17 Springer Series in Solid-State Sciences

Edited by Peter Fulde

Springer Series in Solid-State Sciences
Editors: M. Cardona P. Fulde K. von Klitzing H.-J. Queisser

Volumes 1-39 are listed on the back inside cover

40 Semitonductor Physil:S An Introduction 59 Dynamital Protesses and

3rd Edition By K. Seeger Ordering on Solid Surfates
41 The LMTO Method Editors: A. Yoshimori and M. Tsukada
Muffin-Tin Orbitals and Electronic Structure 60 Excitonit Protesses in Solids
By H. L. Skriver By M. Veta, H. Kanzaki, K. Kobayashi,
42 Crystal Optits with Spatial Dispersion, Y. Toyozawa, and E. Hanamura
and Excitons 61 Lotalization, Interaction, and
By V. M. Agranovich and V. L. Ginzburg Transport Phenomena
Editors: B. Kramer, G. Bergmann,
43 Resonant Nonlinear Interactions of
and Y. Bruynseraede
Light with Matter
ByV.S. Butylkin, A.E. Kaplan, 62 Theory of Heavy Fermions
Yu. G. Khronopulo, and E.I. Yakubovich and Valente Fluctuations
Editors: T. Kasuya and T. Saso
44 Elastit Media with Microstructure 0
Three-Dimensional Models 63 Electronit Properties of
By I. A. Kunin Polymers and Related Compounds
Editors: H. Kuzmany, M. Mehring, and S. Roth
45 Electronit Properties of Doped Semiwnductors
64 Symmetries in Physil:S Group Theory
By B. I. Shklo'l'skii and A. L. Efros
Applied to Physical Problems
46 Topologital Disorder in Condensed Matter By W. Ludwig and C. Falter
Editors: F. Yooezawa and T. Ninomiya 65 Phonons: Theory and Experiments 0
47 Statil:S and Dynamits of Nonlinear Systems Experiments and Interpretation of
Editors: G. Benedek, H. Bilz, and R. Zeyher Experimental Results By P. Briiesch
48 Magnetit Phase Transitions 66 Phonons: Theory and Experiments 01
Editors: M. Ausloos and R. J. Elliott Phenomena Related to Phonons
49 Organit Molewlar Aggregates, Electronic By P. Briiesch
Excitation and Interaction Processes 67 Two-Dimensional Systems: Physil:S
Editors: P. Reineker, H. Haken, and H.C. Wolf and New Devites
50 Multiple DitTraction of X-Rays in Crystals Editors: G. Bauer, F. Kuchar, and H. Heinrich
By Shih-Lin Chang 68 Phonon Stattering in Condensed Matter V
Editors: A.C. Anderson and J.P. Wolfe
51 Phonon Stattering in Condensed Matter
Editors: W. Eisenmenger, K. LaBmann, 69 Nonlinearity in Condensed Matter
and S. Dottinger Editors: A.R. Bishop, D.K. Campbell,
P. Kumar and S. E. 1hillinger
52 Superwndnctivity in Magnetit and Exotit
Materials 70 From Hamiltonians to Phase Diagrams
Editors: T. Matsubara and A. Kotani The Electronic and Statistical-Mechanical
Theory of sp-Bonded Metals and Alloys
53 Two-Dimensional Systems, Heterostructures, By J. Hafner .
and Superlattites
71 High Magnetit Fields in Semitonductor Physil:S
Editors: G. Bauer, F. Kuchar, and H. Heinrich
Editor: G. Landwehr
54 Magnetit Excitations and Fluctuations 72 One-Dimensional Conductors
Editors: S. Lovesey, V. Balucani, F. Borsa, By S. Kagoshima, H. Nagasawa,
and V. Tognetti and T. Sambongi
55 The Theory of Magnetism 0 73 Quantum Solid-State Physil:S
Thermodynamics and Statistical Mechanics Editors: S. V. Vonsovsky and M. I. Katsnelson
By D. C. Mattis
74 Quantom Monte Carlo Methods in Equilibrium
56 Spin F1nctuations in Itinerant Electron and Nonequilibrium Systems
Magnetism By T. Moriya Editor: M. Suzuki
57 Polycrystalline Semitonductors, 75 Pseudo Potential for the Electronit Structure of
Physical Properties and Applications Semiwnductors
Editor: G. Harbeke By M. L. Cohen and J. R. Chelikowsky
58 The Retorsion Method and Its Applitations 76 Electronit Properties of Conjugated Polymers
Editors: D. Pettifor and D. Weaire Editors: H. Kuzmany, M. Mehring, and S. Roth
Daniel C. Mattis

The Theory
of Magnetism I
Statics and Dynamics

With 58 Figures

Springer-Verlag Berlin Heidelberg New York

London Paris Tokyo
Professor Daniel C. Mattis, Ph. D.
Department of Physics, The University of Utah, 201 North Physics Bldg.
Salt Lake City, UT 84112, USA

Series Editors:
Professor Dr., Dres. h. c. Manuel Cardona
Professor Dr., Dr. h. c. Peter Fulde
Professor Dr. Klaus von Klitzing
Professor Dr. Hans-Joachim Queisser
Max-Planck-Institut fUr Festkorperforschung, Heisenbergstrasse 1
D-7000 Stuttgart 80, Fed. Rep. of Germany

1st Edition 1981

Corrected 2nd Printing 1988

lSBN-13:978-3-540-18425-6 e-lSBN-13:978-3-642-83238-3
DOl: 10.1007/978-3-642-83238-3

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. Duplication of this publication or
parts thereof is only permitted under the provisions of the German Copyright Law of September 9, 1965, in
its version of June 24, 1985, and a copyright fee must always be paid. Violations fall under the prosecution
act of the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1981

The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of a
specific statement, that such names are exempt from the relevant protective laws and regulations and
therefore free for general use.

2153/3150-543210
To N oemi, for a multitude
of reasons
Foreword

The challenge of writing a clear, concise and exciting account of a field

as complicated as the theory of magnetism is one which few would
choose to undertake. The author, in his first edition, wisely chose to
discuss models of sufficient simplicity, that the reader could follow in
detail the theoretical development of model systems which lay a funda-
mental foundation for understanding magnetism in physical systems. In
this considerably revised and expanded second edition of "Theory of
Magnetism" the author has again achieved a remarkable degree of clari-
ty while maintaining physical relevance throughout the text. The treat-
ment of many subjects of active research interest, such as the Kondo ef-
fect, itinerant ferromagnetism and magnetic solitons bring the reader in
contact with some of the most recent work in this active field. As in the
first edition an economy of formalism and an elegance of style make
reading the text a great plea:sure. It is with equal pleasure that I find
that my lifelong friend and colleague has again found the time and the
will to share his insights into this fascinating subject. We eagerly look
forward to the effects of finite temperature promised for the second
volume in this treatment.

Santa Barbara, California, March 1981 1.R. Schrieffer

Preface to the Second Corrected Printing

This book is based only in small part on the author's earlier work "The
Theory of Magnetism: an introduction to the study of cooperative
phenomena"!, which has recently gone out of print. At the time of
writing the first version, there were no contemporary books on research
in magnetism theory, a topic barely touched upon in survey courses in
solid-state physics. Since that time we have witnessed an explosion in
research and understanding in this field, which has become central to
what is now known as "condensed-matter physics". As an example, the
topics of thermodynamics and statistical mechanics and critical phen-
omena which occupied only two chapters of the earlier book, now re-
quire so much space that the entire companion volume to be entitled
"The Theory of Magnetism, Part II: Thermodynamics and Statistical
Mechanics" will be devoted to this study. The present "Part I: Statics
and Dynamics" deals with all the concepts and methodology that are
required for a coherent discipline - the nature of the many-body
ground states, the construction of suitable operators, the nature of the
elementary excitations (the "dynamics" of the title) - that is a micro-
cosm of the present world of theoretical physics.
In the writing of this book, I have kept in mind a single purpose:
to convey the intellectual content of a field that has, not without good
reason, fascinated so many creative minds. Magnetism is a marvellous
combination of the useful and the fanciful. The theory of interacting
spins has been a clarifying concept in modern physics, and has served
as a laboratory and tool for theoretical work in all disciplines. The
Ising model has branched out into biology and economics as well. But
in our own areas, the young particle physicist or field theorist of today
will be likely to know, or wish to learn, as much about the "anisotrop-
ic Heisenberg model" of magnetism, for example, as his professor of
solid-state theory knew, though his studies date back to an earlier,
simpler time. The condensed-matter physicist must know all this too,
as well as scattering theory, composite particles, solitons and vortices.

1 Harper and Row, New York 1965

X Preface

There is now but one physics, and our book will be devoted to such
ideas as can cut across interdisciplinary lines.
In the first chapter, which is to a large extent reproduced from the
earlier book, where it evoked a gratifying response from readers and
critics, we review the history of magnetism and see how this mirrored
the general evolution of physics and the intellectual development of
western thought. This review has been extended to bring it more near-
ly up to date. Obviously it is impossible to list the important research
let alone divine which of it will survive the test of time. Still, we endeav-
or to convey the flavor of the contemporary work which makes the
study of magnetism exciting, although we leave to the next volume the
description of the most exciting developments of all: the new field of
"critical phenomena".
Chapters 2-4 lay the foundations. We review one-, two- and many-
electron wavefunctions, their symmetry properties and quantum num-
bers. We develop the theorems due to Lieb and Mattis concerning the
nonexistence of a magnetic ground state in one dimension. Angular
momentum is studied in detail in Chap 3. Operator- and field-theoretic
representations are stressed, after a careful development starting from
first principles. Chapter 3 can provide some of the material for a short
course on angular momentum.
Chapter 5 deals with the elementary excitations of interacting spin
systems: magnons, solitons, vortices, bound magnons, etc. Various
linearized approximations and nonlinear improvements are developed
and criticized. Ferromagnetism and antiferromagnetism are contrasted,
ferrimagnetism is seen as a blend of the two. The effects of surfaces on
spin waves arecalculated.The one technique used consistently is scatter-
ing theory, and the identification of bound states. It was thought that
this is didactically sounder than the use of quantum-mechanical Green's
functions, which have waxed and then waned since the earlier book. We
believe the present approach is closer to first principles and easier for
the beginner to understand, and reserve the use of Green's functions to
the second volume, where their advantages will be more evident.
In Chap. 6 the methods developed in the earlier chapters are ex-
tended and applied to the study of delocalized electrons in a metal. The
conditions for them to exhibit magnetic properties are discussed, then
the low-density electron (or hole) gas is solved by the scattering form-
alism previously elaborated. The study of magnons in metals again in-
volves identification of bound states, similar to what we have already
seen in connection with the interacting spin system in Chap 5. Finally,
the appearance or disappearance of local moments in metals - the
 Preface XI

Friedel-Anderson models of a barely magnetic impurity - are solved in

great detail and generality by these methods, although the Kondo prob-
lem which results cannot be studied by the methods of this volume and
is relegated to the next volume. Such modern concerns as the existence
of a spin-glass ground state are touched upon.
Chapter I is written for all who enjoy history and are thoughtful
about the lessons it may teach us. For the other chapters, only elemen-
tary quantum mechanics is needed, and a belief that the space lattice,
always underlying magnetic systems, will supply a simplifying magic.
Solid-state people have always known that the properties of matter on
a regular lattice are simpler than in the continuum; field theorists are
only just rediscovering this. I hope physicists of all persuasions will find
topics of interest herein. To help self-study, a small number of prob-
lems is included as well as copious illustrations, tables and graphs, and a
bibliography.
In this corrected second printing a number of misprints and errors
have been corrected, however, the text remains essentially unchanged.

Salt Lake City, September 1987 Daniel Mattis

Acknowledgments

A number of people have helped with this book, or provided materials,

photographs, or ideas. We are particularly grateful to A.H. Bobeck,
M. Pomerantz, H. Fogedby, M.-B. Stearns, J.A. Tjon, J.A. Wright, and
D. Betts.
The earlier book benefited from similar help notably from P.W. An-
derson and a kind preface by J. Bardeen. Subsequent to publication,
suggestions from M.E. Fisher, H.S. Leff and M. Flicker, R.L. Peterson,
C.F. Squire and K. Yamazaki were gratefully received. Comments on
this new version are similarly welcomed. To those people named above,
to J.R. Schrieffer who has provided the foreword, and to Noemi Mattis
who co-authored the first chapter, go my deepest appreciation. The
reader will agree that the editorial work, for which thanks go to Dr. H.
Lotsch, Mr. R. Michels and their staff, has been superb.

Daniel Mattis
Contents

1. History of Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Physics and Metaphysics. . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Gilbert and Descartes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Rise of Modern Science. . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Electrodynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 11
1.5 The Electron. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 17
1.6 The Demise of Classical Physics. . . . . . . . . . . . . . . . . . . .. 19
1.7 Quantum Theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 23
1.8 Modern Foundations. . . . . . . . . . . . . . . . . . . . . . . . . . . .. 26
1.9 Magnetic Bubbles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 30
1.10 Ultimate Thin Films ............................. 33
1.11 Dilute Magnetic Alloys . . . . . . . . . . . . . . . . . . . . . . . . . .. 35
1.12 New Directions ................................. 37

2. Exchange. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 39
2.1 Exchange Equals Overlap .. . . . . . . . . . . . . . . . . . . . . . .. 41
2.2 Hydrogen Molecule. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 44
2.3 Three Hydrogen Atoms. . . . . . . . . . . . . . . . . . . . . . . . . .. 49
2.4 Nonorthogonality Catastrophe. . . . . . . . . . . . . . . . . . . .. 56
2.5 Method of Lowdin and Carr. . . . . . . . . . . . . . . . . . . . . .. 60

3. Quantum Theory of Angular Momentum. . . . . . . . . . . . . . . . .. 67

3.1 Kinetic Angular Momentum. . . . . . . . . . . . . . . . . . . . . .. 67
3.2 Spherical Harmonics. . . . . . . . . . . . . . . . . . . . . . . . . . . .. 70
3.3 Reason for Integer I and m . . . . . . . . . . . . . . . . . . . . . . .. 74
3.4 Matrices of Angular Momentum. . . . . . . . . . . . . . . . . . .. 76
3.5 Pauli Spin Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 78
3.6 Compounding Angular Momentum. . . . . . . . . . . . . . . . .. 79
3.7 Equations of Motion of Interacting Angular Momenta . .. 81
3.8 Coupled Boson Representation. . . . . . . . . . . . . . . . . . . .. 82
3.9 Rotations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 85
3.10 More on Compound Angular Momentum . . . . . . . . . . . .. 86
3.11 Other Representations. . . . . . . . . . . . . . . . . . . . . . . . . . .. 88
3.12 Spins One-Half . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 90
XIV Contents

3.13 Spins One. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 92

3.14 Quadratic Fonns. . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. 93

4. Many-Electron Wavefunctions. . . . . . . . . . . . . . . . . . . . . . . . . .. 95
4.1 Slater Determinants. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 96
4.2 Antisymmetrization............... . . . . . . . . . . . . .. 99
4.3 States of Three Electrons . . . . . . . . . . . . . . . . . . . . . . . .. 100
4.4 Eigenfunctions of Total S2 and SZ ... . . . . . . . . . . . . . .. 102
4.5 Ground State of Two Electrons: A Theorem ........... 106
4.6 Hund's Rules ................................... 109
4.7 p3 Configuration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 112
4.8 p2 and p4 Configurations ......................... 117
4.9 Independent Electrons ........................... 122
4.1 0 Electrons in One Dimension: A Theorem . . . . . . . . . . . .. 125
4.11 The Wronskian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 128
4.12 Theorem in Three Dimensions. . . . . . . . . . . . . . . . . . . . .. 129
4.13 Ordering Theorem Versus Hund's Rule ............... 132
4.14 Second Quantization ............................. 133

s. From Magnons to Solitons: Spin Dynamics ................. 138

5.1 Spin Waves as Harmonic Oscillators .................. 139
5.2 One-Magnon Eigenstates in Ferromagnets ............. 146
5.3 Two-Magnon States and Eigenstates in Ferromagnets .... 147
5.4 Bound States in One Dimension. . . . . . . . . . . . . . . . . . .. 155
5.5 Bound States in Two and Three Dimensions. . . . . . . . . .. 1.57
5.6 One-Magnon Eigenstates in Heitler-London Solid ....... 160
5.7 Nonlinear Spin-Wave Theory ....................... 162
5.8 Perturbation-Theoretic Correction . . . . . . . . . . . . . . . . .. 166
5.9 Antiferromagnetic Magnons: The One-Dimensional XY
Model ........................................ 169
5.10 Bethe's Solution of One-Dimensional Heisenberg Anti-
ferromagnet. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 175
5.11 Linearized Antiferromagnetic Magnons. . . . . . . . . . . . . .. 181
5.12 Nonlinearities in An tiferromagnetism . . . . . . . . . . . . . . .. 188
5.13 Ferrimagnetism ................................. 189
5.14 Effects of Surfaces on Spin-Wave Amplitudes .......... 192
5.15 Vortices...................................... 195
5.16 Solitons: Introductory Material. .................... 198
5.17 Solitary Wave Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
 Contents XV

6. Magnetism in Metals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

6.1 Bloch and Wannier States ......................... 208
6.2 Tight-Binding .................................. 209
6.3 Weak Magnetic Properties ......................... 215
6.4 Exchange in Solids: Construction of a Model Hamiltonian 219
6.5 Perturbation-Theoretic Derivation of Heisenberg
Hamiltonian .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 227
6.6 Heisenberg Hamiltonian in Metals ................... 229
6.7 Ordered Magnetic Metals: Deriving the Ground State .... 232
6.8 Kondo Effect .................................. 239
6.9 Spin Glasses .................................... 245
6.10 Magnetism Without Localized Spins: Preliminaries ...... 250
6.11 Low-Density Electron Gas ......................... 252
6.12 Quasi-Particles .................................. 257
6.13 Nagaoka's Model ................................ 258
6.14 Degenerate Bands and Intra-Atomic Exchange Forces .... 262
6.15 Magnons in Metals ............................... 266
6.16 Marginal Magnetism of Impurities ................... 272
6.17 Correlations and Equivalence to s-d Model ............ 280

Bibliography. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

References .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288

Subject Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 297

1. History of Magnetism

It was probably the Greeks who first reflected upon the wondrous properties of
magnetite, the magnetic iron ore FeO-Fe203 and famed lodestone (leading
stone, or compass). This mineral, which even in the natural state often has a
powerful attraction for iron and steel, was mined in the province of Magnesia.
The magnet's name the observing Grecians drew
From the magnetick region where it grew. [1.1]
This origin is not incontrovertible. According to Pliny's account the magnet
stone was named after its discoverer, the shepherd Magnes, "the nails of whose
shoes and the tip of whose staff stuck fast in a magnetick field while he pastured
his flocks." [1.2]

1.1 Physics and Metaphysics

The lodestone appeared in Greek writings by the year 800 B.C., and Greek
thought and philosophy dominated all thinking on the subject for some 23 cen-
turies following this. A characteristic of Greek philosophy was that it did not
seek so much to explain and predict the wonders of nature as to force them to
fit within a preconceived scheme of things. It might be argued that this seems to
be precisely the objective of modern physics as well, but the analogy does not
bear close scrutiny. To understand the distinction between modern and classical
thought on this subject, suffice it to note the separate meanings of the modern
word science and of its closest Greek equivalent, E'KtUr7)(J7J. We conceive science
as a specific activity pursued for its own sake, one which we endeavor to keep
free from "alien" metaphysical beliefs. Whereas, E'KtUr7)(J7J meant knowledge
for the Greeks, with aims and methods undifferentiated from those of philoso-
phy.
The exponents of one important school of philosophy, the animists, took
cognizance of the extraordinary properties of the lodestone by ascribing to it a
divine origin. Thales, then later Anaxagoras and others, believed the lodestone
to possess a soul. We shall find this idea echoed into the seventeenth century
A.D.
The school of the mechanistic, or atomistic, philosophers should not be mis-
construed as being more scientific than were the animists, for their theories were
similarly deductions from general metaphysical conceptions, with little relation
2 1. History of Magnetism

to what we would now consider "the facts." Diogenes of Apollonia (about 460
B.C.), a contemporary of Anaxagoras, says there is humidity in iron which the
dryness of the magnet feeds upon. The idea that magnets feed upon iron was also
a long lived superstition. Still trying to cheek on it, John Baptista Porta, in. the
sixteenth century, reported as follows:
I took a Loadstone of a certain weight, and I buried it in a heap of Iron-
filings, that I knew what they weighed; and when I had left it there many
months, I found my stone to be heavier, and the Iron-filings lighter: but the
difference was so small, that in one pound I could finde no sensible declination;
the stone being great, and the filings many: so that I am doubtful of the truth. [1.3]
But the more sophisticated theories in this category involved effluvia, which
were invisible emanations or a sort of dynamical field. The earliest of these is due
to Empedocles, later versions to Epicurus and Democritus. We quote a charming
accounting by the Roman poet Lucretius Carus showing that in the four cen-
turies since Empedocles, in an era of high civilization, the theory had not pro-
gressed:
Now sing my muse, for 'tis a weighty cause.
Explain the Magnet, why it strongly draws,
And brings rough Iron to its fond embrace.
This Men admire; for they have often seen
Small Rings of Iron, six, or eight, or ten,
Compose a subtile chain, no Tye between;
But, held by this, they seem to hang in air,
One to another sticks and wantons there;
So great the Loadstone's force, so strong to bear! ...
First, from the Magnet num'rous Parts arise.
And swiftly move; the Stone gives vast supplies;
Which, springing still in Constant Stream, displace
The neighb'ring air and make an empty Space;
So when the Steel comes there, some Parts begin
To leap on through the Void and enter in ...
The Steel will move to seek the Stone's embrace,
Or up or down, or t'any other place
Which way soever lies the Empty Space. [1.1]
The first stanza is a vivid enough description of magnetic induction, the
power of magnetized iron to attract other pieces of iron. Although this fact was
already known to Plato, Lucretius was perhaps among the first to notice, by'ac-
cident, that magnetic materials could also repel. The phenomenon awaited the
discovery of the existence of two types of magnetic poles for an explanation.
There followed many centuries without further progress at a time when only
monks were literate and research was limited to theological considerations.
 1.2 Gilbert and Descartes 3

The date of the first magnetic technological invention, the compass, and the
place of its birth are still subjects of dispute among historians. Considerable
weight of opinion places this in China at some time between 2637 B.C. and 1100
A.D., reflecting a historical uncertainty principle, no doubt. Many other sources
have it that the compass was introduced into China only in the thirteenth century
A.D. and owed its prior invention to Italian or Arab origin. In any event, the
compass was certainly known in western Europe by the twelfth century A.D.
It was an instrument of marvelous utility and fascinating properties. Einstein
has written in his autobiography of its instinctive appeal:
A wonder ... I experienced as a child of 4 or 5 years, when my father
showed me a compass. That this needle behaved in such a determined way did
not at all fit into the nature of events, which could find a place in the unconscious
world of concepts (effects connected with direct "touch"). I can still remember-
or at least believe I can remember-that this experience made a deep and lasting
impression upon me. [1.4]
Many authors in the middle ages advanced metaphysical explanations of the
phenomenon. However, the Renaissance scientist William Gilbert [1.5] said of
these writers:
... they have lost both their oil and their pains; for, not being practised in
the subjects of Nature, and being misled by certain false physical systems, they
adopted as theirs, from books only, without magnetical experiments, certain
inferences based on vain opinions, and many things that are not, dreaming old
wives' tales.
Doubtless his condemnation was too severe. Before Gilbert and the sixteenth
century, there had been some attempts at experimental science, although not
numerous. The first and most important was due to Pierre P61erin de Maricourt,
better known under the Latin nom de plume Petrus Peregrinus. His "Epistola
Petri Peregrini de Maricourt ad Sygerum de Foucaucourt Militem de Magnete,"
dated 1269 A.D., is the earliest known treatise of experimental physics. Pere-
grinus experimented with a spherical lodestone which he called terrel/a. Placing
on it an oblong piece of iron at various spots, he traced lines in the direction it
assumed and thus found these lines to circle the lodestone the way meridians gird
the earth, crossing at two points. These he called the poles of the magnet, by
analogy with the poles of the earth.

1.2 Gilbert and Descartes

Of the early natural philosophers who studied magnetism the most famous is
William Gilbert of Colchester, the "father of magnetism."
Gilbert shall live till loadstones cease to draw
Or British fleets the boundless ocean awe. [1.6]
4 1. History of Magnetism

The times were ripe for him. Gilbert was born in 1544, after Copernicus and
before Galileo, and lived in the bloom of the Elizabethan Renaissance. Physics
was his hobby, and medicine his profession. Eminent in both, he became Queen
Elizabeth's private physician and president of the Royal College of Physicians.
It is said that when the Queen died, her only personal legacy was a research grant
to Gilbert. But this he had no time to enjoy, for he died a few months after her,
carried off by the plague in 1603.
Some 20 years before Sir Francis Bacon, he was a firm believer in what we
now call the experimental method. Realizing that "it is very easy for men of acute
intellect, apart from experiment and practice, to slip and err," he resolved to
trust no fact which he could not prove by his own experience. De Magnete was
Gilbert's masterpiece, 17 years in the writing and containing almost all his results
prior to the date of publication in 1600. There he assembled all the trustworthy
knowledge of his time on magnetism, together with his own major contributions.
Among other experiments, he reproduced those performed three centuries earlier
by Peregrinus with the terrella; but Gilbert realized that his terrella was an actual
model of the earth and thus was the first to state specifically that the earth is itself
a magnet, "which opinion of his was no sooner broached that it was embraced
and wel-commed by many prime wits as well English as Forraine" [1.7]. Gilbert's
theory of magnetic fields went as follows: "Rays of magnetick virtue spread out
in every direction in an orbe; the center of this orbe is not at the pole (as Porta
reckons) but in the center of the stone and of the terrella" [Ref. 1.5, p.95].
Gilbert dispelled superstitions surrounding the lodestone, of which some
dated from antiquity, such as, "if a loadstone be anointed with garlic, or if a
diamond be near, it does not attract iron." Some of these had already been dis-
proved by Peregrinus in 1269, and even nearer to Gilbert's time, by the Italian
scientist Porta, founder of one of the earliest scientific academies. Let Porta
recount this:
It is a common Opinion amongst Sea-men, that Onyons and Garlick are at
odds with the Loadstone: and Steersmen, and such as tend the Mariners Card
are forbidden to eat Onyons or Garlick, lest they make the Index of the Poles
drunk. But when I tried all these things, I found them to be false: for not onely
breathing and beleching upon the Loadstone after eating of Garlick, did not stop
its Virtues: but when it was all anoynted over with the juice of Garlick, it (Hd
perform its office as well as if it had never been touched with it: and I could
observe almost not the least difference, lest I should seem to make void the en-
deavours of the Ancients. And again, When I enquired of the Mariners, Whether
it were so, that they were forbid to eat Onyons and Garlick for that reason; they
said, They were old Wives fables, and things ridiculous; and that Sea-men would
sooner lose their lives, than abstain from eating Onyons and Garlick [Ref. 1.3,
p.211].
But the superstitions survived the disproofs ofPeregrinus, Porta, and Gilbert,
and have left their vestiges in our own time and in common language. Between
 1.2 Gilbert and Descartes 5

superstition and fraud there is but a thin line, and Galileo recounts how his
natural skepticism protected him in one instance from a premature Marconi:
... a man offered to sell me a secret for permitting one to speak, through
the attraction of a certain magnet needle, to someone distant two or three thou-
sand miles, and I said to him that I would be willing to purchase it, but that I
would like to witness a trial of it, and that it would please me to test it, I being in
one room and he being in another. He told me that, at such short distance, the
action could not be witnessed to advantage; so I sent him away, and said that
I could not just then go to Egypt or Muscovy to see his experiment, but that if
he would go there himself! would stay and attend to the rest in Venice [1.8].
Medical healers of all times have been prompt to invoke magnetism. For
example, mesmerism, or animal magnetism, was just another instance of the
magnetic fluids, which we shall discuss shortly, invading the human body. Still
seeking to disprove such hypotheses, it was in the interest of science that Thomas
Alva Edison, as late as 1892, subjected himself "together with some of his colla-
borators and one dog" to very strong magnetic fields without, however, sensing
any effects.
But how could experiments disprove metaphysics? In spite of their own ex-
tensive investigations, Gilbert and Porta were themselves believers in an animi-
stic philosophy, and such were their theory and explanations of the phenomena
which they had studied. Note in what sensuous terms Porta describes magnetic
attraction:
... iron is drawn by the Loadstone, as a bride after the bridegroom, to
be embraced; and the iron is so desirous to joyn with it as her husband, and is
so sollicitious to meet the Loadstone: when it is hind red by its weight, yet it will
stand on end, as ifit held up its hands to beg of the stone, and flattering of it, ...
and shews that it is not content with its condition: but if it once kist the Load-
stone, as if the desire were satisfied, it is then at rest; and they are so mutually
in love, that if one cannot come to the other it will hang pendulous in the air ...
[Ref. 1.3, p. 201].
His explanation, or theory, for this phenomenon is no less anthropomorphic:
I think the Loadstone is a mixture of stone and iron ... whilst one labors
to get the victory of the other, the attraction is made by the combat between
them. In that body, there is more ofthe stone than of the iron; and therefore the
iron, that it may not be subdued by the stone, desires the force and company of
iron; that being not able to resist alone, it may be able by more help to defend
itself. For all creatures defend their being. [Ref. 1.3, p. 191].
To which Gilbert, picking up the dialogue 40 years later, retorts:
As if in the Loadstone the iron were a distinct body and not mixed up as the
other metals in their ores! And that these, being so mixed up, should fight with
one another, and should extend their quarrel, and that in consequence of the /
battle auxilliary forces should be called in, is indeed absurd. But iron itself, when
6 I. History of Magnetism

excited by the Loadstone, seizes iron no less strongly than the Loadstone. There-
fore those fights, seditions, and conspiracies in the stone ... are the ravings of
a babbling old woman, not the inventions of a distinguished mage. [Ref. 1.5,
p.63].
Gilbert's own ideas are themselves a curious blend of science and myth. On
the one hand he dismisses the effluvia theory of magnetism with cogent reason-
ing, although he admits this concept might apply to electricity. His arguments
are pithy: magnetic force can penetrate objects and the lodestone attracts iron
through solid materials other than air, which should act as deterrents to any
sort of effluvium. Electricity, on the other hand, is strongly affected by all sorts
of materials. But when he comes to give his own explanation for magnetic
attraction, he states it arises because "the Loadstone hath a soul." He believed
the earth to have one, and therefore the loadstone also, it "being a part and
choice offspring of its animate mother the earth. [po 210 of Ref 1.5].
Notwithstanding the shortcomings of his theory, Gilbert had indeed inagu-
rated the experimental method. At the other pole stands Rene Descartes (1596-
1650). Here was a philosopher who ignored the facts, but whose merit it was to
exorcise the soul out of the lodestone, laying the foundations of a rational theory.
Descartes is the author of the first extensive theory of magnetism, stated in his
Principia, Part IV, sections 133-183.
Since Descartes was among the "prime forraine wits" to embrace Gilbert's
hypothesis linking the lodestone to the earth, his theory of ferromagnetism is
accessory to his theory of geomagnetism. Both can be summarized as follows:
The prime imponderables were not specifically denoted "effluvia" but rather
"threaded parts" (parties canneMes). These were channeled in one-way ducts
through the earth, entering through pores in one pole while leaving through
pores in the other. Two kinds of parts were distinguished: those which could
only enter the North Pole and leave by the South, and those which made the
inverse voyage. The return trip was, in either case, by air. The parts find this a
disagreeable mode of travel, and seize upon the opportunity to cross any lodest-
one in the way. So much so, indeed, that if they chance to meet a lodestone they
will even abandon their ultimate destination and stay with it, crossing it over
and over again. This is shown in Fig. 1.1. Vortices are thus created in and around
the material. Lodestone, iron, and steel are the only materials having the proper
channels to accommodate the parts because of their origin in the inner earth.
Of these, lodestone ducts are best, whereas iron is malleable, and therefore the
furlike cillia which canse the ducts to be one-way are disturbed in the process of
mining. The threaded parts-throwing themselves upon the iron with great
speed-can restore the position of the ciIlia, and thus magnetize the metal. Steel,
being harder, retains magnetization better.
With this theory, Descartes claimed to be able to interpret all magnetic
phenomena known to his time. From today's perspective it is hard to see how he
even met Gilbert's objection to the theory of effluvia, stated a generation earlier,
 1.2 Gilbert and Descartes 7

nor how this theory could answer the practical questions which arise in the
mind of anyone working directly with magnetism. However, such was Descartes'
reputation that this theory came to be accepted as fact, and influenced all think-
ing on the subject throughout his century, and much of the eighteenth. Two of
his more prominent disciples in the eighteenth century were the famed Swiss
mathematician Leonard Euler and the Swedish mystic and physicist Emmanuel
Sweden borg.

Fig. 1.1. The first theory: Descartes' threaded parts are shown going
through the earth (center sphere) and in and around other magnetized bodies

Descartes, in his physics and in his philosophy, marks the transition between
metaphysical and scientific thought. First, he re-established confidence in the
power of reason, which was an absolute necessity for the birth of theoretical
science. Second, he postulated a dichotomy of the soul and of the body, which
opened the door to the study of nature on her own terms. In that, he was not
alone. The beginning of the seventeenth century had witnessed a widespread
mechanistic revolution in the sciences, led by such men as Gassendi, Mersenne,
Hobbes, Pascal, Huygens and others. These people often went much further than
Descartes in divorcing physics from metaphysics. Descartes still believed that
physics could be deduced from unprovable first principles, and his mechanism
was thus close to the Greeks'. He invoked metaphysics to ascertain his scientific
assumptions: his argument was that, since God created both nature and our
reason, we can trust that the certitudes He has instilled in us correspond to truth.
The other mechanists, however, were content with a more humble approach to
8 1. History of Magentism

nature: let us describe the phenomena, they would say, and not mind the deeper
essence of things.
Probably the most important contribution of mechanism to modern science
is the adoption of a separate language to describe nature, that of mathematics.
At first, there is an intuition. Galileo had already said in 1590: "Philosophy is
written in a great book which is always open in front of our eyes (I mean: the
universe), but one cannot understand it without first applying himselfto under-
standing its language and knowing the characters in which it is written. It is
written in the mathematical language" [1.9]. Descartes, almost 30 years later,
receives the same idea as an illumination, though he fails to apply it successfully.
The new language having been adopted, physics will receive an impetus from the
invention of calculus (by Newton and Leibnitz, in the 1680's) and from a
"mathematics that continually shifts its foundations and gets more abstract ...
It seems ... that advance in physics is to be associated with continual modi-
fication and generalisation of the axioms at the base of mathematics rather
than ... anyone ... scheme" [3.4]. In this way, mathematics has come to replace
metaphysics.
In magnetism, it is the French monk Mersenne, a friend of Descartes', who is
the first, in 1644, to quantify many of Gilbert's observations [1.10]. Progress in
theory remains slow. The mechanists are reluctant to speculate about deeper
causes, and the field is left to the neocartesians. Even so, by the year 1700 there
is one (as later there will be many) dissident voice singing a new tune. It belongs
to John Keill, Savilian professor of astronomy at Oxford, who in his eighth
lecture of that year observes:
It is certain that the magnetic attractions and directions arise from the struc-
ture of parts; for if a loadstone be struck hard enough, so that the position of its
internal parts be changed, the loadstone wiII also be changed. And if a loadstone
be put into the fire, insomuch as the internal structure of the parts be changed or
wholly destroyed, then it will lose also its former virtue and will scarce differ
from other stones ... And what some generally boast of, concerning effiuvia,
a subtile matter, particles adapted to the pores of the loadstone, etc, ... does
not in the least lead us to a clear and distinct explication of those operations;
but notwithstanding alI these things, the magnetick virtues must still be reckoned
amongst the occult qualities [1.11].

1.3 Rise of Modern Science

It is until the second half of the eighteenth century that we see the beginnings
of a modern scientific attack on the problems of magnetism, characterized by
a flexible interplay between theory and experiment and founded in rational
hypotheses. For a while, theory becomes variations on the theme of fluids. Even
Maxwell was to swim in this hypothesis, and it was not until the discovery of the
electron that magnetic theory could be placed on more solid ground.
 1.3 Rise of Modern Science 9

The fluid theory was originally proposed as an explanation for electricity

after the discovery by Stephen Gray in 1729 that electricity could be conveyed
from one body to another. This was through the medium of metals or other
"nonelectric" substances, i.e., substances which conduct electricity and do not
lend themselves readily to the accumulation of static charge. An early, and
eminent proponent of the "one-fluid" hypothesis was Benjamin Franklin. He
interpreted static charge as the lack, or excess, of the electric fluid. One of his
scientific disciples, a German emigre to St. Petersburg by the name of Franz
Maria Aepinus (1724-1802), applied the one-fluid theory to magnetism. His
theories of electricity and magnetism appear in Tentament Theoriae Electrici-
tatis et Magnetismi, published in st. Petersburg, 1759 [1.12]. This is an important
work for it brought many ideas of Franklin into sharp focus and gave the theo-
ries of efHuvia the coup de grace by dint of mathematical and experimental rea-
soning.
In 1733, Charles Francois duFay, superintendent of the French Royal Bo-
tanical Gardens, discovered that there were two types of electricity. These he
denoted vitreous and resinous, each of which attracted its opposite and repelled
its own kind. As this idea came into competition with the Franklin one-fluid
theory, much thought and many experiments were expended on proving one at
the expense of the other.
Some years after electricity was granted a second fluid, so was magnetism.
The two fluids were denoted austral and boreal in correspondence with the two
poles. It was said that in the natural nonmagnetic state the two fluids saturate
iron equally, but that magnetization parts them and leads them slightly towards
their respective poles, where they accumulate. The Swede Johan C. Wilcke, a
former student and collaborator of Aepinus, and the Dutchman Anton Brug-
mans presented this two-fluid hypothesis independently in 1778.
The best-known proponent of the two-fluid theory was Charles Augustin
Coulomb (l73~1806), who proposed an important modification in the theory,
and whose experiments have immortalized his name as the unit of charge and in
connection with the law of force [1.13]. It was by means of a torsion balance of
his invention that he established with some precision the law which bears his
name: that infinitesimals of either fluid, in electricity as in magnetism, attract or
repel in the ratio of the inverse of the square of their distance. After establishing
this is 1785, he performed many experiments on the thermal properties of
magnets. On the theoretical side, his principal contribution was the realization
that the magnetic fluids could not be free to flow like their electrical counterpart,
but perforce were bound to the individual molecules. Thus, he supposed each
molecule to become polarized somewhat in the process of magnetization. In this
manner it could be explained why the analogue of Gray's effect had never been
found in magnetism, and why two new poles always appeared when a magnet
was cut in twain. Coulomb was also aware that the laws of force he had dis-
covered were not applicable on an atomic scale, that a solid body was not in
stable equilibrium under the effect of the inverse square law of force alone.
10 1. History of Magnetism

However, concerning the unknown laws of molecular repulsion, attraction, and

cohesion, he was to write, "It is almost always more curious than useful to seek
to know their causes" [1.14J, a correct (if defeatist) attitude considering the
nearly complete ignorance of atomic structure in his day.
Simeon Denis Poisson (1781-1840) is the man who eventually became the
best interpreter of the physical constructs which Coulomb discovered. Here was
a brilliant mathematician whose scientific career appears to have been predes-
tined from his early studies, whose school-teacher had predicted, punning on the
verse of La Fontaine:
Petit Poisson deviendra grand
Pourvu que Dieu lui prete vie. l
To magnetism, Poisson brought the concept of the static potential, with
which he had been so successful in solving the problems of static electricity. And
having invented the mathematical theory of magnetostatics, he proceeded by
solving a considerable number of problems in that field. This work dates from
1824 onwards, a very exciting time in physics as we shall see in the following
section. But Poisson ignored all developments that came after the experiments
of Coulomb, and while he gave the full theory of all the discoveries by the master
of the torsion balance, he did not participate in the exciting movement which
followed and which was to lead science into a totally new direction.
We understand the magnetical work of Poisson today mostly in the manner
in which it was extended and interpreted by George Green (1793-1841). Pois-
son's "equivalent volume and surface distributions of magnetization" are but a
special case of one of the later Green theorems. Poisson found it convenient to
consider instead of H(r) the scalar potential V(r) with the property that

VV(r) = H(r) (1.1)

The contribution to V(r) from the point r' in a magnetic material is

(1.2)

and the potential from the entire sample is therefore,

V(r) = f m(r/). V' Ir -1 r I d r/.

vol
I 3 (1.3)

By partial integration, this is transformed into the sum of a surface integral and
of a volume integral, [1.15]

I 1
V(r) = f Ir - r'I m·dS' - f Ir - r'I V/.m(r /) d r'
sur vol
3 (1.4)

I Little Fish will become great if God allows him but to live.
 1.4 Electrodynamics 11

the latter vanishing for the important special case of constant magnetization. In
that special case, the sources behaved as if they were all on the surface of the
north and south poles of the magnet, and from this one might suspect that ex-
periments concerning the nature of magnetic fields surrounding various subst-
ances would never reveal the slightest information about the mechanism within
the material. Therefore, the fluid hypothesis seemed as good a working model
as any. Poisson also extended the theory in several directions. For example, by
means of certain assumptions regarding the susceptibility of magnetic fluids to
applied fields he obtained a law of induced magnetization, thereby explaining
the phenomenon of which Lucretius had sung.
Poisson, like Coulomb, refused to become excited about any speculation
concerning the nature of the sources of the field, which is to say, the fluids. This
reluctance to discuss the profound nature of things was, of course, an extreme
swing of the pendulum away from metaphysics. That attitude was itself deve-
loped into an all-embracing philosophy by Auguste Comte. Positivism, as it was
called, holds that in every field of knowledge general laws can only be induced
from the accumulated facts and that it is possible to arrive, in this way, at funda-
mental truths. Comte believed this to be the only scientific attitude. But, after
Poisson, the main theoretical advances will be made by those who ask why
as well as how. That is, by physicists who will make hypotheses vaster, simpler,
and more speculative than the mere facts allow. The specialization to the facts
at hand answers the how; the vaster theory, the why.

1.4 Electrodynamics

As early as the seventeenth century there was reason to connect the effects of
electricity and magnetism. For example, "in 1681, a ship bound for Boston was
struck by lightning. Observation of the stars showed that "the compasses were
changed'; 'the north point was turn'd clear south.' The ship was steered to
Boston with the compass reversed" [1.16]. The fluids had proliferated, and this
now made it desirable to seek a relationship among them. The invention of the
voltaic pile about 1800 was to stimulate a series of discoveries which would bring
relative order out of this chaos.
During the time that Poisson, undisturbed, was bringing mathematical re-
finements to the theory offluids, an exciting discovery opened the view to a new
science of electrodynamics. In April, 1820, a Danish physicist, Hans Christian
Oersted (1777-1851) came upon the long-sought connection between electricity
and magnetism. Oersted himself had been seeking to find such a connection since
the year 1807, but always unfruitfully, until the fateful day when he directed his
assistant Hansteen to try the effect of a current on a delicately suspended
magnetic needle nearby. The needle moved. The theory which had guided his
earlier research had indicated to him that the relation should manifest itself most
favorably under open-circuit conditions and not when the electric fluid was
12 1. History of Magnetism

allowed to leak away, and one can easily imagine his stupefaction when the un-
expected occurred. On July 21, 1820, he published a memoir in Latin (then a
more universal tongue than Danish), which was sent to scientists and scientific
societies around the world. 2 Translations of his paper were published in the
languages and journals of every civilized country.
The reaction was feverish; immediately work started, checking and extending
the basic facts of electromagnetism. French and British scientists led the initial
competition for discoveries. At first the French came in ahead. The French
Academy of Science of that time was a star-studded assembly, and besides Pois-
son it included such personalities as Laplace, Fresnel, Fourier, and more parti-
cularly active in this new field, Biot, Savart, Arago, and Ampere.
Dominique F. J. Arago (1786-1853) was the first to report upon the news
of Oersted's discovery, on September 11, 1820. Here was a most remarkable
scientist, who had been elected to the Academy 12 years earlier at the age of
23 as a reward for "adventurous conduct in the cause of science." The story
of his dedication is worth retelling. In 1806, Arago and Biot had been com-
missioned to conduct a geodetic survey of some coastal islands of Spain. This
was the period of Napoleon's invasion of that country, and the populace took
them for a pair of spies. After escaping from a prison, Arago escaped to Algiers,
whence he took a boat back to Marseilles. This was captured by a Spanish man
of war almost within sight of port! After several years of imprisonment and
wanderings about North Africa, he finally made his way back to Paris in the
Summer of 1809, and forthwith deposited the precious records of his survey in
the Bureau des Longitudes, having preserved them intact throughout his vicissi-
tudes. In this, he followed in a great tradition, and today's armchair scientists
may well find their predecessors an adventurous lot. Earlier, in 1753, G. W.
Richmann of St. Petersburg, had been struck dead by lightning while verifying
the experiments of Franklin. The effects of the electricity on his various organs
were published in leading scientific journals, and Priestley wrote: "It is not given
to every electrician to die in so glorious manner as the justly envied Richmann"
[1.17].
Shortly after his initial report to the Academy, Arago performed experiments
of his own and established that a current acts like an ordinary magnet, both in
attracting iron filings and in its ability to induce permanent magnetism in iron
needles.
Seven days after Arago's report, Andre Marie Ampere (1775-1836) read a
paper before the Academy, in which he suggested that internal electrical currents
were responsible for the existence of ferromagnetism, and that these currents
flowed perpendicular to the axis of the magnet. By analogy, might not steel
needles magnetized in a solenoid show a stronger degree of magnetization than

2 This memoir refers to the experiment performed "last year," by which Oersted meant "last
academic year," viz., "last April." This has often been misinterpreted, and the date 1819
incorrectly given for the discovery.
 1.4 Electrodynamics 13

those exposed to a single current-carrying wire? Ampere proposed this idea to

Arago, and they jointly performed the successful experiment on which Arago
reported to the Academy on November 6, 1820.
The English were not for behind. It took Sir Humphrey Davy (1778-1829)
until November 16 of that year to report on his similar experiments. Everyone
took particular pains to witness and record important experiments and the dates
thereof, and thus establish priority. No scientist on either side of the Channel
underestimated the importance of Oersted's discovery nor of the results which
flowed therefrom.
After Ampere's death, correspondence to him from Fresnel (one letter un-
dated, the other dated 5 June, 1821) was found among his papers, containing the
suggestion that the "Amperian currents" causing ferromagnetism should be
molecular rather than macroscopic in dimension. Ampere had hesitated on this
point. Fresnel wrote that the lack of (Joule) heating, and arguments similar to
those advanced above in connection with Coulomb's theory, suggested the
existence of elementary, atomic or molecular currents [1.18]. This must have
accorded well with Ampere's own ideas, and he made some (unpublished) cal-
culations on the basis of such a model. This work was carried on subsequently
by W. E. Weber (1804-1891), who assumed the molecules of iron or steel to be
capable of movement around their fixed centers. These molecules, in unmagne-
tized iron or steel, lie in various directions such as to neutralize each other's
field; but under the application of an external force, they turn around so that
their axes lie favorably oriented with respect to this external field. Precisely the
same concept had been stated by Ampere, in a letter addressed to Faraday dated
10 July, 1822, and it proved superior to Poisson's theory in explaining the satura-
tion of magnetization, for example. The final evolution of this idea may be traced
to J. A. Ewing (1855-1935), who pivoted tiny magnets arranged in geometric
arrays so that they might be free to turn and assume various magnetic configura-
tions. If these magnets could be assumed each to represent a magnetic molecule,
and the experimental distance scaled to molecular size, the experiments of Ewing
might have been expected to yield quantitative as well as qualitative information
about ferromagnets. As Digby had written long before, in connection with Gil-
bert's terrella, "any man that hath an ayme to advance much in naturall science,
must endeavour to draw the matter he inquireth of, into some such modell, or
some kind of manageable methode; which he may turne and winde as he
pleaseth ... " But for the unfortunate outcome, see ahead, pp. 27-28.
The fact that current loops had been found by Ampere to behave in every
manner like elementary magnets did not logically prove that ferromagnetism is
caused by internal electric currents. Nevertheless, this was the hypothesis
most economical in concepts which could be put forward, and as it turned out,
the most fruitful in stimulating new discoveries and in creating "insight"
into the "physics" of magnetism. It was Ampere and his followers, rather than
Poisson and his school, who acted in the modern style, the harmonious union of
theory and experiment, economical in hypotheses.
14 1. History of Magnetism

The modern scientific method, which found in electrodynamics one of its

early applications, seeks to imbed every phenomenon into a vaster mathematical
and conceptual framework and, within this context, to answer the question of
why it occurs, and not merely of how it happens. The long list of discoveries by
the electrodynamicists is sufficient tribute to the efficacy of this method, even
though their explanations of the cause of magnetism were to be proved wrong.
If we dwell on the subject, it is because of the popular misconception that
modern science is positivistic, an idea which is not justified by the methods with
which science is carried forward today. Einstein wrote: "There is no inductive
method which could lead to the fundamental concepts of physics ... in error
are those theorists who believe that theory comes inductively from experience."
[1.19].
The nineteenth century was so rich in interrelated theories and discoveries in
the fields of atomic structure, thermodynamics, electricity, and magnetism, that
it is quite difficult to disentangle them and neatly pursue our history ofthe theory
of magnetism. Fortunately many excellent and general accounts exist of the
scientific progress made in that epoch, in which magnetism is discussed in the
proper perspective, as one of the many areas of investigation. Here we concen-
trate only on the conceptual progress which was made and distinguish between
progress in the physical theory of magnetism and progress in the understanding
of the nature of the magnetic forces.
Much is owed to the insight of Michael Faraday (1791-1867), the humble
scientist who is often called the greatest experimental genius of his century.
Carrying forward the experiments of the Dutchman Brugmans, who had dis-
covered that (paramagnetic) cobalt is attracted, whereas (diamagnetic) bismuth
and antimony are repelled from the single pole of a magnet, Faraday studied the
magnetic properties of a host of ordinary materials and found that all matter
has one magnetic property or the other, although usually only to a very small
degree. It is in describing an experiment with an electromagnet, in which a
diamagnetic substance set its longer axis at right angles to the magnetic flux,
that Faraday first used the term, "magnetic field" (December, 1845). He was
not theoretically minded and never wrote an equation in his life. Nevertheless,
his experiments led him unambiguously to the belief that magnetic substances
acted upon one another by means of intermediary fields and not by "action
at a distance." This was best explained by Maxwell, who
... resolved to read no mathematics on the subject till I had first read
through Faraday's Experimental Researches in Electricity. I was aware that
there was supposed to be a difference between Faraday's way of conceiving
phenomena and that of the mathematicians .... As I proceeded with the study
of Faraday, I perecived that his method ... [was] capable of being expressed
in ordinary mathematical forms .... For instance, Faraday, in his mind's eye,
saw lines of force traversing all space where the mathematicians saw centres of
force attracting at a distance: Faraday saw a medium where they saw nothing
 1.4 Electrodynamics 15

but distance .... I also found that several of the most fertile methods of
research discovered by the mathematicians could be expressed much better in
terms of ideas derived from Faraday than in their original form. [1.20].
Faraday's concept of fields led him to expect that they would influence light,
and after many unfruitful experiments he finally discovered in 1845 the effect
which bears his name. This was a rotation of plane-polarized light upon passing
through a medium in a direction parallel to the magnetization. Beside the Fara-
day effect, several other magneto-optic phenomena have been found: Kerr's
magneto-optic effect is the analogue of the above, in the case of light reflected
off a magnetic or magnetized material. Magnetic double refraction, of which an
extreme example is the Cotton-Mouton effect, is double refraction of light
passing perpendicular to the magnetization. But the effect which was to have the
greatest theoretical implications was that discovered by Zeeman, which we shall
discuss subsequently.
By virtue of the physicomathematical predictions to which it led, the hypo-
thesis of fields acquired a reality which it has never since lost. Henry Adams
wrote ca. 1900:
For a historian, the story of Faraday's experiments and the invention of the
dynamo passed belief; it revealed a condition of human ignorance and helpless-
ness before the commonest forces, such as his mind refused to credit. He could
not conceive but that someone, somewhere, could tell him all about the magnet,
if one could but find the book .... [1.21]
But of course there was such a book: James Clerk Maxwell (1831-1879) had
summarized Faraday's researches, his own equations, and all that was known
about the properties of electromagnetic fields and their interactions with ponder-
able matter [1.20J.
With Faraday, Maxwell believed the electric field to represent a real, physical
stress in the ether (vacuum). Because electrodynamics had shown the flow of
electricity to be responsible for magnetic fields, the latter must therefore have a
physical representation as rates of change in the stress fields. Thus the energy
density E2/81t stored in the electric field was of necessity potential energy; and
the energy density H2/81t stored in the magnetic field was necessarily kinetic
energy of the field. This interpretation given by Maxwell showed how early he
anticipated the Hamiltonian mechanics which was later to provide the founda-
tions of the quantum theory, and how modern his outlook was.
The harmonic solutions of Maxwell's equations were calculated to travel with
a velocity close to that of light. When the values of the magnetic and electrical
constants were precisely determined, it was found by H. R. Hertz in 1888 that
these waves were precisely those of light, radio, and those other disturbances
that we now commonly call electromagnetic waves. Later, the special theory of
relativity was invented by Einstein with the principal purpose of giving to the
material sources of the field the same beautiful properties of invariance which
Maxwell had bestowed on the fields alone. One formulation of Maxwell's equa-
16 1. History of Magnetism

tions which has turned out most useful introduces the vector potential A(r,t) in
terms ofthe magnetic field H(r, t) and the magnetization M(r, t) as

vX A = H + 41tM = B. (1.5)

The vector B is defined as the curl of A, and is therefore solenoidal by definition;

that is, V.B = 0, as is instantly verified. The next equation introduces the electric
field in terms of the potentials A(r,t) and the scalar potential U(r) without the
necessity of describing the sources. More precisely,

E=-VU---.
I aA (1.6)
c at
This is but the equation of the dynamo: the motion or rate of change of the
magnetic field is responsible for an electric field, and if this is made to occur in
a wire, a current will flow. The familiar differential form of this equation,

VxE=---
I aB (1.7)
c at
is obtained by taking the curl of both sides. Next, the results of Oersted , Ampere,
Arago, and their colleagues almost all were summarized in the compact equation

V X H = 41t j + ~ aD . (l.8)
c c at

where j is the real current density, and

D = E + 41tP. (1.9)

relates the electric displacement vector D to the electric field E and the polariza-
tion of material substances, P. Equation (1.8) can be transformed into more
meaningful form by using the definitions of Hand D in terms of A, P, and M,
and assuming a gauge V. A = 0:

(- V2 2
-(
+ -cI at 2 )
2
A(r t) = 41t (I.
'
- J
c
+ -cI ap
-at + V X M ). (1.10)

This equation shows that in regions characterized by the absence of all material
sources j, P, and M, the vector potential obeys a wave equation, the solutions
of which propagate with c = speed oflight. The right-hand side ofthis equation
provides a ready explanation for Ampere's famous theorem, that every current
element behaved, insofar as its magnetic properties were concerned, precisely
like a fictitious magnetic shell which would contain it. For if we replace all cons-
tant currents j by a fictitious magnetization M = r X j /2c the fields would be
unaffected. Nevertheless, for the sake of definiteness it will be useful to assume
 1.5 The Electron 17

that j always refers to real free currents, P to bound or quasi bound charges, and
M to real, permanent magnetic moments.

