Runge–Lenz conservation law of the
Coulomb potential.
As we will see below, Pauli’s principle is just what we need to explain chemistry, but
what is the physical origin of his oracular declaration? The answer turns on the subtleties
of identical particles in QM. We have seen above that photons, the particles of which light
is composed, are excitations of a quantized field. As such, it makes no sense to make a
distinction between two states which differ by exchanging two photons, since they correspond
to the same field configuration. Multiparticle states of photons are automatically invariant
under exchange of the labels of individual photon states.

4
This is the fundamental theorem of Density Functional Theory [21]. We will discuss it briefly in Chapter 11.
 Rotation Invariance and the Hydrogen Atom  183

It is natural to be curious about whether the same could be true for other particles
and there is a paradox in classical statistical mechanics which indicates that it must be so.
When we open a partition between a box containing hydrogen and a box containing nitrogen
molecules, initially at some fixed temperature, T , the entropy of the system increases. This
change in entropy is called the entropy of mixing. The statistical mechanics argument for the
entropy of mixing goes through just as well if the two halves of the box contain identical
gases, but this contradicts experimental data. Gibbs concluded from this that one should not
count as distinct, states which differed only by the exchange of particle labels.
Fermi–Dirac statistics involves another subtlety, which is purely quantum mechanical.
The identical particle gauge symmetry says that each permutation must take any multipar-
ticle state into a physically identical one, but in QM the overall phase of a state cannot be
measured. It is a famous statement in group theory that every permutation can be written
as a product of transpositions, which simply exchange a particular pair of particles. The
square of a transposition is the identity, so if a state picks up a phase under transposition,
it must be ±1. Every permutation will thus multiply the state by either ±1. The group of
all permutations is a complicated object, and we have to ask whether there are consistent
ways to assign plus and minus signs to the action of all transpositions in such a way that the
transformations indeed satisfy the multiplication table of the group.
One obvious way to make a state invariant up to a phase under all permutations is to
assign (−1) to every transposition. Then every permutation which is a product of an even
number of transpositions gets a plus sign, and every odd product gets a minus sign. The
set of all even permutations is a subgroup of the permutation group called the alternating
group AN . Every odd permutation has the form P = Tij A, where A is in AN and Tij is a
transposition of any fixed pair of particles. It turns out that in space dimensions greater than
or equal to three, this assignment of phases is the only one compatible with the statistics gauge
symmetry, apart from choosing all phases equal to one, which is Bose–Einstein statistics. We
will get a deeper understanding of this when we study the Aharonov–Bohm effect.
The nontrivial even/odd choice is called Fermi–Dirac statistics, and the corresponding
particles are called fermions. Fermions obey Pauli’s exclusion principle. If we consider a two
fermion state, ψ(x1 , x2 ) = −ψ(x2 , x1 ) and so the state ψ(x1 )ψ(x2 ) is not allowed. More
generally, for an N fermion state, the simplest thing we can build from N independent one
particle wave functions ψi is the Slater determinant
1 (i1 ...in )
ψ(x1 . . . xN ) = det ψi (xj ) =  (j1 ...jm ) ψ[i1 ] (x[j1 ] ) . . . ψ[in ] (x[jn ] ).
n!
We have discussed Fermions and Fermi–Dirac statistics in Chapter 5, where we realized it
as an alternate way to quantize the creation and annihilation operators of a collection of
harmonic oscillators (aka a quantum field with a linear field equation). The field theory
approach is by far the most efficient way to deal with both bosons and fermions, but in this
184  Quantum Mechanics

chapter, we will stick to multiparticle language, since most older textbooks, much of the old
literature on QM and the Graduate Record Exam, use this language.
A crude estimate of the self-consistent potential is obtained by writing
1 Z Z
Z
Vef f (x) = [− + d3 y n(y)], (7.81)
4π0 |r| |r − y|
where
ψi∗ (x)ψi (x),
X
n(x) = (7.82)
is interpreted as the electron density at the point x. This is called the Hartree Approxima-
tion.5 The potential is called self-consistent because the single particle eigenfunctions, which
appear in its definition, are obtained by solving the Schrödinger equation with that poten-
tial. One proceeds by starting with an ansatz for the potential, solving the equation, and
computing the refined potential, then iterating. The procedure converges fairly rapidly.
As we said above, this crude approximation does not get the details right, but we can
use it to understand the qualitative properties of atoms. Fermi–Dirac statistics, in the form
of the Pauli principle, says that electrons must be added to higher and higher energy levels.
The level structure has bands, corresponding to the degeneracies of the hydrogen atom, but
split by the fact that the potential is no longer Coulomb. It is spherically symmetric, so the
exact angular momentum degeneracies remain, but the accidental degeneracies are split.
We can understand the nature of that splitting, by noting that the Schrödinger equa-
2 l(l+1)
tion for angular momentum l has a repulsive ~2m er
2 term in it, which keeps higher angular
momentum states further away from the origin. Also, states of higher n have smaller prob-
ability to be near the origin. As we fill in the lower energy levels, the electron density near
the origin screens the nuclear Coulomb potential, so the higher n levels, and the higher l
levels for fixed n feel a less attractive potential. Their energies are thus higher (they are
less bound). Thus, we should expect that, for a given n, the splittings of the exact Coulomb
degeneracies will make higher l states have higher energy. Indeed we may expect that some
of the bands of levels may overlap, with the higher l states at some value n being higher in
energy than the low l states at the next higher value.6
Note that all of the wave functions will have exponential tails, though falling at different
rates due to the fact that they see a screened value of Z. The last electron added will, at
long distance, see a Zef f = 1 attractive Coulomb potential, so it will still be bound. The
potential will deviate from the Coulomb form near the origin, but the wave function of the
last electron (for reasonably large Z) has low probability to be near the origin.
5
See Chapter 18 for an explanation of the relation of the Hartree approximation to variational estimates
of the many electron ground state energy. See Chapter 11 on Atomic, Molecular and Condensed Matter
physics for the description of Density Functional Theory, which replaces the Hartree approximation.
6
Recall also that the splitting between the nth and n + 1st level of the hydrogen atom goes to zero as n
gets large.
 Rotation Invariance and the Hydrogen Atom  185

The final idea that we need to understand the structure of the periodic table is the fact
that electrons actually have two spin states, and that, in the approximation we are using,
the energy is independent of the spin. We will talk more about electron spin below, but this
is all we need for now. For Z = 1, we of course have hydrogen. For Z = 2, the self-consistent
potential is no longer Coulomb, but it still has a nondegenerate ground state level, and we
can put both electrons in this level, by giving them opposite spin. The ionization energy of
helium is almost four full Rydbergs, somewhat reduced by the screening effect. This makes
helium very stable, which is to say, chemically inert. Atoms make chemical combinations by
trading electrons, either “swapping” as in ionic bonding, or “sharing” (this is called covalent
bonding). It costs a lot of energy to tear an electron away from a helium atom, less than can
be gained by swapping or sharing it with another nucleus.
At Z = 3, which is called lithium, one electron must go into the n = 2 shell. It will
choose the state with l = 0, because of the screening effect we have discussed. That is, while
the n = 2, l = 0, 1 levels of single electron ions are degenerate, the actual potential felt
by the electrons in the n = 2 level of Lithium differs from the Coulomb potential because
of the screening due to the electrons in the n = 1 shell. The lower angular momentum
states have probability distributions closer to the origin, and so have lower energy. Lithium
is chemically active because a Li+ ion puts the extra electron in the same l = 0 shell (because
of spin degeneracy), which has fairly low energy. It will tend to form ionic bonds with other
atoms. Z = 4, beryllium puts its fourth electron in that same shell. It is more deeply bound
than the Li+ ion because of the net charge on the latter. Beryllium is not a noble gas like
helium, because the l = 1, n = 2 shell is not split by very much from the l = 0, n = 2 shell, so
it is easier to excite Beryllium’s electrons out of their ground state. Boron, with Z = 5, now
has to have an electron in this shell, and because of the combined spin and orbital angular
momentum degeneracies, we can proceed up to Z = 10 (neon) before the n = 2 shell is filled.
Neon, like helium, is not terribly reactive, because the next highest energy level is quite far
away. On the other hand Z = 9, Fluorine is highly reactive because the F l− ion has low
energy and so is relatively stable. We can begin to see how the chemical activities of atoms,
and which ions they like to form, are explained naturally by the independent electron picture
of atomic structure.
The next row of the periodic table n = 3 is the place we first see dramatic effects
of electron–electron repulsion. The n = 3, l = 0 states are sodium and magnesium (quite
reactive, as expected), followed by six states (aluminum through argon) filling up n = 3, l = 1.
However, the repulsion has pushed the “expected” n = 3, l = 2 states up, so that they are
higher than the n = 4, l = 0 states. Thus, argon, with a relatively large gap to the next
available level, is a noble gas, and ends the third row of the periodic table with only 8
elements.
For the n = 4 row, we start by filling the l = 0 level with potassium and calcium, then
get to the 10 elements corresponding to the n = 3, l = 2 levels (scandium through zinc), and
186  Quantum Mechanics

finish with gallium through krypton, filling in the n = 4, l = 1 levels. Again, the gap to the
next set of levels makes krypton a noble gas, and we go on to the next row. You can find
the rest of the story in a number of atomic physics texts [22]. Reading those texts, especially
the more modern ones, you will find that our detailed quantitative understanding of the
physics of multielectron atoms is far from complete, and involves much more sophisticated
analysis than the simple model we have presented.
The states of multielectron atoms are described by chemists and atomic physicists using
a notation that combines the old spectral names for the different orbital angular momentum
states, with the independent particle picture of atoms, which we have just sketched. Thus,
for example, the ground state of neon is denoted by

(1s)2 (2s)2 (2p)6 . (7.83)

This means, “There are two electrons in the 1s orbital (opposite spin), two in the 2s orbital
and six in the three 2p orbitals.” The word orbital refers to the single particle wave functions
that are used in the Slater determinant approximation to the many electron wave function.
The labels on the orbitals refer to the fact that they are solutions to the self-consistent
single particle Schrödinger equation, with a given value of orbital angular momentum (the
s, p, d, f . . . label) and a given number of radial nodes. Often this is shortened by just writing
the symbol for the noble gas preceding the element in the periodic table, followed by the
orbital configuration of the electrons above that. Like many archaic notations, this one is
not so much a reflection of reality, but of the human desire to catalog things neatly. As such,
there are sure to be questions about it on the physics GRE.

7.7 THE SPIN OF THE ELECTRON

The suggestion that electrons have intrinsic angular momentum came from Uhlenbeck and
Goudschmidt [23], who invoked it to explain the splitting of levels of atoms in external
magnetic fields. As we have seen, the existence of spin, and the fact that the Hamiltonian
of nonrelativistic electrons commutes approximately with the spin operator, is crucial to
understanding the structure of the periodic table. In this section, we will outline Pauli’s
theory of the electron spin.
We already have all the mathematical apparatus to do this, since the spin of a single elec-
tron is the quintessential example of a two state system. Recall that the Pauli matrices satisfy

σa σb = δab + iabc σc . (7.84)

It follows that
σa σb σc
[~ , ~ ] = iabc ~ , (7.85)
2 2 2
 Rotation Invariance and the Hydrogen Atom  187

which are the angular momentum commutation relations for the generators Ja = ~ σ2a . The
Casimir operator of the representation is
3
~s j(j + 1) ≡ Ja2 = ~2 , (7.86)
4
so the spin j = 12 .
Before proceeding, let us examine more closely what we mean by the spin of a particle,
which comes down to a question of what we mean by a particle. Democritus defined atoms as
little indivisible objects, out of which everything else is built as a composite. Today, we know
that atoms themselves are composites, and even that the proton and neutron in the nucleus
of the atom are composites of particles called quarks. We discovered this first by finding that
the charge distributions inside atoms and protons and neutrons were not structureless, and
later by breaking these objects up into their constituents.7
A composite object will have many excited states in which there is orbital angular momen-
tum in the rest frame of the center of mass. One of these states would be the lowest energy
level, and it need not have zero total angular momentum. At an energy scale below that of
the excited levels, we will see an apparently pointlike object with a fixed value of j. This
is the definition of spin of an “elementary” particle. We can never know for sure whether a
particle is elementary or composite, because our experiments might not have enough energy
to excite the higher rotational levels. So, we define an elementary particle as one in which,
at our current level of accuracy, we see states at rest transforming in some particular “spin
j” representation of angular momentum.
This is not the end of the story though. Orbital angular momentum is quantized in integer
units, because wave functions are required to be periodic under 2π rotations. However, we
have seen that the mathematics of the rotation group allows for half integer spins. This is
allowed, if we make a rule that no physical operator can change the number of half integral spin
particles from even to odd or vice versa, because quantum expectation values and transition
matrix elements will be periodic. Note that one cannot obtain half integral spin from internal
orbital angular momentum of constituents. As it turns out, the spins of electrons and quarks,
those particles which have not yet revealed any inner structure, are half integral. One might
be tempted to say that these might be truly elementary, or that at least their structure is at
a scale where our notions of space might themselves break down.
Let us leave these heady speculations and come back to the theory of electron spin.
Electrons interact with the electromagnetic field because they are charged. They experience
Coulomb and Lorentz forces in the presence of external electric and magnetic fields. It was well
known to 19th century physicists that even neutral particles can interact with the magnetic
7
In the case of quarks inside nucleons, there is a confining force, a tube of chromoelectric flux, which does
not let quarks exist as free particles, but we have done experiments which let the constituent quarks in a
nucleon fly far apart from each other.
188  Quantum Mechanics

field by virtue of having magnetic multipole moments, the most important of which is the
dipole moment. Classically, a magnetic dipole moment can be modeled by a term in the
energy of the form
δH = µ · B, (7.87)
where µ is the magnetic dipole. It transforms as a vector under proper three-dimensional
rotations, and a pseudovector or axial vector (like the magnetic field) under reflections. In
QM, µ will become an operator. Throughout most of this book, we will make the assumption
that the electromagnetic field is in a coherent state and replace the field operator B by
the classical field that parametrizes that coherent state. We will neglect the small quantum
fluctuations of the field around this classical value. Thus, the magnetic dipole term in the
Hamiltonian is approximately an operator in the Hilbert space of a single particle. Since
it is invariant under Galilean transformations (once we figure out the proper form for the
magnetic field in each frame), we can work out what this operator is in the rest frame of the
particle.
The states of a particle at rest transform as a representation of the rotation group. As we
have said, for an “elementary” particle like an electron we assume that the representation is
irreducible, which is to say that it corresponds to a fixed value of the spin, j. The magnetic
dipole operator transforms just like the angular momentum. The most general operator in
the space can be constructed as a polynomial in the angular momentum operators. Using the
commutation relations, and the formula Ja2 = ~2 j(j + 1), we can write the general operator
as a sum of totally symmetrized tensors Ja1 ...ak , which are traceless in each pair of indices.
The space of operators in the spin j Hilbert space can be considered the tensor product of
the Hilbert space with itself. The rules for addition of angular momentum tell us that this
contains every integer spin between 0 and 2j, exactly once.
The bottom line is that the only operator transforming like a magnetic moment is the
angular momentum generator Ja itself. Thus,
µa = γJa . (7.88)
For a model of an electron as classical spinning charge distribution, with charge and mass
distributed uniformly around the axis of spin, one derives such a formula with
e
γ=− . (7.89)
2me
γ is called the gyromagnetic ratio. The correct quantum value of γ for an electron is not
predicted by this classical formula and is a free parameter in the nonrelativistic quantum
theory of the electron, but is predicted by quantum electrodynamics (QED), the relativistic
quantum field theory of electrons interacting with the quantized electromagnetic field. QED
multiplies the classical result by a dimensionless number called the electron g factor, which
can be computed in a power series in the fine structure constant. QED gives
 Rotation Invariance and the Hydrogen Atom  189

α α
g = 2(1 + + o(( )2 )),
2π π
1 2
e
where α ∼ 137 is the fine structure constant ~c . Many higher order corrections have been
computed, for both the electron and the muon and the agreement between theory and exper-
iment is better than any other result in the history of science.
The Hamiltonian we have written for the hydrogen atom commutes with the electron spin.
Now let us consider what happens when we subject the hydrogen atom to a small uniform
magnetic field, pointing in the three direction. The orbiting electron will experience Lorentz
forces due to this field. We will study these in Chapter 9, but they give rise to very small
effects compared to the Coulomb attraction of the nucleus. For small field B3 , the dominant
effect will be that the previously degenerate spin states are split in energy. We can in fact
calculate this splitting exactly because the spin Hamiltonian Hs = γ2 σ3 B3 commutes with
the Coulomb Hamiltonian. The entire effect (which by the way is called the Zeeman effect)
is to add ± γ2 B3 to each energy level, with the sign determined by the 3 component of the
spin.
For multielectron atoms, the Zeeman effect is more complicated, because the spin Hamil-
tonian is i γ2 σ3 (i)B3 . Thus, for example, in the ground state of Helium, the two electrons are
P

combined into the antisymmetric spin singlet state and the spin Hamiltonian has no effect.
The Zeeman effect is usually analyzed by applying the single electron spin Hamiltonian to
the unpaired electrons in an atom.
In fact, things are even more complicated than this, because the relativistic corrections
to the Schrödinger equation do not conserve spin and orbital angular momentum separately,
but only the combined rotation generator
J = L + S. (7.90)
The leading term that is invariant under rotations, but not separate spin rotations, has the
form
δH = kL · S. (7.91)
2
This interaction comes from two effects of order vc2 . The first, called the Larmor effect, is
simply the relativistic transformation of the Coulomb field of the nucleus into the instanta-
neous rest frame of the orbiting electron. This gives rise to a magnetic field, which interacts
with the magnetic moment of the electron. The second effect, first computed by Thomas [24],
has to do with the fact that the orbiting electron is accelerating, so the rate of precession of
its spin in a given field is not the same as it would be in an inertial frame. The effect of the
so-called Thomas precession is to reduce the Larmor energy by a factor of 1/2.
The result of these two relativistic corrections is a coefficient k given by
1 dU (r)
k= , (7.92)
2rme c2 dr
190  Quantum Mechanics

where U (r) is the nuclear potential. Using 2L · S = J2 − L2 − S2 , we see that the full
Hamiltonian in the absence of an external magnetic field, commutes with J2 , L2 , S2 and J3 ,
but not S3 or L3 . Thus, when the L − S coupling Hamiltonian is comparable to or larger
than the Zeeman Hamiltonian, the simple description of the Zeeman effect as splitting two
degenerate spin states is no longer valid.

7.8 SPIN PRECESSION AND SPIN RESONANCE

Electron spin resonance (ESR) is a technique that uses the simple Zeeman effect to identify
materials with unpaired electrons. It has applications throughout chemistry and biology.
It is closely related to nuclear magnetic resonance (NMR)8 , where the splitting of nuclear
spin levels in an external field is used. Nuclear magnetic moments are inversely proportional
to nucleon masses, and are thus ∼ 2, 000 times smaller than the electron moment. For a
fixed magnetic field, the energy splittings are thus much smaller in the nuclear case, which
means that NMR signals involve electromagnetic radiation of much lower frequency than
ESR signals.
The basic idea of ESR or NMR experiments is to subject a sample containing a large
number of atoms with the same unpaired electron/nuclear spin to both a source of monochro-
matic radiation and an external magnetic field B0 . The magnetic field produces a splitting
∆E = gµB B0 between two spin states (here g is a g-factor, which might be different than the
free electron ge ∼ 2 because of the environment in which the unpaired electron sits). At finite
temperature, the average number of electrons in each state will differ, by a Boltzmann fac-
∆E
tor e− kT (see Chapter 12 on Statistical Mechanics). One then tunes the field B0 so that
∆E = ~ω, the photon energy of the monochromatic beam. At this resonance frequency the
electrons can move between the higher and lower state by (stimulated) emission or absorp-
tion of a photon. Since the lower state is more heavily populated, this will give rise to a net
absorption of the radiation, which we can monitor by detecting the beam after it is passed
through the sample. For microwaves of frequency ∼ 9, 390 MHz, the resonance (for g = ge )
occurs at B0 ∼ .335Tesla, while for the same magnetic field, NMR occurs for a frequency of
about 14 MHz.
ESR signals can give information about the atomic and molecular structure of the com-
pounds in which the electrons that produce the response are embedded. For example, as we
noted above, LS coupling can change the nature of the electronic spin states and correlate
them with the orbital state of the electron. This effect leads to the replacement of the free
electron ge factor by a matrix. The i-th component of the magnetic moment is given by
σj
µi = µB gij , (7.93)
2
8
NMR is widely used in medical applications, and has been renamed magnetic resonance imaging (MRI)
because of the widespread public fear of anything involving nuclear physics.
 Rotation Invariance and the Hydrogen Atom  191

and ESR experiments measure gij . There are also interesting complications introduced by
interaction of the electron spin with nuclear spins, and these can also be exploited to learn
about the structure and composition of the materials in which the electron is embedded.

7.9 STERN–GERLACH EXPERIMENTS

For free particles, moving at speeds much smaller than the speed of light, the components
of spin are conserved quantum numbers. In principle, that means that we can study states
with fixed values of S3 , the component of spin along some given axis. A general pure state
of the particle spin (for S2 = 12 23 ) will have the form

|ψi = a+ |+i3 + a− |−i3 . (7.94)

A Stern–Gerlach machine is a device for separating out the states of different S3 component.
It exploits the interaction between the magnetic moment of the particle and an external
magnetic field. Stern–Gerlach machines only work for neutral particles that are not subject
to the much stronger Coulomb forces which could mask the S–G effect. S–G filtering is often
used in thought experiments, to prepare pure quantum states with specified properties.
The Hamiltonian for the magnetic moment interaction with a magnetic field pointed in
the 3 direction is
∆H = µB3 (X)σ3 . (7.95)
It modifies the free particle Heisenberg equation of motion from Ṗ = 0 to
P
Ẋ = , (7.96)
m
Ṗ = −µ∇B3 σ3 . (7.97)
The parameter µ is twice the magnetic moment. The full Hamiltonian commutes with σ3 , so
we can solve the Heisenberg equations separately for σ3 = ±1.
Consider a magnetic field
B3 = f (X12 + X22 )X3 , (7.98)
where f is a smooth function, which vanishes when its argument is larger than r2 , and is
approximately equal to 1 for smaller argument. We will neglect effects of the gradient of f .
We choose the Heisenberg wave function of the particle to be
x2 +x2 (x1 −R)2 px1
− 2 3
ψ=e d2 e− d2 ei ~ , (7.99)

where R  r.
The classical picture corresponding to this Hamiltonian is shown in Figure 7.1.
192  Quantum Mechanics

Figure 7.1 The Stern–Gerlach experiment.

There is a magnet centered around the origin of the plane x3 = 0. The initial probability
distribution for the particle’s position is centered around a point in that plane far from the
magnet, but its average momentum sends it toward the region where the magnetic field is
nonzero. The momentum is conserved until the expectation value of the coordinate X(t) is
inside the region where f 6= 0. At that point P˙3 = ±µ and the expectation value of X3 (t)
moves away from x3 = 0 in a direction that is correlated with the sign of the spin component.
The expectation value of X1 (t) continues to move, and eventually exits the region where
f 6= 0.
The effect of the S–G apparatus is to entangle the spin state of the particle with the
X3 position of the particle. If we put detectors above and below the plane at positions
x1 > r, x3 = ±R, then each detector will see particles with definite polarization. Running
the experiment many times, we can measure the probabilities |a± |2 by counting the numbers
of clicks in each detector (we assume we know the sign of µ). More importantly, we now have
a source of particles prepared in pure σ3 eigenstates.

7.10 EXERCISES
7.1 Prove that Li = ijk Rj Pk satisfies the same commutation relations as ~Ki , where Ki is
the generator of rotations.
7.2 Find the form of the operators K± and K3 in spherical coordinates.
l−m
7.3 Find the norm of the state K− |lli.
7.4 The spherical harmonics have the form Ylm (θ, φ) = eimφ Plm (cos θ). Using the form of
the operators Ki derived in Exercise 7.2, find the equation for Plm (w) as a function of
w = cos θ. Argue that Plm = Pl−m . Show that for m > 0, the formula
dm Pl0
Plm (w) = (1 − w2 )m/2 ,
dwm
solves the equation. The polynomials Pl0 ≡ Pl are called the Legendre polynomials.
 Rotation Invariance and the Hydrogen Atom  193

7.5 The classical Kepler problem has the same form as the Coulomb problem, but the
coefficient in the inverse square law is GM1 M2 and the electron mass is replaced by
the reduced mass µ = MM11+M M2
2
. Consider this as a quantum mechanical problem. For
Ze2
the earth sun system, M1  M2 and so we replace 4π 0
by GM m and me by m.
Calculate the Bohr radius for this system, in centimeters and the Rydberg energy, in
joules. Requiring that expectation value of the velocity be less than that of light gives us
a constraint on n, the energy level for which the analysis is applicable. The expectation
value of p2 /2m in the n-th state is of order the binding energy

10188
joules,
n2
and this must be < mc2 = 5.4 × 1041 joules. Thus, n > 1073 .

7.6 Equate the binding energy for the orbit of the earth, as given by classical physics, to
the binding energy given by the Bohr formula, in order to estimate the value of n for
the earth’s orbit. Does it obey the bound of Exercise 7.5?

7.7 We wrote the solution to the radial equation for hydrogen in terms of associated
Laguerre polynomials, Lpq (x) which are solutions of the equation

xLpq 00 + (p + 1 − x)Lpq 0 + qLpq = 0.

Show that the only the equations for q = 0 have a constant solution, and only those
for q = 1 have a linear solution. Find those solutions, each of which should contain an
undetermined multiplicative constant.

7.8 A natural conjecture from the result of Exercise 7.7 is that Lpq is a q-th order polynomial.
Mathematicians like to choose the normalization constants so that the polynomials are
monic: the coefficient of the term with the highest power of x is equal to 1. The Lpk for
k ≤ q are a basis for all polynomials of order ≤ q. xLpq is a new linearly independent
polynomial and so Lpq+1 must be a linear combination of this and the Lpk with k ≤ q. It
turns out that this relation is rather simple. Show that

(A + Bx)Lpq + CLpq−1

satisfies the equation for Lpq+1 for appropriate choice of the constants.

7.9 The polynomials Lpq for fixed p can be considered eigenfunctions, with different eigen-
value, of the operator
d2 d
x 2 + (p + 1 − x) .
dx dx
194  Quantum Mechanics

Show that this operator is Hermitian on the space of functions on the interval [0, ∞]
with scalar product defined by
Z ∞
hf |gi = ds xp e−x f ∗ (x)g(x).
0

Show that the Laguerre operator is Hermitian with respect to this scalar product so
that the Lpq for different q are orthogonal. Compute the norm of the monic polynomials.
It is easiest to do this by using the explicit form of the polynomials that we derived in
the text.
d d
7.10 The Legendre polynomials Pl are eigenfunctions of the operator D = dw ([1 − w2 ] dw ),
with eigenvalues −l(l + 1). Find a scalar product
Z 1
hf |gi ≡ dw µ(w)f ∗ (w)g(w),
−1

on the space of functions on the interval [−1, 1], such that D is an Hermitian operator.
Argue that the different Legendre polynomials are orthogonal with respect to this scalar
product.
7.11 Orbits in the Coulomb potential are ellipses and the semimajor axis of the ellipse stays
constant in time. This indicates the existence of a new vector conservation law E, which,
unlike angular momentum, changes sign under reflection. To understand how it arises,
note that the equation of motion has the form
d2 r a
m = 2 ê,
dt2 r
where
Ze2
a≡ ,
4π0
and ê is the unit vector from the origin to the position of the particle. Recalling that
the motion is in a plane with coordinates (r, φ), and defining eφ to be the unit vector
in the direction of the velocity, we have
ėφ = −φ̇ ê.
This just says that the acceleration is always directed toward the center. Also, the
magnitude of the angular momentum is l = mr2 φ̇. Show that we can write the equation
of motion as
dv a mr2
=− 2 ėφ .
dt mr l
Show that this implies a new conservation law. Call the new conserved vector h, for
Hamilton, the person who discovered it.
 Rotation Invariance and the Hydrogen Atom  195

7.12 Construct the first three Legendre polynomials by insisting that they be orthonormal.

7.13 Since h and L are conserved, so is h×L. Show that for bound orbits this is proportional
to the Laplace–Runge–Lenz vector, the vector defined by the semimajor axis of the
ellipse.
P∞
7.14 Define the generating function of Legendre polynomials by G(w, s) = l=0 sl Pl (w).
Show that
d dG d2 (sG)
([1 − w2 ] ) = − .
dw dw ds2
Show that a solution of this equation is

G(w, s) = (1 − 2sw + s2 )−1/2 .

7.15 Prove that the nth solution of the radial wave function for hydrogen has n nodes: i.e.,
n places where the wave function vanishes.
 CHAPTER 8

Scattering Electrons on a
Nucleus

8.1 INTRODUCTION
This section contains a brief description of the scattering solutions of the hydrogen atom, and
a brief introduction to scattering theory. The material is of much less general utility than
our discussion of the bound state spectrum, so our treatment is somewhat cursory. More
material on scattering theory can be found in Chapter 16.

8.2 POSITIVE ENERGY EIGENFUNCTIONS AND SCATTERING AMPLITUDES

In celestial mechanics, we are familiar with the fact that gravitational attraction can form
bound systems like the solar system. We also know about other trajectories, in which two
objects, subject to gravitational attraction, suffer a close encounter, but then one of them (in
the rest frame of the other) wanders off into space, never to return. In quantum mechanics,
the analog of these “close encounters of the second kind” are called scattering states. If
you have done all the exercises of Chapter 4, you have already encountered these in a one-
dimensional context. In this chapter, we want to study the scattering states of electrons in
hydrogen-like ions.
These are solutions of the Schrödinger equation with positive energy:

Z
[−∇2 ψ − ]ψ = k 2 ψ. (8.1)
r

k 2 is the energy in units of the absolute value of the Rydberg energy, and k is also the wave
number in units of the Bohr radius. We can make an angular momentum decomposition of
this eigenvalue problem, and the resulting radial equation is

197
198  Quantum Mechanics

2 l(l + 1) 2Z
[−∂r2 − ∂r + − ]ψlm (r) = k 2 ψlm (r). (8.2)
r r2 r
r
The Z dependence can be removed by scaling r → 2Z and k → 2Zk, just as in the bound
state problem.
Let us use these rescaled variables and analyze the behavior of the solutions as r → ∞.
Dropping the angular momentum and potential terms, which vanish in this limit, we have

2
[−∂r2 − ∂r ]ψlm (r) = k 2 ψlm (r). (8.3)
r

We can get rid of the first derivative term by writing ψ = ra φ so that

2
[−∂r2 − ∂r ]φ(r) − a(a − 1)r−2 φ(r) − 2(a + 1)/r∂r φ(r) − 2a/r2 φ(r) = k 2 φ(r). (8.4)
r
±ikr
For a = −1, both the first derivative and 1/r2 terms cancel and φ ∼ e r . Thus, the
radial equations for all angular momenta, have two solutions, which behave at infinity like
0 pm ±ikr 0±
ψlm ∼ e r , where we take k to be positive. The functions ψlm are solutions of the free
2
particle Schrödinger equation. When combined with the time-dependent factor e−i~k t , they
look like incoming and outgoing radial waves, with incoming corresponding to eikr . They
have continuum normalization, with the scalar product of two different values of k, and the
same value of the ± index, being δ(k − k 0 ). We would like to understand what these solutions
have to do with the problem of scattering. For the moment, however, note that the general
solution of the Schrödinger equation has a singularity at r = 0, because both the Coulomb
potential and the angular momentum barrier are singular there. The Hamiltonian will only be
a Hermitian operator if we restrict attention to the linear combination which is nonsingular
−ikr
at the origin. As a consequence, at infinity, if we normalize the coefficient of e r to 1, the
solution will behave like
e−ikr e+ikr
+ A(k) . (8.5)
r r
Conceptually, we can formulate the problem of scattering as follows. Start with an incoming
wave packet
d3 k
Z 2 k2
ikk̂·(x−x0 ) −i ~2m (t−t0 )
ψin (x, t) = [ψin (k)Y in (k̂)e e ]. (8.6)
(2π)3/2

We want the initial angular wave function Yin to be concentrated in a small solid angle. It
does not change with the free time evolution. For a scattering problem, we want to take
t0 → −∞ and |x0 | → −~k 0 t0
2m , where k0 is the center of the momentum space wave packet
 Scattering Electrons on a Nucleus  199

ψin (k). The idea is that, in this limit, because the potential falls off at infinity,1 , this function
is a better approximation to the solution of the full time-dependent Schrödinger equation,
when t is large and negative. The problem of scattering is to find the evolution of this solution
as t → ∞.
Physical intuition tells us that two possible things can happen to a classical charged
particle thrown into a Coulomb field of opposite charge from infinity. It can be captured into
a bound orbit, or it can follow a hyperbolic orbit back out to infinity. For a two body system,
which of these two things happens is entirely determined by conservation laws. If the energy
is positive, the particle scatters off to infinity, and if it is negative, it becomes bound. If more
than two particles are involved in the scattering, more complicated things can occur. Only the
total energy is conserved, so an incoming scattering state of 3 or more particles can transform
into one in which a two particle subsystem has negative energy, but the total is made up by
the positive energy of the third. Thus, the complete set of asymptotic states of such a system
will include not only the original particles, but bound states of pairs, or more complicated
subsystems of them. In the real world, things are even more complex, because of the existence
of massless photons. In relativistic quantum mechanics, particles can be created from empty
space if sufficient energy is available. A single electron can scatter off a single ion, and be
absorbed into a negative energy bound state, with energy conservation guaranteed by the
emission of a positive energy photon. However, in our model problem, there are no photons,
and only a single electron scattering off an ion. Positive energy in the past then implies that
the state in the future is also a scattering state. No bound states appear in the final state.
If we now return to the exact eigenfunctions of the positive energy Coulomb Schrödinger
equation, we can state the scattering/bound state problem more clearly. The spectral decom-
position of the Hilbert space for fixed l is given by the resolution of the identity
X Z ∞
1= |nihn| + dk|k ih k|. (8.7)
n 0

The continuum states |ki appearing in this decomposition are the eigenstates of the radial
Hamiltonian for the appropriate value of l. The Hamiltonian is a second-order ordinary
differential operator, with a singularity at r = 0. Only one of the two linearly independent
solutions gives a normalizable solution in the vicinity of this singularity.2 As r → ∞ this
1
Actually, the Coulomb potential does not fall off fast enough at infinity to justify this assumption. The
formalism of scattering theory requires modification to deal with this problem. We will ignore the issue,
but if you are bothered by it, imagine turning the Coulomb potential into a Yukawa potential with a range
of 10 billion kilometers. Formal scattering theory, as explained in Chapter 16, will apply to this problem,
as all of the issues will have to do with certain quantities that do not have finite limits as the range goes to
infinity. For any experiment in a laboratory of size much smaller than the range, these must be irrelevant.
2
For l = 0, as in the bound state problem, both solutions are square integrable near r = 0 but only one is
sufficiently well behaved that no probability current flows into the singular point.
200  Quantum Mechanics

unique eigenstate wave function is a linear superposition of incoming and outgoing radial
waves:
eikr e−ikr
ψl (k 2 , r) → A(k, l) + B(k, l) . (8.8)
r r
We define the incoming state by setting A = 1 and the outgoing state by B = (−1)l+1 . The
two are obviously related by multiplication by a phase
B(k, l)
Sl (k) = (−1)l+1 ≡ eiδl (k) . (8.9)
A(k, l)

The prefactor (−1)l+1 arises because the amplitude for no scattering is the amplitude to take
an incoming wave vector k to an outgoing wave vector −k, since the orientation of incoming
and outgoing vectors on the sphere at infinity is opposite. The prefactor takes into account
the transformation of spherical harmonics under this spatial reflection, so that the amplitude
for this nontransition is 1 if the phase shifts δl (k) vanish.
The amplitudes A(k, l), B(k, l) of course depend on the precise nature of the incoming
wave packet. The actual response of the system is encoded in the ratio B/A, which, together
with the prefactor, we have called the scattering matrix element Sl (k). The actual physical
setup of scattering involves localized incoming and outgoing wave packets, focused around
two directions Ωin/out . The amplitudes for scattering a particular incoming wave packet into
a particular outgoing wave packet will be matrix elements of an operator in the space spanned
by a complete set of either the incoming or outgoing wave packets. This is the Scattering
Operator and its matrix is the S-matrix. The S-operator commutes with the Hamiltonian, the
total momentum and the total angular momentum, for any translationally and rotationally
invariant system. The fact that for a two body rotationally invariant system, the Schrödinger
equation has only one solution for each value of energy and angular momentum shows that
the S-matrix is diagonal in the basis where energy and angular momentum are diagonal.
The phases eiδl (k) are its eigenvalues. In Chapter 16, we will discuss scattering theory for
potentials that are not spherically symmetric, and in that case the S-matrix will not be
diagonal in any easily accessible basis.

8.3 ANALYSIS OF THE COULOMB SCATTERING AMPLITUDES

If you search through a book of special functions or Google “Coulomb Scattering Wave Func-
tions,” you will find that the solutions of the positive energy Schrödinger equation are special
cases of Confluent Hypergeometric functions. A lot is known about these functions, including
exact integral representations. From these, one can extract the Coulomb phase shifts:

Γ(l + 1 + ki )
e2iδl (k) = , (8.10)
Γ(l + 1 − ki )
 Scattering Electrons on a Nucleus  201

where k is the wave number in Bohr radii, which corresponds to the incoming energy.
The Euler Gamma function has poles when its argument is a nonpositive integer. This
happens for negative energy, and the poles are precisely where the bound states of the
hydrogen atom sit. We have seen this phenomenon for the square well potential in Chapter
4 and we will see in Chapter 16 that this is a general phenomenon: bound states show up as
poles in scattering amplitudes when they are analytically continued to “unphysical” regions
where the energy variable is incompatible with the kinematics of scattering. The effect of
these complex poles on scattering can be dramatic, if the imaginary part of the energy
is small.
In a scattering experiment, the detector measures the number of particles per unit time,
N (Ω) that are scattered into a solid angle Ω relative to the direction of the incoming particle.
The differential cross section is
dσ N (Ω)
= , (8.11)
dΩ J
where J is the flux of particles in the incoming beam. J is computed from the probability
~ ∗
current −i m ψ ∇ψ + c.c. which, when integrated over a small surface area, gives us the total
probability per unit time, that a particle with wave function ψ will cross that surface. For a
plane wave ψ = eik·r , J is just ~k
m for a surface orthogonal to k.
The scattering eigenfunction for an initial plane wave can be computed by doing the
expansion of the plane wave in spherical coordinates and using the scattering solutions for
fixed angular momentum. To compute the cross section, we need only the asymptotic behavior
of the scattered wave. You will do this carefully in Exercises 8.1 and 8.2. The asymptotic
behavior of the full in state wave function is

e−ikr
ψin = eik·r + f (θ) , (8.12)
r

where
∞ 2iδl (k)
X e −1
f (θ) = Pl (cos(θ). (8.13)
l=0
2ik

There is no dependence on the azimuthal angle φ because the potential is rotation invariant
and the choice of incoming plane wave is invariant under rotation of φ.
The current of outgoing particles can be separated out by going to a nonzero scattering
angle θ. It is
~k
Jout = Ω|f (θ)|2 . (8.14)
mr2
Here we have neglected interference terms between the incoming plane wave and the scattered
wave. If, instead of a plane wave, we would chosen a wave packet concentrated around the k
202  Quantum Mechanics

direction these would cancel almost completely. The area element around Ω is dA = r2 dΩ.
If we dot the scattered current into Ω and compute N (Ω) we get
dσ
= |f (θ)|2 . (8.15)
dΩ
To get a feeling for the meaning of these formulae, let us consider scattering of classical
particles in a central potential V (r).
In classical scattering, the particle follows a fixed trajectory, which is a hyperbola for the
Coulomb potential. The angle of scattering is determined by impact parameter b(θ), which
is the distance of closest approach to the center. θ is the polar angle of a coordinate system
whose north pole is determined by the direction of the incoming particles. By rotational
invariance, the impact parameter is independent of the azimuthal angle φ. If we have a beam
of particles, with incoming flux J, the number of particles scattered per unit time between θ
and θ + dθ is equal to the number of incident particles per unit time between b and b + db.
The number of particles scattered into solid angle Ω is
db
N (Ω)dΩ = 2πN sin(θ)dθ = J2πb dθ. (8.16)
dθ
See Figure 8.1 for the geometry leading to these equations.
The differential cross section is defined to be the ratio between the number of particles
per solid angle scattered into the direction Ω and the incident flux. It has dimensions of area,
which explains the name cross section. Classically we have
dσ b db
= | |. (8.17)
dΩ sin(θ) dθ
In Exercise 8.3, you will verify that for a Coulomb potential
dσ Z 2 e4
= . (8.18)
dΩ classical (16π0 )2 sin4 (θ/2)

θ
S

dσ
dΩ

Figure 8.1 The geometry of a scattering experiment.

 Scattering Electrons on a Nucleus  203

Quantum mechanically, we have seen that

dσ
= |f (θ)|2 . (8.19)
dΩ
We can relate f (θ) to the partial wave phase shifts, by expressing the asymptotic form of the
wave function as a sum over spherical harmonics. If we take the direction of the incoming
wave k as the polar angle, then
∞
X
ik·r ikr cos(θ)
e =e = jl (kr)Pl (cos(θ)), (8.20)
l=0

where
d2 2 d l(l + 1)
(− 2
− + )jl (kr) = k 2 jl (kr). (8.21)
dr r dr r2
The solution is a function of kr because we can set k = 1 by rescaling r. The solution of this
equation, which is nonsingular at r = 0 is called the spherical Bessel function of order l. You
will explore some of its properties in Exercise 8.4.
Putting together the results of that exercise with our discussion of the wave function for
fixed l, we conclude that
∞
1 X
f (θ) = (2l + 1)[e2iδl (k) − 1]Pl (cos(θ)). (8.22)
ik l=0

For the Coulomb potential this gives

∞
1 X Γ(1 + l + iη)
f (θ) = (2l + 1)[ − 1]Pl (cos(θ)). (8.23)
ik l=0 Γ(1 + l − iη)

The Coulomb scattering amplitude is very singular at θ = 0, and the total cross section
dσ
Z
σ= dΩ , (8.24)
dΩ
is infinite. The total cross section is the probability per unit initial flux that particles are
scattered by the potential. The reason it diverges is the long range nature of the Coulomb
potential. No matter how large the impact parameter, there will be a nonzero deflection of
the charged particle trajectory.
The infinite total cross section is a symptom of an even more serious problem that arises
when we take into account the quantization of the electromagnetic field, of which the classical
Coulomb potential is just the classical, nonrelativistic remnant. Classical charged particles
radiate when they accelerate, and the fact that scattering amplitudes do not fall off with
204  Quantum Mechanics

impact parameter means that particles are always radiating photons, no matter how far
apart they are. Moreover, the classical radiation field of an accelerated charge is a coherent
state, which has an amplitude to consist of an arbitrarily large number of photons. These
facts make the essential hypothesis of scattering theory that particles become free when far
apart, false.
This problem has little practical importance, because most of the photons radiated have
very low energy and we can just define inclusive probabilities, where we sum over states of
photons with energy less than some arbitrary cutoff and define a density matrix for other
particles. These probabilities are all that an experimentalist with limited energy resolution
can ever measure. Nonetheless, the problem is not completely solved at the conceptual level.
A similar problem arises for gravitons, the hypothetical particle connected to Einstein’s
gravitational field, and it may lie at the heart of the nature of the quantum theory of gravity.

8.4 EXERCISES
8.1 The equation for positive energy Coulomb wave functions is
2 l(l + 1) 1
[−∂r2 − ∂r + − ]ψl (r) = k 2 ψl (r).
r r2 r
Show that the substitution ψl = rl e±ikr φl takes this into an equation of the form
[∓r∂r2 + (b ± ir)∂r + a]φl = 0.
What are the values of a and b?
8.2 The fact that r appears linearly in the equation for φl in the previous exercise, motivates
us to write Z d
φl (r) = dz ezr fl (z).
c
The integral is taken over some open contour in the complex plane, with the indicated
endpoints. Derivatives become powers of z and linear terms in r can be converted, via
integration by parts, into first derivatives of fl with respect to z. If we choose the end-
points to be places where fl (c) = fl (d) = 0, the surface terms in the partial integration
vanish. Find the appropriate contour and solve for fl . Show that the endpoints are finite
points in the plane and that fl is nonsingular on the contour of integration. Prove that
φl is nonsingular at finite r, and use this integral formula to extract its behavior as
r → ∞ and thus the phase shifts.
8.3 Show that the classical formula for the Coulomb scattering cross section (first derived
by Rutherford) is
dσ Z 2 e4
= .
dΩ (16π0 )2 sin4 (θ/2)
 Scattering Electrons on a Nucleus  205

8.4 In terms of the variable z = kr, the spherical Bessel equation is

d2 2 d l(l + 1)
(− 2
− + )jl (z) = jl (z).
dz z dz z2
Show that near r = 0, the right-hand side is negligible and the solution has the power
law form z a . Find the values of a and argue that only one of them is acceptable. As
r → ∞, show that only the first term on the left-hand side, and the term on the
right-hand side are important, so that the solutions are e±iz . Eliminating one of the
powers at the origin implies that we get a specified linear combination of the two
exponentials. The equation is invariant under z → −z so we can always choose even
and odd combinations. Show that the value of l determines the combination eiz ± e−iz .

8.5 Solve for the power series expansion of jl (z).

 CHAPTER 9

Charged Particles in a
Magnetic Field

9.1 INTRODUCTION
This chapter discusses Landau’s solution of the motion of a charged spinless particle in a
constant magnetic field. This classic problem has become more and more important as time
passed and really became a key topic in condensed matter physics with the discovery and
explanation of the Fractionally Quantized Hall Effect.
The basic feature of Landau’s solution is that the eigenstates are grouped into discrete
bands with infinite degeneracy. This is directly related to the guiding center solutions of the
corresponding classical problem: charged particles spiral around a fixed position in the plane
perpendicular to the field. In the limit of large fields, all but the lowest Landau level get very
large energy, but the physics of that degenerate level, once electron interactions are taken
into account is intricate and fascinating.

9.2 THE LORENTZ FORCE AND LANDAU’S SOLUTION

The Lorentz force on a charged particle moving in a time-independent magnetic field is

mẍ = q ẋ × B. (9.1)

In order to write these in the Euler–Lagrange form, we must introduce the vector potential
B = ∇ × A. The classical Lagrangian coupling a charged particle to a static magnetic field
with vector potential A(x) is
1
L = mẋ2 + q ẋ · A(x). (9.2)
2
207
208  Quantum Mechanics

In Exercise 9.1, you will show that the Euler–Lagrange equations of this Lagrangian are
simply the Lorentz force equations. Applying the rules of analytical mechanics to the above
Lagrangian leads to a Hamiltonian
1
H= (p − qA(x))2 . (9.3)
2m
In classical mechanics, only the equations of motion matter, and these are the usual Lorentz
force equations, which contain only the magnetic field B = ∇ × A, so classical mechanics is
invariant under gauge transformations A → A + ∇Λ. In quantum mechanics (QM), this is
less obvious because A appears in the Schrödinger equation
(−i~∇ − qA)2
ψ = i~∂t ψ. (9.4)
2m
The key to understanding what is going on is the realization that the overall phase of the
wave function has no physical meaning in QM. The transformation ψ → eiqΛ ψ, with constant
Λ, is a redundancy or gauge symmetry of ordinary QM. Without a magnetic field, or if the
particle has no charge, this is not true of phase transformations with variable Λ. However, it
is easy to see (Exercise 9.2) that if we combine a variable phase change Λ(x) with the gauge
transformation A → A + ∇Λ, then the Schrödinger equation is invariant.1 This is equivalent
to the statement that under the combined transformation,
q
Dψ ≡ (∇ − i )ψ → eiqλ/~ Dψ (9.5)
~A
D is called the covariant derivative operator.
For multiparticle systems, the corresponding transformation of the multiparticle wave
function is P qi
ψ(x1 . . . xn ) → ei i ~ Λ(xi ) ψ(x1 . . . xn ), (9.6)
where qi is the charge of the i-th particle. Note that the gauge function and vector poten-
tial depend only on three space coordinates, while the wave function depends on the 3N
coordinates of all the particles. In the formalism of Quantum Field Theory, which we have
alluded to in Chapter 5, and will discuss a bit more in Chapters 11 and 12, we have operator
valued quantum fields, which are functions of a single position coordinate, but which describe
the entire collection of any number of identical particles with fixed quantum numbers.2 The
Hamiltonian is built from these field operators and gauge invariance is implemented by replac-
ing derivatives by covariant derivatives for each field operator. The full gauge invariance of
1
The gauge function Λ must be independent of time, unless we introduce a scalar potential to make a covari-
ant time derivative. In the gauge where the scalar potential is zero, time-dependent gauge transformations
of A generate physical electric fields E = Ȧ.
2
In relativistic quantum field theory, a single field describes both particles and antiparticles.
 Charged Particles in a Magnetic Field  209

Maxwell’s equations is incorporated into QM by adding the scalar potential Φ(x, t) and let-
ting the vector potential depend on time as well. The coupling of particles to Φ is dictated
by replacing time derivatives by covariant time derivatives

Dt = ∂t − iqΦ.

In this formalism, if one wants to describe the full QM of the electromagnetic field interacting
with charged particles, then all one has to do is add the Hamiltonian of Maxwell’s theory
and quantize the electromagnetic field according to canonical rules. The quantization is a
bit subtle, because gauge invariance tells us that not all the fields are physical variables, but
there are various ways of dealing with this [17].
Let us return to single particle QM, and solve the simplest problem involving magnetic
fields, namely the constant B field pointing in the 3 direction. The most convenient gauge
for the vector potential is to take A3 = 0 and A1 = 12 x2 B, A2 = − 12 x1 B. The Schrödinger
equation is
−~2 2 qB qB
[∂3 + (∂2 + i x1 )2 + (∂1 − i x2 )2 ]ψ = Eψ. (9.7)
2m 2~ 2~
q
~ ~2 qB
Define xi = 2B wi and E = 2mc . In terms of the new variables, we have
w2 2 w1
−[∂32 + (∂1 − i ) + (∂2 + i )2 ]ψ = ψ. (9.8)
2 2
The equation is obviously invariant under translation in the w3 direction, so there are
solutions that are simultaneous eigenfunctions of K3 = −i∂3 , which have the form
eik3 w3 χ(w1 , w2 ), where χ satisfies
w2 2 w1
−[(∂1 − i ) + (∂2 + i )2 ]χ = [ − k32 ]χ. (9.9)
2 2
The covariant derivative operators Di ≡ ∂i − iij wj satisfy

[iD1 , iD2 ] = i, (9.10)

just like canonical position and momentum operators iD1 = Q, iD2 = P , so we can write
the Schrödinger equation in operator form as

(P 2 + Q2 )χ = [ − k32 ]χ, (9.11)

which is the equation for a harmonic oscillator. Remember Feynman’s dictum: the same
equations have the same solutions. There are no springs attached here, but we can use
our knowledge of oscillators to solve the equations. The equations of motion (classical or
quantum) are
Q̈ = −4Q, (9.12)
210  Quantum Mechanics

so that the dimensionless frequency (which is the same as the dimensionless energy) is ω = 2.
Thus, the quantized energy levels are

 = n + 1 + k32 , (9.13)

where n is a nonnegative integer and k3 is a real number. The wave functions in terms of Q
are just the familiar Hermite
√ functions, concentrated around Q = 0. In terms of the original
variables wi , the equation 2a|χi = Q + iP |χi = 0 reads
1
[i∂1 − ∂2 + (−w2 + iw1 )]χ(w1 , w2 ) = 0. (9.14)
2
There are an infinite number of degenerate solutions to this equation, which have the form
1 2 2 ∗
f (z)e− 4 (w1 +w2 ) = f (z)e−zz , (9.15)

where f is an analytic function of the complex variable z = 21 (w1 + iw2 ). A basis of normal-
izable (but not orthogonal) solutions to this equation is gotten by choosing f (z) = z n . These
functions have fixed angular momentum around the point w = 0 in the plane. On the other
∗ ∗
hand, if we look at f (u, z) = e2zu −uu , then we obtain a solution which is not localized at
the origin. We can get a better idea of the properties of this wave function by multiplying it
∗ ∗ ∗
by the gauge transformation ez u−zu to get e−(z−u) (z−u) . The probability density is
∗ +z ∗ u−zz ∗
P (z, z ∗ ) ∝ ezu , (9.16)

and is peaked at
z = u. (9.17)
The ground state wave functions centered at z = u are called guiding center solutions.
We can understand the infinite degeneracy we have encountered by thinking about the
translation symmetry of the problem. A constant magnetic field is translation invariant in all
three directions in space. This is not apparent in the vector potential, which must depend on
w in order to generate a nonzero field. We chose a potential which was manifestly rotation
covariant around the point w = 0, but there are other gauges, where the same field is
generated by a potential whose rotation symmetry is around the point z = u instead. The
Schrödinger equation in a fixed gauge is not gauge invariant. What we showed above is that
by multiplying the wave function in one gauge by a position dependent phase factor, we
obtain the wave function in another gauge.
The different guiding center solutions are not, however, gauge transformations of each
other. They are physically different solutions of the same equation, in a fixed gauge. They are
not, individually, translation invariant, but they do have the same energy eigenvalue. This is,
in general, all QM tells us about symmetries. The translation operator commutes with the
 Charged Particles in a Magnetic Field  211

Hamiltonian, but this just tells us that it takes any state into a state with the same energy.
Of course, we can try to superpose all of these different states, to obtain another degenerate
state which is translation invariant. However, if the symmetry in question is translation, we
run into a problem. The functions f (u, z)e−z̄z are all normalizable. The unique translation
invariant superposition of them Z
d2 u f (u, z)

is not normalizable (Exercise 9.3), and so is not an allowed state in the Hilbert space.
The real question is, why are there localized normalizable3 eigenfunctions for a particle
moving in a constant field? The answer can be found by combining the classical physics
of this system with the uncertainty principle. Classically, if we start a particle off with
nonzero velocity, the Lorentz force accelerates it perpendicular to that velocity. This means,
in particular, that the work done on the particle by the field is zero, so that its kinetic energy
remains unchanged. The particle will move in a circle, called a Larmor orbit, whose radius
is determined by the equation
mv 2
= qBv,
r
which says that the centripetal acceleration necessary to the circular motion comes from
the Lorentz force. Thus, particles in a constant magnetic field perform helical motions at
constant energy, moving freely in the x3 direction while performing Larmor orbits in the
plane perpendicular to the field. The center of the Larmor orbit can be located anywhere in
the plane.
In classical physics, the velocity is a continuous parameter and can be taken to zero. In
QM, because of the uncertainty principle, a particle bound in a finite region of the plane
cannot have a completely certain velocity. Indeed, the Heisenberg equations of motion tell
us that the velocity vi is proportional to pi − qAi , so that its two planar components do not
commute with each other in a constant B field! As usual, confinement of the motion to a
finite region (normalizability of the wave functions) leads to quantization of the energy. The
guiding center states that we have found are the quantum analogs of Larmor orbits and the
reason for the absence of a translationally invariant ground state is now clear.
This problem was first worked out by Landau [25] and the quantized degenerate states
for fixed k3 are called Landau levels. In particular, when B is large and the gap between
levels grows, most of the physics is described by the lowest Landau level.
We have now gone about as far as we can go with single particle QM. We will still use
it to understand a variety of approximation methods, but much of the physics to which QM
applies involves multiple particles in an important way. We have already seen this in our
3
None of the actual eigenfunctions ψ is actually normalizable, because of the free motion in the x3 direction.
We are really talking about normalizability of the wave function χ for fixed k3 .
212  Quantum Mechanics

discussion of Fermi statistics and the periodic table. Before starting in on the discussion of
atomic, molecular and solid state physics though, we should take a step back to discuss some
of the conceptual issues surrounding probability, measurement, and the nature of classical
reality, which we touched on but did not resolve in the first chapter. Students of a practical
cast of mind may want to skip the next chapter on first reading, and go on to real physics.
It is worth coming back and finishing this “philosophical” chapter at some point. Too many
physicists, who know how to use QM, still feel uncomfortable with its conceptual basis. This
discomfort is unnecessary.

9.3 EXERCISES
9.1 Show that the Euler–Lagrange equations of the Lagrangian

1
L = mẋ2 + q ẋ · A(x)
2
are the Lorentz force equations.

9.2 The Schrödinger equation for a charged particle in a magnetic field is

~2
(∇ − iqA(x))2 ψ + (V − E)ψ = 0.
2m

Show that if we transform ψ → eiqΛ(x) ψ, then this becomes

~2
(∇ + iq∇Λ − iqA(x))2 ψ + (V − E)ψ = 0.
2m
Thus, a position dependent phase is equivalent to a gauge transform of the vector
potential, and the equation is invariant under the combined operation of multiplying
by a phase and doing a gauge transformation.

9.3 Show that the translation invariant superposition of guiding center solutions is not
normalizable.

9.4 Solve the problem of a particle in a constant magnetic field using the gauge A1 = 0 for
the vector potential.

9.5 The natural form of the solutions to Exercise 9.4 are states that are eigenstates of the
translation operator P2 . Show that the infinite degeneracy of the eigenstates is now
understood in terms of translations of the x2 coordinate in this gauge, and interpret
the nonnormalizability of translation invariant states in this gauge.
 Charged Particles in a Magnetic Field  213

9.6 Show that the lowest Landau level eigenstates discussed in the text are instead eigen-
states of the angular momentum operator X1 P2 − X2 P1 . Write the angular momentum
eigenstates as linear combinations of the gauge transformations of the P2 eigenstates.
Be careful to do the proper gauge transformation to convert the wave functions in one
gauge into those in the other.
 CHAPTER 10

The Meaning of Quantum

Measurement and the
Emergence of a Classical
Reality

10.1 INTRODUCTION
In this chapter, we will outline the way in which a classical world can emerge as an approx-
imation to quantum mechanics (QM), and assess the likelihood of measuring the quantum
corrections to the classical behavior of macroscopic objects. We will see that these correc-
tions are extraordinarily small in typical states of these objects, a fact which accounts for
the difficulty we have with coming to an intuitive understanding of quantum rules. Those of
you who want to get on with the mathematical formalism of QM, and its use in atomic and
condensed matter physics, can skip this chapter on a first reading, but it is worth your while
to come back to it.
The key word in the previous paragraph is macroscopic, and we begin by defining precisely
what we mean by that. We have seen that the mass of the electron and the strength of the
electromagnetic coupling define a natural length scale for atomic structures, called the Bohr
radius. Its size is about 10−8 cm. A cubic sample of solid material, slightly less than 0.1 cm
on a side thus has about N > 1020 “unit cells” of atomic dimensions. In gaseous or liquid
states of matter, the atoms are further apart so the same number of degrees of freedom take
up somewhat more space. We will talk here about models of solids but similar estimates work
for liquids and gases as well.

215
216  Quantum Mechanics

The interactions between the N unit cells are local, each cell is strongly coupled to only
a few nearest neighbors, because long range Coulomb forces are screened. We will see that
in typical local models of a large number of variables, the number of states with almost
degenerate energies, in energy bands far above the ground state, increases like eN . We will
also see that collective coordinates, averages over of order N variables, have uncertainties that
scale like N −1/2 and the number of states with a fixed expectation value of those coordinates
is also exponential in N . As a consequence, we will show that QM defines, with exponential
precision, probabilities for histories of the collective coordinates, which satisfy Bayes’ rule
and all of our expectations for a classical probability theory. In other words, there is a small
subset of variables in any large quantum system with local interactions, which behave exactly
as we would expect a classical system subject to small random perturbations to behave. Since
all of our coarse grained observations of the world are observations of collective coordinates,
QM can explain the apparently classical nature of the world. In particular, the above results
allow us to condition predictions for the future of the entire world, on a particular history
of some of the collective coordinates. This is the procedure which has become known as
“collapse of the wave function.”
We will also talk briefly about the phenomenon of unhappening, in which the disintegra-
tion of some macrosystem into elementary particles forces us to give up predictions based
on a particular history of its collective coordinates (unless some other macroscopic system
has made a macroscopic record of that history), and return to the original uncollapsed wave
function to make predictions about the future.
Finally, we will discuss Bell’s theorem and further results which show that any attempt
to find a more classical explanation of quantum probabilities will have to have truly bizarre
features.

10.2 COUNTING STATES AND COLLECTIVE COORDINATES

The simplest possible model, which suffices to see how the counting works, assigns a two-
dimensional quantum Hilbert space to each point of a lattice with a spacing of order the
atomic scale. For simplicity of visualization, take the lattice to be cubic, though we will not
really use that in our estimates. A complete set of variables for this system is a set of Pauli
matrices σ(i), one for each lattice site. The Pauli matrices on different sites commute with
each other. A simple, soluble Hamiltonian for this system is
X X
H = −J σ3 (i)σ3 (j) + B σ3 (i). (10.1)
<i,j> i

J and B are positive constants. The symbol < i, j > means that we sum over all pairs of
the nearest neighbors on the lattice. We will call σ3 (i) the local spin of the system. We can
see by inspection that the first term likes all of the “spins” to be aligned, while the second
 Meaning of QM and Emergence of a Classical Reality  217

term prefers them to be negative. Low lying excitations of this ground state are gotten by
flipping a single spin, at a cost in energy of Jc + B, where c is the number of different nearest
neighbors. If there are N different lattice points, then there are N different states of this type,
depending on which spin we flip (use periodic boundary conditions to avoid edge effects at
the “end” of the lattice).
Next we can flip two spins. If they are not nearest neighbors, then the energy cost is
twice that of flipping a single spin, while if they are, we get an energy lower by J. There
are N (N − 1) excitations of this type, most of them exactly degenerate. Similarly, there
are k!(NN−k)!
!
states with k spins flipped, again mostly degenerate in energy. For large N and
1  k  N , the degeneracy is of order ekN . It is this exponential behavior of the number
of states with the volume N , which is responsible both for the success of the methods of
statistical mechanics and the emergence of classical behavior.
Now imagine perturbing the Hamiltonian by a small term
X
δH =  σ1 (i)σ1 (j). (10.2)
<i,j>

The problem is no longer exactly soluble but when   J, B, we can see the qualitative nature
of the effect on the spectrum fairly easily. We will systematize this kind of QM perturbation
theory in Chapter 13, but for the time being we need only qualitative results. For states
between which the unperturbed problem gives a gap in energy of order J or B, the gap will
only be changed a little. By contrast, for the exactly degenerate states, the leading order
correction to the energy is gotten by diagonalizing the perturbation δH in each degenerate
subspace. This gives a matrix whose size is order the degeneracy and whose eigenvalues are
all within something of order ± of the original degenerate energy level. If we start in some
particular state, for example one of the original degenerate eigenstates of σ3 (i), it will evolve
under the action of the perturbed Hamiltonian around the subspace of erstwhile degenerate
states, in a fairly random manner.
It is difficult to make mathematically rigorous statements, which quantify the degree of
randomness in this evolution, but the assumption of randomness is the key hypothesis in
both the explanation of the thermal properties of matter1 and the emergence of classical
physics.

10.3 COLLECTIVE COORDINATES

The second important concept in the explanation of classical behavior is the notion of a
collective coordinate of a macroscopic system. In our model, these are variables like
1 X
Σ≡ σ(i). (10.3)
N
1
Unfortunately, this explanation is far beyond the scope of the current elementary textbook.
218  Quantum Mechanics

They are sums over the lattice, or a reasonably large fraction of the lattice points, of terms
involving only a few lattice sites, close to each other. These particular collective coordinates
satisfy commutation relations
i
[Σa , Σb ] = abc Σc (10.4)
N
and equations of motion (we set δH = 0 for this equation).
X
Σ̇a = Ba3b Σb + J ab3 σb (i)σ3 (j). (10.5)
<i,j>

The equation of motion relates macroscopic operators to other macroscopic operators, with
N independent coefficients. The commutators of macroscopic operators are very small, so
that their quantum fluctuations are of order √1N . This follows in a very general way for
operators that are defined as sums over all lattice points of products of local spins at only a
few neighboring points.
A further general property is that these macroscopic operators take on the same values
for an exponentially large number, ecN of the (large k) eigenstates of the unperturbed Hamil-
tonian. This is easy to see for the operators Σ, but is true for some large class of operators.
It is easy to see that our estimates did not depend on the detailed nature of the system, but
only on the following two properties

• The system consists of a large number N of mutually commuting variables, which

live on some kind of lattice (which need not be regular as long as it is regular when
averaged over large enough sizes). The Hamiltonian is a sum over the whole lattice, of
local operators; operators that depend on only one site and a few nearest neighbors.
For high enough excitation energy, with E − E0 scaling like N , the number of states
with energy between E and E ± δ scales like ecN .

• Collective coordinates Ci are operators which are averages of local operators over a large
fraction of the lattice. The generalized uncertainty relations imply that the uncertainties
in collective coordinates scale like √1N . The Heisenberg equations of motion for the
Ci (t), close on the Ci themselves, because of the local form of the Hamiltonian. For
a given initial state, which is a minimal uncertainty state for a set of Ci satisfying a
closed set of equations of motion, if we write

Ci (t) = hψ|Ci (t)|ψiI + ∆i (t),

with I the identity

√ operator, then ∆i (t) is an operator whose matrix elements remain
small for t  N . The expectation values ci (t) then satisfy an approximately closed
system of classical equations over these time scales. These define a classical history for
 Meaning of QM and Emergence of a Classical Reality  219

the system. It is important that the number of variables appearing in these classical
equations is finite as N → ∞. It is also clear that the number of states that share the
same classical history, in a band with high enough energy, scales exponentially with N .

As an example, we can consider the center of mass coordinate conjugate to the total momen-
tum of a bound system consisting of a large number of atoms. This satisfies

i
[Xi , Ẋj ] = δij ,
M~

∂V
M Ẍi = − .
∂Xi
The equation of motion for the expectation value of Xi is just the classical Newton equation.
Since M ∼ N and the potential is a sum of terms acting on individual atoms, the classical
motion has a time scale of order 1 and minimal uncertainties of order N −1/2 .
For bits of matter made of even the lightest elements, we have

10−4 cm2
δXδV ≥ .
N sec

A solid cube of side 10−1 cm will have N ∼ 1020 . Even if our initial state exceeds the minimal
uncertainty by four orders of magnitude, we can have a position uncertainty of 10−6 cm and
a velocity uncertainty of 10−14 cm/s in the initial state. The position uncertainty will grow
to the size of the object in a time of order 1013 s, if the bit of matter moves without the
influence of external forces. As we have seen, the uncertainty will grow even more slowly in
the presence of a potential.
Quantum uncertainties are thus quite small even for quite small macroscopic objects, but
don’t seem beyond the realm of measurement. The real problem is proving that the fluctu-
ations that we see are quantum mechanical in nature. Indeed, the first direct experimental
verification of the theory of atoms came from observations of Brownian motion; the small
fluctuations of the positions of dust particles suspended in a fluid. A perfectly adequate classi-
cal theory of these fluctuations was developed independently by Einstein and Smoluchowski.
It satisfies Bayes’ rule for conditional probabilities.
The classical theory of Brownian motion is “wrong,” since it treats atoms and molecules
as classical particles, but it is a valid emergent theory, because the quantum treatment of
the same problem reproduces the classical theory with incredible accuracy. To distinguish
between the two, we would have to observe not just uncertainty, but interference phenomena,
and the violation of Bayes’ rule, for the collective coordinates of the dust mote.
220  Quantum Mechanics

10.4 INTERFERENCE OF CLASSICAL HISTORIES

In the energy regime which we are discussing, interference is suppressed by factors of order
e−cN , rather than the mere power laws by which quantum uncertainty is concealed. The
suppression is related to the large number of independent quantum states that have the
same classical history. By arguments just like those we used to count the degeneracy of
energy eigenstates of high enough energy, the number of linearly independent states that
have the same classical initial conditions is exponentially large.
Now let us consider the time evolution of the system, starting from some generic state
in some large energy eigenspace of the unperturbed Hamiltonian. Label the collective coor-
dinates by Ci . Their expectation values will satisfy
d
hψ(0)|ei(H+δH)t Ci e−i(H+δH)t |ψ(0)i = hψ(0)|fi (Cj (t))|ψ(0)i, (10.6)
dt
and involve only functions of the other collective coordinates. Ci (t) is the Heisenberg operator
equal to Ci at t = 0. Because the quantum fluctuations of the Ci are small, these equations
are, with accuracy √1N , classical equations, which relate the expectation values to themselves.
This defines a classical history of the collective coordinates. For a given classical history of the
collective coordinates,√the time for which fluctuations falsify the predictions of the classical
equations is of order N t0 . The natural time scale of atomic physics is about 10−10 seconds,
so this is a time scale of order seconds for a piece of solid material 0.1 cm on a side. These are
very rough estimates, and for some collective coordinates, the time is much longer. However,
as we will see, the real parameter that determines classicality is not the size of the fluctuations.
Indeed, we know from the classical theory of Brownian motion that even a classical physicist
studying the interaction of a macroscopic variable with a system composed of many atoms
expects to see statistical fluctuations, coming from the unknown state of the atomic system.
To distinguish the quantum fluctuations from those predicted by such a classical model, we
have to observe interference phenomena between different classical histories.
In the N → ∞ limit, one can argue that there cannot be such interference. The equation
hψ(0)|Ci (t)|ψ(0)i = Cicl (t) (10.7)
is quadratic in |ψ(0)i and so does not define a subspace of the Hilbert space. However, when
N → ∞, one can define orthogonal subspaces of the Hilbert space in which the Ci (t) are all
simultaneously diagonal and equal to their classical values.
The claim is that when N is finite and large, a typical state with some classical history has
overlap e−N with one that has another history. There are three different kinds of arguments
for this, the last of which predicts an even smaller overlap.

• The first argument is pretty rigorous, but involves an assumption. It is easy to argue
that the number of linearly independent states with a given history is of order ecN .
 Meaning of QM and Emergence of a Classical Reality  221

For example, in our spin model, the constraint N1 hΣ3 i = M is one equation on the 2N
eigenstates of σ3 (i). We have seen that constraining the energy in a band high enough
above the ground state still leaves an exponential number of states satisfying the energy
and magnetization constraints. As long as the classical equations that determine the
history, effectively close on a finite number of expectation values  N , this counting
continues to work.

• Two typical states with the same expectation value ci (t), defining a classical history,
will have exponentially small scalar product. This is a simple consequence of the fact
that the space of states with the same history has an exponentially large dimension. The
scalar product of two randomly chosen vectors is thus exponentially small. Consider
some collective coordinate Ci whose local form involves only k sites, with k  N .
There is a basis of states, which have the form2
N/k
⊗p=1 |ψp i, (10.8)

where each state is acted on only by the operators in one of N/k disjoint clusters of
k sites. Ci is a sum of terms that act only within a cluster, and some that couple
different clusters. Our argument will ignore the coupling terms and is therefore not
rigorous unless k = 1. To change the expectation value of Ci by an amount of order
1, we have to change each of the states |ψp i into |ψp0 i with |hψp0 |ψp i| = cp < 1. The
QN/k
overlap of the two product states is then of order p=1 cp ∼ e−bN . Roughly speaking,
this argument says that two states with different classical histories will always have
exponentially small scalar product.

• The time scale of microscopic change of the quantum state of our system ranges from
much shorter than the time scale for the classical motions ci (t) to much longer. Time
scales in QM are determined by energy differences and in atomic systems, these range
from thousands of electron volts to e−cN 1, 000 eV (because so many states have to fit
into an energy band of typical atomic scale). This has two consequences. First, it is
almost never the case that the classical motion is so slow that it doesn’t affect the
microstate (so that the second item above becomes relevant). Second, the interference
terms in a QM probability calculation are time dependent, and average to zero over
the shortest time scale of the classical motion.

• Finally we want to mention the notion of environmental decoherence. We have been

talking about a small piece of solid matter, with N ∼ 1020 . Such systems are rarely
isolated from their environment very well, and coupling to the environment makes the
2
Recall the notion of tensor product from the chapter on Hilbert space.
222  Quantum Mechanics

effective value of N in the above estimates much larger. This is true even if we imagine
that the system discussed above was the needle on a dial on a device whose scale is of
order 10 cm and imagine that the device itself is sealed inside a completely shielded
laboratory, in a vacuum. When the needle moves from one position to another, it is
interacting with a different part of the device. So in most cases, taking N ∼ 1020 in
the estimates above, grossly overstates how big the interference terms are for different
histories of the expectation values of collective coordinates. Many discussions of deco-
herence emphasize the vastly more important effect of environmental decoherence, and
do not treat the decoherence of the collective coordinates of even tiny bits of matter,
20
due to their own internal structure. We prefer to emphasize the relatively small e−10
effect of internal structure because it shows how remote quantum interference effects
are, even for quite small systems.

10.5 SCHRÖDINGER’S CAT AND SCHRÖDINGER’S BOMB

We can now use these remarks to understand the mysterious “collapse of the wave function”
and “Schrödinger’s cat paradox” that haunt so much of the literature on the interpretation
of QM. Consider a microsystem like the two state Ammonia Molecule of Chapter 1, in a
superposition of states with the two orientations of the electric dipole

|ψiAmmonia = a|+i3 + b|−i3 . (10.9)

The dipole moment is uncertain in this state, and if we evolve the system with the Hamil-
tonian H = σ1 , we cannot describe the time-dependent probability distributions in terms
of probabilities for histories of the value of σ3 . This is because of interference terms. The
matrix elements of products of Pauli matrices in this state depend on a∗ b + b∗ a as well as
the probabilities |a|2 = 1 − |b|2 .
Von Neumann [26] was the first to realize that a model of the measurement procedure
was the coupling of this state to a macroscopic system, via perfectly unitary QM, to obtain
what is called an entangled state

a|+i3 ⊗ |C+ i + b|−i3 ⊗ |C− i, (10.10)

where C± refer to two positions of a macroscopic needle on a measuring apparatus. The

work on Decoherence theory [27] in the 1970s and 1980s made the crucial observation that
there were an exponentially large number of linearly independent states in the ensemble of
states that have the same expectation value for the needle position. As a consequence, for the
reasons sketched above, once the entanglement between the state of the ammonia dipole and
the apparatus has occurred, no future measurement will be able to detect the interference
between the two eigenstates of the dipole position. The ammonia Hamiltonian still acts,
 Meaning of QM and Emergence of a Classical Reality  223

but it is negligible compared to the interaction between the molecule and the apparatus. It
cannot change the position of the needle.
In other words, once a microsystem becomes entangled with a macrosystem in a way
that correlates its state with the expectation value of a collective coordinate, then Bayes’
rule and the emergent notion of probabilities for histories become applicable to the particular
ammonia molecule that was measured in a particular run of the experiment. QM predicts
that the needle will go up with probability |a|2 and we can decide to condition our predictions
for future experiments on the outcome of one particular experiment where the needle went
up. You can decide to get married as a consequence of the needle going up in a particular
run of the experiment to verify the quantum predictions for the ammonia molecule. The
quantum description of this sounds weird if you think of QM telling you about what goes on
in particular experiments. You are in a superposition of being married and not married.
This is a correct description of the quantum state of the world, but the quantum state is
nothing but a probability distribution. We have to ask whether there is anything about the
superposition of married/not married that is more disturbing than if you decided whether or
not to get married depending on whether or not Hurricane Katrina hit New Orleans (which, if
you lived in the Big Easy on the day of the hurricane, might have been a very relevant factor
in your decision about getting married). And since your marriage is in fact correlated with
the position of a macroscopic needle on a dial, Bayes’ rule is valid with incredible accuracy.
In addition, the position of the needle is predicted to be pretty close to its expectation value.
The fluctuations in the position of the needle are  10−10 of the central value, and because
of the validity of Bayes’ rule, they are identical to those of a theory that attributes the
fluctuations to unmeasured classical “hidden variables.”
From a practical point of view, you can treat the quantum prediction of your marital state
as you would the predictions of the weather equations for hurricanes: there are probabilities
for histories, and once we do an experiment, we know which of the histories “really occurred”
and we can throw away our probabilistic prediction for things that “did not happen,” and
renormalize our probability distribution so that, e.g., a = 0 and b = 1, if you did not
get married. The way we do this in QM is called “collapse of the wave function.” Classical
probabilists exploit the linearity of their equations for probability distributions to prove
Bayes’ rule and replace the probability distribution P1 + P2 , after a measurement consistent
with P1 but not P2 , by P1 (1 − P2 )−1 . In QM, we replace aψ1 + bψ2 by ψ1 (recall that
R

|a|2 + |b|2 = 1) to achieve the same end. And that end is to compute the probabilities for
future events conditioned on the result of some particular event that affected a collective
coordinate of a macroscopic object.
The philosophical stance behind the use of Bayes’ rule in classical probability theory is,
however, very different than that which underlies the collapse of the wave function in QM. In
classical physics, we subscribe to the belief that we could have predicted the precise behavior
of the system, if only we had been able to determine the initial state of all the microscopic
224  Quantum Mechanics

variables. The theory is supposed to predict the exact history of the system, in principle. In
this context, the use of Bayes’ rule is supposed to represent a refinement of our knowledge
about that exact history, and the act of throwing away the part of the probability distribution
that disagrees with what actually happened in a particular experiment is the correct thing
to do. It represents what we would have done at the outset, if we had known enough about
all the original initial conditions to make the correct prediction in the first place.
In QM, the conceptual role of Bayes’ rule/collapse of the wave function is quite different.
There are no exact histories. Probability is intrinsic and inescapable, and does not result
from our inability to measure all the relevant variables. It results from the fact that there
are no states of the system in which all the variables that appear in the equations of motion
have definite values. There are, however, macroscopic collective coordinates for which the
probability predictions of QM obey Bayes’ rule with fantastic accuracy. Quantum predictions
for probabilities conditioned on values of the collective coordinates are the same as those
in a classical statistical theory, up to corrections that are, in principle, too small to be
measured. Furthermore, the statistical uncertainties in these variables are small, so that
their expectation values determine approximate histories. Thus, we can use Bayes’ rule and
collapse the wave function, as long as we are always talking about the behavior of these
almost-classical variables.
There is a little thought experiment, which is a sort of combination of the double slit
experiment and the Schrödinger’s cat experiment, which illustrates how careful one must be
about the use of Bayes’ rule in QM. Consider an experiment, taking place inside a small
isolated laboratory out in intergalactic space. Like Schrödinger’s cat experiment, this one
consists of the correlation of the dipole moment of an ammonia molecule, with a macro-
scopic collective coordinate, this time the minute hand on a macroscopic clock. After the
measurement, the state of the system is

a|+i3 ⊗ |3 : 00i + b|−i3 ⊗ |3 : 30i. (10.11)

As before, there is an exponentially large number of clock microstates compatible with each
position of the minute hand. Decoherence makes it impossible to see interference between
the two parts of the wave function, and would encourage us, as in the classical argument for
the double slit experiment, to conclude that we can make arguments about predictions to the
future of this measurement, by saying, “Either the clock reads 3 or 3 : 30. The probability
of something happening in the future is the sum of the probabilities for what would have
happened given one of those two exclusive alternatives.”
However, what we have not yet told you is that the clock is actually the timer on an
explosive device, which is set to go off at 4 o’clock. It is a chemical explosion, which will blow
the entire laboratory into its constituent atoms, leaving no macroscopic trace. However, it is
not powerful enough to give nuclei in the debris relativistic velocities.
 Meaning of QM and Emergence of a Classical Reality  225

In order to understand the actual predictions of QM in this system, it is convenient

to use the language of Feynman diagrams. We have seen in the case of ammonia, that
atomic systems, consisting of charged particles, can emit photons. In the theory of quantum
electrodynamics, there is a probability amplitude for a charged particle like an electron,
propagating through space, to emit a photon, which is then absorbed at a later time by
another charged particle, let us say a proton. This amplitude is small, and the probability
for all possible things a pair of charged particles can do is approximately calculable in terms
of this single elementary pair of processes, with which Feynman associated a space-time
diagram, and a set of rules converting each element in the diagram to a part of the calculation
of the probability amplitude. The diagram is shown in Figure 10.1.
The electron propagates from its original position to a space-time point labeled by a
four-vector x, where it emits a photon. The electron then propagates to its final space-time
position, and the photon propagates to a point y, where it is absorbed by the proton. The
proton then propagates to its own final position. The points x and y are arbitrary, and part of
Feynman’s rule is that one must integrate the amplitude over all possible values of both x and
y. In particular, we integrate over the times the emission and absorption events could have
happened. Feynman’s prescription is backed up by a host of solid theoretical arguments, and
leads to a theory that provides the most precise agreement with experiment in the history
of science.
Let us apply Feynman’s prescription to our little thought experiment, asking for the
probability, far in the future, of an event in which a proton originating in the explosion
encounters a photon emitted in the explosion and scatters from it. The process of proton (or
electron) colliding with a photon is called Compton scattering. If there were only one time
at which the particles could have been emitted, then we would predict that this event could
never occur far in the future. The photons propagate away from the space-time event of the
explosion, at the speed of light, while the protons are much slower. At very late times, all
protons are separated from all photons by a large spatial distance.
However, Feynman’s prescription, assuming the initial state was the entangled state
above, with nonzero amplitudes for both positions of the hand of the clock, says that we
must sum the probability amplitudes over the two possible times for the explosion and then

e− e−

p p

Figure 10.1 Feynman diagram for virtual photon emission and absorption.
226  Quantum Mechanics

square to get the probability. There is an amplitude for a proton to be emitted if the explo-
sion occurred earlier, and an amplitude for a photon to be emitted in the later version, and
therefore there is an interference term in the amplitude for all possible things that can occur,
in which the early proton, scatters off the late photon. That amplitude would be proportional
to a∗ b + b∗ a and would vanish if we assumed that the wave function had actually collapsed
via some real physical process, at the time of the measurement.
In the statistical interpretation of QM that we have advocated in this book, there is no
real problem with interpreting this result in a way that is consistent with both our classical
intuition and the correct prediction about photon proton scattering probabilities. Given the
initial state of the ammonia molecule in a superposition of two directions for the dipole,
we get an unambiguous prediction of a nonzero scattering amplitude. However, we also see
that if we ask the question: what is the probability of proton photon scattering, given that
the explosion occurs at a particular time, we find the answer is zero. What this apparent
contradiction means, is that, after both explosions have definitely occurred, there is no way
to condition measurements on the explosion having occurred at a particular time.
If we insert another correlation with a macroscopic object into the system, then we have
a new set of variables to which we can apply the rules of Bayesian conditioning. That is,
if the explosion debris makes macroscopic changes in some distant piece of matter, then
we can again define conditional probabilities based on the time at which those changes
occurred. Amplitudes projected in this way will show no photon proton scattering amplitude
at late times, but this is because they are answering a different question: not, “what is the
unconstrained amplitude for photon proton scattering at late times given the initial state of
the ammonia molecule,” but “what is the amplitude constrained by the condition that the
distant piece of matter registers a macroscopic effect of the explosion at a particular time.”
If we interpret QM as saying something about definite results of particular experiments,
all of this begins to seem problematic. If we interpret it statistically then, as long as we specify
precisely what we are talking about, nothing weird occurs. The theory is tested by doing
multiple runs of the experiment and comparing the results to carefully phrased theoretical
calculations. We simply cannot use classical logic, and the idea that one of two macroscopic
things definitely happened in any one of the runs of the experiment to conclude what the
theory predicts will happen after all macroscopic trace of those things has disappeared. Or
rather, we can use such logic, but it pertains to uncheckable claims. The fact that a particular
run of the experiment led the clock to go off at a later time is something that we can use
to condition predictions about future experiments. It is sensible to use it as long as some
macroscopic record, for which the usual notions of history and conditional probabilities for
particular events in a history make sense. We could insist on continuing to use only the
branch of the wave function that predicted that particular sequence of macroscopic events
even after all macroscopic record of those events has disintegrated into microscopic particles.
It will predict correctly, that if there was a chain of macroevents in the past, in which the
 Meaning of QM and Emergence of a Classical Reality  227

explosion took place at a particular time, then, no matter which time it was, there is a
vanishing amplitude for any proton photon scattering long after the explosion.
If, on the other hand, we ask about the amplitude for postexplosion Compton scattering
of debris, with no conditions about what happened to macroscopic objects at intermediate
times, then there is a finite amplitude for the scattering to occur. The complexity of the
intermediate state, and the fact that our description of it in terms of collective coordinates
is very coarse grained, is not relevant to this computation. The complexity of the interme-
diate state is reflected in correlations between the single photon and proton states, which
participate in the scattering, and the multiparticle wave function of all of the other debris.
These correlations imply, with probability (1 − e−cN ) that the initial states of the proton and
photon are maximally uncertain, consistent with the fact that they collide,3 but still give a
probability of collision proportional to a ∗ b + b ∗ a.
This kind of experiment is called unhappening, a term which dramatizes the fact that our
assumption that the equations of physics, tell us in principle about things that are “really
happening” in a given run of an experiment is wrong. There can be no interpretation of
QM which retains the notion of probabilities for histories,4 except as an emergent concept,
applicable to the collective coordinates of macroscopic objects as long as those objects exist.
Given an initial state during the period such objects exist one can use Bayes’ rule to collapse
the wave function according to the observed behavior of the collective coordinates.
Once the collective coordinates disappear then predictions about the future must revert
to the initial uncollapsed wave function. This is completely bizarre, acausal and nonlocal if
one insists on thinking about the wave function as a real object, or even as a probability
distribution for histories of something. As long as one thinks about it as a probability dis-
tribution for instantaneous values of things that cannot all be certain at the same time, no
logical contradictions or bizarre behavior occurs. Things are never in two places at the same
time, they merely have probabilities of being in two places at the same time. As we have seen,
if one is too glib about the description of experiments, without acknowledging that exper-
iments always involve macroscopic objects, and paying careful attention to the differences
between probabilities conditioned on some macroscopic behavior and probabilities which are
not so conditioned, then one gets into conceptual trouble. It is our macroscopic condition-
ing; thought processes evolved in a world where everything of importance seemed to be a
macroscopic collective coordinate, that gets in our way of having an intuitive understanding
of quantum phenomena.

3
This means that, with probability of o(1), with no exponential suppression, the proton and photon are
in wave packets that collide with each other, but the exact form of the packet and the spin states of the
initial particles, are completely uncertain.
4
Unless, with the followers of Bohmian mechanics, one abandons Bayes’ rule for probabilities of histories.
This is a stance I find logically contradictory—a mere playing with words. See Appendix A on Interpre-
tation for a more detailed discussion.
228  Quantum Mechanics

In some distant future, if the dreams of both quantum computation and artificial intelli-
gence can be realized, the human race might find itself in conversation with sentient beings
who have an intuitive understanding of the rules of QM. Perhaps they would be able to
explain things better than I have. On the other hand, they might find themselves confronted
with the same problem we encounter, if we contemplate teaching calculus to a chimpanzee.
I will reserve further comments about the interpretation of QM to an appendix.

10.6 THE EINSTEIN–PODOLSKY–ROSEN (EPR) PARADOX, BELL’S THEOREM,

AND OTHER ODDITIES
In relativistic QM, elementary particles can decay into others because there is no separate
conservation law of mass. In particular, a massive particle at rest can decay into two light
particles with equal and opposite momenta because we can have
q
2
mc = 2 p2 c2 + m2e c4 , (10.12)

where we have called the light particle mass me because we want to take them to be electrons.
We will also make an assumption about the production mechanism of the electron pair,
namely that they are emitted with zero orbital angular momentum w.r.t. to the point of
emission, in the rest frame of the decaying particle. It follows from conservation of angular
momentum that the spin state of the electron pair is a singlet, if the decaying particle has
spin zero. The singlet state can be written in the basis where the 3 components of the spins
are diagonal as
1
|ψ0 i = √ (| + −i − | − +i). (10.13)
2

The electron and positron are in wave packets traveling in opposite directions at a speed
s = √ 2 pc 2 2 . They will travel a distance d = st in a time that the width of their packets
p +me c
spreads by an amount of order sm~de ∆  d. The fact that the pair has zero orbital angular
momentum means that there is an equally likely chance for them to be traveling in any
particular direction. We will be studying many decays, so we simply place a pair of Stern–
Gerlach machines at positions ±d along the 1 axis and do our measurements only on those
pairs which happen to pass through the machines. The thought-experimental setup is shown
schematically in Figure 10.2.
Recall from Chapter 7 that a Stern–Gerlach machine scatters incoming particles with
different spin components into different directions in space. Let us add to each machine an
absorber such that only electrons or positrons with n · σ = 1 get to our detector. When we
choose the unit vector n in the three direction we find the following interesting anticorrelation:
 Meaning of QM and Emergence of a Classical Reality  229

Figure 10.2 Experimental setup for Bell’s inequality thought experiment.

in each incident of decay5 only one of the detectors goes off, although conservation laws tell us
that particles entered both detectors. Which detector goes off is completely random, like the
flip of a fair coin. Einstein Rosen and Podolsky [30] considered this experiment to somehow
violate the principle of locality, since measurement of the spin of the electron immediately
determines what the spin of the positron was a space-like distance 2d away. According to the
principle of relativity, no signal can propagate between the two detectors in the arbitrarily
small time between the two measurements. In fact, we can duplicate the results of this
experiment with a purely macroscopic system, obeying the rules of classical statistics with
arbitrary accuracy. We simply place one black and one white ball in a sequence of boxes. The
box contains a random spinner, which rotates the two balls around inside and then opens its
±e1 faces and ejects whichever ball happens to be sitting there in the appropriate direction.
The ball ejected in the plus direction will be black, with probability 1/2 and there will be
an exact anticorrelation between the colors of the balls received at distant detectors in each
run of the experiment.
J.S. Bell pointed out what appears to be a more serious apparent violation of locality by
studying what happens when the unit vector n of one of the detectors points in an arbitrary
direction. The projection operator
1
P ≡ (1 + n · σ) (10.14)
2
is the quantity that determines whether the particle going through the rotated detector hits
the screen or not. Without loss of generality, we let the detector detecting the particle whose
spin appears to the right be the rotated one. There are now four possible outcomes: both
particles are detected, neither is detected, the unrotated detector detects a particle and the
rotated one does not, and vice versa. The corresponding projection operators are
5
We ignore the fact that the decay probability is isotropic in this part of the analysis. It can be taken into
account, with a lot of words, and no extra enlightenment.
230  Quantum Mechanics

1
(1 ± σ3 ) ⊗ (1 ± n · σ) ≡ P± ± .
4
Their expectation values are
1 2 θ
P+;+θ = P−;−θ = sin ( ), (10.15)
2 2
1 1 2 θ
P+;−θ = P−;+θ = − sin ( ), (10.16)
2 2 2
where θ is the angle by which the detector is rotated. By rotation invariance, we can think of
these probabilities as those for two detectors rotated into direction n1,2 with n1 · n2 = cos(θ).
In QM, one considers these probabilities as the probabilities that pairs of particles created
in decays of the heavy particle will register in two detectors oriented in the given directions.
This calculation is inconsistent with the assumption that one can assign a probability to
the history of every component of the spin of the electron for a given initial state. Suppose
we assumed that in every run of this experiment, the electron had some definite value of
n · σ, for every choice of the unit vector, which the theory could not predict, but to which it
assigns probability. We could classify each run as definitely having the value ±1, for every
value of θ. Then we would have probabilities P (±; ± θ) to take on each of these values, given
the value of σ3 for the other particle.
Now consider, according to the rules of classical probability theory, the probability
P (+; +θ = π/2) that the first particle can pass the unrotated detector and the second
can pass at θ = π/2. According to classical statistical thinking, we can divide all runs of the
experiment up according to the value of the spin of the particles in directions with θ = π/4,
even when we do not have an apparatus setup to measure the spin at this angle. Now according
to classical reasoning, we should have

P (+θ1 = π/4; +θ2 = π/2) + P (+θ1 = 0; +θ2 = π/4) ≥ P (+θ1 = 0; +θ2 = π/2). (10.17)

Here are the assumptions in the argument. In each run of the experiment, the spin component
of each particle in each direction has some definite value, with the two values anticorrelated
by angular momentum conservation. We are interested in runs that have the second particle
spin up in the π/2 direction, while the first is up in the 0 direction. Among runs satisfying
the former condition, the spin of the first particle in the π/4 direction could be either up
or down and the spin of the second is always anticorrelated with it. These particles have a
large space-like separation at times less than but of order d/s. But the probability P (+θ1 =
π/4; +θ2 = π/2) might have nonzero contributions when the spin in the θ1 = 0 direction is not
up, and P (+θ1 = 0; +θ2 = π/4) might have contributions when the spin in the θ2 direction
is not up. So, the inequality follows, since every run that contributes to the probability on
the right-hand side is included in one of the probabilities on the left.
 Meaning of QM and Emergence of a Classical Reality  231

If we plug in the quantum formulae for these probabilities, using the rule P (±θ1 ; ±θ2 ) ≡
P±θ1 ;±θ2 we get

P+θ1 =0;+θ2 =π/4 + P+θ1 =π/4;+θ2 =π/2 = sin2 (π/8) = .1464, (10.18)

while the left-hand side is

1 2
P+θ1 =0;+θ2 =π/2 = sin (π/4) = .2500. (10.19)
2
The inequality is clearly violated. Since it is violated by an amount of order 1, no quibbling
about imprecision of measurements will save us from this conclusion. If we insist on sticking
to a formalism in which there are probabilities for histories, one in which the statistical fluc-
tuations in the observed properties of particles are due to interactions with some “hidden
variables” whose state we do not observe, then those hidden variables must mediate interac-
tions between particles at space-like separation, in order to reproduce the quantum results.
At the time that Bell first announced these inequalities, experiments where the particles were
sufficiently space-like separated had not yet been done. Brilliant work by Aspect and others
[31] soon closed this loop hole. Quantum predictions are correct, and their interpretation in
terms of some sort of classical statistical theory must involve bizarre nonlocal interactions.
Anyone who has followed the exposition of QM in this book knows what is wrong with
the classical statistical reasoning that led to Bell’s inequality. QM produces probability dis-
tributions for every normal operator in Hilbert space. But it is impossible to find a state
in which all operators have definite values. Operators that do not commute cannot take on
definite values in the same state. This is the content of the generalized uncertainty relation.
(1) (2)
Our computations were done in a state where the operator σ3 σ3 = −1 and the operators
σ (1) + σ (2) = 0. Other combinations of Pauli matrices, such as the spins at various angles,
have only a probability distribution. The meaning of those distributions is that if we set up a
correlation between eigenvalues of one of those uncertain operators, and the collective coor-
dinates of a macroscopic system, then, by doing repeated runs of the experiment and keeping
track of the history of the macrosystem (which has an emergent meaning, married only by
terms in the quantum predictions that are exponentially small in the number of atoms in
the system), the quantum distribution reflects the frequency with which the macrodetector
will respond in a certain manner. The basic assumption of the classical analysis, that we
can assign values to all variables, independent of their coupling to a macrosystem, is simply
wrong. QM is an intrinsically statistical formalism, which only allows for the classical statisti-
cal notion of probabilities for histories as an emergent concept valid in limited circumstances,
which we have outlined in the early parts of this chapter.
Bell’s inequality shows that any attempt to duplicate the results of QM with some sort of
classical statistical framework will require the introduction of interactions that are nonlocal
in space and violate the principle that signals cannot travel faster than light. One might ask
232  Quantum Mechanics

whether Bell’s result implies some sort of nonlocality in quantum theory itself, and many
popular accounts claim that that is the case. This is simply wrong. Although we cannot
explore it in this book, the formalism of quantum field theory allows for a straightforward
generalization to take into account the constraints of special relativity and causality [17].
Roughly speaking, all one has to do is to take field equations covariant under Lorentz trans-
formations, and quantize them following the prescriptions in this book. The fundamental
postulate of these relativistic theories is that operators can be localized in regions of space-
time, and that the operators in two space-like separate regions commute with each other.
This implies that one can never set up EPR correlations, which will allow for superluminal
propagation of information.
There is an even simpler exposition of the basic quantum vs. classical conundrum revealed
by Bell’s inequality. This is due to Greenberg Horn and Zeilinger (GHZ), though the version
given here is due to David Mermin [32]. Extensive references to the literature on these exam-
ples can be found in Mermin’s article. The example is a simplification of earlier constructions
by Kochen and Specker, and by Peres. Consider three independent sets of Pauli spins, which
we can consider to be the spins of a widely separated triplet of spin 1/2 particles. We want
to ask the question of whether there could be some classical statistical explanation found
for the apparently random values found in experiments measuring various components of
the spin. So, we imagine that there are some other “hidden variables,” and that if we knew
the values of those variables, we would have a definite prediction for the numerical value of
every normal operator on the eight-dimensional spin space. We will call v(A) the value of
an operator A without labeling it by the values of the hidden variables. Our argument will
work for any choice of hidden variables.
QM predicts that mutually commuting operators can all have simultaneous values. Let
us consider the following group of 10 operators (we label the vector components of the spins
by x, y, z, and the particle labels by 1, 2, 3.

σx1 σy2 σy3 , σx1 , σy2 , σy3 . (10.20)

σy1 σx2 σy3 , σy1 , σx2 , σy3 . (10.21)

σy1 σy2 σx3 , σy1 , σy2 , σx3 . (10.22)

σx1 σx2 σx3 , σx1 , σx2 , σx3 . (10.23)

We have written 16 operators above but there are duplications and you can check that there
are only 10 independent ones. The reason for arraying them as above is that each row consists
of four mutually commuting operators. QM insists that these can all have simultaneous values,
and furthermore, that the product of the values in each row is 1. Any assignment of values
to the 10 operators, for fixed values of the hidden variables must satisfy this constraint, or
 Meaning of QM and Emergence of a Classical Reality  233

disagree with QM about the results of actual experiments. QM asserts in addition that each
of these 10 operators can take on only the values ±1.
On the other hand, the four triple products of spins all commute with each other as well
and their product is −1. Thus, we must have

−1 = v(σx1 σy2 σy3 )v(σy1 σx2 σy3 )v(σy1 σy2 σx3 )v(σx1 σx2 σx3 ) (10.24)

= v(σx1 )v(σy2 )v(σy3 )v(σy1 )v(σx2 )v(σy3 )v(σy1 )v(σy2 )v(σx3 )v(σx1 )v(σx2 )v(σx3 ) = 1. (10.25)
The last equality follows because the value of each individual spin appears twice. This con-
tradiction is avoided in QM because the different mutually commuting sets do not commute
with each other, and so cannot have simultaneous values.
In a hidden variable theory, one can predict probabilities for histories of the combined
quantum and hidden variables. In each of those histories, all QM operators are supposed
to take on values compatible with their allowed values in QM, and satisfying some of the
relations that QM predicts. Since QM has states in which all operators in a commuting set
take on definite values, it seems like a minimal requirement that the values assigned to these
operators at some time, in a particular history, satisfy relations like those above, which are
true in any state, for each possible collection of commuting operators. We have just seen that
this is impossible in a Hilbert space with eight states,6 and so it is impossible in any larger
Hilbert space.
The EPR/Bell arguments are based on a somewhat different criterion of what a minimal
set of requirements for a sensible hidden variable theory must be. They emphasize the require-
ment that one’s theory should not allow information to be transferred faster than light. Bell’s
inequalities show that this is impossible, and the three spin example above demonstrate this
in an even simpler manner, if we assume that the three spins are far removed from each other
in space. Let us assume we are in a state such that

σx1 σy2 σy3 |ψi = σy1 σx2 σy3 |ψi = σy1 σy2 σx3 |ψi = |ψi. (10.26)

This is possible because the three operators commute and each has eigenvalues ±1. The
product of the three operators is −σx1 σx2 σx3 , so |ψi is also an eigenstate of this operator,
with eigenvalue −1. Now imagine that the statistical ensemble defined by |ψi allowed for
the concept of histories of all of the operators σai (t) plus some hidden variables HA (t). The
probabilities are obtained by averaging over the hidden variables.
The state |ψi has correlations between the spins of the distant particles. One knows the
value of the y component of any one particle’s spin if one knows the x component of one of the
6
In fact it is already impossible for three states, though the proof is much harder. There is a simple four
state analog of what we have presented [32], but that example does not lead to a Bell/EPR paradox with
locality.
234  Quantum Mechanics

far away particles and the y component of the other. Such nonlocal correlations in space can
only be created, if there is no superluminal propagation of information, by events in the past
history of the system, and in a hidden variable theory, each complete past history gives us
definite values for all of the components of all three spins. The apparent quantum fluctuations
are supposed to come from averaging over the unmeasured behavior of the hidden variables.
The values of the spin components of each individual particle can be established by local
measurements, unlike, for example, the operator σx1 σy2 σy3 . Call these values sia . Furthermore,
these values, in any particular history which contributes to the statistical ensemble defined
by |ψi, must satisfy all of the exact relations between mutually commuting variables that
the quantum variables satisfy. Thus,

s1x s2y s3y = s1y s2x s3y = s1y s2y s3x = 1 (10.27)

and
s1x s2x s3x = −1. (10.28)

These are inconsistent, because the product of the four triplets of spins is equal to one since
each spin is ±1.
The conclusion of discussions of Bell’s theorem is usually stated as “there can be no local
hidden variable theory,” implicitly suggesting that some kind of nonlocal hidden variable
theory, consistent with the idea of probabilities for histories of microscopic variables might
be found in the future. The GHZ form of Bell’s theorem seems to rule out this possibility. It
says that even if we restrict attention to variables that we know can be measured, the spin
components of localized particles, we cannot consistently assign values to those variables in
any particular history, which are consistent with the values that QM assigns to collections
of mutually commuting (but nonlocal) operators, which we can construct as products of the
local variables. We have demonstrated this contradiction in a particular state, and there are
a variety of other states where the same contradiction exists. In more general states, one
must resort to inequalities of Bell’s type to find a contradiction.
In an Appendix, we will examine a variety of interpretations of QM, which purport to
make the theory more compatible with our intuition. None of them resolve the paradoxes
above.

10.7 THE RELATION BETWEEN THEORY AND EXPERIMENT IN QM

In our classical theory of the low energy ammonia molecule, the theory of measurement is
so trivial as not to require discussion. The system has only two states, which are completely
determined by the value of the quantity σ3 . Even if we have some uncertainty about the
initial state, the probability distribution p(σ3 ) is completely determined by doing repeated
 Meaning of QM and Emergence of a Classical Reality  235

measurements of σ3 . On the other hand, in the quantum theory, the general pure state is of
the form
α1 |+i3 + α2 |−i3 , (10.29)
where (α1 , α2 ) is a complex unit vector in two dimensions. In general, a choice of initial state
does not make a prediction for the value of σ3 . Repeated measurements of identically prepared
states, measuring only the value of σ3 , only serves to determine |αi |2 and do not determine the
relative phase. The reason for this, as we have just discussed, is that the term “measurement
of σ3 ” means entanglement of eigenstates of σ3 with the values of pointer variables of a
macroscopic system. By definition, pointer variables are shared by large ensembles of states,
with the property that the overlap between states with different values of the pointer variable
are doubly exponentially small. In the entangled state, it is, in principle, impossible to recover
the phase difference between α1 and α2 . Thus, in order to determine experimentally what
the quantum state is, one must not only do repeated experiments, but also measure different
noncommuting variables. Remarkably, this is a problem even if the initial state is almost an
eigenstate of σ3 . Consider the state
q
|+i3 + 1 − ||2 |−i3 , (10.30)

where  is a very small complex number. Measurements of σ3 are expected to give −1, with
probability 1 − ||2 . That means that we have to do of order (1 − ||2 )−1 measurements of σ3
in order to determine that  6= 0. On the other hand, measurements of σ1 will give a nonzero
average of order ||, which depends on the phase of . Thus, it is almost never possible to
actually determine the quantum state of a system by measuring only a single variable.
This discussion might make one skeptical of the possibility of ever preparing an initial
quantum state in a reproducible way. In fact, for quantities like the electric dipole of the
ammonia molecule, this is not an issue, because of rotation invariance. Given the assumption
that one can build an apparatus that can entangle the σ3 eigenstates with macroscopic
pointers, rotation invariance implies that we can do the same for every operator n · σ. This
then gives us a method for preparing eigenstates of n · σ for any direction in space, by
simply rotating our measuring apparatus, and throwing away runs of the experiment in
which the wrong value appears. Such polarizers have been known for photons since the mid-
19th century. The Stern–Gerlach experiments described in Chapter 7 show us how to prepare
quantum spin states of neutral molecules. Modern experiments in high energy physics have
managed to prepare quantum states of the spin of the electron.
Theoretical physicists often talk as if we can do the same thing for every Hermitian
operator in an arbitrarily complicated quantum system. This is highly unlikely to be true.
Nonetheless, it is tautologically true for any operator that can be measured. To say this
more precisely, A is an operator in some quantum system, and that system interacts with a
macrosystem in such a way that the joint system evolves from αi |ai i ⊗ |Readyi to
P
236  Quantum Mechanics
X
αi |ai i ⊗ |Pi >, (10.31)

where |Pi i are states in the ensembles with values Pi for collective coordinates of the
macrosystem. We have seen that the quantum predictions for the probability distributions
of such collective coordinates are, with incredible precision, consistent with Bayes’ rule and
the notion of probabilities for histories of the Pi . As a consequence, it is entirely consistent
with QM to consider a given run of an experiment, and define probabilities for the future of
that run conditioned on the value of Pi that appeared in that run. This procedure prepares
the state of the microsystem in the eigenstate |ai i of A.
Note that while we have used the word “measured” in the previous paragraph, no actual
laboratory, and certainly no conscious observer, is necessary to such state preparation. All
that is needed is a Schrödinger’s cat correlation between a property of the microsystem, and
the values of a collective coordinate of some macrosystem. A lot of the frustration of critics
of QM with the standard Copenhagen interpretation of the subject7 has more to do with
the use by the founders of the subject, of words like measurement, and observable rather
than with the actual physical content of that interpretation. Readers wishing to find a more
learned discussion of these issues along the lines presented here, but with a wealth of extra
detail and many references, would profit from the recent review article by Hollowood [33].
For the purposes of this textbook, it is now time to leave these interpretational issues and
return to the use of QM to describe phenomena in the world in which we live.

10.8 EXERCISES
10.1 Analyze Feynman’s description of the double slit experiment from the point of view of
this chapter. Is the wave phenomenon shown pictorially in Feynman’s lectures a real
wave, or a summary of the results of repeated double slit experiments with identical
initial conditions? Give a careful explanation of why the interference pattern disap-
pears when one, in each of those experiments, places a detector in the system with the
resolving power to tell which slit the electron went through.

10.2 Hyperion is one of the satellites of the planet Saturn. It is an irregular shaped rock
of size about 140 km, and its intrinsic rotational motion is chaotic. The tidal forces
exerted by Saturn and its other moons make the motion unpredictable. A simplified
model Hamiltonian for the motion of Hyperion was given by Wisdom et al. [35]:

L2 3π 2 a 3
H= − 2( ) (I2 − I1 ) cos(2φ − 2θ(t)). (10.32)
2I3 T r(t)
7
I feel that the account of QM in this textbook follows the Copenhagen interpretation, with additions by
the inventors of the theory of decoherence.
 Meaning of QM and Emergence of a Classical Reality  237

The Ii are the moments of inertia, with I2 − I1 > 0, a is the semimajor axis of the
orbit around Saturn and r(t), θ(t) describe the orbit, and are considered to be fixed
functions, determined from the observed orbit of the satellite. L is the third component
of the internal angular momentum of Hyperion. If R is the average radius of Hyperion
L
and we define x = Rφ and p = R , then because R is large (∼ 300 km), p is almost
continuous, even in QM.
Chaos of the classical motion implies that if we take a disk in x, p space, it turns into
an “amoeba” of equal area in a time of order tc . The length of the boundary of the disk
t
grows like e tc . For Hyperion, the orbital period T is a few days and tc is about 100
days. In the classical theory, an initial probability distribution becomes very stretched
out in the x variable in a time of order tc .
In QM, we can define coherent states of width δ
∞
δ2 m~ 2
eimφ e− ~2 (p− )
X
|x, pi = N R . (10.33)
m=−∞

If δ is small, these states have small position uncertainty. N is a normalization constant,

which guarantees that these are normalized states. Compute it.
If we take the momentum uncertainty in this quantum state to be of order the classical
~
thermal fluctuations in momentum on the surface of Hyperion, then δ = √mT . For
19 55 −2
Hyperion, m ∼ 10 kg ∼ 10 eV, and T ∼ 10 eV. Thus, the position uncertainty in
such a state is tiny, while the momentum uncertainty is what one might have expected
from classical physics. Such a state is quasiclassical, while a state with large δ has large
quantum fluctuations of the position coordinate.
Define a phase space density for any quantum wave function by

ρph = |hx, p|ψi|2 . (10.34)

Compute this density for |ψi = |x0 , p0 i, where both states have the same δ.
If the quantum motion stayed semiclassical for all times, as happens for harmonic oscil-
lator quantum states, then we would remain in a coherent state with small δ. We could
view the initial phase space distribution as representing classical uncertainty, i.e., lack of
knowledge about which coherent state the system was in, which would then propagate
forward in time using the classical equations for the coherent state parameters. For a
chaotic system, this is not consistent. Compute the time tq for the classical uncertainty
of the position to grow to size R, its maximal allowable value. At that time, either the
classical and quantum phase space densities are very different, or the system is no longer
in a coherent state with small δ. Since the experimentally observed distribution does
238  Quantum Mechanics

indeed spread, we conclude that Hyperion must be in a superposition of many small δ

coherent states at time tq . Compute tq and show that it is short enough that the actual
satellite we observe, must in fact be in such a quantum superposition. Argue that,
because the phase space variables x, p are collective coordinates of a very large object,
the quantum predictions of the superposition of coherent states are almost identical to
those of a classical statistical model. Estimate the size of the interference corrections
to those predictions. More details on this discussion of the dynamics of Hyperion can
be found in [33] and the references cited there.
 CHAPTER 11

Sketch of Atomic, Molecular,

and Condensed Matter Physics

11.1 INTRODUCTION
In this chapter, we will sketch the most important applications of quantum mechanics (QM),
to atomic and molecular physics, and the theory of complex materials, also called condensed
matter physics. These are huge subjects, and most of the literature on them is devoted to
either sophisticated approximation schemes or advanced theoretical discussions of special
corners of the field like high-temperature superconductivity or the fractional quantum hall
effect. In an elementary text, we can do no more than outline some major topics and ideas,
without much computational or theoretical depth.
We will begin by introducing the basic Bohr–Rydberg units, which control the energies
and length scales of all of these problems. Once we have written everything in terms of
dimensionless variables, the only large parameters in the system are the ratios between the
nuclear masses and the electron masses, as well as the charges of any large Z nuclei in the
material. In the limit where the latter are large, it is intuitively clear that only the last
few energy levels of the atom, where the electrons see only a screened Coulomb field, can
participate in any sort of collective low-energy excitation of the system. These are called the
valence electrons.
The Born–Oppenheimer approximation takes into account the lowest order in an expan-
sion of the energy in powers of the ratio of the masses of the electron and the nuclei of atoms,
numbers which are all less than 10−3 . At leading order, the nuclei are frozen into fixed posi-
tions, which are determined by minimizing the sum of their mutual Coulomb repulsion and
the ground state energy of electrons for fixed nuclear positions. This gives rise to the notion of
classical shapes for molecules and crystals. Many low-lying excitations of the system can be
understood as rotations of the entire molecule or vibrations of the nuclear positions around

239
240  Quantum Mechanics

me
their minimum values. These have quantized energy levels with spacings of order mN and
q
me
mN ,
in Bohr–Rydberg units, respectively. In crystals, the vibrational excitations of long
wavelength can have much lower energy.
Solving the problem of interacting electrons in an external potential is hard, and a number
of sophisticated approximation methods have been invented to solve it. The most commonly
used method, which is amenable to explanation at the level of this book, is Density Functional
Theory (DFT). We will give a rough sketch of DFT. We will also introduce a simple model
to explain the band structure of crystalline solids and the distinction between conductors,
insulators, and semiconductors. Finally, we will give the reader a hint of Landau’s Fermi
Liquid theory, a simple model with a remarkably wide range of applicability.

11.2 THE BORN–OPPENHEIMER APPROXIMATION

In the nonrelativistic limit, the Hamiltonian for charged point particles interacting via elec-
tromagnetic fields is given by
X P2 e2 X Zi Zj
i
+ , (11.1)
i
2mi 8π0 i6=j |Ri − Rj |

where −Zi is the charge of the i-th particle in units of the electron charge, and mi its mass.
In atomic, molecular, and condensed matter physics, the relevant particles are the nuclei of
the atoms in question, and the electron.
The simplest system of this type is the hydrogen atom, and we have seen that it defines
2
a characteristic energy scale, the Rydberg |E1 | = 2m~e a2 = 13.6057 eV, and a characteristic
B
2
−11
length scale, the Bohr radius aB = 4π
me e2 = 5.29177 × 10
0~
m. e is the charge of the electron
in SI units. Using these scales, we can rewrite the Schrödinger equation for any condensed
matter system in terms of dimensionless variables, which we denote by ri . The dimensionless
canonical momenta satisfy
[ra , pb ] = iδab . (11.2)
The rescaled Hamiltonian operator is
X me X 2Zi Zj
H= pi 2 + . (11.3)
mi i6=j
|ri − rj |

Its eigenvalues are energies in Rydberg units.

P
In a neutral condensed matter system, with N electrons, we have ZI = N, where ZI
are the nuclear charges. The kinetic terms of the nuclei are suppressed by the small numbers
me
mI . This suggests the following approximation, due to Born and Oppenheimer [36]:
 Sketch of Atomic, Molecular, and Condensed Matter Physics  241

• Fix the nuclear positions rI and solve the Schrödinger equation for the electrons, finding
the ground state energy for fixed positions E(rI ).
• Minimize E(rI ) w.r.t. the positions of the nuclei plus the Coulomb repulsion of the
nuclei.
• Expand E(rI ) around the minimum, obtaining a system of coupled harmonic oscillators.
The kinetic terms of the oscillators are of order me /mI and the potential terms are
o(1). Therefore
p the frequencies and the quantized energy levels of the nuclear motion
are of order me /mI < 2 × 10−2 in Bohr/Rydberg units. This is a small perturbation
to the value of E(rI ) at the minimum.
• We also find that the wave functions of
p the nuclei are Gaussian, centered around the
equilibrium positions, with width ∼ me /mI , so that we can think of the nuclear
positions as roughly fixed, giving the system a “classical shape.” This is the origin
of the little ball and stick models of molecules and solids, which used to populate
chemistry labs. Recall how these models led us to our first view of QM in the physics
of the ammonia molecule.
• Depending on the shape of the atom, molecule,
p or solid,1 there may be other energy
levels whose energy is much lower than me /mI . These levels are related to global
rotations and translations of the system.
We see that much of the quantum physics of nonrelativistic systems, interacting solely via
electromagnetism, is related to solving the problem of electrons interacting with themselves,
and with an external potential generated by the nuclei, viewed (approximately) as point
sources. To first approximation, we evaluate the ground state energy of the electrons for fixed
nuclear positions and minimize the resulting Born–Oppenheimer potential, VBO (rI − rJ ) with
respect to the nuclear positions. This results in a “classical shape” for the system. Corrections
to this are calculated by expanding the Born–Oppenheimer potential around its minimum,
leading to system of coupled oscillators with a Hamiltonian
KIa,Jb (rI − rI∗ )a (rJ − rJ∗ )b ,
X X
Hnuc = (me /mI )pI 2 + (11.4)
I I,J

where the stars indicate the positions of the minima. a, b are the spatial components of the
vectors in parentheses. The frequencies of classical motion of the normal modes of these
oscillators are obtained by diagonalizing the matrix
me me 1/2
[Ω2 ]Ia,Jb = ( )1/2 KIa,Jb ( ) , (11.5)
mI mJ
1
We are talking about the ground state of the system, for which almost all systems of large numbers of
atoms are in the solid phase.
242  Quantum Mechanics

me 1/2
and are of order ωtypical ∼ ( m I
) in Bohr/Rydberg units. The quantized energy levels are
me 1/2
half integer multiples of these frequencies. The ratio ( m I
) varies between .022 for Hydrogen
and ∼ .0014 for U238 .
Atomic energy differences, of order a Rydberg, correspond to emission of light in the
visible to ultraviolet part of the spectrum and can range into the X-ray region for atoms with
fairly large nuclear charge.2 This is the reason that we are able to identify elements, even on
distant stars, in terms of their visible spectra. It also indicates that the physical mechanism
responsible for vision in most animals must involve excitations of electronic energy levels in
the atoms of the visual receptors.
In contrast, the vibrational spectra of molecules are smaller by a factor ∼ .01 and cor-
respond to infrared (IR) light, which we feel as heat. This makes molecules potent agents
in the heat balance of planetary atmospheres. Solar radiation striking the rocky surface of
the inner planets is reradiated into the atmosphere, mostly as IR radiation because higher
energy photons are efficiently absorbed in solids that are not transparent (this is the funda-
mental physics behind the colloquial notion of transparency). That IR radiation will either
propagate out into space, or be absorbed by excitation of the vibrational levels of gases in
the atmosphere. It turns out that certain molecules, like CO2 and methane, have vibrational
level spacings that correspond to much of the IR radiation from the planet’s surface, and so
act as potent green house gases. Changes in the atmospheric concentrations of these green
house molecules will thus have dramatic effects on the energy content of the atmosphere,
which determines both the average temperature on the surface and the amount of energy
available for the creation of storm systems.

11.3 COLLECTIVE COORDINATES, ROTATIONAL LEVELS, AND PHONONS

Some of the oscillation frequencies of molecules and atoms are in fact identically equal to zero
if the system is isolated in infinite space. The original Hamiltonian of interacting electrons
and nuclei is invariant under simultaneous rotations and translations of all particle coordi-
nates. The Born–Oppenheimer potential, as a consequence, shares these symmetries. That
is, the potential is constant in the multinucleon configuration space, along the directions
where all the nuclear positions are translated or rotated simultaneously. These flat direc-
tions correspond to zero frequency oscillations in the expansion of the potential around its
minimum.
Viewed more globally, beyond the small oscillation approximation, the motions along the
flat direction of the potential are free motion and free rotation of the entire system. The
energies associated with collective translational and rotational motion come from the kinetic
term (me /mI )pI 2 . The translational energy can be set to zero, by making a Galilean boost
P

2
Recall that binding energy of the ground state of a heavy atom scales like Z 2 in Rydberg units.
 Sketch of Atomic, Molecular, and Condensed Matter Physics  243

−1
to the rest frame of the system. The rotational motions come from HR = (me /mI )Iab La Lb ,
where Iab is the moment of inertia tensor calculated from the classical Born–Oppenheimer
shape of the system and La the components of total angular momentum in the rest frame. The
energy levels are quantized, and we can understand the scale of the quantization by examining
the symmetric case where Iab ∝ δab . In this case, the rotational Hamiltonian is proportional
to the square of the angular momentum operator, whose p eigenvalues are l(l + 1). Thus, the
rotational splittings are down by another factor of me /mI compared to the vibrational
levels. For many molecules, including water at temperatures close to room temperature, this
gives photon frequencies in the microwave energy regime. The ubiquitous microwave ovens,
which inhabit most of our kitchens, operate primarily by exciting the rotational levels of the
water in various foods.

11.3.1 Water
The water molecule H2 O contains 10 electrons, so solving its Schrödinger equation is highly
nontrivial, even in the Born–Oppenheimer approximation. There are many sophisticated
calculations, which demonstrate that the B–O shape of the molecule lies in a plane, with a
bond angle between 110 and 100 degrees, in good agreement with experimental data.
The rough scale of rotational energy levels of water is mm e
H2 O
× 10 eV, or about 10−4 eV,
which is in the microwave range. Water vapor indeed accounts for much of the atmospheric
absorption of electromagnetic radiation in this range. Absorption of microwaves by water is
also the basic physical principle behind the microwave oven. Most of our foods contain a lot
of water and the ovens excite the rotational levels of those water molecules, uniformly in the
food sample. The dissipation of that rotational energy is what cooks the food.
Many of the most important properties of water, such as the fact that its solid form is
less dense than the liquid, near the freezing point, or its solvent properties, have to do with
the interactions between water molecules and are not understood on a quantitative level.
However, it is clear that the relatively weak binding of hydrogen to oxygen in individual
molecules is crucial. We know this because heavy water, where hydrogen nuclei are replaced by
deuterium behaves very differently. The larger mass of the deuterium atoms means that their
nuclei are more tightly bound to the oxygen atom. Indeed, there is experimental evidence
that the bond lengths in heavy water molecules are shorter than those in ordinary water
[41]. Note that in the Born–Oppenheimer approximation, the bond lengths are independent
of nuclear masses, as long as they are large. It is possible that the correct explanation of
these observations will involve a study of the effect of rotational levels of water on the
scattering experiments described in [41], rather than a breakdown of theqBorn–Oppenheimer
approximation. The size of the effect looks like something of order me /mproton . These
experiments also show that the properties of the liquid state of heavy water are different than
those of light water. Clearly, even for molecules as simple as water there is much to be done.
244  Quantum Mechanics

Indeed, replacing 25%–50% of a multicelled animals water content with heavy water leads
to a variety of toxic effects, from sterility to breakdown of numerous important reactions
essential to life. The effects are undoubtedly a result of the tighter binding, but detailed
mechanisms have not yet been worked out. Remarkably, single celled organisms like bacteria
seem to tolerate up to 90% replacement of their water supply by heavy water.

11.3.2 Phonons
For an isolated molecule, the translational collective coordinate is not very interesting. It
just describes the free motion of the molecule. We can always use the Galilean symmetry of
nonrelativistic physics, to study the system at rest, and obtain its properties in any other
reference frame by using Galilean transformations. The internal energies are unaffected by the
boost. However, there is a type of system for which excitations associated with translational
invariance are important and interesting. These are solids, which are the almost ubiquitous
state of matter at low enough temperature.
Consider a system of volume V in Bohr units, which contains N nuclei. We want to study
the ground state of this system in the limit V → ∞, with ρ ≡ N/V fixed. The nuclei repel
each other, but the electrons cancel that repulsion. The cancellation cannot be exact,
p because
we have seen that nuclear wave functions are localized at length scales of order me /mI in
Bohr units, while electron wave functions around low Z nuclei are localized only to within
Bohr radii. Even around high Z nuclei, only the innermost electrons are localized within 1/Z
of a Bohr radius.
The rough picture of the ground state of such systems is that the electrons mostly clus-
ter around individual nuclei, forming atoms or ions with small net charge, and the residual
attractive forces between these atomic/ionic constituents form a stable structure. The sim-
plest example is an ionic solid, like sodium chloride, in which ions of charge ±1 form a regular
cubic array. Thinking of the ions as classical point charges, this is clearly a minimum of the
energy. We will discuss the Density Functional Formalism, which can estimate the corrections
to this naive picture, in the next section.
This sort of crystalline structure exemplifies an interesting phenomenon in the limit
V → ∞ at fixed density. In that limit, the underlying Hamiltonian is invariant under con-
tinuous translations, but the crystalline ground state is not. If the crystal has a finite size,
V → ∞ at fixed but large N , then it has a translation collective coordinate, the position
of its center of mass, and the true energy eigenstates are states where the entire crystal has
fixed momentum. However, for the infinite crystal, there is no meaning to moving the system
around. It takes up all of space, and has infinite mass.
Instead, the infinite system has a degenerate set of decoupled ground states, in which
the position of any one nucleus is translated, with the entire lattice kept rigid. These states
are decoupled in the sense that every localized excitation of one of them is orthogonal to all
 Sketch of Atomic, Molecular, and Condensed Matter Physics  245

of the others. This phenomenon, in which a symmetry of the Hamiltonian of a system, is

realized as a transformation which moves one to a decoupled ground state in a Hilbert space
orthogonal to all local excitations of the original ground state, is called spontaneous breaking
of symmetry. It is quite common and pervades much of both condensed matter and particle
physics.
Now, let us think about the modes of oscillation of the nuclei around the stable crystal
lattice. They are described by displacement fields ∆(xI ). ∆(xI ) is the amount by which the
coordinate of the I-th nucleus is displaced from its lattice equilibrium position. If all the ∆’s
are equal, then the energy of the displacement is zero, because this is just the translation to
a new degenerate ground state. It follows that if ∆ varies very slowly over the lattice, these
excitations must have an energy that goes to zero. This spectrum of low-energy vibrational
excitations of a crystal is called the phonon spectrum. ∆, since it is small, will satisfy a linear
equation of motion, which is invariant under lattice translations. We can Fourier transform
the Heisenberg equation of motion for ∆ and get a harmonic oscillator equation, with a
frequency ω(k) that depends on the wave number, and goes to zero as the wave vector goes
to zero. This dispersion relation for the frequency has only the symmetries of the lattice. Thus,
∆ behaves like a quantized field. Phonons, the eigenstates of this field Hamiltonian, behave
like particles, in the same way that photons do, and are our first example of what condensed
matter physics call quasiparticles. Our system is made up of “fundamental” particles called
electrons and nuclei, but it also contains collective excitations with particle like behavior,
which do not have independent existence outside of the material whose low-energy behavior
they characterize.

11.4 THE HYDROGEN MOLECULAR ION

We now turn to a description of the simplest system to which our general remarks about
atoms and molecules apply: the hydrogen molecular ion. This consists of two protons and a
single electron. In our theoretical treatment, they interact only via Coulomb forces. There
are simple variations of this system in which one or both of the protons are replaced by a
deuterium nucleus.
The Born–Oppenheimer potential for the hydrogen molecular ion is invariant under simul-
taneous translation or rotation of the positions of the two protons. As a consequence, it
depends only on the distance between them. There is no potential energy for either the cen-
ter of mass coordinate or the orientation of the relative position vector. In the exercises, you
will show that the Hamiltonian for these variables is (in Bohr/Rydberg units)

me P2cm me 2
Hcoll = + L . (11.6)
2mP mP
246  Quantum Mechanics

The Born–Oppenheimer potential (which we will define to include the internuclear Coulomb
repulsion) will depend only on rp .
To compute the Born–Oppenheimer potential, we must solve the problem of an electron
in the potential
−1 −1
V (x) = + , (11.7)
x − dẑ x + dẑ
and find the ground state energy as a function of rp = 2d. Note that we have chosen the
separation between the protons to lie along the z axis. The electron Hamiltonian is invariant
under rotations around the z axis and we can diagonalize the z component of orbital angular
momentum L3 . It is also invariant under electron spin rotation. The energy is completely
independent of the spin, and depends on L3 through an additive term proportional to L23 ,
with a positive coefficient. The ground state is thus doubly degenerate in spin, and has
L3 = 0. The Hamiltonian reduces to
1 1
H = p2r + p2z − p −p 2 . (11.8)
r2 + (z − d)2 r + (z − d)2

In the two limits, d → 0 and d → ∞, this problem becomes exactly soluble. In the first limit,
it is just the single electron ion problem, with a nucleus of charge 2. The d → ∞ limit is a
little more subtle. If we consider an electron wave function localized near z = d, then the
problem reduces to that of the hydrogen atom in this limit. The same can be said for a wave
function localized near z = −d. The correct limiting wave function for the ground state is
the superposition of the two, with positive sign (see Exercise 11.1). The splitting between
the positive and negative superpositions is exponentially small at large d.
We conclude that the electronic contribution to the Born–Oppenheimer potential ranges
between −4 Rydbergs at d = 0 and −1 Rydberg at d = ∞. You will show in Exercise 11.3
that the increase with increasing d is monotonic. Thus, the electrons produce a net attraction
between the protons. The argument in Exercise 11.3 depends on a theorem due to Feynman
and Hellman, which we will prove in Chapter 13. This is an equation for the change of an
energy eigenvalue under a change in some parameter λ in the Hamiltonian.

∂λ E(λ) = hE(λ)|∂λ H(λ)|E(λ)i. (11.9)

You will calculate the leading correction to the −1 Rydberg at large d in Exercise 11.2. The
attraction cancels the protons’ Coulomb repulsion only partially, leading to a repulsive force
that falls off like 1/d4 . The repulsion also dominates at very small d, but it turns out there
is an intermediate regime where the attraction due to the electrons dominates, and there is
a local minimum at a point d∗ . We will estimate d∗ in the exercises.
Thus, in the Born–Oppenheimer approximation, the system of two protons and an elec-
tron has a bound state, in which the distance between the two protons is d∗ and the electron
 Sketch of Atomic, Molecular, and Condensed Matter Physics  247

wave function is localized in an ellipsoidal cloud whose semimajor axis is of order a few times
d∗ . The protons bind into a hydrogen molecular ion by virtue of “sharing” the electron in
order to lower the system’s energy. Chemists call this kind of binding a covalent bond.
We can make a crude model of molecular hydrogen by simply placing two electrons in
the same position space wave function as that found in the molecular ion, with their spins
in the antisymmetric spin-singlet state √12 (| + −i + | − +i). This model neglects the effect of
the electron–electron repulsion, on the shape of the electron wave function. A better model
might be ψion (x1 )ψion (x2 )f ( |x2a1Bohr
−x2 |
), where ψion is the single electron wave function in the
hydrogen molecular ion, and f is a smooth function vanishing when the distance between
the electrons is less than a few ionic Bohr radii, and is otherwise equal to 1. We will learn
how to optimize the choice of unknown functions in an ansatz like this in the chapter on the
Variational Principal.

11.5 DENSITY FUNCTIONAL THEORY

This section is meant to be a short summary of DFT rather than even an introductory
formal treatment of it. It is intended to outline the logic of the modern approach to atoms,
molecules, and solids. The Born–Oppenheimer approximation reduces nonrelativistic atomic,
molecular, and condensed matter physics to the quantum problem of electrons interacting
via the Coulomb potential, in a background external potential, which screens out the total
electronic charge. We can view this as the Homogeneous Electron Gas—an artificial system
with a constant background positive screening charge density, in an external potential. In
the real world, that external potential is the sum of nuclear Coulomb potentials, but it
is convenient to let it be a general function of position. If one can solve this problem for
every V (x), and calculate the ground state energy E[V (x)], then one can calculate VBO for
the nuclei as a special case, and begin the process of minimizing and expanding about the
minimum, which we sketched above.
No one knows how to solve the homogeneous electron gas exactly. For calculating VBO ,
the perturbation series techniques we will develop in Chapter 13 show us that we only need
to know how to calculate ground state expectation values of products of density operators
hψ0 |N (x1 ) . . . N (x1 )|ψ0 i. (11.10)
Here |ψ0 i is the ground state of the homogeneous electron gas with no external potential.
The density operator is defined by the equation
X
N (x) = δ(x − xp ), (11.11)
p

where the sum is over all electrons. This operator measures the electron density at the point
x in any state. Expectation values of products of it in the ground state of the homogeneous
gas tell us about the expected value of the local electron density, as well as its fluctuations.
248  Quantum Mechanics

In two remarkable papers written in the early 1960s, Hohenberg et al. [21] showed that
one could reformulate the calculation in terms of a functional of the expectation value of the
density operator. To understand them, we will need to borrow some wisdom from Chapter 18.
The variational principle discussed in that chapter shows that E[V ] is equal to the minimum,
over all normalized states, of the expectation value of the Hamiltonian
Z
H = HHEG + d3 x hψ|N (x)(V − V̄ )(x)|ψ0 i. (11.12)

Here V̄ is the integral of V over all space, and HHEG includes electron kinetic energy, electron–
electron Coulomb repulsion, and V̄ , a constant potential which cancels off the total electron
charge. HHEG treats the nuclear charge distribution as a smeared out homogeneous positive
charge density.
The key observation now is that the term in the expectation value of the energy, which
depends on V − V̄ , is sensitive to the quantum state of the electrons only through the
expectation value of N (x),
n(x) ≡ hψ|N (x)|ψi, (11.13)
in the test state |ψi.
This motivates a two step procedure for finding the minimum energy. First look only
at states that have the same expectation value, n(x) for N (x) and minimize the expecta-
tion value of HHEG among those states for fixed n(x). This calculation defines the density
functional F [n(x)]. Then the expectation value of the energy is
Z
E[n(x)] = F [n(x)] + d3 x n(x)(V − V̄ )(x). (11.14)

The ground state energy E[V ] is just obtained by minimizing this functional over all possi-
ble densities n(x). This constrained search for the minimum reduces the problem of atomic,
molecular, and solid state physics to a classical variational problem, once the density func-
tional F [n(x)] is known. What is remarkable about this result is that F is a quantity that
can be calculated in the homogeneous electron gas, without reference to particular choices
of the nuclei, which make up the substance of interest.
Of course, the problem of calculating F is not easily soluble. It was first approached by
Kohn and Sham, using the following sequence of approximations. First one writes a term
F0 [n], which is the value of the functional F for a gas of electrons with no Coulomb repul-
sion. This term incorporates the constraints on the density that arise from the Heisenberg
uncertainty principle. The high cost in electron kinetic energy suppresses densities that vary
on short wavelengths. The second term in the K–S approximation to the density functional
is called the Hartree term. It has the form
n(x)n(y)
Z
FHartree [n] = d3 x d 3 y . (11.15)
|x − y|
 Sketch of Atomic, Molecular, and Condensed Matter Physics  249

This term approximates the Coulomb repulsion between the electron density operators N (x)
by that between the expectation values, n(x) of those operators in the lowest energy state
at fixed n(x). The Hartree term can be justified in certain variational approximations to
the multielectron ground state wave function, which we will discuss in Chapter 18. The
Hartree approximation was widely used at the time K–S wrote their paper, and computer
codes had been written to explore its consequences. In a previous section, we showed how
it gave qualitative understanding of the structure of atoms and solids. It was never a huge
quantitative success.
K–S proposed to improve the Hartree approximation by adding a term
Z
FLDA = d3 x (n(x)), (11.16)

to the density functional. The function (n) is the ground state energy of the homogeneous
electron gas for constant density n. It is not calculable analytically, but both high and low
density expansions of it were known. Over the years, extensive numerical calculation [37] has
given us very reliable estimates of this function over the whole range of densities. The initials
LDA stand for local density approximation. FLDA would be the entire density functional if
spatial variation of the density did not exist in real substances. Roughly speaking, the K–S
idea was that the dependence on spatial variations captured by F0 and FHartree was enough,
together with FLDA to get a good approximation to the density functional for all materials.
To facilitate the calculation of F0 , as well as to exploit existing computer codes for the
Hartree approximation, Kohn and Sham wrote the density as

ψi∗ (x)ψi (x).

X
n(x) =
i

The summation index i runs over the number of electrons, K, in the system and ψi are chosen
to be K orthonormal single electron wave functions. This ansatz is motivated by the Hartree
approximation, in which the multielectron ground state is approximated by a product of
single electron wave functions. The ψi are called Kohn–Sham orbitals. One can show [38]
that if we write
∇2
ψi∗ (−
X
F0 [n] = − µ)ψi , (11.17)
i
2m
R
then the minimum of F0 [n] + FHartree [n] + FLDA [n] + V (x)n(x) with respect to the K–S
orbitals is the same number and gives rise to the same minimizing density, as direct variation
of the density functional w.r.t. to n(x). On the other hand, as you will verify in the exercises,
the variational equations with respect to the K–S orbitals are the same as the Schrödinger
equation for the lowest lying states in a self-consistent potential. Only the equation for the
self-consistent potential is different than the equation we wrote in Chapter 7, based on the
Hartree approximation.
250  Quantum Mechanics

It turns out that the K–S equations do not, in most cases, give results accurate enough for
the needs of chemistry and materials science. In the 1990s, a number of “semiempirical” terms
were added to the K–S expression for F [n] [39], which achieved the required accuracy. These
terms were motivated by theoretical considerations, but contain a variety of free parameters.
Those parameters are fixed by fitting of order 10 well-studied materials, and the resulting
density functionals then give excellent results for a host of other systems. A systematic review
can be found in [37].
DFT is a beautiful idea which, after empirically fit improvements, has become a powerful
tool in the study of materials. We still do not have a systematic way of calculating the density
functional from first principles, so there is a lot of both fundamental and practical work to
be done in this field.
In recent years, a different approximation scheme, called Dynamical Mean Field Theory
(DMFT), has challenged the DFT approach to condensed matter systems. Unfortunately,
the basic concepts underlying DMFT would take up too much time and require a level of
sophistication not expected of most students reading this book. Nonetheless, anyone inter-
ested in the application of QM to the physics of materials should be aware of the literature
on DMFT [40].

11.6 ELEMENTS OF THE THEORY OF CRYSTALLINE SOLIDS

If we study a very large sample of a homogeneous collection of molecules,3 we might expect
the ground state of the system to be approximately invariant under spatial translation.
Translation invariance is broken by the walls of the box containing the sample but if the size
of the box is much larger than the Bohr radius, this is a very small effect. Nonetheless, most
systems do not have translation invariant ground states.
Recall the separation of the electrons in such a system into valence electrons, and electrons
bound to the nuclei making up each molecule. In the Born–Oppenheimer approximation, the
nuclear positions are frozen. Each molecule is in some position. The energy is lowered by
amounts of order the volume of the system by allowing all of the molecules to be bound
together. One way to do this is to have different molecular ions with alternating positive and
negative charges. These are called ionic solids, and ordinary table salt is a common example.
If the separation between the positive and negative ions is too small, they will bind to form
neutral clusters and we will not get a solid. If it is too large, we do not get much lowering of
energy. The compromise is to have the ions sitting on some kind of regular lattice.
In our discussion of the hydrogen molecular ion, we have also encountered covalent bond-
ing, a quantum phenomenon in which two molecules are bound together because an electron

3
In this section, we will let the word molecule stand for either a bound state of some collection of atoms,
or a single atom.
 Sketch of Atomic, Molecular, and Condensed Matter Physics  251

is in a superposition of states close to each molecule. Again, there is a characteristic separa-

tion, of typical atomic scale at which the molecules prefer to sit, in order to lower their energy.
If we have a large collection of molecules, filling a box of volume V , the Born–Oppenheimer
potential will be minimized for a lattice configuration of the molecules.
What is a “lattice”? The Hamiltonian of the electrons and nuclei is invariant under
translations and rotations, which combine to form the three-dimensional Euclidean group. A
lattice is a set of points in space, invariant under some discrete subgroup of the Euclidean
group. A crystalline solid is a system of molecules whose ground state is invariant under
such a discrete subgroup. It is clear that the first step in the study of crystalline solids is the
mathematical problem of classifying all such discrete subgroups. There are 230 different types
of such space groups, which are symmetries of possible crystal lattices in three dimensions.
One of the primary applications of DFT and DMFT has been to determine the preferred
symmetry group for particular substances.
In order to determine the Born–Oppenheimer potential, whose minimization determines
the space group, one must solve the problem of valence electrons propagating in Coulomb
field of nuclei sitting at the minima of the Born-Oppenheimer (BO) potential. This is, in
principle, done by DFT/DMFT, but one also needs to understand that dynamics in order
to understand the nature of low-lying excitations in the system. These determine the crucial
transport properties of the substance: thermal and electrical conductivity, specific heat, etc.
The numerical cost and lack of intuitive insight that is involved in full scale DFT/DMFT
determination of transport properties leads one to search for shortcuts. For a given lattice,
there is a set of useful approximations, mostly developed prior to modern implementations of
DFT/DMFT codes, which gives a determination of electron propagation sufficiently accurate
to study transport properties. One first studies noninteracting electrons, propagating in an
assumed potential with the right symmetries. This leads, as we will see, to something called
the band structure of the material. The sophisticated numerical methods are necessary to find
the details of the band structure, but we can understand its qualitative nature in simpler
models. The energy level spectrum contains bands of states where the electrons are free to
propagate throughout the material and other bands where they are locked near particular
molecules. Depending on where the Fermi surface lies, the material will be a conductor or
an insulator or a semiconductor.
For a given material, if it were exactly pure, the number of valence electrons would
be completely determined by the geometry of the lattice of molecules, and the number of
valence electrons per molecule. However, we can vary the density of valence electrons by a
procedure known as doping.4 One substitutes a density of other molecules at a sublattice of
sites. This changes the average density of valence electrons, and gives us a slightly different
material which may exhibit sharply different transport characteristics. In theoretical models,

4
Which has no connection to the practice common in modern sports and horse racing.
252  Quantum Mechanics

one incorporates doping by adding a chemical potential term to the Hamiltonian for the
valence electrons. One can also induce interesting phase transitions by varying the doping
sites between a regular array and a random selection of lattice points.
Once one has determined the band structure, one uses what is known as the tight bind-
ing approximation. In crystals formed by covalent bonding, the electron wave functions are
superpositions of states localized near particular molecules. One introduces a lattice approxi-
mation to the quantum field theory of electrons, in which there are creation and annihilation
operators for electrons at fixed lattice sites, and one replaces the kinetic energy operator by
Z
ψ † (−∇2 )ψ → ψi† Kij ψj .
X
(11.18)

The hopping matrix Kij has the symmetries of the lattice. Now one also reintroduces inter-
actions. The most important of these is the remnant of the Coulomb repulsion between the
electrons
HHubbard = λ (ψi† ψi )2 .
X
(11.19)
The idea of this term, first introduced by Hubbard, is that the Coulomb repulsion between
electrons separated by a lattice site or more, is mostly screened, but that Coulomb repulsion
wants to keep electrons from occupying the same site. Note that the Hubbard interaction
would vanish if the electron did not have spin, and that for spin one half electrons there are no
higher order single site terms that we can add. In order to model the magnetic properties of
materials, one often adds nearest neighbor interactions between electron spin densities. These
are called Heisenberg interactions, and can explain ferromagnetism and antiferromagnetism.
Hamiltonians of the Hubbard type are known as effective low-energy field theories. They
aim to characterize universality classes of low-energy thermodynamic and transport behavior.
In principle, one would hope to derive them, or at least determine the values of parameters
like λ for a particular substance, by calculations based on more ambitious methods like
DFT/DMFT. In practice, much condensed matter physics is done by fitting parameters in
Hubbard-like effective field theories.
Although it is important to understand the context of these ideas, much of the technical
detail of the analysis of solids lies far beyond the scope of this book. We will be able to give
just a smidgin of the techniques involved in calculating band structure, and nothing of the
technicalities involved in analyzing the Hubbard model.

11.7 BAND STRUCTURE

Let us start with an intuitive picture that helps us to understand the existence of band
structure in crystalline solids.5 Consider a crystalline solid whose unit cells contain some
5
I would like to thank S. Shastry for explaining this to me.
 Sketch of Atomic, Molecular, and Condensed Matter Physics  253

kind of molecule, and imagine introducing an external potential that can rescale the crystal
so that the intermolecular spacing becomes larger or smaller. This is a mathematical device
and does not correspond to something we can arrange using electrical forces in the real world.
When the intermolecular spacing is made much larger than the Bohr radius, it is intuitively
clear that the electrons will want to clump into separate neutral molecules. The system will
be an insulator. Now bring the spacing down to something of order the Bohr radius. If we
consider just two molecules we know, from our discussion of the covalent bonding of atoms
into molecules, that there might be a possibility that the energy is lowered by putting some of
the electrons into linear combinations of wave functions localized near the different molecules.
If we have a whole lattice of molecules, we can lower the energy even more6 by using
superpositions of wave functions localized near all of the molecules. That is, we look at wave
functions of the form
1 X ik·xk
ψ(k) = √ e ψ(xk ). (11.20)
N xn
The position space wave function ψ(xk ) is the bound state wave function of the highest lying
single electron orbital in the molecule. It is bound to the center of mass of the molecule and
falls off exponentially with distance from the center. In principle, the allowed values of k are
restricted only by the boundary conditions at the edges of the large crystal, so they form a
quasi-continuum. The expectation value of the electron current in such an extended state can
be nonzero, if we allow boundary conditions which allow current to escape from the sample.
For some values of k, such extended eigenstates of the single electron problem will exist, and
for others they may not. However, since the k form an almost continuous set, small changes
in k will typically not change whether or not the extended states exist. Small variations in k
will give small changes in energy, so the extended states will lie in quasi-continuous energy
bands separated by quasi-continuous gaps in energy, where no extended states exist.
Turning off our artificial external potential, the same statements are true for the single
electron eigenstates in the self-consistent potential calculated from (some approximate form
of) DFT. Note that there is nothing in the above argument, which requires the potential to
be perfectly periodic, so the existence of band structure is robust, and will be valid in crystals
doped with impurities, where we can vary the electron density. As we do so, we change the
Fermi level and the crucial question is whether the Fermi level lies within a band, within a
gap, or near the edge between a band and a gap.
Materials where the Fermi surface lies in a gap are insulators, while those where it lies
in a band are conductors. In the latter, any small external electric field will induce a flow
of current, because there are a large set of extended states with nonzero current, which lie
close to the ground state and so can be excited by small external perturbations. Something
interesting happens when the Fermi surface lies just above or just below the line separating
6
Here we are implicitly invoking the variational principle.
254  Quantum Mechanics

a band from a gap. Near such a band edge, it is easy to move the Fermi level between
conducting and insulating states by doping. These materials have high electrical resistance,
but allow weak electrical currents to flow. These materials are called semiconductors.
If, in a semiconductor, the Fermi surface is just above a gap, then the number of available
extended states is small. This means that we can think of the excitations of the ground
state as a very dilute gas of individual electrons. The Pauli exclusion principle has little
effect and a classical model of nonrelativistic electron propagation captures the transport
properties very well. Note that this is very different from what happens for a metal, where
the Fermi surface is deep inside the conduction band. There, the charge carrying excitations
(see the section on Fermi liquids below) have only a single effective, almost continuous,
momentum component, perpendicular to the Fermi surface, and their energy is linear in this
momentum. For a semiconductor, the charge carriers are like single free electrons, except
that the effective value of the mass does not have to be equal to the electron mass in empty
space (the electron kinetic energy term is modified by interactions with the other electrons).
This sort of semiconductor is called n-type because the charge carriers carry negative charge.
Surprisingly, when the Fermi surface is at the top of a conduction band, almost identical
physics occurs. Although “all of the ground state electrons are in extended states,” the ground
state is stationary and no current flows. Current is carried by excitations of the ground state.
At the top of a conduction band, there are no low-energy excitations, which carry negative
charge. However, if we dope the material so that the Fermi surface is a little lower, we no
longer have enough electrons to fill the entire conduction band. Instead, we have a small
density of “missing electrons” or holes, which have very low energy, and can be in extended
states. These will also behave like nonrelativistic particles, but will carry positive charge.
This kind of material is called a p-type semiconductor.
Small amounts of doping can change the electrical conductivity of a semiconductor by
factors of order 103 –106 and this sensitivity makes them useful for all sorts of devices. Essen-
tially, though this is a simplification, these materials can act as on–off switches for the flow
of current. All of modern electronics is based on the properties of semiconductors [42].
Although our intuitive argument for the existence of band structure makes it clear that
exact periodicity of the potential is not essential, much of the early literature on the subject
is based on a mathematical theorem discovered by Felix Bloch. The group of symmetries of
the lattice includes an abelian subgroup of translations. The unitary transformations imple-
menting these symmetries on the Hilbert space of a single electron in a potential invariant
under the symmetries have the form eiK·an , where an are the discrete translations that leave
the lattice invariant. K is the wave number operator of Chapter 3. These operators can be
diagonalized along with the Hamiltonian and so we can study the spectrum in each eigenstate
independently. That is, we can restrict attention to wave functions satisfying

ψ(x + an ) = eik·an ψ(x), (11.21)

 Sketch of Atomic, Molecular, and Condensed Matter Physics  255

where k is a fixed number, called the Bloch wave vector. The set of all k forms a three-
dimensional torus, because if we shift k → k + r where r · an = 2πm for any integer m,
then the phase is left invariant. The discrete set of r satisfying this condition is called the
reciprocal lattice and the torus of allowed wave vectors is called the (first) Brillouin zone.

11.7.1 A Simple Model

To get a feel for how band structure works, we will work out the Dirac Comb model, intro-
duced by Griffiths [43], which captures the essential phenomenon. The original toy model for
band structure, the Kronig–Penney model [44] is somewhat more complicated. We consider
a one-dimensional problem, on a circle of circumference N a in Bohr units and a potential for
the dimensionless coordinate 0 ≤ x ≤ N , of the form
N
X −1
V (x) = α δ(x − ja). (11.22)
j=0

The wave function satisfies the periodicity condition ψ(x + N a) = ψ(x), which defines the
Hilbert space on which the Schrödinger operator acts. a is a number of order 1 and N ∼
1023 , as appropriate for a macroscopically large crystal. The solutions of the free particle
Schrödinger equation are eikx , with k = 2πn
N a , where n is any integer. They form an almost
continuous band in wave number space. We will see that the effect of the potential is to open
up gaps in this spectrum.
In the region 0 < x < a, we have the general solution to the free Schrödinger equation

ψ(x) = A sin(kx) + B cos(kx), (11.23)

2
where the eigenvalue in Rydberg units is E = k2 . For −a < x < 0, Bloch’s theorem tells us
that
ψ(x) = e−iKa [A sin(kx + ka) + B cos(kx + ka)]. (11.24)
The Schrödinger equation tells us that the second derivative of the wave function is a delta
function at x = 0, which means that the first derivative is a Heaviside step function and the
wave function itself is continuous. Thus, we get two conditions

B = e−iKa [A sin(ka) + B cos(ka)] (11.25)

and
2αB = k(A − e−iKa )[−B sin(ka) + A cos(ka)]. (11.26)
We solve the first equation by eliminating A in terms of B

A sin(ka) = [eiKa − cos(ka)]B. (11.27)

256  Quantum Mechanics

sin x
f (x) = cos x + β
2 x

−1

−2

0 π 2π 3π 4π

Figure 11.1 The eigenvalue equation for the Dirac Comb.

After substituting this into the second equation, we see that there is a common factor of B in
every term, because the wave equation is homogeneous. Thus, we get a consistency condition
relating the other parameters. After a bit of algebra, this simplifies to
α
cos(Ka) = cos(ka) + sin(ka). (11.28)
k
The periodicity condition on the wave function is solved by K = 2πn N , where n is any integer
≤ N . Thus, the allowed Bloch wave numbers practically form a continuum.
Now consider the solution of the eigenvalue condition for k given a fixed K. A typical
value of n will be  2πN , so the left-hand side of the equation is close to 1, and is certainly
below 1 in absolute value. On the other hand, the function on the right-hand side is not
bounded by 1. One can see the nature of the solution graphically in Figure 11.1.
As shown in Figure 11.2, this will cause gaps in k space where there are no eigenvalues.
These are interspersed with bands of almost continuous eigenvalues. We described in the
previous section how the positioning of the Fermi level of the material with respect to these
bands leads to a qualitative explanation of the variations in the electrical conductivity of
materials.

11.8 THE FERMI LIQUID THEORY OF CONDUCTORS

The first quantum theory of metals treated the electrons as a free gas enclosed in a box of size
L, neglecting both the nuclei and the Coulomb repulsion between the electrons. The ground
state is found by solving the free particle Schrödinger equation in a box, with energy levels
~2
E = 2m (k12 + k22 + k32 ) where each ki runs between 0 and ∞ in units of 2π
L . The restriction
to positive values arises from the fact that we impose vanishing boundary conditions on the
edges of the box, so that the solutions are sine waves. If we have N electrons with N  1,
 Sketch of Atomic, Molecular, and Condensed Matter Physics  257

band

gap

band

Energy
gap

band

gap

band

Figure 11.2 The energy spectrum for the Dirac Comb.

then the set of almost degenerate ground states fills up a volume in k space. Each allowed set
of wave vectors sits at a point in a three-dimensional lattice and we can associate the cube
in continuous k space formed by the three negative unit vectors emanating from that point.
3
The volume of this cube is Lπ 3 . For large N , the volume filled up by the states occupied by
all fermions is approximately an octant of a sphere of radius kF with the Fermi wave vector
given by

πkF3 N qπ 3
= , (11.29)
6 2L3
where the factor of 2 is inserted in the denominator of the right-hand side to account for the
fact that two electron spin states can fit in each state. q is the average number of valence
electrons per atom, a number of order 1. The Fermi wave number is given by

kF = (3ρπ 2 )1/3 , (11.30)

Nq
where ρ = L3 is the electron density. The corresponding Fermi energy is

~2
EF = (3ρπ 2 )2/3 . (11.31)
2m
2 2
The total ground state energy is given by integrating ~2mk over the octant of the Fermi sphere
(Exercise 11.8), resulting in
~2 (3π 2 N q)5/3 −2
E= L . (11.32)
10π 2 m
The fact that this depends on the volume, and decreases as the volume increases, means that
the gas exerts a pressure on the walls of the box, which can be computed via the first law
258  Quantum Mechanics

of thermodynamics (we are in the ground state, so the temperature is zero): dE = P dV . In

Exercise 11.9, you will verify that
~2 (3π 2 )2/3
P = ρ5/3 . (11.33)
5m
This is called degeneracy pressure. Its origin is quantum mechanical, a combination of the
uncertainty principle allowing smaller electron energy quanta in a larger box, and the fact
that there are more states below a given energy to put the fermions in when the box is larger.
This degeneracy pressure is crucial to understanding the bounds on stability of white dwarf
stars [45] (where however the motion of the electrons is relativistic), and is part of the story
of why ordinary matter is stable.
In a more realistic treatment of conductors, one should apply the above reasoning to the
extended electron states in a conduction band of the periodic Born–Oppenheimer potential.
The formula for individual electron energies will be different, but we still label the states by
an almost continuous wave vector k and an energy (k). The function (k) need have only
the symmetries of the relevant lattice. If, as a function of |k| at fixed angle,  is monotonically
increasing, then there will be a Fermi surface |kF |(Ω), analogous to the Fermi sphere of the
free electron system, delineating the maximum wave number at that angle, corresponding to
a filled single particle fermion state in the multiparticle ground state.
The symmetry constraints on  are weak, so we might imagine that shape of the Fermi
surface is quite nonuniversal. Landau [46] argued, however, that many properties of the sys-
tem depended only on the existence of the Fermi surface, and not on its shape. In particular,
the low-energy excitations around the ground state are characterized by a single parameter,
the distance from the Fermi surface, and the energy momentum relation for those excita-
tions is (generically) linear in this parameter. One can extract a lot of information about
the low temperature thermodynamics and the transport properties of the system from this
fact alone. We will compute the low temperature specific heat of a general Fermi liquid in
the next chapter. The “single particle excitations” of Landau–Fermi liquid theory should not
be thought of as single electrons. They are really collective excitations of the entire solid,
which have the quantum number of the electron. Landau gave these excitations the name
quasiparticle.
Landau–Fermi liquid theory has turned out to work with extraordinary universality. It can
be applied to any system build out of fermions, including those whose constituents are neutral
fermionic atoms or molecules. In recent years a much better theoretical understanding of why
Fermi liquid theory works so well [47] has been developed. An analysis too complicated to
summarize in an introductory textbook, shows that almost all possible interaction corrections
to a Fermi liquid picture, become weaker as one approaches the Fermi surface. The single
exception is an attractive two body interactions between quasiparticles. As first shown by
Cooper [48], such interactions lead to a bound pair of quasiparticles, which behaves very
 Sketch of Atomic, Molecular, and Condensed Matter Physics  259

much like a boson, and Bardeen, Cooper, and Schrieffer (BCS) [49] showed that the fluid of
these Cooper pairs undergoes Bose condensation. If the original quasiparticles are charged,
like electrons, this leads to superconductivity. The new, superconducting ground state has
an energy gap and is stable.7 The elegant BCS theory of superconductivity is one of the
triumphs of the application of QM to condensed matter systems.
In recent years, experimental condensed matter physicists have discovered more and more
new materials, where the Landau–Fermi liquid paradigm fails. The attempt to discover a more
general theory of the ground states of these exotic systems is still in its early stages. There is
experimental evidence of new classes of universal low-energy behavior but no comprehensive
understanding of the correct models, nor even of how many different universality classes
might exist. This is one of the most exciting areas in theoretical physics. An entry point to
the literature can be found in [50].

11.9 EXERCISES
11.1 Show that the ground state of the hydrogen molecular ion, in the Born–Oppenheimer
approximation, for large d is an even superposition of hydrogen wave functions localized
around the individual protons.

11.2 Calculate the leading large d correction to the Born–Oppenheimer potential for the
hydrogen molecular ion. Show that it leads to attraction (the energy is lowered by
making d smaller) and determine how it behaves as a function of d.

11.3 Prove that the ground state energy of electrons for fixed proton separation d is mono-
tonic in d.

11.4 In terms of the Born–Oppenheimer potential VBO (d) (assumed known), and the equi-
librium position d∗ estimated in Exercise 11.7, calculate the spectrum of oscillations of
the hydrogen molecular ion around its ground state.

11.5 Estimate the leading large d dependence of the Born–Oppenheimer force on the protons
in the hydrogen molecular ion.

11.6 Given the Hamiltonian

P12 P22 e2
Hion = + + + ER VBO ([X1z − X2z ]/aB ),
2mproton 2mproton 4π0 |X1 − X2 |

7
If the underlying quasiparticles are neutral, there is a gapless bosonic excitation, but a nonzero energy
gap for fermionic excitations.
260  Quantum Mechanics

where ER is the absolute value of the Rydberg energy and aB the Bohr radius, identify
the terms in the Hamiltonian that describe rotation of the molecule around its center
of mass and estimate the quantum mechanical rotation energies.

11.7 Show that the balance between Coulomb repulsion and Born–Oppenheimer attraction
leads to a minimum in the full potential for the protons.
2 2
11.8 Integrate ~2mk over an octant of the Fermi sphere to show that the ground state energy
of the free electron gas is
~2 (3π 2 N a)5/3 −2
E= L .
10π 2 m
11.9 Show that the pressure of the free electron gas is

~2 (3π 2 )2/3
P = ρ5/3 .
5m
 CHAPTER 12

Quantum Statistical Mechanics

12.1 INTRODUCTION
The key difference between classical and quantum statistical mechanics of particles has to do
with identity of particles. In both classical and quantum mechanics (QM), one can write a
Hamiltonian for an N particle system, which has an SN permutation symmetry exchanging
the particles. In both cases, we can consider states that are invariant under the symmetry, as
well as states that are not. There is no fundamental principle, enunciated in our description
of QM, which requires us to reject states that are not invariant under the symmetry group.
Rather, the experimental facts of the world we live in require this rejection.
The first indication of this was found by Gibbs, in his discussion of the entropy of mixing
between two boxes of gas. If a partition is removed between two boxes containing different
gases, then an elementary calculation shows that the entropy increases. There are more states
available. However, if we treat the molecules of a single gas as distinguishable particles, which
is to say particles whose states1 need not be symmetric, even though the Hamiltonian is, then
exactly the same calculation shows that the entropy increases when we remove a partition
between two identical boxes of gas. This does not agree with experimental observations on
the thermodynamics of uniform gases.
The solution, in both classical and quantum mechanics, is to view particles as localized
excitations of fields. In this interpretation, fields are fundamental, particles are special states
of such fields, and the identity of particle states under permutation is a consequence of the fact
that permutation of the particles gives the same field configuration. The details of how this
works in classical and quantum mechanics are quite different. In classical mechanics, particle-
like behavior implies that the state is eternally localized in space, and only special nonlinear
field equations have such soliton solutions. In QM, however, we have seen that eigenstates

1
Recall that in classical mechanics the state of a system is labeled by a point in its phase space, the space
of all solutions of the equations of motion.

261
262  Quantum Mechanics

of the quantized field Hamiltonian, for linear field theories, have a particle interpretation.
The field is an operator which adds or subtracts a particle to/from the system and the field
configuration associated with a particle is interpreted as the matrix element of the field
φs (x) = h0|φ(x)|si between the state |0i with no particles and a single particle state |si. For
linear field theories, this is equal to the single particle wave function of the particle state, and
the spreading of wave packets is understandable in terms of the fact that free particle states
are not localized eternally, because of the Heisenberg uncertainty relations. φs (x) cannot be
interpreted as a classical observable of the system. By contrast, coherent states, which are
superpositions of states with different particle numbers, DO behave classically in the fully
quantized theory.2
The lesson is that ALL particles in quantum field theory (QFT) are indistinguishable,
and experiment shows that the particle-like objects we detect in the laboratory behave like
the identical particles of QFT. From the field theory point of view, the SN symmetry of the
first quantized particle description of particles is a redundancy or gauge symmetry, morally
similar to the Maxwell gauge symmetry of electrodynamics. The profound difference between
ordinary and gauge symmetry transformations is illustrated by the difference between motion
on the plane with a potential invariant under rotations by the discrete angle 2π/N , and
motion on a cone, with that opening angle. The cone is, mathematically, just the plane with
points related by 2π/N rotations identified. We can study motion on the cone by studying
motion on the plane, but we must be careful to choose only solutions of the equations (either
classical or quantum) which are invariant under these discrete rotations. Similarly, if we
consider perturbations of the free motion, then if we are working on the plane we can add an
arbitrary potential, whether or not it obeys the symmetry, whereas on the cone, consistent
perturbations must be invariant. Similarly, an arbitrary perturbation of a free QFT, will
always produce multiparticle states of identical particles, which are invariant (except for
fermionic minus signs) under permutations.
The parenthesis above reminds us that we have already discussed the other odd QM twist
on field behavior, the existence of fermionic fields. The multiparticle states of particles created
by those fields are NOT invariant under permutations of particles, but pick up a minus sign
under odd permutations. This is allowable in QM if all physical operators (those which might

2
Some nonlinear field equations have localized classical solutions that are static or periodic in time. In
Chapter 17, we will learn about the JWKB approximation: if the Lagrangian of a classical system is
proportional to g −2 with g  1, then one can expand the logarithm of the matrix element of the time
evolution operator between two coordinate eigenstates, in a power series in g, the leading term of this
expansion is the classical action. In the context of the small g 2 expansion, the localized solutions of the
field equations can be interpreted as very heavy particles, if the semiclassical expansion parameter g is
small. They remain localized because in the g → 0 limit, they are infinitely heavy and their wave packets
do not spread. The discussion of these so-called soliton solutions lies far beyond the scope of this book.
 Quantum Statistical Mechanics  263

appear as a perturbation of the Hamiltonian) are even functions of fermion fields. In that
case, the minus sign is undetectable in any matrix elements of physical operators.
Fermi statistics is one of the most important features of the world we live in. It is respon-
sible, as we have seen, for the existence of a myriad of atomic nuclei, and for the periodic
table of the elements, which lies at the basis of chemistry. We will learn a little bit more
about the fundamental mechanism of Fermi statistics when we discuss the nonlocal phases
discovered by Aharonov and Bohm, in Chapter 16.
The point of this long introduction was to convince you that we should do the quan-
tum statistical mechanics of particles that have Bose and Fermi statistics, rather than that
which would follow the classical treatment of Boltzmann and Gibbs, who thought of identical
particles as limits of nonidentical particles.

12.2 QUANTUM FIELD THEORY OF FERMIONS

We have understood the proper treatment of bosons via the quantization of fields. A system
of noninteracting bosons with single particle energy eigenstates p has a Hamiltonian
X
H= np p . (12.1)
p

This formula generalizes the one we wrote for the solution of the quantum mechanical
D’Alembert equation. In that case, the labels p stood for three-dimensional momenta, p
and the energies were p = |p|. np is the number operator of the p-th harmonic oscillator. It
takes integer values between 0 and ∞ and can be written

np = a†p ap , (12.2)

where
[ap , a†q ] = δpq , (12.3)
and
[ap , aq ] = 0. (12.4)
The energy eigenstates are
|p1 . . . pn i = a†p1 . . . a†pn |0i, (12.5)
where
ap |0i = 0. (12.6)
For fermions, we want to obey the Pauli exclusion principle, restricting the values of the
number operator to np = 0, 1 and we want the multiparticle states to be antisymmetric
264  Quantum Mechanics

under permutations. The latter requirement is easily satisfied, by replacing commutators by

anticommutators. The anticommutator of two operators is AB +BA ≡ [A, B]+ . The equation

[a† (p), a† (q)]+ = 0 (12.7)

clearly implies antisymmetry of multiparticle states. It also implies a2 (p) = 0 = (a† )2 (p).
This is the Pauli exclusion principle.
Consistency with the last pair of equation implies that the commutator [a(p), (a† )2 (p)] = 0,
which is inconsistent with the commutator between creation and annihilation operators that
we used for bosons, as one can see using the Leibniz rule. The obvious replacement is that

[a(p), a† (q)]+ = δpq. (12.8)

The number operators n(p) = a† (p)a(p) still satisfy

[n(p), a(q)] = −δpq a(q), (12.9)

[n(p), a† (q)] = δpq a(q), (12.10)

so the Hamiltonian X
H= p n(p), (12.11)
p

describes the states of noninteracting fermions, with single particle energies p .

12.3 STATISTICAL MECHANICS OF BOSONS AND FERMIONS

The most general system of noninteracting particles will have a variety of single particle
energy eigenstates, labeled by a parameter p, which can be discrete or continuous and range
over a finite or infinite number of values. The space of p values might also have a geometrical
structure corresponding to some number of spatial dimensions. As an example, p could label
both the three vector momentum P and the J3 component of the spin σ of a nonrelativistic
particle of any spin j. The momentum would be discrete if space were a torus, and would
live on a torus, the first Brillouin zone, if space were a lattice. Denote the energy of the p-th
single particle state by p .
The N -particle states are labeled by N copies of the quantum number p1 , . . . pN , which
are either symmetric or totally antisymmetric under permutations of the copies. The key to
statistical mechanics is the realization that a nonredundant labeling of the states is just to
give the number of particles n(p) occupying each single particle state p. For bosons, n(p)
can be any nonnegative integer, while for fermions, it is either zero or one. For either type
of statistics, we have
[n(p), n(q)] = 0, (12.12)
 Quantum Statistical Mechanics  265

which means that the Hilbert space breaks up into a tensor product H = ⊗p Hp , where n(p)
acts as 1 ⊗ . . . n(p) ⊗ 1 . . . ⊗ 1. That is, n(p) acts in a nontrivial way only on the factor Hp .
Furthermore, because p is a full set of single particle quantum numbers, the states in Hp
are completely characterized by the eigenvalue of n(p). For fermions, Hp is two dimensional,
while for bosons, it is infinite dimensional.
In nonrelativistic physics, the numbers of each type of stable elementary particle (elec-
trons, protons, neutrons inside nuclei which do not undergo beta decay) are conserved quan-
tum numbers. From a relativistic perspective, this is because we never consider transitions
between energy levels differing by more than about 10−4 of the electron mass. For the pur-
poses of this chapter, we will consider only a single type of elementary particle, so that there
is only one such conserved quantum number
X
N= n(p). (12.13)
Conservation means that [N, H] = 0, which is evident from
X
H= p n(p). (12.14)

12.4 THE PARTITION FUNCTION

Macroscopic systems tend to come to thermodynamic equilibrium: given an initial
macrostate, the system quickly settles down into a universal state characterized by a few
parameters like temperature and pressure. The basic principle of Boltzmann–Gibbs statisti-
cal mechanics is that many macroscopic observables of a system composed of a large number
of elementary particles take on the same expectation values in all states which have the same
values of a few macroscopic conservation laws. We have discussed some of the mathematics
behind that in Chapter 10 on quantum measurement, but the derivation of the laws of sta-
tistical mechanics from those of QM is far from complete. Some guides to the literature can
be found in [51]. We will take the Boltzmann–Gibbs hypothesis as a true statement, but it
is worth outlining the basic lines of argument.
Consider a quantum system like the lattice system of Chapter 10. Its Hilbert space is
a tensor product of local Hilbert spaces, and the Hamiltonian operator and other collective
coordinates are sums of density operators localized near points in space. We showed that in
such systems, once we are sufficiently high above the ground state, the number of states in a
small energy band, of order the typical scales Etyp in the Hamiltonian, becomes exponentially
large in the volume of the system measured in atomic units. Consider an initial state |ψi =
cn |En i, where the coefficients are nonzero only in some energy band of order the typical
P

scale or a bit larger, in a region where the density of states n(E) = eS(E) is exponentially
large. The time evolution of the density matrix is
e−i(ωm −ωn )t cm c∗n |Em ihEn |,
X
ρ(t) = (12.15)
m,n
266  Quantum Mechanics

where ωm = Em /~. The time average of this density matrix over some interval τ is also a
−1
density matrix. If we take τ  ωtyp , then each individual term averages to (∆ωmn τ )−1 . So
terms with (∆ωmn τ )  1 become negligible. On the other hand, terms with (∆ωmn τ )  1
have almost no time evolution on time scales of order τ .
Recall that the extreme quasi degeneracy of levels, which leads to S(E) ∼ V /a3B , origi-
nated from the fact that we could make local changes of the state and raise the energy by
amounts independent of the volume. Collective coordinate operators, by their definition, are
insensitive to such local changes, with the sensitivity going to zero as the volume over which
the collective coordinate density is averaged goes to infinity. Let us define coarse grained
R E− ~
energy bands by the projection operators P (E) = E− ~τ de |eihe|. These bands are much
τ
smaller than the full range of energies contained in the initial state. The essential hypothe-
sis of quantum statistical mechanics is that collective coordinate operators act like the unit
operator within each band, up to corrections of order a3B /V . It follows that the time averaged
expectation values of products of these operators approach constants, at least on time scales
τ < t < eS(E) τ . The system is said to approach equilibrium in this time interval. Statistical
mechanics is the study of these equilibrium expectation values. We generally do computa-
tions in statistical mechanics by computing expectation values of local fields. This is a well
defined mathematical procedure, and it is usually used to compute only expectation values
of a product of a small number of fields. Although local field operators are not collective
coordinates, products of a few of them are insensitive to changes of the state of the system
far from the points where the operators act, so should also act, approximately, like the unit
operator in the subspaces corresponding to coarse grained energy bands.
What then is the equilibrium density matrix, to which the system relaxes for the long time
interval τ < t < eS τ . Since our argument did not depend on the details of the coefficients cn ,
one might expect that the resulting state was also independent of those details. This is the
Boltzmann–Gibbs hypothesis. Of course, if the system has conservation laws, then the values
of those conserved quantities cannot be changed by time evolution. The generic conservation
law in a quantum system with time independent Hamiltonian is the projection operator
on an energy eigenstate. We are interested in expectation values of operators that do not
make fine-grained distinctions between states in a given energy band with projector P (E).
Those expectation values are thus independent of the values of most of the conservation
laws. Furthermore, the initial state has been chosen to be a typical superposition of states
with different values of those fine-grained conserved quantities. Thus, one might imagine that
the density matrix depends only on the values of conserved quantities which are themselves
collective coordinates, namely energy, momentum, and angular momentum.
We are typically interested in systems at rest in the laboratory, which are enclosed in a
box, so rotation and translation invariance are broken and the only conservation laws are
energy and the number of each species of particle, if those particles are in a nonrelativistic
 Quantum Statistical Mechanics  267

energy regime, where particle creation can be neglected. We can also consider electric charge,
baryon number, and lepton number to be conserved, even in the relativistic regime.3 Boltz-
mann and Gibbs tell us that we should evaluate equilibrium properties of the system by
averaging the expectation value over all states with fixed energy and particle number.4 In
quantum mechanical language, the Boltzmann–Gibbs hypothesis is that the density matrix
of an equilibrium system is
ρeq = δ(H − E)δN n , (12.16)
where E is a real number and n a positive integer. This is called the microcanonical ensemble.
One simple way to explain why a system might be in such a state is to assume that it
is maximally entangled with a much larger system. If we label a basis of states of the small
system by i and those of the large one by I, then a typical state of the combined system (the
tensor product of the individual Hilbert spaces) is
X
|ψtyp i = ciI |i, Ii, (12.17)

where ciI is a maximal rank complex matrix satisfying

X
|ciI |2 = 1. (12.18)
i,I

We have 1 ≤ i ≤ n and 1 ≤ I ≤ N , with N  n. If we make no measurements on the larger

system, the state of the smaller system is uncertain and its density matrix is

c∗iI cjI .
X
ρij = (12.19)
I

In Exercise 12.1, you will verify this is a positive Hermitian matrix with trace equal to 1.
Now consider the equation
U c ∗ = c ∗ VU , (12.20)
where U is an n × n unitary matrix and VU an N × N unitary matrix. For each choice of
U , the number of equations for VU is much smaller than the number of unknowns, so we can
always satisfy them. Therefore,

U ρU † = c∗ VU VU† cT = ρ. (12.21)
3
It is believed that baryon and lepton numbers are violated by small effects. The equations of the standard
model of particle physics then imply that the sum of baryon and lepton numbers is violated. Many models
that attempt to explain the small mass of neutrinos violate the difference, B − L, as well.
4
For simplicity of exposition, we will assume that our system consists of a single type of nonrelativistic
particle. The generalization to multispecies models should be self evident.
268  Quantum Mechanics

Since ρ commutes with every unitary matrix it is proportional to the unit matrix, which
means that the probability distribution for states of the smaller system is maximally uncer-
tain.
If we choose a random state in the full Hilbert space then the probability that ciI is not
of maximal rank is the probability that N  n vectors of dimension n span a subspace of
dimension < n. This goes to zero like a power of n/N , depending on how small a dimension
we ask for. For macroscopic systems, n is exponentially large in the volume of the system
measured in microscopic units like the Bohr radius. Thus, the expected deviations from a
maximally uncertain density matrix are exponentially small, of order e−(VN −Vn ) , where V is
the volume of the indicated system in atomic units.
We have just sketched a proof of Page’s theorem [52], which tells us that a small system
that has become entangled with a larger system is exponentially likely to be in a maximally
uncertain state consistent with macroscopic conservation laws. We have noted, however, that
most of the conservation laws are broken by the macroscopic box that we put the system in, in
order to isolate it from its environment. Energy conservation is not destroyed by the box, and
energy appears to be conserved on all time scales much shorter than the age of the universe.5
However, the small and large systems are entangled either by a boundary condition at the
beginning of time, or, more reasonably, via some interaction between them. This means that
the energy of the small system itself is not conserved, so that the microcanonical ensemble
is at best an approximate concept.
Boltzmann was the first to study, within classical mechanics, this situation of a system
coupled to a heat bath; a much larger system with only weak interaction with the system
of interest. The reason for the weak interaction can be thought of as a combination of
geometrical and mechanical factors. The two systems interact only at the boundary of the
box in which the substance of interest is contained. Furthermore, one can imagine designing
a box for which energy transport through the relatively thick walls is much less efficient than
processes which reemit the energy from the walls, into the system of interest. Let us write
the total Hamiltonian as
H = Hsys + Hint + HHB . (12.22)
An eigenstate of the total Hamiltonian with eigenvalue E0 can be written as an entangled
state of different eigenstates of Hsys , and by Page’s theorem this is extremely likely to be the
maximally uncertain density matrix on the Hilbert space of the system, constrained only by
the total energy of system plus heat bath. The heat bath has a much larger energy than the
system, so the probability of finding the system with energy E is zero above the eigenvalue
and is a rapidly decreasing function of E as this bound is approached. Indeed, the total
energy of the heat bath is much larger than the system energy E, and its number of states
5
In making this statement, we are taking the origin of time somewhere in the current slowly expanding
period of universal history.
 Quantum Statistical Mechanics  269

close to that energy is also much larger. Thus, we can write the logarithm of the number of
states at energy E0 , each of which is approximately6 the tensor product of an eigenstate of
the heat bath with energy E0 − E, with a particular eigenstate of the system with energy
E as
dL(E0 ) E
l(E) = L(E0 ) − E ≡ L(E0 ) − . (12.23)
dE0 kT
Here L(E0 ) is the logarithm of the number of states of the heat bath with energy E0 and we
have assumed that it increases with energy. This formula is the statistical mechanics definition
of the thermodynamic variable called temperature. k is Boltzmann’s constant, which is equal
to one if we use energy units for temperature, rather than conventional degrees. Thus, the
probability of being in a particular eigenstate of the system, with energy E  E0 varies as
E
e− kT which is Boltzmann’s famous statistical law.
One the other hand, as we have seen in Chapter 10 on quantum measurement, the number
of states of the system with energy E rises rapidly with E. We have also seen that the total
energy of a macroscopic system in a typical excited state is of order the volume V of the
system. Dimensional analysis says that this is multiplied by an energy density, , which has
dimensions of energy per unit volume. The entropy, or logarithm of the number of states of
the system with fixed energy and volume, is also proportional to the volume. If we are in
an energy regime above the scales that appear in the Hamiltonian of the system, then the
entropy density must scale like 3/4 so the number of states in volume V behaves like
 3/4 1/4 ( E )3/4
n(E) ∼ eV ( ~v ) eV ~v .

The velocity v is the maximum velocity at which signals can propagate in the system. It
appears, along with ~, in order to get a dimensionally correct formula independent of most
system parameters, which is what we expect in this energy regime. At lower energies, we
expect similar exponential growth with a fractional power of the energy. These expectations
are based on the spin model of Chapter 10 and many similar models. The falling exponential
Boltzmann weight combines with the rising density of states to make a function sharply
peaked at an energy of order V T 3 (we have set ~ = v = k = 1 here). It can be shown
that statistical averages with this probability distribution give results identical to those in
the microcanonical ensemble at that energy, in the thermodynamic limit in which V goes to
infinity. This new canonical ensemble is much more convenient to use, and much closer to the
actual state of most physical systems. The density matrix associated with it is β ≡ (kT )−1 .

e−βH
ρcan = , (12.24)
Z
6
This approximation is valid because of the weak coupling between the system and the heat bath.
270  Quantum Mechanics

where Z = Tr e−βH , is called the partition function. It turns out to be a convenient tool for
quickly calculating thermal averages. For example,

hHi = −∂β ln Z. (12.25)

The canonical partition function for an M particle system is

ZM (T ) = Tr e−βH PM (N ). (12.26)

Here β −1 = kT , where k is Boltzmann’s constant. This formula is written in what is called

the Grand Canonical Hilbert space in which each of the operators n(p) is allowed to take
on any of its allowed values, so that the number operator N can be any nonnegative integer
(it will have a maximum for fermions if there are only a finite number of values of p). The
operator PM (N ) = δ0, N −M (Kronecker delta) projects on the subspace on which N = M .
The calculation of ZM (T ) is a bit complicated because the projection operator couples
together all of the tensor factors. For large systems, it is also calculating a quantity that
is hard to measure. We never have microscopic control over the particle number, just as
we never have microscopic control over the total energy of a large system. In both cases,
we introduce a control parameter which weighs different energy or particle number sectors
in such a way that the statistical fluctuations away from the mean are small. This is the
mathematical meaning of temperature in the case of energy fluctuations, and corresponds to
replacing the microcanonical by the canonical ensemble. The corresponding parameter for
particle number is called chemical potential. It is denoted by µ and is defined by the Grand
Canonical Partition Function:

Z(µ, T ) = Tr e−β(H−µN ) . (12.27)

This is simple to compute, because the trace of a tensor product of operators is the product
of their traces (Exercise 12.2). Thus,

Tr e−β(p −µ)n(p) .
Y
Z(µ, T ) = (12.28)
p

(1 ± e−β(p −µ) )±1 .

Y
Z(µ, T ) = (12.29)
p

The plus sign refers to Fermi–Dirac statistics and the minus sign to Bose–Einstein. A useful
mnemonic for remembering the sign is that it is the same sign that appears in the commu-
tation relations
aa† ± a† a = 1 (12.30)
for the creation and annihilation operators for fermions and bosons.
 Quantum Statistical Mechanics  271

The equation fixing the expectation value of the total number of particles is

∂ln Z X e−β(p −µ)

hN i = β −1 = −β(p −µ)
, (12.31)
∂µ p 1±e

while that fixing the energy expectation value is

∂ln Z X e−β(p −µ)

hEi = = p . (12.32)
∂β p 1 ± e−β(p −µ)

Note that, in the limit of high temperature, when β → 0, the term in the denominator, which
differentiates between bosons and fermions, is negligible. In this limit, we get formulae that
are the same as those Boltzmann derived in classical statistical mechanics, except that we are
instructed to treat configurations which differ only by the exchange of identical particles as
the same configuration. This inserts a factor of M1 ! in the partition function for M particles,
and corrects the wrong prediction of an entropy of mixing for identical particles.