1.5 The Electron

Once these equations were formulated, permitting a conceptual separation of

cause and effect, of the fields and of their sources, progress had to be made in
understanding the sources. This was tied in with the nature of matter itself, a
study which at long last became "more useful than curious," to turn about the
words of Coulomb. A giant step was taken in this direction by the discovery
ofthe electron, one of the greatest scientific legacies of the nineteenth century to
our own.
While Faraday, Maxwell, and many others had noted the likelihood that
charge existed in discrete units only, this idea did not immediately make headway
into chemistry, and Mendeleev's 1869 atomic table was based on atomic weights
rather than atomic numbers. The first concrete suggestion was made in 1874 by
G. Johnstone Stoney, the man who was to give the particle its name in 1891
[1.22]. We know the electron as the fundamental particle, carrier of e = 1.602 X
10- 19 Coulomb unit of charge, and m = 9.11 X 10- 28 gram of rest mass, building
block of atoms, molecules, solid and liquid matter. But it is amusing to us to
recall that it was first isolated far from its native habitat, streaming from the
cathode of the gas discharge tubes of the 1870's. Certainly it was of mixed
parentage. To mention but two of the greatest contributors to its "discovery,"
Jean Perrin found in his 1895 thesis work that the cathode rays consisted of nega-
tively charged particles, and Thomson had obtained the ratio elm to good preci-
sion by 1897 (curvature in a magnetic field). Its existence was consecrated at the
Paris International Congress of Physics held in 1900 inaugurating the twentieth
century with a survey of the problems which the discovery of the existence of
the electron had finally solved, and of those which it now raised. [1.23].
By this time there already was great interest in the spectral lines emitted by
incandescent gases, for their discrete nature suggested that the fluids that con-
stituted the atoms were only capable of sustaining certain well-defined vibra-
tional frequencies. In 1896 Zeeman had shown that the spectral lines could be
decomposed into sets of lines, multiplets, if the radiating atoms were subjected to
intense magnetic fields. This experiment had disastrous consequences on various
hypotheses of atomic structure which had hitherto appeared in accord with
experimental observations, for example, that of Kelvin. Concerning his "gyro-
static" model of the atom as an electrified ring, Kelvin himself, in 1899, was to
write its epitaph:
No simplifying suppositions as to the character of the molecule, such as the
symmetry of forces and moments of inertia round the axis of the ring, can possi-
bly give Zeeman's normal results of the splitting of a bright line into two sharp
18 1. History of Magnetism

lines circularly polarized in opposite directions, when the light is viewed (in a
spectrograph) from a direction parallel to the lines of magnetic force; and the
dividing of each bright line into three, each plane polarized, when the light is
viewed from a direction perpendicular to the lines of force. Hence, although
from 1856 till quite lately I felt quite satisfied in knowing that it sufficed to
explain Faraday's magnetooptic discovery, I now, in the light of Zeeman's
recent discovery, discard myoId tempting gyrostatic hypothesis for an irrefrag-
able reason .... [l.24].
It was Zeeman's teacher, the Dutch theoretician Hendrik Antoon Lorentz
(1853-1928), who provided the first reasonable theory of the phenomenon, and
he based it on the electron theory. Later he extended his electron theory to give
a physical basis for all of electrodynamics. "The judgment exhibited by him here
is remarkable," wrote Sommerfeld [l.25], "he introduced only concepts which
retained their substance in the later theory of relativity." Consider Lorentz'
formulation of the force exerted by electric and magnetic fields on a (nonrelativi-
stic) particle of charge e and velocity v

(1.11 )

In the Lorentz theory of the Zeeman effect, one assumes the electron to be held
to the atom by a spring (of strength K) and a weak magnetic field H( = B) applied
along the z direction. Thus, by the laws of Newton and Lorentz,

mr = F = - Kr + !:...e i' X H. (1.12)

The parallel motion z(t) proceeds at the unperturbed frequency

z(t) = z(O) cos wot with (1.13)

but in the perpendicular direction there is a frequency shift in the amount

I1w ~± 21 eH + O(H2) (assuming!!"w

me
< wo) (1.14)

according to whether the angular motion is clockwise or counterclockwise in the

plane perpendicular to the magnetic field. If this is equated to the width of the
splittings observed, it yields a value for elm within a factor of 2 from that which
had been established for the cathode rays. This factor was not to be explained
for another quarter century.
The thermal properties of magnetic substances were first investigated in a
systematic manner by Pierre Curie (1859-1906), who found M to be proportional
to the applied field H. He studied X, the constant of proportionality, known as
the magnetic susceptibility, finding for paramagnetic substances:
 1.6 The Demise of Classical Physics 19

. .A C
X= 1I m - = - . (1.15)
H~O H T

Curie's constant C assumed different values depending on the material, with

T the temperature measured from the absolute zero. In diamagnetic substances
X is negative and varies little with T. In all ferromagnetic materials, he found a
relatively rapid decrease of the magnetization as the temperature was raised to
a critical value, now known as the Curie temperature; above this temperature,
the ferro magnets behaved much like ordinary paramagnetic substances [1.26].
While many of these results had been known for one or another material, the
scope of his investigations and the enunciation of these general laws gave parti-
cular importance to Curie's research.
The diamagnetism was explained a decade after Curie's experiments, in a
famous paper [1.27] by Langevin (1872-1946), as a natural development of
Lorentz's electronic theory of the Zeeman effect. As far as Langevin could see,
diamagnetism was but another aspect of the Zeeman effect. Without entering
into the details of his calculation, we may merely observe that this phenomenon
is apparently already contained in one of the Maxwell equations: Eqs. (1.6,7)
indicate that when the magnetic field is turned on, an electric field will result;
this accelerates the electron, producing an incremental current loop, which in
turn is equivalent to a magnetization opposed to the applied field. Thus Lenz'
law, as this is called in the case of circuits, was supposed valid on an atomic
scale, and held responsible for the universal diamagnetism of materials. Para-
magnetism was explained by Langevin as existing only in those atoms which
possessed a permanent magnetic moment. The applied magnetic field succeeded
in aligning them against thermal fluctuations. Standard thermodynamic reason-
ing led Langevin to the relationship

(1.16)

where f = odd function of its argument. Then, in weak fields the leading term in
a Taylor series expansion of the right-hand side yields Curie's law, without
further ado.

1.6 The Demise of Classical Physics

As we know, the great development of physics in our century is indebted mainly

to the invention of quantum mechanics. Progress in the modern understanding
of magnetism has depended, to a great extent, on progress in quantum theory.
Conversely, the greatest contributions of the theory of magnetism to general
physics have been in the field of quantum statistical mechanics and thermody-
namics. Whereas understanding in this important branch was limited in the
20 1. History of Magnetism

nineteenth century to the theory of gases, the study of magnetism as a cooperative

phenomenon has been responsible for the most significant advances in the theory
of thermodynamic phase transitions. This has transformed statistical mechanics
into one of the sharpest and most significant tools for the study of solid matter.
The initial, and perhaps the greatest, step in this direction was taken in 1907
when Pierre Weiss (1865-1940) gave the first modern theory of magnetism [1.28].
Long before, Coulomb already knew that the ordinary laws of electrostatics
and magnetostatics could not be valid on the atomic scale; and neither did Weiss
presume to guess what the microscopic laws might be. He merely assumed that
the interactions between magnetic molecules could be described empirically by
what he called a "molecular field." This molecular field Hm would act on each
molecule just as an external field did, and its magnitude would be proportional
to the magnetization and to a parameter N which would be a constant physical
property of the material. Thus, Weiss' modification of the Langevin formula was

(1.17)

If the molecular field were due to the demagnetizing field caused by the free
north and south poles on the surface of a spherical ferromagnet, the Weiss con-
stant would be N ::::; -41t/3 in some appropriate units. In fact, experiments gave
to N the value::::; + 104 for iron, cobalt, or nickel, as could be determined by
solving the equation above for ~(N,T) and fitting N to secure best agreement
with experiment. This is easy to do at high temperature, where/can be replaced
by the leading term in the Taylor series development. That is,

~= C (H+N~) (1.18)
T

predicting a Curie temperature (where ~/H = (0) at

(1.19)

Above the Curie temperature, the combination of the two equations above
yields

~ C (1.20)
X=-=--
H T- Tc

the famous Curie-Weiss law which is nearly, ifnot perfectly, obeyed by all ferro-
magnets. This agreement with experiment was perhaps unfortunate, for it meant
that the gross features of magnetism could be explained without appeal to any
particular mechanism and with the simplest of quasi thermodynamic arguments.
Therefore any model which answered to those few requirements would explain
 1.6 The Demise of Classical Physics 21

the gross facts, and a correct theory could be tested only on the basis of its pre-
dictions for the small deviations away from the laws of Weiss. It was to be some
time before the deviations were systematically measured and interpreted, butthe
large value of Weiss' constant N was intriguing enough in his day and pointed to
the existence of new phenomena on the molecular scale.
To the mystery of the anomalously large molecular fields was added that of
the anomalous Zeeman effect, such as that observed when the sodium D lines
were resolved in a strong magnetic field. Unlike the spectrum previously des-
cribed, these lines split into quartets and even larger multiplets, which could no
more be explained by Lorentz' calculation than the normal Zeeman effect could
be explained by Kelvin's structures. Combined with many other perplexing facts
of atomic, molecular, and solid-state structure, these were truly mysteries.
A development now took place which should have effected a complete and
final overthrow of the Langevin-Lorentz theory of magnetism. It was the dis-
covery of an important theorem in statistical mechanics by Niels Bohr (1885-
1962), contained in his doctoral thesis of 1911. In view of the traditional obscu-
rity of such documents, it is not surprising that this theorem should have been
rediscovered by others, in particular by Miss J. H. van Leeuwen in her thesis
work in Leyden, 8 years later. In any event, with so many developments in
quantum theory occurring in that period, the Bohr-van Leeuwen theorem was
not universally recognized as the significanf landmark it was, until so pointed
out by Van Vleck in 1932 [1.29).3 Consider the following paraphrase of it, for
classical nonrelativistic electrons: At any finite temperature, and in all finite
applied electrical or magnetical fields, the net magnetization of a collection of
electrons in thermal equilibrium vanishes identically. Thus, this theorem demon-
strates the lack of relevance of classical theory and the need for a quantum
theory. We now proceed to a proof to of it.
The Maxwell-Boltzmann thermal distribution function gives the proba-
bility that the nth particle have momentum Pn and coordinate r n as the following
function:

dP(pl> ... ,PN; rl> ... , r N) = exp [ - klrf7C'(PI> ... ; ... ,rN)]dPI ... drN
(1.21 )

where k Boltzmann's constant

=
r= temperature
f7C' = Hamilton's function, total energy of the system.
The thermal average (TA) of any function F(Pl' ... ; ... r N ) of these generalized
coordinates is then simply

3 Summarizing the consequences of this theorem, Van Vleck quips: " ... when one attempts to
apply classical statistics to electronic motions within the atom, the less said the better ... "
22 1. History of Magnetism

fFdP
(F)TA = fdP (1.22)

with the integration carried out over all the generalized coordinates, or "phase
space."
Next we consider the solution of Maxwell's equation, (1.8), giving the mag-
netic field in terms of the currents flowing. As the current density created
by the motion of a single charge en is in = env n, the integral solution of (1.8)
becomes

(1.23)

where Rn = r - rn' This is the law of Biot-Savart in electrodynamics. The

quantity Vn x Rn which appears above is closely related to the angular momen-
tum, to be discussed in Chap. 3. To calculate the expected magnetic field caused
by the' motion of charges in a given body, it suffices to take the thermal average
of H(r) over the thermal ensemble dP characteristic of that body. For this
purpose it is necessary to know something about Hamilton's function ~, also
denoted the Hamiltonian, for particles in electric and magnetic fields. In the
absence of magnetic fields, this function is

~= I: i-mnv: + U(r"
n
.. , ,rN ) (1.24)

where Vn = Pnlmn = drnldt as a consequence of Lagrange's equations; the first

term being the kinetic energy of motion and the second the potential energy due
to the interactions of the particles amongst themselves and with any fixed poten-
tials not necessarily restricted to Coulomb law forces. oWe observe from Max-
well's equation, (1.6), that Newton's second law enE = Pn assumes once more
the form in which forces are derivable from potentials if we incorporate the
magnetic field by the substitution Pn - Pn + enA(rn, t)le. This is in fact correct,
and the appropriate velocity for a particle under the effects of a vector poten-
tials is

(1.25)

For a more rigorous derivation, see the text by Goldstein [1.30]. Substitution of
(1.25) into (1.24) yields the correct Hamiltonian in a field.
It shall be assumed that a magnetic field as described by a vector potential
A(r, t), is applied to the particles constituting a given substance, and the resultant
motions produce in turn a magnetic field H(r) as given by (1.23). The thermal
average of this field can be computed with three possible outcomes:
 1. 7 Quantum Theory 23

1) (H)TA has a finite value which is independent of A in the limit A - O. In

this case, the substance is evidently a ferromagnet.
2) (H)TA is parallel to the applied magnetic field V X A, but its magnitude is
proportional to the applied field and vanishes when the latter is turned off.
This is the description of a paramagnetic substance.
3) (H)TA is proportional but antiparallel to the applied field. Such a substance is
repelled by a magnetic pole and is a diamagnet.
The actual calculation is simplicity itself. In calculating the thermal average
of H(r), we note that the convenient variables of integration are the Vn and not
the Pn. A transformation of variables in the integrals (1.22) is then made from
the Pn to the V n, and according to the standard rules for the transformation of
integrals, the integrands must be divided by the Jacobian of the transformation

J = det 1I~::nll = l. (l.26)

As is seen, the Jacobian equals unity, and therefore A(r, t) simply disappears
from the integrals, both from the Maxwell-Boltzmann factor and from the
quantity H(r). Finally, the thermal expectation value of the latter vanishes
because it is odd under the inversion Vn - - V n• This completes the proof that
the currents and associated magnetic moments induced by an external field all
vanish identically.
Therefore, by actual calculation the classical statistical mechanics of charged
particles resulted in none of the three categories of substances described by
Faraday, but solely in a fourth category, viz., substances having no magnetic
properties whatever. This stood in stark conflict with experiment.

1.7 Quantum Theory

In the period between 1913 and 1925 the "old quantum theory" was in its prime
and held sway. Bohr had quantized Rutherford's atom, and the structure of
matter was becoming well understood. Spatial quantization was interpretable on
the basis of the old quantum theory, and a famous experiment by Stern and
Gerlach [1.31] allowed the determination of the angular momentum quantum
number and magnetic moment of atoms and molecules. The procedure was to
pass an atomic beam through an inhomogeneous magnetic field, which split it
into a discrete number of divergent beams. In 1911 the suggestion was made
that all elementary magnetic moments should be an integer multiple, of what
later came to be called the Weiss magneton, in honor of its inventor. Although
the initial ideas were incorrect, this suggestion was taken up again by Pauli in
1920, the unit magnetic moment given a physical interpretation in terms of the
Bohr atom, a new magnitude some five times larger
24 1. History of Magnetism

eli
JiB = - = 0.927 X 10- 20 erg/G ( 1.27)
2mc

and renamed the Bohr magneton. In 1921, Compton [1.32] proposed that the
electron possesses an intrinsic spin and magnetic moment, in addition to any
orbital angular momentum and magnetization. This was later proven by Gouds-
mit and Uhlenbeck, then students of Ehrenfest. In a famous paper in 1925 they
demonstrated that the available evidence established the spin of the electron as
1i/2 beyond any doubt [1.33]. The magnetic moment was assigned the value in
(1.27), i.e., twice larger than expected for a charge rotating with the given value
of angular momentum. The reason for this was yet purely empirical: the study of
the anomalous Zeeman effect had led Lande some two years earlier to his well-
known formula for the g factor, the interpretation of which in terms of the spin
quantum number obliged Goudsmit and Uhlenbeck to assign the anomalous
factor g = 2 to the spin of the electron. But a satisfactory explanation of this was
to appear within a few years, when Dirac wedded the theory of relativity to
quantum mechanics.
Developments in a new quantum mechanics were then proceeding at an
explosive rate, and it is impossible in this narrative to give any detailed account-
ing of them. In 1923 De Broglie made the first suggestion of wave mechanics
[1.34], and by 1926 this was translated into the wave equation through his own
work and particularly that of Schrodinger. Meantime, Heisenberg's work with
Kramers on the quantum theory of dispersion, in which the radiation field
formed the "virtual orchestra" of harmonic oscillators, suggested the power of
noncommutative matrix mechanics. This was partly worked out, and transfor-
mation scheme for the solution of general quantum-mechanical problem given,
in a paper by Heisenberg et al. [1.35]. Born and Wiener [1.36] then collaborated
in establishing the general principle that to every physical quantity there corres-
ponds an operator. From this it followed that Schrodinger's equation and that
of matrix mechanics were identical, as was actually shown by Schrodinger [1.37]
also in 1926. It is the time-independent formulation of his equation which
forms the basis for much of the succeeding work in solid-state physics and in
quantum statistical mechanics, when only equilibrium situations are contem-
plated, a generalization of ge = E of classical dynamics:

(1.28)

with the operator Ii V./i replacing the momentum p. in Hamilton's function. By

radiation or otherwise, a system will always end up with the lowest possible, or
ground state, energy eigenvalue Eo. When thermal fluctuations are important,
however, the proper balance between maximizing the entropy (i.e., the logarithm
of the probability) and minimizing the energy is achieved by minimizing an
appropriate combination of these, the free energy. This important connec-
 1.7 Quantum Theory 25

tion between quantum theory and statistical mechanics was forged from the
very beginnings, in the introduction of quantization to light and heat by Planck
in 1900, through Einstein's theory of radiation (1907), its adaptation by Debye
to the specific heat of solids (1912), and most firmly in the new quantum mecha-
nics. For Heisenberg's uncertainty principle,

(1.29)

which expressed the lack of commutativity of a/axn with X n , gave a natural size to
the cells in phase space in terms of Planck's constant 21t1i. More pertinently
perhaps, the interpretation of processes in quantum mechanics as statistical,
aleatory events seemed an important necessary step in applying this science to
nature. This required unravelling the properties of 'II
The importance of the energy eigenfunction'll in the subsequent development
of quantum theory was overriding, and is discussed in some detail in Chaps. 2
and 4 of this volume. Here a bare outline will suffice. In 1927 Pauli invented his
spin matrices; the next year Heisenberg and (almost simultaneously) Dirac ex-
plained ferromagnetism by means of exchange, the mysterious combination of
the Pauli exclusion principle and physical overlap of the electronic wavefunc-
tions. The next development was to consider the wavefunction itself as a field
operator, i.e., second quantization. The importance of this was perhaps not im-
mediately recognized. But the quantization of the electromagnetic field by
Planck, leading to the statistics of Bose and Einstein, had here its precise analo-
gue: the quantization of the electronic field by Jordan and Wigner [1.38] in 1928.
This incorporated Pauli's exclusion principle an led to the statistics of Fermi
and Dirac and to the modern developments in field theory and statistical mecha-
nics. In 1928 also, Dirac [1.39] incorporated relativity into Hamiltonian wave
mechanics, and the first and immediate success of his theory was the prediction
of electron spin in precisely the form postulated by Pauli by means of his 2 X 2
matrix formulation, and an explanation of the anomalous factor g = 2 for the
spin. The anomalous Zeeman effect finally stood stripped of all mystery. The
Jordan-Wigner field theory permitted one loophole to be eliminated from the
theory, viz., the existence of negative energy states. These were simply postulated
to be filled [l.40]. The striking confirmation came with the 1932 discovery by
C.D. Anderson of the positron-a hole in the negative energy Dirac sea.
Side by side with these fundamental developments, Hartree, Fock, Heitler,
London, and many others were performing atomic and molecular calcula-
tions that were the direct applications of the new theory of the quantum
electron. In 1929 Slater [1.41] showed that a single determinant, with entries
which are individual electronic wavefunctions of space and spin, provides
a variational many-electron wavefunction for use in problems of atomic and
molecular structure. Heisenberg, Dirac, Van Vleck, Frenkel, Slater, and many
others contributed to the notions of exchange, and to setting up the operator
formalism to deal with it. And so, without further trying to list individual contri-
26 1. History of Magnetism

butions, we can note that by 1930, after four years of the most exciting and
brilliant set of discoveries in the history of theoretical physics, the foundations
of the modern electronic theory of matter were definitively laid, and an epoch of
consolidation and calculation based thereupon was started, in which we yet find
ourselves at the date of this writing. We note that 1930 was the year of a Solvay
conference devoted to magnetism; it was time to pause and review the progress
which had been made, time to see if "someone ... could tell all about the
magnet. ... "
It is not out of place to note the youth of the creators of this scientific revolu-
tion, most of them under 30 years of age. None of them had ever known a world
without electrons or without the periodic table of the atoms. On the one hand
their elders saw determinism and causality crumbling: "God does not throw
dice!" Einstein was to complain [1.42]. But determinism was just the scientific
substitute for God's prescience [1.43] and the elders forgot that long before, in
the very foundations of statistical mechanics and thermodynamics, such as had
been given by J. Willard Gibbs and Einstein himself, determinism had already
been washed out to sea. The alarm about causality was equally unwarranted, for
it has retained the same status in quantum theory as in classical theory in spite
of all the assults upon it. On the other hand, the young workers only saw in the
new theory a chance finally to understand why Bohr's theory, why the classical
theory worked when they did-and also why they failed when they did. They saw
in quantum theory a greater framework with which to build the universe, no less
than in the theory of relativity; and being devoid of severe metaphy~ical bias,
they did not interpret this as philosophical retrogression. There is a psychological
truth in this which had not escaped the perspicacious Henry Adams: "Truly the
animal that is to be trained to unity must be caught young. Unity is vision; it
must have been part of the process of learning to see [1.21] (italics ours).

1.8 Modern Foundations

At the same time as the theoretical foundations were becoming firmer, the experi-
mental puzzles concerning magnetism were becoming numerous. Let us recount
these with utmost brevity. Foremost was the question, why is not iron spontane-
ously ferromagnetic? Weiss proposed that his molecular field had various direc-
tions in various elementary crystals forming a solid; thus the magnetic circuits
are all closed, minimizing the magneto static energy. Spectacular evidence was
provided by Barkhausen in 1919, who by means of newly developed electronic
amplifiers, heard distinct clicks as an applied field aligned the various Weiss
domains. This irreversible behavior also explained hysteresis phenomena. Mea-
surements of the gyromagnetic ratio gave g ~ 2 for most ferromagnetic sub-
stances, showing that unlike in atomic or molecular magnetism only the spins
participated in the magnetic properties of the solids. In atoms, the mounting
 1.8 Modern Foundations 27

spectroscopic data permitted Stoner to assign the correct number of equivalent

electrons to each atomic shell and Hund to enunciate his rules concerning the
spontaneous magnetic moment of a free atom or ion.
In the study of metals it was found that alloying magnetic metals with non-
magnetic ones resulted in a wide spectrum of technical properties. In metals,
unlike insulators, it was also found that the number of magnetic electrons per
atom was not, in general, an integer. In many respects, knowledge of the
magnetic properties of many classes of solids was fairly definitive by 1930. Only
the class of solids which are magnetic, but nonferromagnetic, ordered structures
remained to be discovered, and only two crucial tools of investigation were
lacking: neutron diffraction, and magnetic resonance, each of which has
permitted modern investigators to study solids from within.
The progress of the theory of magnetism in the first third of the twentieth
century can be followed in the proceedings of the Sixth Solvay Conference of
1930, which was entirely devoted to this topic, and in two important books:
Van Vleck's The Theory of Electric and Magnetic Susceptibilities (Oxford, 1932),
and E. C. Stoner's Magnetism and Matter (Methuen, London, 1934). In the first
of these books, but one chapter out of thirteen is devoted to the study of magne-
tism as a cooperative phenomenon. In the other, barely more emphasis is given
to this field of study; after all, Heisenberg had written "it seems that till now,
Weiss' theory is a sufficient basis, even for the deduction of second-order effects,"
in his contribution to the Solvay conference. Nevertheless, some attempts at
understanding magnetism as a collective phenomenon which had already been
made by this time were to lay the groundwork for our present understanding
of the subject.
First came the Lenz-Ising formulation [1.44] of the problem offerromagne-
tism; spins were disposed at regular intervals along the length of a one-dimen-
sional chain. Each spin was allowed to take on the values ± 1, in accordance
with the laws of Goundsmit and Uhlenbeck. This model could be solved exactly
(vide-infra), and as long as each spin interacted with only a finite number of
neighbors, the Curie temperature could be shown to vanish identically. Did this
signify that forces of indefinitely large range were required to explain ferromag-
netism?
The introduction by Pauli of the spin matrices Sj established spin as a
vector quantity; the requirements that the interaction between two spins be an
isotropic scalar, and arguments based on the theory of permutations, led Dirac
in 1929 to the explicit formulation of the "exchange operator":

( total interaction energy == - ~. Jij Sj . SJ) (1.30)

',J

as the essential ingredient in the magnetic interaction. The coefficient J of this

operator was a result of the electrostatic force between the electrons, a force so
large that the magnetic dipole - dipole interactions (the Amperian forces) could
28 1. History of Magnetism

even be neglected to a first approximation. The theory oj Weber and Ewing

was now rendered obsolete. Quoting Stoner:
'" As soon as the elementary magnets are interpreted as corresponding to elec-
trons in orbits, or as electron spins, it is found that the magnetic forces be-
tween neighbouring molecules are far too small to give rise to constraints of
the magnitude required in the Ewing treatment of ferromagnetism. Atoms or
molecules may in certain cases have some of the characteristics of small bar
magnets-in possessing a magnetic moment; but an interpretation of the pro-
perties of a bar of iron as consisting of an aggregate of atomic bar magnets ceases
to be of value, whatever its superficial success, once it is known that the analogy
between atoms and bar magnets breaks down just at those points which are
essential to the interpretation [l.45].

With (1.30) or its equivalent as the basic ingredient in the magnetic Hamil-
tonian, Bloch [1.46] and Slater [1.47] discovered that spin waves were the elemen-
tary excitations. Assigning to them Bose-Einstein statistics, Bloch showed that
an indefinitely large number of them would be thermally excited at any finite
positive temperature, no matter how small, in one or two dimensions; but that a
three-dimensional ferromagnet possessed a finite Curie temperature. For the
magnetization in three dimensions, Bloch derived his "three-halves' power law"

(1.31)

with Tc calculable from the basic interaction parameters. Now, at last, Ising's
result, Tc = 0, could be understood to have resulted primarily from the one-
dimensionality of his array and not at all from the old-quantum-theoretical
formulation of the spins.
The properties of metallic conduction electrons were fairly well understood by
the 1930's. In 1926 Pauli had calculated the spin paramagnetism of conduction
electrons obeying Fermi statistics; and not much later Landau obtained their
motional diamagnetism; one of the primary differences between classical and
quantum-mechanical charged particles arose in this violation of the Bohr-van
Leeuwen theorem. Bloch then considered the Coulomb repulsion among the
carriers in a very dilute gas of conduction electrons in a monovalent metal. In
the Hartree-Fock approximation, he found that for sufficiently low concentra-
tions (0r sufficiently large effective mass, we would add today) this approxima-
tion gave lowest energy to the ferromagnetic configuration, in even greater con-
trast to classical theory. However, Pauli criticized the calculation:
Under those conditions (of low concentration, etc.), in fact, the approxima-
tion used by Bloch is rather bad; one must, however, consider as proved his
more general result, which is, that ferromagnetism is possible under circum-
 1.8 Modern Foundations 29

stances very different from those in which the Heitler-London method is applic-
able; and that it is not sufficient, in general, to consider merely the signs of the
exchange integrals. [1.48].
More modern work has fully borne out Pauli's skepticism; for example, it is
now widely believed that the charged electrons of a metal behave, in most res-
pects, as ideal, noninteracting fermions. Nevertheless, many detailed studies are
still anchored on this frail Hartree-Fock ferromagnetism due to the natural
reluctance of physicists to abandon the only model of ferromagnetism other
than the Heitler-London theory, which could be characterized as truly simple.
A modern many-body theory took another 20 years to develop (see Chap. 6).
In 1932, Nee! [1.49] put forward the idea of antiferromagnetism to explain
the temperature independent paramagnetic susceptibility of such metals as Cr
and Mn, too large to be explained by Pauli's theory. He proposed the idea of two
compensating sublattices undergoing negative exchange interactions, resulting
in (1.20) with a negative Curie temperature-now known as the Neel tempera-
ture.
In 1936, Slater and Wannier both found themselves at Princeton, one at the
Institute for Advanced Studies, and the other at the University. This overlap
must have had positive results, for in an issue of the Physical Review of the
succeeding year there are two consecutive papers of some importance to the
theory of magnetic solids. In the first of these, Wannier [1.50] introduced the set
of orthogonal functions bearing his name, of which we shall have much more to
say. In the second of these, Slater [1.51] gave a nontrivial theory offerromagne-
tism of metals partly based on the use of Wannier functions. He discussed the
case of a half-filled band which, while least favorable to ferromagnetism, is
amenable to analysis (one electron + one hole is exactly soluble):
Starting with the theory of energy bands we have set up the perturbation
problem and solved approximately the case of a band containing half enough
electrons to fill it, all having parallel spins but one. This problem is a test for
ferromagnetism: if the lowest energy of the problem is lower than the energy
when all have parallel spins, the system will tend to reduce its spin, and will not
be ferromagnetic; whereas if all energies of the problem are higher ... we shall
have ferromagnetism. [1.51].
This followed hard upon his band theory of the ferromagnetism of nickel [1.52].
The study of nickel-copper alloys was strong evidence for the band theory of
ferromagnetism in metals. A single copper atom dissolved in nickel does not
succeed in binding the extra electron which it brings into the metal by virtue of
its higher valency. This electron finds its way into the lowest unfilled band, which
belongs to the minority spins; and therefore decreases the magnetization of the
entire crystal in the amount of one Bohr magneton, homogeneously distributed.
As pure nickel possesses 0.6 Bohr magnetons per atom, the magnetization
30 1. History of Magnetism

should decrease linearly with copper concentration, extrapolating to zero at 60

percent concentration. The experiments accorded beautifully with this hypothesis
[1.53a].
Continuing study of Weiss domains soon indicated that they did not neces-
sarily coincide with the physical crystals, that their size was determined by such
effects as magnetostriction, quantum-mechanical exchange energy, etc., as well
as the magneto static energy. The first complete theory (for Co) was sketched by
Heisenberg in 1931, and subsequent work by Bloch, Landau, Bozorth, Becker,
and many others have laid the experimental and theoretical framework of a
theory of domains. Recent formulations by Brown and collaborators, Landau
and Ginzburg, and others, have produced the interesting differential equations
that regulate the slow distortions in space and time associated with domains,
spinwaves, and the like. This field [1.53b] has even acquired a name: microma-
gnetics, and spawned a new industry: magnetic "bubbles". The solitons discuss-
ed in Chap. 5 are examples of domains. A practical application follows below.

1.9 Magnetic Bubbles

In the 1960's a number of experimental and theoretical studies of strip- and
cyclindrical domains [1.54] were followed in 1967 by Bobeck's suggestion of the
device potential of the cylindrical domains [1.55]. This invention by Bobeck,
Shockley, Sherwood and Gianola has become known as magnetic bubbles; they
are ubiquitous today, with applications in telephone switching, computer me-
mory banks-that is wherever a high density, low energy rapidly switchable
information storage is required. In view of their great intrinsic interest, and as
a simple demonstration of micro magnetics, we now outline some of the basic
facts and theory and refer the reader to the recent monograph by Eschen/elder
[1.56] for supplementary information.
Magnetic bubbles are small, mobile cylindrical domains in thin films magne-
tized perpendicular to the film surface. Bubble memory chips with capacities of
106 bits and data rates of the order ofmegabits/s are now technological common-
place, with other applications in the offing. Yet the field is so recent that two
excellent, modern books on thin magnetic films [1.57] have no mention of this
phenomenon at all! Magnetic thin films are themselves a recent development,
dating to Blois' fabrication of 1000 A thick permalloy films (80% Ni, 20% Fe),
chosen for the low value of magnetocrystalline anisotropy [1.58]. Magnetostatic
considerations favor the magnetization in the plane of the film, and the early
literature [1.57] discussed mainly applications of this type. But already by 1960
there were photographs of bubble domains [1.59] and attempts at a theory [1.54].
By 1967 Bobeck and colleagues [1.55] had invented a bubble device (capable of
generating, storing and counting the units) and shortly thereafter a complete
theory was developed by Thiele [1.60]. Such a theory must predict the range of
parameters (film parameters and thickness, external field parameter) over which
bubbles are stable, and take into account the factors affecting their motion when
 1.9 Magnetic Bubbles 31

subjected to magnetic field gradients. So first, what are the materials and para-
meters?
Magnetic rare earth garnets-orthoferrites such as TmFe03 or DyFe03-
oriented with c axis perpendicular to the film have very small barriers (as small
0.01 Oe) against the motion of a domain wall and have been found eminently suit-
able. Other materials, including amorphous magnetic metal alloys, have been
used.
Figure. 1.2 shows a demagnetized sample, with contrasting shades indicating
domains of opposite magnetization (into or out of the film). In such a configura-
tion, strip domains are most stable, the sole bubble being the exception. An iso-
lated bubble is always unstable and will grow indefinitely unless an external mag-
netic field is present, in such a direction as to collapse the bubble. In Fig. 1.2,
the bubble is stabilized by the demagnetizing field of the neighboring domains.
An external field applied to the demagnetized sample causes the domains in the
direction of the field to grow and the others to shrink until they assume the cir-

<l Fig. 1.2. Magnetic domains in a thin platelet of ortho-

ferrite TmFe0 3 in the absence of any magnetic field. A
lone "bubble" domain is seen-the strip domains are in
stable equilibrium. Any resemblance to Fig. 1.1 is purely
coincidental

"V Fig. 1.3. The left-hand sides of (a) and (b) are as-
grown 6J.1m (YGdTm)3 (GaFe), 0 12 garnet film, the right-
hand sides have been treated by superficial (!J.1m)
hydrogen ion implantation to improve the quality of the
domains. (a) is in zero bias-note the total absence of
bubble domains; (b) is in 170 Oe applied magnetic field,
sufficient to produce only bubble domains. [1.55]

. :-..... . . .
.0
" .......
0,.:-
0
.
-.: . . -:
•• : : : •• : • • •
. I I/ .. \..
• 0
...
.. . . ..•:. •• "

• <p • • I
. .• •: •:.-.
• • •
. ••
O ... f)
• I · · ••
. .•••
· . . .. . ..
· ......
e.
•
•

.. .. .
t!J • • • • • • • • • • •

O """ . . . . ..
.·.. ..-.......
o
•
32 1. History of Magnetism

cular shape denoted as bubbles. Increasing the field causes the bubbles to shrink
until at a second critical field they disappear totally. This is illustrated in Fig. 1.3
[(a) is no bias, (b) is for 170 Oe external field)], which also shows the importance
of the proper choice of materials. The material is 6-Jlm (YGdTm)3 (GaFe)s012
garnet film, untreated in the left-hand side of the pictures and after hydrogen ion
implantation on the right-hand sides.
Thiele's theory was subsequently nicely simplified by Callen and Josephs
[1.61] and it is this version we now present starting with the conditions for static
equilibrium.
We consider a film of magnetic material of thickness h 0.1 mm or less. The
bubble radius r will turn out to be of approximately the same magnitude as h
and both are related to a characteristic length I == u w/41tM2. The magnetization
has the magnitude IM I everywhere, pointing in the + z direction (normal to the
film) within the domain, and - z direction elsewhere. An applied field H
stabilizes the domain.
The total energy ET = Ew + EH - ED comprises the following: the wall
energy,

(1.32)

in which the important parameter is uw , the wall energy per unit surface; the
interaction with the external field

(1.33)

and the demagnetizing energy of a single cylindrical domain in an infinite plate

ED, given by Thiele [1.60] in terms of elliptic functions, for which Callen and
Josephs found an excellent-and simple-approximation. Before proceeding
with the calculation, it is convenient to express all the quantities in dimension-
less form: a dimensionless energy 'iff == E/(161t 2M 2h3 ), a dimensionless field
~ = H/41tM a dimensionless radius x = r/h, and a dimensionless characteristic
length A = //h = uw /41tM 2h.
Then, d'iffT/dx = 0, the condition for bubble stability, yields

~ A + ~x - x/( 1 + ~ x) = 0 (1.34)

where the Callen-Josephs approximation to Thiele's ED yields the simple term,

-x/(l + 3x/2). The equation is solved graphically as shown in Fig. 104. with
the regions of stability being indicated, as derived from the sign of d 2'iffT/dx2.
Above a critical field ~o = 1 + 3AJ4 - (3A)1/2 no bubble domain is stable (the
straight line is tangentto the curve). It is similarly possible to calculate a criterion
for the transformation of bubbles into strip domains [1.60].
In bubble dynamics, one assumes that the rate of change of the energy in
the bubble is balanced by drag (the friction force is proportional to the velocity
 1.10 Ultimate Thin Films 33
Fig. 1.4. Graphical solution of (1.34).
Tilt of straight-line segment depends
on external field. Stable and unstable
solutions are shown, with the direc-
tion of bubble growth or shrinkage
shown by arrows. [1.60]

4
x

so the frictional loss is proportional to the square of x) and by a second mecha-

nism dependent only on the distance traversed per unit time (i.e., proportional
to Ix I). Writing the energy balance per unit wall:

(1.35)

~ is the coercive field for wall motion, and rJ is the wall mobility parameter (in
our units of h/4rcM). The equation of bubble dynamics is then

x = -rJ[ (2~) - (1 + ; xfl + jl" + jl"c sgn(x)] (1.36)

determining the rates at which the bubbles can be grown or shrunk.

A study of the interaction energy between two or more bubble domains does
not seem to have been carried out at this time, perhaps because here cooperative
phenomena-the correlated behavior of several domains-are not important.

1.10 Ultimate Thin Films

The search for the ultimate thin film led Pomerantz and collaborators to the
fabrication of literally two-dimensional ferromagnets [1.62]. Two types are
shown in Fig. 1.5; in either case, the magnetic stratum consists of Mn2+ ions.
Despite this tour deforce, two-dimensional systems stand in peculiar relation
to the rest of physics. Difficult of resolution, they seem capable of exhibiting
34 1. History of Magnetism

Fig. 1.5. (a) Schematic diagram of the

deposition of a literally 2-D magnet by
the Langmuir-Blodgett technique. The
hydrophilic substrate (e.g., quartz) is
pulled through the water surface. A
monolayer of manganese stearate that
was floating on the water adheres to the
quartz to produce a "Structure I" film.
The inset shows the structure of a stea-
rate anion, and a Mn2+ ion. (b) Schema-
(0) tic cross section of a 2-D magnetic of
structure II. The Mn ions are shown in
a plane, but the actual structure may be
nonplanar. [1.62]

some of the properties of the "real" three-dimensional world, but not all. In any
event, they have been of some interest for many years now. In 1942, Onsager
developed an exact solution to the problem of Ising spins in a plane, the "two-
dimensional Ising model" [1.63]. This work stands, to this day, as a pinnacle of
the achievements of theoretical physics of our time. Onsager's solution yielded
the thermodynamic properties of the interacting system, and demonstrated the
phase transition at Te but in a form quite unlike that of Curie-Weiss. In parti-
cular, the infinite specific-heat anomaly at Te is a challenge for approximate,
simpler theories to reproduce. Onsager's discovery was not without an amusing
sequel. The original solution was given by Onsager as a discussion remark,
following a paper presented to the New York Academy of Science in 1942 by
Gregory Wannier, but the paper, based on an application of Lie algebras, only
appeared two years later. [1.64]. However, his formula for the spontaneous
magnetization below T e , which requires substantial additional analysis,

.A' = (l - X- Z)1/8 X = sinh 2JI sinh 2Jz (1.37)

, kT kT

was never published by him, but merely "disclosed". It

... required four years for its decipherment. It was first exposed to the public on 23 August
1948 on a blackboard at Cornell University on the occasion of a conference on phase transi-
tions. Laslo Tisza had just presented a paper on The General Theory of Phase Transitions.
 1.11 Dilute Magnetic Alloys 35

Gregory Wannier opened the discussion with a question concerning the compatibility of the
theory with some properties of the Ising model. Onsager continued this discussion and then
remarked that-incidentally, the formula for the spontaneous magnetization of the two-
dimensional model isjust that (given above.) To tease a wider audience, the formula was again
exhibited during the discussion which followed a paper by Rushbrooke at the first postwar
IUPAP (International Union of Pure and Applied Physics) statistical mechanics meeting in
Florence in 1948; it finally appeared in print as a discussion remark ... However, Onsager
never published his derivation. The puzzle was finally solved by CN. Yang and its solution
published in 1952. Yang's analysis is very complicated ... [1.65].

Of course the Ising model is peculiar in that it does not accommodate spin-
waves. Bloch had already established that in two dimensions, as well as in one,
the number of spin waves excited at finite temperature is sufficient to destroy the
long range order. In 1966, Mermin and Wagner [1.66] succeeded in proving rigo-
rously the absence of long range order in both the Heisenberg ferromagnet and
antiferromagnet in one and two dimensions, i,e., .A = 0 at all temperatures.
This by itself is, however, not sufficient to rule out a phase transition at finite
temperature Te! This interesting possibility came out as a result of numerical
work by Stanley and Kaplan [1.67], later confirmed by a remarkable theory of
the xy model due to Kosterlitz and Thouless [1.68 a], in which vortices - a to-
pological disorder - were seen to be an important ingredient, an adjunct to the
spin-waves which are the more common excitations of a magnetic substance.
The two-dimensional phase transition which they obtained was quite different
from the usual. Indeed, an entire field of physics has now developed around
the properties of low-dimensional systems, principally two- and one-dimen-
sional (fractional dimensionality being also admitted), as discussed in [1.68 b].

1.11 Dilute Magnetic Alloys

At first sight, no topic could appear less intellectually challenging than that of
dilute magnetic alloys. Yet some of the great recent conceptual advances of
of theoretical physics have had their inception in this seemingly modest subject!
The original studies by Friedel [1.69] centered about the question of forma-
tion of a local magnetic moment-how do iron or manganese maintain their
atomic magnetic moments when imbedded in a nonmagnetic host matrix such
as copper metal? Anderson then suggested a model [1.70] with the intention of
resolving or closing this question in a semiquantitative way. Far from it!
Anderson's model, which was not rigorously soluble either, opened an entire
new field of investigation, seeking to explain or predict the occurrence and
non-occurrence of local moments, as in the variety of examples in Fig. 1.6 and
Table 1.l.
If the impurity atom does possess a magnetic moment this polarizes the con-
duction electrons in its vicinity by means of the exchange interaction and there-
by influences the spin orientation of a second magnetic atom at some distance.
36 1. History of Magnetism

CXl14 Fig. 1.6. Magnetic mo-

~ ment on Fe impurity ion as
::L
III 12 function of electron number
c
0 of the host. (From [1.79])
~01 10
0
::2:
.c0 8
III
.~ 6
E
0
::i 4
~
0-
2
C
CIJ
E 0
0
::2: 3
Y
Electron Concentration

Table 1.1. Which "magnetic" atoms retain their moments in dilute alloys (adapted from [1.78])

Magnetic Host Metal

Species Au Cu Ag Al

Ti no ? ? no
V ? ? ? no
Cr yes yes yes no
Mn yes yes yes no
Fe yes yes ? no
Co ? ? ? no
Ni no no ? no

Owing to quantum oscillations in the conduction electrons' spin polarization

the resulting effective interaction between two magnetic impurities at some
distance apart can be ferromagnetic (tending to align their spins) or antiferroma-
gnetic (tending to align them in opposite directions). Thus a given magnetic
impurity is subject to a variety of ferromagnetic and antiferromagnetic interac-
tions with the various neighboring impurities. What is the state oflowest energy
of such a system? This is the topic of a new very active field of studies entitled
"spin glasses" the magnetic analog to an amorphous solid, which we touch upon
in a later chapter, to expand on it in the next Volume.
When the impurity atoms are very few, in the very dilute magnetic alloys,
each impurity atom correlates with just the conduction electrons in its own vici-
nity. This has a surprising consequence: for many decades now, an anomalous
resistivity of high-purity gold and copper samples has been remarked, in which
the resistance increased with decreasing temperature. This contradicts the usual
 1.12 New Directions 37

behavior, for generally thermal disorder decreases as the temperature is lowered.

These anomalies were finally found to be caused by the trace impurities of iron
deposited by contact with the steel used in the fabrication.
The resistance mystery was ultimately pierced by J. Kondo in the early 1960's,
with his explanation of what is now called, the "Kondo effect" [1.71]. It occurs
only when the magnetic species is anti ferromagnetically coupled with the con-
duction electrons, causing a scattering resonance at the Fermi level, which
becomes the sharper the lower the temperature. A completely satisfactory theory
of this many-body effect is not yet known, but recent developments on the
ground state and thermodynamic properties have led some to believe that an
explanation of the transport anomalies is close to hand. First, the thermodyna-
mics has been obtained by analytical, numerical and renormalization-group
type methods [1.72]. The symmetry of the ground state-in many cases, a
nondescript singlet-is known [1.73]. The thermodynamic and transport pro-
perties have been proved, rigorously, to be analytic functions of the temperature
by Hepp [1.74], despite indications, from the initial calculations of Kondo and
his followers, that logarithmic singularities of the form log (TjTK ) were the
signature of this phenomenon down to the lowest temperatures.
Finally, a breakthrough has appeared. Within months or weeks of each other,
in 1980 Andrei [1.75] in the United States and Wiegmann [1.76] of the Soviet
Union have independently found almost identical solutions to the Kondo
Hamiltonian, which they believe to be exact. These solution are broadly based
on a method first proposed by H. Bethe in 1932 to solve the one-dimensional
Heisenberg anti ferromagnet, now known as Bethe's ansatz. This method is
thoroughly detailed in Chap. 5 of this book, but aside from a brief introduction
to the Kondo phenomena in Chap. 6, we defer the detailed study of this problem
until Vol. 2, devoted to finite temperature effects.

1.12 New Directions

To quote Dryden, on the Canterbury Pilgrims: " ... But enough of this; thete
is such a variety of game springing up before me, that I am distracted in my
choice and know not which to follow ... "
The world of magnetism has been expanding. In biological systems, the
minute presence of magnetite-the loadstone-has been found to explain such
remarkable properties as the orientational ability of homing pigeons. Indeed,
the production and detection of minute magnetic fields has become an industry,
thanks to a remarkable development in electrical science, viz., superconductivity.
The theory of superconductivity, jointly invented by J. Bardeen, J. R. Schrieffer,
and L. N. Cooper in 1957 explained how, and why it might be possible to produce
infinitesimal magnetic fields in amperian current loops, and devices acronymed
SQUIDS are now available for this purpose. This many-body theory also
introduced the use of field-theoretic methods into condensed-matter physics.
38 1. History of Magnetism

Nowadays, we find the traffic of information is sometimes reversed; quantum

statistical methods specifically designed for the solution of magnetic problems
have become the prototypes even in field theory and particle physics [1.77].
Recent Nobel prizes have honored the above-named theorists of supercon-
ductivity, also B. Josephson for his theory of pair-tunneling into and out of
superconductors, Louis Neel, P. W. Anderson, and J. H. Van Vleck for their
fundamental contributions to the understanding of magnetism, and Lars
Onsager for his brilliant solution of the two-dimensional Ising model. We hope
to convey the excitement of this work in the pages that follow, and follow up in
[1.68b].
2. Exchange

With the simultaneity which has characterized so many of the greatinventions of

modern science, Dirac [2.1] and Heisenberg [2.2] independently dicovered ex-
change. This effect appeared at first to be mysterious, for its origins in the
Pauli exclusion principle of quantum mechanics had no classical analog. But
it must be remembered that classical mechanics had failed to provide any expla-
nation of magnetism, and now one could see why. Ferromagnetism cannot exist
in the classical correspondence limit 11- 0; it is one of the results of quantum
mechanics, a manifestation of "exchange."
Loosely speaking, exchange was explained as follows. The Pauli exclusion
principle keeps electrons with parallel spins apart, and so reduces their Coulomb
repulsion. The difference in energy between the parallel spin configuration and
the antiparallel one is the exchange energy. This, however, is favorable to
ferromagnetism only in exceptional circumstances because the increase in
kinetic energy associated with parallel spins outweighs the favorable decrease
in potential energy-as is the case in the familiar examples of the helium atom
or the hydrogen molecule. In rare cases, such as with metallic iron, a large num-
ber of parallel spins produces ferromagnetism because the cost in kinetic energy
is not as great as in some other metals, whereas the gain in potential energy is
significant. Thus, the forces which are involved are electrostatic Coulomb forces
and not the far weaker Ampere current, or magnetic dipole, forces. These elec-
trostatic forces regulate the spin configurations in atoms, molecules, and solids
via the Pauli principle, so that in constructing the foundations of the theory of
magnetism, one could to a very good approximation completely ignore the magnetic
fields proper. It is fortunate that cause and effect could be so neatly disen-
tangled.
It is necessary to anticipate formally some of the material in the next chapter
by introducing three common exchange operators, named, respectively, for
Majorana, Bartlett, and Heisenberg. 1

I See, e.g. [2.3]. Note that the operator gm was invented by Dirac [2.4].
40 2. Exchange

where '10 '2 are the spatial and ~Io ~2 the spin coordinates of two electrons.
Clearly,

and later we shall see that on wavefunctions allowable to electrons, ,9H always
has eigenvalue == -I, so that ,9M and ,9B are effectively inverses of one another.
Let us construct the simpler one, ,9B, out of the spin one-half operators at our
disposal. It must be invariant under the inconsequential rotations in spin space
+
and have eigenvalue I in triplet states and - I in singlet states. These require-
ments are satisfied uniquely by the following function of spin operators S, and S2 :

(2.1)

Assuming there is an energy -J'2 associated with the exchange, we obtain

an effective Hamiltonian ex; - J 12 S, • S2 (omitting the uninteresting constant
part). It is also interesting to remember that just such an effective Hamiltonian
had been postulated in the "vector model" of the atom, before the invention of
wave mechanics, and that it is in effect a sort of semiempirical rule. We shall later
see how it correlated with Hund's rules of atomic structure. In the present
chapter we explore molecular structure and the role of exchange, its limitations
and its successes in laying the groundwork of a theoretical understanding of
magnetism.
The work of Dirac, Heisenberg, and others, and particularly the book of Van
Vleck first focused attention on an elementary interaction of the type in (2.1)
as the fundamental object of study of the theory of magnetism. It may be sup-
posed that this is the reason the Heisenberg exchange Hamiltonian [also called
the vector model, or the Heisenberg-Dirac-Van Vleck (HDVV) Hamiltonian]
(2.2)

occupied so many theorists over the span of 50 years, calculating energy levels
and eigenfunctions and the statistical mechanics which flowed therefrom. No
less effort went into calculating the magnitudes of the exchange constants J I }
from atomistic considerations.
In spite of all this effort, and perhaps because of what it revealed, it was not
possible to elevate exchange to the rank of a universal principle, to make of it a
force of nature such as Coulomb's law or Newton's laws. For example, even
though the explanation of Hund's rules by the vector model was one of the
earliest successes of the theory, it turned out that the idealized Hamiltonian of
(2.2) was only semi quantitatively accurate in complicated atoms and that a
more exact Hamiltonian had to be solved in a higher approximation than the one
in which exchange is well defined. And in general, notions of electronic correla-
tions and other effects not directly related to the Pauli principle came to cloud
the naive picture.
 2.1 Exchange Equals Overlap 41

Even today, many difficulties, both in principle and in practice, still plague
those who would derive (2.2) from first principles. This will be made evident in
the present chapter as well as in succeeding ones. As Henry Wadsworth Long-
fellow once wrote, "So nature deals with us and takes away our playthings one
by one."

2.1 Exchange Equals Overlap

In order for exchange to occur, the paths of the interacting electrons must in-
evitably and inextricably overlap. Let us prove this, with the aid of a model in
which electrons are confined to nonoverlapping regions [2.6]. Consider N elec-
trons, with arbitrary interactions, and divide coordinate space into N separate
boxes such that electron number I is confined to box number I, electron number
2 to the second box, etc. The wavefunction is thus subject to the boundary condi-
tion that it vanish whenever a particle reaches the surface of the volume assigned
to it. We illustrate a reasonable choice of boundary conditions in the case of N
electrons + N fixed protons in Fig. 2.1 . Subject to the boundary condition we
have specified , the spatial eigenfunctions obey the Schr6dinger equation

(2.3)

in which the Hamiltonian is generally invariant under any permutation of the

coordinates of the indistinguishable particles, but is otherwise quite arbitrary.
If the electrons were truly noninteracting, an arbitrary solution of(2.3) could
be written as a product function,

(2.4)

..•. '\~. ."

Fig. 2.1. Crosses indicate protons ;

shading, electrons. The wavefunction
is constrained to vanish on the poten-
tial barriers indicated by solid lines,
i.e., the nodal surfaces are imposed
42 2. Exchange

the Hartree product wavefunction. In effect, j;(r,) would be the one-particle

wavefunction of the jth electron, when constrained to the jth box. But before
discussing the product wavefunction, we must stress that it is not an eigen-
function of Sf: For even though the particles are constrained to remain in
separate compartments, they do interact, and their correlated motion cannot
in general be factored. Therefore if we do not wish to sacrifice generality, we
must write the true eigenfunction rp in the most general way as,

rp = rp(I/2/ ... / N) (2.5)

indicating by this notation that particle number 1 is in box number 1, etc. Be-
cause of the symmetry of the Hamiltonian under permutations, we can transpose
particles 1 and 2, and the wavefunction

will also be an eigenfunction of the Hamiltonian, (2.3), as indeed are any of the
N! permutations

p = 1,2, ... ,N! (2.6)

of the N electrons amongst themselves, in the original wavefunction of (2.5).

It is trivial to check that all these rpp functions are mutually orthogonal, as
particles are constrained to be in separate boxes. Every energy level of the
many-body Hamiltonian is therefore intrinsically N!-fold degenerate for the
boundary conditions specified. But all N! are not admissible: for example, the
totally symmetric linear combination of functions rpp'

while it might serve in the theory of a system of similar Bosons, is never suitable
for N ~ 3 Fermi-Dirac particles. This is but one of the consequences of the
Pauli exclusion principle.
Rules for constructing totally antisymmetric wavefunctions' of space and
spin are given in Chap. 4. What is most important in the present analysis, is that
there are two spin degrees of freedom for each electron, therefore 2N orthonor-
mal spin functions which may be combined with the N! space functions, to give
precisely 2N wavefunctions of space and spin obeying the Pauli principle, for every
eigenvalue E. If N is even (odd) the resultant total spin ranges in magnitUde from
Stot= OCD to a maximum of tN. There is no interference befween spatial motion
and spin degrees of freedom.
Therefore, even in an applied external magnetic field H the free energies of
the electronic and spin degrees of freedom are additive: F = Fel + Fspo In
 2.1 Exchange Equals Overlap 43

particular the spin contribution is

F,p = -kTN In [2COSh(gi~:)] (2.7)

identically that of the noninteracting spins studied in a later chapter. Fel contains
a diamagnetic contribution, Fsp is paramagnetic, but neither is ferromagnetic.
Thus, spontaneous alignment of magnetic moments in the absence of an applied
field is impossible in the present context.
When the fictitious potential barrier separating the electrons is removed,
the 2N -fold degeneracy of each eigenvalue E is lifted, as depicted in Fig. 2.2 where
Eo is the ground state, E1 the first excited state, etc. The manner in which Eo fans
out is perhaps the most crucial: if the exchange Hamiltonian is capable of
reproducing the true spectrum of Eo, we shall have achieved a significant simpli-
fication of the problem. (Even if there is considerable overlap between the
spectrum of Eo and that of the higher EJ , the wavefunctions may be disting-
uished, and classified according to whether they represent magnetic degrees of
freedom or electronic degrees of freedom so a whole set of 2N low-lying states
can conceivably be described by the Heisenberg Hamiltonian, i.e., by an
"effective" interaction among the spins.)

Fig. 2.2. Fanning out of 2N -fold degenerate levels, when

potential barriers are removed. Note that levels may cross

In the chapter on the theory of magnetism in metals, we shall examine at least

two instances where this simplification occurs. When electrons are allowed to
co-mingle, and their interactions are treated in low order of perturbation theory,
it is found that the manner in which the degeneracy of Eo is lifted is precisely
described by the Heisenberg Hamiltonian. This is the case both in certain
insulators, and in those metals regulated by the indirect exchange mechanism.
44 2. Exchange

However, we are anticipating too far ahead, and it is important to develop the
plot gradually.
Suffice it to say that there is in general no guarantee that the levels will fan
out in the desired manner, and in many substances they do not do so and there-
fore defy simple-minded descriptions. Hopefully, a detailed analysis based on
physically admissible approximations to the Schrodinger equation will give us
some sharper criteria. It is to this task that we turn our attention in this and
subsequent chapters.

2.2 Hydrogen Molecule

We study molecules first to appreciate some of the difficulties which are absent
in our subsequent study of the atom. But these are by now almost classical
subjects on which many books have been written. The actual treatment will
therefore be addressed to the students who are yet unacquainted with, or have
forgotten atomic and molecular physics ("quantum chemistry", as it is now
known). The more experienced reader can omit all but a few remarks towards the
end of the chapter concerning subjects with which he might be less familiar. The
method with which we break ground was invented by Heitler and London 2 shortly
after the discoveries ofSchrodinger, Heisenberg [2.2] and Dirac [2.1] gave impetus
to the quantitative study of the many-body problem in quantum theory. It is still
the simplest approach to the problems of molecular binding and interatomic
exchange.
Assume nuclei fixed at Ra and R b , with interatomic spacing Rab = several
atomic radii. (We set the mass of the proton = 00 so as to be able to localize it.
Taking the finite mass into account does not change the results much, but
leads to an interesting exchange effect: the existence of two species of H 2 , ortho-
and parahydrogen. The energy splitting is related to the overlap between protons
rotating about a common axis, the overlap due to vibrational motion being ne-
gligible. So when the molecule is hindered in its rotation, e.g., by a crystal field,
the splitting between ortho- and parahydrogen must disappear.) To a first a
approximation each neutral atom exerts no force on the other, and each electron
"sees" only the central force field of its own proton. According to this hypo-
thesis, we simplify the two-particle Schrodinger equation,

fft' 'PI = E'PI

with

(2.8)

2 See [2.7] or any good text on molecular physics or chemistry.

 2.2 Hydrogen Molecule 45

by choosing 'I'1(r" r 2) to be a product function 'Pa(r l )rh(r2), where each factor

obeys the one-particle Schrodinger equation, (there should be no confusion be-
tween the ground state eigenvalue eo and the charge of the electron)

and

(2.9)

At the total Hamiltonian is invariant under the interchange of the two coor-
dinates rl and r 2, an equally good choice must be 'Pn = 'Pa(r2)'Pb(rl). Therefore,
let us diagonalize the Hamiltonian of (2.8) within the subspace ofthese two sim-
ple functions. Assume the atomic orbitals 'P(r) to be normalized, and define
various overlap integrals I, V, U as follows:

(2.10)

Let us take a variational wavefunction

(2.11)

and determine the coefficients CI and Cll so as to make the variational energy
stationary

f d3rl d3r2'P*~'P aE 0
f d3rl d3r2 'P* 'P ' aCI,ll -
var _
(2.12)
Evar = •

The solutions to this are best expressed in matrix notation. Let 'PI correspond to
the vector (1,0) and 'Pn to the vector (0, 1). In this notation, the variational
wavefunction 'P is merely (CI, Cll), and the variational equations can be expressed
in compact matrix form as an eigenvalue problem:

P] [CI ]
I Cll'
(2.13)
46 2. Exchange

As in fact all functions under consideration are (or can be made) real, we shall
henceforth omit the asterisk. It is not difficult to guess that the solutions to this
euqation are the symmetric and antisymmetric functions corresponding to

Cx =± C rr (2.14)

and that the respective eigenvalues are

(2.15)

The space symmetric (+) solution calls for the spin anti symmetric "singlet"
function, and the space anti symmetric function ( - ) for any of the three symme-
tric "spin-triplet" functions. The triple-singlet separation is

VP - U
flE = E_ - E+ = 2 1_ r (2.16)

and can be used to define an effective exchange force in the Heisenberg Hamil-
tonian. For the energy levels of

(2.17)

are -{.J12 in the triplet states, and +V12 in the singlet state. (See Chap. 3). By
comparison with (2.16), the exchange constant is deduced to be

VP - U
J 12 = - 2 1 _ /4 . (2.18)

Ferromagnetism, of an embryonic molecular sort, would occur if the exchange

constant J 12 turned out positive. Antiferromagnetism would be the consequence
of an antiferromagnetic bond J l2 < O. What the actual sign turns out to be
depends on the relative magnitudes of the "Coulomb integral" V, the "overlap
integral" I, and the "exchange integral" U. In the primitive calculation of Heitler
and London unperturbed hydrogen Is orbitals were used in these various in-
tegrals-yielding results in satisfactory agreement both with experiment and
more accurate modern calculations (see Problem 1, below). The exchange con-
stant turned out negative, corresponding to a singlet (i.e. embryonic antifer-
romagnetic) ground state, and varied with internuclear distance Rab as shown
in Fig. 2.3. The triplet state was found, correctly, to be unbound, to have energy
greater than the energy of two separate atoms, 2eo.
A second method for treating the Hz molecule owes its origin to the work of
 2.2 Hydrogen Molecule 47

Fig. 2.3. Exchange parameter 112 as

function of internuclear distance R.b in
3 hydrogen molecule (H 2). Dot shows
equilibrium value

-2
-3

F. Hund and R. S. Mulliken,l and is known as the method of molecular orbitals

(M-O). The one-electron functions are chosen, in this method, to be eigenfunc-
tions of some of the symmetry operations which leave an appropriately defined
zeroth order Hamiltonian invariant. In this case, the eigenstates of the ion Hi
could be chosen as the molecular orbitals. These fall into two classes, the even
ones and odd ones. A further refinement would be to take for the one-
electron functions the solutions of a Hamiltonian in which the ionic potentials
are screened in a self-consistent way. But even the following crude choice gives
reasonable agreement with the exact solution, at the equilibrium value of R ab •
Let

(2.19)

and

(2.l9a)

approximating the even and odd (gerade, ungerade, in the time-honored notation
g, u) eigenfunctions of Ht by linear combinations of the atomic orbitals. This
generates four functions with which to diagonalize the Hamiltonian,

(2.19b)

The M-O method generates four states from two atomic orbitals, twice as
many as either the H-L scheme or the Heisenberg Hamiltonian. Molecular or-
bitals are related to the Bloch functions of solid state theory, discussed elsewhere
in the text. There also exists another set of states, more perspicuous than the
previous, being mutually orthogonal functions that properly reduce to atomic
orbitals in the limit of large separation, 1- O. They are:

3 See, for example, [2.8,9].

48 2. Exchange

'II a = ,y'l + g2 _ 2g1 (2.20)

and

'lib = ,y'l + g2 _ 2g1 (2.20a)

with

g = +(1 - ,y'l - 12)

to ensure orthogonality. (These functions are prototypes of Wannier func-

tions, treated elsewhere in this book). The eigenvalues of the two-body Hamil-
tonian can be found with the aid of the four orthonormal functions,

Fl = IfIirl)'1fI a(r2) F2 = IfI a(rl)'1fIb(r2)

F3 = IfIb(rl ) 'lfIb(r2) F4 = IfIb(r l)'lfIa(r2) (2.20b)

As these functions span the same "function space" as the four M-O functions
(2.19b), the four eigenvalues will be the same as if we worked with molecular
orbitals. F2 and F4 correspond approximately to PI and PH for the nonortho-
gonal orbitals, Fl and F3 to "ionized configurations." The two lowest eigenvalues
will correspond to E± of (2.15); see Problem 2.1.

Problem 2.1. Find the eigenvectors and eigenvalues of the 4 X 4 matrix

In addition to the definitions in (2.10), this requires the definition of additional

integrals Wand X, etc. Define them.
(a) Show that there is one triplet (space anti symmetric) eigenfunction, with
energy E_ = st"z2 - Je"24 = identically as given in (2.15). Find the lowest of the
three space symmetric, singlet, solutions. Prove that it has energy lower than
E+, the H-L ground state.
(b) Second-order perturbation theory can also be used to calculate the singlet
ground state energy,

Express this formula in terms of U, V, W, X, and I, and by comparison with the

exact result found in part (a) determine the physical region of approximate
validity of the perturbation theory.
The importance of this calculation is that it shows that the Heisenberg Hamil-
tonian can be derived qualitatively on the basis of first- and second-order pertur-
 2.3 Three Hydrogen Atoms 49

bation theory. using orthogonalized orbitals; and therefore many of the results
of the theory (exchange constant, spin waves, existence of Curie temperature,
etc.) are valid even when H-L theory is not.

2.3 Three Hydrogen Atoms

The total Hamiltonian is separated into terms appropriate to individual hydro-

gen atoms ~o + the interaction terms ~' :

- '(I
-+-1) - (1'3a
'2a '2c
-+- 1)]
'3b
= {~o} + {~'} (2.21)

and as before, we use products of the nonorthogonal atomic functions of(2.9) of

which there are 3! = 6 in total. Let us label them according to the following
table:

If! 1 = lPa(I)lPb(2)lPi3)
1f!2 = lPi2)lPb(1 )lPc(3)
1f!3 = lPa(3)lPb(2)1P.(1) (2.22)
1f!4 = lPil)lPb(3)lPc(2)
If! s = lPi2)lPb(3)lPc(l)
1f!6 = lPa(3)lPb(I)lPc(2).

How are these related to each other under the permutations of various particles?
For typographical simplicity, let us omit the Greek symbols, and write If!l as 1,
1f!2 as 2, etc. Then we can draw up simple tables (showing into which functions
any of the six transform) under transpositions (permutations of two particles):

1-2,3, or 4
2-1,5, or 6
3-1,5, or 6 (2.23)
4- 1,5, or 6
5 - 2,3, or 4
so 2. Exchange

6- 2,3, or 4

and under nontrivial permutations of all three particles:

1- 5,6
2-3,4
3-2,4 (2.24)
4-2,3
5-1,6
6-1,5

Thus, for equidistant atoms at the vertices of an equilateral triangle (Fig. 2.4)

f ",T"'2 d'r = f ",T"'3 d'r = f ",T"'4 dr: = /2 = f "'~"'5 dr:, etc. (2.25)
and
f ",T "'5 dr: = f ",T"'6 dr: = P= f "'~ "'3 dr:, (2.26)

etc. Or in general if we use these integrals to define an overlap matrix a

a" = f ",t"" dr: (2.27)

then we have simply,

1 /2 J2 J2 P P
J2 1 P P J2 /2
/2 P P J2 J2
0= (2.28)
J2 P P J2 12
P J2 J2 /2 1 P
P J2 J2 /2 J3

a real symmetric, matrix. And as for the matrix elements of the Hamiltonian, we
do not have to examine all 36 possibilities, but in fact just 6:

f ",TfJe""" dr = HI., , i = 1,2, ... , 6 (2.29)

for we can obtain all the others by appropriate permutations. And of these only
three are independent. They are,

HI,I = 3eo + H~,1 (2.30a)

 2.3 Three Hydrogen Atoms 51

H I •2 = H I •3 = H I •4 = 3eo[2 + H;.2 (2.30b)

H I • s = H I •6 = 3eoP + H~.s. (2.30c)

From (2.30a) we also get H 2 •2 = H 3 • 3 = ... = H I • I • From (2.30b) we get the

matrix elements of :Jft" between any function in (2.23) and its transform, for
example, H 2 • S = H I •2 , etc.; and (2.30c) gives the archetype matrix element be-
tween functions of (2.24) and their transform. A slight change of notation will
greatly simplify the appearance of the interaction matrix. We define A, b, and c
by

A = H;.l> b[2·A = H~.2 and cP·A = H~.s. (2.31)

This replaces the matrix elements as parameters by A, b, c, and the previously

defined overlap integral [ [cf. (2.10)]. The Hamiltonian matrix is then particular-
ly convenient to derive: 3eo times the overlap matrix, plus A times an interaction
matrix (which can be derived from the overlap matrix merely by replacing [2 by
bf2 and P by cP in the latter), so that the eigenvalue equation now reads:

b[2 bf2 b[2 cP cP [2 [2 [2 P P

b[2 cP cP b[2 b[2 P P [2 [2

b[2 cP cP b[2 b[2 [2 P P [2 [2

A ·v = (E - 3eo) ·V
b[2 cP cP b[2 bf2 [2 P P
[2 [2 (2.32)
cP b[2 bf2 b[2 cP P
cP bf2 bf2 bP cP 1

We denote eigenvectors by v. We construct the six distinct eigenvectors by use

of the permutation tables given just previously. One starts with If/I, or

o
o
(2.33)
o
o
o
in the vector notation. A totally symmetric un-normalized function is constructed
by adding to this vector all the vectors obtained by permutations of the particles;
for example,

V.ym = (1, 1, 1, 1, 1, 1). (2.34)

52 2. Exchange

(For typographical reasons, we now write the vectors as row vectors. Technical-
ly, therefore, they are left eigenvectors.) For a totally antisymmetric but un-
normalized function, we again start with If/I> subtract all the odd permutations,
which are given in (2.23), and add the even permulations, given in (2.24)

V.- s = (1, -1, -1, -1, +1, +1). (2.35)

We next seek vectors antisymmetric in particles 2 and 3, but not totally antisym-
metric, that is, orthogonal to Va-so We find

V 23
1 2'
= ( 1, 2' 1 -1, -2' 1) and
1 -2 I
V 23 = (0, 1, -1,0, +1, -1). (2.36)

Next, we look for vectors symmetric in particles 2 and 3, but orthogonal to v sym

V23sym = (1, - ~ , - ~ , I, - ~ , - ~ ) and

V;3sym = (0,1, -1,0, -I, +1) (2.37)

These choices are not unique. However, because of the invariance of the
Hamiltonian under permutations, functions of different symmetries do not mix.
The totally symmetric and the totally antisymmetric function both stand alone
in their own symmetry class and therefore must be eigenvectors. The remaining
four eigenvalues are obtained by diagonalizing the matrices of the eigenvalue
equation (2.32), in the 2 X 2 subspaces of the functions of (2.36,37), respectively.
This use of symmetry saves us from the tedium of diagonalizing 6 X 6 matrices:
the importance of the permutation operators should be already abundantly
clear.
In the present problem, hidden symmetries simplify the eigenvalue equation
further. For it happens that each of the last four vectors is simultaneously an
eigenvector of both the overlap and the interaction matrices, and all are de-
generate. The eigenvalues can be found almost by inspection now.

1 + 3bf2 + 2cP
E,ym =.3eo + A 1 + 3[2 + 2P (2.28)

1 - 3bf2 + 2cP
E._. = 3eo +A 1 _ 3[2 + 2P (2.39)

and the four-fold degenerate eigenvalue,

1 - cP
E 23 = 3eo + A 1 _ [3 . (2.40)

Elsewhere we discuss how to combine space with spin to obtain wavefunc-

tions of space and spin, obeying the Pauli principle. At present we need only the
 2.3 Three Hydrogen Atoms 53

following which the reader must accept on faith: V,ym is not an allowable eigen-
function for electrons; V.- s will be used to construct the function of spin three-
halves and V 23 the function of spin one-half, known as quartet and doublet
states, respectively. No other values of total spin angular momentum can be
obtained with three electrons.
How do these results compare with the solutions of the Heisenberg Hamil-
tonian? To the two-spin Hamiltonian of (2.17) which corresponded to the
hydrogen molecule, -J I2 S1 • S2' we now must add two more equal bonds to
connect all sides of the equilateral triangle, as shown in Fig. 2.4. That is,

(2.41)

Fig. 2.4.
Three-atom molecule, or three
equivalent Heisenberg spins,
with solid lines indicating
bonds

We use a superscript (~), however, to warn that the exchange "constant" might
be rather more variable than its name suggests, and that the value we shall find in
the present calculation may not agree with the result previously obtained for two
atoms.
The distinct eigenvalues of the above Heisenberg Hamiltonian can be cal-
culated by diagonalizing .%Hels in the subspace of the three functions belonging
to M = + t. Because of the rotational symmetry, it is found moreover that the
two solutions belonging to M = S = t are degenerate. The quartet solution is
of course unique and, therefore, automatically an eigenfunction of the Hamil-
tonian. The reader is encouraged to construct these states, starting with a basis

(2.42)

in an obvious notation. One can also obtain the eigenvalues more simply by
completing the square in (2.41). Either way, he finds for the solutions of (2.41)

(2.43)

Now this must be set equal to

54 2. Exchange

I - 3bP + 2cP I - CP) 3

E.- s - E 23 = A ( 1 _ 312 + 2P - t=f3 = -2:Ji2 (2.44)

the calculated level separation, so as to obtain finally a value for the exchange
parameter

r = _ 2Af2 (1 - b) + (l + f2)(c - b) (2.45)

12 (1 - [3)(1+ 1- 212)
Unless the overlap is very small, this expression has terms in [3, etc., and is in no
way comparable to the formula of (2.18) which we obtained previously. How-
ever, there is no reason to take the H-L scheme seriously if the overlap is that
great. [Note that if the atoms are exceedingly close, then the eigenfunctions of
atomic lithium are a better zeroth-order set than the three hydrogen atomic
functions, whereas those of helium best approximate the solution of the H2
molecule. In vain can we expect the H-L picture to describe these physical
systems when I""' 1. In solids, the H-L picture breaks down completely once
I approaches or exceeds liz in magnitude, where z = number of nearest neigh-
bors of each atom (z = 6 for simple cubic structure, etc.) because Q becomes
a singular matrix. That is, the wavefunctions cannot be normalized unless I <
liz; see ahead].
For the reasons above, in a microscopic derivation of the Heisenberg Hamil-
tonian and of the exchange constant, we must assume that Rob = many atomic
units, and calculate only to lowest order in the overlap, I. One is powerless to
define interatomic exchange more accurately (although once again in the aton.,
for intra-atomic exchange, the situation is more favorable, as we shall see subse-
quently).
In accordance with these arguments, we now calculate the exchange para-
meters to lowest order in the overlap: From (2.18),

and the calculation of (2.45) giving

~ Ji2 = -Af2 + AJ2b = -H;,I[2 + H;,2

= -12 f d 3rl d 3r2 d3r3qJ~(rl)qJHr2)qJ~(r3)[ (;:b + ;:c + ;:J
 2.3 Three Hydrogen Atoms 55

(2.47)

recalling the definition of the various parameters in (2.31), and assuming also
for simplicity that all the atomic functions are real, and normalized.
Let us try to reduce the second calculation to some of the integrals in the
first, so that we may effect a comparison. First, (2.46): The terms in R"b cancel
exactly, and the remainder can be put in the form

21 JIZ = f d3rl d 3rz (2eribZ - eZ) (P n J2 - Pnex)

rn
(2.48)

defining Pn to be the ordinary electron density qJ;(rl)qJ~(rz) and PIZex to be the

"exchange density" qJ,,(rl)qJb(rl)qJ,,(rZ)qJb(rZ), in an obvious notation.
Some simple manipulations of the terms in the integrals for Jiz enable us
finally to reach the desired result

(2.49)

The correction term is not necessarily negligible compared to the exchange

parameter itself (see Problem 2.2).

Problem 2.2. qJ,,(r) is the unperturbed Is atomic ground-state wavefunction of

an electron belonging to a hydrogen atom at R", and qJb is defined correspond-
ingly. Calculate I(R"b) and J 12(l), JI~(l). Is the exchange parameter increased
or decreased by the presence of the third atom? (Assume R"b to be large). See
also [Ref. 2.12, Eq. (6)].

The presence of the third atom at Rc modifies the exchange bond, as we have
defined it, between spins at R" and R b • If atoms a and b were imbedded in a solid,
the exchange interaction between them would be further modified, and the
molecular exchange constant J 12 or Jl~ would be of little quantitative use. Those
who would calculate exchange parameters must heed the following warning:
It is futile to imagine that exchange, even between similar atoms, can be charac-
terized by a universal parameter J lz (R 12 ) which one might calculate on a mole-
cular model, no matter how refined. It should be regarded as a parameter which
depends on all the other environing atoms as well, even to lowest order. When
the Heisenberg Hamiltonian (2.1) is not applicable, the exchange parameter
loses, of course, even this meaning, and exchange vanishes into the limbo of ill-
defined concepts and ideas.
Herring [2.10] points out that the H-L scheme per se is in fact asymptotically
wrong for R"b > some 50 atomic distances, where it predicts a ferromagnetic
exchange parameter, albeit an exponentially small one. Thus even if the Heisen-
56 2. Exchange

berg Hamiltonian is applicable, and even if an exchange parameter can be

defined and the overlap is small, it still seems a tricky matter to calculate the
exchange parameter properly! In the following sections, the difficulties are
compounded, not to be resolved until a later chapter.

2.4 Nonorthogonality Catastrophe

In the many-body problem, when one unwittingly expands eBN in a Taylor series,

eB N = I + gN + ~ (gN)2 + ... (2.50)

and tries to take the limit N - + 00, he is overtaken by what is mildly termed a
catastrophe. Because of the use of nonorthogonal orbitals in the H-L scheme,
just such a nonorthogonality catastrophe occurs when we try to apply the me-
thods which were used in the two-and three-body problems, to the N-body
problem of an infinite chain or infinite three-dimensional solid. This difficulty,
first pointed out by Slater [2.11] and Inglis [2.12], has for many years been
resolved in principle. Nevertheless it adds considerably to the confusion and
difficulties of the problem and to the headaches of those who seek to under-
stand exchange.
The overlap matrix and the interaction matrix will both be N! X N!, al-
though as we have mentioned not all the eigenfunctions will be admissible, and
only 2N will survive the process of antisymmetrization. A row in the overlap
matrix may include one 1, N terms in [2, N2 terms in f3, etc., and the question will
no longer be, "can we neglect f2 compared to 1, or f3 compared to P?" but rather,
"can we neglect NP compared to I? Or N2[3 compared to NI2?" The answer is
an emphatic "No!" no matter how small the overlap. In the present section we
give a preliminary qualitative, intuitive analysis resulting in an approximate
derivation of the Heisenberg Hamiltonian.
It is not practicable to set down either the overlap or interaction matrices in
the explicit form we have previously used, and an operator formalism is re-
quired. We expand both as series in permutation operators, decomposed into
simple transposition operators and their products, and then take advantage of
the Pauli principle to express these in the form of spin exchange operators.
Assume, for simplicity, that the atomic functions are truncated slightly so
that there is overlap only with nearest neighbors, in a regular geometric array
of Natoms.

f m*(r)m (r) d 3r - {0I if n,h m =.

nearest neighbors
(251)
'l'n 'I'm -
ot erWlse .

Then with the normally ordered functions as the standard, the N! functions can
be obtained by various permutations of the particle coordinates
 2.4 Nonorthogonality Catastrophe 57

"'I = 9'1(r l)9'2(r2) ..• 9'N(rN)

f!/Il2"'l = 9'1(r2)9'2(rl) '" 9'N(rN) (2.52)
f!/Il2f!/12N"'1 = 9'1 (r2)9'2(rN) ... 9'N(rl),

"'I'
etc. The great advantage of the operator formalism is that the eigenvalues of the
matrices we obtain are independent of the choice of the initial function
We can almost guess that the overlap matrix has the following operator re-
presentation:

o = 1 + liP + 1.112/4
2
+ ... = e + ... A/' (2.53)

with
II == ~
(n,m)
f!/I nm (2.54)
-nearest neighb.

and dots (oo.) to hide the real difficulties. There are two types of correction:
"Kinematical" are those due to counting errors, because a product of x factors
of f!/Inm does not always describe a permutation of x pairs of coordinates. For
example, terms of the type

(2.55)

should be subtracted from i"2/4, so that the first kinematical correction is

(2.56)

where z is the number of nearest neighbors of each atom (for example, Z = 2

in the linear chain, and z = 6 in the simple cubic lattice). The next constant
correction is

(2.57)

neglecting terms of order NIB which are smaller than (2.56), not to speak of
(2.57). The fact is, that the principal corrections enter with ever-increasing
powers of N, to form an apparently divergent series. If we add the largest con-
tributions in II, liZ, etc., the overlap matrix can be resummed as
O=Refll'+oo. (2.58)
with R representing a nonconvergent (that is, N-dependent) series

R=~
N (Nz/4)n
- en (2.59)
n-O 4
and the ellipsis (oo.) higher-order neglected terms.
58 2. Exchange

Among further corrections are the "dynamical corrections," which we shall

now define. These are concerned with the nature of the particular law assumed
for the overlap, (2.51) in the present case. Unlike the terms in R, the dynamical
corrections would be subject to change if we assumed a next-nearest-neighbor
overlap, or some other law. They also depend on geometrical niceties, such as:
Can an atom have nearest neighbors which are nearest neighbors of each other?
If the answer is negative (as in the simple cubic lattice) the first dynamical cor-
rection to the overlap occurs in the third term of (2.53), from which we must
subtract

(2.60)

in order to eliminate what amounts to an effective next-nearest-neighbor ex-

change, prohibited by our postulates. Further dynamical corrections consist, in
part, of eliminating from the series for D all the permutations which, incorrectly,
remove an electron to a distance further than its nearest neighbors.
An entirely analogous analysis can be made of the interaction matrix. It is
important to recall that there is no exchange interaction between nonoverlapping
electrons, as was proved in a preceding section; so that the truncation of the ato-
mic functions, cf. (2.51) supra, leaves only a nearest-neighbor interaction. Even
the long-ranged Coulomb force cancels to a large extent because of average
electrical neutrality. The problem is most simplified if we take for the zero of
energy not N times eo, but 3e'I,I. The interaction term, which is the extension to
the present problem of the left-hand side of (2.32), thus can also be guessed to
be a series in powers of h, with corrections of both the kinematical and dynamical
sort, i.e.

Int = u1h + ~ u h + ...

2 2 (2.61)

with
(2.62)

(2.63)

Besides corrections of the same nature as in the overlap matrix, there are addi-
tional dynamical corrections to the un's, particularly when there are successive
interchanges of neighbouring pairs. For example, instead of U 2 as defined above,
the coefficient of the particular permutation f!J112[J1I34 should be precisely

(2.64)

or else one counts double the nonnegligible interactions among the four particles
 2.4 Nonorthogonality Catastrophe 59

when they are nearest neighbors. Again leaving this and other corrections aside,
we find
Int = hUI(Re A1 ' + ... ). (2.65)
Thus, the eigenvalue equation takes the form

(2.66)

with E measured relative to H1,I and rp the appropriate spatial eigenfunction.

However, since the latter is eventually to be multiplied by the appropriate spin
functions and totally antisymmetrized, it is permissible to replace the particle
permutation operators glInm by their spin conjugate operators,

(2.67)

Let us suppose momentarily that the corrections indicated by dots do not diverge
as badly as R, or that they are approximately the same on both sides of the above
equation. The parentheses can be factored in that case, and making the above
substitution into (2.66) we obtain the HDVV (Heisenberg, Dirac, Van Vleck)
eigenvalue equation

(2.68)

with X a spin eigenfunction. and (n,m) are nearest-neighbor pairs. Except for
the trivial additive constant term, the left-hand side of this eigenvalue equa-
tion is the Heisenberg nearest-neighbor exchange Hamiltonian. The exchange
parameter is the integra12uI> in the present case, which can be compared to the
exchange parameters J 12 Jt1, which were obtained in the preceding section.
Although it is of the same general form, the actual numerical value may differ.
Like J 12 , UI may in general be positive or negative. It is normally negative,
which leads to an antiferromagnetic ground state. This is in agreement with the
observed ground state of N atoms of hydrogen (a molecular nonferromagnetic
solid at or near zero absolute temperature).
Although the "catastrophe" has been averted, it has not been shown that the
neglect of the dots is justified. Permutations leading to terms quartic or higher-
order in spin operators may sometime be found to be of importance. 4 In the
following section, we shall discuss a method which has aims somewhat more
modest than the construction of a Heisenberg Hamiltonian, but achieves these
limited goals quantitatively without uncontrollable approximations. In the cha-
pter on magnetism and magnons in metals, the Heisenberg Hamiltonian will

4 For a systematic analysis of this approach see [2.13]. For different approaches cf. the next
section and Chap. 6. Experimental evidence of higher-order spincoupling is found in [2.14],
and also in [2.15].
60 2. Exchange

finally be seen to be the incidental consequence of low-order perturbation theory

carried out for insulators. The neglect of three-body permutations will be seen
as equivalent to the neglect of third- and higher-order perturbation corrections-
However before presenting this simplified interpretation, many other topics
will occupy our interest.

2.5 Method of Lowdin and Carr

The nonorthogonality catastrophe in the Heitler-London theory was introduced

and discussed qualitatively in the previous section. Here we shall present a con-
structive method for calculating directlY the eigenvalues of the H-L Hamiltonian.
For the range of parameters in which this theory gives reasonable results, these
solutions are equivalent to finding the energy of the 2N lowest eigenstates of Z,
which can then be fitted to a Heisenberg Hamiltonian. If the fit is good, it
provides evidence that the procedure of the previous section works. If the
parameters of the Heisenberg Hamiltonian cannot be chosen so as to reproduce
the correct eigenvalues, then the evidence is that exchange processes involving
three or more spins are important, i.e., the higher-order permutations indicated
by dots in the previous section do not factor between the interaction and the
overlap, and are of physical importance.
One advantage of the method of P.O. Liiwdin as discussed by W.J. Carr s is
that it deals with N X N arrays rather than N! X N! arrays of the permutation
space. We shall here obtain the variational energy of two very important states
with it: the ferromagnetic state, for which the spatial function is the totally anti-
symmetric determinantal function, and the Neel anti ferromagnetic state of
alternating spins "up" and "down," which we shall define subsequently. Starting
with the ferromagnetic state for which the many electron wavefunction is a
determinantal function

qJI (r l ) qJ2(ra qJN(r l )

qJI (r 2) qJ2(r2)
1
'1'-- (2.69)
- -IN!

qJI(r N) qJN(r N)
We calculate the energy of this state as simply

E J'P*JflJld3rl ... d3rN

Ferro = J'1'* 'I'd 3rl ... d 3rN • (2.70)

S Lowdin's methods for nonorthogonal arrays were applied to the problem of ferromagnetism
by Carr [2.16]. A variant method was used by Takano [2.17].
 2.5 Method of LOwdin and Carr 61

The Hamiltonian JJe'may quite generally be taken of the form,

(2.71)

such that the principal diagonal in the determinantal function is an eigenfunction

of ~ with eigenvalue eo. Without loss of generality, let us pose eo = 0, which
sets the origin of the scale of energy. Thus,

E = ~ fII£'1 2 U(r,)d'rl •.• d'r N + ~~ fII£'1 2 V(r" rJ)d'rl ... d'r N

Perro, denominator , .,. J denominator .
(2.72)

The denominator may be manipulated as follows:

denominator = ~! J

1P1(rl)lPl(rl) 1P1(rl)1P2(rl)

=J
1P!(r2)1P1 (r2) 1P!(r2)1P2(r2)
(2.73)

1P;(rN )IPI (r N )

The second form of "denominator" is obtained by noting that each term in the
expansion of the left-hand determinant contributes equally. Using this sym-
metry, one obtains exactly the same value of the integral by retaining only the
principal diagonal and multiplying the result by N! By the standard rules of
multiplication of determinants by scalars, each row in the right-hand array can
now be multiplied by a function occurring in the principal diagonal of the first
array, and this results in the third form of (2.73).
The numerators are handled in an analogous manner, although the problem
is somewhat trickier. We may not assume that the various terms in the potential
energy are invariant with respect to interchange of the particle coordinates. How-
ever the total Hamiltonian is invariant under such interchange. We obtain, for
one-body potentials

f II£' 12 ~, U(r,) d'rl ... = ~

~J
f 1P1(r,)U(r,)lPir,) d'r,
X f D~ d'rl ... d'r,_l d'r,+l ... (2.74)
62 2. Exchange

and for two-body potentials

(2.75)

The sum of these terms when divided by denominator, equals the ferromagnetic
energy.
The integrands D} and D:~ are obtained from the third line of (2.73) by strik-
ing out the ith row and jth column to obtain D5 X (_1)1+1; and the ith and jth
rows, Ith and mth columns, to obtain Di~ X (_l)I+J+l+m. The functions which
have been so stricken are explicitly taken into the energy integrals. Now all these
determinants can be easily evaluated by expanding, integrating, and re-collecting
terms. One finds first

Denominator = det ILli I where (2.76)

The integral of D} is seen to be merely the cofactor of LIj in this determinant.

And by the usual theory of determinants, we have

f D~d3rl ...
J = (L-I) (2.77)
denominator Ii .

And similarly,

f Dilim d3rl ... = (L-I) Ii (L-I) mi _ (L- 1)Ii (L- 1)ml. • (2.78)
denominator

Here L - I indicates the matrix inverse to L, the latter being defined as the square
matrix array with elements Lij. In solids which possess translational invariance,
these are cyclic matrices and can easily be inverted. 6
For example, consider a one-dimensional chain with nearest-neighbor overlap
only. Thus, Lil = 1 and LI,i±1 == I, and all other matrix elements vanish. To
make the matrix cyclic, we set N + I = 1, which is equivalent to adopting
periodic boundary conditions. The eigenvectors v of L are the plane waves,

, = (Vb ... , v., ... ), V. = e ik • with k = ± 21integer

t--
N
(2.79)

and the corresponding eigenvalues are

Lk = I + 21 cos k. (2.80)

6 The problem of inversion of the overlap matrix is discussed further in [2.18-20].

 2.5 Method of Lowdin and Carr 63

Thus, it may be verified that Lnm = N-I ~ eik(n-m)Lk, and more importantly,
k
that the matrix elements of the inverse matrix are

= J.- ] d() cos ()(n - m) = d 1n- ml 1 + d 2 (2.81)

2n _It I + 21 cos () 1 - d2 ,
where

d= ~/ (1 - ../ 1 - 4[2), d 2 < 1.

In the second line the sum over k was replaced by the integral appropriate in the
limit N - 00. It is seen that the elements of the inverse matrix decrease ex-
ponentially with distance 1n - m I. Therefore if I is sufficiently small, only 1n -
°
m 1 = will survive, which corresponds (as we shall see) to a Heisenberg nearest-
neighbor interaction.
Let us now illustrate this in three dimensions, in the simple cubic structure.
Using (2.72) and those following, we find that the energy of the ferromagnetic
stage in the three-dimensional array is

(2.82)

Let and RmJ == an' = a(n;, n;, n;) (2.83)

define
Uk = I: e-ikoR" f rpr(r)U(r)rpir) d 3 r
I
(2.84)

and
Lk = I: e- ik • R" L1j = 1 + 2/( cos kxa + cos kya + cos kza). (2.85)
I

Thus if I> i some Lk vanishes and LIj does not have an inverse. Using the
identity Ilf = f~ ds e- s, and the definition of In(z), the Bessel function of ima-
ginary argument, In{z) = 1/2n f~ltd() cos n() exp (z cos(), we find
~

(L-I)mj = L-'(n;, n;, n;) = f ds e-sIn\ (2sl)Inl (2s1)In: (2s1) (2.86)

o

indicating the explicit dependence on distance by the argument (n;, n;, n;) and
assuming I < i. Note that to leading order in powers of I, L -1(0) = 1 + 6[2 == 1;
and L- I(1, 0, 0) = -I; L-'(1, 1,0) and L- I(2, 0, 0) = 0(12); etc. Finally, the
ferromagnetic energy is,
64 2. Exchange

(2.87)

with R, = R, + all, and R, = Rm - all'. The L- 1(0) terms yield the direct and
Heisenberg exchange interactions, denoted Vand U, respectively, in the notation
we used for the hydrogen molecule. The L-I (1,0,0) terms include integrals
reminiscent of the three-atom molecule

L-I(1, 0, O)L-I(O) f [lPtra(1,o,O)(r)IP!(r') - IPtra(I,O,O)(r')IP!(r)]V(r, r')

X 'P,(r)lPm(r') d 3 r d 3r'.

These, and successively more complicated terms cannot be associated with

simple operators of the form S/" SJ, but require such forms as

or (2.88)

in their description. Fortunately, they enter with ever-increasing powers of /,

and therefore may be presumed negligible if / is sufficiently small, Note that there
is no "catastrophe": / must be small compared to 11-, but not to 1/N, in order that
a Heisenberg Hamiltonian, with (at worst) corrections of the type indicated in
(2.88), describe the energy of the ferromagnetic state.
We turn to another important state, the anti ferromagnetic Neel configura-
tkon, defined as follows: Every electron of spin "up" is surrounded by neighbors
of spin "down." We note that because the spin functions of nearest-neighboring
electrons are orthogonal, the one-electron wavefunctions of space and spin are
also orthogonal, and therefore lack of orthogonality of the spatial parts is irre-
levant. There is no "exchange" whatever as U(r) and V(r,r') do not involve
the spins, and the energy is straightaway,

As we shall see in a subsequent chapter, the energy ofthe Neel state is likely with-
in 25 percent of the true ground-state energy of an antiferromagnet in three di-
mensions. Therefore, within the stated accuracy and restricted range of validity
of the H-L scheme, one can determine whether the ground state is ferromagnetic or
antijerromagnetic by comparing (2.87) and (2,89), without in fact any need to
calculate exchange constants or to first set up a Heisenberg Hamiltonian!
One often encounters the following statement: "The Heitler-London scheme
and the band structure theory coalesce for filled bands." As we have an en-
lightening example of this phenomenon in the present section, it is appropriate
to terminate with a brief remark on the subject. Return to the determinantal
wavefunction given in (2.69) and take linear combinations of the columns (this
does not affect the value of a determinant) to obtain
 2.5 Method of Lowdin and Carr 65

'I'k,(rl)
1 'fIk,(r2)
IJ'-- (2.90)
- ../N!

where

(2.91)

is the desired linear combination. It is exactly the Bloch wavefunction for an

electron in the tight-binding limit. The set offunctions in (2.91) is an orthogonal
set, albeit an unnormalized one. For,

(2.92)

from which one may deduce (2.80) and other results in the text. This sort of
procedure has been used by Takano [2.17] to examine the theory of spin waves.

Problem 2.3. Assume a simple cubic structure, quasi constant electron wave-
functions

when r is within the unit cube (a 3 ) centered about R J , or within any of the six
nearest-neighbor unit cubes centered about RJ ± (a, 0, 0), ... , RJ ± (0,0, a),
and

otherwise. Let the potentials be

U(r) = -A

that is, constant, and

V(r, r') = + BJ(r - r').

Using these functions to calculate the various quantitites in the method of

Lowdin and Carr, find the following:
(a) The ferromagnetic and Neel state energies: Determine which lies lower.
How does this depend on the magnitude of A? of B?
(b) In the pure Heisenberg model, simple cubic lattice, there are 3N bonds
66 2. Exchange

and therefore EN601 - Eperro = !NJ12 , where R lz = a = near neighbor distance

and J 12 = near neighbor exchange parameter. Using the energies calculated
in part (a), obtain J 12 • Then use the definition, (2.18), to calculate J IZ ' Explain
the agreement (or disagreement) between the results of these alternative calcula-
tions.
3. Quantum Theory of Angular Momentum

This chapter affords a brief summary of the quantum theory of angular mo-
mentum. As the angular momentum of a charged particle is proportional to its
magnetization, this subject is at the core of the theory of magnetism. We shall
show that motional angular momentum is inadequate, and introduce spin angu-
lar momentum. We shall develop operator techniques expressing angular
momentum or spin operators in terms of more primitive fermion or boson
operators. The topics of spin-one-half and spin-one are treated individually, for
use in subsequent chapters on the theory of magnetism.

3.1 Kinetic Angular Momentum

In the absence of any external forces, the classical angular momentum of a point
particle has a constant value

L=r X P (3.1)

and therefore in quantum theory the analogous quantity, which shall be denoted
"kinetic angular momentum" is an operator,

h
L=rx-;-V (3.2)
1

according to the usual rule for constructing quantum-mechanical momenta:

p" = h/i(olox), etc. The angular momentum associated with a wavefunction If! can
be easily determined if this function is spherically symmetric so that it does not
depend on the angular coordinates rp and (), but only on the magnitude of the
radius vector r. For in that case we can prove that L == 0 by noting that

(3.3)

where ii, = unit vector in the radial direction = rlr, and therefore,

(3.4)
68 3. Quantum Theory of Angular Momentum

A function with nontrivial angular dependence has non vanishing kinetic

angular momentum, although it is not always so simple to discover its magni-
tude. For this requires the solution of an eigenvalue problem; and moreover,
the three components of L = (Lx, L y, L z) do not commute with one another and
therefore cannot be simultaneously specified. Briefly,

With the other two components obtained by cyclic permutations of the above,
we find a set of relations which can most conveniently be described by the
vector cross product, as given by the usual determinantal rule

Ux uy Uz
L x L : Lx Ly Lz = iliL. (3.6)
Lx Ly Lz

While it is true that this rule was obtained by the definition of angular mo-
mentum given in (3.2), it is in fact more general. There are operators (identified
later) which obey (3.6) but not (3.2). It is therefore important that we define
angular momentum as follows: A vector operator is a genuine angular momentum
operator (whether kinetic, spin, or generalized angular momentum) only if it
satisfies (3.6) or the equivalent set of equations (3.10--16).
Example: If L = (Lx, L y, L z) is an angular momentum, then L' = (-L", -Ly,
- L z ) is not. This is related to the pseudovector nature of the classical angular
momentum r X p, and is also directly a consequence of (3.6).
Returning to the kinetic angular momentum, we express it in spherical polar
coordinates
r = (r sin () cos ,p, r sin () sin ,p, r cos ())
in which the cartesian components of L take the form

(3.7)

L + = lIe l ; (i.
a(} + i cot () i.)
a,p (3.8)
 3.1 Kinetic Angular Momentum 69

L- = lie- i ; ( - :0 + i cot 0 :~). (3.9)

Here, for future convenience, we introduced the operators

(3.10)

But if desired, the Cartesian components may be extracted by (what is equivalent

to the last equation):

and (3.11 )

The ± operators are the more useful, though, and have the names of angular
momentum raising (+) and lowering (-) operators, respectively. The reason for
this nomenclature is to be sought in the commutation laws for these operators.
Because it is possible to derive these commutators using only (3.6), without
making use of the actual differential forms for L., etc., given above, the following
equations are generally valid for any angular momentum:

[L., L +] = IiU (3.12)

and
(3.13)

Given an eigenfunction 'Pm of L z with eigenvalue m, that is,

apply both sides of (3.12) to this eigenfunction and after rearranging terms,
obtain

This shows that L + 'Pm is an eigenfunction of L z belonging to eigenvalue m + 1.