12.5 THE LOW TEMPERATURE LIMIT

In the low temperature limit, noninteracting bosons and fermions behave quite differently.
For fermions, the formula for the expectation value of np in the Grand canonical ensemble is

e−β(p −µ)
hnp i = . (12.33)
1 + e−β(p −µ)
Thus, when β → ∞, the expected occupation number for every level with  − µ > 0 goes
to zero. Note that µ is chosen to fix the expectation value of N , the number of fermions.
In the low temperature limit, every state below µ has expectation value 1 for np . This is
the maximum value that operator can take, so we are in a state where that level is filled.
Thus, the fermion ground state, which is unique if the eigenvalues p are nondegenerate, is
the state in which the lowest M single particle states are occupied and all others have no
fermions in them. This is a pure state. The maximum occupied eigenstate is called the Fermi
Level. The concept of Fermi level and the nature of the low lying excitations just above the
Fermi level is among the most important in the physics of materials. Remarkably, it is valid
in a wide range of materials, even though electrons in solids are far from noninteracting.
The underlying reason for this is Landau’s Fermi Liquid theory [46] [47]: one can show that
the low lying excitation spectrum of a system of fermions is determined to a large extent
by the shape of a surface (p) in momentum space. Fermionic excitations near this surface
are called quasiparticles. They are not individual electrons, but a sort of fermionic collective
excitation of the system of interacting electrons, and one can show that the interactions of
272  Quantum Mechanics

these quasiparticles near the Fermi surface are very weak. Unfortunately, this topic is too
advanced for a first course in QM, but serious students of the subject will want to learn
about Fermi liquid theory as soon as they can.
More recently, we have discovered classes of materials, including the superconductors
with the highest known transition temperatures, which are not well described by Fermi
liquid theory. This is an area of intense current interest [50] and it is fair to say that we do
not yet have a comprehensive theory of these materials. The exploration of their properties
is one of the most interesting areas of modern physics.
The behavior of noninteracting Bosons at low temperature is even more remarkable than
that of Fermions. The expectation values of the number operators are
e−β(p −µ)
hnp i = . (12.34)
1 − e−β(p −µ)
In this case, if (p − µ) < 0 the denominator can have poles, and the expectation values blow
up. Of course, as the temperature is lowered at fixed chemical potential, this will happen
first for the single particle ground state.
P
Of course, if the expectation value of the total number operator, N = np is fixed and
finite, we cannot have a divergence in any single occupation number. In real systems, the
number of particles in a finite volume is always finite, and hN i = ∞ is only realized in
the thermodynamic limit of infinite volume. This sort of singularity in the thermodynamic
limit is called a phase transition, and this particular phase transition is called Bose–Einstein
condensation (BEC). Its real meaning for a finite system is that a macroscopic fraction of
the particles in the system are in their single particle ground state. That is hn0 i ∼ V /a3 ,
where a is an atomic length scale of order the Bohr radius.
For many years after the theoretical discovery of BEC in the noninteracting Bose gas,
the only experimental model of the phenomenon was the superfluid behavior of liquid He4 .
He4 is in fact a fairly dense fluid, rather than a dilute gas, and the noninteracting model is
not really a good guide to its detailed properties. In 1995, Cornell, Weiman, and Kepperle
[53] succeeded in constructing apparatus which could trap systems of cold dilute atoms,
whose properties are quite close to those of the noninteracting Bose gas. They received the
Nobel Prize in 2001. Since then, many other experiments like this have been performed, and
these Bose condensed systems have been used to construct systems that behave like a large
collection of idealized textbook problems, and can test rather delicate predictions of QM.
It is worth our while then to study the ideal Bose gas of spinless particles with single par-
p2
ticle Hamiltonian H = 2m . The formulae for the expectation values of the number operators
and the energy are:
β(~2 k2 −2mµ)
d3 k e− eβµ
Z
2m
hN i = V + n 0 . (12.35)
2π 3 1 − e− β(~2 k2m
2 −2mµ)
1 − eβµ
 Quantum Statistical Mechanics  273

β(~2 k2 −2mµ)
d3 k ~2 k 2 e−
Z
2m
hEi = V 3 β(~2 k2 −2mµ)
. (12.36)
2π 2m 1 − e− 2m

Note that we have allowed for the possibility of macroscopic occupation of the discrete zero
momentum mode of the finite volume system, in writing this large V limit of the formulae.
The mass m, ~, and the temperature can be combined to make a parameter λ, called the
thermal de Broglie wavelength because it has dimensions of a length:
s
2πβ~2
λ≡ . (12.37)
m
√
Now define k ≡ 4πλ−1 Ωx, where Ω is a unit vector. Then we can do the angular integrals,
getting a factor of 4π since the integrand is spherically symmetric. We end up with equations
for the number density n and the energy density :

n0 ∞ x2
Z
n= + λ−3 4π −1/2 dx . (12.38)
V 0 zex2 − 1

16~2 π 1/2 ∞ x4
Z
= dx . (12.39)
2mλ5 0 zex2 − 1
We have defined z ≡ e−βµ , which is called the fugacity. If z < 1, the integrands have poles
and these expressions are ill-defined, so physical systems have z ≥ 1. On the other hand,
as a function of z, the integral appearing in the expression for n is bounded by its finite
value at z = 1. Thus, the integral term appearing in the expression for n is monotonically
decreasing with temperature (because of the factor involving the thermal wavelength) for
low enough temperature. For any given temperature, there is a density above which we must
invoke a finite value of n0 /V in order to solve the equations. For any given density, there is
a temperature below which there is a finite value of n0 /V . The line in the (n, T ) plane below
which this condensate appears is called the phase transition to BEC (Figure 12.1).
What is the Bose–Einstein condensate state? It is best to think of a very large but finite
box, and certainly this will be the correct description of any real experiment. The momenta
T (arb. units)

Normal matter
µQ = 0
BEC

0 ρQ (arb. units)

Figure 12.1 Phase diagram of Bose–Einstein condensation.

274  Quantum Mechanics

are discrete. For particles of nonzero momentum, it is simply an ordinary thermal state. For
the discrete zero mode, it is a density matrix of maximal entropy constrained by the equation

T rρa† (0)a(0) = n0 , (12.40)

where n0 ∝ V λ−3 is the density we need to fit the thermodynamic equations above for some
values of n and T below the transition line. In Exercise 12.3, you will show that this is
†
eLa (0)a(0)
ρ= , (12.41)
Tr [eLa† (0)a(0) ]
where
eL
= n0 . (12.42)
1 − eL
As the volume goes to infinity, n0 goes to infinity and L → 0. The density matrix approaches
something proportional to the unit matrix and the entropy becomes infinite.
This is certainly the result in the Grand Canonical ensemble. For the Canonical ensemble
though, where the particle number is fixed, it cannot be right. For the system in a box, with
a fixed finite number of particles N ∝ V , the result of lowering the temperature is that
more and more of the particles go into the ground state. The Grand Canonical calculation
of the limiting density of particles in states with nonzero momentum is still valid for large
V (in de Broglie wavelength units), so there is still a Bose–Einstein condensate below a
certain temperature, but the density matrix for the zero momentum creation and annihilation
operators is quite different. We can use the equation

n0 ∞ x2 n0
Z
n= + λ−3 4π −1/2 dx x2 ≡ + ρ> , (12.43)
V 0 e −1 V
to calculate n0 , the number of zero momentum particles below the transition temperature.
However, the term proportional to V in this number is no longer an expectation value. We
have fixed the total number of particles as an operator constraint. Below the transition
temperature, the term of order V in the number of nonzero momentum particles is ρ> V ,
and the fluctuations of this operator are suppressed relative to this by V −1/2 . This means
that n0 is fixed in the canonical density matrix, up to terms of relative order V −1/2 . The
full multiparticle system is mostly in its ground state, and the entropy is all carried by the
relic particles of nonzero momentum. In the strict T → 0 limit, the term in the entropy
proportional to the volume goes to zero.
For a system of noninteracting bosons confined to a box by a potential V (x), the N -
particle ground state wave function is just

Ψ0 (x1 . . . xN ) = ψ(x1 ) . . . ψ(xN ), (12.44)

 Quantum Statistical Mechanics  275

where ψ(x) is the single particle ground state. If we add a repulsive short range two body
interaction of the form
λX 3
δH = δ (xi − xj ), (12.45)
2 i6=j

then it is plausible, at least at low density, that the ground state is still approximately a
product of single particle wave functions (this is the Hartree approximation—see Chapter
11, as well as Chapter 18 on the variational principle). If we calculate the expectation value of
the energy in states of this form, and minimize with respect to ψ ∗ , we get the Gross–Pitaevski
equation [54]:
~2 2
(− ∇ + V (x))ψ(x) = λψ(x)ψ ∗ ψ(x). (12.46)
2m
The G–P equation is one of the main tools in the modern theory of Bose–Einstein conden-
sates.
Apart from its resolution of Gibbs’ entropy of mixing paradox, quantum statistical
mechanics is primarily useful in the study of systems at very low temperatures. It turns
out that the variety of peculiar states of matter at low temperature is extremely large and
the field promises to remain fascinating into the indefinite future. Landau’s Fermi liquid
theory and the experimental discovery of dilute Bose–Einstein condensates have made the
very simple calculations we have done in this chapter much more useful than one might have
thought. However, the real future of the field is probably the study of more exotic states of
matter, for which we do not yet have a complete set of theoretical tools.

12.6 STATISTICAL MECHANICS OF A FERMI LIQUID

As we have said, a Fermi liquid is an accurate description of the low lying excitations near
the ground state of a wide variety of systems whose underlying constituents are fermions.
Those low lying excitations are almost noninteracting fermions. Their single particle energies
have the form
(p, Ω) = vF (Ω)p, (12.47)
where vF is called the Fermi velocity. p is a continuous parameter, which describes the
distance in momentum space above the Fermi surface, a surface in momentum space, defined
by an equation of the form E(p) = EF . This equation defines a two-dimensional surface
in three-dimensional momentum space. Ω is a solid angle (we have written it as a three-
dimensional unit vector), which tells us where the excitation is on the surface. For a crystalline
solid with a particular lattice, the equation for the Fermi surface will have only the symmetries
of the lattice.
The underlying microscopic physics is a complicated interacting problem, and it is not
easy to calculate the ground state energy. The miracle of Fermi liquid theory is that the
276  Quantum Mechanics

physics of excitations near the ground state is completely characterized by the function
vF (Ω). The grand canonical partition function of a Fermi liquid is given by the general
prescription above
Z Λ Z
ln Z = p−1
F dp d2 Ω [ln (1 + e−β(pvF (Ω)−µ) )]. (12.48)
0

The cutoff Λ on the momentum integral reminds us that the Fermi liquid approximation is
only valid near the ground state. Both Λ and the scale pF might be expected to be of order
~
aB for a system consisting of nonrelativistic electrons and nuclei interacting via the Coulomb
potential.
As long as vF (Ω) does not have any zeroes, the contribution to the integrand from
Λ
momenta near the cutoff is of order e−βΛ = e− kT , so as long as kT  Λ, this is negligi-
ble. If we formally take Λ → ∞, the integral converges, but we should recognize that it only
corresponds to real physics when this temperature inequality is satisfied. In the Λ → ∞
limit, we can extract the temperature dependence by defining βp = y. y has units of inverse
velocity. Then, Z Z∞
ln Z = (βp−1
F ) dy d2 Ω [ln (1 + ze−yvF (Ω) )]. (12.49)
0

In this equation, z = eβµ is the fugacity.

One can show [47] that to leading order at low temperature, thermodynamic quantities
do not depend on the interactions between Landau’s quasiparticles. Thus, we should really
only use these simple formulae to lowest order in T , because higher order terms will have
interaction corrections. We should emphasize that we assume that we are dealing with a
normal metal. Bardeen Cooper and Schriefer showed that if the quasiparticle interactions
are attractive, then the Fermi liquid ground state is unstable to a superconducting ground
state and the thermodynamics is completely different. For large β, the fugacity z is small
(the constraint of fixed average density determines the chemical potential to be negative).
Thus, we can expand out the logarithmR and the leading order result depends on the angle
dΩ
dependent Fermi velocity only through v(Ω) . Thus, since the noninteracting Fermi gas is a
particular example, the leading low temperature behavior of an arbitrary Fermi liquid is the
same as that of a free gas, up to a multiplicative constant. Thus, for example, the specific
heat (per particle) is given by Exercise 12.7:
π2 T
CV = k , (12.50)
2 TF
where k is Boltzmann’s constant. For a free electron gas, the “Fermi temperature” TF is
given by
~2
TF = (3π 2 ρ)2/3 , (12.51)
2mk
 Quantum Statistical Mechanics  277

where ρ is the fermion density. For metals, the free electron models gives m = me . Fermi liquid
theory tells us that we get the same low temperature formula with me → m∗ , the “effective
mass” of the quasiparticle. It is important to realize that unlike the case of semiconductors,
where the charge carrying excitations have a standard Galilean invariant dispersion relation
at small wave number, Landau’s quasiparticles do not behave like nonrelativistic electrons
with an effective mass.

12.7 PLANCK’S RADIATION LAW

The origin of QM was Planck’s resolution of the ultraviolet catastrophe in the statistical
mechanics of the electromagnetic field. A field is just a collection of oscillators with wave
numbers kn , where the kn are the allowed wave numbers with (e.g., periodic) boundary
conditions on a cubical box of side L, kN = 2πL (N), where N is a vector of integers. Each
oscillator has frequency
2π
ω(kN ) = c |N|,
L
where c is the speed of light. The logarithm of the partition function is just the sum of the
logs of the partition functions of these oscillators

ln [1 − e−β~ω(kN ) ].
X
ln Z = −2 (12.52)
k

The factor of two counts the two polarization states of light waves. Because the oscillations
are not coupled, we actually have a separate energy conservation law for each frequency, so
we can calculate the expectation value of the energy for each frequency by having a different
inverse temperature βk for each value of ω(kN ). Taking the limit of large L turns sums into
integrals. k becomes a continuous variable. We can think of the allowed values of k as forming
an infinite lattice in three-dimensional space. Each lattice point can be associated with the
unit cube in this space, which has that point as its largest vertex in all three directions. Each
of these cubes has volume ( 2π 3
L ) . In the limit of large L, the surface in k space with fixed ω
becomes a sphere of radius ω/c in k space.
Using the fact that we have two polarization states for each elementary cube, it is easy
to work out that the density of energy per unit frequency dω is

~ω 3 ~ωβ
ρ(ω) = (e − 1)−1 .
π 2 c3
This is Planck’s famous radiation law, the blackbody spectrum. The ultraviolet catastrophe
of classical physics comes from extrapolating the low frequency limit of this formula ρ =
2
kB T πω2 c3 to high frequencies. The physical reason for it is that classical physics assumes that
278  Quantum Mechanics

one can get arbitrarily low intensity electromagnetic waves, which cost very little energy,
at any frequency. There are more modes at higher frequency, so the statistical sum at any
temperature should be dominated by very low energy, very high-frequency electromagnetic
waves. The solution of the quantum oscillator problem shows that there is a minimum energy
~ω of any excitation of the quantized electromagnetic field. This minimum energy excitation
is called a single photon and this connection between particles and fields is the one we pointed
out in Chapter 5. At any finite temperature, the high-frequency modes will be Boltzmann
suppressed because even one photon carries a very high energy.
Since photons are bosons, one might have thought that one could have a Bose condensate
of photons at sufficiently low temperature. The reason that this does not happen under
normal circumstances is that photons in a box are constantly being absorbed and reemitted by
the atoms in the walls of the box, so that photon number is not conserved. For nonrelativistic
particles, energy and momentum conservation normally forbid interactions which change
the number of particles. Thus, one has two conservation laws, particle number and energy.
Recently, a theory of thermalization processes which could conserve photon number [55] has
been developed, and observations of Bose condensed photon fluids have been reported [56].

12.8 EXERCISES
P ∗
12.1 Show that the matrix ρij = I ciI cIj has the properties of a density matrix.

12.2 Show that the trace of a tensor product of matrices is the product of the traces of the
individual matrices.

12.3 Show that the maximal entropy density matrix for the ground state of a system of Bose
particles with only the average number of particles constrained to be n0 is given by
†
eLa (0)a(0)
ρ= ,
Tr [eLa† (0)a(0) ]

where
eL
= n0 .
1 − eL
12.4 Diamagnetism is a term describing the first-order response of a system to an external
magnetic field. The material gets a magnetic moment. If the magnetic field generated
by that moment points in the direction opposite to the applied field, the system is said
to be diamagnetic. If it points in the same direction, it is called paramagnetic. Show
that in a system composed of classical charged point particles with no intrinsic magnetic
moment, the thermal average of the induced magnetic moment vanishes. This is called
van Leuwen’s theorem.
 Quantum Statistical Mechanics  279

12.5 Consider a system of N atoms, each with an intrinsic magnetic moment µ. Treat the
motion of the particles, as well as the energetics of the dipoles, classically. Using the
result of Exercise 12.4, show that the statistical mechanics of the magnetization in a
constant magnetic field is given entirely by the Hamiltonian HB = −µB N
P
i=1 cos(θi ),
where B is the magnitude of the field and θi the angle between the direction of the
field and the dipole of the i-th atom. Evaluate the magnetization and the magnetic
susceptibility of this system in the canonical ensemble at inverse temperature β =
(kT )−1 . The susceptibility is defined as the derivative of the magnetization with respect
to B. Show that
M ≡ β −1 ∂B (ln Z) = N µ(coth(α) − α−1 ),
where α ≡ µBβ.
χ = βµ2 [α−2 − csch2 (α)].
χ is the susceptibility per atom. Show that χ ∼ β for small β (Curie’s law) and evaluate
the proportionality constant, which is called Curie’s constant.

12.6 In QM, van Leuwen’s theorem is no longer valid. We saw in Chapter 9 that for free
particles, the motion perpendicular to the field has quantized energy levels with spec-
trum e~B
mc (n + 1/2). Each of these levels was infinitely degenerate due to the translation
symmetry. If we put the system in a box, that symmetry is broken and each level has
eB
a finite degeneracy g. One can show that at large volume g = L2 2π~c , where L is the
2
pz
size of the box. The motion along the field has energies 2m , where pz = 2π~ L . Write an
integral formula for both the logarithm of the grand canonical partition function and
the average particle number for noninteracting charged fermions in a constant magnetic
field. Evaluate the formulae in the high temperature limit, compute the susceptibility
per unit volume and show that it is nonzero and obeys Curie’s law.

12.7 Derive the low temperature specific heat for a free fermion gas with single particle
2 k2
energies ~2m .
 CHAPTER 13

Perturbation Theory: Time

Independent

13.1 INTRODUCTION
Much of the rest of this book will be devoted to approximation methods. The simplest of
these is time-independent perturbation theory. Most interesting problems in physics cannot
be solved exactly but many are close to exactly solvable problems and one can write the
solution as a power series expansion in some small parameter. This is called perturbation
theory. The higher-order terms in the series become more and more complicated, but one can
often organize each of them into a sum of individual terms, each associated with a picture or
diagram. Even when there is no small perturbation parameter in the real physical problem,
one can introduce one artificially and try to get insight into more sophisticated approximation
schemes by resumming the perturbation theory, keeping only a certain set of simple diagrams
at each order.
This chapter will give a general set of rules for computing the perturbation theory for
the eigenvalues of a Hamiltonian of the form H0 + gV . These methods are applicable to the
discrete spectrum. The discussion of perturbation theory for the continuous spectrum can be
found in the next chapter and Chapter 16 on scattering theory.

13.2 BRILLOUIN–WIGNER PERTURBATION THEORY FOR NONDEGENERATE

LEVELS
We have seen that there are several quantum mechanical problems, which can be solved
exactly. Let H0 be such a Hamiltonian, and consider

H = H0 + gV, (13.1)

281
282  Quantum Mechanics

where g is a small parameter. The Schrödinger equation can be written

(H0 + gV )(|ψ0 i + |φi) = (E0 + ∆)(|ψ0 i + |φi). (13.2)

|ψ0 i is an eigenstate of H0 with eigenvalue E0 . The projection of the perturbed vector |φi
on |ψ0 i is ambiguous, because the equation is homogeneous. We can choose it so that the
eigenstate is normalized, obtaining a form of perturbation theory first invented by Rayleigh
[57] in his study of sound waves, and adapted to quantum mechanics (QM) by Schrödinger
[58]. Instead, we will adopt a procedure invented by Brilliouin and Wigner [59] and choose

hφ|ψ0 i = 0. (13.3)

The resulting state has a squared norm 1 + hφ|φi, and we must remember to divide by the
norm when computing expectation values.
The eigenstate of H0 with eigenvalue E0 might be degenerate, but we will first consider
the case when it is isolated. Define

P = 1 − |ψ0 ihψ0 |, (13.4)

the projector onto the subspace orthogonal to the unperturbed eigenvalue, and write the two
equations obtained by left multiplying the Schrödinger equation by P and 1 − P .

P gV [|ψ0 i + |φi] + P H0 |φi = P E|φi. (13.5)

|ψ0 ihψ0 |gV (|ψ0 i + |φi) = (E − E0 )|ψ0 i. (13.6)

The second equation is just an evaluation of the perturbed energy:

hψ0 |gV |(|ψ0 i + |φi) = E − E0 . (13.7)

The first equation can be rewritten as

P
|φi = g V (|ψ0 i + |φi). (13.8)
E − H0
We have used the fact that P commutes with H0 . We see the |φi is indeed small when g is
small, and has the power series expansion
∞
X P
|φi = [ (gV )]k |ψ0 i. (13.9)
k=1
E − H0

This is, implicitly, a power series in g, because the equation for E − E0 shows that it begins
with a term
E1 = ghψ0 |V |ψ0 i, (13.10)
 Perturbation Theory: Time Independent  283

with corrections of order g 2 . Thus,

∞
X
E= g k Ek . (13.11)
k=0

Note also that the dependence on E − E0 in the term with k = 1 is an illusion, because
the projection operator kills it. Our relatively simple equation for |φi hides some of the g
dependence. To get an explicit power series in g, we have to expand out E = E0 + ∞ k
P
k=1 g Ek .
To get explicit expressions for both |φi and ∆ ≡ E − E0 , we proceed systematically. The
expression for ∆ at order g k requires the knowledge of |φi at order g k−1 . The expression for
|φik in the expansion |φi = ∞ k
k=1 g |φik involves all the coefficients Ep up to Ek−1 . So we
P

get
E1 = hψ0 |V |ψ0 i. (13.12)
P
E2 = hψ0 |V V |ψ0 i. (13.13)
E0 − H0
P
|φi1 = V |ψ0 i. (13.14)
E0 − H0
P P
|φi2 = (V − E1 ) V |ψ0 i. (13.15)
E0 − H0 E0 − H0
P P P P P
|φi3 = (V − E1 ) (V − E1 ) V |ψ0 i + (E2 ) V |ψ0 i.
E0 − H0 E0 − H0 E0 − H0 E0 − H0 E0 − H0
(13.16)
And so it goes. . .
The terms get more and more complicated as k gets larger. For certain simple systems,
notably when H0 is a collection of harmonic oscillators, the perturbation theory can be
simplified further and pictorial representations of individual terms, called (time-ordered)
Feynman diagrams, allow one to compute to fairly high orders. In actual fact, the time-
dependent perturbation theory we will study in the next chapter, or the associated Feynman
path integral method, provide more convenient expressions for computing higher-order terms.
However, they do not produce direct expressions for perturbed energy levels.
P
Note that the operator E0 −H 0
, which appears everywhere in these formulae, is well defined
because the eigenvalue E0 is nondegenerate. However, it is also obvious that if there are
eigenvalues very close to E0 , the higher-order terms in the perturbation series will be very
large, unless g is very small. We will see that the methods of degenerate perturbation theory
also give us a clue about how to deal with situations like this.
The expression for E2 in case |ψ0 i is the ground state of H0 , is always negative, and
this exemplifies a general principle called the Variational Principle, which we will explore
further in Chapter 18. We have already used it in our discussion of the hydrogen molecular
284  Quantum Mechanics

ion and the Gross–Pitaevski equation. Consider the expectation value of H = H0 + gV in

any normalized state ψi. Expanding ψi in eigenstates of H
X
|ψi = cn |En i, H|En i = En |En i, (13.17)

we obtain X X
hψ|H|ψi = |cn |2 En ≥ Eground |cn |2 = Eground . (13.18)
That is, the ground state energy is the minimum expectation value of H among all normalized
states.
In particular, hψ0 |H|ψ0 i = E0 + gE1 ≥ Eground . On the other hand, perturbation theory
tells us that for small enough g, Eground ≈ E0 + gE1 + g 2 E2 . The inequality then tells us
that E2 must be negative, and this agrees with our explicit formula.
The most straightforward way to evaluate the higher-order terms in perturbation theory
is to insert complete sets of H0 eigenstates and obtain formulae like
X 1
∆2 = |hψ0 |V |ψn i|2 . (13.19)
n6=0
E0 − En

With the notable exception of the important case where H0 is the Hamiltonian of nonin-
teracting quantized fields, where this strategy leads to Feynman diagrams, these infinite
summations can be very tedious. In some cases, there is an alternative, in low orders of per-
turbation theory, invented by Dalgarno and Lewis [60]. These authors observed that if one
can find an operator Ω2 satisfying

V |ψ0 i = [H0 , Ω2 ]|ψ0 i, (13.20)

then
P P
V |ψ0 i = [H0 − E0 , Ω2 ]|ψ0 i = P Ω2 |ψ0 i, (13.21)
E0 − H0 E0 − H0
so that
E2 = hψ0 |V Ω2 |ψ0 i − hψ0 |V |ψ0 ihψ0 |Ω2 |ψ0 i. (13.22)
If Ω is “simple,” then the latter formula is easy to evaluate. We will see some examples of
this below. To extend the Dalgarno–Lewis (D–L) method to next order, we would have to
find an operator satisfying
V Ω2 |ψ0 i = [H0 , Ω3 ]|ψ0 i. (13.23)
This is only possible if E2 = 0 (prove it!). So the D–L trick is limited to second-order
perturbation theory. However, in that context it can be very useful and it deserves to be
explored more than it has been in the literature.
 Perturbation Theory: Time Independent  285

13.2.1 Relation to Rayleigh–Schrödinger Perturbation Theory

The Rayleigh–Schrödinger version of perturbation theory determines the part of the per-
turbed state parallel to |ψ0 i, by insisting that |ψ0 i + |φiRS have length one. Thus,
|φi
|φiRS = p , (13.24)
1 + |hφ|φi
where |φi is the BW state. It is a lot easier to compute the normalized eigenstate from this
formula than to apply the RS rules directly.
There is another aspect of the difference between the two methods of computing the
perturbation expansion which is exposed by the above formula. A perturbative calculation
of |φi gives an infinite number of powers of g in the normalized wave function. Similarly, if
we write the expression for the energy in BW perturbation theory
∞
X P
(E − E0 ) = hψ0 |gV [ (gV )]k |ψ0 i, (13.25)
k=0
E − H0

then at a finite order in the expansion, we get a nonlinear equation for E, whose exact
solution contains an infinite number of powers of g. There has not been much exploration
of whether the resummations of perturbation theory implicit in the BW formalism lead to
better approximations at finite values of g.

13.3 DEGENERATE PERTURBATION THEORY

A major virtue of the Brillouin–Wigner approach to perturbation theory is the ease with
which the formalism generalizes to the case where the energy level E0 is degenerate. The
only change is that the projection operator P is replaced by
X
P =1− |ψ0d ihψ0d |, (13.26)
d

the projector on the degenerate subspace. As in the nondegenerate case, the projector P
makes the inverse operator P (E0 − H0 )−1 = (E0 − H0 )−1 P well defined, and the higher-order
terms in the perturbation series are formally small for small g.1
1
The question of convergence of the perturbation series is much more involved. It is always an asymptotic
series for the actual eigenvalue as a function of g. This means that the difference between the exact answer
and the first n terms vanishes like g n+1 as g → 0. This is much less than convergence. Convergence would
imply that the eigenvalue existed as an analytic function on a disk surrounding the origin in the complex
g plane. It is often easy to see that such analyticity violates physical sense. For example, if we perturb
a harmonic oscillator by gx4 , then it is manifest that the Hamiltonian is not bounded from below for
negative g, so the eigenvalues cannot be analytic functions. We will say a bit more about this in our
chapter on the JWKB approximation.
286  Quantum Mechanics

If |ψ0b i is any state in the degenerate subspace, we write the perturbed state as

|ψ b i = |ψ0b i + |φb i, (13.27)

where |φb i is orthogonal to the subspace. The Schrödinger equation takes the form

(H0 + gV )|ψ0b i + φb i = E b (|ψ0d i + φd i). (13.28)

Then we can solve for |φb i as before

∞
b
X gP
|φ i = ( V )k |ψ0b i. (13.29)
k=1
E b − H0

Now take the scalar product of the Schrödinger equation with |ψ0a i.
∞
X gP
hψ0a |H0 + gV |ψ0b i + hψ0a |g|V ( V )k |ψ0b i = E b δba . (13.30)
k=1
Eb − H0

This equation is only consistent if we choose the basis in the degenerate subspace to be one
in which the matrix
∞
X gP
hψ0a |H0 + gV |ψ0b i + hψ0a |g|V ( V )k |ψ0b i, (13.31)
k=1
Eb − H0

is diagonal. The now split eigenvalues of H are the eigenvalues of this matrix.
This form of degenerate perturbation theory can also handle cases where the level E0 is
not exactly degenerate, but the level splittings in its neighborhood are very small, so that the
perturbation series is a bad approximation unless g is extremely small. One simply defines
the projector P so that it projects on the space orthogonal to all of the closely spaced levels,
and then proceeds as in the degenerate case. The number E0 then becomes a diagonal matrix
in the quasidegenerate subspace, the image of (1 − P ).
The typical case in which such quasidegeneracy arises is for large systems with a volume
V  1 in Bohr units. As we have seen, phonons and other long-wavelength excitations,
corresponding to flows of conserved currents, have energies as low as V −1/3 in Bohr units, so
the ground state is quasidegenerate. In principle, the Brillouin–Wigner method enables one
to construct an energy-dependent effective Hamiltonian, which describes only the dynamics
of these low-energy modes. It turns out, however, that the Feynman path integral formalism
provides a much more efficient way to construct the dynamics of these long-wavelength,
low-energy excitations.
 Perturbation Theory: Time Independent  287

13.4 THE FEYNMAN–HELLMANN THEOREM

Feynman and Hellmann [61] proved a simple theorem about derivatives of energy levels with
respect to parameters, which is often helpful in evaluating the expectation values required for
computing first-order perturbation theory. Let H(λ) be a one parameter set of Hamiltonians,
En (λ) one of its discrete eigenvalues and |ψn (λ)i the corresponding normalized eigenstate.
Assume that we are working at a value of λ where the eigenstate is nondegenerate. The norm
of |ψn (λ)i is independent of λ so
d
hψn (λ)| |ψn (λ)i + c.c. = 0. (13.32)
dλ
If we write
En (λ) = hψn (λ)|H(λ)|ψn (λ)i, (13.33)
and use the previous equation, we find the Feynman–Hellmann result
d dH
En (λ) = hψn (λ)| |ψn (λ)i. (13.34)
dλ dλ
This can be viewed as a generalization of the first-order perturbation theory formula to
Hamiltonians that depend on nonlinear functions of λ.

13.5 EXAMPLES
13.5.1 The Stark Effect
Let us now apply these general and abstract ideas to a simple example: the hydrogen atom in
external constant electric or magnetic fields. Let us do the electric field first. The perturbation
has the form V = −eEX3 , where we have used the rotation invariance of the unperturbed
problem to choose the electric field in the x3 direction. The first-order perturbation theory
formula for the perturbed levels is
(1)
Enlm = En − eEhn l m|X3 |n l mi, (13.35)

where En is the nth Rydberg energy.

The operator X3 = r cos(θ) is invariant under rotations around the three direction.
The unperturbed eigenfunctions have the form Rnl (r)Ylm (θ, φ), so the matrix element of
X3 is a product of a matrix element of r between two radial wave functions, and aqmatrix
element of cos(θ) between a pair of spherical harmonics. Noting that cos(θ) is just 4π 3 Y10 ,
we see that computing the angular matrix element reduces to the problem of addition of
angular momentum. That is, the wave functions Y10 Ylm are just products of spin 1 and spin
l eigenstates, which can be expanded in a complete basis of spherical harmonics. The general
288  Quantum Mechanics

rules of addition of angular momenta tell us that we get a linear combination of states with
spin l and |l ± 1|. Furthermore, we get only states with Lz = ~m.
More generally, we have
Ykn Ylm = c(l0 , m0 ; k, n l, m)Yl0 m0 . (13.36)
The coefficients c(l0 , m0 ; k, n l, m) are called Clebsch–Gordon coefficients, and are entirely
determined by the group theory of the rotation group. The required matrix element of cos(θ)
is thus, r
0 0 4π 0
hl m | cos(θ)|lmi = c(l , m; 1, 0 l, m), (13.37)
3
where l0 ranges between |l−1| and l+1. You will find tables of the Clebsch–Gordon coefficients
on the World Wide Web.
The above discussion is a special case of what is known as the Wigner–Eckart theorem
[62]. In any rotation invariant system, the rotation generators act on the space of operators
via
O → U † (R)OU (R). (13.38)
This formula is analogous to the relation between Schrödinger picture and Heisenberg picture
operators. It gives us the version of an operator appropriate in a rotated reference frame. Since
the space of operators is a linear space, we can decompose it into irreducible representations
of the rotations. Thus, any operator will have a representation
X
O= Ol olm , (13.39)
where the operators Ol commute with rotations, and the operators olm transform under
rotations like the spherical harmonics Ylm . The matrix elements of olm between unperturbed
eigenstates of the rotation invariant system, will be Clebsch–Gordon coefficients, and we will
need only the matrix elements of the invariant operators Ol . The appendix on Group Theory
contains a short proof of the Wigner–Eckart theorem.
For the ground state of the hydrogen atom, first-order perturbation theory gives zero
for the Stark Effect, because of the angular momentum selection rules. The second order
energy is
P
−hψ0 |V V |ψ0 i. (13.40)
E − H0
This is an example where the Dalgarno–Lewis method proves useful. The D–L equation is
[H0 , Ω2 ]|ψ0 i = −eEX3 |ψ0 i. (13.41)
Since H0 and the ground state wave function are rotation invariant, Ω2 transforms like X3
under rotations. If Ω2 commutes with the coordinates it has the form
Ω2 = f (r) cos(θ), (13.42)
 Perturbation Theory: Time Independent  289

where
2 2meE
−(f 00 (r) + 2f 0 (r)∂r )ψ0 (r) + [f (r) − f 0 (r)]ψ0 (r) = − 2 rψ0 (r). (13.43)
r ~
Dividing through by ψ0 and using ∂r (ln ψ0 ) = − a1B , we find

meE 2 aB r2
f (r) = (aB r + ). (13.44)
~2 2
The second order energy is thus,

E2 = hψ0 |(−eEX3 )f (r) cos(θ)|ψ0 i. (13.45)

which is a simple integral. The result is

9a2B
E2 = −(eE)2 . (13.46)
4
In Exercises 13.1–13.2, you will show how to generalize this result in a variety of ways.

13.5.2 The Zeeman Effect

The Zeeman effect is the shift in hydrogen energy levels due to a weak constant magnetic
field. It is fairly straightforward to evaluate it, and we will do so in a moment, but first
we must point out that the question of exactly how weak the field is, becomes crucially
important because of relativistic corrections to the Schrödinger equation, which also have
to do with magnetic fields. We have modeled the electromagnetic field of the nucleus by a
Coulomb field, and this is valid for electrons at rest, but the electron in the hydrogen atom
is moving, albeit at a velocity much less than that of light. As a consequence, it also sees a
magnetic field, which is weak because the electron velocity is  c. The qualitative features
of the Zeeman effect depend on whether the external field is larger than or smaller than the
effective magnetic field seen by the moving electron. We will first study the effect where the
external field is stronger, where it is also known as the Paschen–Back effect.
An external magnetic field acts on both the spin of the electron, and via the minimal
substitution P → P − eA, where A is the magnetic vector potential. For a constant magnetic
field B in the three direction, we can choose the vector potential to be
B
Ai = − ij3 xj . (13.47)
2
To first-order in B we have
(P − eA)2 Pi ij3 xj eB
= eB = L3 , (13.48)
2m 2m 2m
290  Quantum Mechanics

where L3 is the third component of the orbital angular momentum.

By definition, the interaction with the spin is given by a Hamiltonian
eg
−µ3 B = − σ3 B, (13.49)
2
where g is the gyromagnetic ratio, relating the electron magnetic moment to its intrinsic
spin. Dirac’s relativistic theory of the electron gives g = 2. The full theory of Quantum
Electrodynamics gives corrections to this, which are smaller by powers of απ , where α ∼ 137
1

is the fine structure constant.

The full first-order perturbation is thus,
e~
HP B = B(K3 + σ3 ), (13.50)
2me
where K3 is the orbital angular momentum in ~ units. If we consider the hydrogen eigenstates
labelled by (n, l, m, s), with n fixed, this perturbation splits the (2l + 1) degenerate orbital
states and the two degenerate spin states. The m = 0 state splits into two states with energies
e~ e~
En ± 2m B. More generally, we get states with energies En + 2m B(m ± 1).

13.5.3 Fine Structure and the Weak-Field Zeeman Effect

For weaker fields, we have to compute two other (relativistic) perturbations, which compete
with the external field. Normally, to first order, we can simply add up the effect of different
perturbations, but in a degenerate system, one perturbation may be diagonalized in a different
basis than the other. The correct procedure is to diagonalize the sum of all perturbations
which are of the same size. We will concentrate on the weak-field regime, where the relativistic
perturbations dominate over the external magnetic field. They leave over a degeneracy, which
is split by even a tiny external field. The relativistic corrections, which exist even when the
external field vanishes, give rise to the fine structure corrections to the hydrogen spectrum.
The first fine structure correction comes simply from the relativistic correction to the
relation between energy and momentum for the electron. The correct relativistic formula is

E 2 − (Pc)2 = m2 c4 . (13.51)

This says that energy and momentum sit on a three-dimensional hyperboloid in a four-
dimensional space. This relation is obviously invariant under hyperbolic rotations, which are
Lorentz transformations.
In QM, P is still (~ times) the translation operator, so we can write the correction to the
Schrödinger equation when |P|  mc as a term
(P2 )2 P2 2 1
δH = − = −( ) . (13.52)
4m3e c2 2me me c2
 Perturbation Theory: Time Independent  291

The second expression is m1e c2 (H − V )2 , for any potential, and when evaluating its matrix
elements in any unperturbed eigenstate, we can replace H by the corresponding eigenvalue.
The correction is rotation invariant, and commutes with the spin, and so will not break the
angular momentum and spin degeneracies, but it will lift the degeneracies between different
l states with the same n, because the expectation values of r−1 and r−2 depend on l. The
perturbation is already diagonalized in the angular momentum basis. Thus, just like the
Stark effect, this contribution to the fine structure involves matrix elements of rp between
radial Coulomb wave functions.
The second contribution to the fine structure is more complicated, and involves a term
proportional to L · S. Roughly speaking, the orbital motion of the electron produces a mag-
netic field in its rest frame,2 and the electron dipole moment interacts with that field. Naively,
one could evaluate the magnetic field by saying that in the rest frame of the electron, the
proton is moving, thus producing an electric current, which produces a magnetic field. The
problem with such a calculation is that the electron rest frame is an accelerated frame, so
that the simple Lorentz transformation rules of electric and magnetic fields between inertial
frames do not give the correct answer.
L.H. Thomas pointed out that the transformation to an accelerated frame could be viewed
as a time-dependent Lorentz boost transformation. We do not have the space here to elaborate
on this argument, but will simply record that Thomas’ answer was simple: it multiplies the
naive field by a factor of 1/2. The proton current in the Lorentz frame which coincides
instantaneously with the electron rest frame is

P 3
J(X) = −evδ 3 (X) = −e δ (X). (13.53)
me

Solving Maxwell’s equation

1
∇×B= , (13.54)
0 c2
we get
1 e
B= L, (13.55)
4π0 mc2 R3
using
L = R × P. (13.56)
Thomas’ result reduces this by a factor of 1/2.
The interaction of the magnetic dipole with this field is

δH = −µ · B, (13.57)
2
The field is purely electric in the proton rest frame.
292  Quantum Mechanics

and the magnetic dipole moment of the electron is

−eS −eS
µ=g = . (13.58)
2me me

We have used the fact that Quantum Electrodynamics tells us that g = 2, up to a small
correction. The final result is
e2
δH = S · L. (13.59)
8π0 m2e c2 R3
Now
2S · L = (L + S)2 − L2 − S2 = ~2 [j(j + 1) − l(l + 1) − 3/4]. (13.60)
R−3 also commutes with L2 , J2 so the states of fixed j and l are those in which this pertur-
bation is diagonal. The allowed values of j for a given l are l ± 1/2, so this perturbation splits
the 2(2l + 1) states in a degenerate spin/angular momentum multiplet, into two groups of
degenerate states of size 2l + 2 and 2l. Note that the external field perturbation, proportional
to Bext S3 does not commute with J 2 . However, when the field is weak, so that the energy
shifts it induces are small compared to the fine structure, all we have to do is evaluate the
matrix of S3 in each of the subspaces with fixed j l and varying j3 . You will do this in
Exercise 13.3 and see that the matrix elements are Clebsch–Gordon coefficients.

13.5.4 Coulomb Expectation Values of Powers of R

All of the perturbations of the hydrogen atom that we have studied, involve the expectation
values of powers of R in Coulomb wave functions, or matrix elements of such powers between
Coulomb wave functions. For R−p with p = 1, 2 we can evaluate these easily using the
Feynman–Hellmann theorem. For other powers, we will need a relation, called Kramers’ rela-
tion, between the expectation values of any three consecutive powers of r in any unperturbed
hydrogen wave function. One writes the radial equation as

l(l + 1) 2 1
u00nl = [ 2
− + 2 2 ]unl , (13.61)
r raB n aB

to write dr urp u00 in two ways. In the first one uses the equation directly, while in the
R

second one integrates by parts before using it. Since Coulomb wave functions are real the
method relates expectation values of different powers of r. You will prove this relation in
Exercise 13.4.
Kramers’ relation is
p+1 p
hunl |rp |unl i − (2p + 1)aB hunl |rp−1 |unl i + [(2l + 1)2 − p2 ]a2B hunl |rp−2 |unl i = 0. (13.62)
n2 4
 Perturbation Theory: Time Independent  293

In a moment, we will evaluate the cases p = −1, −2 using the Feynman–Hellman theorem,
and Kramers’ relation does the rest of the work.
The effective Hamiltonian for the radial wave function u = rR is

l(l + 1) e2
H = −~2 2me [∂r2 − ] − . (13.63)
r2 4π0 r

Derivatives of the Hamiltonian with respect to l and e give the requisite expectation values
of the first two inverse powers of r. We must be careful in taking the derivative w.r.t. l,
because of the degeneracy of l levels. The principle quantum number n is jmax + l + 1 where
jmax is the highest power in the Laguerre polynomial. If we varied l with n fixed, we would
be making drastic changes in the wave function. Instead, we want to vary l with jmax fixed.
The energies are
me e4
En = − . (13.64)
32π 2 20 ~2 (jmax + l + 1)2
Then the F–H theorem tells us that

1 ∂En me e2
h i = −4π0 2 = . (13.65)
r ∂e 4π0 ~2 n2

1 2 ∂En m2e e4
h i = 2me ~ (2l + 1) = . (13.66)
r2 ∂l (2l + 1)8π 2 20 ~4 n3
Putting everything together, we get a formula for the fine structure of hydrogen energy
levels
fs α2 En2 4n
Enj = 2
(3 − ). (13.67)
2mc j + 1/2
Here j is the integer determining the square of the total angular momentum J = L + S. α is
the fine structure constant ∼ 1/137. The full second order formula for the energy levels is

13.6eV α2 4n − 3j − 3/2

Enj = − 1 + [ ] . (13.68)
n2 n2 4j + 2

If we now put the system in a weak magnetic field, the dominant effect is to split levels
which are degenerate according to the formula just derived for Enj . These are the states
with different mj values at fixed j. The perturbing Hamiltonian for the weak-field Zeeman
e
effect is still 2m e
B(J3 + S3 ), but now we have to evaluate its expectation value in the states
obtained by diagonalizing the spin-orbit plus fine structure perturbations. These are labeled
by the eigenvalues of the operators that commute with the spin-orbit Hamiltonian, which
are J2 , J3 , L2 . Note since the Hamiltonian without the external field is rotation invariant, we
294  Quantum Mechanics

are free to choose the component of J that we diagonalize to be the one in the direction of
the external field. For expectation values in states of fixed J3 we can set
S·J
hmJ |S3 |mJ i = hmJ | J3 |mJ i, (13.69)
J2
because the other components of S · J have vanishing expectation values. Noting that
1
S · J = (J 2 + S 2 − L2 ),
2
we can now evaluate the weak-field Zeeman perturbation as
e~ j(j + 1) − l(l + 1) + 3/4
∆EZeeman = [1 + ]mJ . (13.70)
2me 2j(j + 1)
As promised, the degenerate mJ levels are split. The term in square brackets is called the
e~
Lande g-factor. The prefactor 2m e
≡ µB is called the Bohr magneton. In SI units, it is
5.788 × 10−5 eV/T. T stands for Tesla, the unit of magnetic field.

13.5.5 A Three-Dimensional Example

Consider the matrix  
E1 a1 a2
 ∗
 a1 E2 a3  . (13.71)

a∗2 a∗3 E3
If the ai are small, we can use perturbation theory. The first-order perturbation vanishes
because the perturbation is a purely off diagonal matrix in the basis where the unperturbed
Hamiltonian is diagonal. Assuming there is no degeneracy, the operator whose expectation
value we must evaluate to compute the second-order correction to the ground state E1 is
   
0 a1 a2 0 0 a1 0 a1 a2
P
V = a∗1 0 a3   0 (E − E2 )−1
 ∗
H2 = V a2  a1 0 a3  .
  
E − H0 ∗ ∗ ∗ ∗ −1 ∗ ∗
a2 a3 0 a1 a2 (E − E3 ) a2 a3 0
(13.72)
To compute the expectation value, we only need the 1, 1 matrix element of this product
|a1 |2 |a2 |2
matrix, which is (E−E 2)
+ (E−E 3)
. Thus, to second order, the implicit B–W equation is

|a1 |2 |a2 |2
E = E1 + + , (13.73)
(E − E2 ) (E − E3 )
whose solution through this order is
|a1 |2 |a2 |2
E = E1 + + . (13.74)
(E1 − E2 ) (E1 − E3 )
 Perturbation Theory: Time Independent  295

There are similar equations for the other two eigenvalues. There are obvious problems when
|E1 − Ei | ∼ |ai−1 |2 or smaller.
Let us suppose the quasidegenerate pair are E1,2 . Then, the B–W prescription is to
consider the matrix elements of
   
0 a1 a2 0 0 a1 0 a1 a2
 ∗  ∗
a 0 a 0 0 a   1 0 a3  ,
a (13.75)
 
 1 3  2
a∗2 a∗3 0 a∗1 a∗2 (E − E3 )−1 a∗2 a∗3 0

in the two-dimensional subspace. This is relatively easy to do, because the middle matrix is
proportional to a one-dimensional projector. The answer is
!
1 |a2 |2 a2 a∗3
H2 = (13.76)
E − E3 a∗2 a3 |a3 |2 .

The implicit equation for the quasidegenerate eigenvalues is

(E − Ea )(E − E3 ) = αa , (13.77)

where a = 1, 2 and the αa are the (properly ordered) eigenvalues of the 2 × 2 matrix of the
previous equation. The leading order contribution gives
αa
E = Ea + . (13.78)
Ea − E3

13.5.6 Degenerate Perturbation Theory in a Macroscopic System

We end this discussion with a rather sophisticated example. In our discussion of the physics
of solids, we mentioned briefly the Hubbard model. It consists of fermion operators ψ(i)
sitting on the points of a regular lattice, with anticommutation relations

[ψa (i), ψb† (j)]+ = δij δab , (13.79)

[ψa (i), ψb (j)]+ = 0. (13.80)

The fermions have spin 1/2, and the subscript a labels the spin value ±~/2. The Hamilto-
nian is
kij ψa† (i)ψa (j) + g 2 (ψa† (i)ψa (i))2 .
X X
H= (13.81)
ij i

kij contains only nearest neighbor terms, and the first term in the Hamiltonian describes
processes in which a fermion is destroyed on one site and another created on a nearest
neighbor site. Repeated application of this term allows fermions to propagate throughout
296  Quantum Mechanics

∗
the lattice. Hermiticity of the Hamiltonian implies that kij = kji . The second term is a
repulsion, which tries to forbid two fermions from occupying the same site. They are allowed
to do so, consistent with Fermi statistics, because they have two spin states.
Solution of the Hubbard model consists of finding the ground state, for each value of
the conserved fermion number N = i [ψa† (i)ψa (i) − 1]. We want to study the problem in
P

the limit of large g 2 , and for N = V the total number of lattice sites. This is called the
problem with a half filled band, because, in principle, we could accommodate twice as many
fermions. For g 2 → ∞, states with two fermions on a site are absolutely forbidden. The
constraint on the total number of fermions tells us that there must be exactly one fermion
on each site. Neither of these two constraints tells us what the fermion spin is on each site,
so we actually have a degenerate ground state. Any state, with a fermion on each site, is a
ground state, but the spin state of the fermion on each site is undetermined. The operators
Σm (i) = ψa† (i)(σm )ab ψb (i) are the local spin operators, which act on the degenerate subspace
of the full Hilbert space.
We now want to consider g 2 large but finite, and construct a perturbation theory with the
hopping term V = ij kij ψa† (i)ψa (j) as the perturbation. A single action of V on any state
P

in the degenerate ground state subspace takes us to a superposition of states, each of which
has two fermions on some lattice site, and none on one of its nearest neighbors. It is thus
orthogonal to the degenerate subspace. All of the matrix elements of V in the degenerate
subspace vanish, and there is no contribution to the energy shift in first order in g12 .
The second-order operator
P
V2 = (1 − P )V V (1 − P ), (13.82)
E − H0
is nonvanishing. In order for it to have a nonzero degenerate matrix element, the second
action of V must create a fermion where the first action annihilated one, and annihilate a
fermion on the doubly occupied site created by the first action. The intermediate state has
an energy of order g 2 above the ground state, so the operator V2 has eigenvalues of order
1/g 2 , if E is small compared to g 2 . In Brillouin–Wigner perturbation theory, E is just the
total energy of the state in the degenerate subspace, so it is negligible.3 It follows that, to
order 1/g 2
c X
V2 = − 2 kij ψa† (i)ψa (j)ψb† (j)ψb (i)kji . (13.83)
g ij

We can move ψb (i) through to the left, since it commutes with ψa (j)ψb† (j) (remember that i
and j are distinct points). Moving ψa (j) to the right of ψb† (j), we pick up a minus sign, plus
a term from the nonvanishing anticommutator.
3
Actually, if you have followed the derivation above, E is actually an operator in the degenerate subspace,
but it is still true that it is negligible compared to the leading order energies of the excited states.
 Perturbation Theory: Time Independent  297

This results in
c X
V2 = [ |kij |2 ψa† (i)ψb (i)(ψb† (j)ψa (j) − δab )]. (13.84)
g 2 ij

The second term in round brackets is a sum of local density operators N (x) ≡ ψa† (i)ψa (i),
but these are all equal to 1 on every state in the degenerate subspace, so this term just gives
a constant negative shift in the ground state energy, proportional to gV2 , and does not remove
the degeneracy.
Now write
1
ψa† (i)ψb (i) = [ψ † σ n ψ(σ n )ab + ψ † ψδab ], (13.85)
2
which one can verify by using
tr(σ n ) = 0, (13.86)
tr(σ n σ m ) = 2δ mn . (13.87)
In the preceding equations, we have written the labels on the Pauli matrices as superscripts
in order to distinguish them from the matrix indices a, b.
Putting these equations together, we find that the second-order Hamiltonian is, apart
from the negative constant discussed above
c X
V2 [ |kij |2 (N (i)N (j) + Σn (i)Σn (j))]. (13.88)
g 2 ij

The first term is another constant on the degenerate subspace, while the second gives a
nearest neighbor interaction between local spins, called the Heisenberg Hamiltonian [63].
The sign of the interaction is such that, if the spins were classical, one would lower the
energy by having nearest neighbor pairs to point in opposite directions. This is called an
antiferromagnetic interaction, and the Hamiltonian is called the Heisenberg antiferromagnet.
The actual behavior of its quantum ground state is quite interesting, and depends both on the
dimensionality of space and the type of lattice involved. There are still situations where the
qualitative nature of the ground state is not understood.
It is clear that all of these systems are insulators, called Mott insulators [64]. The com-
bination of the half filling constraint and the large g 2 limit prevents the transfer of charge
through the lattice. On the other hand, for small g 2 , Fermi liquid theory applies, and the
system is a conductor. The Hubbard model has fascinatingly complex behavior as one varies
the coupling, and the nature of the lattice, and it can reproduce the qualitative behavior of
many of the states of condensed matter found in nature.
This concludes our discussion of time-independent perturbation theory for bound states.
We will discuss the application of perturbation theory to the continuum eigenspectrum in
the next chapter.
298  Quantum Mechanics

13.6 EXERCISES
13.1 Determine the energy levels of the anharmonic oscillator

P2 ω2m 2
H= + X + aX 3 + bX 4 ,
2m 2
through second order in perturbation theory.

13.2 Consider the Hamiltonian of Exercise 13.1 with a = 0. Show that one can solve the
Schrödinger equation for the ground state, order by order in a power series in b by an
ansatz of the form ∞
mω 2
bk Pk (x)e− 2~ x ,
X
ψ(x) =
k=0

where Pk is a polynomial. What is the order of Pk ? Show that the Schrödinger equation
leads to a difference equation for the coefficients of the polynomials and the perturbed
energy level E = ~ω + ∞ k
P
k=1 b Ek . Verify that this method of solution agrees with the
results of Exercise 13.1 for the level E2 .

13.3 For a normal matrix, the equation P (a) = det (A − a) is a polynomial in a whose roots
are the eigenvalues ak . Now suppose A = A0 + λA1 , where Ai are normal. Argue that
the roots of the polynomial P (a) are analytic functions of λ except for branch points
where one or more roots coincide.

13.4 Consider a perturbation V = f (r)L · S of the hydrogen atom. Consider all those states
with a fixed value of the principal quantum number n. Describe how V breaks the
degeneracy of those states. Is there any degeneracy left?

13.5 We found the eigenvalues and eigenstates of the harmonic oscillator for any value of
the mass and the spring constant. Now consider an oscillator with a different spring
constant, which is a small perturbation of the original one. Use perturbation theory to
compute the first two corrections to the energy levels and verify that they agree with
the exact formula.

13.6 Repeat Exercise 13.5, for a small perturbation of the mass, rather than the spring
constant. You will calculate different matrix elements, but come to the same conclusion

13.7 Consider a harmonic oscillator perturbed by a constant force, with potential V = −F X.

Solve this problem exactly. Then solve it by perturbation theory, to the first nontrivial
order. You might want to write X in terms of creation and annihilation operators. Show
that the expansion of the exact answer agrees with that calculated by perturbation
theory.
 Perturbation Theory: Time Independent  299

13.8 Consider the nucleus of a heavy atom in a molecule, whose other constituents are much
lighter. The low-lying vibrational excitations of the molecule consist of q oscillations of
ki
that heavy atom around its equilibrium position. with frequencies ωi = m , where ki
are the spring constants in the three principle directions of oscillation. Evaluate the
effect of an external electric field E · X on these levels, to second order in perturbation
theory in the electric field strength.
13.9 When an atom, molecule, or nucleus interacts with an external constant electric field,
the first-order term in perturbation theory typically vanishes. This is due to a symmetry
called parity under which all components of X are reflected. The electric field is a polar
vector, and is odd under parity. If the ground state of the system is nondegenerate, then
it must be an eigenstate of the parity operator U (P ), since that operator commutes
with the Hamiltonian. Show that this means the expectation value of the electric field
perturbation vanishes.
13.10 In Exercise 13.9, suppose the ground state is degenerate. Since U 2 (P ) = 1, we might
have some linear combinations of degenerate states with U (P ) = 1 and others with
U (P ) = −1. In this case, the electric field can have nonzero matrix elements between
states of different parity. Show that if there are k more states of one parity than the
other, then the electric field matrix in the degenerate subspace has k zero eigenvalues.
Show that in the subspace orthogonal to these zero modes, the electric field operator
takes the form (E 1 σ1 ⊗ KN + E2 σ2 ⊗ MN , where MN and KN are N × N matrices,
and N equal to half the dimension of the orthogonal subspace. Show that, in general,
the ground state energy of this system depends on the electric field, to first order in
perturbation theory in the electric field. Since the energy has a term linear in E, we say
that the atom, molecule, or nucleus has a permanent electric dipole moment.
13.11 In our discussion of the Born–Oppenheimer approximation, we showed that molecules
had a definite “shape,” which is not invariant under rotations. This means that the
ground state has a lot of low-lying rotational levels, with energies of order mmmol e
in
Rydberg units. As we saw for ammonia, when the shape is not invariant under parity,
the smallest transition probability between the two parity-reversed shapes led to a
ground state that was a definite eigenstate of parity. Prove that this means that the
expectation value of the electric dipole moment is zero. Thus, the situation outlined
in the previous problem is very special. How is it then that molecules have permanent
electric dipole moments? The answer lies in the large number of low-lying rotational
levels. The typical energies at which we observe molecules are those in which a large
number of these rotational levels can be excited. These levels are calculated by fixing the
orientation of the molecule (e.g., whether the ammonia molecule has the nitrogen atom
above or below the hydrogen plane in some fixed coordinate system) and then quantizing
300  Quantum Mechanics

the rotational motion. The actual state we observe is a time-dependent superposition

of rotational levels. If we consider a general state, allowing for superpositions of the
two orientations of the molecule, there is no reason that the superposition of rotational
levels should be the same for the two orientations. Now consider a small perturbation
of size  mixing the rotational subspaces for the two different orientations. Show that if
we assume random superpositions for the two orientations, the matrix element of the
perturbation is at most of order /N , where N is the dimension of the space of allowed
rotational levels. Show further that these matrix elements are time dependent and that
their time average over times long compared to ~/∆E is of order ~/(∆Et). Here ∆E
is a typical energy splitting between rotational levels.
13.12 The typical scale of rotational energy levels is 10−3 /A in Rydberg units, where A is
the total number of protons plus neutrons in the molecule. For ammonia, how high
does the temperature have to be before we have to take these levels into account in
our description of the physics of the molecule. Recall that room temperature is 273 K,
which is about 1/40 electron volts or roughly 2 × 10−3 Rydberg. Do the same estimate
for water molecules.
13.13 Consider an infinite square well confining a particle to −L ≤px ≤ L and add a harmonic
term V = 21 kx2 . Assume that the oscillator frequency ω = k/m is much smaller than
~
the characteristic frequency of the particle in the well ωw = 2mL 2 . Treat the harmonic

term as a perturbation and calculate the first order perturbed energies of the ground
state and first excited state of the well.
13.14 The problem in Exercise 13.13 can be solved exactly. The solutions of the Schrödinger
equation are parabolic cylinder functions (consult your favorite online math oracle) and
the condition for eigenfunctions corresponds to finding zeroes of these functions. Use the
integral representation or power series solution to show that the exact and approximate
solutions for the eigenvalues exist.
13.15 Repeat Exercise 13.13 for the bound states in a finite square well
k
V (x) = −θ(L − x)θ(x + L)[V0 − x2 ],
2
to lowest order in k.
13.16 Consider a harmonic oscillator on the full real line, perturbed by a finite square well.
Calculate the first-order perturbation of the ground state energy and express the answer
in terms of error functions.
13.17 Evaluate the weak-field Zeeman effect for the states of the hydrogen atom with n = 2.
 CHAPTER 14

Perturbation Theory: Time

Dependent

14.1 INTRODUCTION
Time-dependent perturbation theory is used to study the response of quantum mechanical
systems to time-dependent external fields. However, as a consequence of a trick introduced
by Dirac, it is also a simple way to set up a simple perturbation theory for scattering ampli-
tudes. Dyson introduced a perturbative solution for the evolution operator of a general time-
dependent system. The formulae involve the fundamental concept of a time-ordered product
of operators. The more general concept of a product of operators ordered along some path
in a multidimensional space or space-time is one of the most important notions in modern
theoretical physics.
Time-dependent perturbation theory is the tool of choice for understanding the excitation
and decay of excited atomic or nuclear states by external fields, which is the way we explain
spectral lines. We will derive Dyson’s and Dirac’s formulae, and apply them to a variety of
simple problems.

14.2 DYSON’S FORMULA

Time-dependent Hamiltonians arise in a variety of different ways in quantum mechanics
(QM), most commonly when one is studying a system coupled to another one whose dynamics
is not computed explicitly, but approximated by some external classical time-dependent
couplings in the system Hamiltonian. The fundamental formula for solving time-dependent
problems is a formal solution of the time-dependent Schrödinger equation first written by
Dyson [65]:

i~∂t U (t, t0 ) = H(t)U (t, t0 ). (14.1)

301
302  Quantum Mechanics

Here U (t, t0 ) is the time evolution operator, the unitary operator which maps the initial state
|ψ(t0 )i into the final state at time t
|ψ(t)i = U (t, t0 )|ψ(t0 )i. (14.2)
We also have the obvious boundary condition U (t0 , t0 ) = 1, which allows us to write the
Schrödinger equation as an integral equation
i
Z t
U (t, t0 ) = 1 − dsH(s)U (s, s0 ). (14.3)
~ t0

This may be formally solved by iteration

i
Z t 1
Z t Z s1
U (t, t0 ) = 1 − dsH(s) − ds1 ds2 H(s1 )H(s2 )U (s2 , s0 ), (14.4)
~ t0 ~2 t0 t0

and so on. We can continue to use the integral equation inside the last integral, to expose
more and more explicit powers of H(s). Formally, the answer is written as the sum of an
infinite series of terms, with the n-th term involving integrals over an n-dimensional region
characterized by the inequalities
t ≥ s1 ≥ s2 . . . ≥ sn ≥ t0 . (14.5)
Note that the inequalities ensure that the action of the Hamiltonian operators are such that
operator order reflects time order.
We can write a more symmetric form of this formula by introducing the notion of a
time-ordered product of operators. The time-ordered product is defined by
X
T H(s1 ) . . . H(sn ) = θ(sP (1) − sP (2) ) . . . θ(sP (n−1) − sP (n) )H(sP (1) ) . . . H(sP (n) ). (14.6)
P

P (k) is a permutation of the integers 1 . . . n and we sum over all such permutations.
The n-dimensional hypercube, H, defined by t ≥ si ≥ t0 is completely covered by n! regions
in which the inequality t ≥ s1 ≥ s2 . . . ≥ sn ≥ t0 is replaced by t ≥ sP (1) ≥ sP (2) . . . ≥
sP (n) ≥ t0 . Thus, if T is the hypertriangular region defined by the first inequality, then
1
Z Z
n
d s H(s1 ) . . . H(sn ) = dn s T H(s1 ) . . . H(sn ). (14.7)
T n! H

The formal solution of the Schrödinger equation can thus be written

i
Z t
U (t, t0 ) = T exp(− ds/H(s)), (14.8)
~ t0

where the time-ordered exponential is just the sum of the time-ordered integrals of products
n
of H(si ) over the n-dimensional hypercube, multiplied by (−i/~)
n! .
 Perturbation Theory: Time Dependent  303

14.3 THE DIRAC PICTURE

Now consider a Hamiltonian H = H0 + V , where the perturbation V may depend explicitly
on time. We write the Schrödinger picture time evolution operator as
i
U (t, t0 ) = e− ~ H0 (t−t0 ) W (t, t0 ). (14.9)

Rewrite the Schrödinger equation as an equation for W

i i i i
e− ~ H0 t i~∂t W e ~ H0 t0 = V e− ~ H0 t W (t, t0 )e+ ~ H0 t0 . (14.10)
i i
i~∂t W = e+ ~ H0 V e− ~ H0 t W (t, t0 ) ≡ VI (t). (14.11)
VI (t) is the operator V , with each Schrödinger picture operator replaced by the corresponding
Heisenberg picture operator evolved with the Hamiltonian H0 . VI (t) is called the interaction
picture version of the perturbation V and is time dependent even if the original perturbation
is time independent. The results of the previous section tell us that

i
Z t
W (t, t0 ) = T exp(− ds VI (s)). (14.12)
~ t0

14.4 TRANSITION AMPLITUDES

The quantities of interest in time-dependent perturbation theory are transition amplitudes;
probability amplitudes for one energy eigenstate of H0 to convert into another under the
action of the perturbation. A typical situation is the decay of an excited state of an atom
into its ground state under the influence of a time-dependent perturbation. We will calculate
transition amplitudes to first order in the perturbation. To that order, the amplitude is

i i
Z t i
Z t i
Tab (t) = − hEa |e− ~ H0 (t−t0 ) ds VI (s)|Eb i = e E (t−t0 )
~ a ds hEa |e− ~ (Eb −Ea )s V (s)|Eb i.
~ t0 t0
(14.13)
V (s) is now the time-dependent perturbation in the Schrödinger picture. If we assume that
the time-dependent perturbation vanishes outside of a finite interval, then we can send the
limits of the s integration to ±∞, whenever V (s) vanishes for s > t and s < t0 . The transition
amplitude is then given by the matrix element of the Fourier transform of V (s) between initial
and final states, evaluated at the energy difference between the initial and final states.
An important special case is a monochromatic driving force V (s) = cos(ωs)F (s), where
F (s) is a smooth function which approximates θ(a − s)θ(a + s) with a  ω −1 . If we set
F = 1, the Fourier transform has poles at ω = ± Ea −E ~
b
≡ ±ω0 . When F vanishes outside
the interval [−a, a], those poles are displaced into the complex plane so the result is peaked
304  Quantum Mechanics

around the resonance frequency ω0 . Setting F = 1 and choosing ω close to ω0 , we can write
the transition probability, the absolute square of the transition amplitude, approximately as
|hEa |V |Eb i|2 sin2 [(ω0 − ω)t/2]
Pa→b ≈ . (14.14)
~2 (ω0 − ω)2
Here we have taken t0 = 0 and allowed t to be anything. Note that this expression can
exceed 1 and cannot be correct for ω too close to ω0 except when t is near a zero of the sine.
Higher order perturbation theory becomes important at other times, as you will explore in
Exercise 14.1, and ensures that the transition probability is always less than one.
For finite times, the transition probability oscillates so there are intervals of time over
which the transition is very probable, interspersed with intervals where it is unlikely to occur.

14.5 ELECTROMAGNETIC RADIATION FROM ATOMS

The proper framework for treating the interaction of radiation with atoms is the quantum
field theory called Quantum Electrodynamics or QED. In that theory, the Coulomb term in
the Hamiltonian of atoms, which we have written down is supplemented by an interaction
which allows electrons and protons to change their energy by emitting photons, quantized
excitations of the electromagnetic field. The excited states of atoms are no longer exact
eigenstates of the full Hamiltonian, because they are degenerate in energy with states where
the atom has relaxed to its ground state and one or more photons have been emitted to
preserve the energy balance. Since there are many more states of the second type, and since
the emitted photons quickly propagate far from the atom, the excited state decays. We can
only resuscitate it briefly, by shining light on the atom. Indeed, this is how the spectral
lines associated with excited states are observed. The excited energy levels still have an
approximate meaning, because they are metastable: the probability of a photon emission is
proportional to the fine structure constant α ∼ 1/137 and it takes a long time, on atomic
time scales, for the decay to occur.
In this section, we will describe an approximate scheme for calculating radiative transi-
tion probabilities, which is based on treating the electromagnetic field as a time-dependent
classical perturbation. Unlike most of the approximations discussed in this book, this semi-
classical radiation theory is not the first term in a systematic expansion of the exact transition
amplitudes. Nonetheless, it was of great historical importance, and many of the concepts we
will introduce survive in the systematic treatment of these amplitudes in QED.
The typical spatial extent of an atom is given by the Bohr radius ∼ 10−8 cm. The low
lying electrons in an atom of high Z are closer to the nucleus than this, by a factor ∼ 1/Z,
but radiative transitions in such atoms involve the electrons in outer shells, which feel only a
screened Coulomb potential. On the other hand, the typical energy difference between atomic
levels is ∼ 10eV . This corresponds to a wave length for light emitted in the transition that
 Perturbation Theory: Time Dependent  305

is of order 10−5 cm, so that the spatial variation of the electromagnetic field is negligible
over the size of the atom. This motivates the dipole approximation in which one considers
the electric field of the emitted or absorbed light to be constant in space

E(t, x) = E0 cos(ωt)ê, (14.15)

where ê is the unit vector describing the direction of the field. Taking this to be in the three
direction, we can write the electromagnetic scalar potential

Φ(t, x) = −E0 x3 cos(ωt). (14.16)

This leads to a perturbation of the Hamiltonian of a particle of charge q:

V = −qE0 x3 cos(ωt). (14.17)

Note that we are neglecting the interaction of the electron with the magnetic field of the
electromagnetic wave. This is a consequence of the nonrelativistic kinematics relevant to
atomic transitions. The magnetic field interaction is smaller by a factor of v/c ∼ aB ~me c ∼
1
α ∼ 137 than the electric effect that we are studying.
The transition amplitude is then proportional to the matrix element of the third compo-
nent of the electric dipole operator
d = qx, (14.18)
between initial and final states. Call that matrix element dab . We should note that for X-ray
transitions, where the emitted energy is of order 0.1–100 keV, the dipole approximation is
not reliable.
Our general results on periodic perturbations give us a dipole transition probability
2
|dab |E0 sin2 [(ω − ω0 )t/2]

Pa→b (t) = . (14.19)
~ (ω − ω0 )2

This result is the same whether one is thinking of a transition from a lower energy state
to a higher one or vice versa. The first process, which is called absorption (of light, or of a
photon), is easy to understand intuitively: the electromagnetic field provides the energy to
bump the electron up to a higher state. The second process is there because Hermiticity of the
Hamiltonian forced us to include complex waves with both positive and negative frequencies
e±iωt , so that we get two possible poles in the transition amplitude.1
The second transition is called stimulated emission, and its existence was first pointed
out by Einstein, as well as the fact that the probabilities for absorption and stimulated
1
Never forget that the poles are an artifact of letting the wave exist forever, neglected the damping factor
F (t), which converts the poles into finite enhancements of the transition probability.
306  Quantum Mechanics

emission are equal. At a deeper level, the reason for the connection between these two phe-
nomena stems from fact that the electromagnetic field is quantized, and quantized with Bose
statistics (Chapters 5 and 12). The positive frequency e−iωt terms in the field are multiplied
by photon annihilation operators, and are responsible for absorption processes, while the
negative frequency terms multiply photon creation operators. The creation operator terms
are responsible for both the stimulated emission probability we have just discussed and the
spontaneous emission process in which an atom makes a transition from an excited state to
the ground state (the excited state decays into the ground state plus a photon). The proba-
bilities for all three processes are the same, because the creation and annihilation terms are
related by Hermiticity, and so give rise to amplitudes which are complex conjugates of each
other. Remarkably, Einstein understood all of this [66] in 1917, 8 years before the Schrödinger
equation was invented and even longer before the invention of QED. His arguments2 used
only general notions of probability theory and the Planck distribution for thermal photons,
which was discovered in 1900. The derivation of Planck’s distribution requires QED.
The process of stimulated emission is the fundamental principle underlying lasers and
masers (light or microwave amplification by stimulated emission of radiation). Recall, from
the elementary theory of the harmonic oscillator, that a creation operator acting on a state
√
with n photons gives n times the normalized state with n+1 photons. Now suppose we have
somehow introduced a population inversion into a distribution of identical atoms, so that,
contrary to the expectations of Boltzmann’s statistical mechanics, there are more atoms in
the excited state |Ea i than in the ground state |Eb i. The presence of a low amplitude external
field, whose frequency is tuned near the transition frequency ω0 produces both absorption of
photons, and stimulated emission, with equal probability. Since the population is inverted,
we end up with more photons of the frequency ω0 than we started with, but this means a
stronger field, which enhances stimulated emission, etc. We end up with a final state having
a very strong field, all at the frequency ω0 .

14.6 INCOHERENT PERTURBATIONS AND RADIATIVE DECAY

So far we have studied electromagnetic perturbations of fixed frequency, polarization, and
direction of propagation.3 We now ask how the formulae change for incoherent radiation,
such as one might encounter in a thermal bath. The first step is to write E02 in our formulae
for transition rates in terms of the energy density u in the electromagnetic field.

2
E02 = u. (14.20)
0
2
You can find a clear explanation of Einstein’s arguments in Griffiths’ popular textbook [67].
3
The direction of propagation is transverse to the polarization. It did not appear explicitly because we
made the approximation of a field with no spatial variation.
 Perturbation Theory: Time Dependent  307

We are working in SI units and 0 is the permittivity of the vacuum [68]. In a situation where
the electromagnetic field is an incoherent sum of waves of different frequencies, the energy
density is written as Z
u= ρ(ω)dω, (14.21)
and it is plausible, and turns out to be correct, simply to put the transition rate we have
calculated at fixed ω under the integral sign
2 ∞ sin2 [(ω − ω0 )t/2]
Z
Pb→a (t) = |dab |2 ρ(ω) dω. (14.22)
0 ~2 0 (ω − ω0 )2
If ρ(ω) is a slowly varying, smooth function, and the other term in the integral is
sharply peaked around ω = ω0 , we can pull ρ(ω0 ) out of the integral. Then, introducing
x = (ω0 − ω)t/2, we can evaluate the transition rate approximately (Exercise 14.1) as
π|dab |2
Pb→a (t) ≈ ρ(ω0 )t. (14.23)
0 ~2
Note that this gives a constant transition rate rather than the oscillatory behavior found
for the idealized case of monochrome radiation. This formula is a special case of a very
general result, called Fermi’s Golden Rule: a transition rate is the product of a squared matrix
element, and the density of final states. In many examples, there are many final states and
we must sum or integrate this result over all of them to get what is called the inclusive rate
for the transition. This is what is relevant if we do not measure the detailed properties of the
final state. This is commonly the case for decays of excited states. A common catchphrase
for this sum over possible final states is “integrating over phase space.”
The above formula still assumes fixed directions of propagation and polarization for the
electromagnetic field. The quantity dab is the matrix element of the dot product of the unit
vector of polarization ê with the dipole operator x. Averaging over polarizations4 of the
quantity |ê · dab |2 gives us
|êi · dab |2 = d∗abk Πkl dlab = 1 − |n̂ · dab |2 .
X
(14.24)
i

Here Πkl is the two by two projection matrix on the subspace of three-dimensional space
orthogonal to the direction of propagation n̂. We now average this over the direction of
propagation by doing the integral
1 1
Z
I= sin(θ)dθdφ sin2 (θ) = . (14.25)
4π 3
4
In the current context, where we are discussing absorption or stimulated emission, we do this average
because we assume we are in a state with equal probabilities for the polarizations of the impinging radiation.
This is logically different than the sum over final states we would do, by Fermi’s Golden Rule, if we did
not observe the final polarization.
308  Quantum Mechanics

The final result for the transition rate is

π
Rb→a = |dab |2 ρ(ω0 ). (14.26)
30 ~2

As an example, we can calculate the transition rates between states of a charged harmonic
oscillator, and in Exercise 14.1, you will do similar calculations for transitions in the hydro-
gen atom. A charged oscillator is a crude model of a radio transmitter or other device for
producing electromagnetic waves. In this case, the dipole operator is a linear combination of
a creation and an annihilation operator (of the oscillator, not the electromagnetic field!) so
it has matrix elements only between neighboring states.
s
m~
hn|d|mi = q δn,m−1 ê, (14.27)
2mω

where we have only taken the matrix element to the lower state because we are discussing
emission of radiation in the decay of the excited state. The resonance frequency ω0 is of
course just ω so the transition rate is

mq 2 ω 2
Rm = . (14.28)
6π0 me c3

For constant transition rate, if we start with Nm atoms in the state m, the number left after
time t is e−t/Rm and we define the lifetime of a state to be the time at which only Nm /e of
−1
the atoms are left in the state m, so the lifetime is Rm the inverse of the rate. The half-life,
defined as the time at which half the excited states have decayed away, is used in discussions
−1
of nuclear decays, and is given by Rm ln 2.
The energy radiated in this decay is ~ω and the power radiated is this energy multiplied
by the decay rate
q2ω2 1
P = (Em − ~ω). (14.29)
6π0 me c3 2
The quantum rate of energy emission is of course zero for the ground state. You will calculate
the corresponding power for a classical oscillator in Exercise 14.2. That power is of course
proportional to the acceleration of the charge, and the total power radiated in one cycle of
classical oscillation can be calculated using the equations of motion. It is

q2ω2
Pcl = E, (14.30)
6π0 me c3

and it is the same as the quantum rate when ~ = 0.

 Perturbation Theory: Time Dependent  309

14.7 SELECTION RULES

The rule that oscillator states decay only to the next lowest state is an example of a selection
rule. Selection rules follow generally from symmetries of the unperturbed problem, which
are broken by the perturbation. The transformation properties of the perturbation under
the symmetry of the unperturbed system, then constrain the allowed transitions to lowest
order in perturbation theory. Generally, most of these constraints are lifted in second-order
perturbation theory and none of them are exact (unless of course there is some exact residual
symmetry). In the case of the oscillator, the symmetry is the unitary transformation generated
by the Hamiltonian. The dipole perturbation does not commute with this, but it transforms
simply, being the sum of an energy raising and an energy lowering operation.
A general situation in which selection rules show up is a small perturbation of a rotation
invariant system. The perturbation V can be expanded in terms of operators which transform
as components of some integer spin irreducible representation of the rotation group, which is
to say, like some spherical harmonic. To first order in perturbation theory, we can treat each
of these irreducible components separately. Let us denote the components of an irreducible
perturbation as VA . A runs from −jV to jV , where JV is the spin of the representation under
which VA transforms. Then an irreducible perturbation has the form, Virr = g A VA . Now
consider the action of the angular momentum operators Ja on a subspace of states given
by VA |j, mi, where |j, mi is some collection of degenerate eigenstates of the unperturbed
Hamiltonian which transform in the representation with spin j. Then

J a VA |j, mi = [J a , VA ]|j, mi + VA (J (j) a )mk |j, ki. (14.31)

(j) a
The matrix Jmk is the spin j representation of angular momentum. Now use the fact that

[J a , VA ] = i(J a )AB VB , (14.32)

to conclude that this subspace of states transforms under rotations like the states of a pair
of distinct particles, one with spin j and the other with the spin, call it jV , of the VA
representation. The rules of addition of angular momentum tell us that this contains every
angular momentum between jV + j and |jV − j| exactly once. Thus, if we compute the matrix
elements
hJ, M |VA |j, mi, (14.33)
which induce transitions between eigenstates of the unperturbed Hamiltonian under the
influence of the perturbation, then these matrix elements vanish unless jV + j ≥ J ≤ |jV − j|.
In addition, if the perturbation g A VA only contains certain of the J3 eigenstates in the
representation spanned by VA , then only changes in J3 corresponding to those values, are
allowed.
310  Quantum Mechanics

These remarks are formalized in the Wigner–Eckart theorem [62] which states that
J,jV ,j
hJ, M |VA |j, mi = cM,A,m R(J, jV , j). (14.34)

The coefficients cJ,jV ,j

M,A,m are completely determined by the group theory of angular momentum
(they are called Wigner 3j symbols or Clebsch–Gordon coefficients), while the reduced matrix
elements R(J, jV , j) depend on the angular momentum only in the indicated fashion, but
are specific to the particular system and perturbation. You will find a quick proof of this
theorem in Appendix D on group theory.
Let us apply these rules to first-order electromagnetic transitions in the dipole approx-
imation. The perturbation is a vector and carries m = ±1 in some direction perpendicular
to the direction of propagation of the perturbing wave.5 Thus, dipole transitions only occur
between states whose j values differ by 1 or 0. Furthermore, the value of m must change by
±1 in any transition.
It is an unfortunate fact that transitions that do not obey these rules were called for-
bidden transitions by the early practitioners of atomic QM. They are forbidden only in the
electric dipole approximation. Variation of the electric field across the atom is only down by
a factor of 100 to 1, 000, as are magnetic effects and corrections coming from higher orders
in time-dependent perturbation theory. Thus, forbidden transitions actually occur, albeit at
suppressed rates.

14.8 EXERCISES
14.1 Calculate the rate of electric dipole transitions between two states of the hydrogen
atom.
14.2 Calculate the final state for the dipole transition of a charged oscillator assuming the
initial state is a coherent state |zi. Do this by expressing the coherent state as a sum
of eigenstates, and using the formulae in the text.
14.3 Consider a time-dependent Hamiltonian
H(t) = ~ω(t)a† a,
where a is the usual annihilation operator. Find the selection rules for transitions
between oscillator states if ω(t) = ω + δω(t), where the time-dependent piece is small.
What is the relationship between this Hamiltonian and a harmonic oscillator with time-
dependent frequency?
5
The fact that photons cannot have a longitudinal polarization cannot be understood in nonrelativistic
QM, because it is a special property of representations of the Lorentz group describing massless particles
with spin. However, this fact is encoded in the classical Maxwell equations, so it was understood before a
full appreciation of the properties of QED.
 Perturbation Theory: Time Dependent  311

14.4 Consider a hydrogen atom in its ground state, in the presence of a time-dependent
vector potential (Warning: for the rest of the problems in this chapter, we will use
Gaussian units for electromagnetism.
√ You can convert to the SI units used in the rest
of the book by replacing E0 → 4π0 E0 .)

E0
A(r, t) = c cos(k · r − ωt),
ω
where ω is positive. c is the speed of light. We want to calculate the probability that
this perturbation ejects the electron from the atom into one of the Coulomb scattering
states. This problem will take up the next few exercises. We have broken it into bite
size pieces, but you should do all of them, eventually. In principle, what we are doing is
calculating the matrix element of this time-dependent perturbation between the ground
state of hydrogen and a scattering state with some outgoing momentum p. We are going
to make an approximation based on the fact that for large enough momentum we expect
the outgoing electron to spend most of its time far from the atom, so that the scattering
state can be approximated by a plane wave. To leading order in perturbation theory in
E0 , show that the time-dependent Hamiltonian is
e
V (t) = (ei(k·X−ωt) + h.c.)E0 · P.
2mω
In writing this formula, you must use the Coulomb or radiation gauge for the vector
potential. In this gauge, ∇ · A = 0. Show that the matrix element between initial and
final states of the term shown explicitly gives a delta function δ(Ef − Ei − ~ω) when
integrated over time. Show that the complex conjugate term gives a delta function that
vanishes when Ef > Ei . The complex conjugate term would describe extraction of
energy (spontaneous emission) from the atom, which is impossible for the ground state.
The fact that the two appear with equal strength in the Hamiltonian, a consequence
of Hermiticity, is the fundamental fact underlying the equality of Einstein’s A and B
coefficients.

14.5 In the previous exercise, ~k is the momentum transferred to the atom (which is mostly
carried by the outgoing electron because the nucleus is so heavy) by the space-time-
dependent field. The absolute value of k is equal to ω/c because the field satisfies
Maxwell’s equations. ~ω = Ef − Ei which is R Rydbergs with R > 1 for a liberated
electron. Thus, in Rydbergs, ~k = R/c. On the other hand, the typical scale of momen-
tum in the ground state wave function of hydrogen is ~/aB . Show that the ratio of the
2
momentum transfer to the typical momentum is of order αem R = e~cR ∼ 137 R
. Argue
that this means that we can have a fast moving outgoing electron but still neglect
the term proportional to k · X in V (t). This is called the dipole approximation. Show
312  Quantum Mechanics

that it means that we can treat the electric and magnetic fields as constant in this
approximation.
14.6 Show that putting all of these approximations together, the amplitude for the transition
is
Ef − Ei E0
  Z
Af i = 2πδ − ω N d3 x e−ikf ·x · (−i~∇e−r/aB ),
~ ω
where !1/2
e 1 3/2 1
 
N= .
2m 2π~ πa3B
By integrating by parts show that this is just
E0 · kf √ Ef − Ei
 Z
Af i =N 2πδ −ω d3 x e−ikf ·x e−r/aB .
ω ~

14.7 To do the integral in the previous exercise, we use spherical coordinates

Z Z π Z ∞
d3 x e−ikf ·x e−r/aB = 2π dθ drr2 sin(θ)ei|pf |r cos(θ) e−r/aB .
0 0

Do the remaining integrals.

14.8 The total transition probability is given by the square of the expression for Af i in
Exercise 14.6, after doing the integral. Notice that this includes the square of the energy
delta function, which is infinite. By going back to Exercise 14.4, argue that this infinity
comes from an integral over time, and is therefore due to the fact that we assumed the
perturbation was a plane wave. Argue that if the time is finite but very long we should
view the infinity as simply the length of time over which the wave interacted with the
atom. Show that the transition probability per unit time, which is called the rate is
kf2
!
4a3 e2
Rf i = 2 B4 2 |E0 · kf |2 [1 + (kf aB /~)2 ]−4 δ − Ei − ~ω .
m ~ πω 2m
This formula is an example of Fermi’s Golden Rule.
14.9 The delta function means that this rate is singular (though again this is because we
assumed the time interval was infinite). In a realistic experiment, the detector has a
finite opening angle, and energy resolution, so we should integrate this rate over a small
region between kf and kf + dkf and multiply it by dΩ, the detector resolution in solid
angle. Do the integral over momentum and solid angle, to find the total rate. Show that
it is pf = ~kf
16a3B e2 p3f E02
Rftoti = [[1 + (pf aB /~)2 ]−4 .
3m~4 ω 2
 Perturbation Theory: Time Dependent  313

14.10 The previous Exercise calculated the rate of ionization of hydrogen. The rate of energy
absorption from the beam of light is
dEab
= ~ωRf i .
dt
A plane wave has infinite total energy because it is spread over all of space. Show that if
we put up a perfectly absorbing screen of area σ transverse to the beam, it will absorb
dE screen σcE 2
energy at a rate abdt = 8π 0 . We therefore define the absorption cross section σab
for ionization of hydrogen by the ratio of the real absorption rate to that of a perfect
absorber. Show that it is given by

128a3B πe2 p3f

σab = .
3m~3 ωc[1 + p2f a2B /~2 ]4

The cross section is a useful object because it is independent of the characteristics of

the beam. A real experiment would involve a collimated beam of light rather than a
plane wave. As long as the beam is constant over the size of the atom, the same cross
section will be found.

14.11 Calculate the cross section for ionization of hydrogen when the emitted electron has
5, 10, 50 Rydbergs of kinetic energy. Compare these cross sections to the geometrical
size of the atom, which is defined by the region where the probability distribution for
the electron is not exponentially small, and is of order 4πa2B .

14.12 The cross section falls off rapidly for large pf . The electron inside the atom does not
have a fixed pf because of the uncertainty principle. Argue that the rapid falloff is due
to the fact that it is improbable to find a large electron momentum in the unperturbed
atom. Note that the cross section does not have as fast a falloff as the probability
distribution because the interaction Hamiltonian is proportional to pf and because the
number of final states with momentum pf grows with pf .

14.13 If we repeat the ionization calculation for an ion of charge Z, the cross section scales
like Z 2 . Go back over the derivation and estimate the range of momenta over which
this calculation is valid.
 CHAPTER 15

The Adiabatic Approximation,

Aharonov---Bohm, and Berry
Phases

15.1 INTRODUCTION
The adiabatic approximation deals with time-dependent Hamiltonians whose variation is
slow compared to the oscillation frequencies in the quantum state. In a typical situation,
the gap between the ground state and the first excited state is much larger than ~ω where
ω is any frequency in the Fourier spectrum of the time dependence. The adiabatic theorem
says that the solution of the time-dependent Schrödinger equation is then a phase times the
time-dependent ground state. Michael Berry showed that an important part of the phase has
topological properties. Berry’s phase is responsible for a number of the most bizarre phenom-
ena in quantum theory. The most famous is the Aharonov–Bohm effect, where the quantum
phase of the wave function of a charged particle can measure the magnetic field of a solenoid,
even though the particle is never in a region with nonzero field strength.

15.2 ADIABATIC ENERGY EIGENSTATES

We have seen that the effect of a time-dependent perturbation on an eigenstate of the sys-
tem is predominantly to cause transitions between levels whose energy difference is of order
the frequency of the time-dependent field. In this chapter, we will consider time-dependent
Hamiltonians whose frequencies are much smaller than the energy difference between a pair
of levels. The most common situation to which such an approximation is applicable is systems
where the ground state is separated from the first excited state(s) by a relatively large gap.

315
316  Quantum Mechanics

If the Hamiltonian is time dependent, and is not a small perturbation of a time-

independent system, what we mean in the previous paragraph by the “levels” and the “ground
state” are the eigenstates of the time-dependent Hamiltonian

H(t)|Ei (t)i = Ei (t)|Ei (t)i. (15.1)

It is the gap in this time-dependent spectrum to which we were referring. For a general time-
dependent situation, these states exist but are not terribly interesting, but if H = H(ωt),
with ~ω  |Ei − Ej |, then the adiabatic theorem shows us that they are the right states to
consider. For large systems, it is virtually impossible to satisfy this inequality for all pairs of
states. If the characteristic energy scale of the problem is Echar , then in a large system, we
will have energy splittings of order e−s(E) Echar , where s(E) is the entropy or logarithm of the
number of states with energy ∼ E. However, the gap between the ground state and the first
excited level is of order Echar , so the adiabatic theorem will be valid whenever ~ω  Echar .
We will state the adiabatic theorem for the ground state, and hope that the interested reader
will see that it can apply in more general contexts.
The adiabatic theorem is the statement that if we begin in the adiabatic ground state
|E0 (t = 0)i then the system, to a good approximation, will evolve into
Rt E0 (s)
ehE0 |∂t |E0 i−i 0
ds ~ |E0 (t)i.

To understand how the state |E0 (t)i actually evolves, we differentiate the eigenvalue equation:

Ḣ(t)|E0 (t)i + H(t)d|E0 (t)i/dt = Ė0 (t)|E0 (t)i + E0 (t)d|E0 (t)i/dt. (15.2)

If we differentiate the normalization equation for |E0 (t)i, we get

0 = hE0 (t)|∂t |E0 (t)i + ∂t (hE0 (t)|)|E0 (t).i (15.3)

This shows that the component of the time derivative of the adiabatic eigenstate, along the
eigenstate itself, is purely imaginary.
Introducing the projection operator P (t) on the subspace orthogonal to |E0 (t)i, we can
write
∂t |E0 (t)i = hE0 |∂t |E0 i|E0 (t)i + P (t)∂t |E0 (t)i (15.4)
and we get

1 1
P (t)∂t |E0 (t)i = P (t)Ḣ(t)(1 − P (t)) − (1 − P (t))Ḣ(t) P (t)|E0 (t)i.
E0 (t) − H(t) E0 (t) − H(t)
(15.5)
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  317

Note that the second term annihilates |E0 (t)i. We add it to make the evolution operator δH,
defined below, a Hermitian operator.1 The component of Ḣ|E0 (t)i in the |E0 (t)i direction
is, by the Feynman–Hellman theorem, equal to Ė0 |E0 (t)i and so cancels from the above
equations.
Now let us try to solve the Schrödinger equation with an ansatz

|ψ(t)i = e−iφ |E0 (t)i. (15.6)

The result is
E0 (t)
(φ̇ − ihE0 (t)|∂t |E0 (t)i)|E0 (t)i = |E0 (t)i + δH|E0 (t)i, (15.7)
~
where
1 1
δH = i~[P Ḣ(1 − P ) − (1 − P )Ḣ P ], (15.8)
E0 − H E0 − H
and we’ve used the eigenvalue equation [H(t) − E0 (t)]|E0 i = 0. This equation is inconsistent,
unless the term involving δH is negligible. When δH is negligible, it is the proof of the
adiabatic theorem.
If H(t) is a smooth function of ωt with ~ω  Egap , then δH is of order ~ω/Egap . Note also
that at large eigenvalue, a region we might worry about if the spectrum of the Hamiltonian
is unbounded, this operator is still bounded by something of order ω. Indeed, we can write
X
H(t) = αi (ωt)Hi ,

and at large eigenvalue P 0

P
α Hi
δH = ω P i ,
αi Hi
where the prime denotes derivative with respect to the argument of the function. The operator
multiplying ω is bounded by something of order 1 in the large eigenvalue region. For example,
p2
if we had H(t) = 2m(t) + V (x, t), then the large eigenvalue region is dominated by the kinetic
term and we can approximate Ḣ ṁ
H ∼ − m ∼ ω. It should be clear that nothing in this argument
actually used the fact that |E0 (t)i was the adiabatic ground state. We used only the fact
that it was separated from all other adiabatic eigenstates by a gap  ~ω.
A simple example of the adiabatic theorem, of remarkably general utility, is a two state
system with Hamiltonian X
H(t) = Ba (ωt)σa , (15.9)
1
We use the fact that P (t) commutes with H(t)−E0 (t) to show that the operator appearing in the equation
for the time derivative of |E0 (t)i is anti-Hermitian.
318  Quantum Mechanics

with ~ω  |B|. The adiabatic eigenvalues are ±|B|(t). The adiabatic eigenstates are
solutions of
X
Ba (ωt)σa |E± (t)i = ±|B|(t)|E± (t)i. (15.10)

In the basis where σ3 is diagonal, these equations read

√ !
±
!
±
!
1 − BB∗ √ B∗ a 1 a 1
=± ± , (15.11)
B − 1 − BB∗ a±2 a2

B1 (t)+iB2 (t)
where B = |B(t)| . The solution of these equations is

B∗
a±
1 = −√ a±
2, (15.12)
1−B∗B±1

which determines both coefficients up to an overall phase, when combined with the normal-
ization condition |a± 2 ± 2
1 | + |a2 | = 1.
If we think of this Hamiltonian as a description of a spin in a time-dependent magnetic
field, then the simplest way to describe the adiabatic approximation is to say that the state
of the spin lines up with the magnetic field (we have incorporated the dipole moment into
the vector Ba , so in the case of a real magnetic field, we could have antialignment as well).
The operator that controls the corrections to the adiabatic approximation is

Ba0 σa
δH = ωP , (15.13)
−|B| − Ba σa

where P = 1 − |E− (t)ihE− (t)|. It is an operator of norm 1 and its eigenvalue is bounded
0|
by ω |B
|B| , which is much smaller than one, when the conditions for validity of the adiabatic
theorem are valid.