It is similarly shown that L-'Pm belongs to eigenvalue m - 1, whence the termi-
nology of "raising" and "lowering" operators. A final commutation law which
can be obtained from (3.6) is

(3.14)

There is only one important operator which has not yet been introduced;
it is the scalar associated with the vector angular momentum, its length.
70 3. Quantum Theory of Angular Momentum

v = L~ + L: + L~ (3.l5a)

= ~ (L+L- + L-L+) + L; (3.15b)

= UL- + Lz(Lz - Ii) (3.15c)

= L-L+ + Lz(Lz + Ii) (3.15d)

The various alternative forms for V are obtained from each other by means of
the commutation relations of(3.14), and the definitions (3.10, U). Some will be
more useful than others when dealing with eigenfunctions of L .. for example.
It is trivial to prove that V commutes with all the components of L, that is,

(3.16)

This is seen by inspection for Lz as V does not involve the angle,p, and there-
fore must commute with aja,p. It follows by rotational symmetry, that it also
commutes with both other components of L.
Because V and L z commute, they have a complete set of eigenfunctions in
common, which can be labeled by the eigenvalue of each of these two operators.
Because Lx and Ly (and also L±) do not commute with L .. they cannot be
simultaneously diagonalized with the former two operators, and therefore their
eigenvalues cannot also be specified. But there is no loss of generality in singling
out the component L .. since the z axis can be picked along any arbitrary direc-
tion. In the case of kinetic angular momentum, the desired eigenfunctions will
be the well-known spherical harmonics, as we shall now discover.

3.2 Spherical Harmonics

Because the radial dependence of the wavefunction plays no role in the deter-
mination of angular momentum, we may dispense with it altogether and consider
functional dependence on the angles only. The wavefunctions are normalized on
the unit sphere,

27f 7f

fo d,p f dO sin OIp*(O, ,p)Ip(O, ,p) = 1

_7f
(3.17)

and the two simultaneous eigenvalue equations are

Lzlp
Ii a
== T a,p Ip = limlp (3.18)

and
L2
Ip
.,.2[ 1 a(. oa) 1 a2] .,.21
= -" sin 0 ao sm ao + sin2 0 a,p2 Ip = " Alp (3.19)
 3.2 Spherical Harmonics 71

with eigenvalues, respectively, of m (the "magnetic quantum number") and A.

We still show the dependence on the quantum-mechanical unit of angular
momentum Ii, but eventually it will be most convenient to use units such that
Ii = 1, and such units will be assumed throughout most of the book.
The L z equation can be integrated directly, with the result

(3.20)

A boundary condition must now be invoked to determine tpe quantum numbers

m, an obvious choice being to require If/(O, rp) to be single-valued on the unit
sphere. Therefore we require

(3.21)

which is satisfied by the choice

m = integer = 0, ± 1, ±2 .... (3.22)

Although other boundary conditions are possible, they would violate other
requirements of quantum theory which we have not yet discussed, which shall
be examined in the following section. The same boundalY condition of (3.21)
has also been used by Dirac (see Note at end of section) to prove that the elec-
tric charge is quantized, in units of a fundamental charge q.
The solution, (3.20), may now be introduced into the second differential equa-
tion, (3.19), which is the eigenvalue equation for £2. One finds directly

(3.23)

which is the equation obeyed by the associated Legendre polynomials, the pro-
perties of which are well established. But if they were not known, the following
constructive procedure would solve this eigenvalue equation by elementary
means:
Assume that the wavefunction If/CO, rp) = e1m;If/(O, 0) is a solution of the
following first-order partial differential equation,

(3.24)

Therefore, by (3.15d), '" is an eigenfunction of £2 with eigenvalue

A = m(m + 1). (3.25)

The value of m is a special one, by virtue of the previous equation, L +", = O.

It cannot be stepped up. It is the maximum value of the azimuthal quantum
number for the calculated value of A. Therefore we shall denote it by the symbol
I, defined by
72 3. Quantum Theory of Angular Momentum

A = 1(1 + 1) (3.26)

and let m continue to represent the azimuthal (or "magnetic") quantum number.
For the function which obeys (3.24), 1= m. By repeated application of L-, we
can generate functions always belonging to the same value of A or I, but with
decreasing magnetic quantum numbers m = I - 1, m = I - 2, .... The series
terminates with m = -I, by (3.15c).
The first eigenfunction, with m = I, is obtained explicitly by integrating the
linear first-order differential equation (3.24). Given

(3.27)

divide by (cos ()'(lfI), and integrate to obtain

lfI = (sinl ())(e il ¢) const. (3.28)

Except for normalization factors, the remaining functions of the set having the
common eigenvalue I are found by repeated applications of the operator L- to
this solution. These functions can be normalized and given standard phases, and
are denoted the spherical harmonics, YI,m.

Y
I,m
«()~) = (-1 y+m
, 21l! 'V
/ (21 + 1)(1 - m)! • (sin
4n(l + m)!
()m (_a_)
acos ()
/+m (sin ()21e 1m(

(3.29)

These are, then, the associated Legendre polynomials of (cos (), multiplied by
exp(im~) and suitably normalized over the unit sphere.) These functions form
a complete, orthonormal set (see Table 3.1).
Note: Dirac assumed the existence, somewhere in the universe, of a magnetic
monopole. Although such objects have never been found on earth or in astro-
nomical observations, which have so far always indicated that the magnetic di-
pole is the fundamental source of magnetic fields, the search goes on for this
elusive particle,2 and we may follow Dirac in assuming the existence of at least
one such source-of strength 110. The magnetic field of such a pole is

pointing in the radial direction, which implies a vector potential (in spherical
polar coordinates)

A = (Ar> A o, A;) with AT = Ao = 0 and A; = ~Otan( ~ ()).

I The phase (-l)1+m is chosen in conformity with [3.1,2].
2 See, e.g. [3.3]
 3.2 Spherical Harmonics 73

Table 3.1

m Y, ... (normalized on unit sphere)

1
0 0

I
2.J7t
-;-[ - [£(X + iy)] = -aSin8el,
87t
I 0 1-[£z
r 47t = [£cos8
47t

2 ~[ J 31 i.7t(x + iY)Z] = J 327t

15 sin z 8e zl,

~[ - J ~;z(x + iy)] = - J 15 cos 8sin 8eltp

21 0 ~[Jl~7t(3ZZ - rZ)]
87t
= J 5 (2 cosz 8 - sinZ8)
167t

3 ~[ - &(x + iy)3] = - J 35 sin' 8e31,

647t
2 ~[ ./ffiZ(X + iy)l] = JlO5 cos 8 sin z 8ell,
327t
3
~[- &(5Z rl)(x + iy)]
Z - = - J 21 (4 cos l 8 sin 8 - sin 3 8)e1,
647t
0 ~[Jl~7t(5z1 - 3r l)z] = ~ (2 cos38 - 3 cos 8 sinz 8)
167t

4 1-[..l
r 16 j!J(X
4 27t
+ iy).] = nJJtt sin4 8e·l,

3 ;'[-tJ~ z(x + iy)3] = -t JJf: sin3 8 cos Be31,

4
2 ;.[t [f;(7z1 - rl) (x + iy)l] =t If 27t
(6 sinl 8 cosl 8 - sin4 8)eZI,

f.-[ - !&(7z1 - 3r )z(x + iy)]

Z = - ,Ji;f(4 sin 8 cos38 - 3 cos 8 sin38)e1,
647t
0 1-[_3 -(35z4 _ 30r zzz + 3r.)]
r. 16-v'1t
= 16~(8 cos· 8 + 3 sin· 8 - 24 sinl 8 cos18)

Note: ~ (x4 + y. + z· - ~ r·) = 4~[ Y..o + J~ (Y 4 •• + Y 4• _.)]

Y,.-.. = (-1)" Yr. ..

This may be verified by the relation V x A = H, defining A. As we know, the

Hamiltonian, as well as any other dynamical operator of a ± charged particle,
is modified in a magnetic field because of a change in the momentum of the
particle
e
p-p±-A
c

with c = speed of light. But notice that A - 00 along the line () = n, in just
74 3. Quantum Theory of Angular Momentum

such a way that the integral along a path enclosing this line has a definite value,
namely,

A particle which in the absence of the monopole had magnetic quantum number
m, now has magnetic quantum number

m' = m ± (e f A 2~:e) = m ± 2:lieJjo41t.

The wavefunction must still be single-valued, hence m' must be an integer, from
which it follows that ± e must be an integer multiple of the "fundamental
charge"

lie
q=- Q.E.D.
2Jjo
By showing moreover that there is no fundamental unit of length associated
with the magnetic monopole, Dirac proved that there could not be any bound
states of a charged particle in the field H, so that the main physical effect of the
monopole is the quantization of charge [3.4].

3.3 Reason for Integer land m

It is a straightforward matter to show that I and m are all integers, or all half-
odd integers. The choice between these two sets is, however, not so simple.
Let us prove the first statement by recalling the boundary condition, (3.21),
by which we required the wavefunctions to be single-valued on the unit sphere.
This is too strict; in fact, the only physical requirement is that the probability
density-a physical observable-be single-valued in an arbitrary state. That is,
if we take a wave function which depends on ~ as follows:

"'(~) = ~ Am(e 1m;

m
+ eiCm+ + eiCm- + ...)
ll ; ll ;

and require: (3.30)

then the only way to satisfy this in general is to have all the m's integers, or all
the m's half-odd integers. It is a simple exercise to verify that if both are admitted,
or other choices of m are made, then some probability densities can be con-
structed which are not single-valued.
 3.3 Reason for Integer I and m 75

The choice between the integers and half-integers is now made on the basis
of the following reasoning. Physical quantities must be independent of the choice
of coordinate system (whether Cartesian, spherical, or whatever) and of the
the origin of that coordinate system. For example, even for particles not under-
going circular motion, the complete set of eigenfunctions of angular momentum
must be adequate to describe the angular part of the motion. And conversely,
purely spherical motion must be describable in terms of other complete sets of
wavefunctions, such as the plane-wave states. Let us take this for an example.
The well-known formula for the expansion of plane waves in spherical harmonics
IS

(3.31)

where the coefficients

(3.32)

are spherical Bessel functions, here defined in terms of the ordinary Bessel func-
tions Jp(z). If e and l/> are the angles of the k vector and () and tjJ are the angles
of the coordinate vector r, the expansion formula is further reduced by ex-
pressing it in terms of the relative angle 0),

cos 0) = : : = cos () cos e + sin () sin e cos (tjJ - l/» (3.33)

using the so-called addition formula,

(3.34)

While we make no attempt to prove (3.31-34), they are well established in

elementary texts in electromagnetic and quantum theory. The Legendre poly-
nomials, for example,

(3.35)

are particularly well known (almost elementary) functions, constituting a com-

plete set of orthonormal polynomials on the interval -1 S x S + 1.
The expansion of the plane waves given above or of any other useful complete
set of nonspherical wavefunctions in a complete set of spherical wavefunctions
requires only the integer values of I and m, and this is what finally fixes our
choice.
76 3. Quantum Theory of Angular Momentum

Note that the plane waves are a complete set of states of the infinitesimal
translation operator. The half-odd integer states must therefore be reserved for a
different space, one in which only rotations are permitted, called spin space.

3.4 Matrices of Angular Momentum

The spherical harmonics are the orthonormal eigenfunctions of L2 and Lzo and
by using them we can calculate the matrix structure of such operators as L" and
L", or L±. For example, supplementing the relationship we had already derived,

L+Y", = 0,

we can obtain

L+Y"m = h-J(l- m)(l + m + 1) Y"m+l (3.36)

by operating directly on the spherical harmonics given in (3.29). The following

relations are also very useful:

L-Y"m = h-J(l- m + 1)(1 + m) Y"m-l (3.37)

LzY"m = hmY"m (3.38)

and
VY',m = [L2z + ~(L-L+
2 + L+L-)Y',m
= h2[m 2 + ~ (/- m)(I + m + 1) + ~ (1- m + 1)(1 + m)]Y"m
= h /(1 + I) Y"m'
2 (3.39)

Note that in this last equation, no special property of the spherical harmonics
was used. The eigenvalue of V can be calculated using the previous three
equations, that is, by knowing the matrix structure of L± and L z • But this matrix
structure, also, can be derived without explicit recourse to the spherical har-
monics, by using the commutation relations, (3.12-14). Thus it would apply
also to the half-odd-integer solutions in which we shall soon be interested. To
distiuguish these, we shall want to extend our notation.
The conventional notation of L±, Lzo and V, I, and m, will continue to be
used when we are dealing with kinetic angular momentum. For general or ar-
bitrary angular momentum, when only the basic commutation relations are in-
voked but the arguments are not restricted to integer values of 1 and m, J will
replace Land j will replace I, although the eigenvalue associated with J will still
be denoted m. We shall also use S interchangeably with J. The general matrix
structure of the genera) angular momentum operators can be obtained using only
 3.4 Matrices of Angular Momentum 77

(3.12-16) and the construction given in the first section. One obtains the follow-
ing results:

(j'm'l J+ Ijm) = c)J.J'c)m+I,m,h,.jU - m)U + m + 1) (3.40)

(j'm'l J-Ijm) = c)j,j'c)m-I,m,h,.jU - m + I)U + m) (3.41)

(3.42)

(3.43)

Because of their common factor c)J,J" the general angular momentum opera-
tors can be represented by (2j + 1) X (2j + 1) square matrices (irreducible
representations) which act in a subspace of the 2j + 1 linearly independent ei-
genfunctions of J2 belonging to a given j value. Such matrices satisfy the basic
commutation relations (3.6) or (3.12-16), by construction, and are therefore
genuine angular momenta in their own right.
Larger matrices can also be found which are angular momenta, but then they
have sub-blocks with the above matrix structure, and are called reducible re-
presentations. The most compact matrices having the properties of angular
momentum are irreducible representations of the angular momentum operators.
For example, the irreducible representation of angular momentum I = 1 is

(3.44)

For angular momentum I = 2 there are 5 X 5 matrices, for angular momen-

tum I = 3 there are 7 X 7, etc. It is a striking fact, however, that all the even-
dimensional representations are missing from the list if we insist on the integer
values of angular momentum possessed by the spherical harmonics. The missing
matrices, the even-dimensional ones, are the spin angular momentum operators
for j,m = half-odd integers. Upon including these, the series is made complete.
78 3. Quantum Theory of Angular Momentum

3.5 Pauli Spin Matrices

Here we shall discuss the smallest of the even-dimensional irreducible represen-

tations of angular momentum, the matrices of spin one-half. They are very
similar to the 2 X 2 matrices, which were invented by Pauli precisely for the
purpose of describing the intrinsic spin angular momentum of the electron.
In this smallest subspace, the spin operators are, according to (3.40-43),

S+ =h [~ ~l s- = h [~ ~l
and

S, ~ h r: -~j (3.45)

and

S2 = h2 .l [1
4 0 ~l (3.46)

The Pauli spin matrices are obtained from S., S" = HS+ + S-), and Sy =
(S+ - S-)/2i by the relation

(3.47)

and so are explicitly

ax = [~ ~l
and the unit operator: 1= [~ ~l (3.48)

The eigenvectors of S., and a z are the two-component spinors

and x- = [~l (3.49)

corresponding to m = ± t, respectively. Conventionally, a± are defined by

a± = S±.
The higher-dimensional even matrices, 4 X 4, 6 X 6, etc., can be similarly
constructed using the general set of rules for the matrix elements given in (3.40-
 3.6 Compounding Angular Momentum 79

43) and will represent the higher spins 3/2, 5/2, etc. There is no need to give these
explicitly, particularly because matrices become unwieldy with increasing size,
and because we shall find far more convenient operator representations.

3.6 Compounding Angular Momentum

It is frequently necessary to compound constituent angular momenta into a total

angular momentum operator and, conversely, to decompose and simplify opera-
tors with complex structure. Let us go into this matter briefly and directly, even
at the expense of omitting formal proofs.
Given two angular momenta II and 1 2 , it seems certain (intuitively) that
I = II + 12 will be an allowable angular momentum, obeying the commutation
laws of Eqs. (3.6) and those following. However, no other linear combina-
tion of II and 12 will do (cf. Problem 3.1).

Problem 3.1. Given angular momenta II and 12 , show that all + bl2 is itself
an angular momentum only if a = b = I; or a = I, b = 0; or a = 0, b = 1.

How does one express the eigenfunctions of II and 12 in terms of the eigen-
functions of I and vice versa?
Supposing we start by knowing a complete set of angular momenta eigen-
functions of II and 12 , whether spinors or spherical harmonics, or whatever,
which we label by the two sets of quantum numbers

(3.50)

For fixedjl andj2 there are (2jl + I)· (2j2 + 1) orthonormal eigenfunctions of
this type corresponding to the various choices of ml and m2 • Each one is, there-
fore, also an eigenfunction of Jz, with eigenvalue m = ml + m 2 • (Henceforth,
subscripts will label individual angular momenta, and superscripts their com-
ponents; for example, In
This has a maximum valuejl + j2, therefore it is the
maximum value of m, which is by definitionj, the quantum number of P [recall
(3.24-26)]. By repeated applications of the operator J- = J 1 + J 2 to this state
of maximal m, we generate a total of 2(j1 + j2) + 1 orthogonal states all be-
longing to the same j value, but with m = jl + j2, jl + j2 - I, ... , -jl -j2.
(Note that j is not changed, because J- commutes with P.) Only one of these
states has m = jl + j2 - 1, for indeed, every m value occurs only once in this
list. But there are two of the original states which belong to this eigenvalue of
JZ: /jljl - lj2j2) and Ijdlj2j2 - 1). The function which has been included is
in fact the sum of these two functions,

(3.51)

because J- = J1 + J 2 . This leaves us free to consider the difference of the two

80 3. Quantum Theory of Angular Momentum

(3.52)

This function is orthogonal to the previous one, it belongs to the same m value,
and when J+ = It + Jt is a applied to it the result vanishes. Therefore the m
value to which it belongs must be maximal, and (3.52) must be an eigenfunction
belonging to j = jl + j2 - 1. Repeated applications of J- to this function,
(3.52), results indeed in 2 (jl + j2 - 1) + 1 orthogonal functions, corresponding
to all the attainable m values.
Two of the three states belonging to m = jl + j2 - 2 are thus accounted
for, and the third can now be used to construct the set of functions belonging
toj = jl + j2 - 2. This procedure may be continued until all (2jl + 1)·(2j2 + 1)
initial states are exhausted. Such a stage is reached whenj = UI - jo I, which is
therefore the minimum magnitude of the compound angular momentum, just
as jl + j2 is the maximum magnitude. The proof of this so-called "triangle in-
equality" is given in Problem 3.2 and by an operator method in (3.75, 76).

Problem 3.2. Show that the procedure indicated in the text does in fact use every
one of the initial functions, by proving the identity

+ 1) = + 1)·(2~h + 1).
}~,
2.; (2j (2jl
}=Ih-}zl

Having constructed the complete set of eigenfunctions of J2 and JZ from the

set of eigenfunctions of J{ and Ji, one may formalize this procedure somewhat.
The two sets of orthonormal states are related by a canonical (unitary) trans-
formation, that is, we can express the new wavefunctions lihi2jm) in terms of
the old by

Ud2jm) = ~ Ulmd2m2)(jlmd2m2Uddm)
mimi

where the matrix elements of the transformation operator

(3.53)

are the vector-coupling, or Clebsch-Gordan coefficients. A somewhat symmetric

form for these coefficients was derived first by Wigner using group-theoretical
methods, and later by Racah and by Schwinger. 4 (Later in this chapter there is
an introduction to Schwinger's operator technique in angular momentum.) We
quote the result:

(jlmd2 m2Iiddm) = om.+m"m

.[(2
j + 1)(j1 +j2- j)!(jl- j2+ j)!( - iI +j2+ j)!(jI+ml)!(jI-ml)!(j2+ m2)!
(iI+j2+j+l)!

4 For references to Wigner's and other work, and for detailed formulas see [3.1].
 3.7 Equations of Motion of Interacting Angular Momenta 81

+ m)!(j - m)!J
IIZ
·(jZ - mZ)!(j
• ~(-I)Z[z!UI + jz - j - Z)!UI - ml - Z)!UZ + mZ - Z)!
Z

(3.54)

This unwieldy formula is evaluated in a form useful in problems involving

angular momenta! or 1 in Table 3.2.

Table 3.2. Nonvanishing Clebsch-Gordan coefficients for iz = ! and 1

.
} =1J
. +.1
2 +jf' + m +! +ji' - m + t
2i, + 1 2i, + 1
. .
} =1J -
.1.
2 _ji' - m + t +ji' +m+t
2i, + 1 2i,+1

+ m)(J, + m + 1) +jU, -m+ l)U, +m+ 1) +jU, - m)U, - m + I)

i, + 1 +jU,(2j, + 1)(2i, + 1) (2j, + I)U, + 1) (2i, + 1)(2i, + 1)
_j(j, + m)(j, - m + 1) m +jU, - m)(j, + m + 1)
2i,(j, + 1) + .jj,(j, + 1) 2j,U, + 1)
j, - 1 +jU, - m)(;, - m + 1) _jU, - m)U, + m) +jU, + m + 1)U, + m)
2it(2j, + 1) it(2j, + 1) 2it(2j, + 1)

For the purposes of nuclear and atomic physics, a symmetrized form of these
coefficients, the Wigner 3-j symbols, is simpler to manipulate. However, it does
not serve any purpose to introduce them here, and we refer the reader to
Edmonds' book. [3.1].
We shall return to the subject of compound angular momentum, after
introducing an operator technique for expressing the angular momenta.

3.7 Equations of Motion of Interacting Angular Momenta

An interaction Hamiltonian involving two or more angular momenta (e.g.,

spin-orbit coupling of a valence electron in an atom) must be free of unwar-
82 3. Quantum Theory of Angular Momentum

ranted dependence on an arbitrary coordinate system, so must be a scalar. For

two angular momenta, the Hamiltonian must therefore be a function only of
J 1 • J 2 • The Heisenberg exchange operator, Eq. (2.2), is the simplest example of
this. The time development of all other operators are determined by the Hamil-
tonian, through the equations of motion (Poisson bracket equations of motion
classically, commutator bracket equations, quantum mechanically), therefore
we are interested in the commutation relations of the components J b , J ZY , ••• , J Z6
with the bilinear scalar. They are, compactly

(3.55)

Permutation of the two operators yields a similar equation for Iz, and the anti-
symmetry of the vector product then establishes that the total angular mo-
mentum operator, II + Iz, commutes with 11·lz. This motivates representa-
tions such as the one introduced in the preceding section.

3.8 Coupled Boson Representation

In semiclassical theories of magnetism it is common to approximate the unwieldy

spin operators or matrices by harmonic-oscillator operators, for the general
matrix structure of the spins, (3.40-43), resembles in many respects the matrix
structure of harmonic oscillator operators. That this is no coincidence has been
proved by Schwinger in his theory of angular momentum [3.5] based on coupled
harmonic-oscillator fields.
As is well known, one may take linear combinations of momentum and
coordinate operators, to obtain harmonic-oscillator "raising" and "lowering"
operators a* and a; for example,

*= 1 (Ii a .)
a ""21i T ax + IX
Let us make clear the reason for this terminology. The commutation relations
which may be derived for the operators a, a* are those of Bosons

(3.56)

with i, j referring to different particles, from which we deduce also

(3.57)

introducing B, = a1a, = occupation-number operator, referring to the degree

of excitation (occupation) of the ith harmonic oscillator. One may construct a
complete set of states, labeled by occupation numbers, first introducing a "vacu-
 3.8 Coupled Boson Representation 83

urn." This is the ground state of all the harmonic oscillators, the no-particle state,
denoted 1 0), which is annihilated by every lowering operator

adO) == 0 for all i. (3.58)

The one-particle states are clearly

ar 10)
and the (normalized) two-particle states are

araf 10) or <Jr 10),

etc. The general formula for the many-particle, normalized state is

(annl(an n, '" 10)

(3.59)
,.)nl !n2! ...

as proved in Problem 3.3. In these states, the occupation-humber operators nl

are diagonal, with non-negative integer eigenvalues n l = 0, 1,2, .... The total
occupation of a state is given by the sum of the eigenvalues,

Problem 3.3. Prove that the state in Eq. (3.59) is normalized by evaluating (01 ...
(a 2 )"'(altl(at)nt(af)n, ... 10), using only the definition of the vacuum, (3.58), and
the commutation relations of (3.56, 57). Show that n/ is the eigenvalue of n/.

Schwinger showed that with the aid of only two harmonic oscillators the entire
matrix structure of a single angular momentum, as summarized in (3.40-43),
could be exactly reproduced. The advantages are great, even if no new results
were obtained, most especially for large values of j for which the angular-
momentum matrices are large and unwieldy. New features can also be studied
with Schwinger's approach, due to the great understanding which has been
achieved concerning the harmonic-oscillator fields as compared to the more
obscure angular momentum-operators. Labeling the two oscillators by sub-
scripts 1 and 2, let us introduce the spinor operators

a+ = (ar, an and (3.60)

which are merely two-component vectors with operator components. If, more-
over, we contract these operators with the Pauli spin matrices, we obtain the
desired representation. That is, let
84 3. Quantum Theory of Angular Momentum

and similarly for the other components, with the following compact result:

(3.61)

One-half times the contraction with the unit operator will be denoted the j
operator

(3.62)

It may be verified that the eigenvalues of this operator are indeed j = 0, i, 1,

j., ... , where J2 has eigenvalue fj2j(j + 1).

Problem 3.4. Show that J+ = fjara 2 and J- = fjafa l • (These formulas and the
ones in the text may be remembered by associating a change 11m = + i with
each type-I particle, and 11m = - i with each type-2 particle.)
Problem 3.5. Prove the operator identity

J2 = fj2j(j + 1).
Problem 3.6. Quantization of the electric and magnetic fields of light waves is
the first step in quantum electrodynamic theory. Assuming these are harmonic-
oscillator fields, let subscript 1 refer to the electric field and subscript 2 to the
magnetic field in operators ai' af and a2 , af (e.g., for a particular plane-wave
mode). Use the fact that half the total energy must be contained in each field to
prove that the eigenvalues j can only take on integer values, thus showing that
the photon has quantized spin of unity. Compare with (3.103, 104).

Normalized eigenfunctions of a single angular momentum, possessing definite

m and j eigenvalues shall be denoted Ijm), and are simply

'm = (af)J+m(an J- m 0
11) ""'(J + m)!(j - m)! I ).
(3.63)

The coupled-Boson operators have more flexibility than the original angular-
momentum operators. For example, we can make use of the extra degrees offree-
dom to construct the so-called hyperbolic operators which conserve m, but
raise or lower j

and KZ = ~ (nl + n2 + 1). (3.64)

They obey the commutation relations,

 3.9 Rotations 85

and
[K+, K-] = -2I1Kz. (3.65)

Only the last of these differs, by a sign, from the commutation relations of the
angularmomentum operators. The following equations may also be verified:

(Jz)2 - J..-1I2
4
= (Kz)2 - J..-(K+ K-
2
+ K-K+) (3.66a)

= KZ(Kz - II) - K+K- (3.66b)

= KZ(KZ + II) - K-K+. (3.66c)

3.9 Rotations

The study of rotations of coordinate systems and of particles is intimately tied

in with the theory of angular momentum. For example, given an arbitrary
function/(?) of the azimuthal angle ?, its argument may be rotated through an
angle a by applying a differential operator

This formula may be checked by a Taylor series expansion of both sides, in

powers of a. The operator which is exponentiated will be recognized as pro-
portional to L', and is just one of the three differential operators, or linear
combinations thereof, which may be exponentiated to effect desired rotations in
the coordinate system. The most general rotation is expressible in terms of the
three Euler angles a, fl, and y, and takes the form of the unitary operator

(3.67)

One writes J instead of L because the rotations are not limited to the integer an-
gular momenta. Of course, the three-Euler-angle formula is equivalent to a single
rotation about a suitably chosen axis, with its direction along a unit vector ii,
and so

D{afly) = eialN (3.68)

for a suitable angle a' . Thus, two or more successive rotations can be expressed
by a single one. One of the practical consequences of this equality, after the
rotations are expressed in terms of a complete set of spherical harmonics (for
integer angular momentum), is a very useful formula for expressing products of
spherical harmonics as a linear combination of spherical harmonics, viz.,
86 3. Quantum Theory of Angular Momentum

(3.69)

We shall not prove this formula here, but proceed instead to show in what
way the rotations are related to the unitary transformations of Boson operators.
The transformation which takes a operators into a linear combination of each
other is a unitary canonical transformation if it preserves the Boson commu-
tation relations, and the Hermitean nature of Hermitean operators. If it also
preserves the eigenvalue j (that is, commutes with j), it corresponds precisely to
a rotation. In particular,

aT - [e+(i/2) (a+y) cos i + [e(i/2)

Plat (y-a) sin i p]a!
a! - - [e(i/2) (a-y)sin t PJaT + [e(i/2) (a+y) cos t p]a!
al - [e- 2) (a+y)
U/ cos i PJal + [e-(i/2) (y-a) sin t p]a2 (3.70)
a2 - - [e- U / 2) (a-y) sin t p]a l + [e-<il2) (a+y) cos t p]a2

is the transformation which results if in (3.67) we express the components of J

in the coupled-Boson representation, and calculate

For complex angles, this is no longer a unitary but a similarity transformation,

which preserves the commutation relations but not Hermiticity. One may also
imagine unitary or similarity transformations which do not conservej, and which
mix the a operators with a* operators. Such transformations mix the angular-
momentum operators with the hyperbolic ones and are generated by including
the components of K as well as of J in D.

3.10 More on Compound Angular Momentum

The addition of the spin of an electron to its mechanical angular momentum,

the addition of the angular momenta or spins of two distinct electrons, and in
general the coupling oftwo or more angular momenta of various origins, require
mathematical techniques beyond the simple formulation given so far. Here we
shall show some of the machinery established by Schwinger for handling these
problems, without, however, entering into the details. We shall study the
coupling of two angular momenta, and it might appear that by induction one
may use this theory to couple an arbitrary number of angular momenta. How-
ever, in practice this is not quite so, and even for three angular momenta cer-
tain additional simplifications must be sought if the problem is to remain
manageable.
 3.10 More on Compound Angular Momentum 87

Two angular momenta require four Bose particles for their description. But
with four Boson operators, far more than the components of II and 12 can be
constructed: we can obtain I = II + 12 , and also all the relevant hyperbolic
operators. The Clebsch-Gordan coefficients, rotations in one or the other
angular-momentum space, and all the other possible subjects of interest also may
be investigated by studying the Bose operators and their eigenfunctions.
Let the Bose operators referring to angular momentum 1 be labeled al and a2,
and those referring to angular momentum 2 be labeled bl and b2 • Any a operator
commutes with any b operator. The notation and terminology will be that
introduced in (3.56-63). This takes care of the usual operators, the components
of J t , and J;. In addition, the following new ones are required:

1+ = lia+b where (3.71)

(3.72)

etc., in the spinor notation, and

The following relations may be verified by substituting the definitions given

above:

J2 + J; + J;
= ]1;,
= [z(/' _ Ii) + 1+[- (3.74a)
= IZ(/' + Ii) + 1-1+ (3.74b)
= KZ(KZ - Ii) - K+ K- (3.74c)
= KZ(Kz + Ii) - K-K+. (3.74d)

Using the definition of Iz (3.71), and the fact that 1+[- in (3.74a) is a positive
semidefinite operator (so is 1-1+), one proves directly thatj(j + 1) S; UI - j2)
UI - j2 - 1). The choice jl S; j2 [if j2 < jh then use (3. 74b)] establishes
(3.75)

Similarly, the two equations involving K can be used to prove

(3.76)

and thus to establish the "triangle inequality" first introduced in the discussion
following (3.52) and in Problem 3.2.
A state of definite eigenvalues jlml and j2m2 is

(3.77)
88 3. Quantum Theory of Angular Momentum

A state of definite j, m, j. and j2 may equally well be labeled j, m, II, and v, where

and (3.78)

and denoted

(3.79)

where II ::;; j and v 2 j + I. For m = j and v = j + 1, the appropriate state is

(a *)l+P(b*)l-P
... +
IJ]IlJ 1) = • •
,JU+Il)!(j-Il)!
10). (3.80)

Application of the spin-lowering operator yields the states Ijmllj + 1), and
finally, arbitrary states are given by

Ijmllv) = / ~2j + 1).! (K+)V-J-1ljm lli + 1). (3.81)

'Y (v + J)!(v - J - 1)!

The inner product of these wavefunctions with the Ij.m.j2m2) of (3. 77) yields the
Clebsch-Gordan coefficients by Schwinger's method.
The Bose operator method may be extended in various directions to recover
well-known results (e.g., for the spherical harmonics) or to discover new ones.
A connection with the continuous representations may be made using the
operator identity

(3.82)

which may be used to construct generating functions for various sets of ortho-
gonal polynomials. But this subject is properly outside the scope of the present
elementary treatment.

3.11 Other Representations

The first formal justification of the Bloch theory of spin waves may be found in
the work of Holstein and Primakoff[3.6]. Bloch had naturally assumed that spin
waves obey Bose-Einstein statistics, but these authors showed how spin opera-
tors could be expressed in terms of true Bose fields, much as we shall show it in
the chapter on spinwave theory. The Holstein-Primakoff representation is best
understood as a special case of the Schwinger coupled-Boson representation,
that is, as an irreducible representation of the latter in a subspace of fixed j.
Recall (3.62), which we rewrite as
 3.11 Other Representations 89

In a subspace of fixed eigenvalue j, this equation is solved by treating a2 and its

conjugate as two diagonal operators,

a2 = a! = (2j) 1 12.J I - ~ hence, (3.84)

J+ = lia*(2j)1/2.J1 - ~ J- = 1i(2j)1/2.JI - ~a and

l ' = Ii(n - j) (3.85)

omitting the subscript (I) on a and n. The formalism is incorrect whenever n

exceeds 2j, but within the allowed range, it may be verified that the basic commu-
tation relations (3.12-16) are correctly obeyed. (The rationalization of the
square root is discussed, following (3.101) in the section on "spins one.")
If all that one requires is the satisfaction of these commutation laws, and if
one relaxes the requirement that J+ and J- be Hermitean conjugate operators,
then he may perform the so-called Maleev similarity transformation to a new set
of operators

J+ = a*(2j) 1 12 ( I - ~)Ii J- = (2j) 1/2ali and l' = (n - j)li. (3.86)

The obvious advantage, the rationalization of the square root, is somewhat offset
by the complications introduced by the non unitary transformation.
A variant of the above is due to Villain [3.7]. Assuming I 2 lIz I , he writes
I± as

and (3.87)

from which (3.15d, 40-42) flow easily. The correct commutation relations are
ensured by requiring, as in (3.7), or (3.42),

[tp, Jz] = iii, i.e., tp = Iii a/aJz or Jz = -ilia/atp. (3.88)

Thus, exp (-ixtp) is a translation operator for Jz and

(3.89)

In the semi-classical limit, one just drops the terms ±±Ii in (3.87). The extreme
quantum limit j = ! or I deserves separate investigation, and will be discussed
in the following sections.
90 3. Quantum Theory of Angular Momentum

3.12 Spins One-Half

For spins one-half, n = 0, I in the preceding and the Holstein-Primakoff and

Maleev representations are identical within the physically allowed subspace.
Problems involving N interacting spins one-half may be more easily for-
mulated with the aid of one of several Fermion representations. This takes us
back a number of years, to the historic paper of Jordan and Wigner on second
quantization [3.8] in which the Fermion antic om muting operators were explicitly
constructed out of Pauli spin matrices. It is, however, the converse which is
of present interest.
Fermion operators, which we denote by the letter c, are a set of anticommut-
ing operators; for example, for any state 14»,

cTc~ I 4» = -c!cT I 4» (3.90)

and therefore, care must be taken of the order in which the operators are
written, see Chap. 4 for further discussion. The vacuum is defined, just as for
spins,. as the state annihilated by all c's,

(i = I, ... , N) (3.91)

and is the state in which the particle number operators

(3.92)

all have zero eigenvalue.

The anticommutation relations are indicated by curly brackets and are

{c~,cn =0 (3.93)

Setting i = j in the above equations yields the relations

(3.94)

which are identically the equations obeyed by the Pauli spin matrices a±. It is
"*
only the fact that Pauli spin matrices referring to different particles (i j) com-
mute with one another, which differentiates them from the anticommuting
Fermion operators above. This may be remedied by introducing a set of drone
operators dt and d~, equal in number to the original set of c's, which anticom-
mute with the latter and obey amongst themselves anticommutation relations
entirely analogous to (3.93). As a consequence,

and {dt + d~, d, + dn = 0 for i "* j (3.95)

so that finally
 3.12 Spins One-Half 91

is the desired set of spin one-half operators, which commute when i =1= j.
-_ .... _-_ .. -_._--------_ .... _--_ ............. __ ..................................__ ........... __ .. _-_._--_ .. --.. --..... _-- ... --- ... __ .__ .__ .................. .
Problem 3.7. Prove that the various representations in this section obey (3.12-
16), and also that different spins commute:

[S~, Sf] =0 for i =1= j

for all components k, k' = x, y, z of the spin vectors.

A second useful representation is the exact analogue of the coupled-Boson

picture, but replaces the Bosons by Fermions. The paired Fermions, instead of
being labeled 1,2, will be labeled by t and ,J, to make more explicit the role of
each operator. Thus,

and (3.97)

where the anticommutation relations are

This is a very important representation, corresponding to the second quantiza-

tion of electrons cum spin and we shall return to it in Chap. 6.
A final representation of the spins one-half brings us closest to the work of
Jordan and Wigner. The technique which we shall discuss has been useful in the
solution of one-dimensional problems, and also in the two-dimensional Ising
model, an important subject in the statistical mechanics of magnetism. There-
fore, last but not least, set

st = IlcJ'Q, Si = IlQlc , and S: = ~ (2cJ'c, - 1), (3.99)

where
Q, = QJ' = Qil = exp (in ~ c!cJ) (3.100a)
J<i
= exp(in ~ stsj) (3.100b)
J<I
= II (c!
J<I
+ cJ)(c! - cJ), (3.100c)

etc. In actual problems, the choice of ordering i = 1, ... , N is crucial, because the
phase factors Q, introduce great complexity into a problem unless means are
found to eliminate them. When this is possible, however, then the representation
above is the simplest of all, because it is the only one to establish a one-to-one
correspondence between spins and Fermions, and their respective eigenstates.
92 3. Quantum Theory of Angular Momentum

3.13 Spins One

There is no special representation for spins one, except for that which deals
specifically with vector fields, such as the electromagnetic field. However, the
Holstein-Primakoff representation may be used after rationalizing the square
root. For, in the physically admissible range n = 0, 1,2, the equation

.Jl - n/2 == 1- (~ - .JT) n- ~ (.JT - 1) n2 (3.101)

is exact, and only fails for n ~ 3, which is outside the domain of validity of this
particular representation. In fact for any j, a polynomial of order 2j can be
found which has the same structure as the Holstein-Primakoff root, but is quite
tedious to construct for large j. Because the Taylor series expansion of the root,

becomes asymptotically correct in the limit j = 00, it is commonly used in

approximate theories.
We have not dealt with vector fields, except in Problem 3.6, where it was sug-
gested that the "intrinsic spin" ofthe photon was unity. There are good reasons
why this should be so. When angular momentum generates infinitesimal rota-
tions of the coordinates, it changes not only the arguments of a vector field, as
it does a scalar field (e.g., ordinary wavefunctions), but also mixes the various
components of the field amongst themselves. The total effect might be described
by an operator

(3.102)

of which the first term takes care of arguments, and the second of the rotations
of the vector field components. Similar expression for the two other Cartesian
components results in an operator J = L + S. Recall that J can be a true angu-
lar momentum only if Land S each are, in their own right. Therefore, let us
investigate this new angular-momentum operator,

S = ill(ii" X ,iiy X , ii. x) (3.103)

specifically constructed to operate on vector fields, such as the vector potential

A(r). Calculating S2 = S~ + S; + S:, we readily find

(3.104)

and therefore the spin magnitude s = 1.

 3.14 Quadratic Forms 93

Field amplitudes may then be expanded in eigenfunctions of Sz, the spheri-

cal unit eigenvectors,

and (3.105)

with eigenvalues,

r=-I,O,1. (3.106)

These are particularly useful in problems with some spherical symmetry, e.

g., the field of a radiating atom. Linear combinations of functions having definite
J and total azimuthal quantum number M are called vector spherical harmonics

'YJ/M(O, rf» = Lmr Y1m(O, rf»e,(lmlr IfUM). (3.107)

In addition to J and M, only the f of the spherical harmonic need be indicated,

for the spherical unit vectors invariably have unit angular momentum, as we
have shown.
The vector spherical harmonics form a complete, orthonormal set for the
description of vector fields. The normalization integral, including scalar product,
is
21t It

f drf> f dO sin O'YfIm(O, rf»·'YJflfMf(O, rf» = JJJfJIl'JMMf.

o _It
(3.108)

3.14 Quadratic Forms

We have already discussed the scalar product of two angular momentum opera-
tors, say SI·S2. We now obtain the eigenvalues of this operator, and find them
to be parameterized only by invariants.

We start by completing the square

(3.109)

and replace S1 by its magnitude SI(SI + I), S~ by SlS2 + 1), and SI + S2 by the
compound operator Stot, obtaining

(3.110)

with Stot taking on any possible integer value ranging from lSI - s21 to a maxi-
mum (SI + S2). Substituting in the above we find for the maximum and mini-
mum values of SI . S2
94 3. Quantum Theory of Angular Momentum

(3.111)

assuming S2 S; s,. The maximum, occurring when both spins are parallel, could
have been predicted classically. The minimum, corresponding to antiparallel
orientation, is lower by a fraction lis, than the classical estimate; this foresha-
dows the complications of antiferromagnetism compared to ferromagnetism.
For the special case of spins one-half only, the quadratic form can be dia-
gonalized by a nonlinear transformation [3.9]. We introduce two new spin one-
half operators denoted J and P, constructed out of the components of S, and
S2 in the following manner (Ii = 1):

(J", Jy, Jz) = (S,,,, 2S,yS2zo 2S'zS2z), and

(P", Py, Pz) = (S2" 2S,,,S2y, - 2S,,,S2,,). (3.112)

The inverse of these relations is

(Six' Sly, S'z) = (l", 2JyP", 2JzP,,) and

(S2", S2y, S2z) = (-2PzJ", 2PyJ",P,,). (3.113)

In the new representation, the quadratic form is diagonal - Ising-like -

(3.114)

where Jz = ± -! and P z = ± f in the units Ii = 1. For some problems this

formulation may be superior to (3.110).
4. Many-Electron Wavefunctions

Although the old quantum theory of Niels Bohr had great success in the interpre-
tation of the line spectra of the hydrogen atom, it did not provide an adequate
framework for the interpretation of the complex spectra of the many-electron
atoms and molecules. Nevertheless the old theory had many clever proponents,
who with a series of particularly shrewd guesses nailed down the structure of the
periodic table and the theory of line spectra some time before the new wave
mechanics. Notable was the exclusion principle of Pauli, which included a fourth
quantum number in addition to the three which are associated with the orbital
motion of each electron [4.1]. This was the key to explaining why there are two
electrons in an s shell, 6 in p shell, etc., and toward a correct theory of the build-
ing-up principle of the periodic table. Almost immediately thereafter, Uhlenbeck
and Goudsmit [4.2] published their evidence that the fourth quantum number
referred to the spin of the electron, a concept that helped clarify much of the
subsequent thinking on the subject. Indeed so much was understood on the
basis of the old theory alone, that it might even appear that the 1926 theories of
Schrodinger and Heisenberg, far from solving old problems, only raised new
difficulties. The reason is that it proved so awkward and difficult to introduce
spin into the new quantum mechanics.
Once it was established that the probability density is equal to 11JT12 the
square of the wavefunction, then the following paradox imposed itself: The
probability function for N indistinguishable particles must be invariant under
any permutation of the particles. Moreover, as any permutation can be achieved
by a succession of transpositions (permutations of two particles at a time), one
need only investigate the effects of transpositions. The wavefunction, it is then
found, can be allowed to be even or odd under a transpotion, but no other choice
leads to the correct, totally symmetric probability density. (For two electrons,
hydrogen molecule or helium atom, it was recognized from the first that odd
functions describe triplet states, and even functions the singlet states; the agree-
ment of the calculated energy levels with experiment was extraordinary. On the
other hand, the energy levels of the even or odd functions of three or more elec-
trons did not agree with spectroscopic data, or even with elementary notions of
how atoms were constituted. Surely doubts must have been raised regarding
whether quantum mechanics was applicable to more than two electrons even in
principle.) But the experimental spectra of systems of more than two electrons
could only be understood in terms of wavefunctions which were neither odd nor
even under arbitrary transpositions, but, rather, which had complicated proper-
96 4. Many-Electron Wavefunctions

ties under the various permutations. This was totally incompatible with the no-
tion of a probability density symmetric under the interchange of identical
particles. Nor was it obvious to what values of the total spin, i.e., to what
multiplicity these complicated space functions were associated. One can readily
guess that the totally anti symmetric function corresponds to the state of maxi-
mum multiplicity, all spins parallel; but before the invention by Pauli of the spin
matrix operators [4.3] the calculation of general symmetry properties ofthe space
functions corresponding to arbitrary multiplicity posed many such logical and
practical problems.
A remedy was provided by Slater, with his invention of the determinantal
wavefunction of space and spin [4.4]. Spin was an observable; therefore, the
spin coordinate belonged in the wavefunction on the same footing as the space
coordinate. The exclusion principle could now be correctly stated as follows:
"A wavefunction for a set of identical electrons must be antisymmetric under the
interchange of any two of them; by which is meant, the interchange of the space
+ spin coordinates of the two electrons."
We shall examine the Slater determinants, insofar as they shed further light
on the molecules of two and three hydrogen atoms we have studied so far. We
shall then progress in reverse, it might seem, and ask to which functions of space
alone, they correspond for the various multiplicities. We shall not now have
any difficulty in finding nor in reconciling these with the antisymmetrization
principle (supra). This is in contrast with the elaborate theory which was devel-
oped for this purpose before spin was finally understood.
We then put this knowledge to use, proving a theorem due to Lieb and Mattis
[4.5] which seemingly contradicts Hund's rule that states of highest multiplicity
lie lowest. Fortunately this paradox is also soon removed, for the theorem and
the rule are found to have distinct domains of applicability. In the reconciliation
of opposites, new light is shed on the theory of ferromagnetism. Naturally, this
eventually leads us back to the interesting and central subject of "exchange,"
thence to manybody theory.
It is essential that the reader have familiarity with some notation and
theorems of angular momentum provided by the previous chapter. Beyond
this, we develop the many-electron theory up to, and including second quanti-
zation in this rather far-ranging chapter.

4.1 Slater Determinants

It might be well to clarify the notion of "spin function" by analyzing the Hilbert
space of ordinary space functions and bringing out the similarities. The student
will not be surprised, then, to see space and spin treated on equal footing in the
determinantal wavefunction, and will obtain insight into the possibilities and
limitations of the form. For a more precise notion of meaning of spin and spin
 4.1 Slater Determinants 97

coordinates, the reader may wish to consult the preceding chapter on angular
momentum, particularly the first five sections. (See also [4.3]).
Consider a complete set of functions, assumed orthonormal,

h(r),fz(r), ... ,f,,(r), ... (4.1)

in which an arbitrary function g(r) can be expanded,

g(r) = I; gnf,,(r) (4.2)

n

by analogy with a Fourier expansion. The expansion coefficients gn play the role
of Fourier coefficients (or Fourier transform). The function g(r) can be re-
presented as a vector in Hilbert space,

g = (gh ... , gn ... ). (4.3)

The overlap integral,

c == Jg*(r )h(r )dr (4.4)

is then the scalar dot product between two vectors

C = I; g:hn = goh. (4.5)

n

Further discussion is best left to mathematical treatises. However, recall that it

is just this strong analogy which unites the matrix mechanics of Heisenberg and
the wave mechanics of Schrodinger.
In the space of the Pauli spin matrices, the basis vectors are the spinors

(1,0) and (0, 1). (4.6)

Similarly, in the functional Hilbert space, the basis vectors are

(1,0, ... ,0, ... ,0, ... ), (0,1,0, ... ,0, ... ), ... (4.7)

etc., as we discover by writing the functions of (4.1) in the vector notation of

(4.3). The spaces spanned by (4.6, 7) may have different dimensionality, but the
vector analysis is similar. The apparent difference between continuous and dis-
crete variables as in fact been removed.
But one can equally well reverse the process and associate two orthonormal
functions to the two basis vectors of (4.6), for example,

to (1,0) spin "up" (4.8)

98 4. Many-Electron Wavefunctions

X-(~) to (0, 1) spin "down". (4.9)

The X's have properties of ordinary functions. If we wished ~ to be a discrete

two-valued variable, we could allow the two functions to become delta func-
tions in some appropriate way, and still formally preserve the analogy with the
functions f.(r).
If now the f.(r) functions describe electronic states, a product function
describing three particles for instance, might be

(4.10)

with each of r, s, t, assuming the value +

or -. This is not an allowable wave-
function for identical particles since the probability density

P(l, 2, 3) == 'P* 'P (4.11)

is not invariant under permutations of the three particles, except if

j=k=n and r=s=t (4.12)

an equality which would not have been allowed for electrons according to the
Pauli exclusion principle, in the old quantum theory. In the new quantum theory,
moreover, we are supposed to consider not 'P(l, 2, 3), but the antisymmetrized
version,

'Pil, 2, 3) = 'P(l, 2,3) + 'P(2, 3,1) + 'P(3, 1,2)

- 'P(2, 1, 3) - 'P(3, 2, 1) - 'P(l, 3, 2). (4.13)

If a product function violated the old Pauli principle, such as in (4.12), it would
also automatically be projected out by the anti symmetrization procedure, for
as we shall see, 'PA would vanish identically. This shows that in the new me-
chanics, electrons cannot violate the Pauli principle even in an arbitrary state-
let alone in an eigenstate described by good quantum numbers.
To extend the procedure to N particles would require writing out N! terms.
But following Slater, we recognize in (4.13) the rules for writing out a deter-
minant, i.e.,

jj(rl)Xr(~I)

'PASI'ter = ",,3!
} det I' ( )
lk
;:)
rl,x.(.,1

f.(rl)X,(~I) (4.14)

The determinant is conventionally divided by ,J3l for purposes of normali-

zation. This ensures that in addition to being totally symmetric, the probability
 4.2 Antisymmetrization 99

distribution of (4.11) calculated with Slater determinants of (4.14) is normalized

to unity, ifthe/(r) and X(~) functions are members of an orthonormal set.
It is by means of such determinants as (4.14) that one forms a complete,
orthonormal set of states for three electrons, and not by means of the product
functions of (4.10). The number of such linearly independent determinants is
always far less than the number of product functions, as all states which cannot
be anti symmetrized are "excluded."