15.3 THE BERRY PHASE

The phase factor φ decomposes into a dynamic Rt
phase, which depends on the adiabatic energy
level of the system, and a term γ0 (t) = 0 ds ihE0 (t)|∂t |E0 (t)i, called the geometric phase.
The geometric phase suffers from a certain degree of arbitrariness, because one is always free
to redefine the eigenstate |E0 (t)i by multiplying it by an arbitrary phase factor |E0 (t)i →
eiθ(t) |E0 (t)i. This changes the integrand of the geometric phase by ∂t θ. No measurement at a
single time t can be sensitive to such a change. Furthermore, the geometric phase difference
at two different times depends only on the wave functions at those times and not on the
intervening history. For 60 years, the geometric phase was considered unobservable.
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  319

It was the genius of Michael Berry [69] to recognize that the ambiguity in the geometric
phase was a kind of U (1) gauge invariance.2 Let us assume that the time-dependent change
in the Hamiltonian is a change in the coefficients of k different operators in an expansion
H = Y i Hi in some canonical basis of operators. In our two state system, the Y k would be
P

the three components of the magnetic field Ba . Then

d|E0 i/dt = Ẏ k ∂Yk |E0 i (15.14)

and Z s
γ= Ẏ i (s)Ai (Y (s)), (15.15)
0

where the Berry Connection or Berry Vector Potential3 is defined by Ai = ihE0 |∇i |E0 i. The
phase can therefore be thought of as the line integral of a generalized vector potential. If
the parameter space of the Y k has the topology of flat k dimensional space,4 then the line
integral of Ai around a closed path is equal to the integral of ∂i Aj −∂j Ai , the Berry Magnetic
Field or Berry Curvature, over any two-dimensional surface whose boundary is that path.
This is a generalization of Stokes’ theorem from electromagnetism.
A simple way to generate a nonzero Berry phase is to split a beam of particles initially
prepared in the same quantum state, and guide the two halves of the beam around two
different paths in space, which intersect at some point in the future. Subject one of the
beams to an adiabatic change in its Hamiltonian, by changing some external fields localized
around that beam’s path and turned off before the beams cross. Then the particles Rin one
t E(s)
beam have wave function ψ, while those in the other have a wave function ψeiΓ e−i 0 ~ ,
where Γ is the geometric phase. The interference of these two beams can measure Γ, if one
can separate out the contribution from the dynamic phase. It turns out that this is possible
experimentally, and the Berry phase has been measured. We will discuss a particular example
of this in the next section.
In the case of a two state system, the Hamiltonian depends on three “magnetic field”
components Ba (t). The Berry potential is thus a function Aa (Ba ) of three variables, just like
a vector potential in electrodynamics. Thus, given any Berry potential with nonzero curl,
there will be closed curves Ba (t) in the space of couplings in the two state Hamiltonian, such
2
Berry’s gauge invariance should not be confused with the U (1) gauge invariance of electromagnetism,
although the Aharonov–Bohm effect, which we will explore in the next section, does lead to a conflation
of the two. They are, however, logically distinct concepts, which become intertwined in that particular
example.
3
The choice between these two names depends on one’s attachment to physical or mathematical terminology.
Mathematicians would call this the connection in the line bundle over the parameter space Y k defined by
the state vector. Physicists tend to think of it as a k dimensional generalization of a vector potential.
4
If the parameter space has noncontractible closed loops, like a torus, then the line integral can be nonva-
nishing even when the Berry Curvature is zero.
320  Quantum Mechanics

that the Berry phase is nonzero. In particular, if the magnitude |B| of Ba (t) is fixed, the
closed curve will lie on the surface of a sphere of radius |B| and the Berry potential will only
have two components, and its curl only one component.

15.4 THE AHARONOV–BOHM EFFECT [70]

Let us return to the problem of a charged particle in a magnetic field, this time the field of
an infinite solenoid. The B field of such a solenoid vanishes outside of a cylinder whose cross
section is a small circle of radius r0 in the x1 − x2 plane. We will be interested in the wave
function of charged particles in regions separated from that circle by a distance R  r0 . For
example, we can put the electrons in a square well potential (V0 > 0)
V (r) = −V0 [θ(R + a − r) − θ(R − r)].
r is the two-dimensional radius and θ(x) is the Heaviside step function, equal to 1 when its
argument is positive and 0 otherwise. We will consider electrons with energy E + V0  V0 .
For any value of r, the Schrödinger equation is that of a free electron in the solenoid
field, with E → E ± V0 . For r outside the negative region of the potential, the wave function
is small and we will neglect it (make V0 very large). Just as in the constant field case, the
motion in the x3 direction decouples, and we will ignore that as well. It is convenient to work
in cylindrical coordinates.
Stokes’ theorem from classical electrodynamics tells us that the line integral of A around
any curve encircling the solenoid once is equal to the magnetic flux F through the solenoid.
These line integrals around closed curves are completely gauge invariant. In cylindrical
coordinates, a potential with this property is
F
(Az , Ar , Aφ ) = (0, 0, 1). (15.16)
2πr
The Schrödinger equation with energy E in this background field, plus an external potential
V (z, r, φ), has the form
~2 2 1 qF 2
− [∂ + ∂r2 + 2 (∂φ − i ) ]ψ = (E − V )ψ. (15.17)
2m z r 2πr
The solutions of this equation have the form
qF
ψF = ψ0 ei 2π~ φ , (15.18)
where ψ0 is a solution with vanishing flux.
Now begin with a beam of electrons approaching the origin along the trajectory φ = 0, in
the plane z = 0, and consider a potential, which includes a term describing a device that can
split a beam of electrons at some point (z, r 6= 0, φ = 0). On average, half the electrons are
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  321

given velocities v = (0, ± v1 , 0) in Cartesian coordinates. The potential away from the beam
splitting point attracts the particles back toward the origin. Each electron in the beam is in
the same initial spatial wave packet, but with time delays, so that we can neglect electron
interactions. However, the potential is asymmetric between left and right, so that there is
a finite probability that an electron which moves to the right after the beam splitting, will
come back to φ = π at the same time as one which is diverted to the left. The scattering
amplitude between the two electrons will include a phase
F
R 2π
SAB = eiq 2π~ 0 , (15.19)
which comes from the phases accumulated by the two electron wave functionsR as each tra-
verses half the circle. Note that the logarithm of this phase is proportional to dθAθ . Since
the magnetic field vanishes everywhere except at r = 0, this phase is, by Stokes’ theorem from
electrodynamics, independent of the detailed paths followed by the two electrons. Only the
topology of the closed curve formed by the trajectories of the two electrons is relevant to the
phase in scattering, which is also completely gauge invariant. If the closed curve circles the
line r = 0 in three-dimensional space, along which the magnetic flux is concentrated, then
we pick up a flux-dependent phase, otherwise we do not.
The Aharonov–Bohm (AB) effect has been verified experimentally, first in the 1960 exper-
iment of Chambers [71], and in many subsequent experiments, but astonished physicists when
it was first proposed, because it is an electromagnetic effect on the electron in regions where
all electric and magnetic fields vanish. Only the vector potential is nonvanishing and only
the nonlocal topological integral of the vector potential appears in the AB scattering phase.
Aharonov told me that his Ph.D. thesis committee brought in R.F. Peierls, an outside expert,
to debunk the claims of Aharonov and his advisor Bohm. After grilling Aharonov for hours,
Peierls said to him “Young man, I know that you’re wrong, but you’ve defended your claim so
well that you deserve the Ph.D.” Nowadays, the AB effect is considered obviously correct and
lies at the root of our understanding of the generalizations of electrodynamics that describe
the Standard Model of particle physics.
As Berry first pointed out in his seminal paper, the AB effect is an example of a Berry
phase. To see this, we look at a different setup than the original AB experiment. The potential
is now taken to be an infinite square well, confining a single electron to a box sitting at some
position Z in three-dimensional space, whose r coordinate is nonvanishing. The size of the
box is much smaller than the r coordinate of Z. As before, we can solve the Schrödinger
equation in the presence of the flux by
F
ψ = eiq 2π~ φ ψ0 . (15.20)
Since ψ0 vanishes outside the box, all that matters in this formula is the value of φ in the
box. Now consider an adiabatic variation Z(t) of the position of the box, which starts and
ends at the same point. The Berry phase is
322  Quantum Mechanics
Z
γ=i hψ|∇Z ψi · dZ. (15.21)

This has two terms, the Berry phase from ψ0 and that coming from the variation of φ in the
flux term, as we vary Z. Since φ comes back to the same point, the latter is just the AB flux.
The wave function of any stationary state in the square well potential is a function of
x − Z, so the gradient w.r.t. Z that appears in Berry’s formula may be replaced by a gradient
w.r.t. x. The latter is proportional to the momentum operator so the ψ0 part of the Berry
phase is proportional to the expectation value of the momentum in an eigenstate of the square
well. The square well eigenstates all obey Dirichlet (vanishing ψ0 ) boundary conditions at
the walls of the box. That is, they are standing waves and have zero momentum expectation
value. Thus, the AB phase is just the Berry phase for this special case.

15.5 ANYONS, FERMIONS, AND THE SPIN STATISTICS THEOREM

The Aharonov–Bohm effect gives us insight into the meaning of Fermi statistics, but to
understand this insight, we have to take a detour into a world with one less spatial dimension
than our own. This is called 2 + 1 dimensional physics: a three-dimensional space-time with
one time dimension.
In 2 + 1 dimensions, an infinitely thin solenoid is just a point object, so we can imagine
point particles, which carry “magnetic” flux, as well as charge. We put magnetic in quotation
marks, because the gauge field in question is not really electromagnetic. It will be called the
“statistical gauge field,” because, as we will see, it determines both the particle’s statistics
and its spin. More precisely, we will see that coupling a collection of identical particles, with
Bose statistics, to such a statistical gauge field, leads to a correlated change of both spin and
statistics. In 2 + 1 dimensions, we can have noninteger spin and forms of statistics different
from either Bose or Fermi, but we will see that this is incompatible with rotational symmetry
in higher dimensions. Only Bose and Fermi statistics are possible in higher dimensions, and
a change of statistics leads to a change of spin by 1/2.
In the real world, the connection between spin and statistics is very tight. Fermions all
have half integer spin and bosons all have integer spin. We can explain this by postulating
a set of bosons with integer spin, some of which are coupled to the statistical gauge field,
giving them half integer spin and Fermi statistics. However, there is nothing in the formalism
of nonrelativistic quantum mechanics which prevents us from postulating the existence of
bosons with half integral spin.
The 2 + 1 dimensional “electric field” is a two-vector E i and the “magnetic” field is an
antisymmetric tensor Bij . They can be combined together into a three-dimensional anti-
symmetric tensor. Let us introduce the speed of light, simply to convert time units into space
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  323

units: ct ≡ x0 . This does not imply that we are writing a relativistic theory. For applica-
tion of these equations to some condensed matter system, we might choose instead to use
some characteristic propagation speed c in the material under study. Then the field strength
tensor is
Fµν = ∂µ Sν − ∂ν Sµ , (15.22)
where Sµ is a three-dimensional vector potential, which we call the statistical gauge potential.
We have F0i = ∂0 Si − ∂i S0 ≡ Ei , and Fij = ∂i Sj − ∂j Si ≡ ij B.
We say that a particle has statistical charge q if there is a term in its classical action5 of
the form
dxµ
Z
δS = q ds Sµ (x(s)). (15.23)
ds
In this equation, we have introduced an arbitrary parameter s to describe the path of the
particle xµ (s) in space-time. The term δS in the action does not depend on the choice of this
parameter, as you will demonstrate in Exercise 15.1.
In nonrelativistic physics, there is a natural choice of time parameter, s = t, the absolute
time of Newton’s Principia. For this choice, x0 = ct. The full action for a statistically charged
particle is
mẋ2 dxi
Z
dt + S0 (x(t)) + Si (x(t)). (15.24)
2 dt
The equation of motion following from the condition of stationarity of this action is

dSi (x(t)) dxj

mẍ + = ∇i S0 + ∇i Sj , (15.25)
dt dt
which is the same as the Lorentz force equation.
To define carefully what we mean by a particle with both charge and flux, we model
such an entity as a charged particle stuck to the wall of a circle surrounding a pointlike flux
Φ, and then shrink the circle to zero size [72]. It can be shown [73] that the results we will
obtain are valid for a much wider range of models of what a particle with both charge and
flux looks like. If we rotate our model charge–flux composite by 2π, the charge picks up an
qΦ
Aharonov–Bohm phase ei ~ . On the other hand, the response to this rotation is described
by the action of the rotation operator e−2πiJ on the quantum state of the composite system.
We conclude that
qΦ
J =m− , (15.26)
2π~
where m is an integer. As usual, the fractional part of this is considered to be internal spin.
qΦ
The simplest composite would have J = − 2π~ .
5
See Chapter 4 for the definition and properties of the classical action.
324  Quantum Mechanics

Now consider two such charge–flux composites at positions ±x in the x, y plane. We can
exchange them with each other by doing a rotation of π around the origin. This is an adiabatic
motion of the particles. Its only effect comes through the statistical Aharonov–Bohm phases
particles pick up by moving charges around fluxes. The particles remain stationary with
respect to their own fluxes during this motion, so the phases come from the motion around
the other particle’s flux. Assuming that all particles/fluxes are identical, these add up to
exactly the phase experienced when a single particle traverses around its own flux in a 2π
rotation. This shows that there is a relation between the change in statistics of a particle due
to interaction with Sµ and its change in spin. In 2 + 1 dimensions, no principle prevents qΦ
from being an arbitrary, even irrational, number in units of 2π~. Particles with general spin
and statistics are called anyons [74].
qΦ
For general values of the angular momentum J = − 2π~ , the multibody constraints on
the wave function become very complicated if we also insist that the wave function satisfy the
Schrödinger equation for a free particle. They have only been solved in the limit of infinite
mass, where the particle’s kinetic energy is dropped from the Hamiltonian. This limit turns
out to be relevant to the collective excitations of matter in a certain kind of insulating phase
called the fractional Quantum Hall regime [75]. The Hall Effect is a well known phenomenon
in classical electrodynamics, in which an electric field applied to a material in one direction,
leads to a flow of current in a perpendicular direction. For a planar sample, this is summarized
by an equation
J i ∝ ij Ej . (15.27)
This equation is related by boosts to an equation relating charge density J0 to magnetic flux
density B = ∂1 A2 − ∂2 A1
J0 ∝ 0ij ∂i Aj . (15.28)
Indeed, if we introduce a time coordinate x0 = vt, where v is some characteristic velocity of
the material under study, then we can write the two equations as
Jµ ∝ µνλ ∂ν Aλ , (15.29)
where we have written the scalar potential of electrodynamics as A0 . The Greek indices run
over 0, 1, 2.
This equation resembles the equations for the statistical gauge field, but involves the
electromagnetic potentials instead of the statistical one. The theory of the fractional quantum
Hall effect (FQHE) is based on localized collective excitations of effectively planar materials,
which have negligible kinetic energy. They behave like anyons with a variety of fractional
values of the statistics parameter qF and also carry fractional electric charge. We do not
have space here for a full description of the FQHE, but students should be aware that
it is one of our most beautiful illustrations of quantum effects, and of the importance of
Aharonov–Bohm–Berry phases in particular.
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  325

Returning to 3 + 1 dimensions, we can of course obtain the same effects by attaching

infinite statistical flux tubes to particles. If these were real electromagnetic fluxes, they
would contribute an infinite energy to the particles, but the statistical gauge field equations
do not have Maxwell like terms in them. They define a local relation between the flux and
the particle currents.6 So there is no infinite energy associated with these flux tubes. Still,
they appear to introduce a violation of rotation invariance into the description of particles,
by picking out a line in space associated with the tube. This intuitive observation is made
mathematical by noting that the spin-statistics connection we have derived is valid in four
dimensions for rotations in the plane perpendicular to the flux tube. Thus, the only way that
a particle with flux attached can have the same rotation properties as a normal particle is
if its Aharonov–Bohm statistical spin takes on one of the values allows by the full rotation
group SO(3). We have learned that the eigenvalues of J for rotations in any plane must be
integer multiples of 1/2, so the only allowed fractional spin/statistics is fermionic.
To summarize: we can change the statistics of particles in d space-time dimensions, or
more properly of the quantum field which changes the number of particles, by coupling it
to a statistical gauge field which attaches a flux tube of dimension d − 2 to each point-
localized excitation of the field. If d ≥ 4, this is consistent with the rotational properties
of particles, only if the statistical phase is either trivial or fermionic. The argument that
quantizes the statistical phase comes from quantization of angular momentum. This way of
viewing fermions, as bosons coupled to a field that generates only Aharonov–Bohm phases
of ±1,7 and has no other physical effect, seems quite intuitive, and allows for an explanation
of the spin-statistics connection which is transparent and independent of special relativity or
the existence of antiparticles.
However, nothing in our argument so far prevents us from starting the process of statistical
flux attachment with bosons of spin one half. We would then obtain integer spin fermions,
by adding an interaction to a statistical gauge field. Thus, although the idea of a statistical
gauge interaction is an attractive way to understand the origin of Fermi statistics, it is not
enough to prove the tight connection between spin and statistics, which we observe in the
world. To derive the spin statistics connection, we observe in the real world, we have to
make the additional assumption that bosons with half integer spin cannot exist. In fact, this
assumption follows from combining quantum mechanics, special relativity and the principle
that Einstein–Podolsky–Rosen (EPR) correlations cannot be used to send messages faster
than the speed of light.8

6
In dimensions higher than 2+1, the flux is an extended object of dimension d−2, where d is the space-time
dimension. The corresponding “electric and magnetic fields” form an antisymmetric tensor of rank d − 1
in space-time and the flux-current relation is J µ ∝ µν1 ...νd−1 Fν1 ...νd−1 .
7
Such a gauge interaction is called a gauge theory with Z2 gauge group, or Z2 gauge theory for short.
8
See Chapter 22.
326  Quantum Mechanics

The Z2 statistical gauge field interpretation of Fermi–Dirac statistics implies that fermions
can only be created locally in pairs. Notice that this also follows from the fact that, in a theory
invariant under rotations, half integer spin particles can only be created in pairs, because the
only operators carrying half integer spin are creation/annihilation operators of such particles.
Relativity ties together these two very similar restrictions on particle interactions. We note
that the discussion of anyons here has been very condensed. The reader is urged to consult
the excellent lecture notes of J. Preskill [72] for more details.

15.6 EXERCISES
µ
ds Sµ (x(s)) dx
R
15.1 Prove that ds is independent of the choice of the parameter s.

15.2 Let us consider the Hamiltonian H = −x · σ, for a two state quantum system in more
detail. The three parameters x can be slowly varying functions of time, so that we can
apply the adiabatic approximation. The adiabatic eigenstates are denoted |nxi, where
n = 1, 2. The two states have different eigenvalues generically, but at x = 0 they are
degenerate. We will see that this degeneracy shows up as an interesting structure in the
Berry potential. The Berry potential for the state n is

A(n) = i~hn(x)|∇|n(x)i.

Use the fact that for all x

hn|mi = 0 = hn|H|mi,
to show that
hn|∇H|mi
hn|∇|mi = .
Em − En
This result is true for any finite dimensional Hilbert space, not just the two state system.

15.3 Compute the “magnetic field” of the Berry vector potential using the result of Exercise
15.2. It will be convenient to insert a complete set of states in this calculation and use
the fact that there are only two states and therefore, one energy difference.

15.4 Consider a Hamiltonian H(α), for any number of particles, depending on parameters
αi . Assume the wave function in some basis is real for all values of αi . Show that the
Berry phase vanishes.

15.5 Consider a system with νN electrons in a very strong magnetic field. We know from
Chapter 9 that the lowest Landau level for noninteracting electrons in this field is highly
degenerate. N is the maximum number of electrons in the lowest Landau level and ν is
the fraction of that number in our sample. The lowest Landau level is parameterized by
 Adiabatic Approximation, Aharonov--Bohm, and Berry Phases  327

any analytic function F (z1 , . . . , zνN ), which is antisymmetric in the coordinates. The
full wave function is P ∗
B
F e− 4~c i zi zi .
Now reintroduce the electron interactions. Since the gap to the next Landau level goes
to infinity with the strength of the external field, and the strength of the interactions
is independent of the field, the ground state of the system is given by some particular
choice of F (assuming the interactions lift the degeneracy). Note that the electron spins
are polarized by the field so every electron has a single preferred spin state in the Landau
1
level. Laughlin guessed that when ν = 2k+1 the ground state wave function was

νN Y
Y
FLaughlin = (zi − zj )2k+1 .
i=1 j<i

Note that when a pair of electrons are close, the wave function vanishes rapidly, thus
minimizing the Coulomb repulsion. Numerical studies and spectacular agreement with
the qualitative properties of real systems in strong magnetic fields have convinced every-
one that Laughlin’s guess was right, and he was awarded the Nobel Prize. Laughlin also
proposed that localized excitations, called quasiholes, of the ground state had the form
νN
Y
Fquasihole = (zi − z0 )FLaughlin .
i=1

A state with two quasiholes, one at the origin, and one at z is

νN
Y νN
Y
(zi − z) (zi )Flaughlin .
i=1 i=1

Show that if we adiabatically move z in a circle around the origin then the wave function
picks up a Berry phase eiνπ . Thus, Laughlin’s quasiholes are anyons. The existence
of these excitations (which also carry fractional electric charge) has been verified by
experiment [75].
 CHAPTER 16

Scattering Theory

16.1 INTRODUCTION
The phrase scattering theory refers to a set of general results about systems of particles with
a Hamiltonian H0 + V , where H0 describes the free motion of the particles, and V falls
off rapidly in the regions where particle coordinate differences rij = |ri − rj | are all large.
−3
Typically, “falls off rapidly” means at least as fast as rij . We have seen in Chapter 8 that
some aspects of scattering theory apply even to the Coulomb potential. We always imagine
that our system is translation invariant, but that we have diagonalized the total momentum
of the system. If the system is also Galilean invariant, we can go to the rest frame of the
center of mass.
Scattering theory can be thought of in two different ways, both coming from an analysis
of the Schrödinger equation in the asymptotic region. The eigenstates of H0 are free particle
wave functions of all possible momenta. In particular, we can divide them into momenta that
are pointing further in the asymptotic direction (outgoing) and those which point toward
smaller values of rij (incoming). In a given part of the asymptotic region, the incoming and
outgoing states look independent, but they are not. If there is no interaction, we know that
the exact eigenfunctions are plane waves, and in incoming plane wave at e.g., x3 → −∞ is
the same as an outgoing plane wave at x3 → ∞. So also in the interacting case, the incoming
and outgoing states are simply two different bases of the Hilbert space.
The physical description of scattering is to start at time t0 → −∞ with localized incoming
wave packets for each particle, whose Fourier transforms are smooth functions of momentum
i
ψin (ki ). One then imagines acting on this state with the evolution operator of the interaction
picture
R∞
−i ds V (s)
S = Te −∞ . (16.1)

This scattering operator or S-matrix tells us what any incoming free particle state evolves
into. The hypothesis of scattering theory, which can be proven rigorously for some class of

329
330  Quantum Mechanics

interactions V , is that the asymptotic states are complete (see the discussion of bound states
below). Then we can compute an S-matrix, from the matrix elements of S in the incoming
basis. It is a nontrivial fact that this same matrix can be computed as the unitary matrix
describing the transformation between the incoming and outgoing bases of the Hilbert space

hi (in)|S|j (in)i = hi (out)|j (in)i. (16.2)

In terms of this second formula for the S-matrix, we will formulate the Lippman–Schwinger
equations for scattering. Our explicit discussion will apply only to single particles scattering
from a potential, which is equivalent to two particles interacting via a translation invariant
potential. We will introduce a perturbative series for the S-matrix, called the Born series,
and show that for smooth potentials, the first term dominates at high energy. For spherically
symmetric potentials, we can solve the scattering problem for each fixed value of angular
momentum. The S-matrix is then diagonal and written in terms of phase shifts eiδl (E) . We
will show that at low energy, the l = 0 phase shift dominates.

16.2 GENERAL FORMALISM

For two body systems, the bound state and continuous spectrum are disjoint from each other,
and we cannot have transitions between scattering states and bound states, but for systems
of three or more particles, we can form bound states of some subset of particles even when
the whole system has positive energy. This is sometimes called the slingshot effect in the
classical mechanics of astronomical systems. A satellite falls into a negative energy bound
orbit around a planet by dumping energy into a third system of particles, which escapes to
infinity. In scattering theory, we consider such true bound states as new particles and include
their free asymptotic motion in what we call H0 . Thus, in general, the number of each type
of particle does not have to match between initial and final states.
Let us consider the subspace of the Hilbert space with a fixed positive energy E, and
let |ψ0 , ki be a basis for the subspace of the Hilbert space with that same eigenvalue of H0 .
The label k stands for all of the momenta of any collection of free particles (including freely
moving bound states) with that energy, as well as the spins and other quantum numbers
of those particles. We expect there are eigenstates of the full Hamiltonian H, which can be
labelled by those same quantum numbers. That expectation is based on the fact that we
can solve the Schrödinger equation approximately in the region where all rij are large, by
simply dropping V from the equation.1 We call |ψ, k±i the basis of H eigenstates, which
asymptotically approach |ψ0 , ki in the limit rij → ∞. We will see the reason for the extra ±
label in a moment.
1
It is important to note that we are implicitly including effects of V when we include bound states in the
list of free asymptotic particles described by H0 .
 Scattering Theory  331

The Schrödinger equation is

(H0 + V − E)|ψ, k±i = 0. (16.3)

We write
|ψ, k±i = |ψ0 , ki + |φ, k±i, (16.4)
and rewrite the equation as

(H0 − E)|φ, k±i = −[V |ψ0 , ki + V |φ, k±i]. (16.5)

The operator H0 − E is not invertible, because it has zero eigenvalues, but the operators
H0 − E ± i are invertible for arbitrarily small positive . These operators are not Hermitian,
but they are normal. Consider for a moment using them as time evolution operators. They
give rise to evolution of the form
it
E ± = e− ~ [(H0 −E)±i] . (16.6)

We have avoided the usual notation U (t) because these are not unitary operators. As t →
−∞, E+ sends all states to zero, while E− annihilates all states at t → ∞. Now define
1
|φ, k±i = [V |ψ0 , ki + V |φ, k±i]. (16.7)
E − H0 ± i
That is to say, |φ, k±i are the solutions of these equations, which are called the Lippmann
Schwinger equations. In the limit  → 0 |ψ, k±i, solve the Schrödinger equation with energy
E and define two different bases of the eigenspace with energy E. |ψ, k+i is called the out
basis and |ψ, k−i is called the in basis. The S-matrix or scattering matrix is the matrix of
overlaps between these two bases.

Sk0 k = h−ψ, k 0 |ψ, k+i. (16.8)

This is the matrix of the unitary S-operator, which transforms in states to out states

S|ψ, k, −i = |ψ, k+i. (16.9)

16.3 POTENTIAL SCATTERING

The details of multiparticle scattering are quite intricate, and we will content ourselves with
a description of scattering theory of a single particle from a potential. Equivalently, this
formalism defines two body scattering in the rest frame of the center of mass, when the
interaction is translation invariant. With that interpretation, the mass parameter appearing
in the equations is the reduced mass of the two body system.
332  Quantum Mechanics

Let us write the Lippman–Schwinger equation for this system:

1
Z
ψ(x) = φ(x) + d3 y hx| |yiV (y)ψ(y). (16.10)
E − H0
2 2 2
It is convenient to write the potential as V = 2m ~
U , the energy as E = ~2mk , and scale a
similar factor out of φ which is a solution of the free Schrödinger equation. Then we get
Z
ψ(x) = φ(x) − d3 y G0 (x − y)U (y)ψ(y), (16.11)

where
1 eik|x−y|
G0 (x − y) = − (16.12)
4π |x − y|
is the Green function of the Helmholtz operator H = ∇2 + k 2 .
Since U falls off rapidly at infinity, we can write an approximate solution at large r = |x|,
using
eikr −ikx̂·y
G0 (x − y) ≈ e , (16.13)
r
where x̂ is the unit vector in the x direction. We choose the normalization of φ to be simply
φ = eik·r and get
1 eikr eikr
Z
ψ−φ=− d3 ye−ikx̂·y U (y)ψ(y) ≡ f (x̂, k). (16.14)
4π r r
f is called the scattering amplitude. It is the matrix element of the operator 1i (S − 1) ≡ T ,
which is called the transition operator.

16.4 THE BORN APPROXIMATION

There are two methods of approximating this formula, which have proven useful. The first,
called the Born approximation or Born series, is useful when the potential is weak, but also
when k is very large. In this limit, if U is a smooth function of position (has finite derivatives
of all orders), then the integral falls off rapidly at large k, so even if the potential is large
somewhere, it is a small perturbation at large k. Then, we can solve the equation by iteration,
as we did for the equations in the interaction picture. This gives us a series in powers of U ,
the first term of which is
1
Z
fBorn (kx̂) = − d3 y eikx̂·y U (y). (16.15)
4π
q = kx̂ is the momentum transferred to the particle by the scattering event. It is worth
remembering that the first Born approximation to the scattering amplitude is the Fourier
transform of the potential. The higher order terms in the Born series depict the particle
scattering from the potential multiple times, with free propagation between encounters.
 Scattering Theory  333

16.5 PHASE SHIFT ANALYSIS

For spherically symmetric potentials, there is another general method of analysis, which
is best suited to low-energy scattering. This is called phase shift analysis, and we have
already encountered it for the Coulomb potential in Chapter 8. For any spherically symmetric
potential, which falls off sufficiently rapidly, the asymptotic Schrödinger equation is just
2 l(l + 1)
[∂r2 + ∂r − + k 2 ]ψ = 0. (16.16)
r r2
The large r behavior of the solutions is captured by the JWKB approximation (see Chapter
±ikr
17): e r . The minus sign corresponds to incoming and the plus sign to outgoing spherical
waves. On the other hand, near r = 0, the equation is singular, and only one of the two linearly
independent solutions to the equation belongs to a Hilbert space where the Hamiltonian is
Hermitian.2 Thus, the unique solution is a fixed linear combination of the incoming and
outgoing wave. The equation is real, so the real and imaginary parts of the wave function
are independent solutions, but both obviously obey the required regularity condition at the
origin. Thus, the solution that is part of the Hilbert space behaves at infinity like
lπ
sin(kr − 2 + δl (k))
ψ → al . (16.17)
kr
The phase shift δl (k) is defined so that it vanishes when the potential is zero. The shift by
− lπ
2 comes from the fact that for free motion an incoming plane wave exits the sphere at
an antipodal point. Spherical harmonics of odd (even) l pick up a minus (plus) sign under
antipodal reflection.
Since the space of solutions at fixed l, m is one dimensional, the incoming and outgoing
wave solutions of the Lippman–Schwinger equation coincide up to an overall constant. The
S-matrix is a one-dimensional unitary operator, with logarithm equal to the phase difference
sin(kr− lπ +δl (k))
between the incoming and outgoing parts of r
2
minus the phase difference that
would have been there for free motion. In other words, Sl (k) = e2iδl (k) . The scattering ampli-
tude f is a function only of k and θ, the polar angle measured from the incoming direction.
It is obtained by expressing the T -matrix, S = 1 + iT , in the angle “basis,” summing over
all spherical harmonics. Thus,
∞
1 X
f (θ) = (2l + 1)Pl (cos θ)[e2iδl (k) − 1]. (16.18)
2ik l=0
Referring back to the general formula for the scattering amplitude, for the case of spherically
symmetric potentials, we see that the expansion of the amplitude in powers of k corresponds
to integrating the wave function against powers of ŷ.
2
The proof of Hermiticity requires integration by parts and the surface term at r = 0 must vanish.
334  Quantum Mechanics

1 eikr eikr
Z
ψ−φ=− d3 ye−ikx̂·y U (y)ψ(y) ≡ f (x̂, k). (16.19)
4π r r
But the l-th power of ŷ contains only spherical harmonics up to order l, so the low-energy
expansion of the scattering amplitude keeps only low order terms in the angular momentum
expansion. Scattering amplitudes at low energy are dominated by the s-wave, or l = 0, phase
shift.

16.5.1 The Effective Range Approximation

Let us consider the low-energy s-wave phase shift for a potential that falls off rapidly at infin-
ity. If R(k, r) is the radial wave function, then u(k, r) ≡ rR(k, r) satisfies a one-dimensional
Schrödinger equation. Let A(k, r) be the asymptotic form of the solution A(k, r) = sin(kr+δ(k))
sin(δ(k) .
It satisfies the free Schrödinger equation. The derivative of the Wronskian of two solutions
satisfies
∂r (f (k, r)∂r f (0, r) − f (0, r)∂r f (k, r)) = k 2 f (k, r)f (0, r). (16.20)
This is true for f = u and for f = A. If we subtract the relation for u from that for A and
integrate from 0 to R, where R is much greater than the range of the potential, then three
simplifications occur. On the left-hand side, the value of the difference of Wronskians at R
is practically 0, because the exact solution u has approached its asymptotic form A since
the potential is almost zero. Secondly, at r = 0 the Wronskian of u solutions must vanish in
order for the radial Schrödinger operator to be Hermitian. This is manifestly not true for the
A functions, and it need not be, since these only represent the asymptotic large r behavior.
Indeed, the nonzero phase shift shows us that this is not the solution of the free Schrödinger
equation that is regular at the origin. Thus, the left-hand side is evaluated purely in terms
of A. On the right-hand side, we can take the upper limit of integration to infinity since
A(k, r)A(0, r) − u(k, r)u(0, r) ∼ 0 for r > R. The result is
Z ∞
2
k cot(δ(k)) − limp→0 p cot(δ(p)) = k dr [A(k, r)A(0, r) − u(k, r)u(0, r)]. (16.21)
0

We have had to be careful about the zero energy contribution because δ(0) vanishes, so the
term involving the zero energy phase shift approaches a finite limit. This was emphasized by
Fermi and Marshall and the limit is called −1/a where a is the scattering length.
The coefficient of k 2 on the right-hand side approaches a finite limit as k goes to zero.
Thus, we can write
1
ka cot(δ(k)) = −1 + k 2 ar0 ,
2
where r0 is a new parameter with the dimensions of length, called the effective range. Given
a measurement of the phase shifts at low energy, we can fit this formula with any short range
 Scattering Theory  335

potential having two parameters. In particular, a spherical well or barrier with a variable
depth and width, can fit the low-energy data for scattering from any short range spherically
symmetric potential.
The effective range approximation has been used in a wide variety of physical situations.
It was invented in early studies of nuclear physics and has recently been utilized to study
the interactions of cold atoms. In the latter context, it has been found to fail, even at
quite low energies, near narrow resonances in the scattering cross sections. Nonetheless, the
approximation has been so ubiquitous and useful that researchers often parameterize the
phase shifts in terms of an “energy dependent scattering length and effective range.”

16.6 RESONANCES
We have already included an extensive discussion of scattering theory in Chapter 8, for the
special case of the Coulomb potential. Rather than repeating that discussion, the reader
should probably to reread Chapter 8 at this point, before proceeding.
dσ
The differential scattering cross section dΩ , when multiplied by the flux of particles,
given as a number per unit area per unit time, tells us the probability that a particle will
be scattered into a solid angle Ω with respect to its incoming direction. From Chapter 8, we
recall the formulae
dσ
= |f (θ)|2 . (16.22)
dΩ
∞
1 X
f (θ) = (2l + 1)[e2iδl (k) − 1]Pl (cos θ). (16.23)
ik l=0
The total cross section σ(k) for scattering is the integral of the differential cross section over
the sphere. Using the orthogonality of the Legendre polynomials on the sphere, we can write
X
σ(k) = σl (k)(2l + 1)2 , (16.24)
l

where
σl (k) = |al |2 = |e2iδl (k) − 1|2 . (16.25)
This takes on a maximum if the phase shift goes through π/2, which is called a resonance.
Writing the amplitude as 1/(cot δl (E) − i), then near the resonance we have
(M − E)
cot(δl (E) − i) ∼ 2 , (16.26)
Γ
where M is the position of the maximum and Γ, which is called the width of the resonance,
has dimensions of energy. The amplitude near the peak can thus be written
Γ/2
al (E) = , (16.27)
E − M − iΓ/2
336  Quantum Mechanics

Figure 16.1 The Breit–Wigner cross section.

which is called the Breit–Wigner form. In mathematics, |al |2 is called the Cauchy distribution
(when normalized so that its integral is one), though physicists tend to call it the Lorentzian
distribution. It has the form shown in Figure 16.1.
Its Fourier transform, which is a function of time (when ~ is inserted) has the form
−i M Γ
t −~
e ~ e t , which suggests that a resonance should be thought of as an unstable state, whose
energy would have been M if the width were zero. If Γ  M , the Breit–Wigner–Cauchy–
Lorentz shape will produce a pronounced bump in the partial cross section σl .

16.7 A PARTIALLY WORKED EXERCISE: THE δ SHELL POTENTIAL

The δ shell potential has the form
~2 k0 ~2 k 2
V (r) = − δ(r − a) = U (r). (16.28)
2m 2m
The radial equation is
2 l(l + 1)
(−∂r2 − ∂r + − k 2 )ψl (r) = k0 δ(r − a)ψ(r). (16.29)
r r2
Defining z = kr, this is
2 l(l + 1)
(−∂z2 − ∂z + − 1)ψl (z) = b(k)δ(z − ka)ψl (ka). (16.30)
z z2
b(k) = kk0 .
For z < ka, the only allowed solution is Ajl (z), where jl is the solution of the spherical
Bessel equation regular at the origin. For z > ka, it can be a linear combination of hl (z) the
 Scattering Theory  337

iz
spherical Bessel function that behaves like i−(l+1) ez at infinity, and its complex conjugate.
This is a normalized outgoing spherical wave. Thus, we have

ψl = θ(ka − z)Ajl (z) + θ(z − ka)[hl (z) + Bh∗l (z). (16.31)

Continuity at z = ka implies that

Ajl (ka) = B[hl (ka) + h∗l (ka). (16.32)

Note that 2jl (z) = hl (z) + h∗l (z). The correct discontinuity at ka is obtained if

hl (ka) h∗l (ka)

(A − 2)(h0l (ka) + ) + (A − 2B)(h∗0
l (ka) + ) = b(hl + h∗l )/2. (16.33)
ka ka
An equivalent approach to this problem is to solve the Lippman–Schwinger equation. The
Lippman–Schwinger equation for the radial wave function for angular momentum l is
Z ∞
ψl (r) = jl (kr) + ds Glk (r, s)U (s)ψ(s)ψl (s), (16.34)
0

where Glk is Green’s function of the radial Schrödinger operator that is nonsingular when
either r or s is at the origin, and is such that the incoming radial wave at infinity is propor-
1
tional to − 2ikr [e−i(kr−lπ/2) ]. This is

Glk (r, s) = −ik[jl (kr)hl (ks)θ(s − r) + jl (ks)hl (kr)θ(r − s)], (16.35)

where hl (z) is the solution of the spherical Bessel equation which behaves like (recall that
jl (kr) is the solution that is regular at zero and that it behaves like a cosine at infinity). In
Exercise 16.6, you will verify all assertions made here in more detail.
For the case of the delta shell potential, the integral in the Lippmann Schwinger equation
collapses to a point and the equation becomes a simple algebraic consistency condition for
ψl (ka, k), whose solution (Exercise 16.6) is

jl (ka)
ψl (ka, k) = . (16.36)
1 − ikk0 a2 jl (ka)hl (ka)

ψl (k, r) = jl (kr) + ik 2 a2 ψl (ka, k) × [θ(a − r)jl (kr)hl (ka) + θ(r − a)jl (ka)hl (kr)]. (16.37)
Taking the large r limit of this, we get

ψl (k, r) → jl (kr) + ik 2 a2 jl (ka)hl (kr). (16.38)

338  Quantum Mechanics

We can then identify the phase shift as the coefficient of the outgoing spherical wave, and
obtain the partial wave scattering amplitude (Exercise 16.6)

kk0 a2 jl (ka)2
fl (k) = eiδl (k) sin(δl (k)) = . (16.39)
1 − ikk0 a2 jl (ka)hl (ka)

Recalling that bound states are given by poles of the scattering amplitude at positive imagi-
nary k, and that Bessel functions are entire functions, we find that the condition for a bound
state is
1 + |k|k0 a2 jl (i|k|a)hl (i|k|a) = 0. (16.40)

If we ask for the minimum coupling strength that will bind a given angular momentum, we
can expand this complicated equation around k = 0 because the bound state energy will be
just below zero. We find the criterion to be

k0 a = 2l + 1. (16.41)

Note that this increases with l as might be expected because of the repulsive angular momen-
tum barrier. Thus, we expect all lower angular momenta to have bound states for this value
of k0 .
For l = 0, the spherical Bessel functions are simple and the exact bound state equation
becomes
1 e−|k|
= . (16.42)
k0 a 2|k|

As k0 gets larger, the binding energy grows as well. Note that for a repulsive potential k0 < 0
there are no solutions, and by our remarks above, there will be no bound states for any l.
Now let us examine scattering states in the limit k0 a → ∞. In this limit the phase shift,
for either sign of k0 approaches arctan( hjll(ka)
(ka) ) which is the phase shift of an impenetrable
hard sphere (Exercise 16.5). On the other hand, we know that our wave functions for k0
positive are nonzero inside the sphere. As part of Exercise 16.6, you will show that this is
only true for quantized values of k such that jl (ka) = 0. Thus, in the limit of infinitely strong
attractive coupling, the scattering states decouple from a set of states bound inside the well.
As the final part of Exercise 16.6, you will investigate what happens to leading order when
k0 a is large and finite. Do this for l = 0 only. You should find zeroes of the denominator
of the scattering amplitude in the complex plane, with negative imaginary part of k. These
are resonances. Compute the scattering amplitudes when the energy is close to one of these
complex resonance poles.
 Scattering Theory  339

16.8 EXERCISES
16.1 Develop a set of diagrammatic rules (Figure 16.2) for the Born series: the diagrams
should have k vertices at each order, each one involving one factor of the potential,
followed by free propagation of the particle between vertices. Write rules, associating
algebraic/integral expressions translating each such diagram into the corresponding
contribution to the scattering amplitude.

16.2 The nonsingular solutions of the free Schrödinger equation with angular momentum
l are called spherical Bessel functions. We studied them in Chapter 8. Show that the
spherical Bessel equations for different k but the same l all have the same solution when
expressed in terms of the dimensionless variable kr.

16.3 Consider a spherically symmetric potential, V (r), which vanishes identically for r > a
and is constant inside that sphere. In the region r > a, one has to solve the spherical
Bessel equation, but since we are not near r = 0, we can include the solution yl (z),
which is singular at the origin, usually called the Neuman function. Show that the
correct linear combination of solutions is

Rl (z) = eiδl [cos(δl )jl (z) − sin(δl )yl (z)].

To solve this problem, you need to remember the definition of phase shift in terms of the
large z behavior of the wave function, and use a table or mathematical search engine
to find the large z behavior of the spherical Bessel functions. Express the function hl
that we used in the delta shell exercise, in terms of jl and yl .

16.4 Use the result of Exercise 16.3 to find an expression for the phase shift in terms of the
R0 (z)
logarithmic derivative Ll = Rll (z) of the radial wave function, evaluated at z = ka. The
expression you should find is

jl0 (ka) − Ll (ka)jl (ka)

tan(δl ) = .
yl0 (ka) − Ll (ka)yl (ka)

··· φ
V G V G V G G V

Figure 16.2 Graphical representation of the solution of the Lippman–Schwinger equation.

340  Quantum Mechanics

16.5 Argue that the logarithmic derivative of the wave function at ka is continuous, and so
the phase shift can be calculated by solving the Schrödinger equation for r < a, apply-
ing a nonsingular boundary condition at r = 0, and then calculating the logarithmic
derivative by evaluating that solution at z = ka. The argument that the logarithmic
derivative is continuous is similar to the continuity arguments we used to solve the
one-dimensional square well potential. Calculate all of the phase shifts for an infinitely
repulsive barrier inside the sphere.
16.6 Solve the worked exercise on the delta shell potential more explicitly than we did in
the text. Verify all the assertions we made there.
16.7 In the case l = 0, the spherical Bessel functions are simple. Show that in that case the
phase shifts for the spherical well satisfy
k tan(k0 a) + k0 tan(kr0 )
tan(δ0 (k)) = .
k0 + k tan(kr0 ) tan(k0 r0 )
~2 k02
Here the potential is written as V = − 2m θ(a − r).
16.8 Find the approximate s-wave scattering amplitude when ka  1 and compute the
partial wave cross section.
16.9 Note than when k0 r0 = nπ the cross section vanishes, and the well becomes invisible.
This is the Ramsauer–Townsend effect, which we explored in the exercises of Chapter 4.
16.10 The l-th partial wave cross section is defined by
4π 1
σl = (2l + 1) .
k2 1 + cot2 (δl (k))
It has a maximum when
cot(δl (k)) = 0.
If this is happening by an increase of δl through a half integer multiple of π as the
energy is varied, then we have what is called a resonance. Near the resonance
2(ER − E)
cot(δl (k)) = .
Γ(E)
The function Γ is slowly varying. Show that the partial cross section is given by
4π Γ2 (ER )
(2l + 1) .
k2 4(E − ER )2 + Γ2 (ER )
This is called the Breit–Wigner resonance formula. Draw a graph of the Breit–Wigner
function and explain why the number Γ(ER ) is called the width of the resonance.
 CHAPTER 17

The JWKB Approximation

17.1 INTRODUCTION
How does quantum mechanics give rise to classical mechanics? We have given a detailed
discussion of this in Chapter 10. The Jeffreys–Wenzel–Kramers–Brillouin (JWKB) approx-
imation gives us another angle on the problem, for systems described by Hamiltonians of
the form g 2 Pi2 + g12 V (Xi ). For such systems, the leading term of the small g 2 expansion
P

of the Schrödinger equation is the classical Hamilton–Jacobi equation for the same system.
The solutions of the classical equations determine the quantum mechanical wave functions.
Remarkably, the expansion is also valid in regions where classical motion is forbidden by
energy conservation. In those regions, one uses classical solutions in imaginary time. Wave
behavior turns into exponential damping. The JWKB approximation always breaks down in
the transition region between classically allowed and forbidden motion, but often one can
solve the Schrödinger equation exactly in the transition region.
We will discuss applications of the JWKB approximation to the derivation of the Bohr–
Sommerfeld rules for determining energies, and to the phenomenon of quantum mechanical
tunneling and the decay of metastable states.
The validity of the JWKB approximation is not enough to guarantee real classical behav-
ior. JWKB wave functions corresponding to different classical motions can have large inter-
ference. It is only for collective coordinates describing the average behavior of large complex
systems, that decoherence, and therefore real classical behavior occurs.

17.2 THE JWKB EXPANSION

Consider a classical system with action

(q̇ i )2
Z
S = g −2 dt [ − V (q)]. (17.1)
2
341
342  Quantum Mechanics

The dimensionless parameter g does not affect the equations of motion, but it will appear in
the quantum mechanical Schrödinger equation.

~2 g 2 2
i~∂t ψ = [− ∂ + g −2 V (q)]ψ. (17.2)
2 i
The Jeffreys–Wenzel–Kramers–Brillouin (JWKB) approximation [76] is a way of obtaining
a systematic small g approximation to the solution of the Schrödinger equation.
We write the solution as
− i S
ψ = e g2 ~ (17.3)
and obtain
(∂i S)2 i~
+ V − g 2 ∂i2 S.
∂t S = (17.4)
2 2
It is now clear that S has an expansion

S = S0 + g 2 S1 + · · · .

(∂i S0 )2
∂t S 0 = + V, (17.5)
2
∂t S1 = ∂i S0 ∂i S1 − i~∂i2 S0 . (17.6)
The first equation is called the Hamilton–Jacobi equation. You will recognize it if you have
had an advanced class in classical mechanics.
No matter if you have not. To solve it, note that the equation tells us how to construct
S0 (t, q i ) starting from S0 (0, q i ). Its value at any point q i at time t can be constructed from the
value of pi ≡ ∂i S at that point and a slightly earlier time, and the value of V (q i ). The value
i
of pi is equivalent to knowing the value of dqdr(r) ∂i S along any trajectory q i (r) in the space
of q i (the configuration space of the system). r is just a parameter specifying the trajectory.
We can choose it to be anything, and we will choose it to be the time coordinate. Thus, we
can write
S0 (t, q i ) = −s(t) + S0 (0, q i (t)), (17.7)
where s(0) = 0 and q i (t) is some trajectory beginning at q i (0) = q i . The equation now looks
like
˙ − 1 pi (t)pi (t) − V (q(t)),
ṡ(t) = pi q i (t) (17.8)
2
which is solved by
t
˙ − 1 pi (s)pi (s) − V (q(s))],
Z
s(t) = ds[−qi (s)pi (s) (17.9)
0 2
 The JWKB Approximation  343

where we have done an integration by parts. The left-hand side (LHS) of this equation is inde-
pendent of q i (0), so we must insist that the same is true of the right-hand side (RHS). When
differentiating the RHS w.r.t. q i (0), we encounter terms with ∂[pi (s)/∂q j (s)]∂q j (s)/∂qk (0) =
[∂ 2 S0 /∂qi ∂qj ]∂q j (s)/∂qk (0), as well as terms with only ∂q j (s)/∂qk (0). The first kind of term
depends on the initial condition S0 (0, q k ), while the second does not, so they must vanish
independently. In Exercise 17.1, you will verify that this leads to the two equations

pi = q̇ i , (17.10)

and
∂V
ṗi = − . (17.11)
∂q i
These are the classical equations of motion for the system, so the JWKB approximation is
often called the semiclassical approximation. The quantity S that appears in the logarithm
of the wave function is, in leading approximation, just the classical action.

17.3 THE JWKB APPROXIMATION IN ONE DIMENSION

In one dimension, we can work all of this out pretty explicitly. The Hamilton–Jacobi
equation is
1
∂t S0 = (∂x S0 )2 + V (x). (17.12)
2
Define S0 (t, x) = Et + S0E (t, x). Then
1
E + ∂t S0E = (∂x S0E )2 + V (x). (17.13)
2
This is solved by taking S0E to be time independent, and to be a solution of
1
E = (∂x S0E )2 + V (x). (17.14)
2
These solutions to the H–J equation are analogous to the solutions of the Schrödinger equa-
tion with initial conditions equal to one of the eigenfunctions of the Hamiltonian. The
Schrödinger equation is linear, and we can get any solution as a superposition of these special
solutions.
To understand the general solution of the H–J equation start with an arbitrary function
of x, S0 (0, x). The equation tells you how to find the value of S0 (t, x) for infinitesimally
small t, and we get the full solution by iterating this procedure from time slice to time
slice of the (t, x) plane. Given a solution S(t, x) of the H–J equation, consider a path in a
three-dimensional space
(x(u), p(u), pt (u)), (17.15)
344  Quantum Mechanics

where p(u) ≡ ∂x S0 (t, x) and pt (u) ≡ ∂t S0 (t, x). u is a parameter along the path, which we
initially take to be independent of t. The statement that S0 satisfies the H–J equation at
every point along the path is equivalent to
1
pt (u) = p(u)2 + V (x(u)) = 0, (17.16)
2
all along the path. Differentiating w.r.t. u, we get
˙
∂u pt = p(u)∂u p(u) + ∂x(u) V (x(u))x(u), (17.17)

where the dot denotes derivative w.r.t. u. This will be satisfied automatically if

∂u p = −∂x(u) V (x(u)), (17.18)

∂u x = p, (17.19)
∂u pt = 0. (17.20)
We recognize the first two of these as the equations of motion of a classical particle with
energy E = 12 p2 + V, which is conserved as a consequence of the equations of motion. The
third equation is just a statement of this conservation of energy, at fixed x. One can now
identify the path parameter u with time, but for the third equation this must be done with
care.
Going back to the original equation, we can calculate

∂t2 S0 = ∂x S0 ∂tx
2
S0 = ∂x (E(t, x)) 6= 0, (17.21)

because we are free to set S0 (0, x) equal to an arbitrary function. If we view this in terms
of the classical particle dynamics, we have a trajectory emerging from each initial point on
the t = 0 surface, but they are not all required to start with the same energy. The equation
with ∂u ∂t S is always true, but we cannot identify t and u in this equation unless we start
from initial conditions where the value of p(0) at each initial x(0) is fixed by demanding
that the energy is x independent. These solutions are the ones that correspond to the time-
independent H–J equation.

17.4 COMMENTS ON THE SOLUTIONS

Note that the wave function we have constructed depends on many different classical trajec-
tories, with different initial conditions and nothing about it picks out a particular trajectory.
Indeed, in leading order approximation, the wave function is a pure phase if S0 (0, q i ) is real
and the probability of being at any particular point in configuration space is the same. The
 The JWKB Approximation  345

first contribution to the imaginary part of the logarithm of the wave function comes, in this
case, from the first-order term S1 , as we will see in a moment. We can, if we wish, choose
S0 (0, q i ) to be complex, so that the initial wave function can be chosen peaked around some
particular position. We are, however, familiar with the uncertainty principle, which leads us
to suspect that we still cannot insist on a particular classical history. The equation

pi = ∂qi S0 , (17.22)

is particularly disturbing if we want to describe a narrowly peaked wave function. First of

all, it has an imaginary part, in the region where the imaginary part of S0 is suppressing
the probability. Secondly, a narrowly peaked e−iS0 implies large gradients of S0 so there is
no sense in which the initial value of pi behaves like a unique classical velocity for a particle
which will remain on a fixed trajectory. Therefore, the validity of the JWKB approximation
alone is insufficient to explain the emergence of classical physics from quantum mechanics.

17.5 THE JWKB APPROXIMATION FOR THE PROPAGATOR

Indeed, it is best not to think of applying the semiclassical approximation to the initial wave
function. This frees us to take the limit of singular wave functions that are concentrated at
a point,i.e.,
ψ0 (q) = δ N (q i − q i (ti )), (17.23)
and ask for the probability amplitude that at time tf , one finds the system at q i (tf ). From
the point of view of the Schrödinger equation, this amplitude, called, the propagator is a
Green’s function, which finds the influence of a delta function source at the initial time. In
operator language, it can be written
H
G(q i (tf ), q i (ti )) = hq(tf )|e−i ~ (tf −ti ) |q(ti )i. (17.24)

If there is a classical solution which gets to q(tf ) at tf starting from q(ti ) at ti , then the
leading semiclassical approximation for the propagator is
i
S[q(tf ),q(ti )]
G(q i (tf ), q i (ti )) = e g2 ~ . (17.25)

In classical physics, action is like virtue: a quantity more often talked about than seen. So it
is worth our while to compute it for the harmonic oscillator. The action is
Z tf 1
dt [mq̇ 2 − mω 2 q 2 ]. (17.26)
ti 2g 2

The complex variable z = √ 1

(mωq − ip) evolves as z(tf ) = e−iω(tf −ti ) z(ti ). We insert
2~g 2 mω
the factor of g 2 here so that the quantum version of z has the commutation relations of
346  Quantum Mechanics

creation and annihilation operators. No classical motion can change the value of z ∗ z, which
is proportional to the energy, but the boundary conditions on the propagator fix only Re z.
For any initial value of q, we can reach any final value of q at any specified time tf by choosing
the energy appropriately. We have
p(ti )
q(t) = cos[ω(t − ti )]q(ti ) + sin[ω(t − ti )] , (17.27)
m
and
p(ti )

q̇(t) = ω − sin[ω(t − ti )]q(ti ) + cos[ω(t − ti )] , (17.28)
m
p(ti )
at all times. We solve the first equation at t = tf to determine vi ≡ m . The Lagrangian is

m 2 2 2 mω 2 2
L= [ q̇ − ω q ] = [v − qi2 ] cos(2ωt) − vi qi sin(2ωt).
2g 2 2g 2 i
Doing the integral and simplifying terms using the solution for vi , we obtain the action
mω
S= [(q(ti )2 + q(tf )2 ) cos(ω∆t) − 2q(ti )q(tf )], (17.29)
2g 2 ~ sin(ω∆t)
where ∆t = tf − ti . Note that when ∆t → 0 all trace of ω disappears from the formula. We
can understand this physically, and see that it is a general result, by noting that for very
short times, one can only get between two positions with a finite separation if one has a very
high velocity. This means that the kinetic term dominates the potential term and the system
can be treated as if it were moving freely. This observation will be important later, in our
discussion of the Feynman path integral.
In summary, the JWKB approximation gives us an explicit solution of the Schrödinger
equation, written in terms of the collection of all solutions to the classical mechanics problem
which gives rise to that Schrödinger equation. The position space propagator G(q(tf ), q(ti ) =
i
hqf |e− ~ H(tf −ti ) |qi i is described, to leading order in g 2 in terms of a single classical solution.
From this, we can get the general solution of the Schrödinger equation via
Z
ψ(q, t) = dN q(0) G(q(t), q(0))ψ(q(0), 0). (17.30)

17.6 THE JWKB APPROXIMATION FOR ENERGY LEVELS

To study energy levels, we first rewrite the problem of finding energy levels in terms of
the propagator, and then use the JWKB approximation for the propagator. The resolvent
operator
1
R(z) ≡ (17.31)
z−H
 The JWKB Approximation  347

exists for all complex values of z that are not in the spectrum of the Hamiltonian H. Formally,
the trace of the resolvent is given by
ρ(E)
Z
T (z) = Tr R(z) = dE . (17.32)
z−E
RB
A dEρ(E) is the number of states in the energy interval from A to B, if the eigenstates are
discrete. ρ(E) is called the density of states. It is the function that appears in the spectral
decomposition of the Hamiltonian
Z
H= dE E ρ(E)|EihE|.

The trace formula might diverge in regions of E where there are an infinite number of states,
typically at very high energies. In that case, we insert projection operators to make it finite
and avoid those regions of integration. If the divergence is only power law, then a sufficiently
high derivative of T (z) is finite, without any such cutoff. T (z) can be constructed in terms of
this derivative, up to a polynomial ambiguity, which will not affect the considerations below.
In the application of the JWKB approximation, we will be interested in the vicinity of a
particular finite energy level, and none of these technical problems will be important.
We can see from the (Cauchy) integral formula that T (z) is analytic away from the spec-
trum of H, with poles at the discrete spectrum and cuts along the continuous spectrum. Our
goal will be to find the position of those poles, in the JWKB approximation, for Hamiltonians
P g2 p2i 1 i
of the form H = 2m + g 2 V (q ). To do this, we write

1
Z ∞ z−H
=i dt ei ~
t
. (17.33)
z−H 0

The integral converges, as long as z has a positive imaginary part.

We can now take the trace in the position basis, and use the JWKB approximation for
the propagator to write
Z ∞ Z Z ∞ Z
izt N izt
T (z) = i dt e d q G(q(t) = q, q(0) = q) =JW KB i dt e dN q eiS(q(t)=q(0)=q) .
0 0
(17.34)
The classical solutions which contribute to T (z) are all periodic solutions, with any period
t. Writing L = pi q̇ i − H, we write the answer as
R
pi (q)dq i
X
eitS (z−ES ) ei . (17.35)
periodic solutions S

The integrals over t and q have been subsumed into the sum over periodic solutions, so
the divergences we worried about above appear, if at all, in that sum, and do not affect the
348  Quantum Mechanics

contribution from any particular solution. tS and ES are the period and energy of the periodic
solution. The integral in the last exponential is taken over the closed path in configuration
space traced out by the solution.
Give such a closed path, there are an infinite number of periodic solutions, which have
the same energy, and trace out that path n times for any nonnegative integer n. The n-th
solution has period ntS . Summing over all these solutions, we get a contribution
1
TS (z) = R . (17.36)
pi (q)dq i
1 − eitS (z−ES ) ei
If Z
pi (q)dq i = 2πK, (17.37)

where K is an integer, then this contribution to T (z) has a pole at z = ES . We have derived
the Bohr–Sommerfeld quantization rule! We will work out some examples below, and you
will do more in the problem sets.

17.7 THE JWKB APPROXIMATION TO THE WAVE FUNCTIONS

OF EIGENSTATES
i g2S~
It is easy to see that the insertion of the ansatz ψ = e into the time-independent
Schrödinger equation leads to the equation
1
(∇S)2 + ig 2 ~∇2 S + V (q) = g 2 E. (17.38)
2m
Assuming E0 ∼ g12 , and expanding S in powers of g 2 , the leading order approximation to
this is what is known as the time-independent Hamilton–Jacobi equation. In simple words,
it is just the energy equation with the substitution pi → ∂i S0 .
We see immediately that ∇S0 is real or imaginary, according to whether the energy is
larger or smaller than V /g 2 . These inequalities divide the configuration space into regions
where classical motion is either allowed or forbidden. But what exactly do we mean by
forbidden? If we examine the equation
dq i
pi = ,
dt
we see that we can get imaginary momentum, if we analytically continue the time to imag-
inary values. If we do this at the boundary of the forbidden and allowed regions, where
V = g 2 E, and pi = 0, we can expect a smooth continuation of the real-time classical solu-
tion, which “bounces off the wall” and an imaginary time solution which “propagates” in the
forbidden region. The semiclassical approximation is valid in the classically forbidden region!
 The JWKB Approximation  349

Despite the smoothness of the classical solution, one can see that the JWKB approxi-
mation breaks down in the vicinity of the zero momentum wall V = g 2 E. The point is that
all of the big terms in the equation for S are going to zero on this wall, according to the
leading order solution. Thus, when we get close enough to the wall that the leading order
terms are of order g 2 , it is no longer a valid approximation to drop the order g 2 ~∇2 S term.
A better approximation in this region is to expand the potential around its value at the wall.

V (z, y i ) = E + zu(y i ). (17.39)

Here the y i are a set of n − 1 coordinates parameterizing the wall V (q i ) = E, and z is

a coordinate locally perpendicular to the wall. In general, the wall is a complicated curved
surface in the configuration space, so these are curvilinear coordinates and we have to rewrite
the gradient terms in this curvilinear system. For this reason, we will restrict our attention
to one-dimensional problems when we write explicit formulae.
Before doing that, let us explore the behavior of the wave function in the classically
forbidden region. As in the allowed region, the value of the logarithm of the wave function
at any point q in Rthe forbidden region can be obtained from the value on the V = E surface
by integrating ± 0t dsLE (q̇(s), q(s)). Here LE = m 2
2 q̇ + V (q), and q(s) is the solution of the
imaginary time equations of motion, which interpolates between a point q(0) on the V = E
surface and the point q(t) = q. We are exploring a regime where LE is positive so that the
two solutions either grow or fall off exponentially as the parameter t increases. The quantity
Z t
± dsLE (q̇(s), q(s))
0

is called the Euclidean Action of the solution.1 Physical intuition suggests that the wave
function is small in this region of configuration space. We are working in the semiclassical
approximation and classical physics tells us that a particle of energy E cannot penetrate into
this region. If the small g expansion is a good guide to the correct behavior, then it cannot
predict an exponentially large probability to be in this region. This implies that the coefficient
in front of the exponentially growing solution, must be at least exponentially small, so that
the exponentially growing term never gets larger than the exponentially falling one.
This intuition is indeed borne out when one solves the Schrödinger equation with the
correct boundary conditions. The details depend somewhat on the actual problem at hand.
One general class of problems deals with bound states in infinite space. In this case, for a
potential with a single minimum, the semiclassical bound state energies are determined by the
Bohr–Sommerfeld conditions applied to periodic real-time classical solutions. The behavior of
the bound state wave functions in the regime outside the large r turning point of the periodic
1
The terminology comes from relativistic quantum field theory (QFT), where the analytic continuation of
time turns Minkowski space into a four-dimensional Euclidean space.
350  Quantum Mechanics

solution is determined by the imaginary time solution obtained by analytic continuation of

the periodic solution. In this case, the boundary condition implied by normalizability of
the wave function implies that we must keep only the exponentially falling solution. The
matching of the two wave functions in the region near the turning point, where the JWKB
approximation breaks down, leads to a small correction to the Bohr–Sommerfeld conditions.
You will explore this in Exercises 17.2–17.5.

17.7.1 Examples
Let us use the JWKB approximation to calculate the bound state energies of the harmonic
oscillator and the hydrogen atom, for which we know the exact answers. For the oscillator,
the general classical solution is

x(t) = A cos(ω(t − t0 )). (17.40)

and
p(t) = −Amω sin(ω(t − t0 ))). (17.41)
q
2π k
The solutions are all periodic with period ω . Here ω = m and V = 21 kx2 . The energy is

p2 1
E= + kx2 , (17.42)
2m 2
so that r
1
p= 2m(E − kx2 ). (17.43)
2
The period solutions go between x = ± 2E
k and a full period traverses this interval twice.
q
2E
Defining x = k y, the action for a period is
Z
E
Z 1 q
E
S= pdx = 4 dy 1 − y 2 = 2π = 2πn~, (17.44)
ω −1 ω
where the last equality is the Bohr–Sommerfeld quantization rule. Thus, we obtain the correct
spectrum of the exact quantum oscillator except for the ~ω/2 shift of the ground state. This
is actually picked up by the first correction to the JWKB approximation, and all higher
order corrections vanish. It turns out that this is true for all Hamiltonians quadratic in the
canonical variables, a fact which we will understand when we learn the Feynman path integral
formula.
It is somewhat more surprising that the first correction to the JWKB approximation
for the hydrogen atom also gives the exact result. This is connected to the fact that the
system has a complete set of conservation laws, which are smooth functions of the canonical
 The JWKB Approximation  351

variables: the full equations of motion are equivalent to the conservation laws. Such systems
are called completely integrable. There is a vast literature on completely integrable systems.
An elementary introduction to the topic can be found in [18].
The conserved energy for the hydrogen atom is given by
(p)2 a
E= − . (17.45)
2m r
2
e
Here a = 4π 0
. Angular momentum conservation implies that the motion takes place in a
plane, and for a bound orbit, conservation of the Eccentricity vector tells us that the orbit
is an ellipse. The equation for an ellipse is
R2
r2 = . (17.46)
1 + κ sin2 (φ)
Here r is the distance from one of the foci and φ is the angle with that focus as center. The
action integral we have to compute in order to evaluate the Bohr–Sommerfeld equation is
s
Z Z r+ l2 a
S= pr dr + pφ dφ = 2πl + 2 dr 2m(E − 2
+ ). (17.47)
r− 2mr r
We have used the fact that the angular momentum pφ = l is a conserved constant. The radial
integral runs between the two zeros of the integrand and the factor of 2 is there because a full
cycle covers this radial path twice. Equating this to 2πk~, we get the quantization condition.
For a = 0 (circular orbits), where r+ = r− , this reduces to Bohr’s condition that the angular
momentum is an integer multiple of ~. You will do the integral in the exercises and verify
that the quantization condition gives
1
E=− Rydberg. (17.48)
(k + l)2
The derivation of the Bohr–Sommerfeld condition from the quantum theory in the JWKB
approximation allows us to conclude that l is an integer even for noncircular orbits, because
the wave function is periodic in φ. This formula looks exact, and we can think of k = jmax +1,
where jmax is the highest power in the relevant Laguerre polynomial, but there is some illegal
trickery being pulled in making that statement. In the exact formula, l is the integer defined
by the square of the total quantum angular momentum operator, l(l + 1)~2 . In the Bohr–
Sommerfeld rule, l~ is the angular momentum in the plane of the orbit, which one would
want to identify with the quantum number m. For the classical orbit, these are the same
thing, and both are quantized as integers, but they are not the same. These problems with
the leading order result are resolved by the first correction to the JWKB approximation, but
we do not have space to go into that here.
352  Quantum Mechanics

For general one-dimensional potentials, one can write the Bohr–Sommerfeld condition as
Z x+ q
dx 2m(E − V ) = πk~, (17.49)
x−

where the integral is taken between two consecutive turning points of the classical motion.
The factor of 2 difference between this formula and our previous statement of the Bohr–
Sommerfeld condition comes from the fact that the classical periodic solution goes back to
x− after being reflected at x+ . Notice that this condition does not refer to the region of space
beyond the turning points, and so cannot distinguish between two potentials with different
behavior outside the turning points. In particular, the classical motion may be in a local but
not global minimum of the potential, but the B–S condition only recognizes that when k gets
large enough that there is no longer a periodic motion in that local well. In the quantum
theory, such states are not true eigenstates of the Hamiltonian, but rather are long-lived,
metastable states. In the next section, we will use the JWKB approximation to compute the
decay rate of such a state.

17.8 THE DECAY OF METASTABLE STATES

Another general class of problems, to which the JWKB approximation is often applied, has
a barrier interpolating between two classically allowed regimes. One is interested in the fate
of a wave function initially concentrated in a well near what we can choose as the origin
of coordinates. The potential either goes to a constant at infinity, or has other wells which
are deeper and/or wider. It is then very often the case that a normalized wave function
concentrated in the well near the origin is an unstable situation. The probability density
propagates outward, ending up concentrated near a different well, or flowing out to infinity.
In the extreme semiclassical approximation g  1, the fraction of probability concentrated
in the original well is either exponentially small, or vanishes, as time goes to infinity.
On the other hand, let us suppose that there is a semiclassical bound state near the origin.
That is, a periodic solution of energy E, which stays trapped near the origin, even though
there are other regions of configuration space, separated from the origin by a finite barrier
which it could in principle explore. Then, in the limit g  1 we expect to find a metastable
state, with a life-time that goes to infinity as g → 0. The semiclassical approximation allows
us to compute the lifetime of the state, as a systematic expansion around g = 0. We call this
the decay of the metastable state, and our aim is to calculate the decay rate of this state;
the probability per unit time that it will decay. The mechanism of decay, in the semiclas-
sical approximation, is called quantum mechanical tunneling through a barrier, or quantum
tunneling for short.
To get a feel for the intuition behind this nomenclature, it is best to watch the following
video.
 The JWKB Approximation  353

www.youtube.com/watch?v=cV2fkDscwvY
It shows the modulus of the wave function, for a particular solution of the time-dependent
Schrödinger equation, hitting a barrier higher than the energy of the incoming particle. One
can see how the wave function “tunnels through the barrier.”
It is important to understand that there is no violation of energy conservation in tun-
neling processes. In the decay of a metastable state, the initial wave function, related to the
periodic classical solution whose energy is below the barrier height, is not an eigenstate of
the Hamiltonian. The true eigenstate of the Hamiltonian at that value of the energy is either
part of a scattering continuum (if the potential simply asymptotes to a constant on the other
side of the barrier), or is concentrated in some other well of the potential.
We have seen that in the JWKB approximation, the behavior of the wave function in
the tunneling region (the region “under the barrier”) is determined by the solution of the
imaginary time classical equations of motion. The surface V = E in configuration space
has two disconnected components, which are distinguished by their distance from the origin.
In the simplest case, they have the topology2 of spheres in configuration space. The imaginary
time solution, which interpolates between two real-time classical motions on either side of
the barrier, has q̇ i = 0 when it hits either of the surfaces V = E. If it did not, we could not
match the imaginary derivative under the barrier to the real derivative inside it. There is a
solution for each choice of a pair of points on the two disconnected components V = E. Each
of these solutions has a different Euclidean action. Since all the Euclidean actions are ∝ g12 ,
all these contributions are exponentially suppressed compared to that of minimal action.
Therefore, to leading exponential order as g → 0, only the minimal action path contributes
to the tunneling amplitude. This most probable escape path [77] has been called an instanton3
following the work of [78].
i i
We can avoid the search over all pairs of points (qinner , qouter ) on the V = E surface by
noting that since q̇ = 0 on this surface, we can find another solution of the imaginary time
i i i
equations which starts at qinner bounces off qouter and returns to qinner , by simply retracing
the original solution. So all we have to do is find the minimal action bounce solution, which
i i
starts and ends at qinner and then minimize over the choice of qinner . The action of this
bounce solution is twice that of the original, so the square of the wave function, which gives
the tunneling probability, is

Ptunneling ∼ e−Sbounce , (17.50)

2
But not generally the geometry, unless the potential is spherically symmetric in the full configuration
space.
3
Again, the terminology comes from QFT, where the relevant solutions are localized in space as well as
time. Static classical solutions, which are localized in space, can be interpreted as heavy particles and were
dubbed solitons. ’t Hooft therefore invented the term instanton to refer to an imaginary time solution
localized in all dimensions.
354  Quantum Mechanics

for the minimal action bounce. Things become even simpler, if we are talking about the
metastable ground state, which corresponds to the static classical solution localized at the
i
origin. In this case, the inner V = E surface collapses to a point. Since qinner = 0, we just
have to find a single bounce solution. For the decay of states other than the ground state,
we have to find the bounce for all values of q inner and find the minimum action.

17.9 MORE EXAMPLES

A very general situation in which JWKB formulae are useful is the determination of the
asymptotic behavior of the wave function of a one-dimensional system with a confining
potential. This is a potential that grows at infinity, so that the system only has bound states.
The equation for the logarithm of the wave function (ψ = e−S ) of an eigenstate of energy E is
" #
~2 dS d2 S
− ( )2 − 2 = E − V. (17.51)
2m dx dx

In the asymptotic region V  E. We make the ansatz that the second derivative of S is
negligible and find r
dS 2m
=± V. (17.52)
dx ~2
The second derivative is then smaller than the first. The ratio to the first derivative term is
V 0 /V 3/2 . Unless V has rapid and growing oscillations (e.g., V ∼ xa sin(xb+1 ) with b > 3a/2),
this term is indeed negligible. In all cases of practical interest then, the JWKB approximation
is valid
R√in the asymptotic regime. It gives rise to two linearly independent solutions, behaving
± V
as e . Only the falling solution is normalizable.
Another very common situation is the decay of a metastable minimum of the potential.
Generically, the minimum will be a quadratic potential, centered around a point that we can
take as the origin of coordinates. The Euclidean (imaginary time) equations of motion are

ẍ = V 0 (x). (17.53)

These are the ordinary classical equations of motion in the upside down potential −V (x). The
nature of the solutions is completely determined by “energy” conservation in this upside down
potential. The metastable minimum at x = 0 is a maximum of the upside down potential.
The bounce solution for the metastable ground state must begin at this maximum at time
tE = −∞, and return to it at tE = ∞. This implies that the velocity at this point is zero
(i.e., asymptotes to zero as tE → ±∞), so the Euclidean “energy” of the solution is equal to
V (0). The conservation law implies that the solution passes through the minimum of −V (the
maximum separating the metastable minimum from the true minimum of the potential) with
 The JWKB Approximation  355

nonzero velocity, and climbs the opposite side until it reaches a point where V (xm ) = V (0)
(see Figure 17.1).
We then have dx/dtE (xm ) = 0, by “energy” conservation, so the solution can turn around
and return to the origin. An additional consequence of the vanishing velocity at xm is that
the Euclidean solution can be analytically continued into a real-time, real-valued solution
at this point. That solution gives the JWKB phase of the wave function in the classically
allowed region.
The bounce solution satisfies
1 2
ẋ − V (x) = −V (0), (17.54)
2m
so that the Euclidean action is
Z ∞ 1 2
Z ∞
S= dtE [ ẋ + V (x)] = dtE [2(V − V (0))]. (17.55)
−∞ 2m −∞
R
We have dropped a term V (0) which would have been present also for the classical solution
where x(t) = 0. This has to do with the proper normalization of the wave function. The
integral over the constant term diverges and would lead to a vanishing wave function. It
must be subtracted out. The general rule, when computing decay amplitudes in the JWKB
approximation is to compute the difference between the Euclidean actions of the bounce
solution, and the static solution sitting at the metastable minimum.

V(f)

A C
B

Figure 17.1 The upside down potential for an instanton calculation.

356  Quantum Mechanics

1 2
The equation 2m ẋ − V (x) = −V (0) allows us to convert the tE integral to an integral
over x. The presence of a point where ẋ = 0 means that the tE integral covers the x-axis
twice. This is a consequence of computing the action of the bounce solution. We end up with
a formula for the decay probability
2
R xm √
Pdecay = e− ~ 0 2m(V (0)−V (x))dx
. (17.56)

In principle, the interpretation of this quantity as a decay probability is valid for cases in
which the potential to the right of xm is such that the system has no discrete eigenstate
localized at finite x. In that case, a Gaussian wave function, initially localized near the
origin, eventually has a probability distribution that flows out to infinity. Pdecay gives the
approximate probability per unit time that the Gaussian will decay to a state-localized near
infinity.
If, on the other hand, the potential has a discrete eigenstate whose eigenfunction is
localized at some x0 > xm , then there is a finite probability that a particle in that eigenstate
will actually be found at the origin. In this case, Pdecay corresponds roughly to that probability.
An interpretation, which is always valid, is that Pdecay is the probability per unit time that a
Gaussian wave function, initially localized near the origin, with a width given by the harmonic
oscillator ground state of the potential 21 V 00 (0)x2 , will be found to the right of xm .

17.10 THE JWKB APPROXIMATION FOR PHASE SHIFTS

For large angular momentum, the effective potential
~2 l(l + 1)
U (r) = V (r) + , (17.57)
2mr2
in the radial Schrödinger equation for ul (r) = Rl (r)/r, satisfies the criteria for validity of the
JWKB approximation. Furthermore, in the expression
∞
1 X
f (θ) = (2l + 1)Pl (cos θ)[e2iδl (k) − 1], (17.58)
2ik l=0

we can substitute the large l expansion of the Legendre polynomials

2i
Pl (cos θ) → p cos((l + 1/2)θ + π/4), (17.59)
2πl sin(θ)
to write the large l part of the sum as
∞
1
[ei(2δl −(l+1/2)θ−π/4) − ei(2δl +(l+1/2)θ+π/4) ],
X
f (θ) = p (17.60)
k 2π sin(θ) l=l∗
 The JWKB Approximation  357

where l∗ is large enough for the expansion of the Legendre polynomial to be accurate. In the
JWKB approximation, the phases of the wave function are large, so all the terms in this sum
are rapidly oscillating and the sum is dominated by the stationary phase point
dδl
2 = ±θ. (17.61)
dl
The JWKB approximation to the wave function has the phase
s
1
Z r ~2 l(l + 1)
φJW KB = π/4 + dy ~2 k 2 − 2mV (y) − . (17.62)
~ r∗ y2
The JWKB approximation to the phase shift is the limit as r → ∞ of the difference between
φJW KB and the phase kr − lπ/2 of the free wave function. For potentials satisfying the
postulates of scattering theory, the difference is finite:
s
1
Z ∞ ~2 l(l + 1) 1
δlJW KB = dy [ ~2 k 2 − 2mV (y) − 2
− k] + π(l + 1/2) − kr ∗ . (17.63)
~ r∗ y 2
r∗, the turning point, depends on l, but that dependence cancels against the last term in the
formula. Thus, the stationary phase equation becomes (to leading order in l, with L ≡ ~l
the “classical angular momentum”)
Z ∞ dy 1
L p = (π ± θ). (17.64)
r∗ y 2 2m(E − V ) − L2 /y 2 2

We have written ~2 k 2 = 2mE so that all terms in this formula have a classical interpreta-
tion. This equation is the same as that we could have obtained in classical mechanics, by
computing the scattering angle in terms of the angular momentum. These two variables are
quantum mechanically complementary, but the stationary phase approximation allows them
to have fixed values simultaneously. The stationary phase equation says that δl ∼ lθ and our
derivation assumed that δl was large, so the criterion for the validity of this approximation
is that lθ be large.

17.11 EXERCISES
17.1 Show that the condition that
t
˙ − 1 pi (s)pi (s) − V (q(s))]
Z
ds[−qi (s)pi (s)
0 2
is independent of the initial value q i (0) leads to the classical Hamilton equations for
the system.
358  Quantum Mechanics

17.2 The JWKB approximation for eigenfunctions breaks down when one is near a turning
point of the classical motion, a place where V (x) = E. Near a generic turning point we
have V (x) − E ∼ ax. Show that in this region you can solve the Schrödinger equation,
which is called the Airy equation, exactly. Show that the Fourier transform turns the
equation into a first order equation. This solution is called the Airy function Ai(x).
Show that the second solution has the form Bi(x) = b(x)Ai(x), where

b00 Ai0
= −2 .
b0 Ai

17.3 Find the behavior of the two solutions to the Airy equation when x → ±∞. In this
limit, you can do the Fourier transform integral by stationary phase approximation.

17.4 Consider a periodic solution of the classical equations and the eigenfunction of the
Schrödinger equation for each of the Bohr–Sommerfeld levels. For simplicity, consider a
potential symmetric around x = 0, with bound states in the region near the origin. The
bound state wave functions are either even or odd, so we can concentrate on turning
points of the classical motion to the right of the origin. To the right of the turning
point, the solution must fall off exponentially. Show that only one linear combination
of solutions to the Airy equation has this property. Match that solution to the JWKB
wave function to the left of the turning point and show that the match can only be
accomplished by a shift in the energy level, compared to that computed by Bohr–
Sommerfeld.

17.5 Consider the Schrödinger equation in the complex plane and show that the JWKB
approximation can be used to connect directly between the two sides of a turning
point without ever solving the Airy equation. Show how to obtain the energy level shift
described in the previous exercise.

17.6 Finish the computation of the Bohr–Sommerfeld quantization condition for the hydro-
gen atom.

17.7 In this exercise, we will use units in which ~ = c = 1. Mass, energy, and momentum
then have the same dimensions. The natural energy scale of the strong interactions is
∼ 100–200 MeV. The modern theory of the strong interactions is based on a relativistic
QFT known as QCD (short for quantum chromodynamics). The theory contains a
variety of particles known as quarks, which have different masses. They interact via
forces generated by other fields, called gluons. If a very heavy quark–antiquark pair is
separated by a large distance, the theory is supposed to generate a confining potential,
eV )2
which varies linearly, V = kr, with k ∼ (100M ~c . At short distances compared to
 The JWKB Approximation  359

~c
100 MeV the potential is supposed to be approximately Coulomb-like V αc ~c r . αc is a
dimensionless constant.
Low lying bound states of quarks and antiquarks that are much heavier than this
scale should be treatable in a nonrelativistic approximation, because the kinetic energy
and the binding energy are of the same order of magnitude, and much less than the
mass. There are three types of quarks, charm, bottom and top, to which this analysis
might apply. However, top quarks decay (via the weak interactions) into bottom quarks
too rapidly to form bound states. Assuming the nonrelativistic approximation, one can
try to compute the spectrum of quark antiquark bound states as a function of the quark
mass, and the parameters αc and k. The JWKB approximation gives a handy analytic
tool for making a first estimate of the bound state spectrum. We will make an even
more drastic approximation, and replace the Coulomb term by a constant V0 , of order
100 MeV. Write the Bohr–Sommerfeld quantization rule for zero angular momentum
in this potential. Show that knowledge of the ground state, the first radial excitation,
and the quark mass enable you to calculate the masses of higher excitations. The mass
of a bound state is twice the quark mass plus the (negative) binding energy. For the
charmed quark system, the quark mass is 1.5 Gev c2 and the rest energies of the ground
state and first radial excitation (these are called the ψ(1S) and ψ(2S) particles) are
3.1 and 3.7 GeV. The bottom quark rest energy is 4.18 GeV and the υ(1S) and υ(2S)
particles have rest energies 9.46 and 10.02 GeV. Give the JWKB estimate for the energy
of the next radial excitation in each case. Compare to the values you can find in the
Particle Data Group compilation [79].

17.8 If a one-dimensional potential rises at infinity (e.g., like a power |x|p ), then there are
an infinite number of solutions of the Bohr–Sommerfeld condition
Z q
dx 2m(E − V (x)) = 2πn~. (17.65)

Show that as n → ∞, these JWKB energies are closer and closer to the exact eigenval-
ues.

17.9 Model the strong and electromagnetic forces between an alpha particle (Helium nucleus)
and a nucleus by
2Ze2
V (r) = θ(r − a) . (17.66)
4π0 r
a is of order 10 times the pion Compton wavelength a ∼ 10−12 cm, or 14 1
inverse MeV in
natural units. This is the potential for an l = 0 or s-wave bound state. For more general
l there is an additional repulsive centrifugal potential. The idea behind this potential
is that strong attractive nuclear forces dominate the Coulomb potential for r < a and
360  Quantum Mechanics

that they are of such short range that we can just model them by a constant potential.
Argue that this problem has no bound states.

17.10 Consider the scattering states in the potential above. For small enough positive energy,
the well near the origin in the potential would create a metastable resonance in which
the particle was trapped for some time in the region r < a. Classically this resonance
would be stable, but there is a quantum tunneling amplitude for the particle to escape
to infinity for positive energy, and you argued in the previous exercise that there were no
true negative energy-bound states. Calculate the tunneling probability in this potential
as a function of the energy E.

17.11 Calculate S2 , the third term in the JWKB expansion of the logarithm of the wave
function of an energy eigenstate.

17.12 Show that the formula

1
Z π p
Pl (x) = dφ (x + i 1 − x2 cos φ)l
π 0

satisfies the Legendre equation and that it is square integrable on the interval −1 ≤
x ≤ 1. Argue that this means it is a representation of the Legendre polynomials.

17.13 Use the formula of Exercise 17.12 to derive the large l limit of the Legendre polynomials,
which we used in our semiclassical analysis of phase shifts.
 CHAPTER 18

The Variational Principle

18.1 INTRODUCTION
We have already used the variational principle in Chapters 11 and 13, but here we will give a
more careful exposition and describe some more general examples. The basic idea is simple:
the expectation value of the Hamiltonian in any normalized state is given by
X X
|ck |2 Ek ≥ E0 |ck |2 = E0 , (18.1)

where E0 is the ground state energy. Thus, if we start from any state, and make changes
that reduce the expectation value of the energy, we get a better approximation for the
ground state. The art in employing the variational principle lies in using one’s intuition to
find an ansatz for the ground state wave function, which captures important features of the
physics of the system, while giving rise to calculations that are relatively simple. A lot of
the words that are used to describe complicated physical systems are actually derived from
clever variational approximations. The most famous example is the concept of single electron
orbitals in complicated systems with interacting electrons.
In this chapter, we expose some general properties of the variational method and do a
number of examples. The variety of variational approximations in the literature illustrates
the creativity of physicists, and cannot be captured in a single book.

18.2 GENERAL PROPERTIES

We will begin by making some general remarks about the variational principle, and then give
some examples.
Given a Hamiltonian H, we can get a variational bound on its ground state energy by
choosing any exactly soluble Hamiltonian H0 and noting that

G(H) ≤ hψ0 |H|ψ0 i = G(H0 ) + hψ0 |(H − H0 )|ψ0 i. (18.2)

361
362  Quantum Mechanics

We have introduced, for this chapter only, the notation G(H) to denote the ground state
energy of a Hamiltonian. The second form of the variational estimate suggests the possibility
of finding corrections to the estimate by treating H − H0 as a small perturbation of H0 . This
can sometimes be useful, even though there is no apparent small parameter in the problem.
This consideration can be generalized to statistical mechanics. Write the partition func-
tion as
Tr e−β(H−H0 ) e−βH0
Z ≡ Tr e−βH = Tr e−βH0 . (18.3)
Tr e−βH0
Now it is a very general fact about functions that heX i ≥ ehXi for any probability distribu-
tion.1 To prove this, consider the expectation value of esX as a function of s. We have
!2
sX hX 2 esX i hXesX i (h(X − a)2 esX i
∂s2 ln he i= − = , (18.4)
hesX hesX i hesX i
sX
where a = hXe i
hesX i . That is, the log of the expectation value is a convex function of x. A
convex function is always larger than the linear approximation to that function around any
point, so the inequality is true. Since the free energy is defined by Z = e−βF , we have
F ≤ F0 + h(H − H0 )i, (18.5)
where the expectation value is taken in the thermal density matrix Z0−1 e−βH0 .

18.3 EXAMPLES
With these general ideas behind us, let us try some particular problems. As a first example,
let us consider a one-dimensional problem with a monomial potential V = g 2 x2q . Work in
~2
units where 2m = 1. Let us try a Gaussian ansatz for the ground state wave function
a 2 a
ψ0 = e− 2 x √ . (18.6)
π
This satisfies
−∇2 ψ0 = 2a2 − 4a4 x2 ψ0 , (18.7)
so the expectation value of the Hamiltonian is
hψ0 | − ∇2 + g 2 x2q |ψ0 i = 2a2 − 4a4 hx2 i + g 2 hx2q i. (18.8)
The integral
√
2q − 1 2 π
r
d d π 1
Z Z
2q −a2 x2 −bx2
x e = (− )q [ e ]b=a2 = (− )q 2 = ... . (18.9)
db db b 2 2 a2q+1
1
The notation hAi just means the expectation value of A, not necessarily a QM expectation value.
 The Variational Principle  363

Thus,
1 2q + 1 −2q
hx2q i = ... a (18.10)
2 2
and
1 1 2q + 1 −2q
hHi = a2 + g 2 . . . a . (18.11)
2 2 2
This is minimized at
1 g2 1 2q + 1
= q 2(q+1) . . . (18.12)
2 a0 2 2
2
and the bound on the ground state energy is of order a20 ∼ g q+1 .
1
By scaling x = g q+1 y, we can see that the variational estimate has exactly the same
dependence on g as the exact answer (Exercise 18.1). There are a number of other general
remarks about variational estimates, which are illustrated nicely by this example. The quan-
tum ground state energy of a bound state of some collection of particles is determined by a
competition between the tendency of the potential to collect all of the particles in one place,
and the kinetic energy which the confined particles have by virtue of the uncertainty princi-
ple. A Gaussian ansatz for the bound state wave function incorporates both of these features,
with the width of the Gaussian determined by the relative balance between potential and
uncertainty principle kinetic energy. In our example, we see that as g is decreased, the width
gets larger and the energy lower. Similarly, if we increase q for fixed g, the width gets larger
because the potential gets flatter and flatter for x < 1 and then rises abruptly when x > 1.
Note that the shape of the Gaussian wave function at infinity is completely wrong. Indeed,
as x → ∞ we are in a regime where the JWKB approximation of the previous section is
valid and the logarithm of the exact wave function behaves like
ln ψ0 ∼ gxq+1 . (18.13)
Thus, although the variational ansatz gives a good approximation to the ground state energy,
it does not give accurate answers for questions about the probability of finding a particle
at large distances from the origin, given that the system is in its ground state. The reason
for this discrepancy in accuracy is that in the ground state, the distant regions of x do not
give much of a contribution to the integral that computes the total energy. More generally,
it is true that variational estimates give us a much better approximation to the ground state
energy than they do to the ground state wave function.
A second example to which we can apply the variational technique is the ground state of
the helium atom. In the Bohr–Rydberg units, which we used in our discussion of atomic and
molecular physics, the Hamiltonian is

p21 + p22 2 2 1
H= − − + . (18.14)
2 r1 r2 |r1 − r2 |
364  Quantum Mechanics

We have neglected the differences between the electron mass and the actual reduced mass
of the electrons. There are two simple variational estimates of the ground state energy. The
first uses our general observation above relating the ground state energies of two different
Hamiltonians
G(H) ≤ G(H0 ) + h(H − H0 )i, (18.15)
where the expectation value is taken in the ground state of H0 . For helium, we take
p21 + p22 2 2
H0 = − − . (18.16)
2 r1 r2
The ground state energy of H0 is −8 Rydberg, since H0 is just two copies of the Hamiltonian
of the doubly charged hydrogenic ion. The ground state wave function is the spin singlet,
multiplied by the product of two hydrogen ground states with aB → aB /2:
8
ψs1 /s2 (r1 , r2 ) = s1 /s2 √ e−2(r1 +r2 ) . (18.17)
2π
The expectation value of H − H0 is
32
Z ∞ 1
h(H − H0 )i = 4π 2 dr1 dr2 (r1 r2 )2 e−4(r1 +r2 ) dφdθ sin(θ) q .
π 0 r12 + r22 − 2r1 r2 cos(θ)
(18.18)
The first factor of 4π comes from the overall angular integration. θ is the angle between the
vectors r1 and r1 . To do the angular integration, it is convenient to write
1 1
Z ∞ 2 2
q =√ dss−1/2 e−s[r1 +r2 )−2r1 r2 cos(θ)] , (18.19)
r12 + r22 − 2r1 r2 cos(θ) π 0

where we have used the by now familiar identity

Z
√
Γ(1/2) = dss−1/2 e−s = π.

Now we use the change of variables

Z π Z 1
sin(θ)dθ f (cos(θ)) = dx f (x)
0 −1

to perform the angular integral. The result is

256
Z ∞ Z ∞ 1 2 2
h(H − H0 )i = √ dr1 dr2 (r1 r2 )2 ds s−1/2 e−4(r1 +r2 ) (e−s(r1 −r2 ) + e−s(r1 +r2 ) ).
π 0 0 2sr1 r2
(18.20)
 The Variational Principle  365

The remaining integrals are all elementary, or reduce to Euler Gamma functions. The end
result of the computation is that

E0helium ≤ −75eV. (18.21)

The experimental result is −78.975 eV. Note that the original approximation, which neglected
the electron repulsion, was not a variational approximation, and so undershot the right
answer. In fact, we can prove that it undershot, because it is obtained by neglecting a positive
term in the Hamiltonian. Thus, these simple calculations show that the helium ground state is
between −109 eV (8 Rydbergs) and −75 eV. More elaborate calculations have been done, with
up to 18 variational parameters and they reproduce the correct answer within experimental
error. The simplest improvement is to use a variational ansatz for the wave function that is
the product of two single electron ion wave functions but with the “effective nuclear charge,”
Z treated as a variational parameter. In Exercise 18.2, you will show that this puts an upper
bound on the ground state energy of helium of −77.5 eV.

18.4 THE HARTREE AND HARTREE–FOCK APPROXIMATIONS

Let us begin by recalling the quantum field theoretic (also called second quantized) treatment
of multiparticle states of bosons and fermions. We start with a single particle Hilbert space
with some orthonormal basis |ii, and introduce a set of operators a†i , one for each element of
the basis. These satisfy
[ai , a†j ]± = δij , (18.22)

[ai , aj ]± = 0. (18.23)
The plus sign is for fermions and the minus sign for bosons. The Hermitian conjugate of the
second equation is also valid, and tells us that the states

1 †
|i1 . . . in i = ai1 . . . a†in |0i, (18.24)
N
where
ai |0i = 0, (18.25)
for all i, are a basis for the symmetrized (bosons) or antisymmetrized (fermions) n-fold
tensor product of the single particle Hilbert space with itself. The state |0i is called the
no-particle state or vacuum state. These tensor product Hilbert spaces are the Hilbert spaces
of n independent bosons or √ fermions. The normalization factor N differs from 1 only for
bosons. It is the product of nk !, where nk is the number of bosons in the state |ki of the
single particle Hilbert space.
366  Quantum Mechanics

The most general Hamiltonian describing noninteracting particles has the form

H0 = a†i hij aj , (18.26)

where we use the summation convention. These are called one body operators because they
are a sum of terms, each of which acts on only a single particle. The Hilbert space on which
the creation and annihilation operators act, contains any number of particles. An operator
which does not change the total number of particles will be a sum of terms, each of which is
a monomial containing equal numbers of creation and annihilation operators. We can always
write operators in normal-ordered form, with all the creation operators to the left of all the
annihilation operators.
Operators of the form
a†i a†j Vij,kl ak al (18.27)
are called two body operators. Written in terms of the states |i1 . . . in i in the tensor product
Hilbert space, they have matrix elements which act on only two particles at a time. Many
model Hamiltonians for interacting particles, like the Coulomb interaction, can be written
in terms of such two body operators. The Hartree and Hartree Fock approximations, which
we are about to introduce, can be applied to Hamiltonians that contain operators affecting
k particles at a time (k annihilation and creation operators), but are simplest for simple two
body interactions.
For a one body Hamiltonian H0 , the ground state is simple. For bosons it takes the form
1
|Ψ0 i = √ (a†i φi1 )n |0i, (18.28)
n!
where
hij φj1 = 1 φi1 . (18.29)
ψ1i |ii
P
In other words, the state is the ground state of the single particle Hamiltonian whose
matrix is hij . The ground state energy of the n particle system is just n1 .
For fermions, because of the Pauli principle, which is implemented by the anticommuta-
tion relations between the creation operators, we cannot put all of the particles in the same
single particle eigenstate, but instead must choose the n lowest states of hij and put one
fermion in each of them. The multifermion ground state is called the filled Fermi sea and has
the form X † X † i
|Ψ0 i = ain φinn . . . ai1 φ11 |0i. (18.30)
i
The wave function (coefficients) of this state in the tensor product basis |i1 . . . in i is det [φij ]
and such states are called Slater determinants.
Note that in the boson ground state, the individual single boson states are unentangled,
while in the fermion ground state there is a very simple pattern of entanglement, total
 The Variational Principle  367

antisymmetrization, required by Fermi Dirac statistics. The basic idea of the Hartree and
Hartree–Fock approximations is to use multiparticle states with these simplest patterns of
multiparticle entanglement, as trial variational states for more complicated Hamiltonians,
which continue k-body operators with k ≥ 2. Equivalently, given a Hamiltonian like

H = a†i kij aj + a†i a†j Vij,kl ak al , (18.31)

we search for the one body Hamiltonian H0 whose ground state gives the lowest expectation
value of H.
For bosons, the expectation value is simply computed. Annihilation operators operating
to the right, and creation operators operating to the left on |Ψ0 i vanish unless they have a
projection on the single particle ground state φi1 :
1 n
aj √ (a†i φi1 )n |0i = √ (a†i φi1 )n−1 φj1 |0i. (18.32)
n! n!
1 n(n − 1) † i n−2 j k
aj ak √ (a†i φi1 )n |0i = √ (ai φ1 ) φ1 φ1 |0i. (18.33)
n! n!
This gives
hΨ0 |H|Ψ0 i = nφ∗1 i kij φj1 + n(n − 1)Vij,kl φ∗1 i φ∗1 j φk1 φl1 . (18.34)
We are instructed to minimize this w.r.t. the coefficients φi1 , subject to the constraint
φ∗1 i φi1 = 1. This leads to a nonlinear equation for the coefficients. It is the general form
of the Gross–Pitaevski equation for dilute Bose gases, which we discussed in Chapter 12.
These are the Hartree equations.
To evaluate the expectation value of the Hamiltonian in a ground state of independent
fermions, we note that such a state is determined by an n-dimensional subspace of the single
particle Hilbert space, defined by the first n levels of the Hamiltonian H0 . The fermion ground
state is invariant under unitary transformations in the single particle space which leave this
subspace invariant. Under such a transformation, the creation and annihilation operators go
into
ai → Uij aj , (18.35)
a†i → Uij∗ a†j . (18.36)
We want to evaluate
hΨ0 |a†i aj |Ψ0 i, (18.37)
and
hΨ0 |a†i a†j ak al |Ψ0 i. (18.38)
These expectation values are collections of pure numbers. The second set of numbers is
antisymmetric under interchange of either of the pairs i, j or k, l. They also have to vanish
368  Quantum Mechanics

whenever, considered as a vector in any of the indices i, j, k, l, that vector is orthogonal

to the special subspace. The only collections of numbers invariant under all the unitary
transformations and having these orthogonality and symmetry properties are

hΨ0 |a†i aj |Ψ0 i = APij , (18.39)

and
hΨ0 |a†i a†j ak al |Ψ0 i = B(Pik Pjl − Pil Pjk ). (18.40)
A, B are numbers and
φ∗n i φjn
X
Pij = (18.41)
n
is the projection matrix on the single particle subspace corresponding to the ground state.
We can determine the coefficients A, B by noting that if we contract the first equation with
δij and the second with δjk we get (using δij Pij = n and the fact that P is a projection)

hΨ0 |N |Ψ0 i = An, (18.42)

and
hΨ0 |a†i N al |Ψ0 i = B(1 − n)Pil . (18.43)
In these equations, N is the number operator, which simply counts how many particles are
in the state. The first equation thus gives An = n while the second implies that

(n − 1)A = (1 − n)B,

so that
A = −B = 1.
Let us write these equations explicitly for the case of electrons interacting with nuclei in the
Born–Oppenheimer approximation. The index i labeling the single particle Hilbert space is
replaced by the electron spin index a and its position x. The Hamiltonian in Bohr–Rydberg
units is
∇2 1
Z Z
H= 3
d x [ψa† (x)[− + V (x)]ψa (x) + d3 x d3 y ρ(x) ρ(y). (18.44)
2 2|x − y|
The density operator is defined by

ρ(x) = ψa† (x)ψa (x). (18.45)

This Hamiltonian is not in normal ordered form, but we can put it in that form by moving
ψa (x) to the right of ψa† (y) using the relation

[ψa (x), ψb† (y)]+ = δab δ 3 (x − y). (18.46)

 The Variational Principle  369

The singular term obtained by this maneuver represents the self-interaction of the electrons
and is usually omitted in the first quantized representation of the Coulomb interaction. The
justification for this omission comes from the theory of Quantum Electrodynamics. There it
is shown that the effect of these self-interactions is felt only through a shift in the electron
mass. We take them into account, by using the experimental value of the mass.
The expectation value of the normal ordered Hamiltonian in a Slater determinant state is

−∇2
Z
φ∗n a (x)[
X
hΨ0 |H|Ψ0 i = d3 x + V (x)] (18.47)
n 2

1
Z
[φ∗n a (x)φan (x)φ∗m b (y)φbm (y) − φ∗n b (y)φan (x)φ∗m a (x)φbm (y)]
X
+φan (x) d3 x d3 y .
m,n 2|x − y|
(18.48)
We must vary this w.r.t. φ∗k c (z), subject to the constraints
Z
φ∗n a φam = δmn .

Imposing those constraints with a Lagrange multiplier λmn , we get

−∇2z ρ(x)δkm − ρkm (x) c

Z
0=[ + V (z)]φck (z) + λkn φcn (z) + d3 x φm (z), (18.49)
2 |x − z|
where
ρkm (x) = φ∗a a
m (x)φk (x), (18.50)
and X
ρ(x) = ρnn (x). (18.51)
n

These are the Hartree–Fock equations for atomic, molecular, and condensed matter physics.
The term involving the matrix ρkm but not its trace is called the exchange term. It would
be absent if the particles satisfied only the Pauli exclusion principle (no two particles in the
same state), rather than the full requirement of antisymmetrization. If we drop it, we get
Hartree’s self consistent field approximation. That is, the equations look like an ordinary
Schrödinger equation, with a potential
ρ(x)
Z
VHartree = V + d3 x . (18.52)
|x − z|
The intuitive idea behind this approximation is that each electron feels a potential equal
to the sum of the nuclear potential plus the Coulomb repulsion of all the other electrons,
taken into account in a sort of classical probabilistic average, with the charge density derived
370  Quantum Mechanics

from the quantum probability of finding each electron at a given position. The full Hartree–
Fock equations have no such intuitive picture associated with them, but they are the correct
expression of Fermi statistics in the variational approximation of the ground state by that of
a Hamiltonian with only single particle terms.
It is interesting to compare the Hartree–Fock approach to the Density Functional
approach. In the Hartree approximation to Density Functional Theory, the Kohn–Sham
orbitals satisfy the Hartree equations without the exchange term. It is not really correct
to think of the antisymmetrized Kohn–Sham orbital wave function as the actual electronic
wave function of the problem, though many authors do so. Exchange effects and more com-
plicated corrections to the Hartree approximation are taken into account by the corrections
to the Hartree approximation to the density functional. In principle, if one had the exact
density functional, one would still solve for the minimum energy in terms of “Exact Kohn–
Sham orbitals,” which satisfy single particle Schrödinger equations, but the energy is not
computed as the expectation value of the Hamiltonian in an antisymmetrized state made
from those orbitals.
The standard way to solve either the Hartee or Hartree–Fock equations is to start with
some choice for the φn (x) (e.g., hydrogen wave functions for an atomic physics problem),
compute the matrix ρmn (x) and then solve for the eigenstates of the new linear operator
defined by this choice. Then iterate. It turns out that the iteration is numerically stable,
because one is seeking a minimum of the energy.

18.5 THE LANCZOS METHOD

There is a general approach to improving variational approximations to the ground state
energy, which is related to a method invented by Lanczos[80], for finding all of the eigenvalues
and eigenstates of a Hermitian matrix by iteration. Imagine that we have found a decent
variational approximation to the ground state |ψ0 i, which is not an actual eigenstate. It has
an eigenstate expansion
X
|ψ0 i = cn |En i. (18.53)

Generally, apart from constraints due to exact symmetries, all of the cn will be nonzero.
Consider the quantities
X
E(k) ≡ hψ0 |H k |ψ0 i = |cn |2 Enk . (18.54)

The vectors |vk i = H k |ψ0 i for 0 ≤ k ≤ N − 1 are not orthonormal, but will generally be a
basis of an N -dimensional subspace of the Hilbert space. Indeed the scalar products between
those basis elements are given by

hvp |vq i = E(p + q). (18.55)

 The Variational Principle  371

Now let us express a vector |ψN i in the N -dimensional subspace as a linear combination of
the |vk i.
N
X −1
|ψi = an |vn i. (18.56)
n=0

The coefficients are arbitrary complex numbers, so they are independent of their complex
conjugates. The expectation value of the Hamiltonian in this state is
a∗ am [E(n + m + 1)
P
Pn . (18.57)
a∗n am E(n + m)

Varying this w.r.t. a∗n (remember that a complex variable and its conjugate are independent)
N
X −1
[E(n + m + 1) − hHi0 E(m + n)]am = 0. (18.58)
m=0

The second term in brackets comes from varying the denominator, which accounts for the
minus sign and the fact that it is proportional to hHi. Let us call hHi0 ≡ E0 and introduce
a collection of real N -dimensional vectors
 
E(k)

 E(k + 1) 

Ek =  .. . (18.59)
.
 
 
E(k + N − 1)

Thinking of the coefficient an as a complex N vector a, we can write the variational equations
as
Pm+1 ≡ Em+1 · a = E0 Em · a, (18.60)
so that
Em · a = E0m E0 · a. (18.61)
The vectors Ek for 1 ≤ k ≤ N will generally be a basis for the space of all N vectors so that
N
X
E0 = ck Ek . (18.62)
k=1

We should emphasize that given a choice of Hamiltonian and |ψ0 i, the ck are relatively easily
determined in terms of E(m) for 1 ≤ m ≤ 2N + 1. The variational equations now read
N
X
Pm = E0m cn Pn . (18.63)
n=1
372  Quantum Mechanics

The consistency condition for this equation to have solutions is

N
X
E0m cm = 1, (18.64)
n=1

and the estimate for the ground state energy is the lowest solution of this polynomial equa-
tion. This approach, and the more elaborate Lanczos method for finding all the eigenvalues in
the subspace generated by acting with the Hamiltonian N − 1 times on |ψ0 i, have been used
extensively in a variety of problems. An entry to this literature can be found in[80]. The par-
ticular approach to the ground state described here does not require one to orthonormalize
the vectors |vk i, and is likely to be computationally simpler than the general Lanczos tech-
nique. One general problem with the Lanczos method is that for systems with a ground state
energy proportional to a large volume V , the estimates do not automatically scale linearly
with V . One must find a regime in V with stable linear scaling, for each N , and establish
that that regime goes to infinite volume as N goes to infinity.

18.6 EXERCISES
18.1 Show that the exact eigenvalue for the potential gx2q scales with g just like the Gaussian
variational answer.

18.2 Calculate the ground state energy of helium using the variational ansatz that the wave
function is the product of wave functions for a single electron ion with a “shielded”
charge Z. Use Z as the variational parameter.

18.3 Consider a particle moving in two dimensions with Hamiltonian

H = p2x + p2y + V0 {θ(|x| − 2a)[1 − θ(a − |y|)] + θ(|y| − 2a)[1 − θ(a − |x|)],

in the limit V0 → ∞. The particle is forbidden to enter the regions where the potential
is nonzero. Classically it can run out to infinity in the x direction if its y position
satisfies |y| < a and its y velocity vanishes. Use a variational argument to show that,
quantum mechanically, this can only occur above some threshold energy. That is, there
are positive energy bound states in the system.

18.4 Obtain a variational bound for the ground state energy of the Yukawa potential V (r) =
−kr
g e r . Try a variational wave function of exponential form, like the ground state of
hydrogen.

18.5 Define the “harmonic atom” by replacing the Coulomb potential by harmonic oscillator
potentials. Show that this problem is exactly soluble. For simplicity, do this problem
 The Variational Principle  373

in one dimension. You will find it easiest to do this problem without using quantum
field theory. The Hamiltonian (in units where ~ and the mass are set equal to 1 is
H = Pi2 /2 + 1/2Ω2 Xi2 − 1/2ω 2 ij Xi Xj , which is a collection of coupled oscillators.
P P

You solve it by finding the normal modes, but then you have to impose the constraints
of Fermi statistics. Work in the limit N  1, Ω  Ω − N ω > 0.

18.6 Define the Harmonic Hartree approximation to the harmonic atom by looking for the
2 P
single particle harmonic potential ν2 Xi2 centered at the origin, whose ground state
for N electrons is the best variational approximation to the Hamiltonian of Exercise
18.5. The ground state is the Slater determinant made from the N lowest eigenstates
of this oscillator. Compare the result to the exact answer.

18.7 Consider a Hamiltonian of the form i<j Kij σ a (i)σ a (j). i and j are some finite set
P

of “sites,” whose total number is N . Consider the ground state of a Hamiltonian

P a a
i h (i)σ (i) as a variational approximation to this problem. Find the equation deter-
mining the best values for the “local magnetic fields” ha (i).

18.8 Let us apply the Lanczos method to the Hamiltonian P 2 /2m + mω12 X 2 /2, starting
from the ground state wave function of the oscillator with frequency ω0 . Compute the
expectation values of the first two nontrivial powers of the Hamiltonian H 2 and H 3 . Use
these to evaluate the first three moments h0|H k |0i. In doing this, you will be helped
by noticing that if P 2 acts to the left or right on the oscillator ground state, it can
be replaced by ~ω − ω 2 X 2 . Also use the fact that X 2 acting on the ground state is
a linear combination of that state and the second excited state. This enables one to
reduce the calculation to expectation values of powers of X. Evaluate the first Lanczos
approximation to the ground state energy and compare to the exact answer.
 CHAPTER 19

The Feynman Path Integral

19.1 INTRODUCTION
Feynman’s path integral approach to quantum mechanics (QM) is the most useful way to
approach most QM problems. It is based on the Lagrangian approach to mechanics, and thus
manifests symmetries much more clearly than the Hamiltonian formalism. This is particularly
true for symmetries like Lorentz boosts, which do not leave the Hamiltonian invariant.
The path integral is a formal solution to the equations of QM in terms of an infinite dimen-
sional integral. It avoids many subtle issues having to do with the domains of unbounded
operators, and directly computes observable quantities. It is also relatively straightforward
to discretize path integrals and use sophisticated numerical integration routines to evaluate
them.
Indeed, the virtues of the path integral formalism are so numerous that one is tempted
to rewrite all quantum textbooks in path integral language. The one drawback is that it is
much harder to explain the probability interpretation of the formulae in the path integral
formula. For example, if one adds a term ẋ4 to the Lagrangian of a free particle, the (imag-
inary time) path integral is perfectly well defined as long as the coefficient of this term is
positive. However, the quantities computed from the new path integral cannot be interpreted
as expectation values of operators in a Hilbert space. For this and other reasons, we have
hidden the path integral chapter at the back of the book. Perhaps one can think of it as
saving the best for last.

19.2 TWO DERIVATIONS OF THE PATH INTEGRAL FORMULA

We are all used to the fact that the equations of physics become simpler over very short time
intervals. We can write closed form differential equations for many problems whose finite
time evolution is extremely difficult to figure out. The insight behind the Feynman Path
Integral formulation is that, in QM at least, the simplification of very short time intervals

375
376  Quantum Mechanics

leads to a “reduction to quadratures” of the finite time solutions. That is, one can write a
very explicit solution of the problem in terms of an integral formula. The catch is that the
integral is always infinite dimensional, if we want the exact answer. Most infinite dimensional
integrals are very hard to compute. On the other hand, if one is willing to live with finite
precision, one can evaluate the integrals with powerful numerical integration techniques. The
path integral (also called functional integral) technique has led to many important insights
into QM problems and is virtually the universal tool of choice in the study of many particle
problems.
The cleanest derivation of the path integral is obtained by analytically continuing to
imaginary time. The fact that quantum amplitudes are analytic is a rigorous theorem for
finite dimensional Hamiltonians. For Hamiltonians that are bounded from below, e−τ H is a
very well behaved operator. τ is the analytic continuation of it~ to positive imaginary values.
From this point on, in this Chapter, we will set ~ = 1. The infinite dimension of the Hilbert
space almost always comes from very high-energy states.1 If the high energy density of states
grows less rapidly than an exponential,2 the exponential damping of imaginary time makes
all infinite sums over states convergent.
Now let us write
τ
hψ(T )|e−τ H |ψ(0)i = hψ(T )|(e− N H )N |ψ(0)i. (19.1)

At this point, we insert a complete set of intermediate states between every pair of operators,
and write the initial and final states in terms of their wave function in the chosen basis.
Different choices of basis lead to different forms of the path integral. If we actually use a
countable basis of normalizable states, we get a path sum rather than a path integral. This
is useful for fundamental variables like spin operators, which operate in finite dimensional
P2
Hilbert spaces. However, we will begin by studying a Hamiltonian H = 2m + V (X) and
inserting position eigenstates. We are then led to the computation of
τ
hx(ti )|e− N H |x(ti−1 )i, (19.2)

with an accuracy of order τ /N in the exponent.

Recall the Zassenhaus formula from Exercise 6.3
t2 t3
et(W +Y ) = etW etY e− 2 [W,Y ] e 6 (2[Y,[W,Y ]]+[W,[W,Y ]])
t4
× e− 24 ([[[W,Y ],W ],W ]+3[[[W,Y ],W ],Y ]+3[[[W,Y ],Y ],Y ]) . . . . (19.3)

1
Most infinite degeneracies associated with low energy can be regulated by putting the system in finite
volume.
2
This is the case for all known systems, which do not involve gravity or string theory.
 The Feynman Path Integral  377

P 2
When W = 2m and Y = V (X), and t = −τ /N , the logarithms of the terms with powers of t
higher than 1 are o(1/N 2 ) or smaller, and therefore should not contribute to the final answer
as N → ∞. On the other hand,
s
τ
− 2mN τ
P2 −N V (X) 2mN − mN (x(ti )−x(ti−1 )2 − τ V (x(ti−1 )
hx(ti )|e e |x(ti−1 )i = e 2τ e N . (19.4)
~2 τ
You can derive this easily (Exercise 19.1) by inserting a complete set of momentum eigen-
states. Remember that the x(t) variables are integration variables. As N → ∞, the first
exponential factor suppresses integration regions for which the x variables at neighboring
times are not close to each other. So we can think of x(t) as a continuous function of time.
For differentiable functions, the product of exponentials from all the infinitesimal time inter-
vals converges to the classical action (for imaginary time). Dirac [81] was the first to notice
this, but Feynman was the person who exploited this observation to do new physical calcu-
lations.
Formally, ignoring the prefactors and the fact that not all continuous functions are dif-
ferentiable, we can write the answer as
Z
−τ H
hx(T )|e |x(0)i = [dx(t)]e−S[x(t)] , (19.5)

where the functional integral is over all continuous paths, which have fixed endpoints. We can
deal with the nasty prefactors by noting that they depend only on N and τ and so are the
same for a free particle as they are for any potential, and are independent of the endpoints.
So we can write the ratio of amplitudes for any potential and the free particle, as a ratio of
two path integrals and not worry about overall position-independent factors in the individual
path integrals.
What is really going on here is that in the limit of small times, the system is completely
dominated by free motion, as long as the potential is not wildly varying with position. For
continuous paths, the contribution of the potential to the short time motion is more or less
constant. This must be rethought for some singular potentials, but it is a rule of very wide
general applicability.
The foregoing was Feynman’s derivation of the path integral formula. What follows is a
derivation, of a similar formula, which follows an argument due to Schwinger. Consider the
P2
Hamiltonian 2m + V (X) again, but now subject it to a time-dependent perturbation

δH = −J(t)X. (19.6)

Following our discussion of time-dependent perturbation theory, the evolution operator of

the perturbed system in the interaction picture can be written in terms of the Heisenberg
operators X(t) of the unperturbed system:
378  Quantum Mechanics
Rt
i J(s)X(s)
UI (t, t0 ) = T e t0
. (19.7)

Taking any matrix element of this operator, we get a complex valued functional Z[J] of the
source J(t). The physical meaning of that functional is clear: it is the amplitude that the
source J(t) induces a transition between a pair of Heisenberg states of the original system.
We have not indicated the dependence on the initial and final state. In most applications,
they are identical and we take the thermal average over them at some temperature T . As
T → 0, the thermal density matrix becomes the projection on the (possibly degenerate)
ground state. Z is then called the ground state persistence amplitude.
The time-ordered exponential is defined by its power series expansion in powers of J, so
if we make a small change in the source J → J + δJ, we find that

δZ
Z
δZ ≡ δJ(s), (19.8)
δJ(s)

which defines the functional derivative of Z. The functional derivative of Z is equal to

1 δZ R
= hf inal |T [X(t)ei J(s)X(s)
]| initiali. (19.9)
i δJ(t)

Using the Heisenberg equations of motion, we have

R d 1 δZ
hf inal |T [P (t)e−i J(s)X(s)
]| initiali = m . (19.10)
dt i δJ(t)

This equation is correct, but we have to be careful of its derivation, because there is implicit
t dependence in the definition of the time ordering symbol, through factors like

θ(t − s)X(t)X(s) + θ(s − t)X(s)X(t).

Fortunately, when we take the derivative of the Heaviside step functions, we get two terms,
which combine to give the commutator of X(t) with itself at equal times. This is a result of
the identity
d d
θ(t − s) = δ(t − s) = − θ(s − t). (19.11)
dt dt
We now repeat this exercise for the second time derivative, this time picking up a term from
the commutation of X and P at equal times. Since this commutator is just a number, it
comes out of the time-ordered product. In Exercise 19.2, you will show that

d2 1 δZ 0 −i
R
J(s)X(s)
m + hf inal |T [V (X(t))e ]| initiali = −J(t)Z. (19.12)
dt2 i δJ(t)
 The Feynman Path Integral  379

Exercise 19.2 Give a careful derivation of the equation directly above.

We can rewrite this as

" #
d2 1 δZ 1 δ
 
0= m 2 +V0 + J(t)Z . (19.13)
dt i δJ(t) i δJ(t)
This equation is both familiar looking and peculiar. It looks like the classical equation
of motion, except that the classical position x(t) is replaced by the functional differential
δ
operator 1i δJ(t) , inserted into the classical equations of motion, and allowed to act on the
functional Z.
If we expand the functional Z out into a functional power series
∞
1
X Z
Z[J] = ds1 . . . dsn Gn (s1 . . . sn )J(s1 ) . . . J(sn ), (19.14)
n=0
n!
the functional differential equation breaks up into an infinite set of ordinary differential
equations for the coefficients. These are called the Schwinger–Dyson (SD) equations [82].

Exercise 19.3 Derive the SD equations from the functional equation for Z[J].

How do we think about, much less solve, this scary set of equations. Everything infinite
is a limit of something finite. We can make these equations look more familiar by breaking
the time interval t − t0 up into N intervals of equal length and replacing the time derivatives
by finite time differences. When we do that, the number of source variables J(tn ) ≡ Jn
(evaluated for example at the midpoint of each interval) becomes finite and the scary looking
δ
functional equations become ordinary partial differential equations via δJ(t) → ∂J∂n . Now we
can think about them.
Our first observation is that despite their obvious relation to classical equations, these
equations are linear equations for Z.3 Secondly, although they may contain high powers of
the partial derivative operator ∂J∂n , they are linear in the source Jn itself. This suggests a
strategy: the Fourier transform turns derivatives into multiplication by the Fourier conjugate
variable. If we Fourier transform the SD equations w.r.t. all of the variables Jn , we will get a
first-order linear partial differential equation (PDE) in the Fourier conjugate variables. We
will call the latter xn . The Fourier transform will involve an integral over all of these variables
xn , which will become a functional integral in the limit of continuous time.
Now that we have got the idea, let us do all of this directly in the continuum limit. We
write Z R
Z[J] = [dx(t)]ei(S[x]+ J(s)x(s)ds) . (19.15)

3
This is connected to the fact that the Schròdinger equation is a linear equation for the time evolution
operator.
380  Quantum Mechanics

The SD equations then become

d2 x
Z R
0= [dx(t)]ei(S[x]+ J(s)x(s)ds)
[m + V 0 (x) + J(t)]. (19.16)
dt2
R
δ(Sc [x]+ Jx)
The term in square brackets is of course equal to δx(t) , where Sc [x] is just the classical
action for the unperturbed system. If we choose the Fourier transform functional S[x] equal
to Sc , then the integrand is just
δ R
ei(Sc [x]+ J(s)x(s)ds)
. (19.17)
iδx(t)

This integral of a total derivative will vanish if the quantity being differentiated vanishes at
infinite values of x(t). We again see the virtue of giving time a positive imaginary part and
defining the real time theory as a limit of this analytic continuation. For purely imaginary
time, the functional integral formula looks like
Z R
Z[J] = [d[x(t)]e−(SE [x]+ J(s)x(s)ds)
. (19.18)

The Euclidean action SE is just −Sc with t → iτ .

In this way of approaching the path integral formalism, we did not have to try to define
what the path integral formula was. It was any linear function on functionals, with the
property that it gives the same result if we replace x(t) by x(t) + y(t). For a finite number of
variables, this defines the integral up to an overall multiplicative constant. That cannot be
right here, because we have implicitly defined Z for any choice of initial and final state. We
also have to deal with the fact that the overall constant is likely to be infinite. It is best to
deal with these issues in the context of an explicit example, so we now turn to:

19.3 THE PATH INTEGRAL FOR A HARMONIC OSCILLATOR

The simplest way
q to do the Euclidean path integral for the harmonic oscillator is to write
2 P
x(t) = xc (t) + T cn sin(nπt/T ), where xc (t) is the solution of

ẍc (t) = ω 2 xc (t), (19.19)

p
with ω = k/m, which satisfies the boundary conditions imposed by the initial and final
position eigenstates. By Fourier’s theorem, cn parameterizes the space of functions satisfying
δx(0) = δx(T ) = 0, so this decomposition is a way of describing all functions which satisfy
the boundary conditions. Since xc (t) minimizes the Euclidean action, the result for the path
integral is
 The Feynman Path Integral  381
Z Z Y P RT 2
−m cn ck d
dt [sin(nπt/T )(− dt 2
2 +ω ) sin(kπt/T )]
[dx(t)]e−SE [x(t)] = e−SE [xc (t)] [dcn ]e T n,k 0 .
n
q (19.20)
2
The functions T sin(nπt/T ) are orthonormal, so
Z Z Y P
m
c2 [( nπ )2 +ω 2 ]
−SE [x(t)]
[dx(t)]e =e −SE [xc (t)]
[dcn ]e− 2 n n T . (19.21)
n

The result is just an infinite product of independent Gaussian integrals. Note that all of the
dependence on the initial and final positions comes from the classical action.
The infinite product is not convergent. However, if we divide the path integral by that
for the free particle (ω = 0), we will see that the result is finite. So we can write

h0|e−tHosc |0i
r
−tHosc −SE [xc (t)] m
hx(T )|e |x(0)i = e . (19.22)
h0|e−tHf ree |0i 2πt~2
The careful reader will want to remember that we have defined Euclidean time with a factor
of ~, so that it has dimensions of inverse energy, when comparing this equation to previous
formulae for the free particle propagator.
The infinite product represented by the ratio is
∞
!
Y ω2T 2 P∞ 2 2
ln(1+ ωn2 Tπ2 )
1+ 2 2 =e n=1 , (19.23)
n=1
n π

which converges. You will evaluate this in Exercise 19.4. In Exercise 19.5, you are asked to
repeat the computation using Feynman’s original derivation of the path integral.
To finish the path integral evaluation of the transition amplitude, we have to compute
the action. The general solution of the classical equations is

xc (τ ) = Aeωτ + Be−ωτ . (19.24)

The boundary conditions imply that

A + B = x(0), (19.25)

AeωT + Be−ωT = x(T ). (19.26)

The action is given by

mω 2 T mω[A2 (e2ωT − 1) − B 2 (e−2ωT − 1)]

Z
SE = [(AeωT − Be−ωT )2 + (AeωT + Be−ωT )2 ] = .
2 0 4
(19.27)
382  Quantum Mechanics

Working this out is a little exercise in 2 × 2 matrix multiplication, which gives

mω sinh(2ωT ) 4x(0)x(T )
SE = [ (x(T )2 + x(0)2 ) − . (19.28)
4~ sinh2 (ωT ) sinh(ωT )

The full propagator is thus

s
mω − mω
[(x(0)2 +x(T )2 ) cosh(ωT )−2x(0)x(T )]
e 2 sinh(ωT ) . (19.29)
2π sinh(ωT )
When analytically continued back to real time, this gives us
s
mω i mω [(x(0)2 +x(t)2 ) cos(ωt)−2x(0)x(t)]
e 2 sin(ωt) . (19.30)
2πi~ sin(ωt)

This formula gives us yet another way to compute the normalized eigenfunctions and the
eigenvalues of the harmonic oscillator. Indeed,
∞
En τ
− Hτ
ψn (x(T ))ψn∗ (x(0))e−
X
hx(T )|e ~ |x(0)i = ~ . (19.31)
n=0

In Exercise 19.6, you will be asked to compute a few of the Hermite functions by this method.
Using this formula, we can compute the transition amplitude between any initial and
final state by integrating the position eigenstate result against ψ ∗ (x(T ))ψ(x(0)). However, if
both the initial and final states are the ground state, there is a simpler formula. Let us look
at this first from the Hamiltonian point of view, in imaginary time. It is clear that if we take
the imaginary time to infinity, the ground state contribution dominates, so
Hτ E0 τ
lim hX(T )|e− ~ |x(0)i → ψ0 (X(T ))ψ0∗ (x(0))e− ~ . (19.32)
τ →∞

On the other hand, the path integral computation gives, in this limit
r
− Hτ − ωT m − mω (x(T )2 +x(0)2 )
lim hX(T )|e ~ |x(0)i →= e 2 e 4~ . (19.33)
τ →∞ 2π
Comparing the two expressions, we see that we can read off both the ground state energy
and the ground state wave function (up to the usual overall constant phase) from the path
integral computation.
Now let us generalize these computations to the case where the oscillator is subjected to
an external force, constant inRspace but with an arbitrary time dependence. This changes
the action by adding a term J(τ )x(τ ). By reviewing our derivation of the path integral
formula, you will see that the only change is that we have to solve the classical equations
 The Feynman Path Integral  383

with an extra source term. The resulting classical action differs from that with J = 0 by a
term quadratic in J.4 . It has the form

1
Z
δS = dτ dσ [J(τ )G(τ, σ)J(σ)], (19.34)
2
where
d2
+ ω 2 ]G(τ, σ) = δ(τ − σ).
[− (19.35)
dτ 2
This is a Green’s function equation and has many solutions. The one appropriate to the tran-
sition amplitude between position eigenstates is fixed by imposing the boundary conditions
Z T Z T
x(T ) = dσG(T, σ)J(σ); x(0) = dσG(T, σ)J(σ). (19.36)
0 0

What is new and exciting about the external force problem is that we can also get ground
state to ground state transition amplitudes by imposing boundary conditions on the Green’s
function equation.
Our Hamiltonian discussion of the J = 0 problem suggests that we should consider the
T → ∞ limit. In the time-dependent problem, this should be a sensible thing to do if J is
turned off asymptotically in time. To be more precise, we consider the interval [T, −T ] and
consider a force J(τ ) such that J(±T ) → 0 as T → ∞. For large imaginary times, the time-
independent Hamiltonian analysis shows that we should be projecting out the ground state.
So, in the limit of large imaginary time, we are studying the amplitude for the ground state
of the J = 0 problem, to remain the ground state after being subjected to a time-dependent
source, which is turned on only for a finite time. This was the sort of problem we studied in
Chapter 13 on time-dependent perturbation theory. To be more precise, we are studying such
a problem if we take the source function J(τ ) and analytically continue it to a real function
of real time.
The Dirac picture analysis, in imaginary time, tells us that the expression for this ampli-
tude is R∞
i J(τ )x(τ )
Z[J] ≡ hΨ0 |T [e −∞ |0i. (19.37)
On the other hand, the path integral analysis tells us that the same expression is computed
(modulo a prefactor, which depends only on time, and which we have already computed at
finite time) by solving the classical equations of motion with some boundary conditions. The
obvious boundary conditions are that x(τ ) go to the classical minimum of the potential at
T → ±∞. The expression for the classical action as a functional of J(s) is
4
Without even solving the equations, you can see that the possible linear term in J is absent, because the
action is invariant under the simultaneous reflection J → −J, x → −x.
384  Quantum Mechanics

1
Z ∞
SE cl = dτ dσ [J(τ )G(τ − σ)J(σ)]. (19.38)
2 −∞

d2
[− + ω 2 ]G(τ − σ) = δ(τ − σ). (19.39)
dτ 2
We can solve the Green’s function equation by Fourier transformation

dz eiz(t−s)
Z
G(t − s) = . (19.40)
2π z 2 + ω 2
Since the integrand has two poles at z = ±iω, different choices of contour in the z plane will
give different answers. In Exercise 19.7, you will show that these differences correspond, as
they must, to different homogeneous solutions that can be added to any particular solution
of the Green’s function equation.
Choosing the contour along the real z axis, we can evaluate the integral by closing the
contour in the upper (lower) half z-plane if τ − σ > 0 (τ − σ < 0). We get
1 −ω|τ −σ|
G(τ − σ) = e , (19.41)
ω
which falls to zero at infinity, as it should. If we now try to analytically continue τ − σ to
negative imaginary values, −i(t − s) to obtain the real ground state persistence amplitude,
we can do this explicitly, or by rotating the contour of the z integration to z = iE. The
resulting real time Green’s function has the expression

dE ie−iE(t−s)
Z
G(t − s) = . (19.42)
2π E 2 − ω 2 + i
 is an infinitesimal positive number. Inserting  is equivalent to deforming the contour of
the real E integration so that the positive ω pole is below the contour, and the negative
frequency pole above it. An equivalent statement of this contour choice is that no matter
what the sign of t − s, only positive energies propagate forward in time.
Comparing the Hamiltonian and path integral expressions for the ground state persistence
amplitude, and expanding to second order in J, we find the equation

hΨ0 |T x(t)x(s)|Ψ0 i = G(t − s). (19.43)

If we pick a time order and evaluate the left-hand side by putting in a complete set of
intermediate states, then we see that indeed only positive energy propagates forward in
time. Higher order time-ordered products of the x(t) operators are evaluated in terms of
sums of products of G(ti − tj ) functions. This result is called Wick’s Theorem, and you will
explore it in Exercise 19.8.
 The Feynman Path Integral  385

19.4 MORE GENERAL POTENTIALS

One can develop a perturbation expansion around the harmonic oscillator problem by expand-
ing a general potential around a local minimum and treating the corrections to the quadratic
term as small. Wick’s theorem allows us to evaluate each term in this series, and there is a
nice pictorial algorithm for computing them, known as Feynman diagrams. You will explore
simple examples in the exercises. The series generated in this way give only asymptotic
expansions, rather than a convergent one. It is worth understanding the origin of this, even
though we do not have space in a course like this one, to go into the details.
Given a function f (z), an asymptotic series ∞ n
P
n=0 fn z is a formal expression, such that
if one takes the polynomial PN (z) formed by the first N terms, then

|PN (z) − f (z)| ∼ |z|N , (19.44)

as |z| → 0 within some wedge of finite opening angle in the z plane. If az0 with a real is a
line in this wedge, then f (az0 ) is infinitely differentiable at 0 w.r.t. the variable a and the
k-th term in the series is k!f (k) (0). Now think about the function e−1/z . As long as the real
part of 1/z is positive, the derivatives of this function at the origin all vanish. So, anywhere
in the wedge with Re z1 > 0, we can add this function to any f (z) that has an asymptotic
series in any smaller wedge, and get a new function with the same asymptotic series. If the
series is really convergent, then among the many functions with the same asymptotic series,
there is a unique one, which is analytic in an entire disk surrounding the origin.
Here is an example of a function with an asymptotic series, which is not analytic.
Z ∞ 2 −zx4
f (z) = dx e−x . (19.45)
−∞

It obviously has an asymptotic series for real positive z, but the coefficient of z n is
(−1)n
Z
2
e−x x4n .
n!
For large n the integral grows like (2n)! so the series diverges. The reason is sort of obvious.
The integral diverges when z is negative, so the function cannot be analytic in a full disk. The
functional integral formulation of QM makes it clear that similar issues will be encountered
in QM perturbation series.
For problems with a small number of variables, there are other methods of computing
the coefficients in the perturbation expansion of eigenvalues and eigenfunctions. In particu-
lar, for single variable problems with potentials well approximated by polynomials, there are
powerful difference equation techniques [83] for computing both the eigenvalues and the eigen-
functions, which are much more efficient than either path integral or Rayleigh–Schrödinger
computations.
386  Quantum Mechanics

Path integral methods really come into their own in systems with many variables, basi-
cally because one can compute Gaussian integrals in any number of dimensions. Feynman’s
diagrammatic series are quite universal in many body problems.

19.5 PATH INTEGRALS AT FINITE TEMPERATURE

−βH
We have seen that the thermal density matrix for a quantum system is ρβ = e Z , where
the partition function is given by Z = tr e−βH . Given a system at thermal equilibrium,
we can consider two different kinds of physical processes. First, in order to understand the
thermodynamics of the system, we want to compute thermal expectation values of various
operators. We can do this by perturbing the Hamiltonian H → H + λa (t)Oa , computing
the partition function Z(λ) and taking derivatives of its logarithm w.r.t. the λa . t is the
Euclidean time parameter.
On the other hand, perturbing the system kicks it out of equilibrium, and we might be
interested in studying the time dependence of various quantities as they decay back to an
equilibrium state. The first kind of computation is a relatively simple variation on things we
have already done, while the time-dependent calculations are intricate and messy. For that
reason, we will treat only equilibrium expectation values.
Our task then is to compute tre−βH , and for a system described by a number of canonical
coordinates Qi , we do this by taking the trace in the basis where these coordinates are
diagonal. Thus, Z
Z= dN qhq|e−βH |qi. (19.46)

The integrand is just the Euclidean time continuation of the amplitude to start at a particular
point in configuration space and return to that point in a fixed time.
The matrix element of the time evolution operator between two points has, as we have
seen, a path integral representation over all paths that start and end at those points in the
required time. In this case, the points are the same, so the paths are periodic with period β,
and we integrate over the end point, so we get all periodic paths. Thus,
Z
Zβ = [dq(t)]e−SE [q(t)] , (19.47)

where we integrate over all paths in imaginary time, which are periodic, with period ~β.
Thermal expectation values are computed by taking logarithmic derivatives of this formula
with respect to perturbing parameters, and then setting those parameters equal to zero.
For the harmonic oscillator, we have already computed the path integral for
s
−βH mω mω
hy|e |xi = exp[− ([x2 + y 2 ] cosh(~ωβ) − 2xy)]. (19.48)
2π~ sinh(~ωβ) 2π~ sinh(~ωβ)
 The Feynman Path Integral  387

To evaluate the thermal partition function, we have to set x = y and integrate over x. The
integral is Gaussian
π
Z
2
dxe−Ax = ( )1/2 , (19.49)
A
with
mω
A=2 (cosh(~ωβ) − 1).
2π~ sinh(~ωβ)
Thus, s
1
Z= . (19.50)
2[cosh(~ωβ) − 1]
On the other hand, straightforward evaluation of
∞
1
e−~ωβ(n+ 2 )
X
Z= (19.51)
n=0

yields
1~ωβ
Z = e− 2 . (19.52)
1 − e−β~ω
Simple algebra shows that these two expressions are equal.
One can also do the imaginary time path integral with an external source added to the
Lagrangian via Z
δL = ds j(s)x(s). (19.53)
The source must be periodic in imaginary time. Dividing by the partition function Z, we get
the generating functional Z[j] for thermal expectation values
tr e−βH x(s1 ) . . . x(s2 ). (19.54)
The answer follows from calculations that we have already done and we obtain
1
R
dsdt j(s)j(t)G(t,s)
Z[j] = e 2 , (19.55)
where
d2 1
[−2
+ ω 2 ]G(t, s) = δ(t − s). (19.56)
dt m
The solution with periodic boundary conditions is
n(t−s)
∞
X e2πi β~
G(t, s) = 2πmn2
. (19.57)
2
n=−∞ mω + β 2 ~2

Below, we will formulate path integrals for anticommuting variables (fermions) and we will see
that the finite temperature results require antiperiodicity rather than periodicity in Euclidean
time.
388  Quantum Mechanics

19.6 PATH INTEGRALS AND THE JWKB APPROXIMATION

The path integral method provides an instantaneous derivation of the leading order JWKB
iHt
approximation to a variety of quantum amplitudes. As an example, let us consider hx|e− ~ |yi,
where x, y are a short hand notation for any configuration space with a Lagrangian whose
kinetic term takes the form
1
Lkin = ẋi Mij ẋj , (19.58)
2
with Mij a constant symmetric matrix. In Exercise 19.9, you will see what happens when M
depends on x. The path integral formula for this amplitude is just
Z Scl [x(s)]
− iHt −1
hx|e ~ |yi = N [dx(s)]ei ~ , (19.59)

where the integral is over all paths x(s) which go from x(0) = y to x(t) = x. The normaliza-
tion factor N is determined by the requirement that
iHt
limt→0 hx|e− ~ |yi = δ(x − y). (19.60)

In the semiclassical approximation, this functional integral is evaluated as

iHt Scl [xcl (s)]
hx|e− ~ |yi = ei ~ det−1/2 (D/D0 ), (19.61)

where D is the matrix differential operator

−d2 ∂2V
Dij = Mij + (xc (s)), (19.62)
ds2 ∂xi ∂xj
2
and D0 is a constant times Mij −d ds2 , with the constant determined by the normalization
condition. xc (s) is the classical solution of the equations of motion satisfying the boundary
conditions. The determinants come from doing the Gaussian integral for fluctuations around
the classical solution, which have boundary conditions ∆x(t) = ∆x(0) = 0. There is a rather
general and beautiful theorem for calculating these functional determinants, due to Gelfand
and Yaglom [84].
Gelfand and Yaglom showed that in the case of a single x variable, the functional deter-
minant could be calculated by solving the initial value problem
−d2 d2 V
[ + (xc (s))]v(s) = 0. (19.63)
ds2 dx2
dv
with boundary conditions v(0) = 0, and dx (0) = 1. Then
2 d2 V
det [ −d
ds2 + dx2 (xc (s))] v(t)
2 = . (19.64)
det [ −d
ds2 ]
v0 (t)
 The Feynman Path Integral  389

Here v0 (s) is the solution of the same boundary value problem for the action with vanishing
potential. You will prove this result in Exercise 19.10. Generalizations of it can be found in
the review by Dunne [84].

19.7 PATH INTEGRALS FOR SPIN AND OTHER DISCRETE VARIABLES

In his famous book on the Path Integral formulation of QM [85], Feynman lamented the fact
that he had not found a path integral formulation for single nonrelativistic particles with
spin. Thinking about this problem leads one in a number of interesting directions. A first
approach is just to note that we can use the path integral strategy of dividing time up into
small intervals for Hamiltonians that depend on spin. When we do that, we end up inserting
complete sets of intermediate states of both spin and position, at each time. Suppose the
spin-dependent Hamiltonian takes the form of a magnetic dipole moment in an external
magnetic field Hspin = µBa (X)Ja . Ja are the spin matrices of the particle. For short times,
we can write
τ τP 2 τ τ
e− N H = e− 2mN e− N V (X) e− N Ba (X)Ja . (19.65)
When we sandwich this between X eigenstates, we get the discrete approximation to the
imaginary time path integral over x(τ ) from the first two factors, while the second gives the
discrete approximation to
Z τ
Z[B] ≡ T exp[− Ba (x(τ ))Ja (x(τ ))].
0

This derivation illustrates one of the most useful features of the path integral formalism,
the decomposition of a quantum problem involving two interacting systems into individual
problems, in which each of the systems evolves in a time-dependent background determined
by the path integration variable of the other. Feynman called this the influence functional
method. In the case of spin, it tells us that if we can find a path integral formula for the
problem of a spin in a time-dependent background field Ba (τ ), then we have a path integral
that will apply to the spinning particle (or any other problem in which the spin Ja interacts
with another quantum system).
The key question is thus how to write path integral formulae for time-ordered products of
operators in a finite dimensional Hilbert space. If we follow Feynman’s procedure of evaluating
matrix elements of the short time evolution operator in some particular basis, then we get
a “path sum” formula, in which there are no obvious simplifications in the short time limit.
On the other hand, if the evolution operator U (t, t0 ) is a continuous function of t, then the
vector in Hilbert space does not change very much when we make a small change in t. This
suggests that we use an overcomplete basis, consisting of all the vectors in Hilbert space to
write the path integral for a finite dimensional system. The quantum state of such a system
is a set of N complex numbers zi satisfying i |zi |2 = 1. The overall phase rotation zi → eiθ zi
P
390  Quantum Mechanics

does not change the state. Equivalently, we can think of the space of states as unnormalized
complex N vectors, with the identification zi ∼ λzi , where λ is a general nonzero complex
number. The space of quantum states is just a complex projective space. The conventional
mathematical name for this space is CP N −1 , because it has N − 1 complex dimensions.
One way to think about the space of states of a quantum system is that it is the degenerate
eigenspace of a Hamiltonian which is equal to zero on every state. We can write a classical
Lagrangian for N complex variables, which gives vanishing Hamiltonian, as follows
i
L = (zi∗ ∂t zi − zi ∂t zi∗ ). (19.66)
2
The canonical commutation relations following from this Lagrangian are
[zi , zj∗ ] = δij , (19.67)
which are the relations for N creation–annihilation operator pairs. If we impose the constraint
zi∗ zi = 1,
X
(19.68)
then there are only N states which satisfy it, namely the states where one of the harmonic
oscillators built from the creation and annihilation operators is excited to its first level, while
all the others are in their ground state.
This constrained subspace of states can be viewed as the result of doing the path integral,
with the above Lagrangian, with the variables restricted to live on the compact manifold
CP N −1 . That manifold is obtained from complex N -dimensional space by imposing the
“gauge equivalence” zi ≡ λzi , with λ an arbitrary complex number. This just means that
the parameterization of CP N −1 by N complex numbers is redundant. We partially fix the
ambiguity by imposing the constraint zi∗ zi = 1, which shows that the manifold is compact.
P

What remains is a phase ambiguity zi ≡ eiθ zi . In a path integral, we can multiply zi (t) by a
different phase θ(t) at each time. If our integral is really over variables defined on CP N −1 ,
P ∗
the action should be invariant under such a change. Indeed, since zi zi = 1, the action
changes to Z Z
L→ (L + ∂t θ) = θ(t2 ) − θ(t1 ). (19.69)
i
The propagator between the two points z1,2 on the manifold of complex unit N -vectors will
thus change under a gauge transformation by G(z2 , t2 ; z1 , t1 ) → ei(θ(t1 )−θ(t2 ) G(z2 , t2 ; z1 , t1 ).
The solution of the time-dependent Schrödinger equation is
Z
ψ(z2 , t2 ) = dN z1 G(z2 , t2 ; z1 , t1 )ψ(z1 , t1 ). (19.70)

We see that the phase ambiguity in the path integral is equivalent to the statement that the
phase of the wave function is unphysical.
 The Feynman Path Integral  391

We can now add a general Hamiltonian for the system by adding a Hamiltonian function
via the standard prescription L → L0 −H(zi∗ , zi , t). This will be a well-defined Lagrangian on
the manifold CP N −1 , if H is invariant under the phase transformation zi → eiθ zi . Note that
we have allowed for explicit time dependence in the Hamiltonian to accommodate interactions
between our N -dimensional system and other variables, as explained above.

19.8 FERMIONS AND GRASSMANN INTEGRATION

When N = 2n , there is another, often more convenient, way of representing finite dimensional
systems in terms of fermionic path integrals. The Hilbert space of dimension 2n can be
presented in terms of n operators satisfying

[ψk , ψl ]+ = 0, (19.71)

[ψk , ψl† ]+ = δkl . (19.72)

Start from a state satisfying
ψk |0i = 0, (19.73)
and construct
ψk†1 . . . ψk†p |0i, (19.74)
for 1 ≤ p ≤ n. The total number of states is 2n . As you will prove in Exercise 19.11, we can
write these operators in terms of 2n Hermitian operators γa satisfying

[γa , γb ]+ = δab . (19.75)

Given a time-dependent Hamiltonian built out of these variables, we can construct matrix
elements of time-ordered products of γa (t). We define the time-ordered products with minus
signs, so that they are totally antisymmetric under interchange. For example,

T γa (t)γb (s) = θ(t − s)γa (t)γb (s) − θ(s − t)γb (s)γa (t). (19.76)

A generating functional for such totally antisymmetric products can be constructed with the
help of Grassmann numbers. Grassmann numbers are complex linear combinations of a finite
number of generators ηi , which satisfy the multiplication rule

[ηi , ηj ]+ = 0.

We will need to talk about Grassmann valued functions. These are simply defined by letting
the complex parameters be functions of time. Since there are an infinite number of linearly
independent functions, we will need an infinite number of Grassmann generators.
392  Quantum Mechanics

Grassmann introduced Grassmann variables as part of the mathematical theory of

differential forms. They have properties attributed to infinitesimals in less rigorous pre-
sentations of that subject. They square to zero, and a product of k of them requires k
different “dimensions” and can represent an infinitesimal k-plane. We will be interested in
them because we can define notions of derivative and integral for Grassmann algebras, just
as we can for ordinary functions.
Functions of a finite number of Grassmann generators (i.e., general elements of the Grass-
mann algebra) are finite order polynomials. The coefficients are antisymmetric tensors. So,
we can define the derivative, ∂η∂ i , by a simple algebraic rule: it is zero if ηi does not appear
in the monomial one is differentiating. If it does appear, it appears linearly, so we can define
the derivative by simply dropping that variable, and multiplying by (−1)P where P is the
number of other Grassmann variables you have to move the derivative through in order to
“get to” ηi . For example,
∂
(η1 η2 η3 ) = η2 η3 . (19.77)
∂η1
∂
(η1 η2 η3 ) = −η1 η3 . (19.78)
∂η2
∂
(η1 η2 η3 ) = η1 η2 . (19.79)
∂η3
Of course the minus sign depends on whether you start differentiating from the left or the
right and some people define both left and right Grassmann derivatives. We will stick with the
left derivative. Once we have defined the derivative for monomials, we extend it to arbitrary
polynomials by insisting that it act linearly.
Linearity is also the key to defining an integral over Grassmann variables. For a single
Grassmann variable, the most general function is a + bη. By linearity, we must have
Z Z Z
dη(a + bη) = a dη1 + b dηη. (19.80)

There is no analog of an indefinite integral for Grassmann numbers. For the definite integral,
we want the integration by parts rule

∂
Z
dη f = 0. (19.81)
∂η
∂η R
Since 1 = ∂η , we must have dη1 = 0. We normalize the only nonzero result by
Z
dηη = 1. (19.82)
 The Feynman Path Integral  393

So the integral is just the derivative! For multiple Grassmann variables, we define the integral
by iteration. The N -dimensional integral is defined by doingR
each one-dimensional integral
in turn, taking care to pick up minus signs as we move dηi through ηj . Obviously the
order in which we do the integrations matters to the overall sign. It is also obvious that the
only monomial in a general Grassmann function, which survives integration, is the last term
proportional to η1 . . . ηN .
We define the order by Z
dN η(ηa1 . . . ηaN ) = a1 ...aN . (19.83)

The most important Grassmann integral for quantum mechanical applications is the Gaussian
Z
1 aA b a
I[χ] = dN η e 2 η ab η
eχa η , (19.84)

where χa are an independent set of Grassmann variables, and A = −AT is an antisymmetric

matrix. Using the fact that Grassmann integrals are invariant under shifts of the variables,
and assuming N is even and det A 6= 0, we can write
1 −1 )ab
I[χ] = e 2 χa χb (A I[0], (19.85)
1
Z
I[0] = ( )N/2 dN η [η a1 η b1 Aa1 b1 . . . η aN/2 η bN/2 AaN/2 bN/2 ]. (19.86)
2
Doing the integral we get
1
I[0] = ( )N/2 a1 b1 ...aN/2 bN/2 Aa1 b1 . . . AaN/2 bN/2 ≡ Pf(A). (19.87)
2
The right-hand side is the definition of the Pfaffian of an even dimensional antisymmetric
matrix.
Now let us double the number of Grassmann variables by inserting an index ηia with
i = 1, 2 and write Z
1 a b
I [0] = d2N η e 2 ηi Aab ηi = Pf 2 (A).
2
(19.88)

Define the complex Grassmann numbers

1
ψ a = √ (η1a + iη2a ). (19.89)
2
1
ψ̄ a = √ (η1a − iη2a ). (19.90)
2
The Jacobian of the transformation is one so
Z
aA b
I 2 [0] = dN ψ dN ψ̄eψ̄ ab ψ
= det(A). (19.91)
394  Quantum Mechanics

The evaluation of the complex Grassmann integral is the same as that for two independent
real Grassmann integrals, as one can verify by elementary algebra. We have proven a cele-
brated theorem, namely that the square root of the determinant of an antisymmetric matrix
is a polynomial in its matrix elements.
We have also gotten the formula for complex Grassmann integration
Z
aM b a +ψ̄ χa −1 )ab
I[χ] = dN ψ dN ψ̄eψ̄ ab ψ
eχ̄a ψ a
= eχ̄a χb (M det(M ). (19.92)

This is just like the ordinary complex Gaussian integral except that we have the determinant
rather than its inverse. Note that we have written this formula for a general matrix M ,
because antisymmetry is no longer necessary.
To apply these formulae to QM, we have to have complex Grassmann functions of time
ψ(t), which requires an infinite dimensional Grassmann algebra.
∞
X
ψ(t) = ψn fn (t). (19.93)
n=0

fn (t) are some complete set of functions. The functional integral is

Z R
ds (χ̄(s)ψ a (s)+ψ̄a (s)χa (s)
I[χ(t)] = [dψ(t)][dψ̄(t)]eiS+ . (19.94)
where Z
S= ds[ψ̄i∂s ψ − h(ψ(s), ψ̄(s))]. (19.95)
To see the relationship between this formula and a quantum system, we consider a single
fermion creation–annihilation operator pair
[Ψ, Ψ† ]+ = 1; Ψ2 = Ψ† 2
= 0. (19.96)
We can realize this operator algebra on a Hilbert space whose kets are functions of a single
complex Grassmann variable ψ but with the rule that the corresponding bra is the complex
conjugate function of ψ̄. The Hilbert space is two-dimensional, since the most general function
is a + bψ. The operator Ψ acts as multiplication by ψ (by analogy with ordinary position
coordinates)
Ψ[a + bψ] = aψ, (19.97)
while Ψ† is d
dψ , that is
Ψ† [a + bψ] = b. (19.98)
Grassmann differentiation is easy, since the most general function of a finite number of
Grassmann variables is a polynomial. One only has to be careful about order, because, for
consistency we have to have
[∂ψa , ψ b ]+ = δab , (19.99)
 The Feynman Path Integral  395

if there are multiple Grassmann variables. So, for example, taking a derivative from the
left might give a different answer than taking it from the right. We always think about
differentiating from the left.
The scalar product in this Hilbert space has the form
Z
hf |gi = dψdψ̄µ(ψ, ψ̄)f¯(ψ̄)g(ψ). (19.100)

We choose the weight function µ so that the two functions 1 and ψ form an orthonormal
basis. Thus,
Z
dψdψ̄µ(ψ, ψ̄) = 1. (19.101)

Z
¯ = 1.
dψdψ̄µ(ψ, ψ̄)psiψ (19.102)

Z
dψdψ̄µ(ψ, ψ̄)ψ = 0. (19.103)

Z
dψdψ̄µ(ψ, ψ̄)ψ̄ = 0. (19.104)

Thus,
µ = 1 + ψ̄ψ = eψ̄ψ . (19.105)

Evaluating the time evolution operator over a sequence of infinitesimal intervals, for this
simple quantum system with Hamiltonian h(Ψ, Ψ† ), reproduces the path integral formula
above. The procedure generalizes easily to multiple fermions, Ψa .
In principle, we can use fermions to describe any system with a finite dimensional Hilbert
space. Simple embed the N -dimensional space into the smallest fermion Hilbert space with
2k > N . The N -dimensional subspace satisfies some linear constraints

Lp |si = 0, (19.106)

if and only if |si is in the subspace. Consider the full fermion system, but with a term

αp L†p Lp ,
X
δH = (19.107)

added to the Hamiltonian, with positive coefficients αp . When αp → ∞ only states in the
N -dimensional subspace survive as finite energy states.
396  Quantum Mechanics

19.9 FURTHER EXERCISES

19.4 Evaluate the infinite product
∞
!
Y ω2T 2 P∞ 2 2
ln(1+ ωn2 Tπ2 )
1+ 2 2 =e n=1 .
n=1
n π

19.5 Evaluate the path integral for the harmonic oscillator using Feynman’s original deriva-
tion.

19.6 Use our evaluation of hy|e−iHt/~ |xi for the harmonic oscillator, to compute the first
three Hermite polynomials.

19.7 Prove that the differences between different choices of contour for the Fourier transform
representation of the Green function obey the homogeneous equation.

19.8 Use the evaluation of the path integral for the harmonic oscillator coupled to a source,
to compute the higher time-ordered products of any number of X(t) operators.

19.9 Develop the path integral formalism for a Lagrangian of the form q̇i Mij (q)q̇j . Note that
you must choose an operator ordering prescription for the Hamiltonian, to define the
quantum theory.

19.10 Prove the Gelfand–Yaglom theorem:

2
det [ −d
dx2 + V (x)] u(1)
2 = .
det [ −d
dx2 ]
u0 (1)

u(x) is the solution of " #

−d2
+ V (x) v(s) = 0.
dx2
dv
with boundary conditions v(0) = 0, and dx (0) = 1. v0 (s) is the solution of the same
boundary value problem for the Hamiltonian with vanishing potential. In this problem
x ∈ [0, 1] and the eigenfunctions have Dirichlet boundary conditions ψ(0) = ψ(1) = 0.

a. The Riemann ζ function is defined by

∞
n−s ,
X
ζ(s) =
n=1
 The Feynman Path Integral  397

for values of s for which the sum converges, and then by analytic continuation to
the complex s plane. For a Hermitian operator with discrete spectrum, we define

ζM (s) = TrM −s .

Show, formally, that if the sums converge for some values of s, then
0
detM = e−ζM (0) .

b. Suppose that the eigenvalues of M are bounded from below. We can always add a
constant such that the bound is 0. Let f (λ) be a function with simple zeroes at the
eigenvalues λ = λn and nowhere else. Suppose further that f is analytic in the λ
plane with a cut along the negative real axis. Then the function L(λ) = dlnfdλ has
simple poles at the eigenvalues, with residue 1. By Cauchy’s theorem, we can write
1
Z
ζM (s) = dλλ−s L(λ),
2πi C

where the contour C starts at ∞ − i, encircles the origin and returns to ∞ + i.
Show that if one can neglect the contribution from a circle at infinity, one can move
the contour to encircle the cut of λ−s at s = 0 and then
sin(πs)
Z −∞
ζM (s) = λ−s L(λ),
π 0

so that
0 f (0)
−ζM (0) = ln .
f (−∞)
2
d
c. Let M = − dx 2 + V (x), with boundary conditions ψ(0) = ψ(1) = 0. Define the

function u(λ, x) to be the solution of

d2
[− + V (x)]u = λu,
dx2
with boundary condition u(λ, 0) = 0, u0 (λ, 0) = 1. This is like a classical mechanics
problem in a time-dependent harmonic oscillator and always has a solution. On the
other hand u(λ, 1) vanishes precisely at the eigenvalues λ = λn . For λ → −∞, the
potential becomes negligible, but u does not have a limit. This annoying problem
goes away if we divide by the determinant of the operator M0 , with V = 0. Thus
M u(0, 1)
det = = u(0, 1).
M0 u0 (0, 1)
This remarkable result can be generalized in a number of ways.
398  Quantum Mechanics

d. Verify the Gelfand–Yaglom theorem by direct computation of the spectrum in the

case where the potential is a constant.

19.11 Given 2n Hermitian operators satisfying

[γa , γb ] = δab ,

show that the complex linear combinations a1 = γ1 + iγ2 , a2 = γ3 + iγ4 , etc. satisfy the
algebra of fermionic creation and annihilation operators.
 CHAPTER 20

Quantum Computation?

20.1 INTRODUCTION
This chapter is meant as a very brief introduction to the ideas of theoretical quantum com-
putation. It is supposed to be enough to allow you to begin reading serious literature on the
subject [86]. The most important part of the subject of quantum computation is experimen-
tal and practical. If it proves impossible to build a machine that can perform large quantum
computations efficiently and reliably, then much of the theoretical work will prove useless. On
the other hand, theoretical quantum computer science is extremely important, both because
it has already shown that a quantum computer can, in principle, solve certain problems
much more efficiently than a classical computer, and because it provides clues (topological
quantum computing [87], etc.) to how one might build a real quantum computing device. In
addition, the theory of quantum computing sheds light on the peculiar properties of quan-
tum entanglement and has led to important developments (the theory of tensor networks
[88]) in the attempt to construct approximate ground state wave functions for complicated
condensed matter systems.
A modern digital computer stores data in physical systems that can reliably encode a
binary number with k digits. There are 2k such numbers, and k is called the number of bits
in the computer. A byte is a subsystem with k = 10 and the largest computers ever built
have of order 1015 bytes of Random Access Memory (the part of the machine on which active
computation takes place) and a total memory that is about 1018 bytes. So we are talking
about k ∼ 1018 at current technological limits.
Computation consists of transformations among these 2k numbers. Conceptually, the
simplest example is a program, which computes the values of some function i → f (i). The
art of writing computer code consists of figuring out efficient ways to convert the problem
you want to solve into a sequence of operations on binary numbers. Among those operations,
there are reversible ones, namely operations that take the list of numbers in some canonical
order into a permutation of that order.

399
400  Quantum Mechanics

In the first chapter of this book, we learned that any finite list of data can be viewed as a
list of the ortho-normal basis vectors of a finite dimensional Hilbert space. It should be clear
to someone who has read this book up to this point, that the Hilbert space of states of a
classical computer is a k-fold tensor product of two-dimensional Hilbert spaces. Choose the
(p)
classical computational basis to be the basis defined by the eigenvalues of σ3 in the p-th
factor of the tensor product, given some choice of ordering for the factors. The binary num-
ber representation is reproduced by writing the p-th digit of the number as the eigenvalue
(p)
1+σ
of 2 3 . Examples of reversible classical computational operations, permutations, are the
(p)
operators σ1 , which flip the p-th bit and leave the others alone. Computer scientists call
this operation NOT, and denote it by X. We will stick to the standard physics notation, but
if you want to read the quantum computation literature you will have to get used to their
notation and terminology.

20.2 QUANTUM INFORMATION

We will begin by discussing quantum information science, which is a collection of general
results about quantum mechanics (QM) that are useful in quantum computation, but have
more general interest. Recall that a general quantum state is given by a density matrix
ρ, which is a positive Hermitian operator whose eigenvalues sum to one. We will stick to
finite dimensional Hilbert spaces, so we do not have to worry about technicalities in defining
concepts like traces, tensor products, etc.
A density matrix is pure if ρ2 = ρ, which means that ρ is the projector on some specific
quantum state vector |si. In such a state, in a Hilbert space of dimension D, there are
D independent operators, whose value is predicted with certainty. These are ρ itself, and
D − 1 other one-dimensional projectors, on vectors orthogonal to |si. If the system is in an
impure state, then less is predicted with certainty. There is an interesting and extremely
deep connection between this notion of maximal information, and the entanglement of two
independent quantum systems.
Recall that we can think of the Hilbert space of a pair of independent systems as a tensor
product of the Hilbert spaces of the individual systems
H = HA ⊗ HB . (20.1)
In quantum information theory, we call a Hilbert space with such a preferred factorization a
bipartite system. Without loss of generality, we take the dimension of HA to be less than or
equal to that of HB . There is a special subgroup of unitary transformations on a bipartite
system, which has the form U = UA ⊗ UB . If the initial state of the system has the form
ρ = ρA ⊗ ρB , this will be preserved under conjugation by that subgroup. Such states are said
to be unentangled. Any vector in H can be written in terms of ortho-normal bases of the
individual spaces as
 Quantum Computation?  401
X
= CiJ |ai i|bJ i. (20.2)
iJ

The coefficient matrix C is generally rectangular, and satisfies

X
|CiJ |2 = 1. (20.3)
iJ

This can be read as the statement that the two positive Hermitian matrices ρA = CC † and
ρB = C † C, both have trace 1 and therefore define states on the individual systems. You will
prove in Exercise 20.1 that these states are pure only if

CiJ = ci dJ , (20.4)

which is to say that the systems are unentangled and the state on H is a tensor product.
These matrices are the matrices of density operators ρA,B in the bases which we used to
define the coefficients CiJ .
Entanglement thus implies that the probability of finding the A system in some particular
state, which can be determined experimentally by doing measurements of operators of the
form OA ⊗1, is correlated with the probability that the second system is in one of its particular
states J. The most famous such correlation is the one explored in the paper of Einstein Rosen
and Podolsky [30]. Entangled states can appear bizarre and nonlocal, because they can occur
even when the two systems are very far apart in space. If, however, we admit that the
entanglement occurred because of some interaction in the past, there is nothing that violates
causality in the existence of entanglement between distant objects. Indeed, the factorization
of unitary operators, U = UA ⊗ UB which preserves lack of entanglement is precisely what
we would expect from a Hamiltonian evolution of a system whose Hamiltonian had the form

H = HA ⊗ 1 + 1 ⊗ HB , (20.5)

characteristic of two noninteracting systems. More general time evolution will create entan-
glement for an initially unentangled state. Of course, we cannot rule out the possibility that
someone started the universe off in a state which had entanglement between distant degrees
of freedom but one usually assumes that any entanglement revealed by an experiment has a
causal explanation.
In a bipartite system, we can always predict the probabilities of single system measure-
ments in terms of the reduced density operator ρA of that system. This is unlikely to be pure,
and we can define a measure of its purity called the Shannon–von Neumann entropy

Tr[ρnA ]
S ≡ −Tr[ρA ln ρA ] = limn→1 = . (20.6)
1−n
402  Quantum Mechanics

For n 6= 1 and integer, the quantities on the right-hand side are called Renyi entropies. The
Shannon–Von Neuman entropy is always nonnegative, and vanishes only when the state is
pure. It depends only on the spectrum of the density operator pi . It is also bounded from above
by the entropy of the maximally mixed (also called maximally uncertain) density operator
ρmax ≡ D1A 1, which is the logarithm of the dimension of the Hilbert space HA . Finally, the
entropy is a convex function of the density operator. Given a finite set of density operators
P
ρi , one can form a new one via ρ = ti ρi where ti are nonnegative numbers summing to
one. Then S(ρ) ≥ ti S(ρi ), as you will prove in Exercise 20.2. You will also prove (Exercise
P

20.3) that the entropies of the two reduced density matrices derived from a pure state in a
bipartite system are equal. These two equal entropies of a pure state in a bipartite system
are called the entanglement entropy of one part of the system with the other.
Given a density operator ρA on HA , we can ask the question: is it possible to find a pure
state ρ in a larger bipartite system, such that ρA is the reduced density matrix. Consider
any Hilbert space HB with dimension larger than or equal to that of HA . Write the spectral
decomposition X
ρA = pi Pi , (20.7)
where Pi is a complete set of commuting one-dimensional projectors on states |ai i. Now let
|bi i be a basis of a subspace of HB whose dimension is equal to DA . Consider the pure state
X
|Si = αi |ai i ⊗ |bi i. (20.8)
i

Then the reduced density operator in HA is

X
ρA = |αi |2 Pi , (20.9)
i

1/2
so we need to only choose αi to have absolute value pi . This procedure is called the Schmidt
decomposition, and the resulting pure state is called a purification of the original state. It is
clear that the purification is highly nonunique. We can perform any unitary transformation
in HB on the states |bi i and get a new purification, or choose any other subspace of the larger
space with the same dimension.

20.3 PAGE’S THEOREM REDUX, MONOGAMY, AND CLONING

We have already encountered entanglement in both our discussions of measurement theory
and of statistical equilibrium. Let us first recall Page’s theorem, from the latter discussion.
The context of the theorem is a bipartite system with DB  DA . The reduced density
∗
matrix ρA = CiJ CJi can then be analyzed statistically. Let us choose the basis in which
ρA is diagonal. The DA diagonal matrix elements pi of ρA are each written as a sum of a
squares of DB complex numbers subject to the constraint iJ |CiJ |2 = 1. This constraint is
P
 Quantum Computation?  403

invariant under unitary transformations in the group U (DA )⊗U (DB ) and in particular under
permutations in the indices i. Therefore, if we choose the complex coefficients randomly, then
on average, we will have all√pi equal to 1/DA . The root mean square deviation of each pi
from the average will be 1/ DB , so if DA  DB we get the maximally uncertain density
matrix with very high probability.
In general, any purification of the maximally mixed density operator will have the form
1 X
|P uri = √ ( |ai i ⊗ |bi i), (20.10)
DA i

where we have chosen some orthonormal basis of a DA -dimensional subspace of HB . For any
such state, we say that system A is maximally entangled with system B. The significance of
this phrase is that a measurement of |bi ihbi | will automatically tell us what state system A is
in, even though system B and system A might be separated by a million light years. As we
have emphasized repeatedly, no real information transfer is involved here. We have simply
prepared a quantum state in which the probability of system A being in any of the states
|ai i is equal, but is correlated exactly with the probability that system B is in |bi i. There is
nothing spooky or nonlocal about this, if we can explain the entanglement by some earlier
causal contact between the two systems.
Maximal entanglement is monogamous. Suppose we ask that the system is simultaneously
maximally entangled with two different bases of HB . Then
X X
|ai i ⊗ |bi i = |ai i ⊗ |ci i. (20.11)
i i

Applying the projection operator |ak ihak | ⊗ 1 to this equation we get

|ak i ⊗ |bk i = |ak i ⊗ |ck i, (20.12)

for each k, which means that |bk i = |ck i. The monogamy of maximal entanglement is very
important to the problem of quantum encryption of data, to be discussed below.
Another important application of these ideas is the so-called no cloning theorem. The
question is the following. Given a system A in state |ai can we allow it to interact with
another system in such a way that system, B carries away with it a clone of |ai? Translated
into mathematics, this is the question of whether there exists a unitary transformation in a
bipartite system, which will transform the state |ai ⊗ |bi into |ai ⊗ |ai. Since either of these
two states is a normalized vector in the full bipartite Hilbert space, we can obviously do this
for any given state |ai. The no-cloning theorem refers to the fact that it is impossible to find
a single unitary transformation, which will perform this operation for every state |ai. Thus,
we ask whether
|ai i ⊗ |ai i = U (|ai i ⊗ |ei), (20.13)
404  Quantum Mechanics

for all |ai i. Consider the scalar product of two different states on the right-hand side of these
equations, before the action of U . Since |ei is normalized, this is just ha1 |a2 i in the Hilbert
space HA . Since U is unitary, this is the same as the scalar product on the left-hand side,
so that

ha1 |a2 i2 = ha1 |a2 i. (20.14)

Thus, the claim that one can clone two states with the same unitary is possibly valid only if
the two states are the same, or if they are orthogonal. Thus, one could clone all states in a
given orthonormal basis, but not superpositions of them. This last sentence is the statement
that in a classical theory, where one forbids superpositions of the preferred basis, one can
clone arbitrary states with the same “machine.” The nonclassical principle of superposition
defeats this attempt.

20.4 QUANTUM KEY DISTRIBUTION

While we have generally tried to steer away from discussing human intervention in QM, the
subject of quantum encryption is all about human agents trying to send messages to each
other, and trying to protect them from being read by an enemy. We will therefore follow the
practice in the field and assume that Alice wants to send a secret message to Bob. A standard
practice in cryptography is the one time pad. Secret messages are encoded by constructing a
map between the clear version of the message, written in some well-known human or computer
language, and an encoded version. The map is constructed with the use of a key, some string
of bits which, in the simplest version of encryption tells us which symbol to substitute for
each symbol in the clear message. If Alice uses the same key multiple times, her encryption
is vulnerable to analysis of letter frequencies, etc., so if multiple messages are to be sent,
she wants the ability to make frequent switches of the key. However, in attempting to send
the multiple keys to the recipient, Alice runs the risk of interception of her key distribution
messages by an eavesdropper, Eve.
Alice could encode her key distribution message as a set of bits in a classical computer,
which we represent by introducing N spins. A message consists of a choice of the state in the
2N -dimensional Hilbert space of the system. N will be taken fairly large, in order to make
statistical reasoning about the results of quantum measurements a reliable tool.
The problem we are trying to solve is to discern the possibility that, in the process
of transmitting the bit string from Alice to Bob, it was intercepted by the enemy, Eve.
With the usual procedures of cryptography, there is really no way to tell if a message has
been intercepted. If, however, we could transmit quantum states, the situation is completely
different. The point is that a quantum state gives only a probability for a given outcome,
unless one knows in advance which complete set of commuting operators is diagonal in that
 Quantum Computation?  405

state. For purposes of encryption, it is sufficient, at least to illustrate the point, to consider
only the operators σ1,3 (i) where i denotes the position along the bit string.
The first step of the encryption procedure is for Alice to choose at random, for each
(i) whether she will encode the bit value 1 as σ3 (i) = 1 or σ1 (i) = 1. Her message is then
encoded in the state of a one-dimensional quantum spin chain. The next step is the part
for which no technology yet exists, except for very small numbers of Q-bits. Alice sends a
physical system prepared in the quantum state of the spin chain she has chosen, to Bob,
without losing any quantum information. One can imagine a quantum wire, which sends N
electrons to Bob, with each electron in a spin state along either the 1 or 3 axes, according to
the random choice made by Alice, with the sign of the spin encoding the message. Bob does
not yet have a message. If he chooses his own random sequence of σ3 (i) or σ1 (i) measuring
devices, he has equal probability of reading the wrong message as the right one, every time
his random choice fails to coincide with that made by Alice.
Now Alice gets on the telephone and reveals (perhaps to Eve as well, if Eve has Alice’s
phone tapped) to Bob what her choice of axes was. It is important to realize that at this
point, the Q-bits have already been sent. If Eve has not already intercepted them, she can no
longer do so. They are sitting safely in Bob’s ultrasecure laboratory. Bob discards those bits
of the message where his axis differed from Alice’s, because there is a 50% probability that
he read that part of the message incorrectly, since the quantity he measured was actually
uncertain in the state sent by Alice. At this point, Alice and Bob publicly compare half of
the bits where their axes coincided.
If Eve did not intercept the message, they should find perfect agreement. If Eve did
intercept the message, she did so by choosing her own random sequence of axes and measuring
either σ3 (i) or σ1 (i) for each bit. In doing so, she changed the quantum state of the system.
To say this more precisely, she entangled the system with her spin measuring devices, which
are macroscopic. The quantum state, conditioned on the fact that those devices read anything
at all, is different from the state sent by Alice. This is the quantum state that Bob actually
queried with his spin measuring devices. Consider those bits where Bob and Alice made the
same choice of axes. For each of those bits, Eve had 50% probability of making the same
choice. When Alice and Bob compare the values they assigned to the bits where their axes
agreed, they will get the wrong answer about 50% of the time, and they will know that Eve
has intercepted the message. Presumably, in this case, Alice would just try again. None of
the textbooks on the subject tells us what to do if Eve intercepts every message. Quantum
encryption can only alert us to the fact that our messages are being read, not get them
through untainted.
However, once Alice and Bob know that Eve has not intercepted the message, Alice can
share the bits of the message that Bob did not get yet, through normal classical channels.
He has learned the value of roughly 1/4 of the bits in the key, randomly distributed in the
message, and Alice and Bob know that Eve can have no clue what those bits were. Even if
406  Quantum Mechanics

she Eve-sdrops on the classical transmission of the remaining data, she is unlikely to learn
N
Alice and Bob’s secret key. The probability that she does guess it is 2− 4 , even assuming
that she has monitored Alice’s classical communication of N/4 bits to Bob. The key, which
is 2N/2 bits long, can now be used to send secure messages.
The bottleneck in the actual implementation of this algorithm is sending the
quantum information from Alice to Bob. The simplest practical implementation of such
a quantum channel is sending free photons through space, or a fiber-optic cable. Coherent
quantum states of photon polarization can be sent over distances of order 102 km. The most
commonly discussed technique for sending quantum information is the quantum teleportation
protocol, to which we now turn.

20.5 QUANTUM TELEPORTATION

The subject of this section has an unfortunate name, since it does not really involve trans-
portation of localized objects from one place to another, and certainly does not involve faster
than light communication. Quantum teleportation is a method for transferring information
about the quantum state of a system, to a remote location. Let us imagine that it is the
state of some two state system, and label a particular basis in its Hilbert space by the binary
numbers |0i and |1i. We assume the state of the system is

|si = α|0i + β|1i. (20.15)

Assume this system is in Alice’s laboratory.

The first step in the quantum teleportation communication protocol has to occur before
the actual teleportation begins. One prepares a pair of spin one half particles, or photons, in
for example the state
1
|B1 iAB = √ (|+iA ⊗ |+iB + (|−iA ⊗ |−iB ). (20.16)
2
The ± signs represent polarization along the three direction, while the subscripts refer to
Alice and Bob, because this part of the protocol also sends one of the entangled particles
to Alice and the other to Bob. In Exercise 20.4, you will verify that this state is part of an
orthonormal basis given by
1
|B2 iAB = √ (|+iA ⊗ |+iB − (|−iA ⊗ |−iB ). (20.17)
2
1
|B3 iAB = √ (|+iA ⊗ |−iB + (|−iA ⊗ |+iB ). (20.18)
2
1
|B4 iAB = √ (|+iA ⊗ |−iB − (|−iA ⊗ |+iB ). (20.19)
2
 Quantum Computation?  407

This is called the Bell basis. Alice and Bob have received the information about the preferred
axis, 3, and the choice of the Bell vector |B1 iAB , from the agent who created and sent the
entangled pair. They each carefully store their precious entangled particle in a way that
prevents it from interacting with anything else.
Now the state of the system in Alice’s laboratory
|si ⊗ |B1 iAB , (20.20)
which consists of the two state system one hopes to communicate, and one of the particles
in the Bell pair, is entangled with the system in Bob’s laboratory (the other particle in the
Bell pair). The actual teleportation consists of two steps. First Alice uses an experimental
apparatus, whose pointer positions are designed to become entangled with the states in the
Bell basis in the tensor product of the state of her spin one half particle and that of the two
state system.
To understand what happens, we should simply expand the premeasurement state in that
basis
1
|si ⊗ |B1 iAB = [|B1 iAC ⊗ (α|+iB + β|−iB ) + |B2 iAC ⊗ (α|+iB − β|−iB )+ (20.21)
2
|B3 iAC ⊗ (β|+iB + α|−iB ) + |B4 iAC ⊗ (β|+iB − α|−iB ). (20.22)
This is the same state we had before, just written in another basis. What is interesting
about it is that if Alice can construct a macroscopic device, which is entangled with the Bell
basis, then for each choice of Bell state, the state of Bob’s particle has the same quantum
information as the quantum state |si of the original system. Indeed, for |B1 iAC the coefficients
of |±iB are the same as those of |0/1i in |si. For other members of the Bell basis, the state
of Bob’s particle is a given unitary transformation on α|+iB + β|−iB .
Recall what it means for Alice to do a measurement. The quantum state of the three
body system is just a probability distribution for all normal operators on the full Hilbert
space. The measurement entangles each component of Alice’s two body subsystem, in the
Bell basis, with the state of one of the macroscopic pointers on Alice’s machine. As always, we
can now apply Bayes’ conditional probability rule to collapse the quantum state to the one
corresponding to what actually happened to the macroscopic needles in a particular run of the
experiment. One can then communicate information about the result of this measurement to
Bob, using macroscopic apparatus, which follows almost deterministic laws starting from the
initial condition of what Alice’s needle registered. This violates unitarity, since Bayes’ rule
always requires us to renormalize probabilities to eliminate those predictions that “did not
happen.” This is why the quantum teleportation scheme can appear to violate the no-cloning
theorem.
Note that independently of which Bell state Alice’s measurement finds, Bob can clone the
quantum state |si in his lab, once he receives Alice’s classical communication. If Alice finds
408  Quantum Mechanics

|B1 iAC , Bob just has |si, whereas if she finds one of the other Bell states, he knows precisely
which unitary operation he must perform on the state of his particle, in order to retrieve
|si. Actually, it is more proper to say that Bob has cloned only the quantum information in
|si since we have taken pains to say that |si was the state of some two state system, not
necessarily that of a spin one half particle identical to the one in Bob’s laboratory. One of the
most successful recent demonstrations of quantum teleportation is [89]. Interested readers
should note the way in which this experiment transfers quantum information back and forth
between nuclear spins, electron spins, and photons, as well as the delicate control needed to
keep the quantum coherence of the system.

20.6 GATES FOR QUANTUM COMPUTERS

Computer scientists use the term gates for operations which change the state of a computer.
For quantum computer scientists, gates are unitary transformations. We have already dis-
cussed the NOT gate which changes the state of the i-th bit by applying the operator σ1 (i).
(i) (i)
1+σ (j) 1−σ
Our next example is the controlled NOT, or c-NOT operation. This is Cij = 2 3 σ1 + 2 3 .
In writing this formula, we have used a compressed notation for operators in a tensor prod-
uct space, which is the notation used in lattice spin systems. We write only the parts of the
operator which act nontrivially on some of the bits, leaving implicit the tensor product with
the unit matrix acting on the rest of the bits. An operator is called p−local if it acts in a
nontrivial way on exactly p bits. Note that C 2 = 1, and C = C † , so that c-NOT is a unitary
and therefore reversible operation. A more compact notation for Cij is
(i) (j) (i)
Cij = P+ σ1 + P− , (20.23)
(i)
where P± are the projectors on the subspaces where σ3 (i) is ±1.
An extremely important operation in computer science is the swap operation Sij ,
which exchanges the contents of the i-th and j-th bits. That is, it takes | . . . 0 . . . 1 . . .i to
| . . . 1 . . . 0 . . .i, for any values of the bits represented by . . .. In Exercise 20.5, you will verify
by operator multiplication that
Sij = Cij Cji Cij . (20.24)
In words, suppose the j-bit is 0, then after the operation of Cij the two-bit system is in
the state |11i or |00i depending on the initial state of the i-th bit. Let us choose the case
|11i. Now the operation Cji changes this to |01i and the second application of Cij leaves this
state intact. Similarly, if the i-th bit were 0, then all three c-NOT operations leave the state
invariant, so we again get the effect of the swap. The same thing happens if the j-th bit is
1. Note that the operators Si,i+1 are just transpositions, and the theory of the permutation
group shows that the transpositions generate all permutations, so the Sij operations form an
overcomplete set of generators.
 Quantum Computation?  409

Another important operator in quantum computation is the Walsh–Hadamard operator,

for which QC theorists have unfortunately chosen the symbol H (i) . This clash with the
standard QM notation for Hamiltonian is mitigated by the fact that in QC one always deals
with discrete unitary evolution, so there are no Hamiltonians. As its symbol indicates, H (i)
acts on the Hilbert space of a single bit, as follows
1 (i) (i)
H (i) = √ (σ1 + σ3 ). (20.25)
2

From the point of view of spin, H is just the Pauli matrix in the direction π/4 in the 1–3 plane.
This means that it takes the state of a single bit in the computational ( i.e., σ3 diagonal)
basis into a superposition of states where the bit is 1 and 0. In fact, those superpositions
are just the eigenstates of σ1 (i), and H (i) is just the unitary transformation between the
eigenbases of σ3 (i) and σ1 (i).
At this point, we should introduce some conventional QC language. A single bit, which
can only be in the 1 or 0 states is called a C-bit, while if we allow superpositions, it is
called a Q-bit. Obviously, from the point of view of Hilbert space, we are talking about
the same two-dimensional space. Insisting on using only C-bits is analogous to our classical
theory of the ammonia molecule. The difference between the classical and quantum theories
of ammonia is whether we allow time evolution to change a C-bit into a more general pair
of orthonormal vectors in Hilbert space. The difference between a quantum and classical
computer program is similarly whether we force the discrete transformation between different
states of the computer to be in the permutation subgroup S2k or allow more general unitary
transformations. H (i) is not allowed in a classical program, but note that

1 P−(i) P+(i)
! ! !
1 H (i) H (i) 1 H (i) H (i)
H (i) H (j) Cij H (i) H (j) = . (20.26)
2 H (i) −H (i) 2 P+(i) P−(i) 2 H (i) −H (i)

Carrying out the rightmost matrix multiplication, we get

(i) (i)
! !
(i) (j) (i) (j) 1 H (i) H (i) σ− − 1 σ+ + 1
H H Cij H H = √ (i) , (20.27)
2 2 H (i) −H (i) (i)
σ+ + 1 1 − σ−

where σ± are the spin raising and lowering operators. It is easy to see that
(i) (i) (i)
Hi σ± = P∓ ± σ± . (20.28)

Using these results, you will verify in Exercise 20.6 that

H (i) H (j) Cij H (i) H (j) = Cji . (20.29)

410  Quantum Mechanics

Switching the control and target bits of a c-NOT operation is an important device in com-
puting. Using only classical operations one can accomplish the same goal by conjugating with
the swap operator Sij . However this is a 2 Q-bit operation which is not a product of two
single Q-bit operations. In hypothetical physical realizations of quantum computers, just as
in classical computers, operations which act on a larger number of bits are more difficult to
construct. The reason for this is locality. The bits, whether classical or quantum, are arranged
on a lattice in some solid. Local Hamiltonians will typically do controlled things only to one
bit, or two nearest neighbor bits.
QC scientists use the term N–Q–bit gate to describe a unitary transformation on N Q-
bits, which is to say an element of U (2N ). Technologically feasible quantum computers will
have to have only one and two Q-bit gates, and programs should minimize the number of
two Q-bit gates they use, in order to be efficient. A general one cubit gate has the form
eiα (n0 + n · σ), where (n0 , n) is a real unit four vector. It has four real parameters. A general
2-Q-bit gate contains 16 real parameters. Now consider the three classes of gates

Ui , Uj , Vi Cij Vj . (20.30)

Clearly the first two classes are independent of each other, and of the third. Suppose that for
two different choices of Vi and Vj the corresponding operators in the third class are equal.