4.2 Antisymmetrization

A simple rule may be helpful to understand the process of anti symmetrization

called for by the Pauli principle.
Assume a function of N variables and all N! of its permutations:

l(l, 2, 3, ... , N),f(2, 1, ... , N),f(N, 3, 1, ... ,2), ... (4.15)

,vhere in the self-evident notation, j is allowed to stand for the jth variable.
To construct a totally anti symmetric function using thef's, there is a systematic
procedure:
Define a function anti symmetric in variables I and 2,

l(l, 2/3, ... ) = l(l, 2, 3, ... ) - 1(2, 1,3, ... ) (4.16)

with the bar (/) to signify that the function is anti symmetrized with respect to the
variables preceding it. Then by induction, we construct a function anti symmetric
in variables 1, 2, and 3,

l(l, 2, 3/4, ... ) = l(l, 2/3, ... ) - l(l, 3/2, ... ) - 1(3,2/1, ... ) (4.17)

In general,
p-l
l(l, ... ,p/p + 1, ... ) =1' - L: .9"pl'
,~l
(4.18)

where
I' =/(1, .. · ,p - l/p, ... )
and .9"p transposes the jth and pth variables. The procedure can be continued
until p = N if desired.
By similar procedures totally symmetric functions, suitable for Bosons, can
be constructed. It is merely necessary to replace all the minus ( -) signs above
by plus (+).

Problem 4.1. Prove by repeated use of (4.18) that the totally anti symmetric func-
tion is given by
100 4. Many-Electron Wavefunctions

I: (-IY~/(1, 2, ... ,N) == 1(1,2, ... , N/)

the sum being over all N! distinct permutations of the N variables, and the ex-
ponent P = 0 for even permutations, + I for odd permutations.

By taking other combinations of permutations, it is possible to generate func-

tions with more complicated transformation properties under permutations of
the various variables. But fortunately we do not need these; as we have already
mentioned, wavefunctions of Fermions must be totally antisymmetric, and of
Bosons totally symmetric, provided, of course, that we are sure to include the
spin variable as well as any other coordinates (e.g., isotopic spin, band index, or
whatever else is required in the complete description of the particle) in the
wavefunction.
Starting with a product function of space and spin coordinates of N electrons,
say 1(l)g(2) ... heN), the antisymmetrization procedure gnerates N! terms,
which have the same sign (+ or - ) as the corresponding terms in the expansion
of a determinant in whichf(l)g(2) ... is the product along the principal diagonal
of the array. If the original product function is normalized, then the determinant
is conventionally normalized by dividing it by --.IN!. But this does not in fact
normalize it unless the various permutations are all orthogonal, as when the one-
electron functions in the original product are orthogonal. (We previously en-
countered problems in nonorthogonality in attempting to extend the Heitler-
London scheme to the N-body problem.)

4.3 States of Three Electrons

Let us postpone the trivial problem of two electrons, and study three electrons
by means of Slater determinants. The most general single determinant is

qJb(r1)xRl) qJb(r2)xR2) qJb(r3)xR3)

qJe(rl)Xt(~I) qJe(r2)Xt(~2) qJcCr3)Xt(~3) (4.19)

aside from a normalization constant. For fixed qJa, qJb, qJe, the two possibilities for
each spin function result in eight distinct determinantal functions in total. If
some of the space functions are not linearly independent, then the number of
independent determinantal functions could be less than eight, and in the
example of (4.12), j = k = n, r = s = t, the number of possible determinantal
functions was zero. But in no case does it exceed eight for three particles. Note
that if there were no exclusion principle, and no requirement of antisymme-
trization, the number of allowed product configurations of the type in (4.10)
would be 6 X 6 X 6 = 216 in the present case. So, a considerable number of
states have been pruned.
 4.3 States of Three Electrons 101

There are two instances of all spins "up" or "down" simultaneously,

qJb(rl)XT(¢I) qJb(r2)XT(¢2) qJb(r 3)XT(¢3)

IPC(rl)XT(¢I) qJk2)XT(¢2) IPk3)XT(¢3) (4.20)

r = + or -. The remaining six determinants are, in turn,

qJa(1)XT(1) IPi2)XT(2) IPa(3)XT{3)

IPb(1 )X.(1) IPb(2)X.(2) qJb(3)X.(3)
qJc(1 )X.(1) qJc(2)X.(2) IPc(3)X.(3) (4.21)

with r = + and s = - , or the converse; and

qJa(1)X.{I) qJa(2)X.(2) IPa(3)X.{3)

qJb(1 )XT(1) qJb(2)XT(2) IPb(3)XT{3)
IPc(1)X.(1) IPc(2)X.(2) IPc(3)X.(3) (4.22)

and finally

IPa(l)X.(1) IPa(2)X.{2) IPa(3)X.(3)

IPb(1 )X.(l) qJb(2)X.{2) IPb(3)X.(3)
IPc{l)XT(1) qJc(2)X/2) qJC(3)XT{3) (4.23)

all with r = + and s = -, or the converse, corresponding to one spin up and

two down, or the converse. Of all these determinants, the spin functions factor
out of only the two quartet states of (4.20). Two more quartet states must
therefore be intimately mixed in with the two sets of doublet states in the deter-
minants of (4.21-23).
We see that the determinanta1 method does not necessarily generate eigen-
functions of S;ot, although by construction all determinants are eigenfunctions
of S~ot"
The doublet states can be projected from the states of (4.21-23) by means
of the obviously constructed projection operator,

(4.24)

which has zero eigenvalue on the quartet states, and eigenvalue unity on the
doublet states. These states are not single Slater determinants, but linear com-
binations of them.
102 4. Many-Electron Wavefunctions

Problem 4.2. Find the eigenfunctions of the projection operator of (4.24) among
the set of determinantal functions of (4.20-23). Expand them in terms of the
space functions "'I> "'2, ... , "'6
of the three-atom problem of Chap. 2.

Trial functions to calculate the low-lying eigenvalues of a Hamiltonian for

three electrons could be made of linear combinations of the eight determinantal
functions. Moreover, if ~ doesn't involve the spins explicitly, we can always
decompose it into noninteracting blocks by using the projection operators D
and I-D. The "quartet," i.e., fourfold degenerate states, Stat = ! belong to the
totally anti symmetric space function. The two sets of doublet states, Stat = t,
give rise to two distinct doubly degenerate doublet levels. (In the case of three
hydrogen atoms which we analyzed previously the two doublets were acciden-
tally degenerate, however, and corresponded to the four solutions of type V 23 •
Another space function which we had constructed (by permutations of space
coordinates alone) was the totally symmetric function, which is not admissible
and does not even enter via the determinantal approach. The sixth function was
the totally antisymmetric quartet state.)

Problem 4.3. Construct the projection operator R which has zero eigenvalue for
triplet states and eigenvalue unity for singlet states of two electrons. Operate with
R on the most general Slater determinant of two electrons, and show that the
singlet state is always symmetric under the interchange of the spatial coordinates
of the two particles, and the triplet antisymmetric.

4.4 Eigenfunctions of Total 8 2 and 8:.

If a function of space and spin coordinates is totally antisymmetric, we shall

denote it a "Pauli function": the Slater determinants are a special case. If in
addition it is an eigenfunction of Si;,t and St~t' then there are severe limitations
on the type of spin and space functions which can be used to construct such a
Pauli function.
Let us assume first that the Pauli function is an eigenfunction of Stat with
eigenvalue M, that is,
(4.25)
indicating the quantum number explicitly as a superscript. Let us now expand
(4.25) in a complete set of spin functions, the product functions
(4.26)
Normally 2N such functions would be needed, but the restriction to a definite M
value means that
M +t N have spin up (+) and - M +! N have spin down (-). (4.27)
 4.4 Eigenfunctions of Total SZ and S'I 103

We can take a typical such function, for example,

MX(¢" ¢2' ... , ¢M+I/2NI¢MH12N+I,"" ¢N) == X+(¢I)

... X+(¢MH/2N) ... X-(¢N) (4.28)

and generate all the others by permutations. The permutations of the spins up
among themselves to not generate a new function, and neither do permutations
of the spins down among themselves. We therefore call these the trivial permu-
tations of the spin function MX(¢" ... 1... , ¢N)' This function is totally symmetric
under the trivial permutations, those of spin coordinates to the left of the vertical
bar amongst themselves, or of the spin coordinates to the right of the bar
amongst themselves.
There are, however, [N!/G-N + M)! G-N - M)!] = nM nontrivial per-
mutations generating nM orthogonal spin functions including the original
function in (4.28), and these comprise the complete set for the expansion of
M",. Note that

H/2N +1/2N N! 2N
2;n- 2; = (4.29)
M=-1/2N M - M=-1/2N( ~N + M)!( ~ N - M)!

the dimensionality of all the M subspaces adding up to the total correct number
2N of orthogonal spin functions.
When M", is expanded in these functions, the coefficient of M X in the expansion
must be a space function of the type,

(4.30)

which is totally antisymmetric under the trivial permutations of coordinates to

the left of the bar amongst themselves, or of coordinates to the right of the bar
amongst themselves. In general, however, it has complicated transformation
properties under the nM nontrivial permutations (see Problem 4.4).

Problem 4.4 In the case of three hydrogen atoms, there were two states for M =
j, notably V 23 and V;3' Express the nontrivial permutations of each of these two
states in terms of the other, and of the partly symmetric functions V23sy and
V;35Y'

If in addition, the Pauli function is an eigenfunction of Stot> with eigenvalue S,

(4.31)

then we can actually figure out the nontrivial symmetries of the space functions
MIS using only simple notions about angular momentum. We do this systema-
tically, starting with -Sj's, which is the coefficient of the spin function of (4.28)
when M = - S. Recall that a fundamental property of angular momentum is
104 4. Many-Electron Wavefunctions

that IMI ~ S, and therefore it should be impossible to further decrease M.

In more formal terms,

(4.32)

This requires that there be some linear relationship among theffunctions and
their nontrivial permutations, and indeed, by use of (4.28, 30, 32), we find

(4.33)

with

i = 1,2, ... , iN - S
and gl1l} being the transposition operator for the coordinates of the particles i
andj. But this is very reminiscent of(4.18) if we interpret the above as an attempt
to increase the number of coordinates to the right of the vertical bar (in whichf
is totally anti symmetric) by the addition of one member. This attempt results in
zero: the set to the right cannot be further increased, i.e., the bar cannot be
moved to the left.
We can construct the f functions belonging to higher M values, using the
-sF which properly obey the equation above. This is done by repeated appli-
cations of the total-spin raising operator,

unnormalized (4.34)

and leads directly to

with
(4.36)

a function which obeys (4.33).

Thus, the attempt to move the bar to the right does not fail. By further appli-
cation of the spin-raising operator, or of the antisymmetrization rules of (4.18),
we progressively generate -s+2fs, ... , up to -sfs, all expressed as linear combina-
tions of the original function -sfs and its nontrivial permutations. Finally, the
angular-momentum rule

(4.37)
 4.4 Eigenfunctions of Total S2 and SZ 105

gives us an additional set of constraints on the f functions, similar to the first

ones we found supra. But now, these constrains are that the bar cannot be
moved farther to the right:

SfS(rl>'" ,rI/2N+slrI/2N+S+I' ... ,rN)

II2N+S
- 2; .9 1,/fS(rl>"" rIl2N+slrI/2N+S+h'" ,rN) = 0 (4.38)
/=1

with
1
q = TN + S + 1, ... , N.
Given an arbitrary space function, it is always a simple matter to generate a
function of the type MJ, that is, the space partner of the spin eigenfunction of
StZot. We merely apply the rules of (4.18), and anti symmetrize in the first set of
p = iN + M coordinates, then anti symmetrize in the set of the remaining
coordinates. Only those permutations which we have denoted as "trivial" are
used in this process, the others being used to raise or lower M, as we have
already seen. Iffor some reasonfis unsuitable, the process of antisymmetrization
will yield identically zero instead of the desired function. For example : iff is of
the form '10 and we wish to construct a function Mf M *- 0 with it, we get zero
instead (i.e., like trying to anti symmetrize a symmetric function).
There is no correspondingly simple rule for constructing functions ofthe type
Mfs from an arbitrary function. It is necessary to projct out the unwanted com-
ponents belonging to values of the total spin other than S by repeated use of the
spin raising and lowering operators S~t' after first constructing MJ, with M = ±
S. But in some special cases this can be done by inspection, as in the following
example:
Example: A product wavefunction describes noninteracting particles, and also
has its uses in variational approximations, as we have seen. Let us use such a
product space function as the raw material in an example where we construct the
proper space functions corresponding to definite M and S eigenvalues.
Starting with such a product as

(4.39)

we can immediately write the Mffunction as a product of two determinants

1/11 (r1)
1/12(r 1)
Mf= det
106 4. Many-Electron Wavefunctions

x det

(4.40)

which is explicitly antisymmetric in the two sets of variables. Now assume

M ::;; O. (The case M > 0 can be handled by simply interchanging the two
determinants.) Thus the first determinant is smaller than the second. Now, if
all the functions qJh qJ2, ••. , qJI/2N+M which appear in the smaller determinant
are also present in the bigger one, then according to the rules of (4.32, 33), the
anti symmetrization procedure of (4.33) will obviously yield zero, and therefore
the function of (4.40) belongs to the definite S value,

S= IMI. (4.41)

Of course this requires that the various qJJ not be all distinct functions. When the
qJJare all distinct, the construction of a function of definite S is possible but more
tedious, but we shall not further investigate this topic. The various qJJ in each
determinant must be distinct or else Mf will vanish identically. But these can
appear in both determinants, which is merely another expression of Kramers'
degeneracy; for the appearance of a qJJ in the first determinant means there is
an electron spin "up" in orbital qJJ' whereas its appearance in the second deter-
minant means there is an electron spin "down" in that orbital.
This completes the demonstration, of how the Pauli wavefunction could be
totally anti symmetric while the space functions, presumably eigenfunctions of a
space Hamiltonian, could have complex transformation properties under the
permutation group. We have also shown that these transformation properties
are a consequence of the rules concerning total spin angular momentum. We
now proceed to specific applications.

4.5 Ground State of Two Electrons: A Theorem

In many problems of differential equations, the lowest eigenvalue can be shown

to belong to the nodeless solution. Can something analogous be proved in the
many electron problem, and if so, does it have any bearing on the problem of
ferromagnetism?
The answer to both queries is affirmative. If we first examine the atoms and
the periodic table, we find that in helium the ground state is nodeless, and we
draw therefrom important consequences. For heavier atoms, conversely, the
ground state has a certain number of nodes, so as to satisy Hund's rules. These
nodes result in atomic spin and angular momentum in the ground state which,
 4.5 Ground State of Two Electrons: A Theorem 107

as we shall see, apparently violate a two-particle theorem. We shall return to

the extension of the theorem to N particles and its apparent violation in three
dimensions in subsequent sections, and then resolve the paradox.
Consider first the helium atom, slightly generalized to arbitrary potential

(4.42)

The Hamiltonian is assumed and real, invariant under the interchange of the
two (indistinguishable) particles. The eigenfunctions f(rl> r2) of Schrodinger's
equation,

(4.43)

are therefore real, and either even or odd under the interchange of the two
particles. Because of the simplicity of the problem, it is not necessary to label
fwith M and S value: recall merely that the even solutions must be associated
with odd spin function.

(4.44)

belonging to Stot = 0, and the odd solutions with one of the even spin functions
[x+(¢I)X+(¢2)] or [x-(¢I)X-(¢2)] or
1
../2 [x+(¢I)X-(¢2) + X+(¢2)X-(¢I)] (4.45)

belonging to Stot = 1, M = 1, -1, or 0. Then the product is a proper Pauli

wavefunction. We shall make use of the variational theorem, stated in the
following terms: The ground-state energy always lies below the variational
energy except when an exact ground-state eigenfunction is used as a variational
trial wavefunction. Explicitly,

(4.46)

and the strict inequality holds except if the trial functiong(r l , r2 ) obeys the differ-
ential equation

(4.47)

for all values of'l and '2' In that case only,

Evar = E grd state' (4.48)

108 4. Many-Electron Wavefunctions

We need consider only functions g which are real, and either even or odd under
the interchange of'l and '2, as in the case of the eigenfunctions f. We can now
state the following:
Theorem. The ground-state eigenfunction is nodeless, ergo it belongs to Stot = O.
Moreover, if V('h '2) is invariant under rotation of the coordinate axes, i.e.
commutes with the total angular momentum, then in the ground state also L tot
= O. Thus helium in the ground state, for example, has precisely zero total
angular momentum and spin and is totally nonmagnetic [4.5].
Proof. We shall first assume that the ground-state eigenfunction has nodes (that
is, has at least one change of sign), and then show that this hypothesis leads to a
contradiction. Let the ground-state eigenfunction with nodes be /0, belonging to
energy eigenvalue Eo. By hypothesis, Eo = E grd .tat."

Fig. 4.1. The Wavefunctionfo and its absolute value

go = 110 I plotted along a direction normal to a nodal
surface of 10

Consider the nodeless function go == 110 I, the absolute value of10. Is it an

admissible variational wavefunction? If10 is constrained to vanish at 00, or is
square integrable, then the same follows for go. Iflo is odd, or ifit is even, go is
even. Since/o is real, so is go. Therefore it indeed fulfills all the requirements for
a variation trial function.
But go is not itself an eigenfunction, for along the nodal surfacelo(rh r 2) = 0,
it has discontinuous normal derivatives. See sketch in Fig. 4.1. Therefore the
strict inequality of (4.46) is applicable, Evar > E grd .tato'
The variational energy is calculated by either of two methods with identical
results: 1) Observe that go obeys Schrodinger's equation with eigenvalue Eo
everywhere except on the nodal surfaces. Because go itself vanishes on these
surfaces, the variational energy is precisely Eo. 2) By partial integration the
variational integral in (4.46) can be brought into the canonical form, from
which Schrodinger initially derived his wave equation

(4.49)

But Igol2 = 1101 2 everywhere, and 18'01 2 = liol 2 everywhere, letting the dot
 4.6 Hund's Rules 109

over the letter stand for any derivative. [In deriving (4.49), we have neglected
only surface terms at 00, where both go andfo vanish by the boundary condition].
By direct substitution, we now also obtain Evar = Eo. Hence Eo > E grd state'
This result contradicts the hypothesis that Eo = E grd state' iffo has nodes; or
the hypothesis thatfo has nodes, if Eo = E grd state' Alternatively, if it is assumed
that the ground-state wavefunction is nodeless, then its absolute value is just
± the same function, and there results no such contradiction.
Consequently, the ground state is nondegenerate. For if there is more than
one eigenfunction belonging to E grd state' the previous arguments indicate that
each must be nodeless. However, there is only one nodeless eigenfunction of %,
because the eigenfunctions of a Hamiltonian form a complete, orthogonal set
of functions, and two nodeless functions cannot be orthogonal.
If the problem has rotational symmetry, then the eigenfunctions also have
definite total angular momentum. The nodeless function must belong to zero
eigenvalue, because it is not orthogonal to the constant function 1, itself an
eigenfunction of total angular momentum, obviously with zero eigenvalue.
Q.E.D.

Problem 4.5. Prove the following unconventional equation for the ground-state
energy Egrdstate and ground-state wavefunction!grdstate of the helium atom

Find an analogous expression for the ground state of the N-Boson problem
[4.6]. For the ground state of three or more Fermions? Hydrogen atom and
molecule?

The actual atomic structures of helium and also of the heavier atoms are just
as well understood theoretically! as experimentally. The well-known calculations
of Hartree and others in the late 19208 and 1930s, and modern high-speed
electronic computers have reduced atomic calculations to what is almost an
exact science The present theorem has merely served to show that the well-
known ground state properties of helium are not the results of a particular
calculation, but are shared by arbitrary two-Fermion systems subject to arbitrary
spatial potentials.

4.6 Huod's Rules

The Hartree-Fock procedure) allows one to solve the many-electron Schrodinger
equation approximately by supposing each electron to move in a self-consistent,
I Even in the old quantum theory, see, e.g. [4.7], or any modern text on atomic structure or
quantum mechanics.
 110 4. Many-Electron Wavefunctions

averaged potentiaP Numerical solution of the effective one-body Schrodinger

equations is still required, as the self-consistent potentials are not strictly Coul-
lombic so that the eigenfunctions are not obtainable in terms of elementary
functions. Still, the one-electron angular momentum is useful to organize the
electrons into shells, and the periodic structure of the atomic table is understood
to be the consequence of the filling up of successive shells.
We have already examined the first closed shell, that of helium and found
that without net spin or angular momentum, it was magnetically inert. Let us add
two more electrons, to make Be. It may be thought the outer shell obeys an
effective two-electron Hamiltonian, the solutions of which must be orthogona-
lized to the core (helium) states. If the theorem of the preceding section holds
for the two new electrons, then the total spin and angular momentum of Be must
also vanish, which is the case.
Next, we add two more electrons to form C, and we find the first surprise.
Carbon has Stat = 1, and L tot = 1, and the theorem fails completely for the
third pair of electrons! Indeed, if we look further, of the first eighteen elements
to have an even number of electrons (which takes us up to krypton, Z = 36),
only eight have no spin or angular momentum in the ground state; and of these,
half are inert gases. Why does the argument go awry? One might argue that the
theorem breaks down for the following mathematical reason: that orthogonality
to core states is equivalent to nonlocal potential forces, for which the theorem is
inapplicable. But there are more convincing physical explanations. How can one
predict the magnetic properties of an atom in the ground state?
The answers are given by Hund's rules. These laws, originally based on the
abundant spectroscopic evidence, were of course confirmed by complete atomic
calculations. 2 Fortunately, there exists a less exhaustive explanation of Hund's
rules in terms of exchange, or the vector model, of which we shall give some
examples.
First Rule: Electrons occupy the 2 X (21 + 1) states of a shell in such a
manner as to maximize the total spin. This may determine the configuration
uniquely. When it does not, then one appeals to the second rule.
Second Rule: Any ambiguities in the first rule are resolved in favor of the
highest value of L tot • But the second rule is not always a unique prescription
either, although these two rules are adequate to describe grosso modo the mag-
netic properties of the atom. Other rules determine whether the spin and angular
momenta are parallel or not, but it is not necessary for us to become too involved
with atomic structure. For with the exception of the rare earths, angular mo-
mentum is most always quenched in the process of chemical binding or in the
solid state, and only the spin and Hund's first rule survive.

2 This variational extension of Hartree's calculations by Fock, and Slater, has proved capable
of very high accuracy in atomic-structure theory.
 4.6 Hund's Rules 111

Qualitatively, both Hund's rules follow from degenerate-state perturbation

theory. Electrons in the outer shells move in the spherical potential ofthe nucleus
and of the filled inert-gas configurations. To a first approximation, all the
configurations within the outer shell are degenerate. It is the Coulomb repulsion
between these electrons which lifts the degeneracy. Now if any two electrons are
in the same orbital state, the Coulomb repulsion is maximal. These configura-
tions should be avoided, just as the ionized configurations were avoided in the
Heitler-London theory.
Therefore, in a first crude attempt to "prove" Hund's rules, or rather to make
them more plausible without too much work, only those configurations of the
outer shell electrons need be considered in which every electron occupies a
different orbital state. This would imply that whenever a shell contains 21 + 1
electrons, all the degenerate orbital states in that shell are occupied: for each ml
there is a -ml; and therefore presumably L tot = O.
The validity of this simple guess is striking: there appear to be no exceptions
in the periodic table. For example, the elements with a half-filled p shell are N,
P, As, Sb, Bi, and all have vanishing orbital angular momentum; the elements
with half-filled d shell, Cr, Mn, Mo, etc., also obey this rule, etc.
For shells less than half-filled, one speaks of electrons, whereas for shells
more than half-filled, one commonly speaks of "holes." There is indeed a great
similarity between the ground terms of elements on either side of half-filled shells
in the periodic table: with a general exception that the total J value (spin plus
orbital angular momenta) equals IS - L I for electrons and S + L for holes.
This difference is caused by spin-orbit coupling, a relatively small energy re-
sponsible for the phenomenon of "magnetic anisotropy" in the solid state.
To understand Hund's rules, the vector model, and particularly their physical
origins, it is important to work out some examples. This is done in the following
sections, but only for illustrative purposes, and the reader who wishes to become
proficient in the true art of atomic calculation and quantum chemistry is re-
ferred to specialized treatises and review articles.
Hund's rules give only the ground state of the atom. The methods of atomic
calculation which we shall illustrate give the (very important) low-lying excited
states as well, together with their energy, spin, and angular-momentum quantum
numbers. These results are central to the theory of magnetism, for it is impossible
to understand magnetic solids without understanding the magnetic atoms and
the magnetic molecules first.
Now we can return to our earlier question: why doesn't the two-particle
theorem hold for the valence electrons of carbon? Are Hund's rules contradicto-
ry to our theorem? The answer, briefly, is this. The orbital states, in atoms
heavier than helium, must be orthogonal to the (inert) core, and therefore cannot
be nodeless. The theorem, particularly the method of proof we used, just does
not apply to these "excited state" orbitals. On the contrary; if the functions are
already required to suffer a number of nodes, and to pay the ensuing price in
112 4. Many-Electron Wavefunctions

kinetic energy, it is then advantageous for the potential energy to be minimized

in exactly the manner predicted by Hund's rules, as we shall now see by actual
calculation.

4.7 p3 Configuration

First, recall that four atomic quantum numbers specify each electron: the
principal quantum number n = 1,2, ... , the orbital angular momentum I S n
- 1, the magnetic quantum number m, = -I, ... , + I, and the spin eigenvalue
m. = ±i. But in the description of an electron, instead of giving 1= 0, 1,2,
3, ... , it is conventional to use the letters s(1 = 0), p(l = 1), d(l = ~,f(1 = 3),
g(1 = 4), and thence in alphabetical sequence. Therefore, n = 2 and I = 0 is a
2s electron. One does not indicate m, or m., and the number of particles with
given n and I is shown as a superscript. For example, two electrons with n = 3,
I = 2 are denoted 3d 2 ; three such electrons by 3d3 • The m values not being
specified, there are many possible wavefunctions for a given configuration. With
the total angular momentum of the configuration indicated by a capital letter ,
for example, S, P, D, F, ... for L = 0, 1, 2, 3, and the total spin Stat via an ex-
ponent: 2Stot + 1 = multiplicity, the various configurations are indicated in
Table 4.3. The ground state has maximum multiplicity.
Consider nitrogen: outside the helium core, two electrons are in the 2s shell,
and three in the 2p shell. The s electrons have no angular momentum or spin,
and we ignore them in zeroth order. The important particles are the three 2p
electrons (the 2p 3 configuration, in atomic notation). If such is the case, the
ground-state and low-lying terms of any atom or ion with a p3 configuration

12 }(252)2p23p

10 2 p _} 2 2 4
4p_ (25 )2p 35 -25 2p

8
>CD
6
....
l
>CD
4 2p
Fig. 4.2. Energy-level diagram of
_20 nitrogen (N). Vector model applies
2 (25 2 ) 2p3 to lowest three levels only; but note
that they are well below the other
- 4S levels. The ground state belongs to
0 maximum S.o. = 3/2
 4.7 p3 Configuration 113

outside of closed shells must be similar. Figure 4.2 and Table 4.1 confirm this,
and show the good agreement between theory and experiment.
We first tackle this problem by the same technique used in the study of three
hydrogen atoms. The more conventional approach is harder, and is discussed in
the next section. First we write out the spatial basis functions. Instead of atoms
a, h, c, with which to associate the electrons, there are now the three p states
specified by the quantum numbers m l = -1,0, +1. We allow each to be oc-
cupied just once: this reduces both the Coulomb repulsion, and the number
of spatial configurations necessary to consider.
The 3! = 6 spatial configurations are

If/I = IfJI(rl) lfJo(r2) IfJ-I(r3)

1f/2 = IfJI(r2) lfJo(rl) IfJ-I(r3)
1f/3 = IfJI(r3) lfJo(r2) IfJ-I(rl) (4.50)
1f/4 = IfJI (rl) IfJO(r3) IfJ-I(r2)
If/5 = IfJI (r2) IfJO(r3) IfJ-I(rl)
1f/6 = IfJI(r3) lfJo(rl) IfJ-I(r2) .

The subscripts on the one-particle functions refer to ml, the eigenvalue of the
one-electron operators Lf. There are only two major differences between these
functions and those in the three-atom case: the present functions are ortho-
normal, and they are not all three equivalent. The significance of this will become
immediately clear.
The functions IfJI1IJ(r) are by definition solutions of the best possible self-
consistent one-particle Hamiltonian, ~o. As ~o usually turns out to be spheri-
cally symmetric, we shall assume this to be the case here, and factor the functions
as follows:

IfJm,(r) = Y1,m«(), fJ)R(r) . (4.51)

Fortunately, everything is known about the angular part of the wavefunction,

the spherical harmonics, Y1,m, and as for the common radial function which
is unknown, it will be eliminated from the problem.
The perturbation is the sum of 2-body interactions:

(4.52)

We may write down an eigenvalue equation for the energies in terms of the
matrix elements of ~' between the six basis states, and the overlap matrix. The
justification for this was given in the chapter on exchange, in terms of making
the energies stationary, or diagonalizing ~' within the restricted subspace. The
overlap matrix is in the present case just the unit matrix, by orthonormality. It
114 4. Many-Electron Wavefunctions

might be thought that the 3e'" matrix could be obtained from our earlier result
also by setting the overlap integral = 0, but that doesn't quite work as

(4.53)

because permuting functions with IAm, I = 1 is not quite the same as when
IAm, I = 2. Otherwise, we proceed much as before. We define the diagonal term
as

A == 3e";.1 = ... = 3e"~.6 >0 (4.54)

and nondiagonal elements BIA by

BIA == 3e";.2 = 3e":.4 = ... = 3e";.6

= f e [IjJT(rl)IjJO(rl)IjJ~(r2)IjJI(r2)] d 3r l d 3r 2 .
2
rl2
(4.55)

Other non diagonal elements B2A are defined as,

(4.56)

All matrix elements are proportional to the same constant A, also all turn out
to be real or can be made real. We have made use of the orthonormality of the
one-electron functions to eliminate all the irrelevant terms in 3e'" from these
expressions, and have indicated the difference between the two types of exchange
integrals by the subscript BI or B 2 • The third type of matrix element, for example,
3e";.s, automatically vanishes. One finds this out either by setting the overlap
integral = 0 in the previous problem, or by noticing directly that the exchange
between functions differing by more than a simple transposition must vanish in
the case of two-body forces and orthonormal basis functions. Finally, we obtain
the eigenvalue equation,

1 BI B2 BI 0 0
BI 1 0 0 BI B2
B2 0 1 0 BI BI
3e"'.v = A ·v = Ev.
BI 0 0 B2 BI
0 BI BI B2 1 0
0 B2 BI BI 0 1 (4.57)

Recall that there was a useless totally symmetric eigenvector, and a totally
antisymmetric eigenvector which we identified as the space partner of the
 4.7 p3 Configuration 115

quartet state, Stot = j-. Then there were two doublet states; these are no longer
degenerate, and we must find the proper linear combination to diagonalize Z'.
Let us examine the two states V Z3 and V~3 which were anti symmetric in particles
2 and 3. (They cannot mix with the two functions which are symmetric in those
particles, V Z3sy , and V~3sy). One finds by construction, the new eigenvectors

and (4.58)

(4.59)

Problem 4.6. Find the proper linear combinations of V Z3sy and V~3.y which are
eigenvectors of Z'.

There is an additional operator of which we should like to know the eigenvalues,

that is L;ot:

aot = Li + L~ + Li + 2(LfL~ + LiLj + LjLf)

+ (LtL"i + L!D; + LtLi + H.c.) . (4.60)

All that is required to calculate the matrix elements of this operator on our set
of states is such information as [cf. (3.44)]

L~ = 1(1 + 1) = 2 and (4.61)

(4.62)

for angular momentum 1. Thus in the diagonal matrix element (aot)I,1 there
enter three contributions of the form of (4.61), making 6, minus 2 from 2LfLj,
for a total of 4. The other matrix elements are just as easily found, and finally,

4 2 0 2 0 0
2 4 0 0 2 0
0 0 4 0 2 2
L;ot =
2 0 0 4 0 2
0 2 2 0 4 0
0 0 2 2 0 4 (4.63)

It is now a simple exercise (for the reader) to write the eigenvalues E as functions
of A, Bio and Bz and calculate the value of the angular momentum to which
they belong, with the eigenvectors already given. The spin of these same states
has already been discussed and given several times.
The following fact may be extricated from Condon and Shortley [4.10] or proved
directly by an expansion of (rlz)-I in spherical harmonics (see Problem 4.8):
116 4. Many-Electron Wavefunctions

(4.64)

It simplifies the results even more for the actual problem. Finally one finds for
the eigenvalues

Eep) = A(l +B 1) (4.65)

EeD) = A(l - B 1) (4.66)

EeS) = A(l - 4B 1) . (4.67)

In the usual spectroscopic notation, the multiplicity (2Stot + 1) has been written
as a prefixed exponent to the angular momentum S, P, D, F, G, ... (L tot = 0, 1,
2,3,4, ... ). From these equations, we find for the level spacings,

Eep) - EeD) 2
(4.68)
EeD) - EeS) 3

a ratio independent of A and Bh or of the integrals involving the radial parts of

the wavefunctions.
The agreement with experiment is pleasant. Not only are all p3 configurations
found to possess low-lying terms of the type discussed, and ordered in energy as
we have found them, but the ratio of (4.68) is experimentally well obeyed within
our self-imposed limits of accuracy. In Fig. 4.2 is reproduced the term level
scheme of nitrogen, together with the electronic configurations giving rise to the
various terms. In Table 4.1 is given the experimentally observed ratio of (4.68)
for various substances.

Table 4.1. Configuration and ratio of term splittings in various atoms and ions, and theory

Experimental
Theory N 0+ S+ As

Configuration np3 2p 3 2p 3 3p 3 4p 3
Ratio' 0.667 0.500 0.509 0.651 0.751

• See (4.68)
Source: Condon and Shortley [4.10]

Because each one-electron state is occupied, there is no difficulty in principle

to the construction of an equivalent Heisenberg, or vector model, Hamiltonian
for spins one-half. This is illustrated in Fig. 4.3.
From previous experience, we may almost guess that in ~H.1s

(4.69)
 4.8 p2 and p4 Configurations 117

Fig. 4.3. Labels (m,) and bonds in vec-

tor model of nitrogen atom

the proper choice of exchange constants is,

and (4.70)

which are defined in (4.55, 56). The solutions are given as Problem 4.7.

Problem 4.7. Find the eigenvalues of the Heisenberg Hamiltonian for spins one-
half, (4.69). Use the Pauli principle, and space inversion symmetry considera-
tions (on the true Pauli wavefunction of space and spin) to classify the solutions
of this exchange Hamiltonian according to L tot as well as Stot" Find the ratio
[Eep) - EeD)]/[EeD) - EeS)] and verify that it equals j when Jz = VI'
Hint: define T = S_I + Sh express ~H.is in terms of T and T + So.

4.8 # and p4 Configurations

Carbon, silicon, germanium, tin all have two electrons in the p shell. Oxygen,
sulphur, selenium, and others have two holes in the p shell. All have 3p ground
states. (The term level scheme for carbon, which is representative, is given in
Fig. 4.4.) Can we show this theoretically, and find the low-lying terms as well?
This is an exercise in Hund's second rule, which has not come into play
before. Also, here is a problem in which the Heisenberg Hamiltonian is of
dubious validity since there are only two particles, with three orbital states in
which to put them. This configuration is studied in lieu of the complex transition
series atoms, or rare-earth series, which are of greatest interest in magnetism.
For it is not intended to take the reader into involved computations, regardless
of their relevancy, but only to provide a conceptual introduction to the theory
of atomic structure.
It is essential to reduce the number of configurations that one has to consider
to a manageable few. In the L-S coupling scheme (so-called), states which can
be obtained from each other by repeated applications of St~t' or 4~t> are equiva-
lent except for the splitting due to spin-orbit coupling, (which is usually treated
separately). So, it is only necessary to consider the largest attainable value of
Ms and ML for each term, viz.,
118 4. Many-Electron Wavefunctions

and (4.71)

and in order to determine Stot and L tot and the associated energy eigenvalues, to
remain in the subspace of Ms = 0, 1 and ML = 0, 1,2.
ML = 2: The space function must be

(4.72)

in which M L = m ll + m12 = I + I = 2. As this is symmetric under interchange

of the spatial coordinates, the Pauli principle requires an anti symmetric spin
function, the singlet state

(4.73)

Thus, L tot = 2 and Stot = 0, and this is the I D state. The energy will be com-
puted subsequently, but now we continue the classification.
M L = I: The space function of interest must be

(4.74)

since the other possibility, the linear combination with + sign, is symmetric
under the interchange of the spatial coordinates and is therefore the M L = I
projection of the L tot = 2 state derived above. If the space part is antisymmetric,
the spin part must be symmetric [the spin-triplet X+(¢I)X+(¢2), or X-(¢I)X-(¢2),
or (4.73) with + replacing -]; thus L tot = I and Stot = I, and this is the 3P
configuration.
M L = 0: The three functions with M L = ° have mll = mI2 = 0, or mIl
= -mI2 = ± 1. Two linear combinations of these must be the M L = projec- °
tions of the two states found above, (4.72, 74). The linear combination orthogo-
nal to both of these is the new function of interest,

(4.75)

Lt,t applied to this state yields identically zero, and it follows that this is the IS
configuration, and the list is now complete (see-Table 4.3).
The perturbation is

(4.76)

which we expand in Legendre polynomials,

 4.8 p2 andp4 Configurations 119

where (4.77)

in which, < is the lesser, and '> is the greater of'l and '2, and OJ is the angle be-
tween the two vectors. Because one takes expectation values using p functions,
only terms with n = 0 and n = 2 can survive the integration.
The atomic functions have normalized angular factors (cf. Table 3.1):

Y1,o = (3/41t)1I2 cos () (4.78)

and
YI,±I = =f(3/81t)1/2 sin (}e±i, (4.79)

as well as a common (unknown) radial factor R(r).

The term energies are the expectation values of the perturbation in the
various states listed previously. The leading term in the expansion, correspond-
int to n = 0, is

(4.80)

which does not depend on angles and is therefore the same in all three states.
This can be eliminated by a simple shift in the zero of energy.
The next term in the expansion, n = 1,

(4.81)

vanishes by inversion symmetry, but finally n = 2,

(4.82)

contributes differently in all three states. The radial factor is common to all
three. It is readily seen or it can be proved by the triangle inequality (of p. 87)
that all succeeding terms in the expansion in Legendre polynomials will not con-
tribute. So that apart from an additive constant arising from (4.80), all the
energies will be proportional to the same, albeit unknown, radial integral. And
it remains a straightforward exercise in manipulating trigonometric identities
to show that

EeS) - EeD) 3
(4.83)
Ee D) - Ee P) = 2' .

This relationship is of course independent of the additive energy shift from (4.
80), or of the common radial integral. Its validity depends only on the validity
120 4. Many-Electron Wavefunctions

of the Hartree-Fock procedure and on the validity of our neglect of states outside
the p shell. In Table 4.2 we compare the theory with experiment.

Problem 4.8. Using the procedure indicated in the text, derive (4.83). Hint: no
integrals need be evaluated to establish this result, but various integrals need to
be compared. This can be done solely by use of the mathematical identity,

The agreement is not always as perfect as for the atoms in Table 4.2. For ex-
ample, the failure of one or another of the hypotheses gives for La+ the ratio
18.43, which differs from theory by an order of magnitude. In such instances
more terms must be taken into account, and while such calculations yield

Table 4.2. Configurations and ratio of term splittings in various atoms vs theory

Experimental
Theory' C Si Ge Sn 0

Configuration np3±! 2p2 3p2 4p2 5p2 2p4

Ratio 1.50 1.13 1.48 1.50 1.39 1.14

• See (4.83)
Source: Condon and Shortley [4.10]

agreement with experiment they may become very involved. But at present the
ultimate validity of the many body Schrodinger equation for the atoms is not in
dispute, and even the remarkably simple truncation, such as the organization
into atomic shells, gives reasonably accurate results with minimum calculation.
Thus, atomic magnetism which is the consequence of Hund's rule, is docu-
mented by experiment, explained by elementary theory, and confirmed by
elaborate calculations.
In Fig. 4.4, we show the spectrum of carbon. It is to be noted that as in N, the
terms not described by the simple theory lie several volts above the IS term. As
a consequence, theories of neutral carbon in solids probably need take only the
lowest three terms into account, which in the atom are adequately described by
the sort of first-order perturbation theory just expounded.
Finally, in Table 4.3, there is given a list of possible terms (configurations)
which can be constructed out of a specified number of equivalent electrons. The
lowest, according to Hund's rule, is given in boldface.
Practically all known magnetic substances involve d- or f-shell electrons, the
study of which would take us well beyond the scope of an introductory work.
The interested reader will readily find specialized treatises on this subject, see,
e.g. [4.12].
 4.8 p2 and p' Configurations 121

Fig. 4.4. Energy-level diagram of

10 Carbon (C), as example of Hund's
two rules

o
Table 4.3. Possible terms for a number of equivalent electrons

S 2S
S2 IS

P or p' 2p
p2 or p' ISD 3p
p3 2PD ·S

dor d 9 2D
d 2 or d 8 ISDG 3PF
d 3 or d 7 2PDFGH 'PF
2
d' or d 6 ISDFGI 3PDFGH 'D
22 2 2 2
d' 2SPDFGHI 'PDFG 6S
3 22

[or [13 2F
f2 or [12 ISDGI 3PFH
P or [II 2PDFGHIKL 'SDFGI
2222
f4or[IO ISDFGHIKLN 3PDFGHIKLM 'SDFGI
24 423 2 3243422
f' or f" 2PDFGHIKLMNO 'SPDFGHIKLM 6PFH
457675532 2344332
f6 or f8 ISPDFGHIKLMNQ 3PDFGHIKLMNO 'SPDFGHIKL 7F
4648473422 659796633 3 2 3 22
f' 2SPDFGHIKLMNOQ 'SPDFGHIKLMN 6PDFGHI 8S
25710109975 4 2 226575533

Note: A number under a term symbol indicates the number of different levels of this type.
The lowest level is normally that with the highest L of the highest multiplicity (Hund's rule).
It is indicated in bold type.
Source: [4.11].
122 4. Many-Electron Wavefunctions

4.9 Independent Electrons

It is possible to refine the study of atomic structure far beyond the short intro-
duction provided in the preceding pages. Moreover, we have not yet approached
the subject of magnetic elements. Usually magnetic substances are those com-
posed of rare-earth or transition-series elements, with incomplete (magnetic) f
shell or d shell as the case may be. Because magnetism is a phenomenon that
occurs in the solid state, it is fitting however to interrupt the discussion of
atomic properties, at the point where it begins to become complicated, and turn
to the idealized model of electrons in the solid state: the independent particle
model.
In this section we shall consider independent electrons in solids from a some-
what special point of view, ordering the ground state and elementary excited
states according to their multiplicity, and proving some results which later will
apply to a theorem about interacting electrons. What we shall now observe is
that the degeneracies (which in the atom were lifted by interactions among the
valence electrons, resulting in Hund's rules) will in solids be lifted in a different
manner, without the need for any interactions; and that these noninteracting
electrons will be nonmagnetic, except for second-order para- and diamagnetic
effects which are discussed elsewhere.
Previously (see the section on overlap exchange) we had proved that interacting
electrons are not ferromagnetic without overlap .. here we shall show that overlapp-
ing electrons are not ferromagnetic if they do not correlate.
The Hamiltonian for N independent electrons (for simplicity assume N
= even) is ofthe form

N
~=~h(r,) . (4.84)
1=1

It is not useful to specify further the one-particle Hamiltonian her), which may
include interactions with other electrons on the average, with the periodic array
of atoms comprising the solid, with the surfaces and other external forces or
impurities. In addition to these forces and the kinetic energy of the particle, we
may allow any other potentials ihto her), except those which involve the spin of
the particle explicitly, and still arrive at the desired results. First, solve the
eigenvalue equation,

h(r)J:(r) = enJ:(r) (4.85)

for the one-particle wavefunctions and eigenvalues J: and en, arranging the
eigenvalues in ascending order:

(4.86)
 4.9 Independent Electrons 123

etc. [It should be noted that the density of levels increases with the volume of the
box; for a finite but arbitrarily large box, the levels can always be enumerated
as in (4.86).] The many-electron wavefunctions are Slater determinants of space
and spin functions. By the completeness of the set of fn' the Slater determinants
form a complete set of admissible Pauli wavefunctions for the many-electron
system. However, we recall that they are not automatically eigenfunctions of
S~ot, although they are always good eigenfunctions of Sfot with eigenvalue M.
Recall also the results of (4.30) and those following: the spin function may be
projected out of the Pauli function of definite M, leaving just the space partner

(4.87)

which for independent particles is the product of two determinants. Because the
one-electron functions may occur once in each determinant, the lowest energy
belongs to a function of this type where the lowest possible en's are represented
twice

(4.88)

This is automatically an eigenstate belonging to M = 0, with energy

1/2N
Eo = 2 I: en· (4.89)
n=1

The factor of2 comes ultimately from Kramers' degeneracy.

The function of chosen in this manner has also the important property that
it is an eigenfunction of S~ot with zero eigenvalue (compare p. 105). Therefore, it
should be labeled 0FO according to the notation previously introduced. It is the
absolute ground state, unless there is degeneracy, in which case it is among the
ground states.
If next we inquire about the ground state in the M = 1 subspace, we must
first eliminate °Fo which cannot be raised. But by reasoning similar to the above,
the optimum second choice is found to be

'F ~ J(~ N
1
+ 1)!( ~ N - I)!
det II/'(r,) .. ,/."...,(r u_,) II

x det 1I/J(rIl2N+2) ..• fIl2N-I(r N) II . (4.90)

The energy of this state is given in (4.91). Now it is trivial to construct functions
of the type of and -IF starting with this new function; but the functions of type
uF, which we might attempt to construct with F land its permutations, vanish
124 4. Many-Electron Wavefunctions

identically. This proves in fact that we have found 1FI, an eigenfunction of S;"t
with the indicated eigenvalue. The energy E 1 ,

(4.91)

is the absolute ground-state energy in the subspace of eigenfunctions belonging

to Stot 2 1, although it is greater than Eo.
One proceeds systematically through all the higher M subspaces, establishing
always that the function of lowest energy of type MF is in fact MFM; and the
concomitant result for the energy can be summarized most compactly as

(4.92)

defining Eo(S) to be the lowest among all the energy eigenvalues E(S) of the
determinantal functions or linear combinations thereof, M FS.
We next examine the possible degeneracies. If her) is real and one of the /" is
complex, then by complex conjugation of both sides of (4.85) we see that/: is
also an eigenfunction, with the same eigenvalue en. This is the only degeneracy
possible in all generality. For consider an empty box, with dimensions L", Ly
and L z and a Hamiltonian

liZ
her) = - -Vz (4.93)
2m

(kinetic energy only) which has eigenfunctions e(lk·r) and eigenvalues

(4.94)

We see that if L" =1= Ly =1= Lz there is in fact no additional degeneracy; so that in
the presence of potentials any added degeneracy would be "accidental" and
could be eliminated by distorting the box a little bit, or by some other artifice.
For convenience N is now chosen to be a multiple of 4, which guarantees Eo
to be non degenerate; for

and therefore (4.95)

What is more, the eigenfunction 0FO of (4.88) is real; for it is both possible and
necessary in the ground state that for every complex/" the conjugate/: also be
present in each determinant. And even more compelling, is the argument that a
complex 0FO would lead to a degenerate ground state, in violation of (4.95) above.
 4.10 Electrons in One Dimension: A Theorem 125

The total current operator is

(4.96)

and its ground-state eigenvalue I is given by

(4.97)

Complex conjugation of both sides of this equation, and the reality of the eigen-
value I and of the eigenfunction 0FO establish the result 10 = 0, regardless of the
potential. Thus the current and the spin both vanish in the ground state, and the
system is entirely nonmagnetic. This is the quantal generalization of the Bohr-
Van Leeuwen theorem.

4.10 Electrons in One Dimension: A Theorem

... no one could move to the right or left to make way for passers-by, it
followed that no Linelander could ever pass another. Once neighbors, always
neighbors. Neighborhood with them was like marriage with us. Neighbors
remained neighbors till death did them part [4.13].
Because electrons in one dimension can be ordered like beads on a string, the
topology of their nodal surfaces turns out relatively simply. One can generalize
the result found for two electrons, and prove the Lieb-Mattis theorem in one
dimension, [4.5]: In one dimension, not even interacting and overlapping electrons
can be ferromagnetic. A somewhat milder extension of this theorem, in which
the type of interactions is restricted, can also be proved for N electrons in three
dimensions (as we shall see in a subsequent section).
Before stating the theorem and proceeding to the proof, it is appropriate to
comment on its relevance. This theorem occupies a special place as the guardi-
an of our conscience when we speculate concerning the origins of ferromagnet-
ism. Any fundamental explanation or theory, which conceivably would lead to
ferromagnetism in one dimension, must ipso facto be false! And that is the reason
we have discussed angular momentum, atomic theory, and the origins of Hund's
rules, for these are intimately connected with the three-dimensionality of space,
and, by virtue of the present theorem, are an essential part of the phenomena.
Deferring further such considerations, we now go on with the

Theorem: For interacting electrons in one dimension, the ground state in any
M subspace is nondegenerate, and belongs to Stot = 1M I. From this it follows
that Eo(S) < Eo(S + 1), that the current vanishes in the ground state of any M
subspace, and that electrons in one dimension are entirely nonferromagnetic.
126 4. Many-Electron Wavefunctions

Proof: What we have shown for noninteracting particles in the previous section
will be adapted to the present case of interacting electrons, for which the Hamil-
tonian is

(4.98)

The potential V includes all external forces, periodic or other, and the interac-
tions among electrons which may be taken to be arbitrary. The only exclusions
are as follows: no velocity or spin-dependent forces or nonlocal potentials will
be admitted, and V must be integrable, for reasons which are or will become clear.
We shall adopt a standard boundary condition: If the size of Lineland is D, the
boundary condition will be that the wavefunctions vanish whenever any x"
±-}D. Admissible Pauli eigenfunctions are

M'l'S(l, 2, ... , N) (4.99)

where, for compactness, n = I, 2, ... stands for the couple of space and spin
coordinates (x"' c;,,). First, we write the eigenvalue equation,

(4.100)

noting that the energy cannot depend on M. (Because Jr'commutes with the spin
operators, we may apply spin raising and lowering operators to both sides of
Schrodinger's equation, changing M without affecting the energy eigenvalue).
As usual, we now eliminate the spins, and consider the space functions Mfs,
which themselves obey Schrodinger's equation (4.100), with precisely the eigen-
value E(S). All of them can be brought into the M = 0 subspace, by application
of spin raising or lowering operators if necessary, and therefore we start by
examining this particular subspace in which the ground-state eigenfunction is
surely to be found. Consider an arbitrary eigenfunction, of unknown Stot>

(4.101)

in the region R defined as follows:

(4.102)

The natural boundaries of R occur where the inequalities are just barely violated,
e.g.,

or x" = X,,±I or (4.103)

 4.10 Electrons in One Dimension: A Theorem 127

etc., and it should be noted that 'Y vanishes identically on these boundaries.
Although the total configuration space consists ofthe variables Xlo ... , X N in any
ordering, when 'Yis known in R it is known everywhere, because the other regions
can be found by trivial permutations .9 of the two sets of coordinates. (Recall
the definition of trivial permutations given on p. 103) For example, if

(Xlo ... ) is a point in R (4.104)

and
(x;, ... ) is a point in R' (4.105)

and .9 is a permutation which connects R to R',

.9R= R' (4.106)

then we obtain 'Yin R' by the trivial permutation

'Y(X;, ... ) == (-I)P.9['Y(xl'''')]' (4.107)

And in this self-evident manner, we can find the function everywhere. Indeed,
with the sign convention used above, it is guaranteed that cywill have the proper
antisymmetries under its trivial permutations .9. The nontrivial permutations,
it is seen, are not needed to map R into the entire space; and thus they play only
their usual role, which is raising and lowering M.
Now, we state and prove a lemma which is at the core of the theorem: The
eigenfunction of ~which belongs to the lowest energy eigenvalue and is subject
only to the boundary condition that it vanish on the boundaries of R is nodeless
inR.
The proof proceeds exactly as in the case of two electrons. If the ground-
state function had nodes, then its absolute value would be a nodeless function,
which would have identically the same variational energy but which would not
itself be an eigenfunction. Since the variational energy of any but a ground-state
eigenfunction must exceed the ground-state energy, there is a contradiction
unless the absolute value of the ground-state eigenfunction is itself a ground-
state eigenfunction; which is only possible if the ground state is nodeless. The
nodeless state is, moreover, nondegenerate. For two nodeless functions cannot
be orthogonal, but all the eigenfunctions of ~can be made mutually orthogonal,
and other than nodeless functions have been excluded by the preceding argu-
ments. This nondegeneracy is stronger than we were able to prove for non-
interacting particles (in the previous section) because we have restricted the
arguments here to one dimension.
The nodeless property is independent of the magnitude of V, and so is the
shape of R. We may therefore compare in R the ground state of a Hamiltonian
~with arbitrary V, to that of the noninteracting Hamiltonian ~o with V = O.
128 4. Many-Electron Wavefunctions

In both cases, the ground state is nodeless; the two corresponding functions can-
not be orthogonal, therefore they share the same quantum numbers I and Stot.
What these quantum numbers are for noninteracting electrons (V = 0) is
something which was already discussed: viz., zero. And by the above it follows
that both these quantum numbers vanish for arbitrary Vas well, in the ground
state.
Next, one introduces the fundamental region suitable for M = 1, which is

(4.108)

By analogy, one proves that the ground-state function in this region is nodeless
and belongs to the same quantum numbers regardless of V. It follows by com-
parison with the results for V = 0 that this function has Stot = 1; the proof for
higher M proceeding in identically the same manner, until maximum S is attained
and the theorem for the interacting particles is proved.

4.11 The Wronskian [4.l4]

A somewhat different approach may serve to clarify the results obtained so far.
Suppose we have a set of real space eigenfunctions of the Schrodinger equation

JJe'J;(Xh ... ) = EJJ;(Xh ... ) (4.109)

with

just as in (4.98), and it is desired to order the eigenvalues EJ in a sequence of

increasing energy.
The lowest energy belongs to the nodeless eigenfunction; since the permuta-
tion operators commute with the Hamiltonian, it can also be established that
this function is totally symmetric under the interchange of any coordinates and
therefore is not of any use for more than two Fermions. But there can only be
one nodeless function (compare proof of analogous statements on p. 109), and the
others can be examined as to the topology of their nodal surfaces. The nodal
surface is defined as the locus off = o.
We shall not go into any detail beyond proving the simple but interesting
lemma:
lfthere are two functions/, andJ; such that a nodal surface S, of/" encloses
a region R, which neither contains, nor is intersected by any nodal surface ofJ;.
 4.12 Theorem in Three Dimensions 129

then Ei > E,. For example, consider any eigenfunctionlt together with the node-
less eigenfunctionJo; the lemma implies Ei > Eo, which is obviously true for all i.
To prove the lemma, it is necessary to multiply both sides of (4.109) on the
left by It; to multiply the analogous equation forlt on the left by h, substract the
first from the second, and obtain

(4.110)

The potential has been eliminated from this expression, and the left-hand side
involves only the kinetic energies or the gradient of the Wronskian, for which
we have used the generalized Laplacian notation

V.V -- V2 = ""-2·
N a
2
(4.111)
-~a
.=1 x.

By assumption, neither function changes sign in Ri so we may, in considering

this region, assume both functions are positive, and if not, make them so. Next,
integrate over R i , which by Green's theorem gives an integral of the Wronskian
over Si on the left-hand side, and a positive integral times the energy difference
on the right.

(4.112)

Butlt = 0 on the nodal surface Si; and the normal derivative of It is negative on
Si because the function is positive inside the volume bounded by the surface and
negative outside (see Fig. 4.1). So the left-hand side is positive, and the equation
can only be satisfied if Ei > E" which proves the lemma.
Evidently, we have used nothing about dimensionality in this demonstration,
therefore it is indeed equally valid for electrons in three dimensions [with
N
V2 == ~ V;, instead of (4.111 )]. But the difficulty of finding nodal surfaces has so
u=1
far frustrated attempts to apply these results to realistic problems for N > 2 in
three dimensions.

4.12 Theorem in Three Dimensions

The great stumbling blocks to proving our theorem in three dimensions are the
impossibility of ordering the particles in any sensible manner and the difficulty
in locating the nodal surfaces along which the wavefunctions vanish. One may
get around both these ifhe restricts the potential to the form

(4.113)
130 4. Many-Electron Wavefunctions

which is to say, "separately symmetric" under permutations of the x, y, or z co-

ordinates alone. The periodic potential typical of the solid state may be included,
as well as electron-electron interactions. For potentials of this class, we may
prove the theorem in the same form as for noninteracting electrons

Eo(8) ::;; Eo(8 + 1) (4.114)

so that for ferromagnetism to occur, even in three dimensions, it does not suffice
for electrons to overlap and to interact, no matter how strong the potential. The
form of the potential shall playa peculiar role.
In the present section, we shall prove the above inequality for a special case
of the potential V, viz.,

(4.115)

the separable potential. Primes are used to distinguish the functions. Extension
to the general class of potentials, (4.113), is straightforward.
When the kinetic energy operators are included, the Hamiltonian with a
separable potential can be written as the sum of three Hamiltonians, one for
each dimension. The space eigenfunctions are therefore product functions:

If! = f(x!> ... )f'(y\, ... )!"(z!> ... ) (4.116)

and the energy eigenvalues are additive:

E = E + E' + E" . (4.117)

We should now introduce the spin functions, minimize the sum of the energies E,
E', and E" subject to the Pauli principle, and finally try to discover the spin of
the lowest allowable eigenfunction. The f(x!> ... ) function obeys a one-dimen-
sional Schrodinger equation which is invariant under permutations of the x
coordinates. Similar statements are true for f' and f". Therefore we may choose
f,f', andf" to be simultaneous eigenfunctions of the permutation operators and
write

(4.118)

and similar expressions for f' and f". The function is antisymmetric under the
interchange of coordinates separated by commas; but has no particular sym-
metry properties under interchange of coordinates separated by one or more
vertical bars.
Whereas the Pauli principle restricts the allowable functions in one dimen-
sions to those with either zero or one vertical bar, in the case of electrons obeying
the Pauli principle in two or more dimensions we must allow for any number of
bars, and therein lies part of the difficulty.
 4.12 Theorem in Three Dimensions 131

Example: Consider a totally anti symmetric space function, F, which belongs to

the totally symmetric (totally ferromagnetic) spin function; it can be of the form

(4.119)

where f' and f" are both totally symmetric, and therefore have N - 1 vertical
bars, and f is totally antisymmetric. Or all three functions could be totally anti-
symmetric; and Fwould be of the form

(4.120)

with no vertical bars at all. Ifwe require Fto have one bar, or if we allow linear
combinations of products off's, then the number of possibilities of bars for the
f's becomes virtually endless.
Proof of Theorem in (4.114): Given an eigenfunction of ~and Stet> which is also
a Pauli function, i.e., a totally anti symmetric function of the coordinates "
= (x" y" z" m,) (let m, = t or ! denote the spin eigenvalue), express it as a
linear combination of product functions of the type

(4.121)

where u, = (x" m,) and v, = (y" z,). In order that a totally antisymmetric Pauli
function might be extractable from (4.121), it is required that g be symmetric in
the variables separated by semicolons if f is antisymmetric in the variables
separated by commas. The function above has the same energy as the original
Pauli eigenfunction. This is because the latter is a linear combination of (4.121)
and its permutations, and of course ~is invariant under the permutations, and
therefore the energy is unaffected.
A simplified proof of the theorem proceeds as follows: for all practical pur-
poses, f is a Pauli function for p one-dimensional electrons Xl> ... , xI" and
separately also for Xp+l> ... , etc.; and if we take linear combinations of j-type
functions so as to give these groups of variables definite spin, it is clear by virtue
of the one-dimensional theorem that the lowest energy is achieved only if the
total spin of the first p electrons is zero (or one-half, ifthe number is odd), and
similarly for all the successive groups separated by vertical bars. Such linear
combinations are of course once more degenerate with the original function,
(4.121), as well as with the original Pauli function from which the latter was
extracted. Notice that the process by which the sets of electrons are given definite
spin does not affect the antisymmetry, so that we may, without loss of generality,
assume that infthe various sets of variables have definite spin.
Finally, a Pauli function is constructed from (4.121) in two steps: first, an
eigenfunction of Stet is constructed with Clebsch-Gordan coefficients; second,
the function so obtained is totally anti symmetrized. The sets with zero spin do
not contribute to the total spin; thus if all the sets contain an even number of
132 4. Many-Electron Wavefunctions

particles, only Stot = 0 is accessible. If N is even, as we usually assume, we may

have none, two, four, six, ... sets with odd numbers of variables, for which the
lowest energy is achieved with spin one-half. Two spins one-half may be com-
bined to give spin zero, or spin one. Four may be combined to yield Stot = 0,
1,2, etc. So long as we stick to the functions with the lowest energy in each
symmetry class, we see that if a function belonging to total spin S can be con-
structed, then so can another function which has the same energy but total spin
S - 1, or less. Because the same cannot be said about S + 1, one immediately
arrives at a statement of the theorem in the form of the inequality, (4.114).3

4.13 Ordering Theorem versus Hund's Rule

The ordering theorem, as we proved it in three dimensions, differs from its one-
dimensional version by not being stated in terms of strict inequalities. And in-
deed the ground state for the interacting three-dimensional electrons could
belong to some nonzero value of the total spin, say S, but then necessarily also
to S - I, S - 2, ... , 1, and O. The maximum value of S in the ground state may
be estimated, on the basis of noninteracting electrons in three dimensions, not
to exceed the order o/magnitude S = iN1 / 2 , regardless of how many electrons
are present. Since ferromagnetism in the solid state requires that

S 4= 0
lim N (4.122)
N-+~

we see that solids with separately symmetric potentials are never ferromagnetic.
Such is not the case for a hypothetical atom with a separately symmetric poten-
tial. The atom is typically a collection ofless than 100 electrons, and a total spin
of the order of S = 5 is not negligible, and indeed is of the order of magnitude
observed in atoms. Moreover, charges in nature interact by Coulomb potentials
which are not "separately symmetric". The difference between the Coulomb in-
teraction and the closest separately symmetric analog to it may be considered
as a perturbation, capable of lifting the degeneracy amongst ground states be-
longing to S, S - 1, ... , etc. In the partly filled shells of atoms and molecules,
the state satisfying Hund's rules does not merely lie among the ground states, but
is found to be the lowest state of all. Simple calculations explaining this feature
have been presented in earlier sections of this chapter.
Isolated atoms and molecules can therefore possess a small "permanent"
magnetic moment, one which requires some additional energy to destroy. What
happens to these individual moments when atoms are aggregated to form a
solid, is a question to be addressed in the following chapters. Before pursuing

3 For much more detailed and constructive proof including the extension to the general se-
parately symmetric potentials, (4.113), see [4.5].
 4.14 Second Quantization 133

this line of inquiry, we shall find it advantageous to review some fairly general
and useful techniques in the study of large numbers of identical particles.

4.14 Second Quantization

The essential information contained in the determinantal function (4.14), is that

3 particles satisfying the Pauli principle occupy (j, r) [using short-hand notation
for the spatial statejj(r) and spin state r = ±~], (k, s) and (n, t). The set of 3
pairs of quantum numbers summarizes this information, but does not explicitly
reveal the anti symmetry of the wavefunction. If however, we introduced 3
operators cj':, c:' and c! with the following property

where b, a are any of the three sets of quantum numbers above, then cJ; c:. c!
is an operator carrying the requisite information and which changes sign upon
interchange of any pair of particles. It is on this observation that the formalism
of second quantization is based.
Second quantization allows one to deal with variable numbers of particles
as conveniently as with a fixed number. The "particles" can be electrons as
stated above, or any other variety of interest such as lattice vibrations (phonons),
and spin excitations (magnons) with differing symmetry properties. We shall
first analyze the case of electrons, then discuss examples not subject to the Pauli
principle. It is convenient to deal with noninteracting particles at first, and then
to express the interactions in the new language.
The unique state of zero particles is denoted the "vacuum" and written,

10) (with (01 the conjugate state). (4.123)

To introduce an electron of spin s at the spatial point r one uses the wave-
operator 1J'.*(r) as follows:

1J':(r) 10), with (0 11J'.(r) the conjugate state. (4.124)

The requirement that this state be normalized and orthogonal to that of a

particle having a different spin index or at a different point of space is the fol-
lowing:

(0 11J'.,(r')IJ':(r) 10) = o•.•,o(r - r') (4.125)

using the Kronecker delta for the discrete index and the Dirac delta for the
continuous one. Equation (4.125) has two interpretations, the first as the inner
product of a state (4.124) with a similar (but conjugate, such as column to row
134 4. Many-Electron Wavefunctions

vectors) state; and the second as the inner product of the (conjugate) vacuum
with the state

'PAr') 'P~(r) 10> . (4.126)

For s = s' and r = r' this must be proportional to the vacuum, thus (4.126)
must not contain any particles. Hence the interpretation: successive applications
of '1'* raise the number of particles in a state by 1, successive applications of 'I'
lower it by 1. (For the conjugate states, the opposite is true.) As the vacuum has
the lowest possible occupancy, we find

'P.(r) 10> == 0 (or <01 'Ps*(r) = 0). (4.127)

The Pauli principle comes into play with 2 or more particles

'Ps*(r) 'P;I;(r') 10> . (4.128)

It requires a change of sign under interchange of two particles, i.e.,

'1':: (r') 'Ps*(r) 10> = - 'P:(r) 'Ps~(r') I0> (4.129)

and this must be true even if 10> is replaced by an arbitrary state. This require-
ment, and the normalization (4.125) can be conveniently combined into a com-
pact set of anticommutation relations as follows:

{'P.(r), 'Ps,(r')} = 0 = {'P:Cr), 'P:;(r')}

{'P.(r), 'P:;(r')} = t5ss,t5(r - r') (4.130)

in which the antic om mutator bracket {A, B} is short-hand for AB + BA.

The linear combination of '1'* which corresponds to a given spatial wave-
function, e.g., to one of the states her) used in the Slater determinantal wave-
function early in the chapter, is simply

(4.131)

and thus the second-quantized equivalent of the three-particle wavefunction

(4.14) is precisely

cJ;c:,c:, 10> . (4.132)

Problem 4.9. Prove that 'P~later(rl> r 2, r 3) of (4.14) is given by:

[Use (4.127, 130, 131)].

 4.14 Second Quantization 135

Problem 4.10. Prove the anticommutation relations

and

by use of (4. 130, 131).

Given a state such as (4.128) or (4.132) with a certain number of particles, we

can define the particle number operator n as

(4.133)

Given an Hamiltonian ~ for noninteracting particles with a complete set of

one-particle eigenstatesjj(r), we can always invert the type of relation (4.131) to
obtain

(4.134)

Inserting this into (4.133) yields for the number operator

(4.135)

The occupation of the state (j, s) is given by the eigenvalue ofthe operator c'!.Cj..
often abbreviated njl.
Like the number operator, the Hamiltonian of noninteracting particles is a
quadratic form in field operators

(4.136)

making use of (4.134) and Hjj(r) = Ejjj(r), the time-independent Schr6dinger

equation for the jth eigenfunction. The interpretation is fairly direct: the energy
ofthejth state is the product of the energy (Ej) and the occupancy (nj + + nj_).
The simplicity ceases when two-body forces are introduced. Where in first-
quantization they might be written

(4.137)

they now appear as

(4.138)

In terms of the raising and lowering operators of the unperturbed Hamiltonian

(4.136), C'!. and Cj.. the interaction Hamiltonian takes the form
136 4. Many-Electron Wavefunctions

(4.139)

with
(4.140)

The first-quantized operators and wavefunctions refer to the explicit number

of particles [e.g., the 3 X 3 determinant (4.14) for 3 particles], whereas the
second-quantized operators are valid for all possible occupation numbers. The
variational and perturbation-theoretic formalisms are equally valid in the new
language as in the old; indeed, the pictorial Feyman diagrams are often used to
describe the various terms in the perturbation-theoretic studies of H 2op' For
example, (4.139) is represented pictorially as a vertex with four arrows.
For further details, we refer to any specialized text, see, e.g. [4.15].
A very important result, known as Wick's theorem [4.15], allows the com-
putation of the matrix elements of H 20P and higher powers of this operator, in
terms of the simple contractions such as (0 ICl,s cl,s I0), their products and
sums, e.g., determinants. This theorem is available for fermion (anticommuting)
operators of the type we have been discussing, for boson operators as we shall
now discuss, but not for spin operators which, as previously studied in Chap. 3,
have fairly complex commutation relations.
Particles subject to Bose-Einstein statistics include He 4 atoms (even numbers
of fermions) as well as any system of particles for which the wavefunction re-
mains symmetric under permutations of two or more of the particles. The main
modification one makes to the relations developed above for fermions is the sub-
stitution of commutation relations for the anticommutators of (4.130). Thus,

rfJ(r)rfJ(r') - rfJ(r')rfJ(r) == [rfJ(r), rfJ(r')] = 0

and similarly,

[rfJ*(r), rfJ*(r')] = 0
together with

[rfJ(r), rfJ*(r'] = c5(r - r') (4.141)

writing the wave-operator as rfJ for the bosons, and omitting the spin index, if
any. Among fields satisfying these commutation relations we can also count
quantized lattice vibrations (phonons), photons, and quantized spin-waves
(magnons). Concerning the latter we shall have much more to say in later sec-
tions of this text. Particle-conservation laws are far less stringent for bosons
than for fermions, and it is not uncommon to have odd powers of rfJ or rfJ* in
the Hamiltonian in addition to the terms discussed above for the fermions.
 4.14 Second Quantization 137

The field operators, whether for fermions or bosons, are particularly useful
in the construction of Green functions-functions of time, space and tempera-
ture which contain all the thermodynamic and transport-theoretic information
relevant to a given Hamiltonian.
5. From Magnons to Solitons: Spin Dynamics

There exist striking similarities between the elementary excitations of a fer-

romagnet and those in an elastic solid. In the latter, we know that an atom
displaced from equilibrium position will oscillate with the motion and fre-
quencies of the normal modes of the crystal. The effect of quantum mechanics
on this motion is to quantize the amplitudes of the individual normal modes,
into units known as phonons. The analogous normal modes in the magnetic sys-
tem are the spin waves; when the quantum mechanical nature of spins is taken
into account, these also are quantized, with the basic unit being the magnon.
In the ferromagnet, for which the ground state has all spins parallel, the small
amplitude oscillations are accurately described by harmonic-oscillator type
dynamical variables. The interaction between neighboring oscillators is respon-
sible for the dispersion, the dependence of frequency w of a normal mode, on the
wave vector k (the wavelength .Ie = 2rt/k). Indeed, as the ground state of a ferro-
magnet is known exactly, we are able to extend the analysis to obtain the exact
one- and two-magnon eigenstates. The latter number bound states and scattering
states, the interactions being caused by an inherent nonlinearity in the spin
dynamics, due to the coupling of the motion in various directions by the com-
mutation relations elaborated in Chap. 3. The analysis of 3- or more-magnon
states requires the solution of a full many-body problem.
In the antiferromagnet (a solid in which neighboring spins tend to be
antiparallel), the inherent nonlinearities even prevent us from obtaining the
exact ground state or elementary excitations, with the exception of one dimension
(and there, only for spins 1/2 or spins s - 00, the classical limit). Nevertheless,
spin-wave theory is as useful for an approximate analysis, as are the phonons
in the study of the anharmonic solid. The ground state and elementary excited
states, and the ground state correlation functions, can all be computed with
spin-wave theory. A check on the accuracy exists, in a variety of numerical
experiments on representative systems.
There are instances where magnons yield only part of the answers, perhaps
not even the qualitatively significant part. In two-dimensional magnets, the
vortex has been found to be the most significant topological excitation, and
in one-dimensional magnets, magnons and solitons are both required in the
dynamics. Such relatively new developments are treated in this chapter which,
with slight exceptions, is consecrated to the study of the Heisenberg Hamiltonian
with nearest-neighbor interactions, i.e., to insulators.
 5.1 Spin Waves as Harmonic Oscillators 139

5.1 Spin Waves as Harmonic Oscillators

The approximation of spin waves by harmonic-oscillator functions, as opposed

to a description "based on creation and annihilation operators, is to a consi-
derable extent only a semantic one, but nevertheless is probably of use to those
readers to whom harmonic oscillators are more intuitive than the techniques
of quantum-mechanical field theory.'"
For concreteness, let us assume (for the present) a ferromagnetic Heisenberg
Hamiltonian with nearest-neighbor interactions, in an external homogeneous
magnetic field H

.%= -J L;L; SloS, - Hgll B L; Sj J>O (S.l)

I>j~ I
nearest neighb.

where

Beside the isotropic interactions which are included therein, we may also con-
sider anisotropic interactions of various origins. Those of lowest order are of
"dipolar" structure,

(S.2)

which represents the magnetic dipole-dipole interaction (the magneto static

potential of each spin in the field of the others) with

(S.3)

But there may be additional contributions to D I" principally those arising from
the spin-orbit coupling mechanism [S.2]. These, like the exchange force itself,
tend to be very short-ranged. For concreteness, let us limit the present discussion
to cubic materials and the anisotropic interactions to (5.3) in lowest order.
In Chap. 3 on angular momentum, we have shown how rigorous representa-
tions of spins may be constructed with the aid of harmonic-oscillator operators;
and in particular, the Holstein-Primakoff representation, and approximations
to it, were introduced. Let us review the latter from a new point of view. First,
the spins are defined by the three nonvanishing matrix elements:

I The present section has been modeled on an excellent review of spin waves by Van Kranendonk
and Van Vleck [5.1] from which this quotation is taken.
140 5. From Magnons to Solitons: Spin Dynamics

<nil Sf Inl + I) = <nl + 11 Sri nl)* = - ~ i,j(n l + I)(2s - nl ) (5.4)

<nIIS:lnl) = S - nl, O:s:;; nl:S:;; 2s, Ii = 1•

The quantum numbers of other spins (nJ ) have been suppressed, because the
components of SI cannot change these; but all occupation numbers nh n2 , ••• ,
nN must be specified in a definite state of the N spins, and therefore there are
exactly (2s + I)N such states if all spins have the same magnitude s.
Matrix elements of harmonic-oscillator operators have a structure very si-
milar to the above. For example, if we symbolize the integrals

where IfIn(X) is the nth harmonic-oscillator function, by means of the Dirac

bracket notation, <nl(op)lm), then for the very important operators XI
coordinate and PI momentum, we have

<nllxllnl + I) = <nl + Ilxlln)* = J2!0)(nl + 1)

(5.5)
<nllpllnl + I) = <nl + Ilpllnl)* = -iJIi~O)(nl + 1)
and the quantum number nl may also be defined as the eigenvalue of an operator,
now the energy operator

p~
2m + "2 mO) XI -"2l l0)i lnl) =
122
nl i0)
i inl) . (5.6)

The connection between harmonic oscillator and spin is now estabished, by use
of the dimensionless canonical variables P, Q, defined by

QI=xl-YT
rmw PI = PI / I (5.7)
-YlimO)

by comparison of (5.4) and (5.5).

In terms of the P, Q the matrix elements are

<nIIQI,jslnl + I) = <nl + IIQI,jslnl)* = ~ ,j(nl + I)2s

<nIIPI,jslnl + I) = <nl + IIPI,jslnl)* = - ~ i,j(nl + I)2s (5.8)

<ntl s - ~ (P; + Q; - I) Inl) = s - nl ,

 5.1 Spin Waves as Harmonic Oscillators 141

quite similar to (5.4) for small values of the n/. Note that for values of n/ ::::: s,
the harmonic-oscillator approximation becomes quantitatively incorrect, al-
though the matrix structure is still qualitatively similar to the correct one. The
error however becomes catastrophic only when some n/ equals or exceeds the
value 2s, for although (5.8) continues to define matrix elements for the harmonic
oscillators, there is no corresponding structure in angular-momentum space.
So it must be understood that the whole theory which will be developed on
the base of similarity between spins and harmonic oscillators will only be valid
for low occupation numbers of every harmonic oscillator.
In the Hamiltonian of (5.1) we now make the substitutions,

s: = s- ~ (P;+ Q7 - 1) (5.9)

and in accord with the above instructions, systematically discard cubic and
quartic terms. One obtains the "linearized" Hamiltonian (i.e., the equations of
motion are linearized, the Hamiltonian itself is of course quadratic)

~in = Eo + gJiB(H + Ho) t: ~ (P; + Q7 - 1) - Js ne~st (p/P j + Q/Qj) (5.10)

neighb.

where

H _Jsz (5.11)
0- gJiB

(z = number of nearest neighbors of any given spin) is precisely the molecular

field constant which is introduced in the molecular field approximation, in a later
chapter. The constant energy term

(5.12)

is the energy of the completely saturated state in which all spins are parallel to
the applied field.
The linearized Hamiltonian bears an obvious resemblance to the Hamil-
tonian of lattice vibrations, which also involves interactions among neighboring
harmonic oscillators. An important difference arises from the presence of
"velocity-dependent forces" PiP}, to which may be attributed the difference
between the spectra of spin waves and that of sound.
We now make the plane-wave transformation to the new set of generalized
coordinates and momenta
142 5. From Magnons to Solitons: Spin Dynamics

note

Qt = Q-k etc., and (5.13)

[p* Q ] _ J kk,
k> k' - i

where the components kx, ky, k z are integer multiples of 21t/L. A discussion of
this is given following (5.112). This transformation diagonalizes (5.10) for which
we find

~In = ~ 21 (PtPk + QtQk - l)liw(k) + Eo = ~ nkliw(k) + Eo. (5.14)

k k

It may be verified that the operators, Ph Qt" etc., which refer to different
momenta, commute, and so we have N new oscillators, which are uncoupled
in the above approximation. For each of these, the quadratic "number
operator"

(5.15)

has integer eigenvalues = 0, 1, 2, .... Fortunately, it is not at all necessary that

each of these eigenvalues be restricted to the range ~ s, and it is sufficient for
the validity of the linearization procedure merely that ~ nk ~ Ns.
As for the energies of the plane-wave "magnons,"

liw(k) = HgJ1B + Js ~6 (1 - cos k.a) (5.16)

:::: HgJ1B + Jsa k + O(k4)

2 2

the sum runs over the vectors ~ which connect a typical spin to the z nearest
neighbors with which it interacts; a = [a [ for the simple-cubic (sc) lattice, and
the wave vectors k are restricted to the range [k x [, [ky [, [k z [ < 1t/a.
Let us now include the dipolar interactions, to see what is their effect on the
magnon energy spectrum. Defining the direction cosines, a, p, and y of the unit
vector joining the spins,

(5.17)

we find three contributions to the dipolar Hamiltonian of (5.2), after omitting

cubic and higher terms outside the scope of the linearized theory

(5.18)
 5.1 Spin Waves as Harmonic Oscillators 143

(S.20)

For the present, it is convenient to imagine that our (cubic) ferromagnet has
the shape of an ellipsoid with one of the principal axes in the z direction. The
The zeroth order term above is then readily interpreted in terms of the classical
demagnetizing factor Nn

Nz =
, I,
a3 L: ,\ (1 - 3yt) + 431t (S.21)

with a and p replacing y above in the similar definitions of N" and Ny" The sum in
(S.21) being independent of '1, provided this point is within the crystal, we may
express the zeroth order dipolar contribution in terms of the demagnetizing fac-
tor, introducing

Mo = gp:s = saturation magnetization per unit volume (S.22)

a

where a 3 = volume per spin, and we find

1
~d.O = -ZVMo(41t
"3Mo - NzMo ). (S.23)

The factor in brackets is the sum of the Lorentz field (41t/3) Mo and the de-
magnetizing field (which is uniform in an ellipsoid ally shaped material) - NzMo,
and is therefore the "effective field" acting at a given lattice site; V = Na 3 is the
total volume.
Aside from this constant energy term and the constant Eo in the exchange
Hamiltonian, (S.lO), the dynamic terms are the nontrivial terms in (S.lO) plus
~d.l and ~d.2' which all together are of the form

~=
I., (VIjPIP, +
L: WIjQIQ, + UI,PIQ,)

- 3S 3/2 L: [(L: DljaljYIj)Q,

j I
+ (L: DljPI,YIj)P,]
I
(S.24)

allowing Vlj' Wlj , and U lj to indicate schematically the coefficients in (S.lO, 20).
The linear term may be eliminated altogether by means of the canonical trans-
formation,

and (S.2S)
144 5. From Magnons to Solitons: Spin Dynamics

where j;,gl are appropriately determined constants. As it happens, perturbation

theory also converges for these linear terms, and the shift in ground-state energy
which can be calculated by treating %d,1 by second-order perturbation theory
agrees identically with the result of the exact canonical transformation above.
Clearly the result is small, for it is second order in such quantities as

(5.26)

which vanish at any point'l which is a point of symmetry. This would be at every
point not on the surface of the cubic material if DI} were not of such long range.
Calculation of this term is left as an exercise for the reader, a tedious one.
The complete elimination of %d,1 still leaves the first line in (5.24), that is
~in + %d,2 unaffected; and this is now to be diagonalized by a transformation,
first to running waves, (5.13),

1
% - "2 ~ [A(k)Q:Qk + B(k)P: P k + 2C(k)Q: Pi] + const (5.27)

where, with liw(k) given in (5.16),

A(k) = liw(k) + Axx(k) - Azz(O)

B(k) = liw(k) + Ayy{k) - Azz(O) (5.28)
C(k) = Axy{k).

The various A's are seen by comparison with the original equations, (5.10, 20),
to be dipolar lattice sums,

A xx (k) =
N L:
~ i,)
D I} (1 - 3a2 .)e ikoR "
I}
(5.29)

and Ayy , Azz are given similarly by replacing al} by PI} or YI}' Also

(5.30)

and Ax" etc., can also be obtained by obvious permutations. These dipolar sums
occur in several other problems of interest in solid-state physics, and have been
investigated numerically, quite thoroughly, by Cohen and Keffer [5.3].2 As an
example of the size effect, these authors give Axik) calculated for a spherical
'1
sample of radius R, and near the center of the crystal

(5.31)

2 Is the energy properly extensive (oc N)? This is answered in the affirmative by Griffiths [5.4].
 5.1 Spin Waves as Harmonic Oscillators 145

(omitting constant multiplicative factors). We see a not entirely unexpected

phenomenon: as k is decreased to a value ~ lOjR , the dipolar sums in a finite
crystal begin to differ significantly from their value in an infinite material, and
the shape of the surface, and the position of the origin become of importance.
However, precisely at k = 0, this lattice sum approaches a well-defined limit
related to the demagnetizing factor defined in (5.21), and we shall make use of
this fact.
Assuming, then, that the coefficients A(k), B(k), and C(k) are known, it is a
straightforward matter to perform a canonical transformation to a new set of
normal modes

(5.32)

choosing the numerical coefficients ale, ... , die so as to diagonalize the Hamilto-
nian yet preserve the canonical commutation relations,

(5.33)

all other commutators = O. The final result is,

7C' = ~ i (P;* p ~ + Q~*Q; - 1)/iw'(k) + const (5.34)

with
/iw'(k) = ,jA(k)B(k) - C2(k) (5.35)

an anisotropic function of the direction of k, as is seen from the definition of

A(k), B(k), and C(k) given in (5.28), and as plotted in Fig. 5.1.
In ferromagnetic resonance an oscillating electromagnetic field is applied to
the sample. UsualIy, the electromagnetic wavelengths are sufficiently long that
only the k = 0 mode need be considered, and we therefore obtain a resonance
phenomenon at applied frequencies in the neighborhood of w'(O). To calculate

Fig. 5.1. Spectrum of en·

ergies liw' (k) = .jA (k)B(k)
- C2(k). Spread, and gap
at k = 0, are caused by
L...-...--------Ikl anisotropy
146 5. From Magnons to Solitons: Spin Dynamics

the magnitude of this frequency, it is reasonable to set A"y(O) = 0 [note the

resemblance of this quantity, defined in (5.30), to the small term discussed in
(5.26)]; whereas the other coefficients may be expressed in terms of the various
demagnetizing factors to obtain

(5.36)

This differs from a well-known empirical law

Iiw(O) = gJLs../[H + (N" - Nz)M][H + (Ny - Nz)M] (5.37)

only when the true magnetization M becomes far less than the saturation magne-
tization M o, as happens near Tc
It should be noted that the effects of dipolar forces (and of pseudodipolar
anisotropy which may be treated similarly) are limited to the long wavelengths.
Thus the predictions of the theory based on the exchange Hamiltonian alone
are sufficient for obtaining many of the gross properties of the ferromagnet, and
also for calculating the spin-wave energy over most of the Brillouin zone ex-
cluding the neighborhood of k ::= O. See, for example, Fig. 5.1 for a plot of w'(k) ,
where a typical frequency spectrum is given. The actual,complicated anisotropic
curves quickly become asymptotic to the idealized parabolic dispersion law of
a pure exchange ferromagnet. This is a justification for the systematic neglect of
dipolar and other anisotropic effects in many fundamental investigations of
ferromagnetism and in the remainder of this chapter. For spherically shaped
samples, N x = Ny = N z = 4rc/3 and therefore only the applied field H (if any)
appears in the expression for the frequency of the long wavelength magnons. In
the remainder of this book, the geometry will always be assumed to be spherical
because of this convenient simplification of the formulas, which more than
compensates for the loss of generality.

5.2 One-Magnon Eigenstates in Ferromagnets

So far, we have nQt taken advantage of the unique fact that the ferromagnetic
ground state of :;;r'is known, where, reversing the sign of the applied field,

(5.38)

and for this direction of applied field, the state in question consists of all spins
"down" and will be hereinafter denoted

1J'0 = 10) energy Eo = -NHgJLss - ~ NzJs 2 (5.39)

 5.3 Two-Magnon States and Eigenstates in Ferromagnets 147

the "vacuum" state. There are N different orthogonal and normalized states
containing one spin deviate each

(5.40)

corresponding to all the choices of R I • Now, amazingly enough, one finds that
when Zis applied to the state 1fI1' it generates other states ofthe same type but no
states with two or more spin deviations are introduced. This would not be the
case if we had included the anisotropic interactions in the Hamiltonian, and it
is a simplifying property of the Heisenberg ferromagnetic exchange Hamiltonian.
Thus Z may be diagonalized within the N-dimensional subspace of the func-
tions of (5.40).
If, as we shall assume, there is translation invariance (and periodic boundary
conditions), the energy is immediately diagonalized by the introduction of plane
waves

III = _1_"
'Ylc - /-..::....
eilc.R'1/I
'YI and (5.41)
'" N i

and the energy eigenvalues are

E(k) = Eo + liw(k) (5.42)

with Eo and liw(k) exactly as defined in (5.12, 16). This is precisely the result we
would have obtained in the harmonic-oscillator approximation to the isotropic
Hamiltonian by setting n" = 1 and all other n", = 0 in (5.14).

5.3 Two-Magnon States and Eigenstates in Ferromagnets

We have found that the ferromagnetic ground state, and the one-magnon eigen-
states have precisely the energies predicted in the linearized harmonic-oscillator
approximation. This is very encouraging, but not entirely unexpected in view of
the agreement with the earlier semiclassical analysis. One might even be tempted
to predict that states of two or more plane wave magnons are approximate ei-
genstates with an error of no more than O(1/N). But this prediction IS only partly
correct, as we shall see in the present study of two-mag non states. For although
most of the two-plane-wave magnon states undergo negligible scattering and
energy shifts (which may be ascribed to the nonlinear corrections to the har-
monic-oscillator approximation), a number of bound states appear. These are
totally unexpected on the basis of our previous considerations, although they
are of considerable importance in the demise of the spin-wave picture at or
below the Curie temperature. Fortunately, the two magnon problem can be
solved exactly an any number of dimensions, and for any magnitude of the
148 5. From Magnons to Solitons: Spin Dynamics

spins s, and so the relative importance of these bound states can be estimated
under the various regimes [5.5-7].
Let us introduce a notation, in which to develop qualitative arguments
for why nonlinearities should be insignificant at long wavelengths. This will be
followed by an exact analysis. Let the normalized, two-spin-deviate basis states
be

IfIIJ = e'J Sf st I0) = IfIJI • (5.43)

For s> t, there are N(N + 1)/2 states in this orthonormal set. For s = 1/2,
the nonexistence of states of type ifill reduces the number to N(N - 1)/2. In any
case, the energy and wavefunctions of N of these may be found without further
calculation, using only previously derived results. To see this, consider the two-
plane-wave states

(5.44)

where e is an appropriate normalization constant. When k = 0, (5.44) is just

a one-magnon eigenstate, to which the total spin raising operator S+ = LSt
has been applied

(5.45)

Now while S+ (as well as S- and SZ) commutes with the exchange part of the
Hamiltonian, it is a raising operator for SZ. (The total spin is, of course, a good
angular momentum).

[S', S+] = S+ (for 11 = 1) (5.46)

that is,

Thus, for the Hamiltonian :;r of (5.38),

(5.47)
and therefore,

E(Ok') = Eo + IIw(k') + gflBH = Eo + IIw(k') + IIw(O). (5.48)

[n the absence of an external field H, w(O) = 0, and this particular two-magnon

state is degenerate with the one-magnon state to which it is related by a spatial
rotation. Since it is in any case an eigenstate, there is no "scattering," and there
is no energy shift in a magnetic field beyond the simple change in Zeeman
energy. By continuity, we may expect that if k and k' are both reasonably small,
 5.3 Two-Magnon States and Eigenstates in Ferromagnets 149

E(kk') = Eo + liw(k) + liw(k') + 1aE(kk') (5.49)

with aE(kk') - + 0 when either k or k' - + 0; and this is indeed the computed re-
sult, if the states, (5.44), are taken as a set of variational states, and the energy
is computed by the variational formula

(5.50)

But such a computation would not be exact-nor even adequate, for several
important reasons, as follows:
I) The functions IfIkk' are not orthogonal. Their failure to be orthogonal,
although it is only to O(lIN), leads to what Dyson has denoted 3 the "kinematical
interactions" which must be taken into account in solving for the energy eigen-
values. Note that for s = i, the set of two-magnon states is overcomplete: there
are i;N(N + 1) functions to describe i;N(N - 1) physical states, which shows
that the functions in that case cannot all be orthogonalized, even in principle.
2) These functions do not diagonalize the Hamiltonian, except when k or k'
vanish; to the extent that both wave vectors are finite, there is scattering, i.e.,
"dynamical interactions" in Dyson's language, which may even lead to bound
states. To see this, it is necessary to go beyond the approximate treatment and
diagonalize ~ exactly within the proper subspace of states 1fI1}. Fortunately, the
Hamiltonian has no matrix elements connecting any of these states with a state
outside the subspace, so that the problem is quite well defined, as it was in the
case of the one-magnon eigenstates.
3) If the number of magnons ~nk is not 1 or 2 but is O(Ns), the magnon-
magnon interactions add up to a finite contribution to the energy of each mag-
non, and must be taken into account in a correct statistical mechanics even at
low T. Equation (5.49) can be generalized to this purpose, by writing

(5.51)

an expression that must be valid at finite, but low, density of excitations, with
aER the real part of the binary interaction energy [calculated in (5.78)]. The
imaginary part of this term also has a direct interpretation in terms of scattering
lifetimes. We define the inverse scattering lifetime of a given magnon (k) in the
presence of a finite, but low, density of other excitations as

2
l/T(k) = N ~ a£I(kk')n k , • (5.52)
k'

Thus, all the important properties of the low-density many-body magnon

3 The interaction of magnons was first studied correctly by Dyson [5.8]; this work was greatly
expanded by Boyd and Callaway and placed on a rigorous footing by Hepp [5.9].
150 5. From Magnons to Solitons: Spin Dynamics

fluid may be expressed in terms of the complex binary interaction function

oE(kk') that we shall calculate in this section.
The exact 2-magnon eigenstate is a linear combination of all possible con-
figurations of 2 spin "flips", i.e., is of the form
(5.53)

Let us solve the eigenvalue equation,

~rp = Erp (5.54)

for the amplitudes/,}. There is no need to use the normalized configurations as
the normalization constants may be considered absorbed in the /'}' Also,
eigenstates belonging to different energies are automatically orthogonal. One
takes the inner product of both sides of the SchrOdinger equation, (5.54), with
every basis state

(OIS,Sj,
to obtain the equations obeyed by the amplitUdes /')' which are,

(E - Eo - 2gf.lBH - 2sJz)/,) + s ~ (J"}/,,, + JI,,!n})

"
(5.55)

where each of the bonds J I}, J,,}, J I" = J, when their respective scripts W) (nj)
and (in) are nearest neighbor pairs, and vanish otherwise. This equation is
correct for all values of s. Even when s = t, the unphysical amplitudes/'I' etc.,
cancel between both sides of (5.55). In that case, nevertheless, it is convenient to
define fictitious /'1 in order to simplify the calculation. There are at least two
manners of doing this: The way chosen by Bethe in his solution of the linear chain
problem (see section on his one-dimensional solution, ahead) was to set/'I + fJJ
= /,} + fil when j, i are nearest neighbors, a sort of "boundary condition"
ensuring that the homogeneous equation is obeyed at every pair of sites. But in
general, it is more satisfactory to treat the special case s = t on the same footing
as s > t, and allow (5.55) with i = jto define the unphysical amplitude/'I' Note
that in any event, J 1} vanishes unless i, j are nearest neighbors, so that with this
exception, the right-hand side of (5.55) vanishes.
Periodic boundary conditions require

where R} - RJ == (L, 0, 0) or(O, L, 0), (5.56)

etc., for a volume £3 in three dimensions, a square of area £2 in two dimensions,

or a length L in one dimension. And if the right-hand side of (5.55) did in fact
vanish everywhere (instead of nearly everywhere), then the above boundary
conditions could be met by the plane-waves
 5.3 Two-Magnon States and Eigenstates in Ferromagnets 151

with the Cartesian components of the wavevectors

k ;., ... an d k 'x, ... = 21t .

L .mteger. (5.57)

These are the eigenstates when either k or k' = O. For nonzero k, k', it is ne-
cessary, because of the interaction, to take linear combinations of such plane
waves to obtain the eigenstates. It is convenient to separate out the center of
mass motion by introducing the center of mass wave vector, and coordinate,

K=k+k' (5.58)

and a relative wave vector, and coordinate,

k - k'
q=-- (5.59)
2

and in terms of these, to express JI} in the form

ft, = AeiK•R{L: eir·1"(q)} == AeiK.RF'(r) (5.60)

q

where A is a normalization constant, required for L: 1ft, 12 = I.

Because spin operators on different sites commute,

F(r) = F( -r) or equivalently,f( -q) = J(q). (5.61)

In writingft, in the simple form above, one takes advantage of the fact that
the interactions J I, depend on the coordinates RI and R, only through the relative
coordinate r and therefore the total momentum (wave vector K) is a constant of
the motion. We shall solve for F(r), or rather,f(q), at fixed K. Once the solutions
are obtained, they can be expressed in the original variables k, k' by means of
(5.58,59).
Insertingft, in the form above into the set of difference equations (5.55), we
have

(E - Eo - 2gpB H - 2sJz)F(r) + 2SJ~(cos ~ '~)F(r +~)

= J(r)[ (cos ~ .r)F(O) _ F(r) +2F(-r)] (5.62)

in which the vectors ~ connect a spin to its z nearest neighbors for which the
interaction J(~) = J =1= O. At all other distances J(r) vanishes. Next, we multiply
152 5. From Magnons to Solitons: Spin Dynamics

the equation for P(r) by exp( -iq.r), and sum on all r to obtain the equation for
f(q)

[E - Eo - 2gpB H - 2sJ~( 1 - cos f·1 cos q.I)J (q)

= ~ ~ cos q.1 ~ (cos ~.I - cos k.l) f(k). (5.63)

We have made use of the usual identity,

..L "" eHk-k')"

N L.! _ i:
(5.64)
, - Uk,k"

For brevity, define a symbol which will be used repeatedly:

(5.65)

which is, basically, the "incoming" 2-magnon energy Eo + Ilw(k) + Ilw(k') of

(5.49) Proof: assume a simple-cubic (sc) lattice in 3D, a square (sq) one in 20,
and for either of these (as well as for the linear chain in 10) we have the obvious
symmetries yg(q) = y_g(q) = yg( -q). In terms of the original k,k' we can
write

h(q) = y(kk') = Eo + 2gp H + sJ~(2

B
, - cosk.l- cosk'.I)

= Eo + Ilw(k) + Ilw(k'). QED. (5.66)

Bound states, if they exist, must be outside the continuum of states lying between
the maximum and minimum vaue of h(q) and so at fixed K must lie either lower
than

Eo + 2gp H + 2sJ~(1
B -Icos f·ll)
or higher than

Eo + 2gp H + 2sJ~( 1 + Icos

B f .11)· (5.67)

Let us start by an analysis of the scattering solutions in the continuum. It is

conventional to distinguish between "incoming" and "scattered" waves. This is
easily done in (5.63) by singling out

(5.68)

to be the amplitude of the "incoming" wave and solving for the remaining
amplitudes f(k) [= 0(1)] as a smooth function of the continuous variable k in
 5.3 Two-Magnon States and Eigenstates in Ferromagnets 153

the large N limit. Equation (5.63) can be put in a most compact form by defining
D-dimensional vectors as follows:

C(q) = (cos q",I\, ... ) (5.69)

and
v = (V"" ... )

where

v'" = ~~(cos~"'c5", - COSk",.c5",)f(k) k =1= ± qo. (5.70)

This yields for (5.63):

(5.71)

for q =1= ±qo and

(5.72)

at q = ± qo. The procedure is now to solve the integral equation (5.71) for
f(q) and V, then to obtain the l/N correction to the energy, (5.72). In so doing,
we run into expression such as

(5.73)

with a, p = x, y, or z. In such a sum the value of E accurate to O(l/N) given in

(5.72) can be used, but the denominators can vanish at values of q =1= ± qo (in
fact, everywhere "on the energy shell") therefore it is convenient to write E =
yx(qo) - ie, with e an infinitesimal positive quantity that we may allow to vanish
at the conclusion of the calculation. It is now permissible to replace the sum in
(5.73) by an integration.
The integrals are complex

Ma,p{qo) = 2J(2:r f dDq(cos 1/2Kaoa - cos qa c5a) cos qpop

X [Yx(qo~~'Yx(q) + i1to[Yx(qo) - yx(q)]] (5.74)

where we use the identity,

lim e - O{l/x-ie) = P.P.(I/x) + i1to(x)

1S4 S. From Magnons to Solitons: Spin Dynamics

with o(x) the D-dimensional Dirac delta function and P.P. (l/x) == x/(x 2 + e2).
Thus the contribution to the integral on the energy shell is entirely imaginary,
the real part coming from the entire Brillouin Zone.
We combine (5.70,71), and with M the D X D matrix of which Ma.P is the
matrix element, and C and V column vectors, we have

V = M.[C(1/2K) - C(lJo) + V]. (5.75)

This is formally solved by multiplication by [1 - M]-I, with 1 the D-dimensional

unit matrix. Inserting the result into (5.71, 72) yields

f(lJ) = ()
2J ( ) C(1J)·[1 - M]-I.[C(1/2K) - C(lJo)]
YI lJo - YI IJ
(5.76)
and

We observe that thisf(lJ) and the l/N correction to the energy both vanish at lJo
= ±iK; these are the special cases k or k' = 0, for which we already anti-
cipated the absence of scattering.
The integrals (5.74) making up the 3 X 3 matrix M are difficult to express
in terms of elementary functions (although they simplify in the long wavelength
limit ka, k' a - 0, precisely the limit studied by Dyson [5.8] because of its ap-
plicability to low-temperature and low-energy behavior). In leading semiclassi-
cal approximation, we just ignore M which is O(I/s), and evaluate

oE(kk') ~ 4JC(lJo)·[C(1/2K) - C(lJo)]

= -(1/s)[hw(k) + hw(k') - hw(k - k') - hw(O)] (5.78)

with hw(k) given in (5.16), as usual.