Vi Cij Vj = Wi Cij Wj . (20.31)

Then there is some pair of single Q-bit unitaries Ui = Wi−1 Vi and Tj = Vj Wj−1 , such that

Ui Cij Uj = Cij . (20.32)

We can write this in 2 × 2 block form as

! ! !
u11 u12 σ1 T 0 σ1 0
= . (20.33)
u21 u22 0 T 0 1

We have omitted the j label on σ1 and T . From this equation, we immediately see that
u12 = u21 = 0, and unitarity then implies that uii = eiαi . But then we also have eα1 σ1 T = σ1
and eα2 T = 1. Thus, T is proportional to the unit matrix, α1 = α2 , and the phase of T is the
opposite of this one. So the only duplication is the transformation that multiplies one Q-bit
by a phase and the other by the opposite phase. We conclude that any two Q-bit gate can
be obtained by concatenating the c-NOT operation with one Q-bit gates.

20.7 COMPUTATIONAL COMPLEXITY

Turing’s mathematical model of a classical computer consists of a transformation in the
Hilbert space of N Q-bits, which takes an initial state into a desired final state, using only
 Quantum Computation?  411

elements of an S2N subgroup of U (2N ). Without loss of generality, we can work in the basis
in which the initial state is a tensor product of N single Q-bit states, each of which is
an eigenstate of σ3 (i). In quantum computing language, the set of all such states is called
the computational basis. The S2N subgroup transforms each computational basis state into
another, permuting the elements of the basis. For practical reasons, one wants to have each
computational step involve only a small number of Q-bits. Turing showed that any element
of S2N could be built by concatenating the single Q-bit operations σ1 (i) and the two Q-bit
operations Sij , as we have discussed above.
We can think of a state in the computational basis as a binary number, with the value
of the projection operator on σ3 (i) = 1 giving the i-th digit of the number. A classical
computation consists of computing the value of some function f (i) for each i. If the computer
has N bits, we are restricted to calculating functions whose values are less than or equal
to the binary number 1 . . . 1 = 2N + 2N −1 + · · · + 1. If the function is invertible, then it
corresponds to some S2N transformation on the N Q-bit Hilbert space. Each permutation
has a characteristic size, viz. the minimal number of 1 or 2 Q-bit gates that it takes to
build that transformation. The running time for a classical computer, which performs its
operations serially, to perform the computation i → f (i) is proportional to the size of the
corresponding S2N transformation.
A parallel computer is a device which can perform some number k of 1 or 2 Q-bit trans-
formations simultaneously on k disjoint pairs of Q-bits. In this case, we define the depth of
a computation as the number of such simultaneous transformations it takes to perform the
computation. The computational complexity of a problem is the depth of the computation
required to solve it. For any given problem, the computational complexity is a fixed number.
However, many problems have a natural scale to them, which can be increased either without
limit, or up to some huge value. For example, we can ask, given some integer M , what its
prime factors are. We would like to know how the computational complexity of this problem
scales with M . Problems whose complexity scales like M p for some fixed p and M → ∞
are called Polynomial or said to belong to class P, whereas those which depend on M expo-
nentially belong to class NP. Obviously, if one hopes to solve a problem for some very large
value of M , one hopes that it is in class P. Conversely, if one is trying to encrypt data with
a code that is hard to crack, one encrypts it using an algorithm that is in NP. It turns out
that for a classical computer, the factorization problem is in class NP. Much of the interest
in quantum computing comes from the fact that some problems, the factorization problem
among them, which are in class NP for a classical computer, are in class P for a quantum
computer.
The complexity of a quantum computational problem is defined in a manner completely
analogous to the classical case. A quantum computation is a unitary transformation in U (2N ).
Since unitary transformations depend on continuous parameters, we are unlikely to be able to
construct one exactly by performing any finite number of one and two Q-bit gates. However,
412  Quantum Mechanics

all we really need to do is to get close to the required unitary. Closeness can be defined in
terms of the quantities
p
1
||U − V ||p ≡ Tr [(U − V )† (U − V )] 2 p
, (20.34)
C
with p = 1 being the usual choice. We insist that ||(U − Ugate )||1 be less than .
One can show that for almost all choices of U , which is to say, with probability one
for a unitary picked from the uniform probability distribution on the space of all unitary
transformations U (2M ), the computational complexity is bounded by
1 1
22M ln ( ) < C < 22M [ln ( )]c , (20.35)
 
where 1 < c < 2 (the exact value of c is unknown). So most quantum computations are
exponentially hard. However, there are some exponentially hard classical computations, which
can be performed in polynomial time on a quantum computer. Since a full exposition of
quantum computer science would require another book the size of this one, and since a
number of excellent books on the subject exist already, we will just illustrate this with a
simple example, known as the Deutsch–Josza (DJ) problem [90].
The DJ problem is one of a class of black box problems, in which one is given a function
f defined on M Q-bits, with certain properties, and asked to verify whether it satisfies some
other property Z. The function is implemented by some program, called the oracle and an
evaluation of it on some particular Q-bit is called a query. The object of the game is to
minimize the number of queries of the oracle one needs, in order to determine whether Z is
true or false. For the DJ problem, the function f is known to map every input state in the
computational basis of the M Q-bit Hilbert space to either 0 or 1, and we are told further
that either it is constant or it takes the value 0 on half the states in the computational basis
and 1 on the other half. In the second case, we are not told the specific states on which it is
zero.
Classically, we would have to call the function 2M −1 +1 times to determine whether or not
it was constant. This is the only way to be sure that the function is or is not constant. If we
make only k calls to the function, then we will fail to produce the right answer with probability
1
≤ 2k−1 . DJ produced a quantum algorithm, which determines the answer definitively with
order M queries.
We use a Hilbert space with M + 1 Q-bits, the last one serving to encode the action of
the function f . We can think of the function f as an operator, Pf , on the Hilbert space of
M Q-bits, diagonal in the computational basis. It is a projection operator on that subspace
where the function f has the value 1. Define an operator in the M + 1 Q-bit space by
Uf = Pf ⊗ σ1 (M + 1) + (1 − Pf ) ⊗ 1. (20.36)
This operator is unitary.
 Quantum Computation?  413

Now consider an input state |s0 i = |+, +, . . . +, −i3 , in the Hilbert space where the first
M Q-bits have σ3 = +1, and the last one has σ3 (M + 1) = −1. Act on |s0 i with the tensor
product of all the Hadamard operators √12 (σ1 (i) + σ3 (i)). When the Hadamard operator acts
on the σ3 = ±1 state it gives the eigenstate σ1 = ±1, so the result of multiplying by all the
H(i) is
|s1 i = |+, +, . . . +, −i1 . (20.37)
Since σ1 (M + 1)|s1 i = −|s1 i, we have

Uf |s1 i = (1 − 2Pf )|s1 i. (20.38)

Note that the operator (1 − 2P ) acts like the unit operator on the last Q-bit.
The state |s1 i is a superposition of all possible eigenstates of σ3 (i), 1 ≤ i ≤ M , with
equal amplitude. If f is a constant function, Pf is just the unit operator. If we now act again
with the product of all Hadamard operators, the state we get is just −|s0 i, so if we measure
the projector on |s0 i, we will get 1. However, if P vanishes on half the joint eigenstates of all
the σ3 (i), this is no longer true. Now, we want to evaluate

hs0 | ⊗i H(i)(1 − 2Pf )|s1 i = hs1 |(1 − 2Pf )|s1 i = 0. (20.39)

Indeed, since |s1 i is an equal amplitude superposition of all the joint eigenstates, and
(1 − 2Pf ) = −1 on half of these eigenstates and +1 on the other half, we get “destruc-
tive interference.”
In carrying out the computation, we have used M +1 gates to construct the state |s1 i, and
“queried the oracle” Uf one time. This represents an exponential speedup of the computation,
relative to a classical computer. Note also that despite the fact that we are using QM, the
final answer is definite. The states Uf |s1 i and |s1 i coincide for the constant function and
are orthogonal for any projector of rank 2M −1 . The Deutsch–Josza algorithm does not in
itself have any practical utility, but it provided Shor [91] with vital clues for constructing an
algorithm for factoring large numbers in polynomial time. Modern methods of encryption of
financial data and military secrets depend on the exponential difficulty of the factorization
problem for classical computers. If a practical quantum computer is ever developed, all data
encrypted by these methods is vulnerable. Quantum computer scientists assure us that if this
ever happens, they will be ready with quantum encryption algorithms, which are impervious
to quantum computation.

20.8 CAN WE BUILD A QUANTUM COMPUTER?

A useful quantum computer must be able to store large amounts of information. We noted
in the beginning that the best current supercomputers have 1018 Q-bits, so a quantum com-
puter with the same storage capacity would be like a macroscopic system. In fact, all current
414  Quantum Mechanics

approaches to real quantum computers involve very large devices, much larger than compara-
ble classical computers. The point is that robust transistors, the bits of a classical computer,
can be made from the collective coordinates of collections of ∼ 106 atoms. However, precisely
because these coordinates are robust, they cannot encode quantum information.
Among the variables that have been proposed for the Q-bits of a quantum computer
are the spins of ions in an ion trap, electron spins, energies or positions in a quantum dot,
and magnetic fluxes in superconductors. As of the date of this paragraph, the number of
Q-bits that have been successfully used for a computation is less than 102 , using any of these
technologies. By comparison, a single computer chip has > 109 transistors. The problem is
the fragility of quantum information. The microscopic states of real substances are typically
changing very rapidly and are entangled with their environment, so that quantum information
is lost. Elaborate theoretical protocols, called Quantum Error Correcting Codes, have shown
that it is theoretically possible to make robust quantum states of a set of Q-bits, by entangling
them with a much larger system, in such a way that most local changes of the state of the
large system will not affect the state of the small Code Subspace. Practical implementations
of these protocols have, so far, not succeeded in getting anywhere close to the number of
Q-bits one would need for a genuine computation.
Quantum computation is thus, as of this writing, a purely theoretical subject. Some
references about the state of the art of building a real quantum computer can be found
here [92].

20.9 EXERCISES
20.1 Show that Sij = Cij Cji Cij , by using the rules for multiplying Pauli matrices. Remember
(i) (j)
that [σa , σb ] = 0 and explain why this is so.
(i) (j)
20.2 Show that Sij = 12 (1 + σa σa ). This relation was originally discovered by Pauli, study-
ing the spin states of two electrons. Sij is called the Pauli Spin Exchange operator in
that context.

20.3 Given a pure state in a bipartite system, prove that the von Neumann entropies of both
reduced density matrices ρA and ρB are equal.

20.4 Prove that the state

1
|B1 iAB = √ (|+iA ⊗ |+iB + (|−iA ⊗ |−iB )
2

has norm 1 and is orthogonal to all of the other vectors in the Bell Basis. Prove that
all of those vectors are orthonormal.
 Quantum Computation?  415

20.5 Prove that

H (i) H (j) Cij H (i) H (j) = Cji .

20.6 The problem of determining whether a large integer N is factorizable into primes is
extremely difficult. Any classical algorithm for solving it requires a number of oper-
ations growing like ecN . This is why the problem is used in RSA encryption. Shor’s
demonstration that this problem can be solved in polynomial time on a hypothetical
quantum computer is one of the most important results in the field. Explaining Shor’s
algorithm would require us to learn too much number theory. The number theory can be
found in many places, among them Mermin’s book on quantum computing. The main
point of it is that the problem can be reduced to finding the periods of a function defined
on binary numbers that are N bits long. If we write the binary number as (x1 , . . . , xN )
where each xi can be either zero or one, then a function is a rule that maps each
N -vector x to another N -vector f (x).1 We can add two such vectors by adding their
components, using mod 2 arithmetic, and define their scalar product by x · y = xi yi ,
P

again mod 2. A period of a function f is a vector a such that f (x + a) = f (x) for every
vector a. Simon’s Algorithm is the task of finding what period of a function f is, given
that one knows such a period exists. Classically, this problem is exponentially hard and
is the core of the factoring problem. Simon showed that there is a quantum algorithm
for solving it in a time linear in N , with probability that goes to 1 as N gets large.
To start our exploration of Simon’s algorithm, introduce a Hilbert space of dimension
22N . Show that a pair of binary vectors x, y corresponds to an orthonormal basis vector
|x, yi in this Hilbert space, according to the rule
i
σ3 (i)|x, yi = (−1)x ; 1 ≤ i ≤ N,
i
σ3 (i + N )|x, yi = (−1)y ; 1 ≤ i ≤ N.
We will always call unit vectors in the Hilbert space states, in order not to confuse them
with vectors of binary numbers. The notation xi denotes the i component of a vector
of binary numbers. Show that the mapping

U (f )|x, yi = |x, f (x) + yi

is a unitary operator on the Hilbert space.

20.7 Now, start with the state |0, 0i for which all of the σ3 (i) operators have eigenvalue 1.
Act on this state with the unitary
1
Mathematically sophisticated readers will have noted that the space of bit strings is really a module over
the ring Z2 rather than a vector space. This does not change the argument. The fact that the sum of two
period vectors is zero is an indication of that fact.
416  Quantum Mechanics

N N
Y 1 Y 1
W = [ √ (σ3 (i) + σ1 (i))]U (f ) [ √ (σ3 (i) + σ1 (i))].
i=1 2 i=1 2
To do this, show that
N
1
[ √ (σ3 (i) + σ1 (i))]|x, yi = 2−N/2
Y X
(−1)xż |z, yi,
i=1 2 z

where the sum is over all possible N bit strings z. Applying W we then get
N
Y 1 X
W |0, 0i = [ √ (σ3 (i) + σ1 (i))]U (f ) (−1)0ż |z, 0i
i=1 2 z
N
Y 1 X
= [ √ (σ3 (i) + σ1 (i))] |z, f (z)i
i=1 2 z
XX
= (−1)zẏ |y, f (z)i.
z y

20.8 Given that we have prepared our system in the quantum state W |0, 0i, we now imagine
entangling the subsystem in the first tensor factor, with macroscopic collective coordi-
nates, so that a fixed value of the collective coordinates corresponds to a fixed value of
y. The remaining subsystem, the second tensor factor, is now in the quantum state
X
(−1)z·y |y, f (z)i.
z

Now use the fact that f is periodic and that the period bit string a satisfies a + a = 0
because of mod 2 arithmetic. Show that this implies that this quantum state is zero
(which means that it had zero probability in the premeasurement state) unless y · a = 0.
That is to say, any measurement in the computational basis will return a value that is
orthogonal to the period vector.
20.9 Now count the number of steps in the above procedure. Argue that implementation of
W takes a number of elementary operations that scales like N . It is important that
we assume the function is given to us as a table of values. It takes no computational
work to evaluate it. The key observation is that acting with N individual σ3 (i) + σ1 (i)
automatically gives us a sum over all the bit strings. So we get a sum over 2N things,
by doing N operations. Now imagine we have done of order kN measurements, with
k > 1 but independent of N . In each measurement we have found a vector in the N -
dimensional space of bit strings that is orthogonal to the period vector. Argue that this
means it is very unlikely that we have not found the period vector.
 CHAPTER 21

L’Envoi: Relativistic Quantum

Field Theory

This book has been about quantum mechanical models compatible with the Galilean
Principle of Relativity (though we usually worked in a fixed inertial frame), and we know that
this must be replaced by Einstein’s Special Theory of Relativity if we want results accurate
to better than v/c where v is the speed of the fastest particle participating in the process
under study. c is of course the speed of light, c = 3 × 108 m/s. In this short chapter, we want
to indicate the remarkable complications that result from insisting that quantum mechanics
(QM) be invariant under Lorentz transformations.
Since Special Relativity introduces a fundamental constant with dimensions of velocity, it
is reasonable to define a time coordinate X0 = ct with dimensions of space, and a four vector
Xµ , which indicates the position of a particle in space-time. The most concise statement of
special relativity is that physical properties involved in traveling between two space-time
points should depend only on the interval

I = (X0 − Y0 )2 − (X − Y)2 ≡ (X − Y )2 . (21.1)

A fixed interval is the locus of points on a hyperboloid in space-time, and Lorentz transfor-
mations are just the hyperbolic rotations that keep the shape of hyperboloids fixed.
If (X − Y )2 > 0, the hyperboloid has two sheets, with opposite signs for X0 − Y0 . Such
an interval is called time-like and Lorentz transformations cannot change which point is in
the future. This behavior persists as I → 0 through positive values. However, if I is negative
then time order is not Lorentz invariant, and detectors traveling at different velocities will
not agree on the causal order of X and Y , when they are space-like separated. Classically
this seems OK. No system traveling at speeds below the speed of light can travel between
space-like separated points.

417
418  Quantum Mechanics

Now let us repeat our quantum mechanical calculation of the probability amplitude for
a free particle to travel between X and Y . Recall that the key to it is that the Hamiltonian
is just a function of momentum. The relativistic relation between energy and momentum is

P02 = P2 + m2 , (21.2)

where we have set the speed of light equal to one so that energy, mass and momentum all have
the same dimensions. As long as m is real, this is the equation for a time-like hyperboloid
(this is the reason we called the energy P0 ), so the sign of P0 can be taken positive. Systems
with energy that is not bounded from below tend to be unstable to the slightest perturbation.
We solve for the time evolution of quantum amplitudes by inserting a complete set of
momentum eigenstates. The only new thing is that we have to integrate over spatial wave
numbers with the Lorentz invariant volume element
!
4 m2 2 d3 k
d k θ(k0 )δ k − 2 = . (21.3)
~ 2k0
~ ~
m = mc = lc (m) p
is called the Compton wave-length of the particle. k0 (m) is a frequency,
given by k0 (m) = k2 + lc (m)−2 . The propagation amplitude is
d3 k
Z
A(x, y) = e−ik(y−x) . (21.4)
2k0 (m)(2π)3
The frequency k0 is positive, so this looks like a properly causal amplitude if y is in the
future of x. It is also Lorentz invariant. The only problem is that it is nonzero when y − x is
a space-like vector.
To see this note that for a space-like vector, we can always find a Lorentz frame where
the two points have the same time coordinate. In this frame,
d3 k
Z
A(x, y) = e−ik·(y−x) . (21.5)
2k0 (m)(2π)3
This is a Bessel function. It vanishes exponentially when the spatial separation is larger than
lc but is never exactly zero. To avoid violating causality, we have to introduce a new principle:
any system that can emit a particle at a point is also capable of absorbing the particle, and
vice versa. We can then write a causally sensible amplitude
d3 k d3 k
Z Z
A(x, y) = θ(y0 − x0 ) e−ik(y−x) + θ(x0 − y0 ) eik(y−x) . (21.6)
2k0 (m)(2π)3 2k0 (m)(2π)3
The step functions in time are Lorentz invariant if x − y is time-like or null, and when it is
space-like the two integrals are identical so the step functions sum to one. Our new principle
enables us to write an amplitude that is both Lorentz invariant and causal.
 L’Envoi: Relativistic Quantum Field Theory  419

Clever readers will call a halt: Wait a minute! What if the particle carried a Lorentz
invariant quantum number like electric charge? Then we could tell where the particle was
emitted or absorbed by examining the charge balance. This is not a bug, but a feature of our
hypothesis: it is the prediction of antiparticles! An antiparticle has exactly the same mass but
opposite values of all conserved Lorentz invariant charges, as the particle. Needless to say,
this prediction has been spectacularly confirmed by experiment, and the equality of masses
of particle and antiparticle (and lifetimes for decaying particles) has been verified to many
significant digits.
The same sort of argument shows that any mechanism that can scatter particles can create
particle antiparticle pairs. Figure 21.1 shows a particle being scattered from an external field.
The amplitude is nonzero when the causal order of the points has the field operating in a
region to the past of both particles. This is creation of a pair. The arrow follows the charge
flow and one can see that the second particle is an antiparticle.
Finally, though we will not be able to prove it here, Lorentz invariance, QM and causality
imply the spin statistics connection (−1)F = e2πiJ3 . We can see a hint of this by thinking of
the amplitude A(x, y) on a plane of simultaneity of two space-like separated points, as the
amplitude for two particles to be present at the two points. It is symmetric under interchange.
We assumed that the particle state was characterized only by its momentum, which means
that its spin was zero. It is straightforward, though a bit formal and tedious, to generalize
this argument to spinning particles, and one finds that the spin statistics connection just
falls out.
One of the most important consequences of these observations is that one must use
quantum field theory to describe relativistic QM. From the point of view of this book, this
remark is not such a big deal. We have advocated the field point of view because it explains
the reason behind particle statistics (which is otherwise an additional postulate tacked on
to the rules of QM), and illuminates the meaning of wave particle duality and of the fact
that the single particle Schrödinger wave function obeys the same equations as a classical

Particle anti-particle production goes into particle scattering under a Lorentz

Figure 21.1
transformation, when points are at space-like separation.
420  Quantum Mechanics

field. However, if the total number of particles is conserved, we could always revert to a
multiparticle Schrödinger wave function. It is occasionally useful to do this, but it is simply
impossible to do it in the relativistic theory, because the number of particles can change once
the energy is larger than the mass.
For example, in the nonrelativistic theory, we can make the initial probability of finding
a particle at position x as narrow a distribution as we wish. If the particle is free, this
distribution will spread, but we can start it as narrowly peaked as we like. In relativistic
quantum field theory, this is not possible. We can make the wave packet narrow by allowing
it to have very high probability at high momentum. But momentum carries energy. Roughly
speaking, if we try to localize a particle in a region smaller than its Compton wave length, by
doing position measurements, then we have a probability of inserting enough energy into that
region to create particle antiparticle pairs. The whole concept of a single particle localized at
a point in space-time no longer has meaning. Any sufficiently localized excitation has some
probability to be a multiparticle state, and the particle number becomes uncertain.
Relativistic quantum field theory has given rise to a model of the world, the Standard
Model of Particle Physics, which explains every experimental result we have obtained, some-
times with shocking precision. For example, the agreement between theory and experiment
for the magnetic momentum of the electron is accurate to one part in 109 . The only gross
phenomena that are not explained by this model are the nature of the gravitational force,
the nature of the dark matter we have observed to permeate the cosmos, and the reason why
the universe contains more baryons than antibaryons.1 There are plausible extensions of the
standard model, still based on quantum field theory, which explain the last two items as well.
Indeed, there are so many of them that we do not know which is correct, if any.
The relationship between Einstein’s theory of gravity and the quantum theory has not
yet been elucidated. Einstein’s field equations resemble the field equations of the standard
model, so one’s initial temptation was to view it as just another quantum field theory. There
are many reasons why this point of view is incorrect. We will mention only the one that
is least well understood, but perhaps holds the clue to the quantum theory of gravity: the
connection between geometry and entropy. In quantum field theory, the dimension of the
Hilbert space associated with a finite region of space-time is always infinite. However, if we
treat the expectation values of the energy and momentum densities of a typical state in this
Hilbert space as sources for the gravitational field, then most of the states lead to a distortion
of space-time geometry called a black hole. The boundary of the black hole is a null surface,
a surface along which light waves can propagate (Figure 21.2). In ordinary flat space-time, if
we follow such a surface into the future, then the cross sectional area grows at later and later
times. For the surface of a black hole, the area remains constant; A = 16π(GM )2 , where G
is Newton’s gravitational constant and M the mass of the black hole.

1
If this were not true, we would not be here to speculate about why it is true.
 L’Envoi: Relativistic Quantum Field Theory  421

on
riz
ho
Figure 21.2The Penrose space-time diagram of a black hole. The tangents to particle trajec-
tories must make and an angle < π/4 with the vertical axis.

When Planck introduced his famous constant ~, he pointed out that when combined
with the speed of light and Newton’s gravitational constant G, it led to a complete system
of natural units. In particular, there is a natural length scale, the Planck length, given by
G~
L2P = ∼ 10−66 cm.2 (21.7)
c3
Bekenstein and Hawking [93] put forward persuasive arguments that a black hole is a thermal
system, with an entropy S = 4LA2 , where A is the area defined above.
P
A causal diamond in space-time is defined by a trajectory that is always pointed toward
the future and two points along that trajectory. The diamond is the region of space-time
to which one can send a signal traveling at speed ≤ c, and receive a reflection back, in the
time interval between the two points. Each such diamond has an area associated with it: the
maximal area on its null boundary. In 1995, Jacobson [94] showed that if one applies the first
law of thermodynamics, dE = T dS, locally to every causal diamond in space-time, with the
entropy expressed as 1/4 of the area in Planck units, then one derives Einstein’s equations
relating the geometry of space-time to the energy and momentum densities of matter.2
There is one more clue to the quantum theory of gravity that can be gleaned from
rather general considerations. If we consider two causal diamonds in a space-time, then it
might be the case that there are no trajectories traveling at a speed ≤ c, which connect
any points in one diamond, to any point in another. Two such diamonds are said to be
at space-like separation. Any operator localized in one diamond, had better commute with
any operator localized in the other. If this were not the case, then an experiment done in
one diamond, could affect the outcome of an experiment done in the other, even though
nothing physical could connect them. The “spooky nonlocality” of QM would not be just the
2
Jacobson did not phrase his argument in terms of causal diamonds, but this formulation is equivalent
to his original one. It should also be noted that Jacobson’s derivation does not capture the cosmological
term in Einstein’s equations. This should be viewed as telling us that the cosmological term is not a local
energy density.
422  Quantum Mechanics

result of mistakenly interpreting probability distributions as physical fields, but would lead
to measureable consequences.
String Theory [95] and the Anti-de Sitter Space Conformal Field Theory Correspondence
[96], derived from String Theory, are the only mathematically well formulated models of
quantum gravity, which exist at the present time. These models apply to space-times with a
cosmological constant Λ ≤ 0, which have asymptotic boundaries containing causal diamonds
of infinite area. None of these models appears to apply to the world we inhabit. Furthermore,
the operators that one is allowed to consider measureable in these models are localized on
the boundaries at infinity and do not tell us directly how to interpret the measurements we
actually do within finite area diamonds. Thus, the challenge of constructing models of QM,
which are compatible with the cosmology we inhabit is still before us. Perhaps some of the
readers of this book will be able to contribute to the task of meeting that challenge.
In the meantime, there is a wealth of QM to be learned, which applies directly to the
earth we live on and the substances inhabiting it. Hopefully, this book has helped to give
you a step in the right direction toward learning that material and applying it.
 APPENDIX A

Interpretations of Quantum
Mechanics

The interpretation of quantum mechanics (QM) championed in this book is an intrinsically

statistical one. Einstein, the most famous critic of QM, was completely satisfied with such
an interpretation, but as an aficionado of classical statistical mechanics, he felt that there
had to be a nonstatistical “reality” underlying QM, which evolved in a deterministic manner,
such that the quantum statistics was obtained by averaging over some unmeasured variables.
The inventors of the Copenhagen Interpretation disagreed with Einstein, but IMHO made
several mistakes in at least their presentation of the disagreement.1 The Copenhagen crowd
insisted vehemently, and correctly, that the wave function (really the density operator derived
from it) was a complete description of a quantum system—there were no hidden variables.
Whatever they meant by this, it is been taken by many modern authors to mean that the
wave function is a “real thing” and that we can take it to be a description of what is “actually
going on” in any given run of an experiment. This is true both for authors who accept the
conventional (Copenhagen plus decoherence) interpretation of QM and those who propose
alternatives.
There is nothing in the actual practice of QM that requires or justifies such an interpre-
tation. The wave function is used as a probability distribution and checks of the predictions
it makes are performed by doing multiple experiments and using the frequentist rule. No one
has ever proposed an experiment or a set of simultaneous experiments, which can measure
the wave function without doing such repetition. Indeed, the EPR argument and those of

1
I am not an historian of science, and have not studied the old literature, so I am not in a position to
defend statements about who said what when. My knowledge of the positions of Heisenberg, Bohr, Born,
Jordan, etc. is mostly limited to the presentations of them in textbooks and modern popular accounts.

423
424  Quantum Mechanics

Bell show that if one could make such a measurement, it would imply that signals could
transfer information faster than the speed of light.
Let us recall Feynman’s discussion of the double slit experiment in order to emphasize
this fact. There are two important parts of that discussion, which we should recall. The first
is that every actual experimental result consists of a click, when some particular detector at
some point of the screen is struck by a particle. The second occurs somewhat later on, when
a picture of an interference pattern is shown, and we a see a phenomenon we have come
to associate with actual physical waves. However, there is no experiment, which actually
detects that wave pattern. Rather, one collects many clicks, and uses the relative frequency
of clicks at different places, via the frequentist rule, to compare the probabilistic predictions
of QM to the statistics of many experiments. The only time we actually see physical waves
corresponding to solutions of the Schrödinger equation is when we look at coherent states of
bosons. We explained all of that in Chapter 5.
Thus, the first challenge one would issue to anyone who interprets the wave function as
some kind of physical phenomenon, like a water wave or an electromagnetic wave, is to pro-
pose a single experiment, done one time, which measures all aspects of the wave function. No
mathematical formula in QM tells us how to perform, even in principle, the measurement of
the wave function at a given time. By contrast, QM does tell us how to measure, in principle,
any set of commuting normal operators at a single time. In order to do so one must couple
the system of interest to a large macroscopic system in such a way that distinct eigenstates
of the set, get entangled with states of the macrosystem labeled by distinct values of the
macrosystem’s collective coordinates. Because the quantum probabilities for the coupled sys-
tem now satisfy Bayes’ rule with accuracy exponential in the entropy of the macrosystem, we
can state that the value of the commuting set of operators at the time of the measurement
becomes part of a decoherent history. As long as the macrosystem continues to exist2 then
the predictions of QM for the future of this particular experiment are identical to those of a
theory which attributed the unpredictable behavior of the macrosystem’s detectables to our
failure to measure some classical hidden variables. In that second kind of theory we would
discard part of the probability distribution which predicted other outcomes, using Bayes’
rule to “update the probability distribution based on new data about what is really going
on.” We do this in QM by “collapsing the wave function” since that is a procedure consistent
with linearity of the Schrödinger equation. As we have emphasized in Chapter 10, this is
a convenient approximation for following the evolution of the system, conditioned on the
assumption that a series of macroscopic systems carry imprints of the original experiment.
Once all macroscopic records of that particular experiment have disappeared, one must, if one
wishes to make correct statistical predictions about the future, go back to the wave function

2
or, if it imprints its history on another macrosystem before it disappears,
 Interpretations of Quantum Mechanics  425

that is a superposition of all possible outcomes of that experiment, because all traces of the
decoherent history have disappeared.
Before summarizing some of the attempts to come up with alternative interpretations
of QM, or alternatives to QM that give rise to “objective wave function collapse,” it is
appropriate to comment on an attitude found all too common among working physicists.
That is, the belief that one can simultaneously adhere to the “Copenhagen interpretation”
of QM and believe that the wave function is a physical thing, rather than a probability
distribution. Presumably the reason for this is that one can find such statements in writings
of some of the inventors of the Copenhagen interpretation. This is an inconsistent point of
view. The only consistent way to interpret the Copenhagen view of QM is the statistical
viewpoint advocated in this book: QM is a new kind of probability theory in which not
all variables can have definite values at the same time. If the equations of motion relate
such mutually indefinite variables, then there are in general no probabilities for histories.
Probabilities for histories arise as approximate descriptions of macrosystems.

A.1 MODAL INTERPRETATION OF QUANTUM MECHANICS

Let us begin our discussion of realist interpretations of QM with a class of interpretations
that appear to do the least damage, both to the formalism, and to our desire to have the
history of the universe be describable by quantities that have definite values at all times.
This vast list of ideas, comes under the name of modal interpretations of QM. We will not
attempt to study all of those proposals.
A simple modal interpretation is based on the assumption that there is some principle
that restricts the initial state of the universe to be some element of a particular privileged
orthonormal basis |ri i (where the letter r for reality is chosen intentionally). If Ri are the
Ht Ht
projection operators on these states, then the Heisenberg operators ei ~ Ri e−i ~ have a time
dependent joint probability distribution, which is determined by the initial probability that
the universe was in one of the states |ri i. In a modal interpretation these are considered
probabilities for histories of variables that are diagonal in the |ri i basis. The probability of
being in the state |ii at time t, given that one was in state j at time t0 is
H
P (i, j; t − t0 ) = |hri |e−i ~ (t−t0 ) |rj i|2 = |U (ri , rj , t − t0 )|2 . (A.1)
Note that the quantum amplitudes U for these transitions have phases eiθ(i,j,t−t0 ) , which are
not determined by these probabilities.
This theory of probabilities for histories of the “real” quantities Ri is not in general
deterministic. That is, even if we begin in a pure state, with only one of the Rj nonzero,
the probability distribution P (i, j; t − t0 ) will be nonvanishing for more than one value of i.
Proponents of a realist interpretation of QM are not bothered by this, and there is no rea-
son why they should be. The results of experiments on microscopic systems show without a
426  Quantum Mechanics

doubt that such systems display randomness. One might want to attribute that randomness
to unmeasured deterministic “hidden variables,” but there is nothing apart from the philo-
sophical stance of 18th century mathematical physicists that makes determinism a sacred
principle.
Similarly, the insistence that the initial conditions of the universe be restricted to elements
of a particular orthonormal basis, and not superpositions of them, is not a source for worry.
If we are talking about the dynamics of a small subsystem of the universe, experiment shows
us that we can prepare it in any superposition of its quantum states. The initial conditions
for the entire universe are not under our control, and we do not yet have a complete theory
of the earliest moments of cosmic history. It might be that such a theory will explain the
restriction on initial conditions
What is disturbing about an interpretation that takes the probabilities for histories of the
specified operators Ri as the underlying reality of quantum theory is that these probabilities
do not determine all other predictions of the quantum theory, because of the undetermined
phases. Furthermore, the probabilities for histories do not obey an intuitively obvious rule,
which one might have thought was required of any realist description. This is the rule
X
P (i, j; t − t0 ) = P (history), (A.2)
histories

where we sum over all histories consistent with the initial and final conditions
The idea behind a realist description is that what actually happens as the universe evolves
is some particular history of the real variables. That is, they have values, even if nothing
intervenes to measure those values. Statistics should enter into the theory only as a restriction
on our ability to predict what those values will be, given their initial values. In our modal
model, we have a probability distribution for such histories. In a realist interpretation, one
would think that one should be able to say that the probability that the universe was in state
|ri i at time t was the sum of the probabilities of all possible histories that had this property.
After all, as realists we must believe that what is actually going on in the world we observe
is some particular history of the real variables ri . If we had not motivated the choice of basis
in our model by an initial condition at the beginning of the universe, we could derive this
sum rule from the frequentist definition of probability, plus the assumption that every re-run
of the world was just a particular history of the ri .
In standard quantum theory, one can compute the probability amplitude for some history
in which at times tk the system was in the state |rk i, given that it was in state |rj i at t = t0 .
It is
A(rk , tk ; rj , t0 ) = U (rN , rN −1 ; tN − tN −1 ) . . . U (r1 , rj ; t1 − t0 ). (A.3)
In the standard interpretation of QM, we consider the absolute square of this number to be
the probability that the system was in state |rk i at time tk , i.e. the probability of a given
history.
 Interpretations of Quantum Mechanics  427

Anyone who’s gotten to this point in the book, will know that in general, the probabilities
for histories defined in this way will not (in general) satisfy the realist sum rule. Rather, by
the linearity of the Schrödinger equation, we must sum up the amplitudes for the different
histories and take the absolute square of the resulting sum. We have gone to great lengths to
argue that if we take variables which are collective coordinates of macroscopic objects, then
the probabilities for their histories do satisfy the sum rule with accuracy exponential in the
size of the object. Part of the demonstration involved acknowledging that these collective
variables gave only a coarse grained description of the objects. They are not candidates
for the variables rk , which are supposed to be a complete basis for the Hilbert space at a
microscopic level.
It appears that there is only one way out of the paradox that our set of probabilities
for histories do not satisfy the sum rule. Namely, we should restrict attention to the special
class of quantum systems where this problem does not arise. For a system with a finite
number of states, this is discrete time dynamics with unitary transformations restricted to
the permutation subgroup SM which transforms every element in the |rk i basis into |rP (k) i,
where the overused letter P is here being made to stand for permutation. For such systems,
the values of the operators Ri are definite for all times and the dynamics is deterministic.
The system is a cellular automaton.
’t Hooft has made the remarkable conjecture [97] that the quantum world we observe
might be described by a model like this. ’t Hooft’s formalism accepts the general quantum
framework in which there are mutually incompatible observables. His conjecture is that the
variables we measure, which show all of the randomness and uncertainty associated with
QM, are simply operators in the Hilbert space of the cellular automaton, which do not
commute with the fundamental operators Ri . The success of the conventional QM description
of matter does not contradict this conjecture if one can show that the action of conventional
Hamiltonians can be realized as an approximation to an evolution under the symmetric
group, perhaps in a system larger than that we associate with conventional theories. For
example, one might attempt to show that some kind of cutoff on quantum field theory could
be embedded in a model with SN dynamics. ’t Hooft has shown [98] that this is indeed
possible for fields satisfying the equation
(∂t ± ∂x )φ = 0, (A.4)
and has tried to connect this result to superstring theory [95], which is a proposed unified
theory of particle interactions and gravitation. Obviously, a thorough investigation of these
claims is far beyond the scope of the present book.
’t Hooft’s approach raises fascinating mathematical questions about which quantum sys-
tems can be well approximated by SN dynamics (perhaps of a larger system). Even if these
questions can be answered in a way that is favorable to ’t Hooft’s program, it is clear that
the proposal does not resolve a number of issues that one might have expected a realist
428  Quantum Mechanics

interpretation of QM to resolve. First of all, like other modal interpretations of QM, the
dynamics of the real variables Ri does not determine the evolution of all quantum operators
in the system. If VD is a unitary operator diagonal in the |ri i basis, then we can replace the
permutation evolution operator UP by

UP →→ VD UP . (A.5)

The Heisenberg operators

t t
Ri (t) = (VD UP )† τ Ri (VD UP ) τ , (A.6)
where τ is the underlying discrete time interval, and t is an integer multiple of it, are equal to
t t
†
Ri (t) = UP τ Ri UPτ . (A.7)

On the other hand, VD does effect the time development of operators which are not diagonal
in the |ri i basis. This ambiguity is essentially the same as the phase ambiguity that we
pointed out above.
The consequence of this ambiguity, which is quite general to modal interpretations, is
that the histories of the “real” variables, do not determine the dynamics of the variables we
actually measure. This observation lends a hollow ring to the claim that the “real” variables
are the underlying reality behind QM randomness. Recall our discussion of the Koopman
model of classical mechanics as QM in Chapter 2. There we pointed out that the quantum
rules reduced to those of classical statistical mechanics for systems with a “Liouville” Hamil-
tonian, if we restrict attention to operators that commute with the commuting phase space
variables p, q. A fancy mathematical way of imposing this restriction is precisely to insist
that the transformations
ψ(p, q) → eiθ(p,q) ψ(p, q), (A.8)
which commute with the Liouville Hamiltonian, are gauge transformations (i.e. redundan-
cies), so that only operators that commute with p and q are physical. A corresponding gauge
principle for ’t Hooft’s model of QM would defeat its purpose. We need the noncommuting
variables to reproduce the experimental successes of the standard interpretation of QM.
Most modal interpretations of QM do not accept the quantum predictions for the prob-
abilities of histories of the ri as part of their definition of reality. Rather, they invent a
stochastic dynamics for the ri , which is designed to reproduce only the probability distri-
butions Tr (ρ(t)Ri ), where ρ(t) is the quantum density matrix. Modal interpreters of QM
accept that the usual quantum predictions for probabilities of histories do represent what will
happen to macroscopic systems whose collective coordinates become completely entangled
with (some subset of) the |ri i3 and attribute the disagreement with the “more fundamental”
3
This is usually expressed in phrases like “this is what will happen as the result of measurements.”
 Interpretations of Quantum Mechanics  429

stochastic probabilities to the “complications of the measurement process.” People who are
comfortable with a theory in which one is unable to probe the most fundamental aspects of a
theory with experiments, will be satisfied with such an approach. However, there is no clear
agreement on a “correct” modal interpretation of QM, and widespread agreement among
advocates that the last word has not yet been said on this subject.
For the present author, the most satisfactory version of modal QM is the one advocated
by ’t Hooft, because it retains the measurable probabilities for histories of the ri as the
fundamental definition of reality, and insists on dynamics which preserves the probability sum
rule for real histories. The ’t Hooft interpretation should be supplemented with a convincing
principle that determines the unitary evolution operator UD , and should realize the goal of
’t Hooft’s program, which is to show that a theory of the world we observe can be derived
from deterministic dynamics.
Even if those goals are realized, there are two further challenges, of perhaps a more
philosophical nature, that must be faced by any interpretation of the modal sort. The notion
of “physical reality” that one tries to recover in modal interpretations of QM is based on our
intuitive understanding of the behavior of the collective coordinates of macroscopic objects.
It seems a bit at odds with the basic philosophical posture of modal interpretations, to derive
that behavior as an approximation to the nonintuitive quantum behavior of operators that do
not commute with the Ri . At the very least, one would want to show that the commutators
[Ri (t), CK (t)] were very small for all i and all collective coordinates of all macroscopic objects.
Further one would want to show that the experimental absence of interference between
different histories of collective coordinates followed not from the kinds of considerations
we presented in Chapter 10, but from the fact that a history of the macroscopic world was
in one to one correspondence with a history of the ri (t).
Remarkably, the phenomenon of unhappening, if it could ever be observed experimentally,
would provide a sharp test of any interpretation of QM, which claimed that real histories of
collective coordinates were in one to one correspondence with real histories of some complete
set of microscopic commuting quantities. In such an interpretation of QM, unhappening
should not occur. Perhaps a clever experimentalist can come up with a testable version of
the exploding laboratory thought experiment, and see whether unhappening really occurs in
nature, as the standard formulation of QM leads one to expect. If such an experiment were
to succeed, one would be, at the very least, forced to conclude that the reality we perceive
in the behavior of macroscopic objects cannot be connected directly to the “real” variables
in any modal interpretation of QM.
The final challenge to any modal interpretation of QM comes from one of the founders of
the subject, J.S. Bell, in his famous generalization of the arguments of Einstein–Podolsky–
Rosen and Bohm. Bell emphasized that the disturbing question posed by the success of QM,
was whether or not one could attribute values to all of the potentially measurable quanti-
ties in a quantum system in histories of the system in which some of them were not measured.
430  Quantum Mechanics

He showed that for simple and experimentally accessible systems, such an assumption con-
tradicted the rules of QM, unless the hidden real variables ri could influence the behavior
of locally measurable quantities, which were at space-like separation. We have reviewed this
argument in Chapter 10. While it is clear that the variables ri (t) of modal interpretations of
QM are not locally related to any measurable properties of particles or fields, there have been
no demonstrations of how such interpretations resolve Bell’s conundrum. Do they provide
histories in which all measurable quantities have definite values, and if so, how? Do they allow
for nonlocal (i.e. faster than light) signaling? One doubts that a “realist” interpretation of
QM, which did not provide definitive answers to these questions would have satisfied Bell or
Einstein.

A.2 BOHMIAN MECHANICS

The earliest attempt to come up with a realist interpretation of QM is the Pilot Wave theory
of De Broglie [99] and Bohm [100]. Its modern incarnation is called Bohmian Mechanics
[101]. Its most basic tenet is that the wave function ψ(Qi , t) is a “real” field in the mul-
tidimensional configuration space of the system. Discrete quantum numbers, like spin, are
treated by introducing multiplets of fields,4 transforming in various representations of sym-
metry groups. One is supposed to consider a fixed history of the world to be determined by a
fixed configuration of this field, developing through time. The field satisfies the Schrödinger
equation. For simplicity, we will ignore spin and other indices in the following
S
Given such a complex field ψ = Aei ~ , with A ≥ 0, one can always introduce a set of
trajectories in the configuration space, via the equation

Q̇i (t) = ∇i S, (A.9)

although this equation is problematic in the vicinity of points where ψ = 0. The Qi (t) are
called the trajectories of Bohmian particles. It would be a mistake to identify these with actual
localized particles in space. Indeed, in the most straightforward extension of this formalism
to quantum field theory [100], the Bohmian particle is really a Bohmian field configuration.
Despite this, much of the literature on Bohmian mechanics is full of phrases conflating the
notion of Bohmian particle with actual particles. Note in particular that for a given history
of the wave function, at any time, there are Bohmian particles going through all points of
space, where ψ 6= 0. Points with ψ = 0 are problematic because the phase S is not defined
there. Since the standard theory of the Schrödinger equation allows us to contemplate initial

4
When multiparticle systems are considered, the tensor product spin indices are appended in a manner
which appears more ad hoc.
 Interpretations of Quantum Mechanics  431

conditions which are smooth functions of compact support5 one must make some sort of rule,
which prevents Bohmian particles from being in regions with ψ = 0.
One way to deal with trajectories that run through all points where ψ 6= 0 is to
assume that the initial conditions for the variables Qi (t) are uncertain, with a probabil-
ity distribution P0 (Q(0)) that vanishes wherever ρ ≡ ψ ∗ ψ(Q(0)) does. The two quantities
P (Q, t) ≡ P0 (Q(t))J(Q(t), Q(0)) and ρ satisfy the same continuity equation:

∂t P = −∇i [P ∇i S]. (A.10)

∂t ρ = −∇i [ρ∇i S]. (A.11)

Here J(Q(t), Q(0)) is the Jacobian of the mapping from the initial values Q(0) to the values
of Q(t) at time t. For ρ, this equation follows from the Schrödinger equation, while for P it
follows from simply solving the differential equation for Q(t) and taking account of the fact
that P is a probability density on configuration space and that the mapping Q(0) → Q(t)
does not usually preserve the Euclidean volume element dN Q.
If we choose an initial probability distribution, which is nonvanishing at points where
the ψ = 0, then the equations of Bohmian mechanics are problematic; the initial velocities
of the Bohmian particles are ill defined. The only apparent way to ensure that this never
happens, is to insist that the probability distribution is a function P (Q) = f (ρ(Q)), where
f is a function which vanishes sufficiently rapidly at ρ = 0. Since P and ρ satisfy the same
linear partial differential equation, this relation can be preserved in time only if, in fact,
P = cρ where c is a constant. Since the Schrödinger equation is linear, we can always choose
a solution that satisfies the probabilistic normalization condition dN Qρ = 1, and conclude
R

that for this solution, c = 1.

To complete a closed set of equations, we should also record the equation for S, which
follows from the Schrödinger equation:

~2 ∇i ∇i ρ
~∂t S = [(∇i S)2 − ] + V (x). (A.12)
2m ρ

If we write the dimensionless quantity S = ~I then ~ appears only in the term involving ρ. If
we neglect this term, we find that I satisfies the classical Hamilton–Jacobi equation, and this
sheds some light on the “reality” of the wave function. In classical Hamilton–Jacobi theory, we
do not consider the classical action I to be a physical field on configuration space. The classical
equations of motion have a plethora of solutions, which depend on initial conditions for the
canonical variables Qi and Pi . The solution of the H–J equation is a device for discussing
all of these solutions at once. Any given situation, in which something definitely happens is
5 ∇2 ψ
Indeed, smoothness is too strong a constraint. We need only require that ψ
have the same degree of
differentiability as the potential.
432  Quantum Mechanics

described by a single trajectory Qi (t) and only involves the variation of the H–J function S,
along that trajectory. One can use the full H–J solution, over all of configuration space, in a
probabilistic situation, where we specify only a distribution for the initial conditions Qi (0).
Then the combination of the H–J equation and the continuity equation for probability tells
us how to solve completely for the evolution of the probability distribution.
In the quantum version of these equations, probability distributions which are sharp
in the Qi (0) give singular contributions to the equations of motion for S. In the linear
Schrödinger equation, a singular initial condition ψ ∝ δ N (Qi (0) − Qi0 ) generates some linear
combination of the retarded or advanced Green functions, which have nonnormalizable but
spread out probability distributions. Thus, the idea of a nonprobabilistic, purely mechanical
interpretation of “Bohmian mechanics,” does not make sense.
The fact that the “quantum Hamilton–Jacobi equation” for S depends nonlinearly on the
probability distribution P = ρ, implies that there is no classical stochastic interpretation of
these equations. By a classical stochastic system we mean equations of the form

Q̇i (t) = f i (t;Q), (A.13)

where the initial conditions Qi (0) and perhaps some coefficients in the forcing functions f i
are random variables chosen from some probability distribution. If the correlations in time
of the random elements in the forcing function are delta functions, then one obtains a local
linear Fokker–Planck equation for the probability distribution P (Q, t).6 Probabilities obeying
such an equation always satisfy Bayes’ conditional probability rule, which means that one
can write the probabilities as sums over probabilities of histories. In the nomenclature used
by many philosophers of QM, and popularized by J.S. Bell, equations like this obey the
principle of “counterfactual definiteness.” The probabilities defined by Bohmian mechanics
do not satisfy this rule, which is not surprising, because they are equal to the probabilities
computed in QM. Mathematically, within the formalism of Bohmian mechanics, the failure
of Bayes’ rule arises from the terms in the evolution equation for S, which are nonlinear in ρ.
If instead of adopting a nonclassical probability interpretation of the equations, we insist
on a stance that the wave function is a “real” field on configuration space, much as the
classical electric and magnetic fields are real fields in space, then the interpretation of the
Bohmian trajectories must be completely different. They are the analog of the Madelung
trajectories of hydrodynamics [102]. If a particle is dropped into a fluid obeying e.g. the
Navier–Stokes equation, it will “surf the wave” along one of these trajectories, with the
initial condition determined by the time and place the particle is dropped in. In this case ρ
is interpreted as the fluid density and has nothing to do with the particle. Although Bohm–
Madelung trajectories exist throughout the configuration space, only the one picked out by
some deus ex machina (DEM) has a particle on it. In this interpretation of the equations,
6
If the time correlations are not delta functions, the resulting Fokker–Planck equation is nonlocal in time.
 Interpretations of Quantum Mechanics  433

there is nothing probabilistic, and no real connection between the wave function’s history,
and that of the particle. One cannot, in such an interpretation, associate the motion of the
Madelung surfer and the measured positions of particles. The former is a deterministic single
trajectory in configuration space, determined by a choice of initial condition, while the latter
are random variables, about which the theory only makes statistical predictions.
One can try to make up an origins story to restore the agreement between this inter-
pretation of the Bohmian equations and experimental facts. At some initial time, the DEM
drops particles at the initial points of Bohmian trajectories, at random. We have seen that
the only consistent way of doing this is to force the initial probability distribution chosen
by the DEM to be the absolute square of the wave function. Thus, we are led back to the
intrinsically probabilistic interpretation of the equations, in which the soi disant probabilities
for histories of the Bohmian particles do not satisfy Bayes rule.
Bohmian mechanics differs from generic modal interpretations of QM, in that it makes
full use of the quantum phase information in defining the histories of its modal variables,
so that those histories determine expectation values of all operators. However, the price to
be paid is that it does not define probabilities for histories which satisfy the realist sum
rule. One cannot imagine that every actual run of an experiment corresponds to a particular
Bohmian history. More general modal interpretations can try to invent a stochastic dynamics,
satisfying Bayes’ sum rule, which reproduces the quantum probabilities for their particular
choice of a commuting set of variables. They need only explain why that dynamics can never
be probed by experiment.
As for Bell’s original complaints about the impossibility of assigning historical values to
unmeasured spin components of actual particles, it does not appear that Bohmian mechanics
says any more about that problem, than a general modal interpretation. Its treatment of
spin is entirely formal, and it does not assign histories to the values of all components of
the spin.

A.3 THE RELATIVE/STATE—MANY WORLDS INTERPRETATION

Probably the most popular of the interpretations of QM that regards the wave function as
an actual physical thing, is the Everett relative state interpretation, which morphed into the
DeWitt–Graham Many Worlds interpretation [103]. Everett insisted that collapse of the wave
function was not real, but just signified a correlation between different subsystems of a global
system representing the entire universe. Subsystems are defined by a tensor factorization of
the Hilbert space. The Hilbert space state is considered real, and satisfies the deterministic
Schrödinger equation for the whole universe. While Everett was aware, at some level, of
the phenomenon of decoherence, he regarded it as a mere tool for deciding which tensor
factors of the Hilbert space were good models of classical observers. He always insisted that
decoherence was not an essential part of his interpretation, and that at some level, the
434  Quantum Mechanics

quantum coherence of superpositions of macroscopically different states was testable. This

agrees with our discussion of the phenomenon of unhappening. Indeed, much of Everett’s
discussion sounds very much like the presentation of QM in this book.
The place where Everett parts ways with statistical interpretations of QM is his insistence
that the fundamental equations stand on their own, without any probabilistic interpretation.
Everett claims that one can somehow derive the Born rule from the Schrödinger equation and
the postulate that operators have definite values in their eigenstates. Many authors, including
this one, find his arguments for this unconvincing. The basic equation of the relative state
interpretation is the von Neumann decomposition of the “state of the universe” in the tensor
product basis of “system plus apparatus”
X
|Ψi = anM |ψn i ⊗ |ψM i. (A.14)
Each term in this decomposition of the normalized state |Ψi is a relative state of the apparatus
relative to the system.
Everett asserts that the natural measure on the set of relative states is simply the squared
norm |anM |2 , essentially putting in the Born rule by hand. There have been a variety of
attempts to really derive the Born rule, but the fact remains that if one regards the wave
function as real, then its coefficients in some tensor product basis are equally real. They should
be measurable by some single experiment at a fixed time. That is, they are not probabilities.
The suggestion that the probability rule might be derivable from the other postulates of QM
is a fascinating one.
In a series of recent papers, Zurek [28] has taken an interesting intermediate stance on
this problem. What follows is my own translation of his arguments into the language of this
book. There is no guarantee whatsoever that it is equivalent to Zurek’s own point of view.
In Chapter 2, we saw that classical mechanics is a particular kind of QM. In the context of
Hilbert spaces with a finite number of states it is a restriction to unitary transformations that
are in the permutation subgroup, which permutes the elements of a particular basis. Zurek
proposes to allow the generalization to arbitrary unitary transformations, without mentioning
the word probability. The probability interpretation of the wave function is to be derived from
the more primitive notion of the decomposition of the system into two subsystems. Zurek
takes as a postulate that this decomposition corresponds to a tensor product decomposition
of the Hilbert space. I would characterize this postulate as IOTTMCO.7 Let H = HS ⊗ HE .
Given bases in the two subfactors, we have the Schmidt decomposition
X
|ψi = ciA |ii ⊗ |Ai. (A.15)
iA

We will now depart from Zurek’s presentation by assuming that the system E, for environ-
ment is vastly larger and that its dynamics is described by a Hamiltonian with the properties
7
Intuitively obvious to the most casual observer.
 Interpretations of Quantum Mechanics  435

of the local lattice Hamiltonians we studied in Chapter 10. In particular, the system E pos-
sesses collective coordinates, with the properties outlined in Chapter 10. It is important to
emphasize that, although we used the language of probability to describe those properties,
they are simply statements about the commutators of operators, and the behavior of “coarse
grained histories,” defined by the mathematical formula

A(Ca (tL )) ≡ hB|P [Ca (tL )]e−iH(tL −tL−1 ) . . . e−iHt1 |Ai. (A.16)

P rob (Ca (tL )) = |A(Ca (tL ))|2 . (A.17)

This formula is written in the Hilbert space HE , and we have set ~ = 1. P [Ca (tL )] is the
projection operator on a range of eigenstates of the collective coordinates with the eigen-
values indicated in square brackets. According to the definition of collective coordinate, we
take the range of eigenvalues around the central value to scale like √1N , and the dimension
of the projector is of order ecN with c of order one. N is the logarithm of the dimension of
HE . The arguments of Chapter 10 show that the numbers P rob (Ca (tL ) have the mathemat-
ical properties we would expect of probabilities for histories, including, up to exponentially
small corrections the classical probability sum rule, the property which allows us to define
Bayes’ conditional probability rule. Given the dynamics of the system, these probabilities are
completely determined by the initial eigenvalue of the collective coordinates. We postulate
in addition that if the system is in an eigenstate of an operator, then the theory predicts
that the physical quantity represented by that operator actually has the value given by the
mathematical eigenvalue.
Now we take the initial state of the composite system to be |ψi and choose the basis
in HE to be eigenstates, with different eigenvalue, of the collective coordinates of B. The
postulated size discrepancy between the two Hilbert spaces assures us that we can do this,
actually in a huge number of different ways. Once we have chosen the basis in this way
we can write ciA = ci δiA . We now assume that the initial state of the world was a simple
tensor product X
|ψinit i = ci |ii ⊗ |Readyi, (A.18)
i
and that some interaction between the small and large systems allowed this to evolve into
the entangled state |ψi. As first noted by von Neumann, this is completely compatible with
unitary evolution.
The squares of amplitudes of decoherent histories have all the properties expected in a
classical theory of probabilities for those histories. The macrosystem has been entangled with
the microsystem in such a way that the value of the initial collective coordinates is correlated
with the values of the projection operators |iihi| on the microstate. Observation shows that
macroscopic collective coordinates perform individual histories, not weird superpositions of
histories, so we have no choice but to interpret the lack of certainty of which history will
436  Quantum Mechanics

actually occur to anything else but a probability theory. The reduced density matrix for
the values of the collective coordinates is a diagonal matrix with entries |ci |2 and so these
must be interpreted as the initial probabilities predicted by the theory, for which history will
occur. Since they are independent of the particular macrosystem, which has been entangled
with the system S and were coefficients in the pre-entanglement state of S, they must be
identified as the probabilities for the microsystem to be in the state |ii given that the value
of the projection operator |ψinit ihψinit | is 1.
So, we have started from the assumption that the state of a system (physical interpreta-
tion not assumed known) is a vector in Hilbert space, which evolves under unitary time evolu-
tion, and that when the state is an eigenstate of an operator, the physical quantity represented
by the operator has value equal to the operator’s eigenvalue. We have added the assump-
tion that composite systems are described by tensor products, and invoked mathematical
properties of collective coordinates of large systems with local Hamiltonians. Entanglement
between a macroscopic and microscopic system then leads to a state which is a superposi-
tion of different initial values of macroscopic collective coordinates. The observational fact
that the world seems to involve only single histories of collective coordinates, combined with
the fact that histories of collective coordinates decohere, leads us to the Born rule for ALL
systems, whether microscopic or macroscopic. The quantum state must then be interpreted
as a probability distribution, not a real physical object. analogous to a classical field.
Another attempt to “derive” the probability interpretation was given in a famous paper
by Hartle [29], which argued that, starting only from the assumption that the value of an
operator A in the state |ai i was ai , then if one studied the ensemble of N identical copies
of the system, each prepared in the state |ψi = ci |ai i, then one could define a frequency
P

operator, which calculated how many times one would obtain the result a1 in simultaneous
measurements of A in each of the systems. The frequency operator is
N
[N −1
X X
f (a1 ) = δ(a1 , aki )]|a1i , . . . aki iha1i , . . . aki |. (A.19)
a1i ...aN
i
k=1

It is obvious that if the ensemble is in a state where A has definite values in each copy of the
system, this operator counts the frequencies of getting any particular result.
Hartle shows that in the limit N → ∞, we have
f (a1 )[|ψi ⊗ . . . ⊗ |ψi] = |c1 |2 f (a1 )[|ψi ⊗ . . . ⊗ |ψi]. (A.20)
Thus, in the limit of an infinite ensemble of identically prepared systems, the frequency
operator, which gives us the fraction of systems in the state |a1 i when it acts on eigenstates
of A⊗. . .⊗A, has vanishing uncertainty on any identical tensor product state of the ensemble
and its value is the Born rule probability |ci |2 . This is certainly an interesting result, though
many have questioned whether it is in any sense a derivation of Born’s rule.
 Interpretations of Quantum Mechanics  437

The Many Worlds Interpretation, is a philosophical expansion of Everett’s point of view,

even though the two are often conflated. It asserts that, at least when the subsystem in
the above decomposition, whose states are labeled by capital letters, represents a collective
coordinate of a macroscopic object, then the decomposition reflects a physical branching of
the world into noncommunicating branches.
The quantum state, according to this point of view, describes not just the universe we
observe, but also an infinite number of alternate universes, all of which are real. The whole
ensemble, which one might call a multiverse (though that term is more often used in a
different context), is a system in which every possibility encoded in a superposition involving
macrosystems, is realized. This version of the interpretation has a much harder time coping
with the phenomenon of unhappening, and an experimental demonstration of unhappening
would rule it out.
The worlds of the MWI are defined by different decoherent histories, which branch from
each other every time a quantum system is entangled with the collective coordinates of a
macroscopic system. Most macrosystems do not leave permanent traces on the world, and
one cannot even decide whether a particular system will leave a permanent trace without
understanding the ultimate fate of the universe. Consider again our laboratory in intergalactic
space, in which we have set up the “Schrödinger’s bomb” experiment, but with a variable
time delay between the two clock settings determined by the outcome of the measurement of
some quantum spin component. I would defy any adherent of MWI, who had not yet heard
about this particular thought experiment, to come up with a description different than “The
spin measurement creates a branch into two alternate worlds, which exist objectively, and
in each of which the time between the measurement and the explosion is different.” Yet we
have seen that the standard interpretation of QM implies that a detector that remains out
of causal contact with the laboratory until after the second possible explosion time, can see
collisions between particles in the debris “from the first explosion” and photons “emitted
in the second.” It is hard to see how to make this compatible with the idea that the wave
function is an element of reality, which is telling us about how a single run of the experiment
behaves. Unhappening is, to use a particularly awkward phrase, the undecoherence of two
temporarily decoherent histories. If we use decoherence as a definition of the coming into
being of a new branch of the multiverse, then unhappening signals “the annihilation of two
worlds, back to a coherent wave function with no classical interpretation.” This has been
illustrated in a somewhat obscure movie [34].
Both versions of Everett’s interpretation of QM give no explanation of why one cannot
describe an experimental procedure, even in principle, which can measure all aspects of the
wave function at a single time. This is probably due to the fact that Everett takes as a
postulate that the quantities in the mathematical formalism of QM, which are identified
with detectable quantities measurable in experiments, are precisely the Hermitian operators
on the Hilbert space. This postulate implies, because not all operators commute, and because
438  Quantum Mechanics

the values of a complete set of commuting operators do not determine the wave function,
that it is impossible to determine the quantum state with experiments done at a single time.
This property, while acceptable and expected for a probability distribution, is unusual for a
normal physical quantity.

A.4 OBJECTIVE COLLAPSE

In the statistical interpretation of QM advocated in this book, wave function collapse is
an application of Bayes’ conditional probability rule to situations involving the histories of
macroscopic collective coordinates. If a wave function is a superposition of states correspond-
ing to two different values of the collective coordinates of a macrosystem, then the interference
terms in the computation of probability from this wave function are exponentially small in
the size of the macrosystem. The quantum theory then predicts a classical trajectory for
the collective coordinates. There are fluctuation corrections to these predictions, which are
inverse power laws in the macrosystem size, and neglecting the exponentially small inter-
ference terms we can attribute these corrections to uncertainties in initial measurements or
interaction with classical hidden variables. Thus we can pretend that the theory makes pre-
dictions about “what will happen in the real world.” The concept of something definitely
happening, in the sense of having a definite history, is an emergent one, applicable only
to collective coordinates. Since Bayes’ rule is satisfied with exquisite accuracy, we can use
it to throw away the piece of the wave function that does not correspond to a particular
macro-observation, renormalize the remainder, and allow it to propagate via the Schrödinger
equation.
Objective collapse theories attempt instead to alter the Schrödinger equation so that,
even at the microscopic level, one can have evolution of pure states into mixed states,
so that the density matrix ρpure = |ψihψ|, where |ψi = α1 |ψ1 i + α2 |ψ2 i, can evolve into
ρmixed = |α1 |2 |ψ1 ihψ1 | + |α2 |2 |ψ2 ihψ2 |. The mixed density matrix is, in a naive reading of
the measurement process, “what is really observed in experiments.” In the statistical inter-
pretation, ρmixed is the reduced density for the microsystem after it has become entangled
with a macrosystem in such a way that the microstates are correlated with different states
of macroscopic collective coordinates.
In objective collapse theories, the evolution from pure to mixed states is supposed to
occur as part of the microscopic dynamics. No one has explained why this makes these
theories any more “objective” than the statistical interpretation. Both of these approaches
admit that Einstein was wrong about God not playing dice: probability is, in either case,
a fundamental feature of the equations describing the world, rather than a consequence of
our lack of information about parts of the dynamical system. Both describe the apparent
collapse of the wave function as an application of Bayes’ rule to a mixed density matrix for
quantum states of microsystems.
 Interpretations of Quantum Mechanics  439

The origin of the unease that objective collapse theorists have with the statistical inter-
pretation stems from the frequent use of the word “measurement” by the founders of the
Copenhagen interpretation. This word seems to imply human agency, and thus would be
“subjective.” On the other hand, anyone who has understood the import of the theory of
decoherence, whether from the presentation of it in this book, or the many more lucid pre-
sentations in the literature, realizes that one can always substitute the phrase “entanglement
with the collective coordinates of a macroscopic object” for the word measurement, in the
discussion of the decoherence of the density matrix of a microsystem.
At any rate, the best thing about objective collapse theories is that they are subject
to experimental test.8 Indeed, if we could actually explore the phenomenon of “unhappen-
ing” experimentally, it would give us an immediate test of objective collapse theories. In our
thought experiment with an exploding laboratory, described in Chapter 10, the statistical
interpretation of QM implies the existence of a proton-photon scattering event, coming from
a superposition of histories where the laboratory exploded at different times. In an objec-
tive collapse theory the actual microscopic density matrix, no longer has matrix elements
connecting the two histories.
More feasible experiments, which have actually been done, already put very strong con-
straints on objective collapse theories. To understand them, we have to decide what math-
ematical form to allow for such theories. Since we want to use Bayes’ rule for the density
matrix ρmixed , it would not make much sense to assume a nonlinear time evolution equation
for the density matrix. The time evolution equation should preserve the nonnegativity of the
density matrix’s eigenvalues, and the fact that they sum to 1. It can be shown that the most
general evolution equation consistent with these rules is the Lindblad equation [104]

i X
ρ̇ = − [H, ρ] + Cab [Oa , [Ob , ρ]. (A.21)
~ a,b

An illuminating example, which gives rise to such an equation, can be constructed [105] by
coupling the Hamiltonian H to time dependent external sources H → H + ja (t)Oa . If the
sources are treated as classical Gaussian random variables, with vanishing expectation value
hhja (t) = 0ii and
hhja (t)jb (0)ii = Cab δ(t), (A.22)
then the expectation value of the density matrix satisfies this equation. In this context,
Cab is positive definite, and the Oa (t) are Hermitian operators, and one can show that, as a
consequence, the von Neumann entropy, −Tr ρ ln ρ is increasing, so that there is no violation
of the second law of thermodynamics.
8
This is in contrast to many worlds or Bohmian interpretations, which do not deviate from the mathematical
formulation of QM.
440  Quantum Mechanics

A disturbing property of the Lindblad equations, is that energy is only conserved statis-
tically. The white noise correlation of the sources shows us that there is a finite probability
for injecting arbitrarily large amounts of energy into the system described by H. Only the
expectation value of the injected energy is vanishing. In the context of quantum field theory
with an ultra-violet cutoff at some experimentally acceptable energy Λ, my colleagues and
I [105] showed that unless the dimensionless9 coefficients in Cab were very small, the Lind-
blad equation fills the universe with a gas of particles with energies of order Λ in a time
scale of order Λ−1 . We argued that the only way to avoid this conclusion, is to have the
external sources couple to very nonlocal operators, and that this would lead to violations of
the locality property of quantum field theory. The locality property states that two quan-
tum operators situated at points in space-time, which cannot communicate via signals sent
at a speed no larger than the speed of light, must commute with each other. If it is not
valid, then Einstein–Podolsky–Rosen experiments can be used to send signals faster than
light. EPR experiments have been done, and no such signals detected. Moreover, nonlocal
modifications of quantum field theory have a hard time accounting for the 12 decimal place
agreement between local quantum field theory and experiment.
Energy conservation has also been tested very stringently by the success of quantum
tunneling formulae in accounting for the lifetimes of highly meta-stable states. We are quite
familiar with the fact that in thermal equilibrium states, with temperature of order the
barrier height, states that are meta-stable at zero temperature decay very rapidly. The most
stringent constraints probably come from the extremely long lifetimes of certain radioactive
nuclei, which you explored in Exercise 17.7. In [105] we estimated that the dimensionless
coefficients in Cab had to be smaller than 10−125 in order not to interfere with the agreement
between theory and experiment in alpha particle decays of nuclei. It is hard to see how
these tiny terms in the Lindblad equation could play the role they are supposed to play in
explaining the apparent evolution of pure quantum states into mixed states, which occurs
when microscopic and macroscopic systems interact over time scales of order seconds or
minutes. This bound assumed that the violations of locality in space induced by the Cab
terms took place only on the Planck scale 10−33 cm. If instead, we allow violation of locality
at scales where experiment assures us that no violation occurs, then the bound is reduced to
10−65 . For each decade in energy that particle accelerators gain, without finding evidence of
locality violation, the bound becomes stricter by a factor of 104 .

9
In quantum field theory with a cutoff, one does dimensional analysis by multiplying each term in the
Λ
Hamiltonian by the appropriate power of 4π . In the text, we are referring to the dimensionless Cab with
Λ
this scaling taken out. In [105] we also assumed that all positive powers of 4π were simply absent from
the Lindblad equation.
 Interpretations of Quantum Mechanics  441

The literature on objective collapse theories contains very few papers, which attempt
to address the arguments of [105]. The reason for this is another defect of the Lindblad
equation, namely its lack of relativistic covariance. In fact, although [105] uses the for-
malism of relativistic quantum field theory, its authors were well aware of this problem.
They considered that the problematic aspects exposed in their work were not likely to be
resolved by a hypothetical version of the Lindblad equation, which was compatible with
special relativity.
We do not have space here to explore the literature on this topic, and readers should not
go away with the impression that objective collapse theories have been proven wrong. What
is clear is that such theories have severe problems with the invariance principles which seem
to hold in the real world (both space time translations, and Lorentz transformations), and
with the fact that signals cannot be sent faster than the speed of light. All of these properties
are subject to experimental checks, and it is certainly worth doing more experiments to check
their validity. Doing so could either confirm the kind of violations of these principles that
the Lindblad equation leads us to expect, or put stricter bounds on the unitarity violating
terms in the equation. Advocates of this sort of interpretation of QM should spend more time
trying to make a version of Lindblad dynamics, which is more compatible with symmetries
and locality.
Although I like the falsifiability of the dynamical collapse approach, from a philosophical
point of view I do not view it as a real advance on the standard statistical interpretation.
Given the modern updating of the ideas of Bohr and von Neumann via the theory of deco-
herence, the statistical interpretation gives us a complete understanding of why the mixed
density matrix ρmixed is equivalent to the true quantum density matrix, with accuracy expo-
nential in the size of the macrosystem, whenever a microsystem’s states become entangled
with those of the collective coordinates of a macrosystem. The real philosophical objection to
this approach is Einstein’s “God doesn’t play dice,” i.e. that probability should enter into a
physical theory only as an admission of ignorance of the complete physical state of the system.
Objective Collapse theories do not deal with this objection. They admit that the most refined
version of physical theory is in fact statistical, in a way that can’t be attributed to an inability
to measure all the properties of the system’s state. The only difference in philosophical pos-
ture is that dynamical collapse theories view the eventual validity of Bayes’ rule, which allows
us to pretend that there is an underlying predicted trajectory in certain circumstances, as fol-
lowing from Lindblad dynamics, rather than the entanglement with macrosystems. Occam’s
razor would suggest that we reject such theories for adding a complicated new set of rules,
which have obvious problems with Lorentz invariance and causality, and do not really solve
the fundamental philosophical problem of QM. QM is incompatible with the idea that the
mathematical equations of physics describe an underlying physical reality, “what really hap-
pens,” to which we have to add probabilistic notions for purely practical reasons. So are
dynamical collapse theories. At a time before the theory of decoherence clarified the actual
442  Quantum Mechanics

nature of wave function collapse, perhaps such theories held some attraction, but that is no
longer the case.
Still, it is preferable to reject physical theories on the grounds that they do not fit exper-
imental data. Dynamical collapse theories, unlike Bohmian mechanics and the Many Worlds
interpretation, can be falsified in this way. I would contend that to a large extent they have
been, but perhaps more sophisticated models of this type, and more precise experimental
tests, will finally put this set of ideas to rest.

A.5 SUMMARY
Attempts to “explain” QM in terms that are compatible with human prejudice about the
nature of the world, are extremely diverse, and it is probably foolish to try to force them all
into a small number of boxes. Nonetheless, in what follows, we try to outline some common
themes, which should be addressed by any such attempt.

• The first question one must ask of a theory is whether it considers quantum amplitudes
hen |ψi, perhaps in some preferred basis, to be real in the sense that we think of solutions
of Newton’s equations or Maxwell’s equations as real. The alternative is to think of
the amplitudes only as a probability distribution for normal operators on the Hilbert
space.

• A theory, which considers the amplitudes to be real, should explain how we can deter-
mine them by a collection of measurements at a fixed time. Our notion of reality is
based on our experience with being able to check whether something is true or not.
Many advocates of realist interpretations of QM accept the conventional wisdom that
measurements are complicated and use this as an excuse to not address the question of
how one actually checks the reality of the wave function. IMHO any notion of “reality”
that is not subject to experimental check should be discarded, by Occam’s razor.

• Any model which treats the quantum amplitudes as merely a probability distribution,
checkable only by doing multiple experiments on identically prepared systems, has to
face the conundra of any intrinsically statistical theory. The frequentist definition of
probability can never be carried out, so even frequentists have to settle for a certain
amount of Bayesian “expectation,” with all of its philosophically unpleasant subjectiv-
ity. This problem is exacerbated for theories of cosmology, where, in principle, one does
not get to repeat the experiment at all.

• Probabilistic models must then give a prescription for dealing with the apparent exis-
tence of a macroscopic “reality,” in which randomness appears to be consistent with the
existence of probabilities for histories of bodies at least as small as Brownian particles.
 Interpretations of Quantum Mechanics  443

The practical interpretation of a probability theory always utilizes Bayes’ rule. Bayes’
rule is a mathematical procedure, which can be implemented for any set of probabilities
for histories which obeys the decoherence sum rule. The classical rationale for Bayes’
rule is that there is an underlying reality with fixed initial conditions, and physical
laws, which predict a definite outcome given fixed initial conditions. Probability only
appears because of our ignorance of some of the initial conditions. However, we can
apply Bayes’ rule to any set of probabilities for histories obeying the sum rule, even
if the sum rule is only obeyed approximately. This is what we do in the statistical
interpretation of QM. We have already commented about the philosophical confusions
and experimental embarrassments of objective collapse theories.

• Advocates of realist interpretations of QM often ask the question “Probabilities for

what?” when confronted with a purely statistical interpretation. The answer to this
question is clear: QM is a theory that describes systems at a microscopic level, but
some systems have collective variables, whose QM properties are exponentially close to
those predicted by a classical statistical theory. The QM predictions for some micro-
scopic variable are predictions about the frequencies of occurrence of states of collec-
tive coordinates of some macrosystem, given that they have become entangled with
the states of the microsystem. This entanglement can occur without any experimental
intervention, human or otherwise. The predicted probabilities are independent of the
nature of the macroscopic object whose collective coordinates are entangled with the
microsystem.

A.6 WHAT IS WRONG WITH FUNDAMENTAL PROBABILITY?

Our experience of the world teaches us that it is a place where things happen over the course
of time. Physicists try to make that intuitive notion precise by assigning numerical values to
each “thing” and “measuring” those values. That is how we know that things happen over
the course of time. The measured values change. When we think about a baseball, two of
the things we can measure are its center of mass position and the rate of change of that
position with time. If we tried to measure them more and more precisely we would find that
we could not actually do it with arbitrary precision. Repeated measurements of accuracy
the inverse square root of the volume of the baseball in Bohr units, would reveal inevitable
random fluctuations, which obeyed the Heisenberg uncertainty relations. However, with an
V
−c 3
accuracy e aB we would be able to account for those fluctuations by classical models of
random forces. So we can get an accurate predictive theory of the motion of the baseball,
and an even more accurate classical statistical theory, which would assert that we could have
gotten precise predictions if only we would carefully identified the source of the random forces
acting on the baseball. Both of those theories would be wrong, but extremely accurate.
444  Quantum Mechanics

There are two different important concepts in the previous paragraph. One is that things
have values at a particular time. If we identify those values with the results of measurements
at that time, this is tautologically true, by the only measure of truth a scientist should accept.
The second concept is that we can make predictions about the future behavior of “things,”
given enough knowledge at a given time.
Predictive theories of physics do not predict initial conditions. Thus, if we just talk
about the values of variables at a fixed time, there is no experimental difference between
probabilistic and predictive theories. Each will predict that any given experimental outcome is
possible, depending on the initial state of the system. The real difference between probabilistic
and deterministic theories has to do with prediction. Probabilistic theories do not predict
the outcomes in the future, given a complete set of initial data in the past10 The set of
probabilistic theories divides into two classes, defined by whether or not one can define exact
probabilities for histories of all variables, satisfying the probability sum rule, which states
that the probability of observing some outcome at time t is the sum of the probabilities for all
histories which lead to that outcome. The only mathematically consistent theory we know,
which does not satisfy this rule is QM.
One can call the vast class of probabilistic theories which do satisfy the probability
sum rule exactly, pseudo-deterministic. That is to say, they are built on some set of variables
satisfying deterministic equations, and then averaging over some of the variables. The theorist
might declare that certain of the variables are hidden; i.e. in principle impossible to measure,
so that we could never verify that the world was following some particular history of the
variables, but the system behaves as if such a history existed.
In QM, we can only have pseudo-determinism for the limited subset of variables we have
called collective coordinates of macrosystems. The phenomenon of unhappening shows that
such pseudo-determinism is likely to be an ephemeral property for all but the most coarse
grained aspects of the universe. However, this pseudo-determinism, limited both in the range
of variables to which it applies, and the time over which it applies to them, is enough to
account for all of the macroscopic phenomena that shaped the evolution of our brains and
our intuitive conception of how the world works.
Since pseudo-determinism is not an exact property of QM, we are truly forced to confront
the question of what a probabilistic theory means. What does it mean, in Einstein’s phrase, for
God to play dice? We are willing to accept our own limitations as data gathering instruments
and use probabilistic theories based on our ignorance, but we are used to thinking of the
laws of nature as an expression of objective rules for describing “what is really going on”

10
The phrase complete set of initial data, means the result of a maximal set of simultaneous measurements
on a system. The word measurement means complete entanglement of a basis of states of the system with
different values of the collective coordinates of macroscopic objects. It is an approximate notion, but the
20
approximations have an accuracy better than e−10 .
 Interpretations of Quantum Mechanics  445

in the universe we observe. How can those rules be based on a formalism whose objective
frequentist definition is defined by a limiting process, which can never be carried out? Any use
of probabilistic concepts inevitably involves subjective language like expectation or confidence.
It appears that all attempts to view the Schrödinger equation and the quantum state as real
things in the sense that term is used in classical physics, are motivated by the feeling that
the subjective notion of probability should not be a fundamental part of physical theory.11
I think there are two rejoinders to this. The first is that the whole notion of prediction,
on which physical theories are based, is a human enterprise. The universe does what it does,
independently of any physical theory. Humans want to know, for both practical and emotional
reasons, what it is going to do before it does it. Classical physical theories are based on a
stance of philosophical omnipotence: human beings can invent mathematical models, which
are a precise map of the mechanism by which the universe evolves from one instant of time
to another, limited only by the ability to gather data on all relevant initial conditions. QM
denies the possibility of doing that. It says we cannot make predictions, except statistical
ones. Experiment is in very broad agreement with that conclusion. Data on microscopic
phenomena are definitely random. When we can do “repeated experiments on identically
prepared isolated systems” (the scare quotes remind you that every adjective in this phrase
refers to an approximation), the statistics of those random events matches the mathematical
predictions of QM. The mathematical formulae of QM do not admit a pseudo-deterministic
interpretation. So what is wrong with that?
All it really means is that we must give up our dreams of omnipotence to a greater
degree than some of us are willing to agree to. We cannot even build a mathematical model
of the workings of the universe that would have predicted everything if only we would been
allowed to know in advance the particular history of the hidden variables that the universe was
following. What a tragedy for human ambition! Given the unprecedented degree of agreement
between theory and experiment that quantum theories have achieved, one is tempted to say,
“Get over it and grow up. You can’t always get what you want, but so far we have gotten
more than we need.”
The second rejoinder is that the quantum theory of probability for noncommutative vari-
ables has a degree of mathematical inevitability and beauty that far outstrips that of classical
mechanics. Given any list of data, supposed to correspond to some measurable aspects of a
physical system, the set of all classical detectables of the system can be viewed as a maximal
commuting family of normal operators in a Hilbert space and any classical probability dis-
tribution is a density operator on the space, which commutes with the family. That density
operator automatically provides a probability distribution for all normal operators. Fur-
thermore, any other density operator, not necessarily commuting with the original maximal

11
In my opinion, all attempts to get around probability have failed. Some of the evidence was presented in
our discussion of specific interpretations of QM above.
446  Quantum Mechanics

abelian subalgebra, defines a probability distribution for all normal operators. Quite remark-
ably, from a point of view about probability that attributes its use to failure to measure all
relevant variables, for pure states all of the nonsharp distributions are predicted with mathe-
matical precision. A 19th century mathematician, familiar with Boole’s logic and the theory
of Hilbert space,12 and a sufficient amount of chutzpah, might have discovered this noncom-
mutative probability theory and proposed that it might account for some of the randomness
observed in experiments in the real world.
If she had been sufficiently insightful, our hypothetical mathematical physicist might have
noticed the beautiful unification of detectable quantities in a system, and operations on a
system, provided by QM, and scooped Emmy Noether by proving the quantum Noether’s
theorem before the classical version. The quantum theorem is simpler than the classical
one, and much more integrated with the fundamental principles of the noncommutative
probability theory. Completely apart from its agreement with experiment, the mathematical
structure of QM is much more compelling and beautiful than its classical counterpart.
Wigner and others have speculated [106] about the “unreasonable effectiveness of math-
ematics in the natural sciences.” While this point of view has some disturbing quasi-religious
overtones, it nonetheless points out a truth, for which we do not yet have a complete explana-
tion. Mathematics is an invention of humans, but mathematical structures have the surprising
ability to become useful in contexts far removed from the original purpose for which they
were invented. The geometry of Hilbert space, a straightforward mathematical extension of
the (approximate!) Euclidean geometry of ordinary three-dimensional space, turns out to be
the fundamental principle underlying physics at its most microscopic level. How remarkable!
Perhaps we just have to accept that the fundamental limitations of our brain’s software
will forever prevent us from having a more intuitive feeling for the mechanism that underlies
the evolution of the universe. We should be thankful that we have complete command of
the mathematics, which allows us to make exquisitely precise predictions of the behavior
of microsystems which are so far beyond our intuitive ken. We have speculated above that
sentient beings whose brain architecture was based on the principles of Quantum Computer
Science, might have a better intuitive understanding than we do, of the intrinsic nature of the
quantum universe. If we ever encounter or build such beings, we could ask them to explain
it to us, but we should worry about suffering the fate of Christopher Kingsley, the hero of
Fred Hoyle’s “Black Cloud.”

12
If the mathematician were willing to content herself with finite dimensional spaces, Boole’s logic would
have been sufficient.
 APPENDIX B

The Dirac Delta Function

Consider functions on the real line, which have one or more of the properties, continuity,
smoothness (continuous derivatives of all orders), or Lp ( dx |f |p < ∞). Each class of
R

functions defines a vector space of infinite dimension over the complex numbers. It is also
true, although we will not prove it in general, that each of these is a topological vector space.
That is, there is a notion of when two functions are close to each other. The dual space of
any vector space is the space of linear functionals, linear maps of the space into the space of
complex numbers. For topological vector spaces, we also require that the linear functionals
be continuous. That is, when two functions f1,2 are close to each other, the values of the map
L(f1 ) and L(f2 ) are close to each other.
The Dirac delta function can be thought of as a linear functional on many of these spaces.
For example, consider the space of square integrable functions. The delta function, δ(x − x0 ),
acting on a function, simply evaluates that function at a point x0 . It is obviously
qRlinear. It is
not, however, continuous. The distance between two functions in this space is |f1 − f2 |2 .
Let f1 − f2 differ from 0 by an amount of order 1, but only in an interval of size  centered
√
at x0 . Then the distance between the functions is o( ) but the difference between δ[f1 ] and
δ[f2 ] is order 1, no matter how small  is. The delta function is continuous if we define the
distance between two functions to be the maximum of their pointwise difference over the
real line.
Similarly, derivatives of the delta function can be defined as functionals, but on somewhat
more restrictive spaces of functions. We simply use integration by parts to write

dk k
kd f
Z
δ(x − x 0 )f (x) = (−1) (x0 ),
dxk dxk

which defines a linear functional on the space of functions that is k times differentiable at
x0 . Depending on the notion of distance in the space of functions, these functionals might or
might not be continuous.

447
448  Quantum Mechanics

The delta function (and its derivatives) concentrated at a fixed point y map the space
of functions on which they act into the complex numbers. If we let y vary, then we have
instead a map from a space of functions into another space of functions, which can often be
identified with the original function space. Indeed, the delta function itself is clearly just the
identity operator in function space. We can think of such distributions as continuous versions
of matrices, called integral operators, mapping a function f into another function g via the
formula Z
g(x) = dy K(x, y)f (y).

This formula makes sense when K(x, y) is an integrable function (with further smoothness
conditions, if we want g and f to have the same differentiability properties), but also when
K is a delta function or one of its derivatives. Given two such integral operators, we can
define their product by Z
K12 (x, y) = dz K1 (x, z)K2 (z, y),

which gives us the operator product in the corresponding linear space.

In quantum mechanics, we are concerned with a space of square integrable functions,
which are, generically, not differentiable. There are, however, bases consisting of infinitely
differentiable functions, so that any function can be approximated (in the L2 norm) by such
functions. The delta function is a globally defined operator on this space, but its derivatives
define unbounded operators, which are only defined on a dense subspace of the Hilbert space.
 APPENDIX C

Noether’s Theorem

Consider a Lagrangian L(q i , q̇ i ), whose action integral is invariant under a one parameter
continuous group of symmetries q i → q i + f i (q, q̇). We have written the infinitesimal form
of the transformation and  is the infinitesimal. The variation of the action integral is

∂L i ∂L ˙i
Z
δS = dt[ f + i f ]. (C.1)
∂q i ∂ q̇

We can write this as

∂L d ∂L d ∂L
Z
δS = dt[ i
− ( i )]f i + ( i f i ). (C.2)
∂q dt ∂ q̇ dt ∂ q̇

If the action is invariant under the symmetry, δS = 0, for every path q i (t) and every time
interval. Invariance of the action does not imply invariance of the Lagrangian. It is enough
that δL = ∂t Λ. Thus, the condition for invariance is

∂L d ∂L d ∂L
0=[ i
− ( i )]f i + ( i f i − Λ). (C.3)
∂q dt ∂ q̇ dt ∂ q̇

The Lagrange equations of motion are

∂L d ∂L
0= i
− ( i ). (C.4)
∂q dt ∂ q̇

d ∂L i
0= ( f − Λ). (C.5)
dt ∂ q̇ i
That is, every continuous one parameter group of symmetries of the action corresponds to a
conserved quantity.