In 1D M is just a number.
Equation (5.74) yields

M - 2JTd (cos 1/2K - cos q) cos q

- 21t _" q 4sJ cos l/2K(cos q - cos qo) + ie

= 2s co! 1/2K (cos ~ K - cos qo)(1 - i cot qo) (5.79)

or, in terms of k and k'

M = s cos 1;(~ + k') sin ; k sin ~ k'[ 1- i cot ~ (k - k')] . (5.80)

Insertion into (5.76) yields the scattering amplitudes characteristic of the con-
 5.4 Bound States in One Dimension 155

tinuum of scattering states, expressed in terms of the unperturbed k values of

(5.57). Later [see (5.150)] we shall re-examine this point; energy and momentum
can be conserved with just two values of k, provided the k values are shifted
somewhat. This is part of the so-called Bethe's ansatz; its applications to date
have been limited to 1D, its relation to the above, somewhat obscure.
We now turn our attention to the two-magnon bound states, and look for E
lying outside the range of values given in (5.67). As there is no corresponding
(real) qo there is no special amplitude (5.68). The eigenvalue equation (5.63) is
now an homogeneous equation, and
J(q) is given by

(5.81)

with

v= M· V, the matrix Map(E) given in (5.73) (5.82)

The bound state energy E is adjusted until V is an eigenvector of M. This deter-

mines V to within a multiplicative constant, required for normalization.

5.4 Bound States in One Dimension

The easiest application of this theory is to the linear chain. The eigenvalue
equation reduces to

1- J
- (I )
1t
L
+" d (cos 1/2K - cos q)cos q
q E - Eo - 2gf1.B H - 4sJ + cos 1/2K cos q
(5.83)

For greater clarity, one introduces dimensionless parameters. Let,

1
A = (E - Eo - 2gf1.B H - 4sJ)/4sJ cos '[K .

and therefore,

1 = - 1" dx 1 (cos x- cos 2x )

2s1t ~ A + cos x cos 1/2K

- 2s
IAI- 1)(1
-~(1- ,JA2 + cos A)
1/2K .
(5.84)

There are solutions only for A < - 1, proving that the bound state exists only
below the continuum, and that there is no bound state above the continuum
(which would correspond to positive values of A). This is not surprising in view
1S6 S. From Magnons to Solitons: Spin Dynamics

of the fact that the effective interaction is attractive: two spin deviations have
lower energy when they are nearest neighbors than when they are farther apart.
On the other hand, one might have thought of the effective interaction when
two spin deviations occur on the same site, as a repulsive (or "hard core") poten-
tial which could lead to bound states above the continuum. The actual calcu-
lation shows this latter point of view to be false.
The equation is easiest to solve analytically when s = -t, in which case we
find

and therefore,

bound state. (5.85)

A similar formula will be obtained when we study solitons, (5.249). Equation

(5.85) may be compared to the lowest energy of a scattering state, i.e., the bot-
tom of the continuum, which occurs at q = 0 [cf. (5.67)]

(5.86)·

The bound state lies lower than this by a "binding energy" JE in the amount of

(5.87)

which ranges from

for K { to J for K = 1t •

A schematic plot is given in Fig. 5.2 for arbitrary value of the spin s.

Fig. 5.2. Energy of 2-magnon states in

linear chain. Shading indicates continu-
K
-'If o 'If um, single line represents bound state
 5.5 Bound States in Two and Three Dimensions 157

S.S Bound States in Two and Three Dimensions

The results obtained by Wortis [5.5-7] show that in two dimensions there is at
least one bound state below the continuum/or all K, and that for some K there
may be two bound states. In three dimensions, he found that there are no bound
states whatever for small K, and it is only for sufficiently large values of K that
one, two, or up to a maximum of three bound states make their appearance.
This is a most important result, for the absence of any bound-state solution over
a significant range of K confers validity to the linear spin-wave theory in three
dimensions, even though the linearization be invalid in one or two dimensions.
Before considering the region of small K, which because of the low spin-
wave and bound-state energies will be of the greatest importance in thermody-
namics, let us first examine the special case when all components of K are equal
to 7t, the maximum possible value, in which case

cos K"
"2 = cos K¥
"2 = cos K.
"2 = cos "2
7t
=0

and all the integrals in the eigenvalue equation become trivial. Because the
denominators are now all constant, the nondiagonal matrix elements M",¥ de-
fined in (5.73) all vanish, while the diagonal ones are all equal. The bound state
will therefore be D-fold degenerate. Note that the continuum has collapsed to
a single point at

y" == y,,(q) == Eo + 2gp H + 2sJz

B (5.88)

and therefore the integral equation reduces to an algebraic one. The bound-state
solutions of (5.82) are easily found to have energy

E=y,,-J, (5.89)

by solving this algebraic equation. The bound state is identified as the complex
formed by keeping the two spin deviates nearest neighbors; and the degeneracy
can be understood as the D distinct directions along which two spins can be
nearest neighbors.
The existence of at least one bound state for every value of K in two dimen-
sion is guaranteed by the behavior of the various integrals near q = O. Let us
denote the energy of the lowest state in the continuum by E min , then the inte-
grands in the vicinity of q = 0 all contribute on the order of

~ f E - numeratorD'q2 q q
E min -
d

for suitable positive constant D', and appropriate nonvanishing numerator. This
contribution, as a function of E, ranges from zero to - 00 as E varies from - 00
158 5. From Magnons to Solitons: Spin Dynamics

to E m,n , and in the limit, E m'n - E = oE - 0, varies as ::::: log oE. Thus one
expects, even without the benefit of explicit calculation, that the "binding
energy" oE will depend exponentially on the various parameters. Indeed, for
K" = K)' ::::: 0, Wortis found

where C = cos(K,,/2); for more general K he obtained the binding energy in

terms of various elliptic integrals.
Turning finally to the more realistic case of a three-dimensional structure,
for which the integrals cannot be evaluated analytically, one may find it ad-
vantageous to represent them by Laplace transform methods. That is, the
substitution
1
f0 dt e-
00

- = Dt
D
combined with the definition of the Bessel functions of imaginary argument

I ..
l,,(z) = 1t [ e Z COS" cos nx dx

permits the replacement of three-dimensional integrals by a rapidly converging

one-dimensional integral. But these substitutions are not required to determine
the threshold for the existence of a bound-state solution.
For this, we may presume K" = K)' = K .. which yields the lowest edge of the
continuum and is more likely to produce the bound state than some less isotropic
direction. In this case the three diagonal matrix elements M", are all equal,
and the off-diagonal elements are also equal to each other. This results in
V" = V)' = Vz and in an eigenvalue equation

1 = M",,, + 2M",),
2J cos q,,[3 cos(K,J2) - cos q" - 2 cos q)'J
=N ~E - Eo - 2gpBH - 12sJ + 4sJ(cos q" + cos q)' + cos qz)cos (K,,/2)
= E-Eo-2f{PBH-12SJSin 2(K)2)[I_ E-Eo-2gpBH-I2sJ L 1 ]
24s 2J cos2 (K,,/2) N r E - yx(q) .
(5.90)

Use has been made only ofthe cubic symmetry, and of the equality of the three
components of K to arrive at the last, simplified expression. The bound-state
threshold occurs for that K at which E just drops below the bottom of the
continuum at yx(O). This occurs for K"o

(5.91)
 5.5 Bound States in Two and Three Dimensions 159

Replacing the sum in (5.90) by an integral in the usual manner, we use the first
line to determine the threshold K"o

1 - -1 + cos 1/2K,,0 (1 W)' . 2 1K _ s

- 2s cos 1/2K,,0 - , I.e., sm 4"" ,,0 - 2s +W- 1' (5.92)

where W is Watson's integral (see Note at end of section)

W--1-fJf dq"dqydqz -1516386 (5.93)

- (27t)3 _It 1 - 1/3(cos q" + cos qy + cos qz) - . .

Thus, the bound state exists in the range

K" 0.516386
0::;; cos "2 ::;; 2s + 0.516386 (5.94)

provided K" = Ky = K z. This results in 1400 ::;; K" ::;; 180 0 for s = t, 157 0 ::;;
K" ::;; 1800 for s = 1, 167 0 ::;; K" ::;; 1800 for s = 2, etc. As K is increased
beyond the minimum, the other solutions make their appearance. For example,
we have already found that there are three solutions at K" = Ky = Kz = 180 0
,

which is the largest total momentum wavevector in the positive direction. This
small range of momentum space over which bound-state solutions are available,
even in the most favorable case of spins one-half, is in sharp contrast with the
results of the one- and two-dimensional calculations; and in the correspondence
limit s - 00, the bound states simply disappear.
In Fig. 5.3, a sketch is given of the regions in which one, two, and three distinct
bound states exist. The energy of the bound state increases with K (as does the
binding energy). Thus (5.91) yields the minimum energy of a bound state in 3D,
with (5.92) giving the value of K"o' At temperatures, low compared with this

Fig. 5.3. Octant of Brillouin zone

showing regions of 0, 1, 2, and 3
bound-state solutions for two spin

°
waves, S = j.. For s > j., the
region of solutions grows, and
for s -+ 00 occupies the entire Bril-
louin zone. (Sketch after [5.5])
160 5. From Magnons to Solitons: Spin Dynamics

minimum energy -7- kB' spin-wave theory with scattering corrections is essen-
tially exact.
Note: Concerning the magnetic origins of G.N. Watson's integral:
... it is strange that anyone of his own free will would have investigated such
integrals. Actually van Peijpe encountered the integral in constructing a theory of
ferromagnetic anisotropy based on spin wave theory. Unable to evaluate it in
closed form, he resorted to graphical integration. H. A. Kramers, van Peijpe's
thesis supervisor, then put the problem ... to R. H. Fowler who communicated
it to G. H. Hardy, whereupon, to quote Watson, "the problem then became
common knowledge in Cambridge and subsequently in Oxford whence it made
the journey to Birmingham without difficulty" [5.10].

5.6 One-Magoon Eigenstates in Heitler-London Solid

It is not to be supposed that the Heisenberg Hamiltonian has exactly the same
spectrum of solutions as the Heitler-London Hamiltonian, due to the nonortho-
gonality of the electronic wavefunctions used in the latter. As this is discussed in
the chapter on exchange, we refer to that chapter for further elaboration, and
proceed forthwith to calculate the one-magnon eigenstates in the H-L theory.
This is simple to do in principle, for by translational symmetry the eigenfunctions
are plane waves. We then obtain the energies as functions of the interaction
and normalization parameters.
Suppose, then, there is a single unpaired electron at each of N sites labeled
RJ , and that we use as the basis product functions
N
'P(ml, ... ,mN) = II rp(rJ - RJ)XmJ(eJ) (5.95)
J=l

where r J and eJ are the space and spin coordinates of the electron, respectively,
and the 2N sets of m/s (±i) provide the same number of degrees of freedom as
in a similar Heisenberg model. In evaluating energy, normalization, etc., the
totally antisymmetrized wavefunctions

(5.96)

must be used. Let us denote the totally ferromagnetic product function by 'Po,

_ (1 1)
-2' ... , -2
'Po = 'P (5.97)

and one spin-deviate states by 'P"

 5.6 One-Magnon Eigenstates in Heitler-London Solid 161

i = 1, ... , N (5.98)

and use the same indices on the corresponding totally antisymmetrized (deter-
minantal) wavefunctions, 'Po and 'PI, constructed from the above by means of
the prescription (5.96). Because of the overlap of the one-electron wavefunctions,
the 'PI do not form an orthonormal set. Therefore, instead of the usual Schr6-
dinger equation, the energy variation principle yields an equation involving an
overlap matrix Q

(5.99)

for an eigenfunction assumed of the form

cP = L.:. It 'PI . (5.100)

I

The Hamiltonian matrix elements are

(5.101)

and the overlap matrix elements are simply

(5.102)

including the spin variables in both integrations.

~presumably depends on the spatial coordinates only, and is translationally
invariant. Therefore both sets of matrix elements depend on i andj only through
the relative distance R lj , and the equations can be solved by setting

(5.103)

Consequently, the one-magnon eigenvalues E(k) are simply ratios,

(5.104)

By measuring these energies relative to the ferromagnetic energy (k = 0), we find

the spectrum of excitations

E(k) _ E(O) = Q(O)[~(k) - ~(O)] - ~(O)[Q(k) - Q(O)] _ Dk 2 0(k)4 +

Q(O) {Q(O) +
[Q(k) - Q(O)]} - 1 O(k )2 +
(5.105)

The various parameters ~j and Qlj may be evaluated by the theory of deter-
162 5. From Magnons to Solitons: Spin Dynamics

minants (cf. section on "the method of Lowdin and Carr" in the nonortho-
gonality problem; however the present integrals are somewhat more difficult to
evaluate than the ones solved in that section). They have reasonable magnitudes
(Le., they are not off by factors of N); and in fact (5.105) would be in exactly the
form of the Heisenberg theory result given in (5.42, 16), were it not for the k-
dependence of the denominator. At long wavelengths this affects k4 (and higher-
order) terms in the expansion of Eel) - £(0), and may eventually lead to some
experimental determination of the importance of effects of nonorthogonality.
It may be presumed that a linearized version of H-L theory can be con-
structed which for many magnons in three dimensions would reproduce some of
the results of "harmonic-oscillator spin-wave" theory in the Heisenberg ferro-
magnet. But this has not yet been attempted. Nor have the two-magnon problem
and the existence of two-magnon bound states been studied in the H-L theory,
although this would involve only slight generalizations of the analysis of the pre-
ceding section.

5.7 Nonlinear Spin-Wave Theory

Ifit is desired to go beyond the harmonic oscillator approximation and calculate
nonlinear corrections to the Heisenberg Hamiltonian, without the rigors of
many-body scattering theory, a fairly direct manipulation of the spin operators
themselves lends itself to this purpose. We start with the Holstein-Primakoff
representation (cf. Chap. 3),

The a's are boson operators, satisfying ordinary commutation relations

(5.107)

with n, = a"ta, the Boson number operator measuring the deviation from spin
"down". The Heisenberg Hamiltonian takes on the form

+ gJlBH~, (n, - s). (5.108)

It is convenient to rationalize the square roots (cf. Maleev transformation,

Chap. 3), by the similarity transformation

and a·,• •, (1 -2s

-112 n 1/2
aj-+ ( 1 - ;;) aj -+a· - ')•
(5.109)
 5.7 Nonlinear Spin-Wave Theory 163

This transformation preserves the number operator as well as the commutation

relations (5.107) within the physically relevant Hilbert space.
Equation (5.108) now becomes

(5.110)

the constant Eo being the usual ground-state energy first defined in (5.12). This
non-Hermitean operator has all the eigenvalues of the Heisenberg Hamiltonian,
plus an infinite number of unphysical ones. Whether an eigenstate is physically
admissible or not can be tested by seeing whether it is an eigenfunction of the
positive semidefinite operator

(5.111)

with eigenvalue 0, or not.

For the purposes of spin-wave theory, one now transforms the above 3r'to
running waves. We now give a discussion ofthis and a definition of the BriIIouin
zone which wiII be useful in the next chapter as well. Let

(5.112)

where n, m, I = integers. It may be verified that this transformation is canonical,

by showing that the ak's obey a set of commutation relations identical to (5.107).
For if one inverts the transformation, he obtains

aI r-- _1_ " e-ilr.R'a I

/-"-' aIr* -- _1_ " eilr'R'a*I
/-"-' (5.113)
'" N i '" N j

and verifies they also satisfy the commutation relations (5.107). It is sufficient in
(5.112) to use the set of smallest integers: n = 0, ± 1, ± 2, ... , ± iN.~, m = 0,
± 1, ... ,±iNy, etc., for a solid N" x Ny x N z. This defines the first Brillouin Zone;
an arbitrary k can always be reduced to the first BZ by subtracting an integer
multiple of (2rc/a, 0, 0) or (0, 2rc/a, 0) or (0,0, 2rc/a), where a= lattice spacing.
Denoting by K" any integer. multiple of these, we note

hence alr+K. = air [by (5.113)], and all relevant operators lie in the first BZ.
The Hamiltonian in terms of air's is
164 5. From Magnons to Solitons: Spin Dynamics

(5.114)

Let us isolate, in this large sum of terms, the contribution of operators diagonal
in the occupation number representation (that is, of operators such as constants
and functions of nk = a'tak) from the nondiagonal terms. Terms bilinear in the
a's, such as the last term in the magnetic field, are automatically diagonal by
virtue of momentum conservation. Quartic terms are not, although a subset of
them, for which

or

(see Fig. 5.4) involve only nk's and so will be included in the important diagonal
part, which we shall call ~D. As the Fourier transform of J lj consistently occurs,
it is convenient to denote it J(k), defined as

J(k) = _1 ~ eik.(R.-RJ)J (5.q5)

2N I,j Ij·

In the cubic structure, J*(k) = J( -k) = J(k). For example, in the simple cubic
structure nearest-neighbor interaction

J(k) = 2J(cos k"a + cos kya + cos kza) . (5.116)

And in this manner one obtains the compact expression, which reproduces ex-
actly the results of(5.51, 78) based on binary scattering

:
k2 k2

DIRECT
k4
~ k4

:
k2 k4
EXCHANGE

k4
~ k2
Fig. 5.4. Direct and exchange contributions
to nonlinear magnon Hamiltonian
 5.7 Nonlinear Spin-Wave Theory 165

$t"o = Eo + ~ [2sJ(O) -
k
2sJ(k) + gf1 H]n k
B

l ~ [J(O)
- N
k,k'
+ J(k - k') - J(k) - J(k')]nknk' . (5.117)

No bound states are found but, as already noted (in Fig. 5.3), none are expected
at large values of s.
The study of $t"o involves only minor generalizations of the theory of the
linear Hamiltonian. The only important question is the following. Given a state
occupied by n magnons of wave vector k I, n2 of wave vector k2' etc., how much
energy is required to add one magnon of wave vector k to this state? Let W(nk)
and W(nk + 1) indicate the eigenvalue of $t"o in the two states Ink) and at Ink);
then

2
- N 2: [J(O) + J(k - k') - J(k) - J(k')]-nk'}at Ink)
k'

(5.118)

and therefore

W(nk + 1) = W(nk) + [2sJ(O) - 2sJ(k) + gf1 H] B

- 2 ~ [J(O) + J(k - k') - J(k) -

N J(k')]n k, . (5.119)
k'

In terms of the magnon energy liw(k) as previously defined, for example, (5.14),

1
W(nk + 1) - W(nk) = liw(k) - N 2: [liw(k)
s k'
+ liw(k') -liw(k - k') -liw(O)]nkl

== e(k) (5.120)

with the sum representing the corrections to linear spin-wave theory. It is im-
portant to note that k' = 0 spin waves do not contribute to this correction, for
this confirms the rotationally invariant nature of the approximations. The non-
linear magnon energies are denoted e(k), and, for nearest-neighbor interactions,

(indicating average over the directions of k by a bar). So, by increasing the

occupation numbers nk, all magnon energies can be reduced, as we shall see
elsewhere in a discussion of thermal effects. However the above inequality does
not hold for arbitrary long-range interactions, and is reversedin nearest-neighbor
jerrimagnets, so that it appears to be a special property of simple ferromagnets.
166 5. From Magnons to Solitons: Spin Dynamics

There is also an approximate superposition principle: the energy required

to add two magnons of wave vectors k and k' to a given state equals the sum of
the energies required for each, 8(k) + 8(k'), to within a correction term O(ljN)
arising from the interaction. The proof is given as Problem 5.1.

Problem 5.1. W(nk +

I, nk, +
1) == W(nk +
1, nk,) +
W(nk' nk" + 1) - W(nk.
nk') +
w. Find the correction term w; verify that it is O(ljN).

Conduding this section, we note that the leading terms in Ijs are exactly the
same as those obtained in the binary collision (2-magnon) analysis given ear-
lier. Correction terms O(ljs2) compared with the leading term are calculable by
perturbation theory.

5.8 Perturbation-Theoretic Correction

So far we have not focused any attention on the non diagonal terms 3f" = 3f' -
3f'n. At his juncture we shall try to estimate the effect of such terms by pertur-
bation theory. In principle, if carried out to all orders, this is a systematic
means of diagonalizing the Hamiltonian. Once this is done, the proper and
improper eigenstates can be separated out by the test that has been established
in (5.111). In practice, the perturbation expansion might not converge, and we
shall only carry it out to second order by the following means.
Consider the similarity transformation generated by T,

op - e-Top e+ T .

This is a unitary transformation if and only if T* = ~T, that is, if T = i X Her-

mitean operator. Such a unitary transformation could not diagonalize the
Hamiltonian, which in its present form is non-Hermite an and which when dia-
gonalized will be explicitly real, diagonal, and therefore Hermitean. The con-
nection between the two is possible only through an unrestricted T. To lowest
order in the interaction, it is determined by expanding,
T2
eT=l+T+-+· ..
2

and choosing T so as to eliminate 3f" from the Hamiltonian. Of what remains,

we shall keep only the lowest-order diagonal terms. That is,

3f' - (1 - T + ~2)3f'n( 1 + T + ~2) + (1 - T)3f"(1 + T)

:::::: 3f'n+ ([3f'n, T] + 3f") - T3f'nT + i (T23f'n + 3f'n T2 )

- T3f" + 3f"T + O(3f"3) (5.121)
 5.8 Perturbation-Theoretic Correction 167

together with the relation

(5.122)

specify the transformed Hamiltonian to the desired approximation. [Compare

(5.122) with (5.126), where this equation is solved for the matrix elements of T).
The quantum numbers of the eigenstates of ~D are the sets of occupation
numbers, and let a, b, ... stand for such sets. One finds for the new eigenvalues
of (5.121) after solving (5.122)

:it'.
a,a = (%, )
D a,a -
"
"T ~;.~~.a
(%,D ) b,b _ (%,D ) Q,a (5.123)

a familiar expression of the second-order perturbation correction to the energy.

Substituting ~' into this expression, one obtains the leading correction to ~D

x ~A(~k=1:-'-+_k----'3'-.--,--k-".2_-_k-:,4~)[,--,J(,---k~2)---,-_J(.;.k-".2_-:-:-k-'I""'))-':::[J:-Oc(k,..=-3)"-.-_J-'.(k-=3c...-----,k:!.<..;4))
W(nk' + 1, nk, - 1, nk, + 1, nk• - 1) - W(nk" nk" nk" nkJ
X [(1 + nk,){l + nk,)nk,nk• - (l + nk,)(1 + nk.)nkh,)). (5.124)

This expression may be verified by use of the following properties of the inter-
action: J(k) = J( - k), and J(k + Kn) = J(k), where Kn is one of the reciprocal
lattice vectors, e.g., as defined following (5.113). The symbol A(k), which is often
used to generalize the Kronecker delta to crystal physics, is defined to vanish
except when its argument k = 0 or K n , in which case it equals unity. The con-
tribution of terms for Kn is called umklapp, after the German meaning to be
brought back into (the Brillouin zone). The denominator equals

Whereas the linear spin-wave theory gave magnon energies proportional

to s; and whereas the nonlinear corrections in ~D' (5.119), were independent of
s; we now find that the perturbation-theoretic correction just obtained vanishes
as '-'S-I and therefore will be negligible compared to ~D in the correspondence
limit s -- 00. This is due to the s-dependence of the denominator. Higher-order
corrections vanish even more strongly as the number of energy denominators is
increased.
Thus there appears to be a well-defined semiclassical limit s -- 00, which in
practice means s } t, where ~D is a sufficiently accurate approximation to the
correct Hamiltonian that it may be used with confidence, provided
168 5. From Magnons to Solitons: Spin Dynamics

%n can therefore be considered as the effective Hamiltonian of nonlinear, semi-

classical spin-wave theory, with perturbation theory and projection operators
required to obtain further corrections.
Note: Even though the elimination of %' is not very important to the low-lying
energies when s is large, it does contribute substantially toward the satisfaction
of the sum rule,

(5.111)

for physical states. That the sum rule is not obeyeed in the original spin-wave
representation, can be seen by expanding the operators in plane waves. Retaining
only the diagonal operators nk> we find there are (2s + I)! ways of pairing the
creation and annihilation operators and thus obtain for the above,

(2s + I)! ( N1 I; n")2,+' (5.125)

which may not be large when I;n" { N, but is nevertheless nonzero.

After the similarity transformation, with

(5.126)

where the denominator equals e(k,) + e(k3) - e(k2) - e(k4 ) + O(l/N), we

obtain for the diagonal part of the sum-rule operator, (5.111),

(5.127)

with F = (1 + n",)(1 + n".)n"tn". - (1 + n".)(l + n".)n",n".. the ellipsis in-

dicating non diagonal and higher-order corrections. The improvement is sche-
matically indicated in Fig. 5.5
While the above analysis gives one confidence in the nonlinear spin wave
theory developed earlier, it certainly does not apply to ID or 2D where the exis-
tence of bound states at arbitrarily low energies ensures the breakdown of
 5.9 Antiferromagnetic Magnons: The One-Dimensional XY Model 169

ordinary perturbation theory. Indeed, careful study will show the situation to be
quite different in the low-dimensional systems from what we have just concluded
in three dimensions. We return to this point after a diversion into the study of
antiferromagnets.

f/)
IAI
~
~'OO% I
....J
«
o I
I

f/) /
/
>- Fig. 5.5. Schematic effect of
~
Q.
Z '" '" '" transformation T on elimination
of unphysical states in nonlinear
~ J...::.';;"-~~_ _ _..L- Ns spin-wave theory. Upper curve is
l4.
o NUMBER EXCITED MAGNONS
before, and lower curve after the
transformation
it

5.9 Antiferromagnetic Magnons:

The One-Dimensional XY Model

The following theorems can be proved about the Heisenberg antijerromagnet

(5.128)

with J 1J > 0 for (ij) = nearest neighbors and vanishing otherwise:

1) With some minor restrictions on the J 1J , and on the lattice geometry, the
ground state is a nondegenerate singlet (Stot = 0). What is more, the lowest
energy of total spin S is lower than that of any state belonging to S + 1.
2) For spins one-half there is no energy gap against excited states, in the
limit N -00. With few exceptions the excited states in a many-body system
(N - 00) form a continuum down to the ground state. The content of theorem
2 is that the continuum of S = 0 states starts at a lower energy than the con-
tinuum of S = I states, which in turn starts lower than S = 2, etc. The proofs
of these theorems are given elsewhere, 4 they are related to the existence of long
wavelength, low energy disturbances-spin waves. For when a spin deviation

4 See [5.11-13]. The reader interested in a general introduction to the various aspects of the
theory of antiferromagnetism should read the extensive review article [5.14].
170 S. From Magnons to Solitons: Spin Dynamics

propagates at very long wavelengths, the relative orientation of nearest-neighbor

spins is maintained to a very good approximation, and consequently the cost in
energy must be low.
In 1931 Hans Bethe, followed by Hulthen and others, succeeded in solving
for the eigenstates of the Heisenberg linear chain model with nearest-neighbor
interactions. We shall discuss this in the following section. Unfortunately, be-
cause of its difficulty, the Bethe solution has not given any insight into the
question of long-range correlations, excited states, etc., and so it must be con-
sidered physically incomplete. But by truncating the full Hamiltonian some-
what, in fact by eliminating the z components of the interactions entirely, one
arrives at a model which is completely soluble in one dimension for spins one-
half and nearest-neighbor interactions. It has many of the features of the Bethe
solution, but it is quasi linear and fully transparent. We call it the XY model. s

~Xy = ~ t. (StS i+l + H.c.), s=2

1
and (5.129)

The vacuum (all spins down) is an eigenstate, with energy O. By translational in-
variance, one may guess that the "one-particle" states (N - 1 spins down, one
spin up) are plane waves

(5.130)

with easily computed energy eigenvalues,

E" = cos ka with Ikal < 1t. (5.131)

Periodic boundary condition exp(ikNa) = 1 results in the discrete set,

21t. 1t(2p)
k = - X mteger = - - p = 0, ±1, ... (5.132)
Na Na

Next, a product of two plane waves does not vanish in the configurations stst,
as it should, but a determinant does. However, a determinant is anti symmetric
under the interchange of the coordinates, whereas spins on different sites com-
mute and therefore have a wavefunction symmetric under interchange. The
following choice thus imposes itself:

(5.133)

5 See [5.12]. Katsura [5.15] has calculated the magnetic susceptibility in the XY model, and
Niemeijer [5.16] the time-dependent spin correlations.
 5.9 Antiferromagnetic Magnons: The One-Dimensional XY Model 171

Whenj is increased to N - I, then to N, and finally to N + I, the second spin

becomes the first and the wavefunction changes discontinuously from the upper
form to the lower unless the proper boundary condition is imposed. Because the
position of the origin of the numbering system is completely arbitrary for the
cyclic problem we are solving, such discontinuities at a particular site are
inadmissible. The resolution of this difficulty is to take, instead of(5.132),

k = n(2p + I) (5.134)
Na

(and similarly for k'), an antiperiodic boundary condition. The generalization to

any number of spins up is straightforward. Let

'file"" ... · -- C ""-'

" F IeI ••••12Ie,.......• S+S+
11 12'" 10) (5.135)
Il, i 2,'"

where C is the normalization constant, and F is the determinant

(5.135a)

8p = + I when the spins are in a natural order, i1 < i2 < i3 < ... , or an even
permutation of this order, and 8 p = - 1 when the spins are arranged in an odd
permutation of the natural order. Ifi. is the farthest spin, then the translation of
i. from N to N + I involves a reordering equivalent to an odd permutation for
n + I odd, and an even permutation for n + 1 even. Thus

n
n +I= odd - k = -(2p
Na
+ I)
and
n
n +I= even-k = Na(2p). (5.136)

This is a very interesting situation. For although the many-spin wave-

functions at first appear to be essentially independent particle wavefunctions,
yet when one "particle" is added (that is, when one more spin is turned up) all
the other plane-wave states are modified. This is in the nature of a cooperative
effect, due to the effective hard core repulsion [from (St)2 = 0] of two nearby
spin deviations. Whether the k's are members of the even or of the odd set, the
energy corresponding to the wavefunction above is

E k •• k , .... • cos kla = L: cos N

= L: n X
{22P + 1 .
I
(5.137)
1=1 I PI
172 5. From Magnons to Solitons: Spin Dynamics

Note that all k's must be distinct, or else F == O. The ground state energy is
achieved by allowing all the states of negative energy to be occupied, that is, all
k's in the range

n k 3n. N 3N
"2:::;; a:::;; "2 I.e., 4":::;; p :::;; 4 . (5.138)

However, we must consider two ground states: the ground state for the even k's
and the ground state for the odd ones. Denote these by E~ and E~', respectively,
with values

n n
E~ = ~cos N(2p + I) and E~' = ~ cos N(2p)

where lim E~ = E~' = - N/n (5.139)

N-oo

with the sums in either set restricted to the range, (5.138). The separation of
energies E~ and E~' depend on whether N is divisible by 4, or merely by 2, or
whether it is odd. In any event, it is of the order of magnitude of the smallest
elementary excitations.

-1T f - - - - - - - , / - - - - - + - - - 4 - - - - j 1T FERM I LEVEL

Fig. 5.6. Energy as a function of ka in XY model, showing occupied and unoccupied states.
Bubble and dot indicate "quasihole" and "quasiparticle" created in elementary excitation

Either ground state may be represented by an energy-level diagram shown

in Fig. 5.6. The Fermi level intersects the cosine curve at ka = n/2 and 3n/2,
and all allowed states below this are filled and all those above it are empty in
either ground state. Excited states correspond to occupying states above the
Fermi level, or emptying states below it. If k is a wave vector of either category
in (5.136), as measured from ± n/2a, the energy of an elementary excitation,
e(q) is described by a doubly degenerate spectrum

e(q) = Isin ql setting q == ka (5.140)

in addition to any correction due to changes in parity of the state. Each element-
 5.9 Antiferromagnetic Magnons: The One-Dimensional XY Model 173

ary excitation represents an anti ferromagnetic magnon, and it should be noted

that unlike the ferromagnetic magnons, their energy is linear in q for small
q. This appears to be a model-independent property of all antiferromagnetics.
These results are sufficient for energetic considerations, such as computing
the statistical mechanical properties of the model at finite temperatures. They are
not adequate for the dynamics. For example, in processes used experimentally
to probe the excitation spectrum of a solid (optical absorption, neutron scatter-
ing, etc.) we must also know the relevant matrix elements. Generally, the re-
sponse of the system is given in terms of the square of the matrix element,
divided by the energy difference between the excited state and the ground state.
The low-lying spectrum may in fact not be observable if the matrix element
connecting it to the ground state vanishes in the particular experiment designed
to probe the spectrum.
To write dynamic operators in the representation in which the Hamiltonian
(5.129) is diagonal it is necessary to use the fermion representation given in (3.99)
of Chap. 3, the operators Cj. Briefly,

(5.141)

and similarly for the Hermitean conjugate expression. With some care taken at
the periodic end (SitS! + H.c.) the Hamiltonian becomes a quadratic form
diagonalized by a transformation to running waves, see (5.112).

Problem 5.2. Carry out the indicated diagonalization, paying special attention to
the operator QN, Eq. (3.100a); at the end of the periodic chain, for the two
distinct cases of total occupancy r
c: C i = even or odd.

A typical dynamic matrix element involves operators such as

N
Tk == N-J/2 1: S".n eikna •
n=!
(5.142)

In the fermion language,

Tk = N-J/2 f.
11=1
21 [c!' exp(in 1: ctc})
i<n
+ H.c.]e ikna . (5.143)

The cn's are then expressed in the running-wave operators in which the Hamil-
tonian is diagonal [analogous to (5.112, 113)]. The exponent involves expressions
which are very complicated (but not impossible) to evaluate.
In this connection, we quote some preliminary results obtained by Taylor
and his associates [5.17, 18] in their study of Praeseodymium chloride (PrCI 3).
174 5. From Magnons to Solitons: Spin Dynamics

The crystal structure of this compound (qh symmetry) is such that the nearest
neighbor Pr ions lie on the c axis, leading to predominantly one-dimensional
interactions between nearest-neighbor Pr ions on each c-axis chain. These
interactions involve the Kramers' doublet ground state of each ion and are
characterized by a Hamiltonian

:%'= ~ JL: (StSi+l + H.c.) + L:g/lBHzSz,1 + L: y(g'tSi + H.c.) (5.144)

where Hz and g'1 are externally applied magnetic and electric fields. The ab-
sorption of photons by the magnons occurs, if momentum and energy are
conserved, via the coupling terms. The momentum of infrared photons is very
small. We need to evaluate the matrix elements of T for very small k between
the ground state and the low-lying excitations, viz., (Tk)a,o and in principle
obtain the absorption line shape. The experimental line is shown in Fig. 5.7,

1.5
[AJ
1.0

os

o Fig. 5.7 Absorption coefficient A for

Prel 3 at various temperatures. [5.17,18]

and is seen to peak at 3 cm- J • Instead of computing this line shape, one can
try to calculate this peak (which corresponds to the average absorption energy
as well). The result is

(5.145)

where m is any fixed site. The numerator is evaluated from the fermion re-
presentation, and yields 1/2rt - cos (ka/rt 2). The denominator was evaluated
numerically [5.17, 18] to yield 0.0386, at k = O. The result is Eo = 1.50J. With
J = 2.0 cm- J , this leads to Eo = 3 cm- J , in excellent agreement with the loca-
tion of the experimental absorption peak. Of course, the evaluation of the
exact matrix element and fitting of the absorption line shape, Fig. 5.7, would
be even more compelling evidence for this theory, but the onerous calculations
have not yet been performed.
 5.10 Bethe's Solutions of One-Dimensional Heisenberg Antiferromagnet 175

5.10 Bethe's Solution of One-Dimensional Heisenberg

Antiferromagnet

Progress in solving the Heisenberg antiferromagnet, even in one dimension, has

been slow. Although Bethe [5.19] gave the wavefunctions in 1931, the ground-
state energy was not obtained until 1938, by Hulthen [5.20]. It was subsequently
generalized by Orbach [5.21], who discussed an anisotropic Hamiltonian, and
Walker [5.22] who studied the analytic properties of the Orbach solutions.
Finally, fully 32 years after the initial progress, Des Cloizeaux and Pearson [5.23]
calculated the one-magnon spectrum and Griffiths the magnetic susceptibility
[5.24].
It is desired to calculate the eigenstates of

1
for s = 2' g = I. (5.146)

g = 00 (Ising model), g = °
Other limits which can be studied by related methods include:
(XY model), g = - I (transformable into fer-
romagnet by trivial rotation of every second spin). For all g > - I the ground
state belongs to M = 0, i.e. contains Ns "particles".
We start with 2 "particles", using a procedure valid only for spins one-half
and assign to the unphysical amplitudes Iii the value determined by

/;,1 + /;+1,1+1 = /;,1+1 + /;+1,1 = 2/;,1+1 (5.147)

where the physical amplitudes/;i = hi for} *- i) obey the equations

(5.148)

This is the left-hand side of (5.55), with s = j, J = - I, H = 0, and z = 2.

(5.147), playing the role of a boundary condition, ensures that the right-hand side
of that equation always vanishes. The homogeneous equation is solved by plane
waves, via Bethe's ansatz:

(j:2: i) (5.149)

with/;j in the range} < i given by symmetry,/;} = hi. The phase factor If! ranges
over the interval -1t, +1t. (In the XY model, (5.133), the effective If! is a con-
stant ±1t). We make no attempt to normalize solutions here, as it is not required
in the calculation of energy eigenvalues and represents some fairly involved
analysis.
176 S. From Magnons to Solitons: Spin Dynamics

Inserting the above form of j';j into the boundary condition and performing
some algebra, we obtain

1 1 1
= cot
2cot -'"
2 2 -k - 2
cot -k' (S.lSO)

with a second, periodic, boundary condition,

(S.lSl)

determining the modified k, k'

k = 1t(2p) + '" and k' = 1t(2p') - '" (5.152)

N N

with p,p' = integers. Thus, a two-magnon state is described by two wave vectors
and a phase shift; there is refraction but no scattering. The sum k + k' is in-
dependent of", and is the center of mass wave vector K which parametrized the
solution in an earlier section of this chapter. The difference k-k' corresponds to
the variable fJ. The bound state that appeared below the continuum threshold
at fJ = 0 is now above the continuum, due to the change in sign of the interaction.
Now, as k approaches k', (S.150) shows that", approaches ±1t, the upper sign
applying if k < k'. From (5.152), if p = p' -1, k = k' and '" = 1t so that
/u= exp [ik(i+ j)] x 2cos+1l == O. Therefore, Ip-p' I ~ 2.
The energy, measured from the ferromagnetic level Er = iN, is given
by the above eigenvalue equation as

E - Er = -(1 - cos k) - (1 - cos k') (5.l53)

just the energy of two noninteracting magnons; the interactions are included
implicitly, via the phase shifts in k, (S.150, 152)
Bethe's ansatz consists partly in the statement that in the many-particle state,
the amplitudes are subject to phase shifts that are simply given, as the sum ofthe
two-particle phase shifts. Thus, for a given ordering i ::;; j ... ::;; n,

all n! - 1
+ ( remaining )J
permutations . (5.154)
of the k's

If we need};,j, ... for any other ordering of the particles, we obtain it from the
symmetry under permutations:};,j, ...,m, ... = };.m, .... }.... so we need to know it only
 5.10 Bethe's Solutions of One-Dimensional Heisenberg Antiferromagnet 177

in the given interval. In this region, the amplitude f consists of a product

over plane-wave factors, summed over all permutations ofthe wave vectors with
the 'fI's antisymmetric in their subscripts and satisfying the equations

(5.155)

1t(2p) +t 'flu'
k= 1 p = 0, ±1, .... (5.156)
N

The sum in this equation, as well as in the next, is over the set of k's present in
(5.154) The energy, measured relative to the ferromagnetic reference energy
Ee = N/4, is

E - Ee = -
,=1
t (1 - cos k,) . (5.157)

Note that if we translate the entire chain by one site, that is, let i,j, ... - i
+ l,j + 1, ... , the wavefunction, (5.154), is multiplied by a phase factor
(5.158)

The exponent is identified as the total momentum, modulo 21t, which is again
independent of the phase shifts 'flu', This is important, because the phase shifts
themselves are now rather large. Each 'fIw is 0(1); there are a number of them
contributing to each k, and the total shift in k is O(n/N) or a substantial frac-
tion of k.
In the two-particle problem, there was no solution of(5.150, 152) for k - k'
:::::; 0. Similarly, in the many-particle state, one must choose the interval between
k's such that no p = 0, and such that for all p and p',

Ip - p'l ~ 2 (5.159)

for a non-vanishing solution to exist. The ground state, for N/2 particles subject
to this restriction, has the set of {p} :

{p} = 1,3, ... , N - 1. (5.160)

Notice an interesting effect of the S;S;+1 "interaction," which is to spread the

set of iN integers p (restricted over the range N/4 < p < 3N/4 in the ground
state of the XY model) to cover the entire range of phase space at present.
The regular spacing permits us to replace sums by integrals in the limit
N - 00, and thus, with x = piN and LJp = iNdx
178 5. From Magnons to Solitons: Spin Dynamics

E - Er = -"2
N
! [1 -
I
cos k(x)] dx = -N! sin 2 Tdx
I k(x)
where (5.161)

2 cot /fI(x, y) = cot k(x) _ cot key) and (5.l62)

2 2 2

+ "2 ! /fI(x, y)dy .

1 I
k(x) = 21tx (5.163)

as 1/fII cannot exceed 1t, (5.162) indicates that /fI has a jump discontinuity of
-21t at x = y. More precisely, the derivative is

a/fl(x, y) = -2!5( _ ) + .l(~2 + 1) dk(x)/dx (5.164)

ax 1t x y 4 1 + 1/4(~ - 1])2

where we use the short-hand notation

1
~ == cot ~ k(x) and 1] == cot 2k(y) . (5.165)

We also find it convenient to introduce the functions

f(~) = -(d~/dx)-I and f(1]) = -(d1]/dy)-1 (5.166)

and to differentiate (5.163) w.r. to x to obtain

+ "2 L 1 + 1/4(~ _
1 +~ f(,,)/f(~)
dk(x)/dx = 1t 1])2 d1] . (5.167)

With the identity f(~) dk/dx = 2(1 + ~2)-1 this becomes

(5.168)

This type of equation is soluble because the kernel is a function only of the
difference (~ - 1]). One may take Fourier transforms,

f dOf(O)e
~

Pk == ik8 and (5.169)

to obtain a simple solution of(5.168)

F'k = (2 cosh k)-I . (5.170)

The ground state energy (5.161) has the form

 5.10 Bethe's Solution of One-Dimensional Heisenberg Antiferromagnet 179

(5.171)

This is a very famous result, being one of the first exact solutions ever obtained
of a nontrivial many-body problem in quantum mechanics.
In the usual terms of energy per spin, the Bethe-Hulthen result for EIN
represents -ln2 + i = -0.443. We can compare this to the ground state
energy per spin in the XY model [(5.139) X liN] which is _1[-1 = -0.318.
While this is a variational upper bound to the Heisenberg model ground state
energy, it is evidently not a very close one. On the other hand, t X ground state
energy of the XY model is a lower variational bound on the Heisenberg model
(really, an "XYZ" model) and this yields -31[12 = -0.477 reasonably close to
the exact answer. By way of contrast, we note that the Ising model (S: SI-tl) has
energy -0.250 per spin in these same units, which is a variational upper bound
(although a poor one) to both XY and Heisenberg models. Moreover, 2 X
(-0.25) = -0.500 is a poor lower bound to the XY model and 3 X (-0.25)
= -0.750 is an even worse lower bound to the Heisenberg energy per spin.
(The reasons for these larger discrepancies may be attributed to the lack of
spin waves in the Ising model).

Fig. 5.8. Quasiparticle spectrum tJ(ka) obtained by Des Cloizeaux and Pearson for the
infinite Heisenberg antiferromagnetic chain, nl2 sin ka (upper curve), compared with their nu-
merical results for chain of 48 spins. Also, spectrum for infinite XY antiferromagnetic chain
(lower, dashed curve). Anderson's Iinerarized antiferromagnons coincide precisely with the XY
model. (Note: q = ka, a e n.n. distance.)

Des Cloizeaux and Pearson [5.23], Fig. 5.8, have extended Bethe's analysis
to obtain the energy of some of the lowest excited states, which are known to be
triplet states [5.11]. NI2 - 1 values of k are chosen by them, to obtain the
eigenstates of lowest energy belonging to a finite total momentum :Ek. This pro-
180 5. From Magnons to Solitons: Spin Dynamics

cedure is very delicate, however, and somewhat too complicated to reproduce

or to justify here. There is an unanswered question of whether bound states
participate among the low-lying excitations. If they do, then Bethe's ansatz
formalism is incapable of describing them since the validity of the various equa-
tions above seems predicated on having real k's. Nevertheless, it is entirely
possible that further mathematical study of (5.152) and those following may
allow the extension of the known results to complex k plane and that a complete
classification scheme will result. For example, following Orbach one might
consider the S:Sf+1 coupling terms with variable parameter g, and study the
energy levels as g is adiabatically increased. At g = 0, we recover the XY model
(of which the solutions are completely determined) and then increase it to g = 1
which is the present Heisenberg model. As g is further increased, the Ising model
is approached. In fact, g = ± 1 are two "critical points": for g > 1 the proper-
ties approach those of an Ising antiferromagnet (e.g., the ground state energy
approaches -ig per spin at large g, the excitation spectrum has a gap), and for
g < -1 they resemble those of an Ising ferromagnet. It is only in the range
-1 :::;; g :::;; + 1 that results are obtained similar to the gapless spectra we have
seen here. For more details, the papers of Orbach [5.21] or Walker [5.22] or the
more recent studies by C.N. Yang and C.P. Yang [5.25] or the generalization
by Baxter [5.26] to the completely anisotropic linear chain can be consulted.
The case g > 1 has been analyzed in [5.27, 28].
Des Cloizeaux and Pearson's results for the one-magnon triplet (S = 1)
excitation spectrum is shown in Fig. 5.8, together with some numerical results
on chains of 48 spins and compared there to the spectrum of the XY model
(g = 0). This comparison will have further significance when we study ap-
proximate theories of antiferromagnetism in the following sections. If we denote
the triplet excitation spectrum 81> it is given by

8, = ~ 7t Isin q I . (5.172)

Yamada [5.29], and Bonner et al. [5.30] noted 8, is most likely the lower edge
of a continuum of triplet excitations, with a curve 82 defining the upper edge

(5.173)

This continuum is illustrated in Fig. 5.9. Undoubtedly, excitations belonging

to S = 0,2,3, ... lie in other continua that partly intersect this. However, in an
experimental situation where one unit of angular momentum is transferred to
the magnetic system, only the triplet continuum can contribute. For such
processes, Hohenberg and Brinkman [5.31] have computed that the matrix
elements peak near the lower, des Cloizeaux-Pearson, threshhold. This topic,
 5.1I Linearized Antiferromagnetic Magnons 181

and its nontrivial extension to the case of an applied magnetic field, is increas-
ingly being studied theoretically [5.32] and experimentally [5.33].
One important result follows immediately from Fig. 5.9, viz., the size of the
Brillouin Zone (BZ) remains 2n for the linear chain antiferromagnet. With only
the spectrum of Fig. 5.8 one might suspect that the BZ could be folded in half,
which is equivalent to the choice of 2 spins as the primitive unit cell. Physically,
this is a very attractive idea: the couple of spin "up" and "down" nearest-
neighbors make an ideal unit cell, and this is indeed the starting point of many
approximate analyses [5.33]. With the complete excitation spectrum, however,
one sees that the translational invariance of the ground state is preserved and
the larger BZ must be retained. 6

Fig. 5.9. The continuum

of magnon (S = 1) excita-
tions for the Heisenberg
XYZ linear chain antiferro-
magnet, indicating the two
bounding curves 8 1 and 82
q given in the text

5.11 Linearized Antiferromagnetic Magnons

The quantization of antiferromagnons in the linear approximation is performed

by analogy with the procedure in the theory of ferromagnetism . Such an ap-
proximate quantum theory, in which only terms quadratic in the P's and Q's
are retained in the Hamiltonian, was originally proposed by Anderson [5.35],
and was very successful in the early interpretation of the dynamics of antiferro-
magnets.' As can be expected , the frequencies are identically the same as found

6 The general topic of correlation functions for the ground state (and also, at finite T) for this
model has been much studied in the recent past. The interested reader may wish to study the
papers in [5.34J.
1 See also [5.14J.
182 5. From Magnons to Solitons: Spin Dynamics

in the classical linearized equations of motion of the antiferromagnet (discussed

earlier); only the amplitudes are quantized as Bosons. Below, we shall indicate in
what vital manner procedures which are satisfactory in ferromagnetism must
be changed for the present problem.
The paramount difficulty is that the ground state of the Heisenberg antiferro-
magnet is not known, with the exception of the linear chain. Alas, it is quite
apparent that the procedure which worked in one dimension is not generalizable
to three because of the heavy reliance on ordering the spin deviates along a line.
The somewhat simpler XY model is similar to the quantum-mechanical hard-
sphere Boson problem, for which also no adequate intermediate- or high-density
theory exists in two or three dimensions.
Fortunately, there exists a perturbation theory of sorts, which gives us a
"handle" on an approximate, variational ground state. It is a series in powers
of lis. The off-diagonal parts of the Hamiltonian are linearized, and diagonalized
as best can be, in the standard way. The starting, or zeroth, point of this pertur-
bation procedure is the classical ground state: the configuration of lowest
energy of the Hamiltonian with spins replaced by classical vectors of fixed length
and position but variable orientations. The choice of this classical ground state
is a soluble problem (see pp. 232, 233). We determine it to be a two-sublattice
antiferromagnet with the following properties:
Antiferromagnetic coupling between nearest neighbors, and a lattice geome-
try such that two spins which are nearest neighbors of a third are not nearest
neighbors of each other. These requirements are met, e.g., in the linear chain,
simple square, simple cubic, and body-centered cubic lattices (but not in the
face-centered cubic lattice). The classical ground state is then the Neel state:
every spin up is surrounded by nearest neighbors which are down, and vice versa.
To give them a name, denote the spins down the A sublattice, and the spins
up the B sub lattice . Perform a canonical transformation on the B (but not on the
A) spins: rotate them by 1800 about the S" axis, to obtain the so-called Neel
state

Sf- + SJ, (j in B) . (5.174)

At first, assume that i is in A and.i is a nearest neighbor of i in B to write the

Hamiltonian in the form

~= 1: 1: [( 21 stst + H.c.)
I~A.J~B
- S~SjJ. (5.175)

It is convenient to take the exchange constant J to. be the unit of energy, so that
it need not be written explicitly. Now, because the last result is explicitly sym-
metric in the two sublattices, one allows i to range over the entire crystal andj
over all nearest neighbors of i, denoted by j(i), and divides by 2 so as not to
double-count bonds.
 5.11 Linearized Antiferromagnetic Magnons 183

Z= 21 2: 2: [21 (stSj + H.c.) - S~SjJ~

all i j(i)
(5.176)

In the "linear" approximation, the Hamiltonian becomes

(5.177)

After the transformation to running-wave Boson operators, it is not yet diagonal.