449
450  Quantum Mechanics

Let’s try this out for the symmetry of time translation invariance. Then f i = q̇ i and
Λ = L. Then the conserved quantity is
∂L i
q̇ − L. (C.6)
∂ q̇ i
Recall that
∂L
≡ pi , (C.7)
∂ q̇ i
the momentum canonically conjugate to q i . The application of Noether’s theorem to time
translation invariance gives us the canonical Hamiltonian as the corresponding conserved
“Noether charge.”
For invariance under a linear transformation of the q i , like rotations, we have f i = ωji q j ,
and Λ = 0, where ωji is the infinitesimal matrix of rotation. The conserved charge for rotations
in the ij plane is
Jij = qi pj − qj pi , (C.8)
where we have used the Euclidean metric δij to lower the index on q i .
There are three ways to think about the quantum mechanical use of Noether’s theorem.
The first is to view it as a way to derive the canonical commutation relations. In quan-
tum theory (QM), symmetry transformations are unitary operators in Hilbert space, which
commute with the Hamiltonian. A one parameter group of symmetries has a Hermitian gen-
erator, Q, which commutes with the Hamiltonian. Noether’s theorem gives us an expression
Q = pi f i (q) (we will only discuss the simple case where f i is independent of pj ) for this
generator in terms of the classical canonical variables q i and pi . The equations
i[Q, q i ] = f i (q) (C.9)
are solved by the canonical commutator
[pi , q j ] = −iδij . (C.10)
In the path integral formalism, the use of Noether’s theorem is even more direct. Restricting
again to transformations for which f i depends only on q, we do a change of variables in
the path integral, which has the form of a symmetry transformation with a time dependent
infinitesimal parameter (t). The measure of integration is invariant under this transforma-
tion1 and the action changes by
Z Z
δS = dt ∂t pi (t)f i (q(t)) = dt ∂t  Q(t). (C.11)

1
Actually, in quantum field theory there are examples where the measure is not invariant. This leads to
quantum mechanical breaking of a symmetry of the classical action, also called an anomaly because the
phenomenon was so confusing when it was first discovered.
 Noether’s Theorem  451

Since we just did a change of variables, the actual value of the integral should not change
and so the effect of this new term in the action should be zero.
Z
hF | dt ∂t  Q(t)|Ii = 0. (C.12)

Choosing  to be a differentiable function with support separated from the ends of the interval,
so that the boundary values are not changed, we conclude, by integration by parts, that ∂t Q
has vanishing matrix elements between any initial and final state of tF > t > tI .
If we apply the same procedure to the evaluation of expectation values of time ordered
products of operators, we get an identity
n
X
∂t hF |T [Q(t)O1 (t1 ) . . . On (tn )]|Ii = δ(t − tj )hF |T [O1 (t1 ) . . . δOj (tj ) . . . On (tn )]|Ii. (C.13)
j=1

Identities like this are called Ward–Takahashi identities and they are very useful in quantum
field theory.
However, it may be that the ultimate point of view about Noether’s theorem in QM is that
it is really an intrinsic part of the quantum formalism, for which we have no need of a classical
derivation. This is, for the most part, the point of view we have followed in the body of the
text. In QM, a symmetry is a unitary operator, which commutes with the Hamiltonian.
On the other hand, every normal operator has eigenvalues and that makes it a potential
candidate for “measurement,” if we can construct a device, which will entangle different
eigenstates with different values of a macroscopic collective coordinate. So symmetries lead
directly to conserved quantities in QM.
The one place where the classical analysis leads to results which have not yet been
reproduced using this more abstract point of view is the construction of local Noether currents
and charge densities in quantum field theory. Noether’s theorem leads directly to operators
representing the amount of conserved quantum number in a local region, and the abstract
formalism has not yet reproduced such a construction.
 APPENDIX D

Group Theory

A symmetry in quantum theory (QM) is a unitary transformation, U , which commutes with1

the Hamiltonian. Given two such transformations U1,2 , their products U1 U2 and U2 U1 are
also symmetries. Furthermore, every unitary transformation is invertible.
The set of all symmetries of a given Hamiltonian thus satisfies the mathematical axioms
of a group, namely a set G equipped with a multiplication rule, and a unit element e ∈ G
such that
eg = ge = g,
for all g ∈ G. The axioms for a group also require that every element has an inverse such
that
gg −1 = g −1 g = e.
A group is defined by its abstract multiplication law: a listing or parameterization of all the
elements, and a rule for determining what the product of any two elements is. In principle,
this rule takes the form of a big table. However, in most cases, one can find a simpler way of
describing the multiplication law. For groups whose underlying set is discrete, this is often
expressed in terms of generators and relations. For example, the group ZN of integers k
modulo N is generated by a single element z = e2πik/N satisfying the relation z N = 1. The
group SL(2, Z) of fractional linear transformations on a complex variable w, with integer
coefficients is
aw + b
w→ ,
cw + d
where a, b, c, d are integers satisfying ad − bc = 1. The group is generated by

T :w →w+1
1
More generally, we could demand that the time evolution operator and the operator U form group with
a finite number of infinitesimal generators. The Galilean or Lorentz groups are examples of this kind of
structure.

453
454  Quantum Mechanics

and
1
S:w→− .
w
These satisfy
S2 = 1
and
(ST )3 = 1.
The group SN of permutations of N elements is generated by transpositions, etc.
Given a group multiplication law, there may be many inequivalent ways to realize it in
terms of unitary transformations on a Hilbert space. These are called inequivalent unitary
representations of the group. Note that given a representation U (g), there is another one
V † U (g)V for every unitary transformation in the Hilbert space in which U (g) acts. Two
representations related in this fashion are called unitarily equivalent.
Some symmetries, like translations and rotations, depend on continuous parameters.
When this is the case, there is always a subgroup which can be continuously deformed to the
identity. This is called the connected component of the identity. Sophus Lie had the insight
that any transformation in the connected component of the identity could be built up by a
sequence of infinitesimal transformations. As a consequence of his seminal work, continuous
groups are now called Lie groups. If ω a are the real continuous parameters on which the
group elements depend (the number of independent parameters is called the dimension of
the group), then in any unitary representation, an infinitesimal transformation has the form

U (ω a ) ∼ 1 + iω a Ta ,

when the parameters are very close to zero. To leading order in ω, we have

U † (ω a ) ∼ 1 − iω a Ta† ,

so that unitarity implies Ta = Ta† .

To leading order in the deviation of ω a from zero, we have U (ω1a )U (ω2a ) = U (ω2a )U (ω1a ).
The leading correction to this relation is given by

U (ω1a )U (ω2a )U −1 (ω1a )U −1 (ω1a ) = 1 − ω1a ω2a [Ta , Tb ].

The right-hand side will be an infinitesimal group element, if and only if

c
[Ta , Tb ] = ifab Tc ,
c c
where the coefficients fab = −fba are real.
 Group Theory  455

The set of infinitesimal generators T a is closed under taking real linear combinations
and commutation. This is the definition of a Lie Algebra.2 The Baker–Campbell–Hausdorff
formula
1
ln (eX eY ) = X + Y + [X, Y ] + · · · ,
2
where the ellipses denote higher order nested commutators, shows that
a a
eiω1 Ta eiω2 Ta .

is a group element, and thus the entire multiplication table of the group is determined by its
Lie algebra.
If the group has a faithful (no element is mapped into the identity matrix unless it is
the identity element) finite dimensional unitary representation, then the parameter space is
a
compact, since the space of all unitary matrices is. Consider the abelian subgroup eiαω Ta
for some fixed vector ω a . The real symmetric matrix tr (Ta Tb ) can be brought to the form
δab , by redefining Ta → Sab Tb with S a real invertible matrix. With this normalization, we
see that the group element is periodic in α with periodicity α → α + √ω2πa ωa .
A maximal set of commuting generators Ta is called a Cartan subalgebra, and the number
of commuting generators is called the rank, r, of the group (the total number of generators
is called the dimension, dG , of the group). If k of those generators commute with all other
generators, we say that G = U (1)k ⊗ GSS . GSS has rank r − k and dimension dG − k. We
can obviously study the full symmetry group by studying the U (1)s and GSS , separately. We
can simplify things further by writing

GSS = G1 ⊗ . . . ⊗ Gn ,

where every element in Gi commutes with every element in Gj , and each Gi cannot be further
factorized into commuting subgroups. Such nonfactorizable groups are called simple, and the
SS on GSS stands for semisimple.
A famous theorem [107] classifies all compact simple Lie groups. They fall into four infinite
families, and there are five exceptional groups G2 , F4 , E6,7,8 . The subscript denotes the rank
of the group. The four infinite families are the groups of all special (i.e., determinant 1)
unitary matrices, SU (n) in n complex dimensions, the groups SO(2n) and SO(2n + 1) of
orthogonal transformations in even and odd numbers of real dimensions, and the groups of
unitary symplectic transformations Sp(n) in 2n real dimensions. Symplectic transformations
2
For a pure mathematician the definition of a Lie Algebra invokes an operation with all the properties of
the commutator of two operators, but is not explicitly realized in some linear space. Such a linear space
realization is called a representation of the Lie algebra. In QM, continuous symmetries are always unitary
transformations, so we are always working with a unitary representation.
456  Quantum Mechanics

are like orthogonal transformations except that they preserve a nondegenerate antisymmetric
bilinear form.
We will not outline the proof of this theorem, but just indicate the main line of argument.
The maximal set of mutually commuting generators of the Lie algebra is called its Cartan
subalgebra. The number of generators in the Cartan subalgebra is called the rank of the group.
Label the Cartan generators by Hi , 1 ≤ i ≤ r. Commutation with the Cartan generators is
a linear transformation on the set of generators not in the Cartan subalgebra. One argues
that this linear transformation is diagonalizable, that is: there exist linear combinations of
generators Er such that
[Hi , Er ] = ri Er .

The r dimensional vectors ri are called the roots of the algebra. The Jacobi identity then
implies that [Er , Es ] is a generator with root vector ri + si , so that the roots form a lat-
tice in r-dimensional space. The constraints on the geometry of such lattices lead to the
abovementioned classification of Lie algebras.
Given a simple compact Lie group, one can ask for all of its inequivalent unitary represen-
tations. There is a trivial way to make new representations from old, somewhat analogous to
the product construction of semisimple groups from simple ones. Given two sets of generators
Ta1,2 , each of which satisfies the group commutation relations and is realized as Hermitian
operators in Hilbert spaces H1,2 then the operators
!
Ta1 0
0 Ta2 ,

acting in H1 ⊕ H2 are Hermitian and satisfy the commutation relations. Note that operators
of the form a1 P1 + a2 P2 , where Pi are the projection operators on the subspaces Hi , com-
mute with all the generators of this reducible representation. The action of the group leaves
invariant the individual subspaces.
A representation is said to be irreducible if the only subspaces left invariant by the group
action are the zero vector and the whole Hilbert space. For finite dimensional irreducible
representations, Schur’s Lemma states that in this case, the only operators commuting with
all the generators are proportional to the unit operator. This means that operators like
Ta Ta are just pure numbers in a finite dimensional irreducible representation. The values
of these Casimir invariants (their general form is tr[Ta1 . . . Tan ]Ta1 . . . Tan ) characterize the
representation. Following the model of our discussion of angular momentum, one uses this
numerical information to find all the vectors in the representation.
Given two unitary representations of a Lie group G in Hilbert spaces H1,2 , the tensor
product space
H1 ⊗ H2 ,
 Group Theory  457

carries a representation with generators

Ta1 ⊗ 1 + 1 ⊗ Ta2 .

Even if the individual factors are irreducible, the tensor product representation is not. The
coefficients expressing the decomposition of a general vector in the tensor product into vectors
in its irreducible pieces are called generalized Clebsch–Gordan coefficients, and an entire
industry is devoted to computing them. The fact that the tensor product representation
has such a direct sum decomposition follows from the Peter–Weyl theorem: for compact
semisimple Lie algebras: every representation is a direct sum of irreducible representations.
This theorem is not true for the noncompact groups we discuss below.
A representation of a group is said to be real if all of the generators Ta can be written as
imaginary antisymmetric matrices in some basis, so that the group action is by real matrices.
For any representation, the matrices T̄ a = −Ta∗ form another representation called the com-
plex conjugate representation. Obviously, in a real representation these two representations
are equal to each other. A weaker requirement is that the two are unitarily equivalent.

−Ta∗ = U Ta U † .

An example where the complex conjugate representation is unitarily equivalent to the original
one, where U 6= 1, is given by the spin 1/2 representation of SU (2).

−σa∗ = σ2 σa σ2 .

Such representations are called pseudo-real. The multiplication rule σa σb = δab + iabc σc
shows that it is impossible to find a representation in which all matrices are imaginary.
Representations which are unitarily equivalent to their conjugates, but are not real, are
called pseudo-real. One interesting property of all real or pseudo-real representations is that

tr [Ta1 . . . Ta2k+1 ] = −tr [Ta∗1 . . . Ta∗2k+1 ] = −tr [Ta†1 . . . Ta†2k+1 ] = −tr [Ta1 . . . Ta2k+1 ],

so that the trace of an odd product of generators vanishes. In the penultimate equality above,
we have used the fact that the traces of M and M T are equal, and that the trace of a product
is invariant under cyclic permutation.
The representation space of a complex or pseudo-real representation is a complex Hilbert
space. That of a real representation is a real Hilbert space, but we can complexify it by
doubling the number of real dimensions. Thus, given a Lie group with real parameters ω a
a
we can look at the set of all operators eiω Ta with ω a = Ra + iI a . These will form a group
(by the Campbell–Baker–Hausdorff formula) called the complexification of the original Lie
a
group. The complexified groups are no longer compact, because the transformations e−I Ta
are not periodic functions of I a . They have unbounded eigenvalues for large values of |I|.
458  Quantum Mechanics

We can consider these complexified groups to be real Lie groups, with parameters Ra , Ia
but in these finite dimensional representations, not all the generators are Hermitian, so the
representation is not unitary. Among the groups we can construct in this way are SO(m, n)
the set of all transformations satisfying

det O = 1,

OT gO = g,

with g a symmetric matrix with m positive and n negative eigenvalues. A particular case is
the Lorentz group SO(1, 3).
These non-compact groups do have unitary representations, but they are all infinite
dimensional. The theory of unitary representations of non-compact groups is quite intricate
and we will not study it, except to give an example for SO(m, n). Consider the generalized
hyperbolae defined by
y i gij y j = ±1.

We can build a Hilbert space of complex valued square integrable functions on one of these
hyperbolae. The scalar product is
Z
dm+n yδ(y i gij y j ± 1)f ∗ (y)g(y).

The action of SO(m, n), via y i → Oji y j induces an action of the group on this Hilbert space,
which preserves the scalar product, and so is given by unitary transformations.
We conclude this appendix with a quick and dirty proof of the Wigner–Eckart theorem.
We study a system with conserved angular momentum J. The Hilbert space breaks up into
a direct sum of spaces, with each term in the sum being the tensor product of an irreducible
representation |j mi of the rotation group and a Hilbert space Hj on which the operators J
do not act. The same is true of the vector space of all operators A on the original Hilbert
space. We can write any operator as
X
A= Ak Okn ,

where
(j)
[J, Okn ] = iJkp Ojp ,

where J (j) is the (2j +1)×(2j +1) dimensional matrix representation of angular momentum.
The expansion of A omits all of the indices that A carries, by virtue of transforming like a
direct sum of angular momentum representations.
 Group Theory  459

To prove the Wigner–Eckart theorem, we insert the above commutation relation between
states |j mi and |j 0 m0 i, and allow the J operator in the commutator to act on the states.
We then equate that to the result of the operator commutator, obtaining
0
0 = [J (j ) − J (j) − J (k) ]hj 0 m0 |Okn |j mi.

The notation here is very compact: each set of angular momentum matrices acts only on the
magnetic quantum numbers associated with its label. This equation is the defining equation
for Clebsch–Gordon coefficients and it determines them completely up to a normalization.
However, the operators Okn also act on the Hilbert spaces Hj , so in general, the matrix
element of a fixed Okn between states in Hj and Hj 0 will be a number Rk,j,j 0 , which depends on
the total angular momenta as well as the choices of individual states in those Hilbert spaces.
This is called the reduced matrix element, and is only defined when the Clebsch–Gordon
coefficient is not zero. You can find a table of Clebsch–Gordon coefficients by consulting
Google.
Group theory is a vast and intricate branch of mathematics, full of beautiful theorems
and a vast array of fascinating formulae. Many mathematicians spend their entire careers
opening up new avenues in the teeming metropolis of groups. This appendix barely scratches
the surface of what is known about the relatively small part of group theory with direct
relevance to quantum mechanics. It is enough for you to understand the uses of group theory
in the text, but no more.
 APPENDIX E

Laguerre Polynomials

Recall that the hydrogen wave functions are written in terms of polynomials in the variable
w = 2ρ, which we called the associated Laguerre polynomials Lak where a = 2l + 1 and
k = n − l − 1. These satisfy

d2 Lak dLak
w + (a + 1 − w) + kLak = 0. (E.1)
dw2 dw

We have followed Griffiths in our normalization of Lak , whereas in this appendix we will follow
the mathematical convention that the coefficient of the constant term in every polynomial
is 1. Thus, (Lak )text = k!(Lak )appendix . For k = 0, the obvious polynomial solution of the
equation is La0 = 1, while for k = 1 a trial linear solution leads to La1 = 1 + a − x. Now
assume we know all of the polynomials up to some value of k and write

(k + 1)Lak+1 ≡ (2k + 1 + a − w)Lak − (k + a)Lak−1 . (E.2)

Acting with the operator wd2w on both sides and using the equations for Lak and Lak−1 , it is
easy to verify that this definition gives a solution of the k + 1th equation, which is obviously
a polynomial of one higher order. The normalization is correct if we remember that in the
math convention, all of these polynomials equal 1 at w = 0.
Using the fact that the hydrogen wave functions unlm are orthogonal, we find that
Z ∞ Γ(k + a)
xa e−x Lak Laj = δjk . (E.3)
0 Γ(k + 1)

To understand the normalization here one must use the formula for normalized wave functions
in Chapter 13, and remember the change of conventions between this appendix and the text.

461
462  Quantum Mechanics

Now let us write the recursion relation defining the associated Laguerre polynomials as

(k + 1)Lak+1 − (2k + 1 + a)Lak + (k + a)Lak−1 = wLak . (E.4)

This equation should be interpreted as defining the (infinite) matrix representation of the
operator w, in the orthogonal basis of Laguerre polynomials. The expectation values of wp
are given in terms of diagonal matrix elements of the p-th power of this matrix.
 APPENDIX F

Summary of Dirac Notation and

Linear Algebra

A finite dimensional Hilbert space can be thought of as all complex linear combinations of a
set of orthonormal basis vectors |ei i (1 ≤ i ≤ N ) whose scalar products are defined by the
formula
hei |ej i = δij .
A general vector |vi is a linear combination
X
|vi = vn |en i,
n

and a general dual vector hw| is

wn∗ hen |.
X
hw| =
n
The scalar product is required to be linear in both of these variables so that

wn∗ vn .
X
hw|vi =
m,n

We can think of |vi (w.r.t. the |en i) basis, as the column of numbers vn and hw| as the
row of numbers wn∗ , in which case the scalar product formula is the same as the formula for
multiplying rectangular matrices. Mathematicians write vectors as simple letters, v, and the
scalar product as hw|vi = (v, w). This is “linear in its first argument and conjugate linear
in the second.” The reason we introduce complex conjugation into the formula for the dual
vectors is that the scalar product of a vector with itself is nonnegative and can be thought of
as the square of a length. In quantum mechanics (QM), the scalar product of a vector with
itself is the probability that if one is definitely in a particular state, then one is actually in
that state, which is of course equal to one.

463
464  Quantum Mechanics

A linear operator A on the Hilbert space is a function from the Hilbert space to itself,
|vi → A|vi, which satisfies the rule

A(a|vi + b|wi) = (aA|vi + bA|wi).

Then X X
A|vi = A vn |en i = A|en ivn .
Taking the scalar product of this equation with |em i, we obtain the expansion coefficients
(Av)m (the column vector) corresponding to the vector A|vi. They are
X
hem |A|en ivn ,
n

which is the rule for acting with a square matrix on a column vector.
The scalar product hw|A|vi is also the scalar product of the vector |vi with another vector
A† |wi, which in Dirac notation has the awkward notational form

hw|A|vi = hA† w|vi.

A† is the Hermitian conjugate operator. The matrix of the operator A† is

hem |A† |en i = hen |A† |em i∗ .

The most useful rule for Hermitian conjugates is that when an operator acts to the left in a
scalar product, it acts like its Hermitian conjugate.
The expression |wihv| defines a linear operator, taking any vector |si into |wihv|si. The
vector is taken into its projection on |vi multiplied into the vector |wi. Schwinger called these
measurement symbols. The special case of a projection operator on a vector of length 1 |ei
just gives us the component vector of |si in the |ei direction. One of the two or three most
important equations in QM is the resolution of the identity
X
1= |en ihen |,
n

where the 1 on the left-hand side stands for the unit operator. In terms of matrices, the
matrix of |en ihen | has a 1 in the n-th row and column, and zeroes everywhere else, so the
identity is obvious, but it is also a valid identity for every other orthonormal basis. The unit
operator looks the same in all bases, and nothing we have said specifies which orthonormal
basis we are using.
An operator is called normal if it has a complete set of orthonormal eigenvectors

A|ai i = ai |ai i.
 Summary of Dirac Notation and Linear Algebra  465

hai |aj i = δij .

These need not coincide with the basis |ei i or some permutation of it. Any basis satisfies the
resolution of the identity, so we can write
X
|ai i = |en ihen |ai i.
n

The matrix hen |ai i is the matrix of the operator U , which transforms the |en i basis into the
|ai i basis. Its inverse is obviously the matrix hai |en i = hen |ai i∗ , because of the properties of
the scalar product. Thus, its inverse is its Hermitian conjugate

U † U = U U † = 1.

Such operators are called unitary. Furthermore, every unitary operator transforms one
orthonormal basis into another. One can show that the set of all unitaries can be exhausted
by thinking of all the transformations that transform a given orthonormal basis into another.
If we transform the diagonal matrix of the normal operator A via A → U † AU , we will get
the matrix of A in the |en i basis, which is not diagonal. It will still satisfy [A, A† ] = 0, since
this equation transforms into U † [A, A† ]U = 0, and we can remove the U ’s because they are
invertible. Commuting with its adjoint is an equivalent characterization of a normal operator.
Every operator with this property is diagonalizable. This is called the spectral theorem and
a proof is sketched in the text. Every normal operator can be written

A = H1 + iH2 ,

where Hi† = Hi and [H1 , H2 ] = 0. Thus, the study of normal operators reduces to that of
Hermitian operators, and most QM texts use only Hermitian operators. Note, however, that
unitary operators are also normal, but rarely Hermitian. Unitary operators, which commute
with the Hamiltonian of a system, are symmetries of the system. Noether’s famous theorem,
which relates symmetries to conservation laws, is valid for discrete symmetries only if we
consider the generally complex eigenvalues of unitary operators as valid conservation laws.
Infinite dimensional systems are defined by letting the vectors have an infinite number of
components, with the restriction ∞ 2
n=1 |vn | < ∞. The Cauchy–Schwarz inequality
P

v
u∞ ∞
uX X
hw|vi ≤ t |vn | 2 |wn |2
n=1 n=1

shows that the scalar products of two such infinite vectors is finite.
A simple way of understanding the limit is to think of a finite dimensional system as
describing the states of a particle that lives on the second hand positions of a clock with N
466  Quantum Mechanics

seconds. The finite number of states correspond to the positions on the clock, and can be
thought of as labeling the possible values of a unitary matrix U , whose eigenvalues are the
N -th roots of unity.
2πin
U |ni = e N |ni.
Define the shift operator V by

V |ni = |n + 1 (mod N )i.

Then
2πi
V N = 1; UV = V Ue N .
Now imagine taking N to infinity through even values. Then V N −k = V † k for 1 ≤ k ≤
(N − 1)/2. Let us work in the basis of eigenstates of V and label them by positive integers
for k in this range, negative integers for k < N above this range, and label the state with
eigenvalue 1 of V by 0. In the limit, we get a basis of states labeled by all integers, and
normalizable vectors must satisfy
∞
X
|vp |2 < ∞.
p=−∞

Now let us try to think about what happens to the eigenstates and eigenvalues of U in this
limit. The eigenstates satisfy
U |θi = eiθ |θi,
and
U |pi = |p − 1i,
as long as p is not near the boundary p = −(N − 1)/2. If we could neglect the boundary
contributions, then a solution for X
|θi = cp |pi,
p

would be cp ∝ eipθ with a p independent constant of proportionality. Such states are not
normalizable. Note also that by neglecting the mod N contributions, we no longer have any
constraint on θ except the periodicity condition which says that θ and θ + 2πK are the same
state for any integer K.
The function of θ, X
δN (θ) ≡ e2πipθ ,
p

is finite for finite N but obviously approaches infinity in the large N limit if θ = 0. To see
what happens for other values, note that when N is finite, it vanishes for all values of eiθ
that are N -th roots of unity not equal to 1. This is just the statement that for finite N all of
 Summary of Dirac Notation and Linear Algebra  467

the nontrivial eigenstates of U are orthogonal to that with U = 1. The N -th roots of unity
become dense on the circle as N goes to infinity, and for finite N these are the only values
of θ that are allowed in our system. So the eigenfunction with θ = 0 must, when viewed as
a function on the circle, go to zero everywhere except at θ = 0. On the other hand, if f (eiθ )
is a continuous function on the circle, then
N
X
δN (θq )f (eiθq ) = f (1),
q=1

for all N . So the function δN approaches what mathematicians call a measure with point
support on the circle and physicists call the Dirac delta function δ(α) on the circle. eiα is the
coordinate on the circle. The eigenfunctions for any other value of θ are simply δ(α − θ). The
orthonormality relation converges to the relation
Z 2π
dα δ(α − θ)δ(α − θ0 ) = δ(θ − θ0 ).
0

This is called delta function normalization. All of this is explained in more detail in Chapter 6.
 APPENDIX G

Answers to Selected Problems

Answer to Exercise 1.1: The Let’s Make a Deal Problem: The possible distribution
of winning and losing doors is LLW, LW L, W LL, where W means the contestant wins and
L that she loses. There is one chance in three that she has picked the right door, and two
chances in three that she hasn’t. Thus, it is twice as probable that she’s picked the wrong
door. Once shown that one of the other doors is a loser, she knows that it is twice as probable
that the remaining door is the winner. The extra information does not change the probability
that she’s chosen the right door, it just leads to the knowledge of which of the other two
doors holds the prize, in the more probable situation in which she’s in fact chosen the wrong
door. So she should always switch her choice, to maximize the probability of winning. The
mistake many people make is to equate this situation to one in which the door with the
booby prize is revealed before the contestant makes a choice. In that case, the sample space
just consists of W L, LW for the two remaining doors and there is a 50/50 probability of
making a mistake by switching.

Answer to Exercise 1.2: Listing the possibilities by birth order we have

BBB, BBG, BGB, GBB, BGG, GBG, GGB, GGG.
Fifty percent of the cases have two girls. Now consider that we know that one of the children
is a girl named Florida. The sample space is
BBGf, BGf B, Gf BB, BGGf, BGf G, GBGf, Gf BG, Gf BGf,
BGf Gf, Gf Gf B, GGGf, GGf G, Gf GG, Gf Gf G, Gf GGf, GGf Gf, Gf Gf Gf.
We have allowed for the possibility that parents will give their children the same name. On
the other hand, Florida is no longer very popular, so the probability that more than one
child is named Florida is very very tiny, and we can neglect it. The sample space now has
ten elements, out of which seven have two girls.

469
470  Quantum Mechanics

Answer to Exercise 1.3: Assuming an equal probability for each direction, the probability
is just the fraction of 2π contained in the wedge, which is just Np . It is independent of k. If
one chooses a different origin, the only thing that is affected is the starting direction in our
choice of drawing lines. If it is taken to be such that one of the lines from the origin goes
through the point of the needle closest to the origin, then the answer is unchanged. If not,
there is a fractional change by the angle between the line between the origin and the tip
of the needle, and the nearest of the radial lines from the origin.

N ±1
Answer to Exercise 1.5: If N is odd, there are two maxima, at K = 2 . Using Stirling’s
approximation to the factorial
√
M ! ≈ 2πM (M/e)M ,

we find the distribution

N −2 2x 2x N 4x2
[ !] [(1 − )−N/2+x (1 + )−N/2−x ] ≈ [ !]−2 e− N .
2 N N 2

Answer to Exercise 1.7: There are n!(NN−n)! !

different ways of choosing n voters out of a
total population of N . We are interested in those ways in which k of the n are chosen from
the population of pN who will vote for Jefferson and n − k are chosen from the population
of (1 − p)N who will vote for his opponent. This number is

(pN )! [(1 − p)N ]!

,
k!(pN − k)! (n − k)![(1 − p)N − n + k]!

and the normalized distribution of the fraction r = k/n is obtained by dividing this by
N!
n!(N −n)! . In the regime of interest, N  n  1, we can use Stirling’s formula to approximate
√
the factorial (aN + b)! ≈ 2πaN e−aN (aN )aN +b . It is then easy to verify that N drops out
of the ratio. The k dependent part of the distribution comes entirely from the denominator of
the ratio of factorials and has the form
1−r
e−n[rln (r/p)+(1−r)ln ( 1−p )
.

Since n  1, this is dominated by the maximum, which is r = p, and the distribution

is approximately the Gaussian expansion around this point, whose width scales like n−1/2 .
The confidence intervals similarly shrink with n, and polling experts believe that n ∼ 103 is
large enough to give reliable results. What is undoubtedly true is that the systematic errors
involved in trying to poll a truly random sample of voters are so large that the additional
precision of larger n (but always  N ) would be somewhat illusory.
 Answers to Selected Problems  471

Answer to Exercise 2.1: According to the general algebraic formula, v i → Mji v j , the
P
j
first element of the vector gotten by acting with
!
0 a
(G.1)
b 0
on !
c
(G.2)
d
is 0c + bd = bd, while the second is ac + 0d = ac, so
! ! !
0 a c bd
= (G.3)
b 0 d ac

Answer to Exercise 2.2: The equation for the inverse is

! ! !
−1 a b e f 1 0
MM = = . (G.4)
c d g h 0 1

The product of the two matrices is

!
ae + bg af + bh
. (G.5)
ce + dg cf + dh

The equations for the inverse are thus

ae + bg = 1 = cf + dh, af + bh = 0 = ce + dg. (G.6)

So, g = −(c/d)e, h = −(a/b)f and 1 = (a − bc ad

d )e = (c − b )f . The last two equations
have solutions if and only if ac − bd 6= 0. This combination of matrix elements is called
the determinant of the matrix M and denoted det(M ). It measures whether the rows and
columns of the matrix are linearly independent vectors. The inverse matrix is
!
−1 −1 d −b
M = (ad − bc) , (G.7)
−c a

and it is easy to verify that M −1 M = 1.

Answer to Exercise 2.3: To show that

σ1 |±i = |∓i,
472  Quantum Mechanics

write the equations as ! ! !

0 1 1 0
= ; (G.8)
1 0 0 1
! ! !
0 1 0 1
= , (G.9)
1 0 1 0
and use the result of Exercise 2.1. Now write
! ! ! !
0 1 0 1 0×0+1×1 0×1+1×0 1 0
σ12 = = = .
1 0 1 0 1×0+0×1 1×1+0×0 0 1

Answer to Exercise 2.4:

! ! ! !
0 1 1 0 0 × 1 + 1 × 0 0 × 0 + 1 × −1 0 −i
σ2 = i =i = .
0 1 0 −1 1 × 1 + 0 × 0 1 × 0 + 0 × −1 i 0

σ22 = i × i × σ1 σ3 σ1 σ3 = σ12 σ32 = 1,

where we have used σ1 σ3 = −σ3 σ1 . It is straightforward to show that the same result follows
from explicit matrix multiplication.

Answer to Exercise 2.5: Using standard probability theory, evaluate the expectation value
of a general polynomial P (E, D) of the energy and dipole moment of the ammonia molecule.
Show that this is equivalent to the formula TrP (E, D)ρ. According to standard probability
theory, the expectation value of P (E, D) is just
2
X
pi P (E i , Di ).
i=1

Since Ei = E and Di = ±d, the most general polynomial reduces to A + BDi , so the
expectation value is
hP (E, D)i = A + Bdp1 − Bdp2 .
To do the same computation in terms of matrices, note that for a given polynomial P (E, D) =
A + BD, where A and B are the same coefficients we got for that polynomial in the classical
computation. Then
Tr (ρ[A + BD]) = A + Bdp1 − BdP2 .

Answer to Exercise 2.6: To show that

abc cde = δad δbe − δae δbd ,

 Answers to Selected Problems  473

note that all three indices must be different in order for abc to have a nonzero value. For a
given value of c, this means that either a = d or a = e, with b being the other value 6= c.
The two possibilities are mutually exclusive, so for given values of a, b, d, e satisfying this
constraint, only one value of c contributes to the sum implicit in the Einstein convention.
Consequently the answer is always ±1 and the a = d and a = e terms have opposite sign.
For a = d we have abc cab = 1, so the formula is proven.
Using this equation to evaluate ∇ × (∇ × V), for a vector function V(x), we have [∇ ×
(∇ × V)]a = abc ∇b cde ∇d Ve = ∇a ∇b Vb − ∇2 Va .

Answer to Exercise 2.7: The diagonal matrix element of A in some orthonormal basis is
the coefficient of |Ei i in the expansion

A|Ei i = Aij |Ej i.

Taking the scalar product with |Ei i, this gives Aii = hEi |A|Ei i. Now insert 1 = |fj ihfj |
P
j
for any choice of basis. Then

Aii = hEi ||fk ihfk |A|fj ihfj ||Ei i.

Performing the sum over i first means we are computing (U U † )jk = δjk , where U is the
unitary transformation between the two bases. This proves that Aii = hfi |A|fi i.
Now note that Aij Bji can be interpreted as Tr AB if we sum over j first or Tr BA if we
sum over i first. The trace of a product of two operators is thus independent of the operator
order.

Answer to Exercise 2.8: We evaluate Tr Aρ by summing its diagonal matrix elements in

the |ak i basis, and inserting the identity ρ = j |pj ipj hpj |. We obtain
P

X
Tr (Aρ) = ak hak |pj ipj |ak i.
jk

This has the form X

ak P (ak ),
k

where X
P (ak ) = pj |hpj |ak i|2 .
j
P
The soi-disant probabilities are obviously positive. They are less than one because j pj = 1
and the quantity multiplying each pj is the norm squared of the projection on |pj i of |ak i.
474  Quantum Mechanics

Furthermore, the sum over k just gives the norm of |pj i, which is 1, and the probabilities
P
P (ak ) then sum to j pj = 1.

Answer to Exercise 2.9: The easiest way to do this problem is by explicit matrix mul-
√ us take na na = 1, and introduce the complex number z = n1 − in2 and
tiplication. Let
n3 = ±w = ± 1 − zz ∗ . We will choose the plus sign. The matrix eigenvalue equation is
! ! !
w z √ a ∗ a
=± √ .
z ∗ −w 1−a a 1 − a∗ a

We have used the freedom to multiply eigenstates by a phase to make the lower component
zero. The absolute square of both sides of the upper component of the matrix equation is
|z|2 (1 − |a|2 ) = |a|2 (w ∓ 1)(w∗ ∓ 1).
Using the relation between z and w, we can write this as
1
|a|2 = [|z|2 ± (w + w∗ )].
2
These are the probability to be in one of the two eigenstates of na σa , if the system is in |+i3 .

Answer to Exercise 2.10: If M = U DU † , where D is diagonal and U U † = U † U = 1, then

[M, M † ] = U [D, D† ]U † = 0.
The most general 2 × 2 matrix has the form
!
a b
M= .
c d

The off diagonal part decomposes uniquely into a symmetric and antisymmetric piece, and
these are proportional to σ1 and σ2 , respectively. The diagonal part has the form A + Bn3 σ3 ,
where a = A + Bn3 and d = A − Bn3 . Now we can compute
[M, M † ] = [A + Bna σa , A∗ + B ∗ n∗b σb ] = iBB ∗ abc na n∗b σc .
This is zero only if na = βn∗a with β a complex number, which can be absorbed into B. In
other words, by redefining B we can make na real. We have seen how to diagonalize na σa for
a real 3-vector, in the previous exercise.

Answer to Exercise 2.11: Write na = βma , where ma is a unit vector so that (ma σa )2 = 1.
Then
V = eina σa = cos(β) + i sin(β)ma σa ,
 Answers to Selected Problems  475

which is diagonalizable according to the previous exercise. Using the algebra of the Pauli
matrices and the fact that their traces are all zero its easy to see that

Tr V = 2 cos(β)

and
Tr V 2 = 2[cos2 (β) − sin2 (β)] = 2 cos(2β).
The eigenvalues are thus e±iβ , which means that V is unitary of determinant one. The extra
factor eiα gives a unitary with general determinant.

Answer to Exercise 2.12:

[A, B]† = [B † , A† ] = [B, A] = −[A, B],

for any two Hermitian matrices. The matrix Gab ≡ Tr (λa λb ) is real and symmetric. If v a is
a real n2 dimensional vector then

v a v b Gab = Tr (va λa )2 .

The square of a hermitian matrix only has positive eigenvalues, so Gab is a positive definite
matrix. This means that by replacing the matrices λa by linear combinations of themselves,
we can make Gab = δab . Now multiply the commutator

[λa , λb ] = ifabc λc .

by λd and take the trace. We get

ifabd = Tr ([λa , λb ]λd ).

But the trace of a product of three matrices is independent of a cyclic permutation of their
order. So we also have

ifabd = Tr ([λd , λa ]λb ) = Tr ([λb , λd ]λa ).

Thus, fabd is antisymmetric under any transposition of indices.

Any matrix can be written M = A + v I λI , where A and v I are complex and I runs over
only n − 1 indices. We have simply identified λn as proportional to the unit matrix. The
commutator of this matrix with its adjoint is

[M, M † ] = fIJL v I v ∗J λL .
476  Quantum Mechanics

Thus, the condition for a matrix to be diagonalizable is

fIJL v I v ∗J = 0.

Answer to Exercise 2.13: Let H = U DU † , where D is diagonal. It follows that H 2 =

U D2 U † , and more generally that H k = U Dk U †. Using the power series expansion as the
definition of the exponential, we have

eiH = U eiD U † .

The matrix eiD is a diagonal matrix of phases, and so is obviously unitary. Thus, we have
written the exponential as a product of three unitary operators, which is unitary.

Answer to Exercise 2.14: We need the result that, for any two matrices, det AB =
det Adet B. As a consequence det U det U † = 1 for any unitary matrix, and det U DU † =
det D. The relation
det D = etrlnD

is obvious for diagonal matrices, and the remarks above prove that this is a basis independent
statement.

Answer to Exercise 2.15:

X
P (E, D) ≡ pnm E0n dm σ3m .

But σ3m = σ3 if m is odd and 1 if m is even. Therefore,

P (E, D) = p1 (E0 , d) + p3 (E0 , d)σ3 .

p1 is the polynomial gotten by taking only even m, i.e., with coefficients pn,2k and p3 uses
only the odd m coefficients.

Answer to Exercise 3.1: As noted in the hint, the fact that the operators U (a) all commute
and are unitary, means that there is a basis where they are simultaneously diagonalizable,
with eigenvalues eiki (a) , with real ki . We have written this in a manner appropriate for
a discrete spectrum. The generalization to continuous spectrum should be obvious. The
multiplication law of the U (a) shows that ki (a) = ki a. The operator K is the Hermitian
operator with spectrum ki .
 Answers to Selected Problems  477

Answer to Exercise 3.2: The statement that the Heisenberg equations of motion remain
invariant under a Galilean boost with infinitesimal parameter δv is

˙ = i ([δH, X] + [H, δX]).

δX
~

˙ = i ([δH, P ] + [H, δP ]).

δP
~
Since δX and δP are both proportional to the unit operator, the second term on each right-
hand side vanishes. We get
~δv = i[δH, X].
0 = i[δH, P ].
The first equation tells us that δH is nonzero, and the second that it commutes with P , and
so is a function of it. The first equation tells us that it is in fact a linear function.

δH = δvP.

However, we also know that

δH = −i[N, H]δv = δvP.
This is
m P m
P = −i[ (X − t), H] = −i [X, H].
~ m ~
P2
This equation implies that H = 2m . We have also proven that

[N, H] = iP = ~∂t N/∂t,

which means that the Heisenberg operator N (t) is time independent.

Answer to Exercise 3.3: Start by computing

i ∂E ∂ ∂E ∂ P
[±i~( − ), X] = ∓ . (G.10)
~ ∂P ∂X ∂X ∂P m

i ∂E ∂ ∂E ∂ ∂V
[±i~( − ), P ] = ± . (G.11)
~ ∂P ∂X ∂X ∂P ∂X
The Heisenberg equations of motion for Hcl are thus the same as Newton’s equations if we
choose the minus sign in the definition of Hcl . After dividing by i~, the Schrödinger equation
is then
dV P
∂t ψ(P, X) = ( ∂X ψ − ∂P ψ).
dX m
478  Quantum Mechanics

The probability distribution ρ = ψ ∗ ψ satisfies the same equation because the equation is first
order in all derivatives.
For this energy function, the classical equations of motion for a function f (X, P ) are
P dV
∂t f = ∂X f − ∂P f .
m dx
Given an initial probability distribution ρ(0, X, P ) for the variables, we compute time-
dependent expectation values via
Z
hf i = dXdP f (X(t), P (t))ρ(0, X, P ).

Alternatively, in a manner analogous to the transition between Heisenberg and Schrödinger

pictures, we can compute the expectation value as
Z
hf i = dXdP f (X, P )ρ(t, X, P ).

In order to capture the motion, the probability distribution must evolve back to the point
where X(t) = X and P (t) = P , so its equation of motion has the opposite sign from that
expected from just following the forward time evolution of a function. Since quantum expec-
tation values of functions of X and P are insensitive to the phase of the wave function, this
quantum theory makes no use of the phase unless we want to discuss operators involving ∂X
or ∂P . If we declare those operators to be “unphysical,” its predictions are entirely equivalent
to those of classical physics.

Answer to Exercise 3.5: The expection value is

(x−x0 )2
dx xn e−
R
4∆2
hX n i = R (x−x0 )2
.
dx e− 4∆2

Shifting variables to y = x − x0 , this is:

y2
dy (y + y0 )n e− 4∆2
R
n
hX i = y2
.
dy e− 4∆2
R

If we expand the integral in powers of y0 , then the coefficient of odd powers of y0k is pro-
2
portional to the integral of y n−k e−ay and vanishes if n − k is odd because of the symmetry
under y → −y. The coefficient of y0k for n − k = 2l is
y2
dy y 2l e− 4∆2
R
n!
y2
.
(k)!(2l)! R dy e− 4∆ 2
 Answers to Selected Problems  479

1
The numerator is (−1)l times the l-th derivative of the denominator with respect to w ≡ 4∆2 .
The denominator integral is
√
y2 4 ∞ 4 π
Z Z
− 4∆ −1/2 −u
dy e 2
= duu e = .
∆ 0 ∆
√
This is 8 πw, and it is easy to take the l-th derivative.

Answer to Exercise 3.7: Let

2πk
Un |ki = ei N |ki,
where k is an integer modulo n. The shift operator is

Vn |ki|k + 1i,

where the addition is mod n, i.e., k + n ≡ k. Then

2π(k+1)
Un Vn |ki = ei N |k + 1i,

and
2πk
Un Vn |ki = ei N |k + 1i,
so the required operator algebra is satisfied.

Answer to Exercise 3.9: By expanding the exponential, it is easy to see that

U k Pθ U −k
U k Vα U −k = eiα ~ .

Now let us evaluate

~ ∂ −ikθ
Pθ U −k f (θ) = [e f (θ)] = −i~ke−ikθ f (θ) + e−ikθ Pθ [f ].
i ∂θ
It follows that
U k Pθ U −k = Pθ − ~k,
so that
U k Vα U −k V−α = e−ikα .

Answer to Exercise 4.1: Consider the unitarity equation U † U = 1. The rule of matrix
multiplication is that the ij matrix element of the product matrix AB is the dot product (no
complex conjugation) of the i-th row of A with the j-th column of B. The i-th row of U † is
the complex conjugate transpose of the i-th column of U , so the ij matrix element of U † U
480  Quantum Mechanics

is the scalar product between the i and j columns of U . Unitarity is just the statement that
these are an orthonormal set. Since their number is equal to the dimension of the Hilbert
space, they are an orthonormal basis.

Answer to Exercise 4.2: If X

U= |ek ihfk |,
k

so that U maps the f basis into the e basis, then U † , its conjugate transpose is

U† =
X
|fk ihek |.
k

Then
U †U =
X X X
|fk ihek | |el ihfl | = |fk ihfk | = 1.
k l k

The last step is the resolution of the identity for the basis f . The opposite order U U † = 1. To
show that any unitary transformation can be written in this way, start with any orthonormal
basis |fk i. The column vectors representing |ek i = U |fk i in the |fk i basis are just the columns
of the matrix of U in that basis. The unitarity condition is the statement that those vectors
form a new orthonormal basis.

Answer to Exercise 4.3:

2 +bx
ψ(x) = e−ax .
dx dx
Z Z
2 (b−ik) 2 1 2
ψ̃(k) = √ e−ikx e−ax +bx = √ e−a(x− 2a ) e 4a (b−ik) .
2π 2π
Now shift and rescale the variable of integration

x − (b−ik)
y= √ 2a ,
a
and obtain
√ Z dy −y2 1 (b−ik)2
a √ e e 4a .
2π
Now Z Z
2 2 2
dy e−y = [ dy1 dy2 e−(y1 +y2 ) ]1/2 ,

and the latter integral can be done in radial/angular coordinates d2 y = dφrdr. The integrand
is independent of π, so we get
Z
2
Z ∞ 2 √
dy e−y = [π d(r2 )e−r ]1/2 = π.
0
 Answers to Selected Problems  481

The end result is

1
q
2
ψ̃(k) = e 4a (b−ik) a/2.
ψ(x) = e−a|x| .
∞
r
dx 1 1 1 2 a
Z
ψ̃(k) = √ [e−ikx + eikx ]e−ka = √ [ + ]= .
0 2π 2π a + ik a − ik π k + a2
2

ψ(x) = sθ(a − x)θ(x − b).

Z ∞ Z a
dx is
ψ̃(k) = s √ e−ikx = √ (e−ika − e−ikb ).
0 2π b 2π

Answer to Exercise 4.4: A smooth function has an infinite number of continuous deriva-
tives. If we use F 2 = R to write f (−x) as the Fourier transform of the Fourier transform of
f (x), we have
dn dk −ikx ˜
Z
n
f (x) = e f (k)(−ik)n . (G.12)
dx 2π
The existence of all of these derivatives implies that these integrals all converge, so that the
Fourier transform falls off faster than any power. The Fourier transform of (x2 + a2 )−b is
∞ sb−1
Z Z
2 +a2 )
f˜(k) = dxe ikx
ds e−s(x .
0 Γ(b)

We can do the Gaussian integral over x, obtaining,

√ Z ∞ 2 k2 sb−3/2
f˜(k) = π ds e−sa e− 4s .
0 Γ(b)

Rescale s = kt so that the integral becomes

√ b−1/2 Z ∞ 2 k tb−3/2
f˜(k) = πk dt e−kta e− 4t .
0 Γ(b)

For large k we can do this integral by the steepest descent method. Its behavior is domi-
nated by the stationary point of the function in the exponential t∗ = 1/4a2 . f˜(k) falls off
exponentially, but the exponential is multiplied by a larger power of k for larger b.

Answer to Exercise 4.5 The Fourier transform is just

a dx 1 eika − 1
Z
θ̃(k) = √ e−ikx = √ . (G.13)
0 2π 2π ik
482  Quantum Mechanics

This vanishes for ka  1. The function is complex and oscillates in the regime ka ∼ 1, while
in the regime ka  1, the first term oscillates so rapidly that it vanishes when averaged over
small intervals in k and so in this regime, we have an almost featureless 1/k falloff. If this
√
is the wave function of a particle at time t = 0, it should be multiplied by 1/ a in order to
have proper normalization. It evolves into

ika
dk ~k2 t e −1
Z
ψ(x, t) = √ e−ikx−i 2m √ .
2π ik a

Factor out √1 and take the derivative with respect to a to get

2πa

dk
Z
~kt
√ e−ik(x−a)−i 2m .
2π

If it is positive this is a Gaussian integral. It is defined by analytic continuation from this

regime to the one where t is positive. So the result is

im(x−a)2
e− 2~t .

Our wave function is the integral of this expression between 0 and a. If |x|  a, the integrand
is very rapidly oscillating and the integral is very small, until ~t ∼ mxa. After that time, the
integral is order 1. Thus, the wave function falls off very rapidly outside the original interval
~
but it spreads to cover the region out to x at a velocity am . This is the velocity corresponding
to the momentum uncertainty in the original wave function.

Answers to Exercises 4.6–4.8: The solution to the Schrödinger equation with energy E is
q
2m(V (x)−E)
± ~2
ψE = e , (G.14)

where V (x) = V0 θ(a−x)θ(x+a). Integrating the Schrödinger equation near the discontinuities
in V (x), we find that the wave function and its first derivative must be continuous there. If
E > 0 we have delta function normalizable solutions, while if E < 0 we can get normalizable
bound state solutions by picking only the falling exponential as x → ±∞. The problem is
invariant under reflections, so for the bound states we can choose to look at even and odd
solutions. We can impose the continuity conditions only at x = a and they will automatically
be satisfied for negative x. In the region near the origin, the wave function depends on the
crucial quantity
 Answers to Selected Problems  483
s
2m(V0 − E)
ik0 = , (G.15)
~2
which is imaginary. The imaginary part of ak0 is equal to the classical action in units of ~ of
a particle traveling from −a to a, if E − V0 > 0. When the opposite inequality holds, k0 is
imaginary. In either case, we define k0 with the positive (imaginary) square root. We define
s
√ 2m(−E)
ik = −E . (G.16)
~2

In addition, define r ≡ kk0 . In the region |x| > a, the wave function grows or falls exponentially
if E < 0. We must choose the falling solution at both positive and negative values of x,
to obtain a normalizable wave function. The continuity conditions for even and odd wave
functions at x = a are

A cosh(|k0 |a) = Be−|k|a , Ar sinh(|k0 |a) = −Be−|k|a , (G.17)

A sinh(|k0 |a) = Be−|k|a , Ar cosh(|k0 |a) = −Be−|k|a , (G.18)

when k0 is imaginary. The hyperbolic sine and cosine are both positive, and these equations
have no solution. Thus, we must have the intuitively obvious condition E − V0 > 0, which
implies a potential well rather than a barrier and asymptotic energy less than the depth of
the well, in order to have a bound state. For k0 real, the matching conditions become

A cos(|k0 |a) = Be−|k|a , Ar sin(|k0 |a) = −Be−|k|a , (G.19)

A sin(|k0 |a) = Be−|k|a , Ar cos(|k0 |a) = −Be−|k|a . (G.20)

Motivated by the bound on E, we define E = yV0 , with 0 ≤ y ≤ 1. The conditions become
y
r p
= − tan(s0 1 − y). (G.21)
1−y
y
r p
= − cot(s0 1 − y). (G.22)
1−y
q
These depend only on a single parameter s0 ≡ a −2mV ~2 . When s0 is small, we can expand
0

the trigonometric functions. The equation for odd solutions becomes y = 1/s0 , which is
inconsistent with 0 ≤ y ≤ 1. The even solution is y = s0 and is consistent. There is a single
√
bound state. When s0 is large, we can satisfy the equation with s0 1 − y ∼ nπ 2 , where n
nπ
is odd for even solutions and vice versa, as long as 2s0 < 1. The point is that the tangent
and cotangent take on any large value in the vicinity of their poles, so that as long as y is
484  Quantum Mechanics

sufficiently close to 1 we can match. These explicit formulae are only valid for large n, but
there are bound states for every value of n satisfying the inequality. As s0 goes to infinity, we
get an infinite number of states and the explicit formula becomes more exact. In this limit,
after adding a constant to make all the energies positive, the spectrum approaches that of
the infinite square well. Note, however, that s0 can be large for a shallow, but very broad,
well also. In that situation, we would still have continuum eigenstates.
Turning now to E > 0, we define
ikx −ikx
ψ + = A+
out e + A+
in e , x > a, (G.23)

−ikx −ikx
ψ + = A+
out e + A+
in e , x < −a. (G.24)
~2 k2
The subscripts in and out refer to the fact that, when multiplied by e−i 2m t , the relevant
part of the solution becomes a traveling incoming or outgoing wave. To write the continuity
±
conditions at x = ±a compactly, it is convenient to define z = eik0 a and αin = A±in e
−ika
±
αout = A±out e
ika
. Then we have

Az + Bz −1 = αout
+ +
+ αin , (G.25)
−1 − −
Az + Bz = αout + αin , (G.26)
−1 + +
r(Az − Bz ) = αout − αin , (G.27)
−1 − +
r(Az − Bz) = − αout + αin . (G.28)

We can solve these equations by adding and subtracting the first and third and also the
second and fourth, obtaining:
+
2αout = (1 + r)zA + (1 − r)z −1 B, (G.29)
− −1
2αout = (1 − r)z A + (1 + r)zB, (G.30)
+ −1
2αin = (1 − r)zA + (1 + r)z B, (G.31)
+ −1
2αout = (1 + r)z A + (1 − r)zB. (G.32)

These equations reflect the fact that there are only two linearly independent solutions of the
Schrödinger equation. The in and out states represent two independent bases for the Hilbert
space. These equations tell us how to transform between them. The matrix relating them is
−1
S = S+ S− , where S± the matrices exhibited above, relating the in and out bases to the
A, B basis. Using the usual formula for the inverse of a 2 × 2 matrix, we get
!
1 (1 − r2 )(z − z −1 ) −4r
S= . (G.33)
(1 − r) z − (1 + r)2 z −2
2 2 −4r (1 − r2 )(z − z −1 )
 Answers to Selected Problems  485

Answer to Exercise 4.9: The reflection amplitude is of course the same ratio of reflected
to incident wave
A− 2
R = (| out |) .
A−in

The out amplitudes are related to the in amplitudes by the S-matrix

+ − −
A+
out = S11 Ain + S12 Ain = S12 Ain .

+ − −
A+
out = S21 Ain + S22 Ain = S22 Ain .

The transmission and reflection coefficients are just the absolute squares of the S12 and S22
coefficients of the S-matrix which is the norm of the vector formed by the second column of
the matrix. Unitarity of the S-matrix guarantees that this is 1. We calculated the S-matrix
in Exercise 4.8
−4r
S12 = .
(1 − r) z − (1 + r)2 z −2
2 2

Recall that
k0
r= , z = eik0 a .
k
Thus,
4
T =| |2 .
2(1/r + r) sin(k0 a) − 4 cos(k0 a)
In the range of energies where k0 is real, but r < 1, this oscillates as a function of energy. At
points where sin(k0 a) = 0, T = 1, exhibiting the Ramsauer–Townsend effect.
Answer to Exercise 4.11: The solution of the Heisenberg equation of motion is X(t) =
X + Pmt . The uncertainty in this operator is the square root of

t t2 P 2 t2
h|X 2 + (XP + P X) + i − (hP i)2 .
m m2 m2
For large t, this is dominated by the uncertainty of P in the wave function ψ. Heisenberg’s
uncertainty relation says that ∆P ≥ ~∆X. Assuming minimal uncertainty, this implies a
position uncertainty at time t of order

t~
∆X(t) ∼ .
m∆X(0)

So the 1 m uncertainty is achieved at time

t = m(1 m)∆X/~,
486  Quantum Mechanics

2
where ~ = 10−34 kg sm . The time in seconds is

t = 1034 m∆Xs,

where the mass is expressed in kilograms and the initial position uncertainty in meters. For
a baseball with ∆X = .001 and m = .5, this is 5 × 1030 s or about 5 × 1023 years. For the
moon, with the same initial uncertainty, the time becomes about 4 × 1046 years.
Answer to Exercise 4.13: Write Pn (x) = nk=0 Cnk xk . By definition, Pn is orthogonal to
P

all of the lower Pk , but since xk for k < n is a linear combination of the lower Pk , we have
Z n
X n
X
k l
dx r(x) Cnk x x = Cnk rk+l = 0
k=0 k=0

for each l with 0 ≤ l < n. This says that the n + 1 dimensional vector whose k-th component
is Cnk is orthogonal to all n of the n + 1 dimensional vectors Rl , where 0 ≤ l < n and Rl is
the vector whose k-th component is rl+k . Cnk is thus given by

Cnk = cn ka1 ...an r1a1 . . . rnan ,

where  is the n + 1 dimensional Levi-Civita symbol. This is equivalent to the determinant

formula quoted in the exercise. The norm of the polynomial is easily computed in terms of
the moment matrices Z n X
dx r(x)Pn2 (x) = Cnk rkl Cnl ,
k,l=0

and this determines cn .

Answer to Exercise 4.15: The multiplication operator by the function c is obviously

Hermitian if c is real, so it is sufficient to study the other two terms. Using integration by
parts (d ≡ d/dx) Z Z
rf ∗ (ad2 g + bdg) = [d2 (rf ∗ a)g − d(rbf ∗ )g].

rg(a∗ d2 f ∗ + b∗ f ∗ ), which would give Hermiticity if a and b were

R
In order for this to be
real, we need
d2 (ra) − d(rb) = 0, and d(ra) = rb.
The two constraints are compatible, and evaluate b in terms of a.

Answer to Exercise 4.17:

d df
Z Z
dx f (x) θ(x) = − dx θ(x),
dx dx
 Answers to Selected Problems  487

as long as f (x) vanishes at the endpoints of the integration. If the integral of integration
does not include zero, this expression vanishes: if it is all on the negative axis θ(x) = 0. If
it is all on the positive axis, then the integral is just the difference of the values of f at the
endpoints, both of which vanish. If the interval includes zero, then
Z
d
Z a df
dx f (x) θ(x) = − dx = f (0) − f (a) = f (0).
dx 0 dx

Answer to Exercise 4.18: The residue theorem states that the integral of a function with
only poles, around a counterclockwise contour surrounding some of the poles is equal to 2πi
times the sum of the residues of the poles inside the contour. In our case, we have a single
pole, at s = i. If x > 0 then in the upper half plane the integrand falls exponentially, so
we can add a circle at infinity to make a closed contour, which is followed counterclockwise.
We thus pick up the residue at the pole, which gives the value 1. If x < 0 we can close the
contour in the lower half plane, and get zero, because there are no poles.

Answer to Exercise 4.19: Since the potential goes to infinity outside the well, the contri-
bution to the expectation value of the Hamiltonian in any state that has ψ ∗ ψ 6= 0 outside the
well will be infinite. Therefore, all finite energy eigenstates vanish outside the well, and they
are all bound states. Now integrate the Schrödinger equation in a tiny interval straddling the
point x = a. We get
dψ
−[ ]a+ ∝ |V0 |ψ(a + ),
dx a−
since we have set the potential to zero inside the well. The derivative vanishes at a +  so
the derivative at the wall is related to the wave function at the wall, by a relation of the
form dψ/dx = cψ. The constant c is a limit of the product of the infinite potential and the
vanishing wave function outside the wall, and so can be any number. Indeed, the Hamiltonian
is a Hermitian operator for any choice of c, as one can verify by integration by parts. A similar
boundary condition must be imposed at x = −a. The solutions inside the well are of course
2 2
eikx with E = ~2mk . The boundary conditions are

ik(Aeika − Be−ika ) = c(Aeika + Be−ika ),

ik(Ae−ika − Beika ) = c(Ae−ika + Beika ).

These lead to two evaluations of B/A, which agree only if

e4ika = 1,

unless c = 0. In that case, both conditions are solved by sin(ka) = 0. In all cases, we have
only a discrete spectrum of energies.
488  Quantum Mechanics

Answer to Exercise 4.21: The solutions have the form A± e±kx with k > 0 for x > a and
x < −a, respectively. We take A− = 1, A+ = A. In the region −b ≤ x ≤ b, the solution
has the form C+ ekx + C− e−kx and in the interval [−a, −b], we have B+ eik0 x + B− e−ik0 x .
s k2
Finally, in the interval [b, a], we have D+ eik0 x + D− e−ik0 x . The eigenvalue is E = − ~2m and
q
2 2
~k0 = 2m(V0 − ~2mk ). Define α = eak , β = ebk , a0 = eik0 a , b0 = eik0 b . We have continuity
conditions at ±a and ±b. These are

−a : α−1 = B+ a−1
0 + B− a0 , kα−1 = ik0 (B+ a−1
0 − B− a0 ),

−b : C+ β −1 + C− β = B+ b−1
0 + B− b0 , k(C+ β −1 − C− β) = ik0 (B+ a−1
0 − B− a0 ),

b : C+ β + C− β −1 = D+ b0 + D− b−1
0 , k(C+ β −1 − C− β) = ik0 (D+ b0 − D− b−1
0 ),

a : α−1 A = D+ a0 + D− a−1
0 , −kα−1 A = ik0 (D+ a− −1
0 D− a0 ).

These can be viewed as matrix equations of the form

M1 B = v, M2 B = M3 C, M4 C = M5 D, M6 D = w.

The vectors v and w are v = α−1 (1, 1) and w = Aα−1 (1, −1). The vectors B, C, D are
composed of (B+ , B− ), etc. We can solve for B in two different ways

B = M1−1 v = M2−1 M3 M4−1 M5 M6−1 w.

This gives us two conditions on the two free parameters A and k, so there will only be discrete
solutions for k. To argue that the ground state energy is lowered when b is decreased with
b − a fixed, we take the derivative

(∂a + ∂b )[hψ(a, b)|H|ψ(a, b)i].

Using the fact that ψ is an eigenstate, this is

E0 (∂a + ∂b )[hψ(a, b)|ψ(a, b)]i + hψ(a, b)|(∂a + ∂b )H|ψ(a, b)i.

The first term vanishes because ψ is normalized to 1 for all values of the parameters. The
second term is the expectation value of

− V0 [θ(x − b)θ(b + x)(δ(a − x)θ(a + x) + δ(a + x)θ(a − x))

− θ(x + a)θ(a − x)(δ(x − b)θ(b + x) + δ(x + b)θ(b − x))].

The first term vanishes because b < a while the second is positive. Thus, the ground state
energy is monotonically increasing as b is increased with b − a fixed.
 Answers to Selected Problems  489

Answer to Exercise 4.23: The wave functions at ±∞ give rise to currents:

J± = i(ψpm ∗ ∂x ψ± − ψ± ∂x (ψ± ) = −2k(A∗ ±A± − B± ∗ B± ).

This current must be the same at plus infinity as at minus infinity because the solution is
time independent and the current conserved. Thus,

A+ ∗ A+ + B− ∗ B− = A− ∗ A− + B+ ∗ B+ ,

which is the equation saying that the S-matrix preserves the norm of vectors. Consequently
S is unitary.

Answer to Exercise 5.1:

hs|H 2 |si = kH|sik2 , (G.34)
since Hermiticity of H implies that hs|H is the bra corresponding to the ket H|si. Thus,
expectation values of H 2 are always nonnegative, and the same is true for sums of squares
of Hermitian operators. Applying this result to the case where |si is an eigenstate of H, we
find that the eigenvalues of H 2 are all nonnegative.

Answer to Exercise 5.3: 5.3a is a trivial exercise in rescaling variables. To solve 5.3b,
write the equation for the logarithm S of the wave function

−(∂y2 S + (∂y S)2 ) + y 2 = 2.

At large y, this is solved by q

∂y S = ± y 2 ,
because, given this ansatz, both the second derivative term and the  term are smaller by
two powers of y.
y2
5.3c: If ψ = e− 2 v, then

−∂y2 v + y∂y v = (2 − 1)v.

P∞ n
Writing v = n=0 vn y , we obtain the recursion relation

(n + 2)(n + 1)vn+2 = (n + 1 − 2)vn .

This shows that we can study even and odd solutions separately. At large n, we get
1
vn+2 ∼ vn ,
n
490  Quantum Mechanics

which is solved by v2k = k!1 in the even case, and the same formula for v2k+1 in the odd case.
Thus, we get back the bad exponential behavior unless the recursion stops. This can only
happen if 2 = n + 1, so we have derived the quantization of the energy levels.

Answer to Exercise 5.5: The oscillator ground state wave function is

mω 1/4 − mωx2
ψ0 = ( ) e 2~ .
π~
The overlap of two such wave functions is a Gaussian integral, which gives
ωω 0
hψ0 (ω)|ψ0 (ω 0 )i = (2)1/2 ( )1/4 < 1.
(ω + ω 0 )2
Answer to Exercise 5.7: Recall that
1
a† ≡ √ (mωX − iP )
2~mω
and
1
a≡ √ (mωX + iP ).
2~mω
Introduce the dimensionless variable y by
s
~
x= y.
mω
Then
1 d
a† ≡ √ (y − ),
2 dy
1 d
a = √ (y + ).
2 dy
Write the wave function of the n-th eigenstate as
mω 1/4 y2
ψn = ( ) Hn (y)e− 2 .
π~
Note that this is not the standard mathematical normalization of the Hermite polynomials,
which takes the coefficient of the highest power of y to be 1. It is also different from what is
called the “physicist’s” normalization in the Wikipedia article on Hermite polynomials.
With this normalization
y2 1 y2 √ 1/2 y2
a† Hn e− 2 = √ (2yHn − Hn0 )e− 2 = n + 1 Hn+1 e− 2 ,
2
 Answers to Selected Problems  491

so √ √
(2yHn − Hn0 ) = n + 1 2Hn+1 .
Similarly, √ √
n 2Hn−1 = Hn0 .

Answer to Exercise 5.9: The uncertainty of A is defined as

hA2 i − hAi2 .

The Hilbert space of the K–G field is the tensor product of individual oscillator Hilbert
spaces, and a coherent state is a tensor product of coherent states of different oscillators. If
we take operators Ai acting in different tensor factors, then

hAi Aj i = hAi ihAj i.

Thus, in computing the uncertainties of N and H, which are sums over modes, we need only
sum the uncertainties for individual modes. The key part of the calculation is
∞
hz|(a† a)2 |zi 1 2
= e−zz̄ [
X
n (z z̄)n ].
hz|zi n=0
n!

Using n2 = n(n − 1) + n we write this as

∞
hz|(a† a)2 |zi 1
= e−zz̄ [
X
(z z̄)2 + z z̄)(z z̄)n ] = (z z̄)2 + z z̄).
hz|zi n=0
n!

We have dropped terms in the sums that are zero and renamed the two summation variables.
A similar and simpler calculation gives

hz|(a† a)|zi
= z z̄.
hz|zi
Therefore, we have
z(k)z ∗ (k),
X
hN 2 i − hN i2 =
k

ω(k)z(k)z ∗ (k).
X
hH 2 i − hHi2 =
k

This calculation is obviously valid for any dispersion relation ω(k) for the waves, and so is
more general than the Klein–Gordon equation. Note that if zz ∗ is large in some range of k
the uncertainties, which are the square roots of the above combinations of expectation values,
492  Quantum Mechanics

are much less than the expectation values of the operators themselves. This is the basis for
the argument that the coherent states behave like classical fields.

Answer to Exercise 5.11: Consider a single annihilation operator. Since a2 = 0, there

must be a subspace of states satisfying a|s0 i = 0. The block decomposition of a in the basis
formed by this subspace and its orthogonal complement is
!
A 0
a= ,
B 0

so that !
† A† B †
a = .
0 0
As a consequence !
AA† ± (B † B + A† A) AB †
aa† ± a† a = .
BA† BB †
So, we must have
AB † = BA† = 0
and
AA† ± (B † B + A† A) = 1 = BB † .
The first of these equations is generally for a rectangular matrix and the second is actually
for two square matrices of (possibly) different dimensions. Since the products of the two
(possibly) rectangular matrices are nonnegative Hermitian operators and BB † = 1, the only
way to satisfy these equations is to set A = 0 and choose the plus sign. B is then a unitary
operator on a space of dimension half the original space and
!
0 0
a=B⊗ .
1 0

Notice that we have also solved Exercise 5.12, by taking the dimension of the space B acts
on to be one dimensional. For multiple fermionic creation and annihilation operators, we just
write ai = αi ⊗ σ3 for i > 1. The αi anticommute with each other, commute with B, and
square to zero.

Answer to Exercise 5.13: Start from an infinite collection of commuting copies of the
Pauli matrices
[σa (i), σb (j)] = 2iδij abc σc (i).
 Answers to Selected Problems  493

Choose an ordering of the indices i as points on an infinite one-dimensional lattice. Define

Q
a(i) = σ+ (i) j>i σ3 (j). Then each a(i) squares to zero and the different a(i) anticommute
because (assuming without loss of generality that j > i) the σ3 (j) in a(i) anticommutes with
σ+ (j). If the indices are arranged on a higher dimensional lattice, then the Jordan–Wigner
construction provides operators that depend on a choice of a line from the point where the
operator a(i) sits, out to infinity.

Answer to Exercise 5.14: The Heisenberg equation of motion is

i
ȧ = [H, a] = −iωa,
~
so that
a(t) = e−iωt a.
Thus,
H H
ae−i ~ t |zi = e−iωt ze−i ~ t |zi. (G.35)

This tells us that the time evolved state is proportional to a coherent state with z(t) =
e−iωt z. Since the time evolved state has the same norm as the original coherent state, the
proportionality constant must be a pure phase. Indeed, that phase comes from the constant
term in the Hamiltonian (which doesn’t change the Heisenberg operator a(t) but just changes
ω
the state by a time-dependent phase). Thus, the phas is just e−i 2 t . Note that the equation
for the coherent state with z = 0 is the same as the equation for the ground state of the
Hamiltonian, and this is consistent with the evolution law above.

Answer to Exercise 6.1: In components, the inequality reads

qX
vi∗ wi | ≤
X X
| vi∗ vi wi∗ wi .

We can divide through by the norms of the vectors, so that the inequality says that the
scalar product of two unit vectors has absolute value less than 1. The absolute value of
the sum is obviously maximized if all terms in it have the same phase, so we can restrict
attention to the case where all components are positive. Now consider that 0 ≤ (vi ±wi )2 =
P
P 2
(vi + wi2 ± 2vi wi ) = 2 ± 2 vi wi . This is precisely the required inequality.
P

Answer to Exercise 6.3: The equations

[A, A† ] = 0
494  Quantum Mechanics

are N 2 complex equations for N 2 complex unknowns. Writing A = H1 + iH2 , where Hi are
Hermitian, normality implies H1 commutes with H2 . The most general Hermitian matrix
has N 2 real parameters. Given such a matrix, the condition that a second Hermitian matrix
commutes with it puts N 2 − N real constraints on those parameters. To see this, note that
the equation
[H1 , H2 ] = 0
looks like N 2 real conditions, but it is automatically satisfied if H2 = H1k . A generic Hermitian
matrix will, by the Cayley Hamilton theorem, satisfy its characteristic polynomial PH1 (H1 ) =
0, but no lower order equation, so the powers k = 0, . . . , N − 1 will be linearly independent
matrices. Thus, the commutator condition is, generically, N 2 − N constraints and a normal
operator has N 2 + N real parameters.
A unitary matrix has N 2 real parameters, but in the formula

A = U † DU,

we can ignore unitaries that differ by U1 = UD U2 , where UD is diagonal in the same basis
as D. There are N independent unitary matrices of this form, each of which depends on one
real phase. The complex diagonal matrix D depends on 2N independent real parameters,
so diagonalizable matrices have the same number of independent real parameters as normal
matrices. They are all normal. The spectral theorem, whose proof is sketched in the text,
shows that all normal matrices are diagonal in some orthonormal basis.

Answer to Exercise 6.5: Insert a parameter t so that we are trying to compute

t2 t3
E1 ≡ et(A+B) = 1 + t(A + B) + (A + B)2 + (A + B)3 + o(t4 ).
2 3!
On the other hand,
t2 t3 t2 t3
E2 ≡ etA etB = (1 + t(A) + (A)2 + (A)3 + o(t4 ))(1 + t(B) + (B)2 + (B)3 + o(t4 )).
2 3! 2 3!
The two expressions agree up to order t, but at order t2 we have
1 2
(A + B 2 + AB + BA),
2
for E1 and
1 2
(A + B 2 + AB),
2
for E2 . To this order, we can fix things by writing
t2
E1 = E2 e− 2 [A,B] ,
 Answers to Selected Problems  495

but at order t3 this gives

t3 3
(A + B 3 + AB 2 + BAB + B 2 A + BA2 + ABA + A2 B)
3!
t3 t2
6= (A3 + B 3 ) + (A2 B + AB 2 − (AB + BA)).
3! 2
This can be fixed by writing
t2 t3
E1 = E2 e− 2 [A,B] e 6 (2[B,[A,B]+[A,[A,B]]) .
The term of order tk involves k fold commutators. The BCH formula can be derived in an
analogous manner. The most interesting case of these identities occurs when the commutator
algebra closes after a finite number of iterations. Such a structure is called a finite dimensional
Lie algebra.

Answer to Exercise 6.7: The matrix of V in the k basis is

 
0 1 0 ... 0 0
0 0 1 0 . . . 0
 
. . . .. .. 
V = . .
. . .. . . .

0 0 0 . . . 1
 

1 0 0 0 ... 0
Its Hermitian conjugate simply has all the ones on the diagonal just below the main diagonal,
except for one in the upper right-hand corner. This is the matrix that rotates the clock in
the opposite direction, so V † = V −1 . The equation
2πi
UV = V Ue N
is simply the mathematical formula that says that V shifts the clock by one unit. The matrix
V k has ones on the diagonals k units above the main diagonal and k −1 units above the lower
2ilπ
left corner. The matrix U l V k multiplies the l-th row of V k by e N . Thus, these matrices are
all linearly independent. Since there are N 2 of them, they form a basis for all matrices.

Answer to Exercise 6.9: The formula

Z 1
hf |gi = µ(x)f ∗ (x)g(x)
−1
1
µ(x)f ∗ (x)f (x) < ∞.
R
defines a scalar product on the space of square integrable functions −1
The monomials xn are linearly independent, and if the moments are all finite we have
hxn |xm i = Mn+m ,
496  Quantum Mechanics

so they are not orthonormal. Define

k
X
Pk (x) = pkn xn .
n=0

Then

hPk |Pl i = p∗kn Mn+m plm .

Considered as a matrix in the indices m, n, Mn+m is real and symmetric, and so can be
diagonalized by an orthogonal transformation. That is

Mn+m = Onp mp Omp ,

with
Onp Omp = δnm .
If we take
−1/2
pkn = p∗kn = Onk mk ,
we have constructed orthonormal polynomials.

Answer to Exercise 6.11: We have

U † (x)U (x) = (x0 − ixj σj )(x0 + ixk σk ) = x20 + xj xk σj σk = x20 + xj xk (δjk + ijkl σl ).

The last term vanishes because jkl is antisymmetric and xk xl symmetric under interchange of
the summation indices. Thus, U † U = 1. Writing x0 = cos(θ), xj = sin(θ)ej , where ej ej = 1,
we can write U = eiθej σj . Then detU = etr iθej σj = e0 = 1. Since the Pauli matrices and the
unit matrix are a complete set, we get the most general matrix by letting xa be four arbitary
complex numbers. Then

U † (x)U (x) = (x∗0 − ix∗j σj )(x0 + ixk σk ) = x∗a xa + i(x0 ∗ xj − x∗j x0 ) + ijkl x∗k xl )σj .

The cross product between the three-vectors xj and x∗j is perpendicular to both of those
vectors, so unitarity requires it to vanish, and also that x∗0 xj = x∗j x0 . This means that the
complex three vector is an overall phase times a real three vector and that the phases of x0
and xj are equal. Thus, the most general unitary 2 × 2 matrix is an overall phase multiplied
by a matrix with real xa . The determinant of the matrix is just the square of the phase and
so is equal to 1 only if the phase is real.

Alternate Answer to Exercise 6.11: By cyclicity of the trace,

Tr [U † (xa )σi U (xa )] = Tr [σi ] = 0.

 Answers to Selected Problems  497

Since any matrix is a linear combination of the unit matrix and the Pauli matrices, and we
have just proven that the coefficient of the unit matrix is zero, we indeed have

U † (xa )σi U (xa ) = Rij σj .

Now consider

δij = Tr[σi σj ] = Tr [U † (xa )σi σj U (xa )] = Tr [U † (xa )σi U (xa )U † (xa )σj U (xa )]
= Rik Rjl Tr [σk σl ] = (RRT )ij .

Answer to Exercise 6.13: Since the subalgebra is closed under Hermitian conjugation
C commutes with both a and a† for every member a of the algebra, which implies that
C † also commutes with every element of the subalgebra. Thus, the Hermitian operators
(C±C † )i1∓1 , commute with every a. The eigenspaces with fixed eigenvalue, of these operators,
would be invariant subspaces, unless all their eigenvalues are equal, which implies that C ±C †
are both proportional to the unit matrix.

Answer to Exercise 7.1:

[La , Lb ] = aij bkl [Ri Pj , Rk Pl ].

Using Leibniz’ rule, and the fact that the only nonzero commutator is [Ri , Pj ] = i~δij , we
have
[Ri Pj , Rk Pl ] = Ri [Pj , Rk ]Pl + Rk [Ri , Pl ]Pj = i~[δil Rk Pj − δjk Ri Pl ].
Now use
aij bkl δil = δak δbj − δab δij ,
and permutations of this identity, to conclude that

[La , Lb ] = i~abc Lc .

Answer to Exercise 7.3: The trick for doing this computation is to note that
l−m l−m l−m−1 l−m−1
hll|K+ K− |lli = hll|K+ (K2 − K32 + K3 )K− |lli.
l−m−1
The state K− |lli is an eigenstate of K2 with eigenvalue l(l + 1) and an eigenstate of K3
with eigenvalue m + 1. Thus,
l−m l−m l−m−1 l−m−1
hll|K+ K− |lli = [l(l + 1) − (m + 1)2 + m + 1]hll|K+ K− |lli.
498  Quantum Mechanics

Iterating this equation, we get

l−m
Y
l−m l−m
hll|K+ K− |lli = [l(l + 1) − (m + j)2 + m + j],
j=0

where we have used the fact that

hll|lli = 1.
Answer to Exercise 7.5: The Bohr radius is
4π0 ~2 ~2
ag = → .
me e2 m2 GM
The analog of the Rydberg energy is

~2 m(mGM )2
= .
2ma2B 2~2

Now GM m = 6.7 × 10−7 M m

kg 2 = 6.7 × 10
48
in joule-cm. m = 6 × 1024 kg and ~ = 10−34
joule-s. This gives a gravitational Bohr radius of 2.3 × 10−138 m, and a Rydberg energy of
1.35 × 10188 joules! In fact this calculation is not valid in the real world. The nonrelativistic
treatment of the “gravitational Bohr atom” is valid only for very highly excited states En
for which the effective Bohr radius is nag . Requiring that expectation value of the velocity
be less than that of light gives us this constraint. The expectation value of p2 /2m in the nth
state is of order the binding energy

10188
joules,
n2
and this must be < mc2 = 5.4 × 1041 joules. Thus, n > 1073 .

Answer to Exercise 7.7: For a nonzero constant solution, the first two terms in the equation
vanish, while the last vanishes only if q = 0. The constant is undetermined. For a linear
solution, A + Bx the first term vanishes so we have

(p + 1 − x)B + q(A + Bx) = 0,

which implies q = 1 and A = (p + 1)B. B is undetermined.

Answer to Exercise 7.9: The Laguerre operator can be written

d2 d d 1 d
L≡x + (p + 1 − x) = [x + (p + 1 − x)] [x ].
dx2 dx dx x dx
 Answers to Selected Problems  499

Now note that, using integration by parts

d d
Z Z
dx [f ∗ xp e−x x g] = − dx [x + (p + 1 − x)](f ∗ )xp e−x g,
dx dx
so that, with respect to this scalar product
d d
x + (p + 1 − x) = −[x ]† .
dx dx
d
Define D = x dx . Then
1
L = −D† D.
x
This shows that L is a negative definite Hermitian operator. There is a subtlety in this
derivation, because x1 is not defined on the whole Hilbert space, since it blows up on constant
functions if p is not large enough. However the action of D to the right, kills the divergence.
To compute the norm of Lpq we recall its explicit power series expansion (this is the expansion
for the monic polynomial, rather than the orthonormal one we described in the text):
q
X Γ(q + p + 1)
Lpq = (−1)k xk .
k=0
Γ(q − k + 1)Γ(p + k + 1)Γ(k + 1)

In evaluating the norm of this function, we have to do the integral

Z ∞
dx e−x xp+k1 +k2 = Γ(p + k1 + k2 − 1).
0

Answer to Exercise 7.11: The fact that the equation of motion is

dv a
= − e˙φ ,
dt l
follows by simply combining Newton’s equation with the kinematic equation for eφ . Since
both left and right sides of this equation are total time derivatives,
a
h ≡ v + eφ
l
is conserved. h is sometimes called Hamilton’s vector, since the fact that it is conserved is
his observation.

Answer to Exercise 7.13:

h × L = v × L + aeφ × L̂,
500  Quantum Mechanics

where L̂ is the unit vector in the direction of the angular momentum. The cross product of
two orthogonal unit vectors (the angular momentum vector is perpendicular to the plane of
the orbit) is the third unit vector in an orthonormal basis, which is just ±ê, depending on
the order of the cross product. Thus,
h × L = v × L − ae.
The LRL vector is usually defined as mh × L, while the eccentricity vector is (h × L)/a.

Answer to Exercise 8.1: We compute

(−∂r2 )[rl χl ] = −[l(l − 1)rl−2 χl + 2lrl−1 ∂r χl + rl ∂r2 χl ].

2
(−∂r )[rl χl ] = −2lrl−2 χl − 2rl−1 ∂r χl .
r
Thus, the equation for χl is
2(l + 2) 1
−∂r2 χl − ∂r χl − χl = k 2 χl .
r r
If we now write χl = e±ikr φl , the k 2 φl term cancels from the left and right of the equation
and we are left with
−r∂r2 φl − [2(l + 2) ± 2ikr]∂r φl − [1 ± 2ik(l + 2)]φl = 0.
Because the independent variable appears only linearly in this equation, if we write φl as a
Fourier or Laplace transform the equation becomes a first order ODE and is exactly soluble.
Thus, we have exact integral representations of the solution. This equation is called the
confluent hypergeometric equation.

Answer to Exercise 8.3: Write the energy in terms of the variable u = 1/r and use the
equations
ṙ = −u̇/u,
and
du lu
= u̇θ̇ = u̇ .
dθ m
Then we get
l2 2
E= [u + u2 ] − κu.
2m θ
Take the derivative w.r.t. θ and assume l 6= 0 to get
2mκ
uθθ + u = .
l2
 Answers to Selected Problems  501

The impact parameter b is defined to be the perpendicular distance between the trajectory of
the incoming particle and the origin, which is r sin(θ) in the limit of infinite r. The constant
on the right-hand side of the differential equation has dimensions of inverse length, so call it
L−1 . The incoming direction of the particle is arctan(b/L).
The solution of the equation is
2mκ
u = u0 cos(θ − θ0 ) − .
l2
Asymptotically u → 0, so θ → 0, π at finite b. This equation defines a hyperbola and u0
is the point of closest approach of this hyperbola, to the origin. The outgoing direction is
π/2 − arctan(b/L). Therefore, the angle of deflection is δ = 2 arctan(b/L). Thus,
b/L = tan(δ/2).
dσ
The scattering cross section dΩ dΩ, is defined as the number of particles scattered into solid
angle dΩ per unit time, divided by the number of incoming particles per unit time per unit
area. Since b and the deflection angle δ are functions of each other we have
dσ
dΩ = πd(b2 ) = 2πbdb.
dΩ
On the other hand, the infinitesimal solid angle is dΩ = 2π sin(δ)dδ, so that
dσ b db
= .
dΩ sin(δ) dδ
In order to write an expression in terms of asymptotic energy and impact parameter, we have
to rewrite L in terms of b and energy. The result of this algebra is
2κ
b= cot(δ/2),
E
so that
dσ κ2
= .
dΩ sin4 (δ/2)

Answer to Exercise 8.5: The spherical Bessel equation is

d2 jl djl
z2 2
+ 2z + [z 2 − l(l + 1)]jl = 0.
dz dz
P∞ n
Write jl = l=0 cn z . Then
∞
X
n(n + 1)cn z n + cn [z n+2 − l(l + 1)z n ] = 0.
 

n=0
502  Quantum Mechanics

Rewrite this as ∞ ∞
X X
[n(n + 1) − l(l + 1)]cn z n + = cn−2 z n .
 

n=0 n=2
n
Equating the coefficients of z on both sides we get

−l(l + 1)c0 = 0,

[2 − l(l + 1)]c1 = 0,
and
1
cn = cn−2 , n ≥ 2.
n(n + 1) − l(l + 1)
Note that this implies c0 = 0 unless l = 0 and c1 = 0 unless l = 1. It is also clear that we can
solve separately for cn for even and odd n. This is because the equation is invariant under
z → −z. The solution for cn becomes singular for n = l unless cl−2 = 0. That is consistent
with the recursion relation only if all the lower values of cn also vanish, so the series for jl
must begin with cl . Thus,
n−l
2
Y 1
cn (l) = cl (l).
k=1
(l + 2k)(l + 2k + 1) − l(l + 1)

Answer to Exercise 9.1: The general form of the Euler–Lagrange equations is (we follow
the convention of Lagrangian mechanics and use raised indices on coordinates):

∂t (∂q̇i L) − ∂qi L = 0.

For the charged particle Lagrangian

∂t (∂ẋi L) = mẋi + qAi ,

∂xi L = q ẋj ∂xi Aj .

The Euler–Lagrange equations are

mẍi = q ẋj (∂xi Aj − ∂xj Ai ).

We recognize the antisymmetric combination of derivatives of the vector potential as ijk Bk

(B is the magnetic field). So,
mẍ = q ẋ × B,
the Lorentz force equation.
 Answers to Selected Problems  503

Answer to Exercise 9.3: The guiding center solutions are

ψ(u, z) = ez̄u+ūz e−z̄z .

The scalar product is given by integration over the complex plane. The translation invariant
superposition is the integral of ψ(u, z) over the complex u plane and the integral does not
converge.

Answer to Exercise 10.1: The wave is not a real wave. It is a way of encoding the predic-
tions of quantum mechanics (QM) for the probability of the particle being at various vertical
positions, at horizontal positions to the right of the slits. Like all probability predictions it is
tested only by doing repeated experiments with identical initial conditions. When one places
a detector capable of discriminating which slit the electron went through, and asks what QM
predicts for such a situation, one is asking a different question, namely the conditional proba-
bility of finding the electron at vertical position y given that the detector, a macroscopic object,
has registered or not. These probabilities do not have detectable interference terms because
the state of the detector that has registered a macroscopic hit has doubly exponentially small
overlap with the state of the detector that has not registered a hit.

Answer to Exercise 11.1: Introduce a basis in the Hilbert space consisting of the even
and odd combinations of hydrogen ground states around the two well separated protons,
plus an infinite set of wave functions orthogonal to both of these. The expectation value of
the Hamiltonian in the “infinity minus two” dimensional subspace is bounded from below by
something of order ten Rydbergs. On the other hand, we will show in a moment that the split-
ting between the two states is of order e−d , so it makes sense to determine the ground state by
restricting attention to the matrix elements of the Hamiltonian in the two-dimensional sub-
space. The two localized hydrogen ground state wave functions are proportional to e−|r±dẑ| .
The matrix elements of the Hamiltonian between these two states are real and equal, because
the wave functions are real, and given by

1
Z
N d3 x [e−|r+dẑ| (−1 − )e−|r−dẑ| .
|r + dẑ|

N is the squared norm of the exponential function and the −1 comes from acting with the
kinetic energy plus potential due to the hydrogen atom on the positive z axis. Both terms
are negative, so the off diagonal part of the Hamiltonian in the two-dimensional subspace
is −σ1 , where  > 0. The ground state is the positive eigenvalue eigenstate of σ1 , which is
the symmetrized state. It is clear that  is exponentially small for large d, because each wave
function is of order e−d where the other is order 1.
504  Quantum Mechanics

Answer to Exercise 11.2: To prove the Feynman Hellman theorem note that any eigen-
state of the Hamiltonian H(λ) is normalized to 1 for all λ. Differentiating the normalization
condition with respect to λ we find that
dψ
hψ(λ)| i = 0. (G.36)
dλ
Now write
E(λ) = hψ(λ)|H(λ)|ψ(λ)i. (G.37)
Differentiating this with respect to λ we get two terms where the state is differentiated, and
its scalar product taken with H|ψ(λ)i = E|ψ(λ)i. The scalar product vanishes, so we are left
with
dE dH
= hψ(λ)| |ψ(λ)i. (G.38)
dλ dλ

Answer to Exercise 11.3: Let |ψ(d)i be the normalized ground state of

1 1
H(d) ≡ H = p2r + p2z − p 2 − p .
r + (z − d)2 r2 + (z + d)2

Then, since the norm is d independent, hψ(d)|∂d |ψ(d)i + ∂d hψ(d)|ψ(d)i = 0. The derivative
of the ground state energy E(d) is

∂d (hψ(d)|H(d)|ψ(d)i) = E(d)hψ(d)|∂d |ψ(d)i + ∂d hψ(d)|ψ(d)i + hψ(d)|∂d H(d)|ψ(d)i. (G.39)

Now
d−z d+z
∂d H(d) = + p +p 3. (G.40)
r2 + (z − d) 23 r2 + (z + d)2
The Hamiltonian H(d) is invariant under reflection in z so its eigenstates can be chosen either
even or odd under this reflection. We have seen that at large d the ground state is even. This
is also true at small d where the wave function approaches the spherically symmetric doubly
charged hydrogen ground state. The ground state is even for all d, by continuity. This means
that in evaluating the expectation value of ∂d H, the terms involving z vanish by symmetry.
Thus,
d d
∂d E(d) = hψ(d)| p 3 + p 2 3 |ψ(d)i, (G.41)
2
r + (z − d)2 r + (z + d)2
which is manifestly positive.

Answer to Exercise 11.5: Since the splitting between the even and odd combinations of
hydrogen wave functions centered around a single proton is exponentially small, it is sufficient
 Answers to Selected Problems  505

to consider the expectation value of the Hamiltonian in either one of these wave functions.
Let us choose the left proton as the origin. Then we have to calculate
Z
1
Z ∞ Z ∞ √ 1
u2 +z 2
− d3 r e−2r q =− du dz [ue−2 p .
2 2
r1 + r2 + (r3 − 2d)2 0 −∞ u2 + (z − 2d)2

For large d, we can approximate the square root by 2d as long as u and z are both  d
in absolute value. When they are this large, the integrand is exponentially small, so this
approximation is valid up to exponential corrections, throughout the region of integration.
There are of course power law corrections to the value of the integral from higher orders in
the expansion of the square root. When we substitute in the large d expression for the square
root, integral just becomes the normalization integral for the hydrogen wave function, times
1
− 2d . Thus, the leading behavior of the of the Born–Oppenheimer potential is a negative
constant plus an attractive Coulomb potential, which exactly cancels the Coulomb repulsion
of the protons.

Answer to Exercise 11.7: We have seen in the text and previous exercises that the electron
ground state energy for fixed d is a monotonically increasing function of d, varying between
−4 and −1 in Rydberg units, as d goes from zero to infinity. The positive Coulomb energy
between the protons is of order 1 when their separation is about a Bohr radius, but goes to
infinity at d = 0 and to zero at d = ∞. Thus, the full Born–Oppenheimer potential has a
minimum at d∗ ∼ 1.

Answer to Exercise 11.9: We use the first law of thermodynamics at fixed entropy dE =
−P dV . The entropy in a system of N q noninteracting electrons is N qln 2, so we want to
vary the volume at fixed N q. Thus,
dE 2E
=− dV = −P dV
V 3V
and
~2 (3π 2 )2/3
P = ρ5/3 .
5m

Answer to Exercise 12.1:

ρ∗ji =
X
cjI cIi = ρij ,
I
so ρ is Hermitian. Compute the expectation value of ρ in any state as

vi∗ ρij vj = vi∗ c∗iI cIj vj ≥ 0,

X X

ij ijI
506  Quantum Mechanics

since it is a sum of absolute squares of complex numbers. Applied to a vector vi , which is an

eigenvector of ρ it implies that all the eigenvalues are positive. The trace of ρ is iI c∗iI cIi = 1,
P

because it is the norm of the full vector in the tensor product Hilbert space. The sum of the
eigenvalues is thus 1, and ρ has the properties of a density matrix.

Answer to Exercise 12.3: We are looking for the maximum of the entropy

−Tr (ρln ρ)

subject to the constraint

TrρN = n0 ,
where N is the number operator for zero momentum bosons. The expectation value is
X
kpi |hk|pi i|2 ,
ik

where |ki are the number eigenstates, and |pi i the eigenstates of the density matrix, while
the entropy is − i pi ln pi , and is independent of the scalar products |hk|pi i|2 ≡ Oki . Let us
P

assume the two operators commute, so that the |ki and |pi i bases are the same. Then the
constraint is ∞ X
p k k = n0 .
k=0
We minimize the energy subject to this constraint by using a Lagrange multiplier, minimizing
X ∞
X
− pk lnpk + L( pk k = n0 ),
k k=0

w.r.t. both pk and L, obtaining

−lnpk − 1 + Lk = 0,

along with the constraint equation. Thus,

pk ∝ eLk .

We determine the proportionality constant by requiring that the sum of all probabilities is
one
eLk
pk = ,
1 − eL
and the value of L by P Lk
ke L eL
n0 = = ∂ L ln(1 − e ) = .
1 − eL 1 − eL
 Answers to Selected Problems  507

This implies
n0
eL = .
1 + n0

Answer to Exercise 12.4: In classical statistical mechanics, the partition function is the
integral over all momenta and coordinates of e−βH(p,x) , where H is the classical Hamiltonian.
A magnetic field is added by shifting pi → pi − ei A(xi ), where ei is the electric charge of
the i-th particle and A is the vector potential whose curl is the magnetic field. Since we
can eliminate the vector potential by a shift of all the momentum integration variables, the
partition function is independent of the magnetic field. The expectation value of the energy
is the derivative with respect to temperature of the energy expectation value and so does not
depend on the magnetic field. Thus, there’s neither diamagnetism or paramagnetism.

Answer to Exercise 12.5: The full Hamiltonian is

H = h(pi − e/cA(xi ), xi ) + HB .

In classical mechanics we have to integrate e−βH over all pi , xi and θi . As in Exercise 12.4,
we can shift the pi integration variable to pi − A(xi ), and the Jacobian of this change of
variables is 1. So the partition function factorizes:
Z
Z = Z0 dθi e−βHB ,

and the first factor has no B dependence. It affects neither the magnetization nor the sus-
ceptibility. The θi integrals are all decoupled and identical so
Z
ln Z = N ln [2π sin(θ)dθ e−α cos(θ) ].

Note that the integration is over solid angles, which accounts for the 2π sin(θ). Since
sin(θ)dθ = d(cos(θ)) this is
Z 1
ln Z = N ln [4π dce−αc ] = N ln [α−1 sinh(α)],
−1

up to terms independent of B. The magnetization and susceptibility formulae are simple

derivatives of this. At small β, α is also small and we can approximate the partition func-
tion by
α2
ln Z ≈ N ln [1 + ].
3!
The magnetization is then N µ βB/3 and the susceptibility is N µ2 β/3. The signs indicate
2

that we have paramagnetism and one can verify that this is true for all β.
508  Quantum Mechanics
q
Answer to Exercise 13.1: To calculate the expectation values we write X = 2mω ~
(a+a† ),
and realize that in calculating expectation values of powers of X in energy eigenstates, we
only have to take into account terms with equal numbers of powers of a and a† . Thus,
effectively
~ 2
X 4 → (a2 a† 2 + a† 2 a2 + a† a2 a† + aa† 2 a + aa† aa† + a† aa† a)( ) .
2mω
Using the commutation relations we can write all of these expressions in terms of the number
operator N = a† a = aa† + 1.
~ 2
X 4 → (a(N − 1)a† + a† N a + N 2 + (N − 1)2 + 2N (N − 1))( ) .
2mω
Using the commutation relations again, the first two terms give two more factors of N (N −1),
so altogether
~ 2
X 4 → (6N 2 − 6N + 1)( ) .
2mω
The perturbed En is thus,
~ 2
En = ~ω(n + 1/2) + b(6n2 − 6n + 1)( ) .
2mω

Answer to Exercise 13.3: The coefficients of the polynomial P (a) are themselves polyno-
mial functions of λ, so even though the operator A0 + λA1 is not Hermitian for complex λ
we can define the roots of P (a) as functions of λ:

P (λ, a(λ)) = 0.

Taking the derivative of this equation w.r.t. λ we get

da ∂λ P
=− .
dλ ∂a P
Both partial derivatives of the polynomial are analytic functions of λ. If we are at a value of
λ for which the root ak (λ) is nondegenerate, then ∂a P is nonzero, so the derivative of ak (λ)
is well defined in all directions in the complex plane, so that ak (λ) is analytic. The equation
a1 (λ) = a2 (λ) is a single complex equation for a complex unknown, so it will generally
have a solution. On the other hand, more than one degenerate level at a fixed value of λ is
nongeneric and might occur only in the presence of symmetries. So, near a level crossing,
the solution of the polynomial equation reduces to a quadratic equation and the roots have
square root branch points at the value of λ where they coincide. For an N × N Hermitian
 Answers to Selected Problems  509

matrix we therefore expect each root to have N − 1 branch points, where it crosses some
other root. Remarkably, this means that if we know the analytic continuation of any single
root to complex λ, then we can find all the other roots as different branches of the same
locally analytic function.
δk2 2 δk2
Answer to Exercise 13.5: The perturbation is 2 X = α2 (a+a† )2 , where α2 = ~
2 2(mω) =
δk2
4k2 ~ω.As in Exercise 13.1, we only have to consider terms with equal numbers of powers of
a and a† . Thus, in first order perturbation theory we can write
hn|V |ni = α2 hn|aa† + a† a|ni = (2n + 1)α2 .
In second order, we have to evaluate
P
α4 hn|(a + a† )2 (a + a† )2 |ni.
~ω(N − n)
Here, the projection operator P = 1 − |nihn|. In this case, because of the projection only the
terms
P
α4 hn|(a2 + a† 2 ) (a2 + a† 2 )|ni,
~ω(N − n)
survive, and in addition we get zero unless we pair the operator a2 on the left with its
Hermitian conjugate on the right, or vice versa. We see that in either case only one of the
states |n±2i is produced, so the Hamiltonian operator in the denominator is just the number
±2~ω. We then have to evaluate the diagonal matrix elements
1 3n
[hn|a2 a† 2 |ni − langlen|a† 2 a2 |ni] = .
2~ω 2~ω
So the perturbed levels are
3n
En = ~ω(n + 1/2) + α2 (2n + 1) − α4 .
2~ω
The exact formula is
δk 2 1/4
En = ~ω(1 + ) (n + 1/2).
k2
Plugging in the value of α we see that the two expressions coincide, to second order.

df
Answer to Exercise 15.1: If we change parameter to s = f (t) then ds = dt dt and dxµ /ds =
dt
dxµ /dt ds = dxµ /dt df 1/dt . The two factors of the derivative of f cancel.

Answer to Exercise 15.3:

(n)
Bi = (∇ × A(n) )i = ijk ∂j (hn|∂k |ni) = ijk ∂j (hn)|∂k |ni.
510  Quantum Mechanics

Now we insert a complete set of states 1 = 2m=1 |mihm|, and use the result of Exercise
P

15.2. We also note the fact that in the formula of 15.2 exchanging m and n is equivalent to
complex conjugation:

(n)
X hn|∂j H|mihm|∂k H|ni
Bi = i~ijk .
m6=n
(Em − En )2

The m = n term doesn’t contribute (avoiding a singularity) because of the antisymmetry

under interchange of j and k. There is really only one term in the sum and the energy
denominator is 2x. Thus, 4x1 2 factors out of the formula and we can restore the vanishing
m = n term to the sum to write this as
(n)
Bi = i~ijk hn|σj σk |ni.

We can evaluate this using the commutation relations of the Pauli matrices. Now the eigen-
states of x · σ have expectation values of σj which are just ±x̂j , so we get

(n) x̂
Bi = ∓~ ,
2x2
which satisfies
~
∇ · B = ∓ δ 3 (x).
2
If we think of this as a “magnetic field” it is the field of a magnetic monopole sitting at the
origin. This is called the Berry monopole. It is a signal of the degeneracy of the two states
at the origin of parameter space.

Answer to Exercise 15.5: The parameter space here is the two-dimensional complex plane
z = reiθ and the relevant component of the Berry potential is
Z νN νN νN νN P
∗ zi∗ zi
(zi∗ ∗
(zi )FLaughlin ]e−
Y Y Y Y
Aθ = i~ FLaughlin −z ) (zi ∗)∂θ [ (zi − z) .
i=1 i=1 i=1 i=1

Since FLaughlin does not depend on θ we have

Z νN νN νN νN
∗
(zi∗ − z ∗ )
Y Y Y Y
Aθ = i~ FLaughlin FLaughlin (zi ∗)∂θ [ (zi − z) (zi )].
i=1 i=1 i=1 i=1

The derivative gives a sum of νN terms

νN
z
Z X
Aθ = ~ |ψquasiholes |2 .
i=1
zi − z
 Answers to Selected Problems  511

The integral around θ picks up the pole and we get

Z
Aθ dθ = 2πν,

which is an anyonic Berry phase.

Answer to Exercise 16.1: The Lippman–Schwinger equation is

1 eik|x−y|
Z
ψk (x) = eik·x − d3 y U (y)ψk (y),
4π |x − y|
2m
where U ≡ ~2 V . Define
eik|x−y|
Z
G0 (x − y) ≡ d3 y .
4π|x − y|
We can solve this by iteration, with the n-th term being
Z
d3 y1 . . . d3 yn G0 (x − y1 )U (y1 )G0 (y1 − y2 )U (y2 ) . . . G0 (yn−1 − yn )U (yn )eik·yn .

The diagram corresponding to this term is shown in Figure 16.2 at the end of Chapter 16.

Answer to Exercise 16.3: The solutions to the Schrödinger equation inside and outside
the well, for angular momentum l are
ψ+ = A+ jl (kr) + B+ hl (kr),
ψ+ = A− jl (k0 r) + B− hl (k0 r),
2m
where k02 = k 2 − ~ 2 V0 .
The second derivative of the wave function is a delta function, so
both the wave function and its derivative are continuous at r = r0 . In the interior, we must
choose B− = 0 since hl (kr0 ) is singular at the origin. Thus,
A+ jl (kr0 ) + B+ hl (kr0 ) = A− jl (k0 r0 ).
k[A+ jl0 (kr0 ) + B+ h0l (kr0 )] = k0 A− jl0 (k0 r0 ).
We can solve these equations for
B+ k0 jl (kr0 )jl0 (k0 r0 ) − kjl0 (kr0 )jl (k0 r0 )
≡ e2iδl (k) = 0 .
A+ khl (kr0 )jl (k0 r0 ) − k0 hl (kr0 )jl0 (k0 r0 )
Answer to Exercise 16.5: In the low energy limit, the formula of the previous exercise
reduces to
tan k0 r0
tan δ0 = −kr0 ( − 1),
k0 r0
512  Quantum Mechanics
q
and k0 ≈ − ~2m 2 V . Remember that V0 is negative for a spherical well. The scattering cross
0
section, which is also the total cross section at these low energies is
4π tan k0 r0
σ0 ≈ 2
sin2 (δ0 (k)) ≈ 4πr02 ( − 1).
k k0 r0
Answer to Exercise 16.7: Turning the formula for the cross section in terms of δl into
the Breit-Wigner form is a matter of simple algebra. The formula obviously has a maximum
at E = ER . When E = ER ± Γ/2, the cross section has fallen to half of its value at the
peak. This is the reason for the name width, short for “full width at half maximum”. If we
take a Fourier transform of this function, we get a function that behaves like e−Γt . Roughly
speaking, this tells us that the physics of a Breit–Wigner peak is that the scattering particle
gets trapped in a meta-stable bound state for a time of order 1/Γ. A more sophisticated
analysis confirms this expectation. The inverse of Γ is therefore called the lifetime of the
meta-stable resonance.

Answer to Exercise 17.1: Differentiating w.r.t. q j (0) we get

t d ∂q k (s)
Z
∂q i (0) s(t) = ds {[q̇ i (s) − ∂qi (s) S]∂qj (0) (∂qi (s) S) − [∂qk (s) V + ∂qk (s) S] j },
0 ds ∂q (0)
where we have used pi (s) = ∂qi (s) S, and integrated by parts in the last term. The vanishing
of the two independent terms in square brackets is precisely the statement of the classical
equations of motion.

Answer to Exercise 17.3: The integrand in the definition of the Airy function is the
exponential of
~2 p3
i(px − ),
6ma
and the stationary phase point is q
p = 2max/~.
Thus, the integral is approximately
2i
√ 2ma
x3/2
Ai ∼ e 3 ~ .
It is important to realize that V − E is going from negative to positive as x increases so that
a is negative and this function falls like the exponential of x3/2 as x goes to positive infinity.
1/2
The logarithmic
q derivative of the Airy function thus behaves like cx for large positive x,
2m|a|
where c = − ~ . The second solution is
Bi = bAi,
 Answers to Selected Problems  513
p 2m|a|
4
x3/2
with b0 = e 3 ~ , so that Bi is exponentially growing at infinity and must be discarded.

Answer to Exercise 17.5: The JWKB solutions are

R √ 2m
(E − V )−1/4 e± dx ~ (E−V ) .

The integral is taken over the real x axis, starting from a point to the left of the turning point
V = E. If the potential V is an analytic function,1 these solutions are analytic functions of
x away from the turning point. Thus, we can connect the behaviors to the right and left of
the turning point, by following a contour that avoids the turning point. Not too far from the
right of the turning point, the exponentially falling piece of the wave function dominates, so
we have to analytically continue this solution. There are two possible contours to choose to
avoid the turning point, either in the upper or lower half x plane. In either case, we pick up
a phase factor from the prefactor (V − E)−1/4 . The phase is eiπ/4 for the upper contour and
e−iπ/4 for the lower contour.
The Schrödinger equation is real in the classically forbidden region, so we can take the
solution to be real. The analytic continuation that preserves this reality property is to take
the average of the results of the upper and lower contours
1
Z x q
−1/4
ψ = C[2m(E − V )] cos( dy 2m(E − V ) + π/4).
~ a

The coefficient C is the coefficient of the exponentially falling solution to the right of the
turning point. Note that the integral is taken from a to x < a because that is the analytic
form of the solution in the classically forbidden region. We can rewrite it in terms of the
conventional direction of integration in the classically allowed region
1
Z a q
−1/4
ψ = C[2m(E − V )] sin( dy 2m(E − V ) + π/4).
~ x

Answer to Exercise 17.7: The Bohr–Sommerfeld condition is

E−V0 r
2m
Z
k
dr (E − V0 − kr) = nπ.
0 ~
E−V0
Defining r = k y, this is
Z 1 q k
dy (1 − y) = nπ √ (E − V0 )−3/2 .
0 2m
1
Actually, for the purposes of this argument, it is sufficient for V to be continuous on a domain in the
complex plane including the contour to be described below. Any such continuous function can be uniformly
approximated by polynomials on that domain.
514  Quantum Mechanics

We solve for the binding energies as

2nπk
En = V0 + ( √ )2/3 .
3 2m
The masses are
m(n+1)s = 2m + En .
Thus,
m1s = 2m + V0 ,
2πk
m2s = m1s + ( √ )2/3 ,
3 2m
m3s = m1s + 22/3 (m2s − m1s ).
For the charmed bound states this gives

m3s = 3.1 + 1.58(.6) = 4.05GeV.

The experimental value is 4.05GeV. For bottom the formula gives

m3s = 9.46 + 1.58(.56) = 10.34GeV.

The experimental value is 10.36.

Answer to Exercise 17.8: The Bohr–Sommerfeld conditions are

Z R q
dx 2m(E − V (x)) = nπ~.
L

We can make the integral as large as we want by taking E large, so that the turning points
L, R are at larger values of |x|. In this limit the action S0 is large, equal to npi~ and S0 /~
is the JWKB approximation to the log of the wave function. The exact equation for the
logarithm of the wave function S/~ is
1
E = (∂x S)2 + i~∂x2 S.
2
The condition for validity of the JWKB approximation is
1
|~∂x2 S|  (∂x S)2 .
2
Applying this criterion to the approximate solution S = S0 , we get
q
~∂x V  2m(E − V )3 /2.
 Answers to Selected Problems  515

Since En is very large, and V is independent of n, this inequality will be satisfied better and
better at large n, except for a region around the turning point, which shrinks in size, as n
goes to infinity.

Answer to Exercise 17.9: A bound state would correspond to a normalizable solution of

the Schrödinger equation. However, it is clear that the expectation value of the Hamiltonian
in any normalizable state is positive, so the bound state would have to be positive energy.
At very large r we can solve the Schrödinger equation with plane wave solutions e±ikr , with
~2 k 2 = 2mE. Thus, there are no bound states, since the plane waves are not normalizable.

Answer to Exercise 17.11: The equation for ~/i times the logarithm of the wave function
is
(∂x S)2 − ig 2 ~∂x2 S = 2m(E − V ) ≡ p2 (x).
Write S = ∞ 2 n 1−n
P
n=0 Sn (ig ~) . Sn has dimensions of [action] , but contains no powers of ~.
∂x S0 is the classical momentum p(x) at point x. The equation for Sn is
n
X
∂x2 Sn−1 = ∂x Sk ∂x Sn+1−k .
k=0

This gives
∂x2 S0 = 2∂x S0 ∂x S1
and
∂x2 S1 = 2∂x S0 ∂x S2 + (∂x S1 )2 .
The solution of the first equation is
1
S1 = ln p(x),
2
so the second equation becomes
1
∂x S2 = −∂x2 [ ].
4p
So
∂x p
S2 = .
4p2
The second order correction to the log of the wave function is real.

Answer to Exercise 18.1: We have

p2
[ + gx2q ]ψ = Eψ.
2
516  Quantum Mechanics

Define x = ay, with

ga2q+2 = 1.
Then the equation becomes
p2y
[ + y 2q ]ψ = a2 Eψ.
2
Since the eigenvalues of the new operator are independent of g, we must have E = a−2 E =
1
g q+1 E, where E is independent of g.

Answer to Exercise 18.3: To use the variational principle to prove the existence of positive
energy bound states, we first show that there is a minimum energy for propagating out to
infinity. This is easy. Go far out in the region x → ∞, a > |y|. In this region the lowest energy
we can have is a plane wave eikx multiplied by the lowest state sin(πy/a) in the infinite square
2
well. This has energy at least πa2 . Thus, if we can show there is a normalizable wave function
with expectation value of the Hamiltonian below this, then we have proven there is a bound
state.
Here’s a trial wave function:

ψ = (a2 − xy)e−b θ(a − x)θ(a + x)θ(a − y)θ(a + y)

+ e−b|y| (a2 − ax)θ(a − x)θ(a + x)θ(y 2 − a2 ) + e−b|x| (a2 − ay)θ(a − y)θ(a + y)θ(x2 − a2 ).

It is not normalized but is symmetric under interchange of x and y, as well as under reflection
of each coordinate. The expression for the expectation value of the Hamiltonian in this state
is a bit complicated, because the action of the Laplacian on this function has delta function
singularities, which must be taken into account. You can find the gory details of the full
solution in the solution manual to the Second Edition of Griffiths (Exercise 7.20).

Answer to Exercise 18.5: The harmonic atom Hamiltonian has the form
X P2 1
i
H= + (Ω2 )ij Xi Xj .
2 2
The matrix Ω2 is given by
(Ω2 )ij = Ω2 δij − ω 2 1ij ,
where the matrix 1ij has a one in every matrix element. its eigenvalues are −N and 0, with
the latter having multiplicity N −1. We thus get eigenfrequencies of oscillation Ω2 −N ω 2 and
Ω2 . The former controls the motion of the center of mass of the electrons, while the latter
describes relative motions. We obviously want to keep Ω2 − N ω 2 positive to stabilize the
system and make sure the center of mass is bound to the “nucleus”. We will however insist
 Answers to Selected Problems  517

2
that 0 < Ω ω 2 − N  N . For distinguishable particles, the eigenstates of the Hamiltonian
are arbitrary excited states of all N oscillators, but we must impose Fermi statistics. Also,
since Ω2  Ω2 − N ω 2 , the ground state will be a state in which the relative oscillations are
excited as little as possible. The relative coordinates xi − xj are redundant. They can all
be expressed in terms of the coordinates xI − xj for some particular value of I. However, if
we first form the Slater determinant det [ψi (xI − xj )] using the first N − 1 eigenstates of
the oscillator with frequency Ω, and then explicitly antisymmetrize under permutations that
exchange I with one of the other labels, then we get a totally antisymmetric function, which
P −1
has energy N k=0 ~Ω(k + 1/2). We then multiply this by the ground state of the center of
mass oscillator, which is invariant under permutations. The total energy is

~
E= (Ω − N ω + N 2 Ω).
2

Answer to Exercise 18.7: We have to compute the expectation value of i<j σ a (i)σ a (j)Kij ,
P

in the ground state of the external field Hamiltonia, which is the state where ha (i)σ a (i) =
−|h(i)|. The expectation value of H in this state is just
X
< H >= Kij < σ a (i) >< σ a (j) > .
i<j

This is because the state is a tensor product over sites and the Hamiltonian has no operator
products of operators on the same site. The expectation values of spins are < σ a (i) >=
−ĥa (i) ≡ sa (i). Note that it is independent of the local field strength. The variational problem
then becomes one for classical spins, unit vectors. It is easy to solve if K is negative definite,
the ground state wants all spins aligned. However, if some of the bonds are positive, they
want to antialign, but the Hamiltonian can want to antialign a spin with two others that are
preferentially antialigned as well. Think of three spins with links on an equilateral triangle
with all positive bond strengths. The technical term for this sort of situation is frustration
and spin problems with frustrated bonds have interesting and complex behavior.

Answer to Exercise 19.1: We insert a complete set of eigenstates of the wave number
operator and write this as
Z
~k2
d3 k hy|kihk|xie−τ 2m e−τ /~V (x) .

The Gaussian k integral now gives

τ m(x−y)2
N (τ )e− ~ [ 2τ 2
−V (x)]
.
518  Quantum Mechanics

In the limit τ → 0 the exponent approaches dτ L. The normalization factor diverges but we
have dealt with this in the text.

Answer to Exercise 19.3: The functional equation is

d2 δZ δ
2
+ V 0[ ]Z = j(t)Z.
dt iδj(t) iδj(t)
Expand
∞ n Z
X i
Z[j] = dt1 . . . dtn [Gn (t1 . . . tn )j(t1 ) . . . j(tn )],
n=0
n!
and equate the coefficients of j(t)j(t1 ) . . . j(tn−1 ) on both sides. The first term on the left-
hand side gives a coefficient
d2
Gn (t, t1 . . . tn−1 ).
dt2
If V 0 (X) = ak X k , then the second term gives
P

X
ak Gn+k−1 (t, t, . . . , t, t1 . . . tn−1 ),

where the number of equal arguments of Gn+k−1 is k. The right-hand side gives
X
i δ(t − tk )Gn−1 (t1 , . . . tn−1 ).

So the SD equations are

d2 X X
2
Gn (t, t1 . . . tn−1 ) + ak Gn+k−1 (t, t, . . . , t, t1 . . . tn−1 ) = i δ(t − tk )Gn−1 (t1 , . . . tn−1 ).
dt

Answer to Exercise 19.7: The real time Green function equation is

(d2t + ω 2 )G(t, s) = δ(t − s),

which has the Fourier transform solution

dν 1
Z
G(t, s) = eiν(t−s) .
2π ω − ν 2
2

The integrand has poles at ω = ±ν and we must choose the contour of integration to avoid
them. There are four independent choices. One can take the contour in the lower half plane,
or the upper half plane, or take it above the negative pole and below the positive one, or
vice versa. The differences between these contours are contour integrals that encircle one or
 Answers to Selected Problems  519

more of the poles, which, by the residue theorem are proportional to e±iωt , both solutions of
the homogeneous equation.

Answer to Exercise 19.9: We will set ~ = 1 for this We will give the path integral for
the operator ordering H 21 (Pi Pj Mij−1 (Q) + Mij−1 (Q)Pi Pj ). Other orderings can be treated in
a similar manner. The strategy is to generalize the Trotter product formula, writing
2A
hq 0 |e−iHt |qi = eitA1 eit 2
...,
and neglecting An n ≥ 2 for small t. The analog of A1 can be computed by expanding the
operator expression to first order in t. We have
1
Z
0
hq 0 |H|qi = dn peip(q −q) ( (Mij (q 0 ) + Mij (q))pi pj .
2
Introducing complete sets of intermediate q and p states at intermediate times tk = tk/N we
have Z Rt 1 −1
0 −iHt
hq |e |qi = [dn q(t)dn p(t)]ei 0 [pq̇− 2 pi pj Mij (q)] .

This is called the Hamiltonian form of the path integral. Going back over the original deriva-
tion, one can see that we actually used this when Mij was just a constant. We can do the
integral over pi (t) exactly, because it is a Gaussian. This gives
Z
i
0 −iHt
hq |e |qi = [dn q(t)detM(q(t))]e 2 qi qj Mij (q) .

So the “position dependent mass” changes the local measure on the path integral.

Answer to Exercise 19.11: Since different ai involve different γ matrices, they all anti-
commute with each other and with each other’s adjoint. Then we have
1
[a1 , a1 ]+ = ([γ1 , γ1 ]+ − [γ2 , γ2 ]+ ) = 0,
2
since γ1 and γ2 anticommute. Similarly
1
[a1 , a†1 ]+ = ([γ1 , γ1 ]+ + [γ2 , γ2 ]+ ) = 1.
2

Answer to Exercise 20.1: [σa (i), σb (j)] = 0 because the operators act on different factors
(i) (i)
1+σ3 (j) 1−σ3
in a tensor product Hilbert space. By definition, Cij = 2 σ1 + 2 , so
(i) (i) (j) (j) (i) (i)
1 + σ3 (j) 1 − σ3 1 + σ3 (i) 1 − σ3 1 + σ3 (j) 1 − σ3
Cij Cji Cij = [ σ1 + ][ σ1 + ][ σ1 + ].
2 2 2 2 2 2
520  Quantum Mechanics

This can be simplified by using

σ12 (i) = 1, σ1 (i)P± (i) = P∓ (i)σ1 (i) P+ (i)P− (i) = 0, P±2 (i) = P± ,
(i)
1±σ3
where P± = 2 . The result is

Cij Cji Cij = P+ (i)P+ (j) + P− (i)P− (j) + σ1 (i)σ1 (j)(P+ (i)P− (j) + P− (i)P+ (j)).

In words, the operation on the RHS does nothing if both Q-bits have the same value, and
changes the value of both Q-bits if they are different. That is, it is precisely the swap operator
Sij .

Answer to Exercise 20.3: The pure state has the form |si = caB |a, Bi, where a and B
P

label bases in the first and the second parts of the system. If the dimensions are unequal, we
choose B to be the larger system. In terms of the rectangular matrix c, the reduced density
matrices are
ρA = cc† , ρB = c† c.
Then
† a2 B(n−1) † a1
X
Tr[ρnA ] = cB1
a1 (c )B1 . . . ca(n−1) (c )B(n−1) .

This is obviously the same as Tr[ρnB ]. It is a simple generalization of the cyclicity of the trace
formula, to rectangular matrices. One can compute the von Neumann entropy as the limit
of Trρn /n − 1. Traces of powers of the density matrix are called Renyi entropies.

(i) (i)
Answer to Exercise 20.5: The definition of the Hadamard operator is H (i) = √12 (σ1 +σ3 ).
This operator is unitary and transforms from the basis where σ3 is diagonal to the basis where
σ1 is diagonal. The easiest way to see this is to note that the columns of the Hadamard
operators in the σ3 basis are the orthonormal eigenvectors of σ1 . The definition of Cij is

Cij = P+ (i) ⊗ 1(j) + P− (i) ⊗ σ1 (j).

Here P± (i) are the projectors on eigenstates of σ3 (i). Writing the unit matrix and the σ1
operators on the j Q-bit as linear combinations of the projectors on σ1 eigenstates.
(1) (1)
Cij = 1(i) ⊗ P+ (j) + σ3 (i) ⊗ P− (j),

where the superscripted P± are projectors on eigenstates of σ1 (j). This looks like a C-not
operation, except that the roles of σ1 and σ3 are interchanged in both Q-bits. Thus, if we
conjugate by the product of the Hadamard operators on both Q-bits, we will in fact get Cji ,
as claimed.
 Answers to Selected Problems  521

Answer to Exercise 20.6: Clearly the number of independent bit strings is the same as
the number of basis vectors in a vector space of dimension 22N . The identification of a given
pair of bit strings with eigenstates of the complete set of commuting operators σ2 (i) shows
us that the vectors form an orthonormal basis. The mapping U (f ) clearly maps every basis
vector into another unit vector. Thus, all we have to do is prove that

hx, y + f (x)|z, w + f (z)i = 0

unless x = z, y = w. For the first argument, this is obvious from the definition of the scalar
product. Then, for the second argument we must have y + f (x) = w + f (x), which implies
y − w = 0.

Answer to Exercise 20.7: We have to prove

N
1
[ √ (σ3 (i) + σ1 (i))]|x, yi = 2−N/2
Y X
(−1)xż |z, yi.
i=1 2 z

i
To do this note that for 1 ≤ i ≤ N , the action of σ3 (i) + σ1 (i) on |x, yi gives (−1)x |x, yi +
|x + 1i , yi. 1i is the N bit string with a 1 in the i-th place and zeroes elsewhere. When we
act with the tensor product operator on all i, we get an N -fold tensor product of sums over
i
two states with a sign (−1)x . Expanding this out by the distributive law for tensor products
of sums of states, we just get a sum over all N bit strings z with a sign (−1)x·z . This is the
key point of the algorithm, we can find a state which is a specific linear combination of all
2N states |x, yi with fixed y by doing only N quantum operations. The rest of this exercise
is essentially done in the statement of the problem.

Answer to Exercise 20.8: For every z, z + a also appears in the sum

X
(−1)y·z |y, f (z)i.
z

The value of f on these two vectors is the same, by periodicity. If y · a = 1, then the two
terms appear with opposite sign, and cancel. So we only get contributions from y orthogonal
to the period vector.

Answer to Exercise 20.9: The product of σ3 (i) + σ1 (i) can obviously be implemented
in N steps. The operation U (f ) on |z, 0i just requires one consultation of the lookup table
defining the function. Thus, the isolation of one string y satisfying y · a = 0 only takes N
steps. Since the space of bit strings in N -dimensional, kN random choices will find a whole
basis in the space for modest values of k, with very high probability. Indeed, for the first
522  Quantum Mechanics

N/2 − 1 choices, the probability of not finding linearly independent strings goes to zero as
N goes to infinity. Afterwards we have to get a bit more lucky but even after we have found
a subspace of dimension N − 1, the probability that 167N further random choices will all lie
within that subspace is very small and goes to zero with N . Thus, in a time linear in N we
will know all bit strings orthogonal to the period.
Bibliography

[1] L. Mlodinow, The Drunkard’s Walk: How Randomness Rules Our Lives, Vintage Books,
Random House, New York, 2009, ISBN: 978-0-3-7-37517-2.
[2] W. Feller, An Introduction to Probability Theory and Its Applications, John Wiley and
Sons, 1957, 1968, Wiley Series in Probability and Mathematical Statistics.
[3] T. Bayes, R. Price, An Essay towards solving a Problem in the Doctrine of Chance. By
the late Rev. Mr. Bayes, communicated by Mr. Price, in a letter to John Canton, A. M.
F. R. S, Philosophical Transactions of the Royal Society of London, 1763, 53(0):370–
418. doi:10.1098/rstl.1763.0053.
[4] T. Bass, The Eudaemonic Pie, Houghton Mifflin Harcourt, 1985.
[5] J. von Neumann, The Mathematical Foundations of Quantum Mechanics, Princeton
University Press, 1955.
[6] N. A. Wheeler (Ed.), Mathematical Foundations of Quantum Mechanics.
https://press.princeton.edu/titles/11352.html
[7] S. J. Gould, The Streak of Streaks, New York Review of Books, August 18, 1988; The
Median Isn’t the Message, Discover Magazine, 1985.
[8] E. Noether, Invariante Variationsprobleme, Nachr. d. Konig. Gesellsch. d. Wiss. zu Got-
tingen, Math-phys. Klasse, 1918, 235–257; English translation M. A. Travel, Transport
Theory and Statistical Physics, 1971, 1(3):183–220.
[9] P. Michel, Mittag-Leffler, Contribution l’tude de la representation d’une fonction arbi-
traire par les intgrales dfinies, Rendiconti del Circolo Matematico di Palermo, 1910,
30(1):289–335. doi:10.1007/BF03014877.
[10] W. E. Boyce, R. C. DiPrima, Elementary Differential Equations and Boundary Value
Problems (8th ed.). New Jersey: John Wiley & Sons, 2005, Inc. ISBN 0-471-43338-1.
Joseph Fourier, translated by Alexander Freeman (published 1822, translated 1878, re-
released 2003). The Analytical Theory of Heat., Dover Publications. ISBN 0-486-49531-
0. 2003 unabridged republication of the 1878 English translation by Alexander Freeman
of Fourier’s work Thorie Analytique de la Chaleur, originally published in 1822.
523
524  Quantum Mechanics

[11] J. -P. Antoine, Quantum Mechanics Beyond Hilbert Space, appearing in Irreversibility
and Causality, Semigroups and Rigged Hilbert Spaces, Arno Bohm, Heinz-Dietrich
Doebner, Piotr Kielanowski, eds., Springer-Verlag, 1996, ISBN 3-540-64305-2.
[12] M. Nauenberg, Quantum wave packets on Kepler elliptic orbits, Physical Review A
1989, 40(2):1133–1136.
[13] F. Cooper, A. Khare and U. Sukhatme, Supersymmetry and quantum mechanics,
Physics Reports 1995, 251:267–385, and references therein.
[14] D.J. Griffiths, Introduction to Quantum Mechanics (2nd ed.). Cambridge University
Press, 2017.
[15] R. Rajaraman, Solitons and Instantons, North Holland Personal Library, Elsevier,
Amsterdam, 1982, ISBN-13: 978-0444870476, ISBN-10: 0444870474.
[16] W. Pauli, The connection between spin and statistics, Physical Review, 1940,
58(8):716–722. Bibcode:1940PhRv...58..716P. doi:10.1103/PhysRev.58.716; G. Lüders,
B. Zumino, Bruno, Connection between spin and statistics, Physical Review 1958, 110:
1450. doi: 10.1103/PhysRev.110.1450.
[17] T. Banks, Modern Quantum Field Theory: A Concise Introduction, Cambridge
University press, 2008; A. Zee, Quantum Field Theory in a Nutshell, 2nd Edition,
Princeton University Press, 2003.
[18] A. C. Newell, Solitons in Mathematics and Physics, CBMS-NSF Regional Con-
ference Series in Applied Mathematics, 1985, ISBN: 978-0-89871-196-7 eISBN:
978-1-61197-022-7.
[19] See the article on Legendre Polynomials by I. Stegun (Chapter 8) in M. Abramowitz
and I. Stegun, Handbook of Mathematical Functions, National Bureau of Standards
Applied Mathematics Series, 55, 1964.
[20] D. R. Hartree, The Calculation of Atomic Structures, Wiley and Sons, New York, 1957.
[21] P. Hohenberg, W. Kohn, Inhomogeneous electron gas, Physical Review, 1964, 136
(3B):B864–B871. Bibcode:1964PhRv..136..864H. doi:10.1103/PhysRev.136.B864; W.
Kohn, L. J. Sham, L. Jeu, Self-consistent equations including exchange and correlation
effects, Physical Review, 1965, 140(4A):A1133–A1138. Bibcode:1965PhRv..140.1133K.
doi:10.1103/PhysRev.140.A1133.
[22] B. H. Brandsen, C. J. Jochain, Physics of Atoms and Molecules, ISBN-13: 978-
0582356924, ISBN-10: 058235692X ; P. W. Atkins, R. S. Friedman, Molecular Quantum
Mechanics, 5th Edition, Oxford University Press, London; M. Born, Atomic Physics,
8th Edition, Dover Books on Physics, ISBN-13: 978-0486659848, ISBN-10: 0486659844.
 Bibliography  525

[23] S. Goudsmit, G. E. Uhlenbeck, Physica, 1926, 6, 273.

[24] L. H. Thomas, Motion of the spinning electron, Nature, 117: 514. Bibcode: 1926Natur.
117..514T. doi:10.1038/117514a0.

[25] Collected Papers of L D Landau, Ed. D ter Haar, NY, 1965 (Reprint of Landaus papers);
Landau, L. D.; and Lifschitz, E. M.; (1977). Quantum Mechanics: Non-relativistic
Theory. Course of Theoretical Physics. Vol. 3 (3rd ed. London: Pergamon Press). ISBN
0750635398 ; H. Murayama, Landau Levels, hitoshi.berkeley.edu/221a/landau.pdf.

[26] J. von Neumann, The Mathematical Foundations of Quantum Mechanics, Princeton

University Press, 1955.

[27] M. A. Schlosshauer, Decoherence and the Quantum-to-Classical Transition, (The

Frontiers Collection) Springer-Verlag, Berlin, Heidelberg, (2007), and references
therein. ISBN-13: 978-3540357735 ISBN-10: 3540357734; C. J. Riedel, W. H. Zurek,
M. Zwolak, The objective past of a quantum universe: redundant records of consistent
histories, Physical Review A, 2016, 93:032126; R. B. Griffiths, Consistent Quantum
Theory, Cambridge University Press, Cambridge, 2002; R. Omn’es, Interpretation of
Quantum Mechanics, Princeton University Press, Princeton, 1994; M. Gell-Mann, The
Quark and the Jaguar, W.H. Freeman, New York, 1994; M. Gell-Mann and J.B. Hartle,
Quantum Mechanics in the Light of Quantum Cosmology, in Complexity, Entropy, and
the Physics of Information, ed. by W. Zurek, Addison Wesley, Reading, MA, 1990, P. C.
Hohenberg, An introduction to consistent quantum theory, Reviews of Modern Physics,
2010, 82:2835–2844; E. Joos and H. D. Zeh, The emergence of classical properties
through interaction with the environment, Zeitschrift fr Physik B, 1985, 59:223.

[28] W. H. Zurek, Environment-assisted invariance, entanglement, and probabilities

in quantum physics, Physical Review Letters, 2003, 90:120404; Probabilities from
entanglement, Borns rule from envariance, Physical Review A, 2005, 71:052105.

[29] J. B. Hartle, Quantum mechanics of individual systems, American Journal of Physics,

1968, 36(8):704–712. doi:10.1119/1.1975096.

[30] A. Einstein, B. Podolsky, N. Rosen, Can quantum-mechanical description of physical

reality be considered complete?, Physical Review, 1935, 47(10):777–780. Bibcode:
1935PhRv...47..777E. doi:10.1103/PhysRev.47.777. M. D. Reid, P. D. Drummond,
W. P. Bowen, E. G. Cavalcanti, P. K. Lam, H. A. Bachor, U. L. Andersen, G.
Leuchs, Colloquium: the Einstein–Podolsky–Rosen paradox: from concepts to
applications, Reviews of Modern Physics, 2009, 81:1727, and references therein.
doi :10.1103/RevModPhys.81.1727.
526  Quantum Mechanics

[31] A. Aspect, P. Grangier and G. Roger, Experimental tests of realistic local

theories via Bell’s theorem, Physical Review Letters, 1981, 47(7):460–463, Bib-
code:1981PhRvL..47..460A, doi:10.1103/PhysRevLett.47.460; A. Aspect; J. Dalibard;
G. Roger, Experimental test of Bell’s inequalities using time-varying analyz-
ers, Phys. Rev. Lett., 1982, 49(25): 1804–1807, Bibcode:1982PhRvL..49.1804A,
doi:10.1103/PhysRevLett.49.1804.
[32] N. D. Mermin, Simple unified form for the major no hidden variables theorems,
Physical Review Letters, 1990, 65:1838.
[33] T. J. Hollowood, Copenhagen quantum mechanics, Contemporary Physics, 2016,
57(3):289. doi:10.1080/00107514.2015.1111978 [arXiv:1511.01069 [quant-ph]].
[34] Coherence, 2014 1h 29m Coherence is an American science fiction film directed
by James Ward Byrkit in his directorial debut. Release date: June 20, 2014
(New York City).
[35] J. Wisdom, S. J. Peale, F. Mignard, The chaotic rotation of hyperion, ICARUS, 1984,
58:137–152.
[36] M. Born, J. R. Oppenheimer, Zur Quantentheorie der Molekeln [On the Quantum
Theory of Molecules], Annalen der Physik (in German), 1927, 389(20): 457–484.
Bibcode:1927AnP...389..457B. doi:10.1002/andp.19273892002.
[37] D. M. Ceperley, B. J. Alder, Ground state of the electron gas by a stochastic method,
Physical Review Letters, 1980, 45:566; J. P. Perdew, Y. Wang, Accurate and simple
analytic representation of the electron-gas correlation energy, Physical Review B, 1992,
45:13244; J. Sun et al., Correlation energy of the uniform electron gas from an inter-
polation between high- and low-density limits, Physical Review B, 2010, 81:085123;
R. G. Parr and W. Yang, Density-Functional Theory of Atoms and Molecules (Oxford
U. Press, 1989); C. Fiolhais, F. Nogueira, M. A. L. Marques (Editors) A Primer in
Density-Functional Theory, Lecture Notes in Physics Volume 620, 2003, Springer-
Verlag, Hedelberg, Berlin, ISBN: 978-3-540-03083-6 (Print) 978-3-540-37072-7 (Online);
J. P. Perdew, K. Burke and M. Ernzerhof, Generalized gradient approximation made
simple, Physical Review Letters, 1996, 77:3865; J. Sun et al. Density functionals that
recognize covalent, metallic,and weak bonds, Physical Review Letters, 2013, 111:106401.
[38] T. Banks, Density Functional Theory for Field Theorists I, cond-mat arXiv:1503.02925.
[39] A. D. Becke, Density functional exchange energy approximation with correct asymp-
totic behavior, Physical Review A, 1988, 38:3098–3100; A. D. Becke, Density functional
thermochemistry. III. The role of exact exchange, Journal of Chemical Physics, 1993,
98: 5648–5652.
 Bibliography  527

[40] A. Georges, G. Kotliar, W. Krauth and M. Rozenberg, Dynamical mean-field

theory of strongly correlated fermion systems and the limit of infinite dimen-
sions, Reviews of Modern Physics, 1996, 68(1):13, Bibcode:1996RvMP...68...13G.
doi:10.1103/RevModPhys.68.13; A. Georges and G. Kotliar, Hubbard model in infinite
dimensions, Physical Review B, 1992, 45(12):6479, Bibcode:1992PhRvB..45.6479G.
doi:10.1103/PhysRevB.45.6479.

[41] A. K. Soper and C. J. Benmore, Quantum differences between heavy and light water,
Physical Review Letters, 2008, 101:065502.

[42] W. L. Faissler, Introduction to Modern Electronics, John Wiley and Sons, New
York, 1991; R. Wolfson, Understanding Modern Electronics Course No. 1162,
www.thegreatcourses.com/courses/understanding-modern-electronics.html

[43] D. J. Griffiths, Introduction to Quantum Mechanics, 2nd Edition, Cambridge

University Press, 2017.

[44] R. Kronig and W. G. Penney, Quantum mechanics of electrons in crystal lattices,

Proceedings of the Royal Society of London A, 1931, A130:499.

[45] S. Chandrasekhar, The density of white dwarf stars, Philosophical Magazine (7th
series), 1931, 11:592–596; The maximum mass of ideal white dwarfs, Astrophysical
Journal, 1931, 74:81–82; The Highly Collapsed Configurations of a Stellar Mass
(second paper), S. Chandrasekhar, Monthly Notices of the Royal Astronomical Society,
1935, 95:207–225; L. D. Landau, On the Theory of Stars, in Collected Papers of L. D.
Landau, ed. and with an introduction by D. ter Haar, New York: Gordon and Breach,
1965; originally published in Phys. Z. Sowjet. 1 (1932), 285.

[46] Collected Papers of L. D. Landau, Ed. D ter Haar, New York, 1965 (Reprint of Landaus
papers) ; Cross, Michael. “Fermi Liquid Theory: Principles” (PDF). California Institute
of Technology, http://www.pmaweb.caltech.edu/ mcc/Ph127/c/Lecture9.pdf.

[47] R. Shankar, Renormalization group approach to interacting fermions, Reviews of

Modern Physics, 1994, 66:129; J. Polchinski, Proceedings of the 1992 TASI Elementary
Particle Physics, Editors J. Polchinski and J. Harvey, World Scientific, 1992.

[48] L. N. Cooper, Bound electron pairs in a degenerate fermi gas, Physical Review, 1956,
104:1189.

[49] J. Bardeen, L. N. Cooper and J. R. Schrieffer, Microscopic theory of superconductivity,

Physical Review, 1957, 106:162.
528  Quantum Mechanics

[50] S. A. Hartnoll, A. Lucas and S. Sachdev, Holographic Quantum Matter, arXiv:1612.

07324 [hep-th].

[51] S. Popescu, A. J. Short, A. Winter, Entanglement and the foundations of statistical

mechanics, Nature Physics, 2006, 2:754; S. Deffner, W. H. Zurek Foundations of
statistical mechanics from symmetries of entanglement arXiv.org ¿ quant-ph ¿
arXiv:1504.02797, New Journal of Physics, 2016, 18: 063013. doi: 10.1088/1367-
2630/18/6/063013.

[52] D. N. Page, Average entropy of a subsystem, Physical Review Letters, 1993, 71:1291.
doi:10.1103/PhysRevLett.71.1291 [gr-qc/9305007].

[53] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, E. A. Cornell,

Observation of bose-einstein condensation in a dilute atomic vapor. Science, 1995,
269(5221):198; K. B. Davis and M. -O. Mewes, M. R. Andrews, N. J. van Druten, D. S.
Durfee, D. M. Kurn and W. Ketterle, Bose–Einstein condensation in a gas of sodium
atoms, Physical Review Letters, 1995, 75:3969. doi = 10.1103/PhysRevLett.75.3969,
url = http://link.aps.org/doi/10.1103/PhysRevLett.75.3969.

[54] E. P. Gross, Structure of a quantized vortex in boson systems, Nuovo Cimento, 1961,
20:454; Hydrodynamics of a superfluid condensate, Journal of Mathematical Physics,
1963, 4:195; L. P. Pitaevski, Vortex lines in an imperfect Bose gas, Zh. Eksp.Teor.
Fiz., 1961, 40: 646 [Sov. Phys. JETP 13, 451(1961)].

[55] Y. B. Zel’dovich and E.V. Levich, Bose condensation and shock waves in photon
spectra, Sov. Phys. JETP, 1969, 28:1287–1290; R. Y. Chiao, Bogoliubov dispersion
relation for a photon fluid: is this a superfluid? Optics Communications, 2000,
179:157–166; E. L. Bolda, R. Y. Chiao and W.H. Zurek, Dissipative optical flow in a
nonlinear Fabry-Prot cavity, Physical Review Letters, 2001, 86:416–419.

[56] J. Klaers, F. Vewinger and M. Weitz, Thermalization of a two-dimensional photonic

gas in a white-wall photon box, Nature Physics, 2010, 6:512–515; J. Klaers, J. Schmitt,
F. Vewinger and M. Weitz, Bose–Einstein condensation of photons in an optical
microcavity, Nature, 2010, 468:545–548. doi:10.1038/nature09567.

[57] J. W. S. Rayleigh, The Theory of Sound Vol. 1, Dover Publications. The book has
been digitized by Google and others and can be found online.

[58] E. Schrödinger, An undulatory theory of the mechanics of atoms and molecules,

Physical Review, 1926, 28 (Schrödinger’s original paper on the wave equation, in
which he reports without details the perturbative calculation of the Stark Effect); E.
Schrödinger, Ann. Phys. 80, 437, (1926).
 Bibliography  529

[59] L. Brillouin, Les problmes de perturbations et les champs self-consistents,

Journal de Physique et Le Radium, 1932, 3:373–389. doi: 10.1051/jphys-
rad:jphysrad0193200309037300; E. P. Wigner, On a Modification of the Rayleigh-
Schrödinger Perturbation Theory, Volume A/4 of The Collected Works of Eugene
Paul Wigner, 131, ed. A. S. Wightman, Springer-Verlag, Berlin Heidelberg, 1997.
isbn=”978-3-642-59033-7”, doi=”10.1007/978-3-642-59033-7-12”.

[60] A. Dalgarno, J. T. Lewis, The exact calculation of long-range forces between atoms
by perturbation theory, Proceedings of the Royal Society of London A, 1995, A233:70.

[61] R. P. Feynman, Physical Review, 1939, 56:340; H. Hellmann, A combined approx-

imation procedure for calculation of energies in the many electron problem, Acta
Physicochimica URSS, 1934.

[62] Google the Wikipedia article on the Wigner-Eckart theorem for proofs, and a host of
references.

[63] W. Heisenberg, Zur Theorie des Ferromagnetismus, Zeitschrift fr Physik, 1928,

49(9–10):619–636. Bibcode:1928ZPhy...49..619H. doi:10.1007/BF01328601.

[64] N. F. Mott, Metal-Insulator Transitions, 2nd edition, Taylor & Francis, London, 1990.
ISBN 0-85066-783-6, ISBN 978-0-85066-783-7, and references therein.

[65] F. J. Dyson, The radiation theories of Feynman, Schwinger and Tomonaga, Physical
Review, 1949, 75:486.

[66] Einstein’s route to the statistical properties of photons is described on pages 351–52
of Introduction to Quantum Mechanics, 2nd Edition, by D.J. Griffiths, Cambridge
University Press, Cambridge, 2017.

[67] D. J. Griffiths, Introduction to Quantum Mechanics, 2nd Edition, Cambridge

University Press, 2017, Section 9.3.

[68] D. J. Griffiths Introduction to Electrodynamics, 4th Edition, Pearson, 2013.

[69] A. Shapere and F. Wilczek, eds. Geometric Phases in Physics, World Scientific,
Singapore, 1989.

[70] Y. Aharonov and D. Bohm, Significance of electromagnetic potentials in the quantum

theory, Physical Review, 1959, 115(3):485–489.

[71] D. Chambers, Shift of an electron interference pattern by enclosed magnetic flux,

Physical Review Letters, 1960, 5:3.
530  Quantum Mechanics

[72] J. P. Preskill, Notes on Quantum Computation, www.theory.caltech.edu/people/

preskill/ph229/

[73] http://www.physics.rutgers.edu/gmoore/511Fall2014/Physics511-2014-Ch2-
Topology.pdf

[74] A. Stern, Anyons and the quantum Hall effect A pedagogical review and references
therein. (You can download the PDF from the Wikipedia page on anyons). Annals of
Physics. 323: 204. arXiv:0711.4697v1Freely accessible. Bibcode:2008AnPhy.323..204S.
doi:10.1016/j.aop.2007.10.008.

[75] D. C. Tsui, H. L. Stormer, A. C. Gossard, Two-dimensional magnetotransport

in the extreme quantum limit, Physical Review Letters, 1982, 48(22):1559. Bib-
code:1982PhRvL..48.1559T. doi:10.1103/PhysRevLett.48.1559; H.L. Stormer (1999).
Nobel Lecture: The fractional quantum Hall effect. Reviews of Modern Physics.
71 (4): 875. Bibcode:1999RvMP...71..875S. doi:10.1103/RevModPhys.71.875; R.
B. Laughlin, Anomalous quantum hall effect: an incompressible quantum fluid
with fractionally charged excitations, Physical Review Letters, 1983, 50(18):1395.
Bibcode:1983PhRvL..50.1395L. doi:10.1103/PhysRevLett.50.1395.

[76] H. Jeffreys, On certain approximate solutions of linear differential equations

of the second order, Proceedings of the London Mathematical Society, 1924,
23:428–436. doi:10.1112/plms/s2-23.1.428; L. Brillouin, La mcanique ondulatoire
de Schrdinger: une mthode gnrale de resolution par approximations successives,
Comptes Rendus de l’Academie des Sciences, 1926, 183:24–26; H. A. Kramers,
Wellenmechanik und halbzhlige Quantisierung, Zeitschrift fr Physik, 1926, 39(10–
11):828–840. Bibcode:1926ZPhy...39..828K. doi:10.1007/BF01451751; G. Wentzel,
Eine Verallgemeinerung der Quantenbedingungen fr die Zwecke der Wellen-
mechanik, Zeitschrift fr Physik, 1926, 38(6–7):518–529. Bibcode:1926ZPhy...38..518W.
doi:10.1007/BF01397171.

[77] T. Banks and C. M. Bender, Coupled anharmonic oscillators. II. Unequal-mass case,
Physical Review, 1973, D8:3366, doi:10.1103/PhysRevD.8.3366.

[78] G. ’t Hooft, Symmetry breaking through Bell–Jackiw anomalies, Physical Review

Letters, 1976, 37:8–11. doi: 10.1103/PhysRevLett.37.8.

[79] The Meson Table, 2018 edition, Volume II. http://www.pdg.lbl.gov/

[80] http://www.cond-mat.de/events/correl11/manuscript/Koch.pdf, and references

therein.
 Bibliography  531

[81] P. A. M. Dirac, Quantum Mechanics, Section 32, 4th Edition, Oxford University Press,
1958.

[82] F. J. Dyson, The S matrix in quantum electrodynamics, Physical Review, 1949,

75:1736. Bibcode:1949PhRv...75.1736D. doi:10.1103/PhysRev.75.1736. J. Schwinger,
On Green’s functions of quantized fields I + II, Physical Review, 1951, 37:452–459.
Bibcode:1951PNAS...37..452S. doi:10.1073/pnas.37.7.452. PMC 1063400, Ibid. 1953,
728, Proceedings of the National Academy of Sciences of the United States of America,
1951, 37:452.

[83] T. I. Banks and C. M. Bender, Anharmonic oscillator with polynomial self-interaction,

Journal of Mathematical Physics, 1972, 13:1320. doi:10.1063/1.1666140.

[84] I. M. Gelfand and A. M. Yaglom, Integration in functional spaces and it applications in

quantum physics, Journal of Mathematical Physics, 1960, 1:48. doi:10.1063/1.1703636;
G. V. Dunne, Functional determinants in quantum field theory, Journal of Physics A,
2008, 41:304006. doi:10.1088/1751-8113/41/30/304006, [arXiv:0711.1178 [hep-th]].

[85] R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals, McGraw
Hill, 1960, ISBN-13: 978-0070206502, ISBN-10: 0070206503.

[86] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information,

Cambridge University Press, 2013; E. Rieffel, W. H. Pack, Quantum Computing :
A Gentle Introduction, MIT Press, Cambridge, MA, 2011; N. D. Mermin, Quantum
Computer Science: An Introduction, Cambridge University Press, 2007; M. M. Wilde,
Quantum Information Theory, 2nd Edition, Cambridge University Press, 2017,
ISBN-13: 978-1107176164 ISBN-10: 1107176166.

[87] J. P. Preskill, Notes on Quantum Computation, www.theory.caltech.edu/people/

preskill/ph229/; E. Dennis, A. Kitaev, A. Landahl and J. Preskill, Topological
quantum memory, Journal of Mathematical Physics, 2002, 43:4452.

[88] R. Orus, Advances on tensor network theory: symmetries, fermions, entanglement, and
holography, The European Physical Journal B, 2014, 87: 280. doi:10.1140/epjb/e2014-
50502-9,[arXiv:1407.6552 [cond-mat.str-el]]; J. C. Bridgeman and C. T. Chubb,
Hand-waving and Interpretive Dance: An Introductory Course on Tensor Networks,
ArXiv e-prints (Mar, 2016) 1603.03039.

[89] W. Pfaff, B. Hensen, H. Bernien, S. B. van Dam, M. S. Blok, T. H. Taminiau,

M. J. Tiggelman, R. N. Schouten, M. Markham, D. J. Twitchen and R. Hanson,
Unconditional quantum teleportation between distant solid-state qubits, Science,
2014, 345:532–535. doi:10.1126/science.1253512,arXiv:1404.4369 [quant-ph].
532  Quantum Mechanics

[90] D. Deutsch and R. Jozsa, Rapid solution of problems by quantum computation,

Proceedings of the Royal Society of London A, 1992. doi: 10.1098/rspa.1992.0167.

[91] P. W. Shor, Polynomial-time algorithms for prime factorization and dis-

crete logarithms on a quantum computer, SIAM Review, 2006, 41(2):303,
http://epubs.siam.org/doi/abs/10.1137/S0036144598347011.

[92] R. Van Meter and C. Horsman, A blueprint for building a quantum computer, Com-
munications of the ACM, 2013, 56:84; B. Lekitsch, S.Weidt, A.G. Fowler, K. Mlmer,
S. J. Devitt, C. Wunderlich and W. K. Hensinger, Blueprint for a microwave trapped
ion quantum computer, Science Advances, 2017. e1601540 Open Access CCBY; G.
Popkin, Quest for qubits, Science, 2016, 354:1090. doi: 10.1126/science.354.6316.1090.

[93] J. D. Bekenstein, Black holes and entropy, Physical Review, 1973, D7:2333.
doi :10.1103/PhysRevD.7.2333; S. W. Hawking, Particle creation by black holes,
Communications in Mathematical Physics, 1975, 43:199. doi:10.1007/BF02345020.

[94] T. Jacobson, Thermodynamics of spacetime: the Einstein equation of state, Physical

Review Letters, 1995, 75:1260. doi:10.1103/PhysRevLett.75.1260.

[95] J. Polchinski, String Theory, Vols. 1–2, Cambridge Monographs in Mathematical

Physics, 2005.

[96] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Large N field

theories, string theory and gravity, Physics Reports, 2000, 323:183. doi:10.1016/S0370-
1573(99)00083-6 [hep-th/9905111].

[97] G. ’t Hooft, Determinism beneath quantum mechanics, in Quo Vadis Quantum

Mechanics? eds. A. C. Elitzur, S. Dolev and N. Kolenda, Springer-Verlag, Berlin-
Heidelberg, 2005. isbn : 978-3-540-26669-3, doi: 10.1007/3-540-26669-0-8; The Cellular
Automaton Interpretation of Quantum Mechanics. A View on the Quantum Nature
of our Universe, Compulsory or Impossible? arXiv:1405.1548 [quant-ph]; Relating
the quantum echanics of discrete systems to standard canonical quantum mechanics,
Foundations of Physics, 2014, 44:406. doi:10.1007/s10701-014-9788-y.

[98] G. ’t Hooft, Beyond relativistic quantum string theory or discreteness and determinism
in superstrings, Subnuclear Series, 2014, 50:243. doi:10.1142/9789814603904-0016.

[99] L. de Broglie, The reinterpretation of quantum mechanics, Foundations of Physics,

1970, 1:5. doi:10.1007/BF00708650 and references therein.
 Bibliography  533

[100] D. Bohm, A suggested interpretation of the quantum theory in terms of hidden

variables, I and II, Physical Review, 1952, 85(2):166; Physical Review, 85(2):180.
doi:10.1103/PhysRev.85.166. doi:10.1103/PhysRev.85.180.
[101] D. Drr and S. Teufel, Bohmian Mechanics–The Physics and Mathematics of Quantum
Theory, Springer (2009); D. Drr, S. Goldstein and N. Zangh”, Quantum Physics
Without Quantum Philosophy, Springer, 2012.
[102] E. Z. Madelung, Quantentheorie in hydrodynamischer Form, Zeitschrift fr Physik,
1927, 40:322. doi:10.1007/BF01400372.
[103] H. Everett III, “Relative State” formulation of quantum mechanics, Reviews of Modern
Physics, 1957, 29:454. doi:https://doi.org/10.1103/RevModPhys.29.454; B. S. DeWitt
and R. N. Graham, eds, The Many-Worlds Interpretation of Quantum Mechanics,
Princeton Series in Physics, Princeton University Press, 1973, ISBN 0-691-08131-X.
[104] G. Lindblad, On the generators of quantum dynamical semigroups, Communi-
cations in Mathematical Physics, 1976, 48(2):119. Bibcode:1976CMaPh..48..119L.
doi:10.1007/BF01608499; A. Kossakowski, On quantum statistical mechanics
of non-Hamiltonian systems, Reports on Mathematical Physics, 1972, 3(4):247.
Bibcode:1972RpMP....3..247K. doi:10.1016/0034-4877(72)90010-9; V. Gorini, A. Kos-
sakowski and E. C. G. Sudarshan, Completely positive semigroups of N-level systems,
Journal of Mathematical Physics, 1976, 17(5):821, Bibcode:1976JMP....17..821G.
doi:10.1063/1.522979.
[105] T. Banks, M. E. Peskin and L. Susskind, Difficulties for the evolution of pure states into
mixed states, Nuclear Physics B, 1984, 244:125–134, Bibcode:1984NuPhB.244..125B.
doi:10.1016/0550-3213(84)90184-6.
[106] E. Wigner, The Unreasonable Effectiveness of Mathematics in the Natural Sciences,
in Communications in Pure and Applied Mathematics, vol. 13, No. I, 1960. New York:
John Wiley & Sons.
[107] H. Georgi, Lie Algebras in Particle Physics, Westview Advanced Book Program,
Perseus Books Group, 1999. ISBN 0-7382-0233-9.
[108] W. Pauli, ber das Wasserstoffspektrum vom Standpunkt der neuen Quanten-
mechanik, Zeitschrift fr Physik, 1926, 36:336–363. Bibcode:1926ZPhy...36..336P.
doi:10.1007/BF01450175.
[109] W. R. Hamilton, The hodograph or a new method of expressing in symbolic language
the Newtonian law of attraction, Proceedings of the Royal Irish Academy, 1847,
3:344–353.
Index

A Anyons, 129, 156, 522

AB effect, see Aharonov–Bohm (AB) Aristotle’s intuitive notion, 15
effect Associated Laguerre polynomials, 461,
Absorption cross section, 313 462
Absorption process, 305 Associated Legendre equation, 171
Action of a matrix, 33 Atoms, and electromagnetic radiation,
Addition of angular momenta, 175–176 304–306
Adiabatic approximation, 315
Adiabatic energy eigenstates, 315–318 B
Adiabatic theorem, 316 Baker–Campbell–Hausdorff (BCH)
Aharonov–Bohm (AB) effect, 5, 129, 315, formula, 455, 500–501
320–322 Band structure, 252–256
Airy function, 526 Bardeen, Cooper, and Schrieffer (BCS)
Ammonia molecule theory, 259, 276
in Born–Oppenheimer Bayesian interpretation, 9, 21
approximation, 152–153 Bayes’ law, 16–17, 19, 53–54, 102,
quantum dynamics of, 45–52 223–224, 432, 438
as two state system, 28–29 BCH formula, see
Analytical mechanics, 23, 208 Baker–Campbell–Hausdorff
Angular momentum, 4, 324 (BCH) formula
addition of, 175–176 BCS theory, see Bardeen, Cooper, and
conservation, 351 Schrieffer (BCS) theory
Anharmonic oscillator, energy levels of, BEC, see Bose–Einstein condensation
518 (BEC)
Annihilation operators, 118, 367, see also Bell’s inequality, 229, 231, 232
Creation and annihilation Bell’s theorem, 228–234
operators Berry connection, 319
Anti-de Sitter space conformal field Berry curvature, 319
theory correspondence, 422 Berry magnetic field, 319
Antiferromagnetic interaction, 297 Berry phases, 315, 318–322
Antiparticles, 419 Berry vector potential, 319, 521–522
Antiquarks, 527 Beryllium, 185
Antiunitary operators, 144 Bessel function, 418

535
536  Index

Bipartite system, 400–402 Breit–Wigner resonance formula, 525

Bits, 399 Brillouin–Wigner perturbation theory,
Black box problems, 412 281–285, 296
Bloch wave vector, 255 Brillouin zone, 66
Bohmian approach, 53 Brownian motion, classical theory of, 219,
Bohmian field configuration, 430 220
Bohmian mechanics, 430–433 Byte, 399
Bohm–Madelung trajectories, 432
Bohr magneton, 294 C
Bohr radius, 166, 215, 505 Canonical momentum, 80–81
Bohr–Rydberg units, 239 Cartan subalgebra, 455–456
Bohr–Sommerfeld condition, 528, 529 Casimir invariants, 456
Bohr–Sommerfeld quantization rule, Cauchy distribution, 336
348–352 Cauchy–Schwarz inequality, 90, 465
Boltzmann–Gibbs hypothesis, 265–267 Cauchy’s residue theorem, 490
Boltzmann’s statistical mechanics, 125 Cayley–Hamilton theorem, 37, 500
Born approximation, 332 C-bit, 409
Born–Oppenheimer approximation, Celestial mechanics, 197
152–153, 239, 240–243, 246, 247, Cellular automaton, 427
250 Changes of basis, 135–138
Born–Oppenheimer potential, 153, 241, Chaos theory, 10
242, 251, 514–515 Chemical potential, 270
for hydrogen molecular ion, 245–246 Classical mechanics, 3, 20, 56, 208, 261
Born series, 523 Classical physics, 28
Born’s rule, 14, 434 Classical probability theory, 10, 11, 15
geometric interpretation of, 41 Classical scattering, 202
and uncertainty, 38–42 Clebsch–Gordon coefficients, 176, 288, 459
Boron, 185 Clock operator, 480
Bose–Einstein condensation (BEC), Cloning, 403–404
272–273 c-NOT operation, 408–410
Bose–Einstein statistics, 3, 322 Coefficient matrix, 401
Bose particles, see Bosons Coherent states, 117–118, 123–124, 262
Bosonic quantized fields, 113 Collapse of the wave function, 54, 222
Bosons, 127, 128, 516 Collective coordinates, 4, 18–19, 54,
ground state, 366–367 217–219, 436
statistical mechanics of, 264–265 classical history of, 220–222
Bounded operators, 140 counting states and, 216–217
Bound states, 103, 107, 530 definition of, 266
Breit–Wigner cross section, 336 of phonons, 244
 Index  537

Commutation relations, 519 Curie’s constant, 518

irreducible representations of, Cyclicity of trace, 59, 496–497, 520
168–175
Compact groups, 155 D
Completely integrable system, 351 D’Alembert equation, 123
Complex conjugation, 157, 463 Dalgarno–Lewis (D–L) method, 284
Compton scattering, 225, 227 De Broglie wavelength, 273
Compton wave-length of particle, 418 Decay of metastable states, 352–354
Computational basis, 411 Decoherence theory, 222, 224, 441
Computational complexity, 410–413 Decoherent histories, 54
Condensed matter physics, 239 Degeneracy pressure, 258
Conditional probability distribution, 13 Degenerate perturbation theory, 285–286
Bayes’ law, 16–17, 19, 53–54, 102, in macroscopic system, 295–297
223–224, 432, 438 Delta function, 447–448, 524
Conditional wave functions, 100 normalization, 93–95, 467
Conductors, Fermi liquid theory of, δ shell potential, 336–338
256–259 Density functional formalism, 244
Confidence intervals, 10, 21, 471 Density functional theory (DFT), 4, 240,
Confluent hypergeometric equation, 509 247–250
Conjugate linear operator, 144 Hartree approximation, 370
Connected component of identity, 454 Density matrix, 35–36, 47, 52, 269, 400
Conservation law, 146, 160, 179 maximal entropy, 516
Continuity of wave function, 103–105, properties of, 515–516
484–486 time evolution of, 265
Continuum eigenvalues, 139–141 Depth of computation, 411
Copenhagen interpretation, 423, 425 Determinant of matrix, 472
Coulomb expectation values of powers of Deus ex machina (DEM), 432–433
R, 292–294 Deutsch–Josza (DJ) algorithm, 412, 413
Coulomb problem, 505 DeWitt–Graham many worlds
Coulomb repulsion, 252, 515, 522 interpretation, 433–438
Coulomb scattering amplitudes, 200–204 DFT, see Density functional theory
Coulomb wave functions, 292, 508 (DFT)
Covalent bond(ing), 247, 250 Diagonalizable operators, 138–139
Covariant derivative operator, Diagonalization, conditions for, 476
208, 209 Diagonal matrix, 35, 36, 43–44, 476
Creation and annihilation operators, Diamagnetism, 517
115–117, 128 Dihedral group, 157–158
anticommutation relation, 497–498 Dipole approximation, 305
Crystalline solids, theory of, 250–252 Dipole transition probability, 305
538  Index

Dirac comb model, 255–256 Electron Hamiltonian, 246

eigenvalue equation for, 256 Electron spin, 186–190
energy spectrum for, 257 Electron spin resonance (ESR)
Dirac delta function, 92, 447–448 experiments, 190–191
Dirac notation, 463–467 Encryption procedure, 404–406
Dirac picture, 303, 383 Energy conservation, 440
Dirac’s notation, 29–31, 91 Energy levels, JWKB approximation for,
resolution of identity, 136 346–348
Dirac’s relativistic theory of electron, 290 Entangled states, 401
Direct sums, 142–143 Entanglement entropy, 402
Dispersion relation, 149, 245 Entropy of mixing, 183
D–L method, see Dalgarno–Lewis (D–L) EPR paradox, see
method Einstein–Podolsky–Rosen (EPR)
DMFT, see Dynamical mean field theory paradox
(DMFT) Equations of motion, 3, 45, 145, 148, 425,
Doping, 251, 254 431
Double slit experiment, 99–102, 424 Euclidean, 354
Dynamical mean field theory (DMFT), Hamiltonian, 81
250, 251 Heisenberg, 71, 75, 146, 167, 218,
Dynamical symmetry groups, 147–149 378, 477, 478, 488, 499
Dyson’s formula, 302 Lagrange, 449
ESR experiments, see Electron spin
E resonance (ESR) experiments
Eccentricity vector, 508 Euclidean action, 349, 353, 355, 380
Effective Hamiltonian, 286, 293 Euclidean equations of motion, 354
Effective low-energy field theories, 252 Euclidean group, 251
Effective range approximation, 334–335 The Eudaemonic Pie, 10
Eigenvalues, 2, 475 Euler Gamma function, 201
continuum, 139–141 Euler–Lagrange equations, 207–208,
of Hamiltonian operator, 3, 59, 493 511–512
of matrices, 36–38 Everett relative state interpretation,
of unitary S-matrix, 106 433–438
Einstein–Podolsky–Rosen (EPR) Exact Kohn–Sham orbitals, 370
paradox, 228–234, 325, 440 Exchange term, 369
Einstein’s field equations, 420 Excitation
Einstein’s special theory of relativity, 417 fermionic, 271
Electric dipole moment (EDM), 48 low lying, 217, 275
Electromagnetic radiation from atoms, in oscillators, 3
304–306 radial, 359
 Index  539

Expection value, 479 Feynman’s prescription, 225

Experimental probability distribution, 10 F–H theorem, 293
Explicit matrix multiplication, 473, 474 Filled Fermi sea, 366
Fine structure corrections, 290–291
F First law of thermodynamics, 515
FD statistics, see Fermi–Dirac (FD) Fluorine, 185
statistics Fokker–Planck equation, 432
Fermat’s principle, 79 Fourier series, 77–79
Fermi–Dirac (FD) statistics, 4, 127–129, Fourier theorem, 74, 92
183–184, 326 Fourier transform
Fermi energy, 257 and Fourier series, 77–79
Fermi level, 271
of position space wave functions,
Fermi liquid theory, 256–259, 271–272, 73–74
275–277
of smooth function, 482–483
Fermion ground state, 366–367
of step function, 481–482
Fermions, 127–129
as unitary operator, 93
and Grassmann integration, 391–395
Four state system, 56–57
quantum field theory of, 262–264
Fractional quantum Hall effect (FQHE),
statistical mechanics of, 264–265
324
Fermi’s golden rule, 307
Free electron gas, pressure of, 515
Fermi statistics, 263, 322
Free particle motion
Fermi surface, 253–254, 258
on discrete/continuous circles, 64–68
Fermi temperature, 276
on infinite continuous line, 68–76
Fermi velocity, 275
Newton’s equation for, 72
Feynman diagram, 225, 385–386
Feynman–Hellman theorem, 287, 317, 513 Free particle wave function, 96–99
Feynman path integral, 375, 535 Frequentist interpretation of probability,
derivation of, 375–380 9, 21
fermions and Grassmann integration, Fugacity, 273
391–395 Functions, vectors and, 44–45
finite dimensional Hilbert space,
389–390 G
at finite temperature, 386–387 Galilean boost, 148, 149
for harmonic oscillator, 380–384 Galilean transformations, 122
and JWKB approximation, 388–389 Galileo’s law, 15
for spinning particle, 389 Gamma function, 96
Wick’s theorem, 384, 385 Gates, for quantum computers, 408–410
Feynman’s dictum, 209 Gauge symmetry, 156, 208, 262
Feynman’s double slit experiment, 512 Gauge transformations, 428
540  Index

Gaussian ansatz, for ground state wave in Bohr–Rydberg units, 368

function, 362–363 for Coulomb interactions, 153
Gaussian integral, 484, 495 with creation and annihilation
Gaussian probability distribution, 10 operators, 116–117
Generalized Clebsch–Gordan coefficients, effective, 286, 293
457 eigenvalues, 3, 59, 493
Generalized uncertainty relation, 143–144 electron spin, 189, 246
Geometric phase, 318 expectation value of, 72, 248, 361,
Gibbs paradox, 124–125, 275 362, 367, 371, 513, 531–532
Gluons, 527 ground state energy of, 361–362
G–P equation, 275 harmonic atom, 532
Gram–Schmidt orthogonalization, 502 Heisenberg, 297
Gram–Schmidt process, 488 Hermiticity of, 177, 296
Grand canonical ensemble, 274 of Hubbard type, 252
Grand canonical Hilbert space, 270 for hydrogen atom, 166
Grand canonical partition function, 270 Liouville, 428
Grassmann integration, 391–395 mechanics, 79–83
Greenberg Horn and Zeilinger (GHZ) for noninteracting particles, 366
form of Bell’s theorem, 232, 234 of path integral, 384, 389–391, 535
Green’s function, 337, 345 rotational, 243
equation, 383–384, 534 of spherically symmetric potentials,
Gross–Pitaevski equation, 284, 367 176–181
Ground state energy, 248, 284, 363–364, spin-dependent, 389
372, 514 as symmetry generator, 147
Ground state persistence amplitude, 378 time-dependent, 301, 315–316
Group multiplication law, 453–454 unperturbed eigenstate, 218, 220,
Group of symmetries, 147, 160 291, 309
center of, 151 Hamilton–Jacobi (H–J) equation, 82–83,
dynamical, 147–149 160, 342, 343–344
projective representations, 149–152 Hamilton’s vector, 508
Group theory, 453–459 Harmonic oscillator, 3
Guiding center solutions, 210, 512 bosonic quantized fields, 113,
Gyromagnetic ratio, 188 127–128
coherent states, 117–119
H creation and annihilation operators,
Hadamard operator, 413, 537 115–117, 128
Hall effect, 324 eigenstates of quantized field, 123–124
Hamiltonian, 51, 76, 166, 199, 517 Fermi–Dirac statistics, 127–129
ammonia, 222 Gibbs paradox, 124–125
 Index  541

Heisenberg equation of motion, 114 mathematical structure of, 27

overlap of wave functions, 495 n-dimensional, 479–480
path integral for, 380–384 with orthonormal basis vectors, 77
quantization of fields, 113–121 real, 457
Schrödinger wave function, 121, separable, 91, 134–135, 141
126–127 spectral decomposition of, 199
second-order equations, 493–494 of square integrable functions, 91
spin statistics theorem, 129 subspace of, 330
wave–particle duality, 117, 121–127 tensor product, 365–366, 536
Hartree approximation, 184, 365–370 theory of, 133
Hartree–Fock approximation, 365–370 two independent bases for, 106
Hartree term, 248–249 unitary transformations in, 147, 158,
Heavy water, 243 454
Heisenberg antiferromagnet, 297 H–J equation, see Hamilton–Jacobi (H–J)
Heisenberg equations of motion, 71, 75, equation
146, 167, 218, 378, 477, 478, 488, Holes, 254
499 Homogeneous electron gas, 247
Heisenberg interactions, 252 Hopping matrix, 252
Heisenberg picture operator, 48 Hubbard model, 252, 295–297
Heisenberg’s uncertainty relation, 144 Hydrogen atom, 165
Helium atom, ground state energy of, addition of angular momenta,
363–365 175–176
Hermite polynomials, 495–496 conserved energy for, 351
Hermitian conjugate, 464, 465 electron spin, 186–190
Hermitian matrix, 476, 500, 519 ground state of, 513
Hermitian operator, 43–44, 138, 235, 489, Hamiltonian of spherically symmetric
491, 504, 535 potentials, 176–181
Laguerre operator as, 506–507 periodic table, 181–186
nonnegative, 497 radial equation for, 506
square of, 493 spin precession and resonance,
Hidden variable theory, 98, 231–233 190–191
Hilbert space, 133, 433, 457–459 Stern–Gerlach experiment, 191–192
finite dimensional, 90, 93, 389–390, units and symmetries, 166–168
463 Hydrogen molecular ion, 245–247
geometry of, 446
Grand canonical, 270 I
infinite dimension, 376 Identical particle statistics, 122
infinite dimensional, 91, 376 Incident amplitude, 487
of Klein–Gordon field, 496–497 Inclusive probabilities, 204
542  Index

Incoherent perturbations, 306–308 K

Infinitesimal generators, 455 Kepler problem, 505
Infinity, notion of, 64–68 Klein–Gordon (K–G) field, 123, 496–497
Influence functional method, 389 Kohn–Sham orbitals, 370
Infrared (IR) radiation, 242 Kramers’ relation, 292
Integral kernels, 142 Kronecker delta, 126, 128
Integral operators, 448 Kronig–Penney model, 255
Interpretations of quantum mechanics K–S approximation, 248–250
Bohmian mechanics, 430–433
Copenhagen interpretation, 423, 425 L
modal interpretation, 425–430 Lagrange’s equations, 80, 449
objective collapse, 438–442 Lagrangian mechanics, 79–83
realist interpretation, 425, 426 Laguerre operator, 506–507
relative state–Many Worlds, 433–438 Laguerre polynomials, 165, 461–462
statistical interpretation, 99, 226, 438 Lanczos method, 370–372
’t Hooft approach, 427–429 Landau–Fermi liquid theory, 258, 271
Intrinsic quantum probability theory, 42 Landau levels, 211, 522
Inverse matrix, 472, 486 Landau’s solution, 207–212
Irreducible representation, 456 Lande g-factor, 294
Laplace–Runge–Lenz (LRL) vector, 179,
J 508
Jeffreys–Wenzel–Kramers–Brillouin Larmor effect, 189
(JWKB) approximation, Larmor orbits, 211
341–343, 526–529 Lattice, crystal, 251
Bohr–Sommerfeld condition, Laughlin’s quasiholes, 522
348–352 The Law of the Excluded Middle, 6–8
decay of metastable states, 352–354 LDA, see Local density approximation
examples of, 354–356 (LDA)
for energy levels, 346–348 Legendre polynomials, 506
for phase shifts, 356–357 Leibniz’ rule for commutators, 114, 118
for propagator, 345–346 Levi-Civita symbol, 150, 489
in one dimension, 343–344 Lie algebra, 455
path integrals and, 388–389 Lie groups, 454–456
Schrödinger equation, 346 Lindblad equations, 439–441
semiclassical approximation, 343 Linear algebra, 5–9, 463–467
validity of, 345 Linear equations, 31
wave functions of eigenstates, Linear functionals, 447
348–352 Linear operator, 31, 137
Jordan–Wigner transformation, 128, 498 Liouville Hamiltonian, 428
 Index  543

Lippman–Schwinger equation, 330, 331, Minimum energy excitation, 278

523 Modal interpretation of quantum
Lithium, 185 mechanics, 425–430
Local density approximation (LDA), 249 Monogamy of maximal entanglement, 403
Lorentz force equation, 207–212, 512 Mott insulators, 297
Lorentz invariance, 417–419 Multiplication law, 453–454
Lorentz transformation rules, 291 Multiplication operators on function
Low temperature limit, 271–275 spaces, 45
LRL vector, see Laplace–Runge–Lenz
(LRL) vector N
l-th partial wave cross section, 524–525 Navier–Stokes equation, 432
Newton’s equation, for free particle
M motion, 72
Macroscopic system, 215
“95% confidence interval,” 10, 471
collective coordinates of, 4, 18–19, 54,
No-cloning theorem, 403
217–219, 436
Noether’s theorem, 4, 49, 149, 449–451,
degenerate perturbation theory in,
465
295–297
Nonrelativistic theory, 420
entangled state, 222
No-particle state, 365
thermodynamic equilibrium, 265
Normalized coherent state, 118, 119
Magnetic dipole moment, 188
Magnetic susceptibility, 517 Normal operators, 2, 27, 38, 138–139, 141,
Magnetization, and susceptibility 464
formulae, 517–518 N–Q–bit gate, 410
Many worlds interpretation, 433–438 n-type semiconductor, 254
Matrices Nuclear magnetic resonance (NMR)
eigenvalues of, 36–38 experiments, 190
eigenvectors of, 2, 7
operations as, 34–36 O
physical quantities as, 30–34 Objective collapse theories, 438–442
pure states, 34 Objective reality, 17–18
trace of, 1 One body operators, 366
Maximal entanglement, 403 One-dimensional particle motion, 89–93
Maximal entropy density matrix, 516 One time pad, 404
Maxwell’s electromagnetic field, 122 Operations, as matrices, 34–36
Maxwell’s equation, 291 Oracle, 412
Maxwell’s theory of light, 8 Orbital angular momentum, 187
Metastable state, decay of, 352–354 Ordinary differential equations (ODEs),
Microcanonical ensemble, 267–269 119
544  Index

Orthonormal basis vectors, 6, 31–32, degenerate perturbation theory,

37–38, 134 285–286
Orthonormal polynomials, 502–503 examples of, 287–297
Feynman–Hellmann theorem, 287
P Peter–Weyl theorem, 457
Page’s theorem, 268 Phase shift(s)
redux, 402–404 analysis, 333–335
Parallel computer, 411 JWKB approximation for, 356–357
Partial differential equations (PDEs), 22 Phase transition, 272
Partition function, 265–271 Phonons, 122, 244–245
Phonon spectrum, 245
grand canonical, 270, 276
Physical quantities, as matrices, 30–34
logarithm of, 277
Physical system, mathematical
Paschen–Back effect, 289
description of, 1
Path integral, see Feynman path integral
Plancherel’s theorem, 74, 79, 93
Pauli exclusion principle, 4, 128, 182, 254,
Planck–Einstein–Compton (PEC)
263–264
formula, 122, 123
Pauli matrices, 57, 150–151, 231, 475,
Planck length, 421, 505
498, 503, 536
Planck’s constant, 50
Pauli spins, 232
Planck’s distribution, 306
PEC formula, see Planck’s radiation law, 277–278
Planck–Einstein–Compton Poisson bracket, 81
(PEC) formula Population inversion, 306
Penrose space-time diagram, 421 Position space wave function, 253
Periodic table, 181–186 Positive energy eigenfunctions, 197–200
Permutation matrix, 13, 159 Potential scattering, 331–332
Perturbation theory, 520 Probabilities for histories, 5, 11–13, 18
time-dependent, 301 Probability amplitude, 8, 14, 39, 54, 225,
Dirac picture, 303 426
Dyson’s formula, 302 Probability density on configuration
electromagnetic radiation from space, 431
atoms, 304–306 Probability distribution, 52, 469–471
Fermi’s golden rule, 307 for finite system, 9
incoherent perturbations, 306–308 time evolution for, 11–13
selection rules, 309–310 Probability theory, 2, 42
transition amplitudes, 303–304 essentials of, 9–14
time-independent, 281 in quantum mechanics, 20–22
Brillouin–Wigner approach, violations of “intuitive” rules, 4
281–285, 296 Product of matrix, 33, 472, 481
 Index  545

Projection operators, 8, 41, 55, 229–230, of single bit, 20

266, 520 spooky nonlocality of, 421
Projective representations of groups, theory of, 14
149–152 unhappening, 19
Propagator, JWKB approximation for, Quantum teleportation, 406–408
345–346 Quantum theory of light, 8
Pseudo-real representation, 457 Quantum tunneling, 352
p-type semiconductor, 254 Quarks, 187, 527
Purification of original state (pure state), Quasiholes, 522
402 Quasiparticles, 245, 271
Pythagoras’ theorem, 2 effective mass of, 277
Query, 412, 413
Q
Q-bit gates, 410 R
Q-bits, 536–537 Radiative decay, 306–308
QCD, see Quantum chromodynamics Ramsauer–Townsend effect, 487
(QCD) Rayleigh–Schrödinger perturbation
QFT, see Quantum field theory (QFT) theory, 285
Quantization of energy, 68 Reduced matrix element, 459
Quantization of fields, 113–121 Reduced matrix elements, 310
Quantum chromodynamics (QCD), 527 Reducible representation, 456
Quantum computer, 413–414 Reflection amplitude, 487
gates operations, 408–410 Reflection coefficient, 487
Quantum computing, 5 Relative state interpretation, 433–438
Quantum electrodynamics (QED), 188, Relativistic quantum field theory, 146,
225, 292, 304 417–422
Quantum error correcting codes, 414 Renyi entropies, 402, 536
Quantum field theory (QFT), 262–264 Residue theorem, 490
Quantum Hamilton–Jacobi equation, 432 Resolution of the identity, 39, 464
Quantum information, 400–402 Resolvent operator, 346
Quantum key distribution, 404–406 Resonance frequency, 304, 308
Quantum mechanics (QM), 1–5 Resonances, 335–336, 525
classical mechanics vs., 56, 261 Roots of algebra, 456
essentials of, 15–19 Rotational energy levels, of water, 243
interpretations of, see Interpretations Rotational Hamiltonian, 243
of quantum mechanics Rotation invariance, 235
mathematics, 22–23 Rotation matrix, 503
prediction, 48–49 Runge–Lenz–Pauli symmetry, of hydrogen
probability theory, 20–22 atom, 149
546  Index

Runge–Lenz vector, 179 Shannon–von Neumann entropy, 401

Rydberg energy, 166, 505 Short time propagation kernel, 533
Simple harmonic oscillator, 113–121
S Simple Lie algebras, 455
Scattering amplitudes Single particle excitation, 258
Coulomb, 200–204 Slater determinants, 183, 366, 369
positive energy eigenfunctions and, Smooth function, Fourier transformation,
197–200 482–483
Scattering length, 334 Soi-disant probabilities, 474
Scattering (S) matrix, 106–107, 200, Soliton solutions, 261, 262
329–330, 487, 493 Space groups, 251
Scattering operator, 200 Spectral theorem, 43, 139, 141, 465, 500
Scattering problem, 68 Spectral weight, 142
Scattering states, 68 Spectrum generating symmetries, 149
Scattering theory, 329–330 Spherical Bessel equation, 337, 338,
Born approximation, 332 510–511
δ shell potential, 336–338 Spherical harmonics, 173–174
effective range approximation, Spherically symmetric potentials,
334–335 Hamiltonian of, 176–181
Lippman–Schwinger equations, 330, Spin precession, 190–191
331, 337 Spin resonance, 190–191
phase shift analysis, 333–335 Spin statistics theorem, 129
potential scattering, 331–332 Spontaneous breaking of symmetry, 245
resonances, 335–336 Spontaneous emission process, 306
Schmidt decomposition, 402, 434 Standard Model of particle physics, 420
Schrödinger equation, 22, 54, 103, 125, Standard probability theory, 473
145, 282, 317, 331, 346, 431, 478, Stark effect, 287–289
484, 527 Statistical gauge field, 322
Schrödinger wave function, 4, 82–83, 121 Statistical mechanics, 266
Schrödinger’s bomb, 55, 222–228, 437 of bosons and fermions, 264–265
Schrödinger’s cat, 55, 222–228, 236 of fermi liquid, 275–277
Schur’s lemma, 456, 504 Step function, Fourier transformation,
Schwarz inequality, 144, 499 481–483
Schwinger–Dyson (SD) equations, Stern–Gerlach experiment, 191–192, 228,
379–380, 533–534 235
Selection rules, 309–310 Stimulated emission, 305–306
Self consistent field approximation., 369 Stirling’s approximation, 470, 471
Semisimple Lie algebras, 455, 457 Stokes’ theorem, 320, 321
Separable Hilbert space, 91, 134–135 String theory, 422
 Index  547

Susceptibility, magnetization and, incoherent perturbations, 306–308

517–518 selection rules, 309–310
Symmetry groups, 147 transition amplitudes, 303–304
center of, 151 Time-independent perturbation theory,
dynamical, 147–149 281
projective representations of, 149–152 Brillouin–Wigner approach, 281–285,
Symmetry operations, 144–146 296
Symplectic transformations, 455 degenerate perturbation theory,
Systematic errors, 9–10 285–286
examples of, 287–297
T Feynman–Hellmann theorem, 287
Teleportation communication protocol, Time-ordered product of operators, 302
406–408 Topological vector space, 447
Tensor product Traceless symmetric tensors, 172
direct sums and, 142–143 Trajectories of Bohmian particles, 430
Hilbert spaces, 365–366, 536 Transition amplitudes, 303–304
matrices, 56 Transition operator, 332
Theoretical quantum computation, 399
Transition rates, 306–308
cloning and no-cloning theorem,
Transition temperatures, 272
403–404
Transmission coefficient, 487
computational complexity, 410–413
Transmitted amplitude, 487
encryption procedure, 404–406
Trotter product formula, 535
gates for operations, 408–410
2 + 1 dimensional physics, 322
monogamy of maximal entanglement,
403
Page’s theorem redux, 402–404 U
quantum information science, Unbounded operators, 139–141
400–402 Uncertainty, Heisenberg’s relation of, 488
teleportation protocol, 406–408 Unentangled state, 400–401
Thermal density matrix, 386 Unitarily equivalent, 454
Thomas precession, 189 Unitary matrix, 136–137, 500
’t Hooft’s formalism, 427–429 orthonormal basis, 481
Tight binding approximation, 252 Unitary operators, 31, 92, 144, 465, 476,
Time-dependent perturbation theory, 301 501–502
Dirac picture, 303 Unitary transformations, 31, 135–138
Dyson’s formula, 302
electromagnetic radiation from V
atoms, 304–306 Vacuum state, 365
Fermi’s golden rule, 307 Van Leuwen’s theorem, 517
548  Index

Variational ansatz, for ground state Water molecule, 243–244

energy, 363 Wave function of particle, 69
Variational principle, 283, 361, 516 Wave number operator, 70
examples of, 362–365 Wave packets, spread of, 97–99
Hartree and Hartree–Fock Wave–particle duality, 113, 117,
approximations, 365–370 121–127
Lanczos method, 370–372 Weak-field Zeeman effect, 290–294
properties of, 361–362 Wick’s theorem, 384, 385
Vector representation, 6 Width of resonance, 525
Vectors, and functions, 44–45 Wigner–Eckart theorem, 288, 458–459
Von Neumann decomposition, 434
Von Neumann entropy, 536 Z
Zassenhaus formula, 376
W Zassenhaus lemma, 500
Walsh–Hadamard operator, 409 Zeeman effect, 189, 190, 289–290
Ward–Takahashi identities, 451 Zurek’s postulation, 434