We eliminate the pair-creation terms by a method first due to Holstein and
Primakoff[3.6]. Subsequently rediscovered by Bogolubov in his theory of Bose
condensation, and now generally known as the Bogolubov transformation, it is:

ak - (cosh uk)ak + (sinh uk)a!k

at - (cosh uk)at + (sinh uk)a-k with (5.178)

One must choose the function Uk so that the terms ata!k are eliminated. It is a
matter of some simple algebra to determine that what is required is

-I
tanh 2uk z 2:
, cos k·~
= - (5.179)

with ~, a vector connecting any spin with any of its z nearest neighbors. This
reduces the linearized Hamiltonian to the diagonal form,

~in = - ~ Nzs(s + 1) + ~ (nk + ~ )(ZS),Jl - tanh 2 2uk • (5.180)

Before discussing the magnon spectrum it is not unwise to digress somewhat

and examine the ground-state energy. In (5.180), it is the energy eigenvalue
when all nk = 0, that is, the vacuum energy. Conventionally, one writes it in
the following form:

(5.181)

The parameter JI may be shown to be bounded between 0 and 1. Indeed, Hul-

then's exact result for the linear chain of (5.171) (z = 2, s = i) yields JI =
0.7726, which is perhaps the largest attainable value. The results in Table 5.1
were computed by the linear theory [5.35] (5.180), for nearest-neighbor interac-
tions. But it is interesting to see the good agreement (within :::::;6 percent) with
the exact result for the linear chain, where the spin-wave theory might have been
expected to fail badly. Although for the one-dimensional magnon spectrum the
184 5. From Magnons to Solitons: Spin Dynamics

Table 5.1 >' for various lattices

Lattice z

Linear chain 2 0.726 (exact: 0.7726 .. for spins 1/2)

square 4 0.632
Simple cubic 6 0.58
Body-centered cubic 8 0.58

agreement will not appear so favorable, it is not farfetched to conceive that the
ground state and excited states of the simple theory may be in good agreement
with the exact result for lattices with higher coordination numbers.
Again for one dimension, using (5.179, 180), we find for the magnon energy,

e(k) = (2s) sin ka (5.182)

with a = lattice spacing. For spins i the result is similar to the exact result for
the XY model, but is too low by a factor 2/re for spins i in the Heisenberg model,
which the spin-wave theory is supposed to approximate. The discrepancy is seen
in Fig. 5.8.
In any number of dimensions, the magnon spectrum

e" = s -J Z2 - CE6 cos k .1)2 (5.183)

is doubly degenerate, and is linear in k for small k[1t should be noted that,
properly speaking, magnons are triplet, not doublet excitations (S = 1)].

Problem 5.3. Derive the magnon spectrum (5.183) from (5.179, 180) evaluated
for nearest-neighbors. Generalize (5.183) to arbitrary range interactions.

The existence of two distinct modes of zero energy, one at k = 0 and the
other at the edge of the Brillouin zone, means that magnons belonging to these
wave vectors can be emitted at no cost of energy, and indicates a high degeneracy
of the approximate ground state. This is a manifestation of the rotational in-
variance of the spin Hamiltonian. By the emission of zero energy magnons, a
state can also be reached in which the A and B sublattices have been inter-
changed. However, the true ground state is nondegenerate [5.11].
The diagonalization of ~in leads to a reduction of the S' spin components
from their saturation magnitude in the Neel state. Calculating this reduction we
find

<<<5S~> == \
1 ~ (s -
IN
t
ISm) = 2Nl ~, ,(-J
I 1 -
1 h 2 - 1) ::::::: 0.078 sc.
tan 2 u"
(5.184)
 5.11 Linearized Antiferromagnetic Magnons 185

The numerical value represents an estimate by Anderson of the integral, for the
simple cubic (sc) structure. Note that for the linear chain this formula yields

lin. chain (5.185)

due to the divergence ofthe integral at long wavelengths. This implies an absence
of long-range order in one dimension even in the spin-wave approximation, a
result in accord with other analyses.
We now compare the results of linear spinwave theory (LSW) with other
calculations, including the most accurate cellular method of Betts and Oitmaa in
a table adapted from one of their recent works [5.37]. LSW is in good agreement
with the last, and presumably best, result.

Table S.2. Comparison of various calculations of the ground state energy parameter eo ==
1 + ~y and of <t5Sr> for the s = ~ Heisenberg antiferromagnet on a square lattice (z = 4)

Method Reference eo <t5S:> y

LSW [5.35] 1.32 0.20 0.64

perturbation theory [5.36] 1.33 0.12 0.66
Variational [5.38] 1.29 0,07 0.58
Variational [5.39] 1.32 0.10 0.64
Cellular [5.37] 1.31 0.25 0.62

The method by which Betts and Oitmaa have been obtaining the ground
state parameters in 2 and 3 dimensions deserves comment. Let us illustrate with
the square lattice, then quote their results for the XY antiferromagnet and for
both Heisenberg and XY models in 3D.
Figure 5.10 illustrates finite cells of 4, 8, 10, 16 and 18 spins on the square
lattice. In each case the infinite lattice can be filled by periodic extensions of the
cell; thus, the wavefunction in the cell is subjected to periodic boundary con-
ditions. It consists of linear combinations of states belonging to the symmetry
appropriate to the ground state. For example, starting with the Neel state, one
overturns a pair of n.n. spins in all possible ways. One then overturns a pair in
any of these states, etc., until all the states which can mix in the ground state are
found. For the 16 spin diamond illustrated in Fig. 5.10 there are 6 generations of
spin flips before the total of 153 basis states is achieved (a great deal less than
the total number of 216 basis states)! For N = 18, 398 states are generated, and
the resulting matrices are easily diagonalized to yield the lowest eigenvalue, the
ground state energy. When plotted vs liN, as in Fig. 5.11, these cellular ground
state energies typically lie on a straight line that extrapolates to liN = O.
From the ground state energy one can readily extract the nearest neighbor
correlation function (t/Jo ISfSj It/Jo) = (t/Jo I SrS~ It/Jo) = (t/Jo I SfSjt/Jo I), with i, j n.n.
186 5. From Magnons to Solitons; Spin Dynamics

<l Fig. 5.10. Finite cells of 4, 8, 10,

r -h 16, 18 spins on the square lattice.
In each case the infinite lattice is
I I l.,..---- \ filled by periodic repetition of the
,L " - f-I
\ cells
l"- '
l\~ V

V \l Fig. 5.11.
s= + (a) Energy per spin of the
Heisenberg anti ferromagnet on
a square lattice (b) Long range order
parameter for XY and Heisenberg
antiferromagnets, showing the exis·
tence of LRO in the ground state
[5.37]

16 0.20

14
----- -------
-~-------- 0.15

--------_.... .----- Xy __ -

/-
1.2 0.10
_--
~-
Heis.
~~
~~
~~
,-~
~~

10 005 -

a b
0.8 L -_ _ _---:-~--_;_----!

-- --
0L-------~1~1~--~~~
.l.
14 1816
1 1
N N

Using the computed ground states, Betts and Oitmaa also obtain the correlation
functions for i,j not n.n. Two spins correlate positively when on the same sub-
lattice, but negatively when on different sublattices. For the antiferromagnetic XY
model, the X and Y correlations are of course equal, but the Z correlation func-
tion is distinct. Calculation of both (901 StSj 190) and (901 S:Sj 190) show them to
decrease surprisingly slowly with distance. These results are evident in Table 5.3
taken with suitable modification from [5.37]. Betts and Oitmaaconcluded thatthe
long range order persists in 2D in the ground states; plotting 21 N ~ ( -1)1-J(St S;>
i,j

vs liN, they obtained the lines in Fig. 5.11 which extrapolate to LRO = 0.116
for the XY model and 0.059 for the Heisenberg antiferromagnet (compared to a
possible maximum of 0.25).
 5.11 Linearized Antiferromagnetic Magnons 187

Table 5.3. Ground state correlations for the 16 spin cell on the square lattice, s = 1/2

XY Heis. A. -F.
4<S~Sj> 4<S1S j> 4<SfSj>

0.182 - 0.562 - 0.468

0.027 0.487 0.285
0.027 0.487 0.285
0.023 - 0.468 - 0.270
0.018 0.458 0.240

LSW theory on the 20 and 3D XY models is extremely simple, although

highly inaccurate for s = 1/2 (see, however, the following section). Starting
with the exact Hamiltonian

:7t' = ~ .L; (st S; + H.c.) (5.186a)

1./

we use the linear approximation to the Holstein-Primakoff operators, and write

the above as

:7t'= ~ s.L; (a1a)

I.)
+ H.c.) (5.186b)

which, after the phase change ai, a1 - -ai' a~ on the B sub lattice alone, yields

:7t' = - ~ s ~ (a1a + H.c.) j (5.186c)

'.1

and after the transformation to running wave operators, (5.112, 113), yields

(5.186d)

in 20, with an analogous expression in 3~.

In the ground state, a symmetry principle requires the eigenvalue of the
operator S:ot to be zero [5.40], thus for s = i the magnon population is:

(5.187)

in the ground state. Evidently, the ground state is achieved by populating the
k = 0 mode only, thus no = iN and nk = 0 for k =1= O. This yields for the
s = i correlation functions
188 5. From Magnons to Solitons: Spin Dynamics

4(~ISfStl~) = (-I)I-J(~I(ataj + afa, + a,aj + ataJ')I~)

= (-I)I-J 1~ (2cos k.R ,j) Uk (5.188)

in any eigenstate~. In the ground state ~o, this sum is N, so (5.188) is ±I de-
pending whether the two spins are on the same sublattice or not. This result is
twice as large as the computed values in Table 5.3 and does not exhibit the rather
mild, but quite definite fall-off with distance of the computed results. Similarly,
the other s = i correlation function is

(5.189)

regardless of sublattice, having the correct sign but some 5 times too large! It
seems that LSW is fortuitously good in the Heisenberg model, for the accuracy
evaporates in the XY model, although it does explain LRO.
Turning to 3D, we compare LSW with other methods (in Table 5.4, adapted
from [5.37]) and in Table 5.5, present the correlation functions in the XY and
Heisenberg antiferromagnets computed by the cell method [5.41]. The LSW
correlation functions for the XY model are the same in 3D as in 2D and remain
in quantitative disagreement with the tabulated values. A striking result of Table
5.4 is that the spin deviation is larger by a factor of 3 or 4 in the cell method than
in the other calculations. Experimental tests to establish the correct value of
this parameter in Nature have not yet been devised.

Table 5.4. Ground state energy parameter)l and spin deviation in 3D Heis. A.-F. spin one-half

s.c.(z = 6) b.c.c.(z = 8)
Method Reference )I <oS~) )I <OSf)

LSW [5.35] 0.58 0.078 0.58 0.059

Perturbation theory [5.36] 0.60 0.064 0.60 0.047
Variational [5.38] 0.54 0.045 0.54 0.033
Variational [5.39] 0.60 0.059 0.60 0.051
Cellular [5.41] 0.4 ± 0.4 0.20 ± 0.02 0.4 ± 0.4 0.19 ± 0.02

5.12 NonIinearities in Antiferromagnetism

In this section, we discuss corrections to the linear spinwave theory (LSW) of

the preceding section.
As previously noted, the only systematic small parameter in which to expand
is I/s. For large s, LSW is capable of yielding the correct answers, the error being
estimated by computation of the next terms in an expansion in I/s. Although this
 5.13 Ferrimagnetism 189

procedure might seem hopeless for s = i, it is possible to make progress there

also. We illustrate by means of the XY model treated above, for which LSW
gave such poor results.
Consider the LSW substitution

(Sf)2 -- ! (a/+ a1)2 = ! (1 + 2n/ + a; + at)

(Sr)2 -- - !(a/ - a1)2 = ! (1 + 2nj - a; - an . (5.190)

The ground state expectation values are t for each, rather than the correct value
i. Thus, if to correct this discrepancy we took "renormalized" operators:
st -- 2- 1/2 a1 and (5.191)

retaining Sf = (ara/ - f) as the counting operator, we would obtain ± 1/2 for

the ground state expectation value of (5.188) instead of ± 1. This is now in
reasonable agreement with the second column of Table 5.3. But this theory still
does not explain the discrepancies between the Tables 5.3 for 2D and 5.5 for
3D, for the ground state correlations we compute by spinwaves in this model
are independent of dimensionality; nor does it explain the observed decay of
correlations with distance R/J •

Table 5.5. Ground state correlations for s = i antiferromagnets in 3D, for a 16 spin cell in
two distinct lattices [5.41]

XY Heis.
RI}/a 4(S1S~> 4(S:S7> 4(S1S~>

Simple cubic (s.c.)

1 0.1230 -0.5464 -0.4327
21/2 0.0297 0.5054 0.3141
31/ 2 0.0246 -0.4926 -0.3107
2 0.0347 0.5174 0.3333
Body-centered cubic (b.c.c.)
(3/4)1/2 0.0965 -0.5395 -0.4167
1,21/2,3 1/ 2 0.0326 0.5163 0.3333

5.13 Ferrimagnetism

Ferrimagnets are generally insulators containing localized spins which are

antiferromagnetically coupled. One example pointed out by NeeI, who coined
the term jerrimagnetism (because of its existence in the jerrites, of which the
lodestone is an example) occurs if the spins on the A and B sublattices of the pre-
190 5. From Magnons to Solitons: Spin Dynamics

vious section are of unequal magnitude SA =/= SB' Other, vastly more complicated
examples of ferrimagnetism exist in theory and in nature,S but their study is a
complex and specialized field. In the present section, we want to accomplish two
goals: to display the spin-wave Hamiltonian for the simple two-sublattice model
of ferrimagnetism, including leading nonlinear terms, and to show the leading
nonlinear terms in antiferromagnetism, obtained from the former by setting
SA = SB' We also show how the same expression, which for unequal spins gives
a magnon energy __ k2 at long wavelengths, will yield the antiferromagnon energy
--k. The mathematics is based on work by Nakamura and Bloch [5.43] and
consists of a straightforward expansion of the square roots in the Holstein-
Primakoff representation. For large spins, agreement with the classical equations
of motion gives some confidence in this procedure, for which there is no other
formal mathematical justification.
Let SA :2: SB' let there be N of each, and put

SA = (l + a)s SB = (l - a)s. (5.192)

Except for the unequal spins, the Hamiltonian is precisely that of (5.175) in the
preceding section. We keep the first three terms in an expansion of the Hamil-
tonian in powers of s:

(5.193)

They are,

%0 = ~ [YO(sBatak + sAbtbk)

+ ,jSASB Yk(akbk + H.c.)] - N ~ SASB (5.194a)

with A(k) defined in the text following (5.124). Finally,

%2 = - 2(4N)2:SASB )3/2 k,···k. ~ (s~Yk,at.bt,bt,bk,b:.bk'

- 2sASBYk,H,-k,at,at.ak,bt,bt.bk,

8 For a comprehensive review and references to the literature, see, e.g., [5.42].
 5.13 Ferrimagnetism 191

where

a, ,-- _1_ '" a e-i".RJ

/-k.. J and b" = J...-u L b/e+I".R, (S.19S)
'" NJSA '" N/9J

and
Yo = z. (S.196)

Next, ~o is diagonalized by the Bogolubov transformation,

a" - a" cosh u" + bt sinh U"

b" - at sinh u" + b" cosh U" (S.197)

which mixes operators of the A and B sites. The degeneracy of the magnon
spectrum in antiferromagnetism is lifted for a =1= 0, and the following two
magnon branches are found:

t.(k) = yos(/,. - a) and (S.198)

where

(S.199)

provided the transformation parameter, Uk> chosen so as to eliminate a"b" +

H.c. from the Hamiltonian, takes the value

tanh 2u" = - y" ..jl - a'- . (S.200)

Yo

In the long wavelength approximation, the magnon energies are

1 - a2
ta(k) ::::: (..ja 2 + (1 - a 2)(ka)2 - a)yos ex: ~(ka)2

tb(k) ::::: [..ja 2 + {l - a 2)(ka)2 + a]yos (S.201)

from which it is easy to see the range dependent on a, over which the lower
branch is quadratic, as in ferromagnets, before becoming approximately linear,
as in antiferromagnets.
The ground-state energy, in this linear approximation, is

(S.202)

(it must be noted that there is a total of 2N spins in the present calculation) with
192 5. From Magnons to Solitons: Spin Dynamics

(5.203)

This quantum-mechanical correction is smaller in ferrimagnets than in antiferro-

magnets, which is not too surprising in view of the resemblance with ferro-
magnets, for which y == O.
The nonlinearities are handled in the following manner. First, ~l and ~2 are
both transformed by the rules of (5.197,200). Then, the diagonal terms in ~l
are combined with the now entirely diagonal ~o to give a first-order diagonal
Hamiltonian which we may denote ~D' by analogy with the treatment of ferro-
magnetism. Finally, the off-diagonal matrix elements of ~l are eliminated by a
canonical transformation, such as T in the ferromagnetic case. The resulting
additional diagonal terms, combined with the diagonal parts of ~z, form the
second-order correction to ~D' The remaining nondiagonal terms are dis-
carded, for their contribution is O(s-Z). These calculations are undertaken in
[5.43] but are too lengthy to be reproduced here. They also analyze the tempera-
ture dependence of the magnon energies (a subject which is treated here for
ferromagnets only, in a separate chapter). Some of their results at T = 0 K are
as follows:
I) There is a small s-independent shift in the magnon energies as given in
(5.198,200).
2) There is a nonlinear diagonal magnon energy, of the form

~ [21r",,(k, k')(ata,)(at;a,,) - r"b(k, k')(ata,)(bt;b,,)

'."
+ ~ rbb(k, k')(btb,)(bt;b,,)] (5.204)

which, together with the linear magnon terms, makes up ~D' One significant
property of the nonlinear Hamiltonian is that the coefficient of ata, turns out to
vanish for k = 0, regardless of the occupation numbers of the other modes.
This expresses the rotationally invariant nature of the approximations leading
to ~D' But differences with the analogous treatment of the ferromagnet can be
noted, due to the presence of two spin-wave branches with different properties;
for example, an increase in the occupation numbers of either branch increases
the magnon energies in that branch and decreases the magnon energies of the
other branch. [However, this appears to be a consequence of the nearest-
neighbor model, and is not a law of universal validity; see the discsusion follow
ing (5.120)].

5.14 Effects of Surfaces on Spin-Wave Amplitudes

The role of surfaces is dominant in the observed properties of magnetic sub-
stances: domains nucleate there, spin waves are "pinned" there by impurities or
 5.14 Effects of Surfaces on Spin-Wave Amplitudes 193

stray fields, and phase transitions at a surface may foreshadow those in the
bulk [5.44]. To study these effects one appeals to a scattering theory which has
some similarities with the two-magnon problem studied above. For definite-
ness, we consider only a single surface (in extremely thin films where the two
surfaces interact, this treatment will have to be modified) perpendicular to one
of the principal crystal axes in a simple cubic structure. The interactions are
ferromagnetic, and J differs from the bulk value 1 only for the surface spins. A
surface anisotropy field h A > 0 is also included, but the bulk ferromagnet is
otherwise isotropic. With the surface located at R; = (X;, Y;,O) and all interac-
tions restricted to nearest-neighbors, we have the Hamiltonian

I
ge= - - ~ ~ S,.S, - J ~ ~ S,.S,
2 Z"ZJ"O z,-o,z... o

-~ J %,=z}=o
~ ~ S,.S, - hA ~ Si.
z,-o
(5.205)

The ground state 10) of all spins "down" has energy

(5.206)

in which we define the bulk energy to be that part which does not depend on the
surface exchange parameters (J, hA ) and we note that each of the N z surface spins
has 4 n.n. in the surface (i.e., two bonds per spin), plus one in the neighboring
plane, for a net total of 3 J-dependent bonds per surface spin. (This should not
be confused with the obvious fact that each surface spin has 5 n.n. rather than 6
in the bulk).
Absent the translational symmetry that enabled us to guess the plane wave
form of the one-magnon states in the bulk, we shall have to construct the set
of excited one-magnon eigenstates, in the form

I",) = (2sN)-1/2
,
~hstIO) (5.207)

and extract from it a set of coupled equations for the amplitudesh by projecting
as follows:

(5.208)

e is the excitation energy, to be compared with liw(k) for the translationally

invariant case. Evaluating the l.h.s. of (5.208) we readily obtain

(Eo + 6s)f(R,) - ,
s ~f(R, + I) (5.209a)

for R, in the bulk; for a spin in the plane adjacent to the surface, the l.h.s. of
(5.208) is, instead,
194 5. From Magnons to Solitons: Spin Dynamics

(Eo + 6s)f(R,) - 5 L:J(R/

6
+ ~) + (J - l)s[f(R/) - f(X/, Y/,O)] (S.209b)

and finally, for a surface spin

(Eo + hA + SJs)f(Xj, Y/,O) - sJ[f(Xj + a, Y/,O) + f(X/ - a, Y/,O)

+ f(X Y j + a, 0) + f(X/,
Io Y/ - a, 0) + f(X Y a)] .
Io lo (S.209c)

There remains translational invariance of sorts, i.e., in the XI> Y/ plane. There-
fore let us seek a solution of the form

j; = C eHk.X,+k,Y,) F(n) (S.210)

where C is a normalization constant, Zj = na and the surface plane is located

at n = 0 (the first interior plane at n = 1, the bulk at n = 2, 3, ... ). Without loss
of generality, we write for the excitation energy of the continuum states

e = 2s(3 - cos k"a - cos kya - cos q) (S.211)

and for the three regions considered above, the Schrodinger equation becomes,

2s(2 - cos k"a - cos kya)F(n)

+ s[2F(n) - F(n + 1) - F(n - I)] = eF(n) (S.212a)

for n ~ 2. At n = 1 we obtain

2s(2 - cos k"a - cos kya) F(l)

+ s[2F(1) - F(2) - F(O)]
+ (J - l)s[F(I) - F(O)] = eF(1) (S.212b)

and at the surface,

[hA + 215(2 - cos k"a - cos kya)]F(O)

+ sJ[F(O) - F(1)] = eF(O) . (S.212c)

As the bulk eigenstates take the form exp(±iqn), we try a linear combination:
let

F(n) = cos (qn + 0), for n ~ 1 (S.213)

in which O(k", k y, q) is a phase shift independent of n. However, this form is not

suitable at the surface, n = o. The substitution (S.213) into (S.212a) solves that
set of (bulk) equations exactly. We are left with (b) and (c), 2 equations in the 2
unknowns, 0 and F(O):
 5.15 Vortices 195

F(O) = cos 0 + (J - 1) cos(q + 0) (5.2 14a)

J J

and

J
F(O) = D cos(q + 0) (5.214b)

where,

D(q) == hA/s + J + 2(J - 1)(2 - cosk"a - cos kya) - 2(1 - cos q). (5.214c)

D(q) is a function of q alone, at fixed hA' s, J, k" and k y but its variation, especial-
ly with k", k y may be important. Solving the coupled equations for F(O), we find
it to be a function of q and indicate this by a subscript

Jsin q
Fq(O) = {(P _ DJ + D)2 sin q + [(P _
2 DJ + D)cos q _ D]2} 1/2· (5.215)

Note that as q - 0, this surface amplitude vanishes

and thus, long wavelength magnons have small amplitudes at the surface, unless
J = D(O). [For this special case, (5.215) yields Fq{O) - 1] .
q-O
When J exceeds D(O), bound states can form below the continuum. Then, the
form (5.213) is not suitable, and one uses F(n) == C exp(-yn). Results by Ilisca and
Gallais [5.44] indicate that when hA < 0 the bound state even acquires negative
energy,8 < 0, and that the surface can no longer be magnetized in the same
way as the bulk.
Recapitulating, we have seen that the discontinuity at the surface, and the
possible surface-specific parameters (change in coordination number, in J,
anisotropy field, etc.) affect the surface magnon amplitudes in a major way.
The same must therefore also be true for crystalline defects: magnons are
expected to be strongly scattered at dislocations, grain boundaries, etc.

5.15 Vortices

Just as dislocations in solids cannot be studied using the theory of small ampli-
tude vibrations, neither can the topological disorder known as a vortex in the
196 5. From Magnons to Solitons: Spin Dynamics

magnetic problem be described in terms of spin waves or magnon excitations.

Vortices have to date been studied only in the classical (s --+ 00) limit, and were
introduced into magnetism by Kosterlitz and Thouless [5.45] in the context of
the 2D classical XY model.
Following those authors, let us consider classical vectors with ferromagnetic
nearest-neighbor interactions, on a square lattice as illustrated in Fig. 5.12. There
we picture a single counter-clockwise vortex of charge + 1. (A similar but clock-
wise vortex has charge -1.) It extends throughout the lattice and because spins
are practically parallel in any neighborhood far removed from the core, a vortex
cannot be "cured" by any finite number of small re-orientations of the spins, nor
can it be caused by them. Once introduced, it is a stable state, although one of
higher energy than the ground state. The vortex represents an extra degree of
freedom of the spin system, independent of those associated with the magnons.
The excitation energy of a single vortex is easily estimated. In 2D one reasons
as follows: the misalignment of 2 neighboring spins is a small quantity, say J().
Starting at some spin at a radial distance R from the core of the vortex and
moving along the circumference in a ccw direction, we must wind through a
total angle () = (2rrRla)J() which is required to be an integer multiple of 2rr, say
2rrn; for the case illustrated in Fig. 5.12, n = + 1. Thus,

J() = naiR. (5.216)

The energy of misalignment at each bond is of the order of J(J())2, with Jthe ex-
change parameter. Thus the total excitation energy or "self-energy" of the vortex
is

= 2rr f dRRa- 2J(nal R)2

L
llE
a

= 2rrn 2Jln(Lla) (5.217)

,
, - "
, ,
, , '"
+
~ i.f ~
J I'" Fig. 5.12. A single vortex of
"charge" + 1. At large distances

~
, , - I
from the core, the spins are lo-
cally almost parallel but there
is still circulation: a spin far to
the left and one far to the right
are anti parallel. The self-energy
is infinite
 5.15 Vortices 197

where we take the lattice to be circular of radius L and area 'ltV, centered about
the core. Because the In is a slowly varying function one does not expect any
dependence on sample geometry and the most convenient one is used.
Two vortices of opposite sign (say a pair +n, -n) at a distance R,} apart have
an energy which is independent of L. Their fields cancel at large distances as
shown in Fig. 5.13; and all that remains is the dipolar near-fields, which result in

(5.218)

Fig. 5.13. A vortex dipole:

effects on spins far from the
cores is much reduced: the
1-1.- - Rij ---t-I energy is finite

Such vortex-antivortex (dipole) pairs can have a strong effect on the finite
temperature thermodynamics [5.45, 46].
One might wonder whether the effect depends on dimensionality. In ID, for
a chain of N spins we have

NoO = n2'lt (5.219)

and a self-energy

f1E = IN(2'ltnjN)2 = J(2'ltn)2jN (5.220)

a negligible value. Moreover, the interaction energy of a vortex-antivortex pair

separated by m sites is

f1E,,} = 1m I(2'ltnjN)2J . (5.221)

What is interesting here is the similarity to the interaction energy of two sheets
of charge in ID, and of course, the energetics of the vortices is analogous to the
flux calculus of classical electrostatics. Thus, in any number of dimensions, the
study of magnetic vortices parallels that of charged fluids.
198 5. From Magnons to Solitons: Spin Dynamics

5.16 Solitons: Introductory Material

It has been observed throughout this chapter that various problems in magnet-
ism become simpler as s, the magnitude of the individual spin, becomes larger
and the quantum fluctuations become correspondingly smaller. A century of
progress has recently culminated in the exact solution of many problems in
classical nonlinear dynamics in one spatial dimension, including those of
greatest relevance to this chapter. A common thread is the identification of at
least two different types of propagating excitations: the ordinary wave (here,
magnon) and the solitary wave, or soliton. The soliton is a particle-like manifesta-
tion which preserves its shape and identity after a long course of travel or period
of time, and after collisions with stationary defects or other solitons. To quote
its discoverer [5.47]:
"I was observing the motion of a boat which was rapidly drawn along a narrow channel by
a pair of horses, when the boat suddenly stopped-not so the mass of water in the channel
which it had put in motion-it accumulated round the prow of the vessel in a state of violent
agitation, then suddenly leaving it behind, rolled forward with a great velocity, assuming
the form of a large solitary elevation, a rounded, smooth and well-defined heap of water,
which continued its course along the channel apparently without change of form or di-
minution of speed. I followed it on horseback, and overtook it still rolling on a rate of some
eight or nine miles an hour, preserving its original figure some thirty feet long and a foot to a
foot and a half in height. Its height gradually diminished and after a chase of one of two
miles I lost it in the windings of the channel. Such, in the month of August 1834, was my first
chance interview with that singular and beautiful phenomenon. . ."

To relate this remarkable phenomenon to magnetism, some concepts will

have to be developed. Chief among these are Heisenberg's equations of motion
for the spin dynamics, such as

(5.222)

with ~the Hamiltonian of (5.1). Abbreviating a/at by (.), we find

(5.223)

The Zeeman contribution is easy to compute by the familiar commutation

relations, Eq. (3.5,6). The bracket coefficient of J mn has been previously cal-
culated in Eq. (3.55). Combining the results, we find

(5.224)

Summing over all spins, we see that the cross terms cancel and the total spin
merely precesses about the external field
 5.16 Solitons: Introductory Material 199

As the quantum unit of action Ii does not appear explicitly in the equations
of motion (5.224), therefore they are valid without modification in the classical
limit s - 00. (In proceeding to this limit, it is necessary to scale J and J.1B as lis
to keep the results finite, while setting Ii = I for dimensional convenience). We
can now represent a spin by a classical unit vector: Sn = SUn' Assuming a simple
lattice [s.c., or sq., or n.n. l.c. with primitive unit vectors ~ in the notation of
(5.16), with J = a the lattice parameter], the equations of motion take the form
of difference equations

(5.226)

with h == gJ.1BH = (0,0, h). For all long wavelength phenomena, whether linear
or not, a lattice may be approximated by a continuum. In (5.226) we have sub-
tracted terms such as Un X Un which vanish in the classical limit, to obtain a form
in which the continuum limit is easily taken. One expands in a Taylor series

(5.227)

Terms in J a or in JaJp(a =1= fJ) vanish in (5.226) by symmetry, leaving, in leading

order, only the following:

;,(R) = Jsa2 u(R) X [V 2u(R)] + u(R) X h. (5.228)

This is better written in terms of u± = Ux ± iu and

y Uz :

(5.229a)

with u- just the complex conjugate of u+, and:

(5.229b)

From our earlier studies we already know the form of a spinwave: it is a solution
of these equations with constant amplitude and plane wave character. Thus,
u± = Aexp [± i(koR - wt)] and (5.229) yield

(5.230a)

and

liz = 0, hence Uz = PA, a constant. (5.230b)

200 5. From Magnons to Solitons: Spin Dynamics

As Ii is a unit vector, the amplitude A must be ../(1 - p~). These results agree
with (5.16) for A --- 0, PA --- 1. At finite amplitude there is a "softening" of the
frequency, in agreement with our expectations on the basis of nonlinear magnon
interactions studied earlier in this chapter, e.g., (5.120). But at finite amplitude,
the excitation energy of the spinwave is extensive, i.e., it is infinite. A finite
amplitude spinwave is the superposition of O(N) magnons. Thus the momentum
and angular momentum are also infinite. This contrasts with the soliton, a
localized excitation whose energy, momentum and angular momentum are all
finite (1 D) or proportional the cross-sectional area of the wave front (2 or 3D);
see Fig. 5.14. To study it further, we shall need explicit expressions for the
excitation energy, momentum and angular momentum. It will also be convenient
to specialize to one dimension.
The Hamiltonian (5.1) can be written

1 ~ N
7C'= 2J~ (Sn+l - Sn)2 - h ~ (S~ - s) + Eo
1 Na [dli(X)]2
--- 2Js2a ~ dx dx + (hsJa) [
Na
dx[1 - uz(x)] + Eo (5.231)

proceeding to both classical and continuum limits. As unit vectors need only
two parameters to specify their direction, we can introduce the two, canonically
conjugate, dynamical variables P, q through the Villain representation, see

x
(a)

( b)
Fig. 5.14.(a) Spin wave (Sz = constant) extends to x = ± 00 • (b) Soliton is a moving pulse
centered about x. = vI. S~ and S. -+ 0 for Ix - x.1 > r. The characteristic pulse width r
and shape are obtained in the text
 5.17 Solitary Wave Solution 201

Eq. (3.87 - 89), in the classical limit

and Uz = p. (S.232)

.
Thus, the excitation Hamiltonian, in the natural units (Js 2a = I) is

H = 3t"0 - Eo = ~ J dx [I ~ p2 (dpjdx)2 + (I - p2)(dqjdx)2]

+ h' J dx(l - p) (S.233)

writing hsja as h'. The equations of motion (S.229) now take on the aspect

-q(x, t) = --p
I I2 (d 2pjdx2)

+ (1 !.p2)2 (dpjdx)2 + p(dqjdx)2 + h' (S.234a)

and

(S.234b)

Along with the Hamiltonian, we identify two conserved quantities [S.48] as, the
total momentum,

p = !.... J dx(l - p) ddq (S.23S)

a x

and the z component of the angular momentum carried by the excitations,

a
J
M. = !.... dx(p - 1) (S.236)

respectively. The latter is naturally negative, as the maximum spin of a ferro-

magnet occurs in the ground state.

5.17 Solitary Wave Solution

Here we set out to construct the most general solutions permitted by the equa-
tions of motion, having a "permanent profile" character, p(x - vt) and q(x
- vt). To qualify as solitons they must moreover be stable against small pertur-
bations, against collisions with other solitons, etc., i.e., they must possess some
quality of permanence.
202 5. From Magnons to Solitons: Spin Dynamics

With time dependence arising only through x - vI and v an adjustable para-

meter, (5.234) simplifies

(5.237a)

and

v(dpjdx) = ~[(P2 - I) : ] (5.237b)

(b) is integrated at once, yielding

dqjdx = v Po - P (5.238)
1- p2

in which Po is a constant of integration. Insertion into (a) now yields a nonlinear,

second-order differential equation for P

This can nonetheless be solved, by the device of setting (dpjdx)2 == F(p), so that
d 2pjdx2 = idFjdp. The above then turns into a linear, first-order differential
equation in the new unknown, F. It is readily solved, yielding

(dpjdx)2 = F(p) = 2h'p(p2 - I) - v2(1 + P5 - 2pop) - PI(p2 - I). (5.240)

Here PI is the arbitrary constant of integration. It should be remarked that this
equation is that of a classical particle of mass i and energy E, in a potential well
V = E - F. There exists the additional constraint, that Ip I ::;; 1. Aside from
this, the motion is confined to the positive region of F, which is a cubic polyno-
mial with asymptotic behavior 2h'p3. The constants PO,I must thus be adjusted
so that the solution is physically allowed. It is possible to integrate (5.240) and
analyze the resulting elliptic functions. It is, however, much simpler to study F
for the behavior between the turning points, from which one concludes that
PO,I must be adjusted to allow one of three distinct patterns shown as (a), (b)
and (c) in Fig. 5.15.
The first curve, (a), satisfies the conditions appropriate to a spinwave of
amplitude PA., previously analyzed. Only P = PA. is allowed, hence (5.237) has
the solution q = k(x - vt) + qo.
The second curve describes the soliton superposed onto an otherwise perfect
ferromagnetic background. The asymptotic value of P is Pc = I, and the largest
deviation from the asymptotic value is at P A. which can lie anywhere in the range
- I ::;; PA < + I, by suitable adjustment of the parameters PO,I'
 5.17 Solitary Wave Solution 203

-I p

( b)

Fig. 5.15. (a) Double root at P., sin-

gle root outside allowed interval (spin-
wave) (b) Single root at -1 < P. < 1,
-I p double root at P = 1 (soliton) (c) P
oscillates betweenp. andpb; this des-
cribes a wavetrain, with characteristic
of (a) if P. and Ph are very close and of
(b) if Pb and P. are very close
(c)

The final curve (c) leads to a periodic repetition of the soliton pulse, i.e., to
a sort of wavetrain. With PB and Pc very close and straddling 1, each pulse in the
infinite train resembles the solution (b), although the period is finite rather than
infinite. With PA and PB very close, the solution reduces to the spinwave of (a).
We thus see the spinwave and the soliton as the two rather opposite limiting
cases of the general behavior exemplified by (c).
The soliton solution" of this problem was first obtained by Nakamura and
Sasada [5.49], later by Lakshmanan et al. [5.50], and had its stability tested by
Tjon and Wright [5.48]. The construction of an F having the shape (b) in Fig.
5.15 requires the right-hand side of (5.240) to have a double root at P = 1.
F(l) = 0 implies Po = 1; then (dFjdp)p_1 = 0 yields PI = v2 + 2h'. Writing
P = 1 - 2sin2 fO and defining a new variable y = (1 - v2j4h')-1/2 sin fO, and a
new independent variable t = x(h' - fV2) 112 , we obtain

(5.241)

a well known ("canonical") equation, known to have the solution y = secht. It

is then a simple matter to substitute the original variables

(5.242)
204 5. From Magnons to Solitons: Spin Dynamics

where r == (h' - iv 2)-1I2, is the width of the pulse. This pulse is centered at Xo

°
(an arbitrary constant, chosen to satisfy initial conditions) at I = 0, travels to the
left (v < 0) or right (v > 0) with the given speed in the range ~ Iv I < 2(h')1/2.
Integration of the equation for q proceeds by similar substitutions, to yield

q = ~o + "21 v(x - vI - Xo) vz -

+ tan- [(4h'
1 1)1/2 tanh (X - vIr - X)]
o.

(5.243)

The distortions associated with the soliton are maximal at v = 0, yielding

the narrowest possible pulse. Conversely, at v 2 - + 4h', the pulse is infinitely broad
and the local distortions infinitesimal.
Several quantities are of interest: the phase shift, the momentum, angular
momentum, and energy. From (5.243) we find, for the total phase shift,

Aq = 2tan- 1 ( vz - 1)
4h' 1/2
(5.244)

Using (5.233) we find the energy density to be

°
which varies from a maximum of 1t at v = to zero as v 2 - 4h' .

H(x, t) 4
= psech 2 (X - r-
vt X)o. (5.245)

The total energy is time independent, of course. Integrating the above, we find it
to be

AE= 8/r.

Restoring the original units

(5.246)

Similarly, the momentum is

(5.247)

and

MI, = -4s/hra. (5.248)

Combining these gives a more perspicuous relation

AE = ~~:; sin (Pa/4s) .

2 (5.249)
 2.5 Two colliding solitons at t=D t =8 t = 12 Fig. 5.16 a-b. Two colliding soli-
r tons: one has velocity v = i and am-
2.0. - cos 191
.._-- Energy density plitude b = sin i8 = 0.8, the other
1.5
1.0. v = -2, b = 0.3. [5.47)
r.
I , (a) t = 0 (before collision)
0..5 /
/
--_.' (b) t = 8 (impact)
-~~r~------- (c) t = 12 (after collision)
C
-10 I I I I I -L-L-J I I I I (d) p = cos () vs x and t.
DO. 10.0. 20.0. 30..0 1.00. 30..0 1.0.0. DO. 10.0. 20.0. 30..0. 1.0.0.
(e) Energy density vs x and t

Two coll id ing sol i tons

VI

--...I

Vl
Q.
~ .
.::!

~
"oVl
C
g.
::I

~
VI
206 5. From Magnons to Solitons: Spin Dynamics

For the spin one-half two-magnon bound state, IHz I = 2, s = t, this formula
agrees precisely with the energy previously derived in (5.85) by quantum-
mechanical analysis. The above also demonstrates the further lowering of the
energy upon binding 3, 4, or more magnons into a bound state "soliton" of a
given total momentum.
Tjon and Wright [5.48] studied the collision of two solitons numerically, some
of their results being displayed in Fig. 5.16. But the crucial progress has occurred
only recently. First Takhtajan [5.51] and then Fogedby [5.52] have come to the
realization that the nonlinear equations of motion (5.228) or (5.229) can be
replaced by a set of coupled linear eigenvalue equations. Difficult though the
linear problem may be to bring to a closed form solution, there is the fact that the
spectrum of eigenfunctions and eigenvalues is complete. This spectrum maps, in
a 'one-to-one correspondence, onto the magnons, solitons and wavetrains we
have encountered. It is therefore no accident that two solitons, having a collision
described in detail in Fig. 5.16, survive this collision with their identities unim-
paired. This is a necessary consequence of the large number of constraints-
constants of the motion-that the equations of motion imply and that a solution
must satisfy.
The interested reader will find the analysis in [5.52] (also, in various stages,
in [5.48-51, 53]), for the mathematics are beyond the scope of the present
volume. Recent applications of this theory include the XY model, modified
somewhat to allow slight oscillations out of the XY plane [5.53] as required for
the experimental applications; future applications promise to explain the out-
standing mysteries about one-dimensional magnets: their response to neutrons,
to time-dependent forces and fields, etc.
In three dimensions, domain walls (which separate oppositely magnetized
regions of a ferromagnet) satisfy equations basically similar to the solitons dis-
cussed in this section. In the simplest cases - when spatial variations are con-
fined to a single direction - the equations are in fact identical. The energy to
create these walls, proportional to their cross-sectional area, hence O(N2!3), is
compensated by the corresponding reduction in magnetostatic energy. Thus,
the existence of domains in ferromagnets at low temperatures in sufficiently
small external fields, is mandated by thermodynamic requirements. While their
study is beyond the scope of the present volume, these concerns are addressed
in a number of accessible texts, such as Eschenfelder's [1.56].
6. Magnetism in Metals

Not all the electrons in a metal participate in the electrical conduction, nor in
other physical, chemical, or magnetic processes; some are core electrons, be-
longing to filled shells tightly bound to the ion and unaware of the metallic
environment. Electrons in unfilled shells have a range of behavior intermediate
between such tightly bound localized electrons and that of quasifree particles
experiencing the smooth periodic atomic potential and participating fully in the
electrical conductivity.
It is easiest to discuss the theory of magnetism in such metals where the
electrons can be clearly divided into distinct sets of tightly bound and quasifree
particles. Hopefully, the results still have qualitative merit when the carriers have
properties intermediate between these two extremes.
The simplest model of magnetism in metals is the following: electrons in well-
localized magnetic d or f shells interact with one another via a Heisenberg
nearest neighbor exchange mechanism, whilst an entirely distinct set of (quasi-
free) electrons in Bloch states accounts for the metallic properties without
partaking of the magnetic ones. Unfortunately, this model is purely fictional;
x-ray data, optical experiments, measurements of specific heat all indicate that
the d electrons in the iron transition series metals are largely conduction elec-
trons, in bands an electron volt wide. Whereas in the rare earths, on the con-
trary, the magnetic f shells are so well localized (~O.3 A) that the overlap
between them (at a distance of approximately 3 A) must be negligibly small,
and there is therefore no Heisenberg nearest-neighbor exchange, to a good
approximation. The observed magnetism in such a case must involve the s-band
conduction electrons, which alone are capable of sustaining correlations over
several interatomic distances: "indirect exchange".
For these and other reasons, I it is not possible to ignore the electrons'
itineracy in any sensible theory of magnetism in metals. 2 In the present chapter
we start with an elementary review of the band theory in the one-electron ap-
proximation with emphasis on tight-binding, the simplest approximation of any
value in the investigation of magnetic properties. Proceeding from there, we
shall see what gives rise to strong magnetic properties, such as ferromagnetic or
antiferromagnetic (AF) behavior. We shall even show that the Heisenberg
Hamiltonian for insulators can be derived on the basis of a band picture entirely

I See the indictment of the Heisenberg and Heitler-London theories in [6.1].

2 For a comprehensive treatment see [6.2].
208 6. Magnetism in Metals

analogous to the band theory of metals. We shall also derive the theory of
magnons in metals for the various models considered.
The case of a half-filled band is easiest. To the intratomic mechanisms
(intra-atomic Coulomb repulsion, exchange corrections thereto, etc.) we can
add the inter-atomic "transfer" matrix elements responsible for the motion of
the electrons through the solid. One or the other of these mechanisms can be
mathematically turned off, to recover the well-known atomic or free electron
limits.
On the other hand, the case of partially-filled (low density) magnetic bands is
far more intriguing. The electrons repel, and are not constrained by density to
occupy the same atomic sites. Turning off the transfer matrix elements would
result in an inhomogeneous localized array. We shall treat this case by a mUltiple
scattering formalism that is exact in the dilute limit, and yields interesting criteria
for the occurrence of magnetic moments in metals. We study it for clues to the
eventual ferromagnetism or AF of the electron fluid.
We include in this chapter a thorough treatment of the magnetic impurity
problem: how does an impurity atom acquire a magnetic moment, and how do
these moments interact with one another.

6.1 Bloch and Wannier States

Consider the one-electron Hamiltonian

(6.1)

where V(r - R/) is the averaged potential due to the nucleus and all other elec-
trons except the one under consideration. In the simple cubic structure, lattice
spacing a, the translation r - r + a(n" nz, n 3 ) leaves :;e:' invariant, and so the
translation operator can be used to provide the eigenfunctions of :;e:'with an im-
portant quantum number, the crystal momentum k. In other lattices, the
translations Ra take a different form, but a crystal momentum can always be
defined and, together with the band index t, it provides the quantum numbers
for the Bloch functions,

(6.2)

which are the eigenfunctions of :;e:'. The function u"k(r) has periodicity of the
lattice and satisfies the eigenvalue equation,
 6.2 Tight-Binding 209

subject to the boundary condition ut,k(r + Ra) = u"k(r), with Ra == a trans-

lation vector of the lattice (see below), previously denoted I in Chap. 5.
The meaning of the band index t is best understood in connection with the
Fourier transform of the Bloch functions, viz., the Wannier functions

(6.4)

which like the Bloch functions, form a complete, orthonormal set offunctions in
the Hilbert space of the Hamiltonian, (6.1). The sum over k is restricted to the
first Brillouin zone, i.e., to N values of k in the region

where Ra = anyone of the smallest translation vectors of the lattice (primitive

translation vectors). In the limit of infinite interactomic separation, the Wannier
functions reduce to ordinary atomic orbitals. In that limit, i identifies the atom,
and t the set of atomic quantum numbers (principal, orbital, azimuthal, spin; the
use of a single index is for typographical simplicity). When atoms are brought
close together, the atomic levels identified by t broaden into a band, unless, as
in the f shell of the rare earths and the Is helium core common to all metallic
atoms, the electrons are still so tightly bound to the nucleus at the observed
interatomic separation that the very concept of one-electron bands remains
inapplicable. But this is not the case of the 3d states, and we note that the first
metal to have a filled 3d band (Cu), and the elements immediately following it
in the periodic table (Zn, Ga, etc.) are nonmagnetic; whereas the iron series just
preceding these, noted for the unfilled d shell in the atom and d band in the
metal, form materials with varied and interesting magnetic properties. One may
rightly suspect the d-band electrons of being particularly important in the study
of magnetism, and the unfilled d band of containing "magnetically active"
electrons.
Before making these notions more precise it is necessary to review some of
the properties shared by all electrons, including the nonmagnetic ones. Some of
these can be studied in the "plane wave approximation," in which we set
V(r - R,) = 0 and ut • k = 1. But we do not wish to sacrifice the band structure;
the qualitative features of which are retained in the tight-binding approximation
which we study next.

6.2 Tight-Binding

The basic premise of this theory is that it is easier to estimate matrix elements
involving Wannier functions (because of their supposed localization about
specified atoms) than to solve the differential equations for the Bloch functions.
210 6. Magnetism in Metals

We illustrate this, using the Hamiltonian ~ defined in (6.1), and form the
Wannier matrix elements within the nth band,

(6.5)

so that Schrodinger's equation reduces to the determinantal eigenvalue problem,

(6.6)

Evidently it does not matter which representation we solve Schrodinger's equa-

tion in, and the eigenvalues E will coincide exactly with the Bloch energies En(k).
Moreover, the eigenfunctions necessarily turn out to be precisely the proper
linear combination,

(6.7)

which make up the Bloch functions, (6.4).

If one uses approximate Wannier functions, however, the interband (n =1= m)
matrix elements H(R'J)n.m need not vanish. It is common practice to use atomic
orbitals instead of Wannier orbitals as a first approximation, and therefore
this method is often known as the LCAO method, for the initials of linear com-
bination of atomic orbitals. The mixing of different orbitals to form the bands in
the solid expresses the well-known fact that angular momentum is "quenched"
(not conserved) due to conflict with the translational symmetry.
It is standard practice to limit the matrix elements H(R'J)n.m to nearest-
neighboring (in extreme cases, perhaps as far as second- and third-nearest-
neighboring) atomic distances R,J• It is not sensible to consider more distant
interactions, for if they become important the tight-binding procedure itself
becomes unwieldy, and other methods such as the quasifree electron approxi-
mation are then simpler and more appropriate.
Consider the band structures derived in the following simple examples. The
determinantal equation yields the energy at all points in k-space, with only the
constant parameters (overlap integrals) required to be numerically calculated.
And if we do not know the atomic orbitals, nor trust them in the particular
crystal under consideration, these constants may be taken as adjustable para-
meters to be fitted either by experiment or by comparison with a few calculated
points given by more accurate band structure calculations. Assume a simple cubic
structure and consider:
s bands: By symmetry, the matrix elements to the six nearest neighbors are
all equal, so that only two parameters enter the problem:

A == H(O) = J 1fI,*(r)~IfI,(r) d r
3
 6.2 Tight-Binding 211

and
-B == H(O, 0, a) = ... = H(a, 0, 0) = f 'fI1*(lr + (0, 0, a)I)~'fItCr)d3r.
In terms of these, the energy eigenvalues are

E(k) = A - 2B(cos k"a + cos kJla + cos kza). (6.8)

Problem 6.1. (a) Assuming nearest-neighbor overlap, prove that in the body-
centered cubic structure the s bands have the form

E(k) = A - B cos k"a cos kJla cos kza

and that in the face-centered cubic structure the appropriate formula is

E(k) = A - B(cos k"a cos kJla + ... + cos kJla cos kza)
(b) Derive the s-band structure for the hexagonal close-packed lattice.

p bands: Again in the simple cubic structure, with nearest-neighbor interac-

tions only, the threefold degeneracy of the atomic orbitals is not lifted. We write
the orbitals as

'fIn(r) = x~(r), y~(r), z~(r)

using 1= 1 functions from Table 3.1. A first parameter,

is the same for all three bands. A second parameter,

-B = f x~*(lr + (0, 0, a)I)~x~(r)d3r = f Y~*(lr + (0, 0, a)l)~y~(r)d3r

and finally a third one

-c = f (z + a)~*(1 r + (0,0, a) I)~z~(r) d 3r

are required. All other integrals may be obtained from the above, except those
for m =t= n; these vanish by symmetry, subject to the restriction to nearest-
neighbor overlap. When the Hamiltonian eigenvalue equation (6.6) is finally
solved, we find three degenerate bands:

EI(k) = A - 2B(cos k"a + cos kJla) - 2C(cos kza) (6.9a)

Elk) = A - 2B(cos kJla + cos kza) - 2C(cos k"a) and (6.9b)

E3(k) = A - 2B(cos kza + cos k"a) - 2C(cos kJla) . (6.9c)
212 6. Magnetism in Metals
Table 6.1. Matrix elements of (6.10) and (6.11) in tight-binding (LCAO) approximation

(sIs) H(ooo)•.• + 2H(100)•.•(X + Y + Z) + 4H(1l0)...(XY + XZ + YZ)

+ 8H(I11)•.• XYZ
(sIx) 2iH(100)•. xX + 4iH(llO)•• xX(Y + Z) + 8iH(1ll)•• xXYZ
(slxy) -4H(1l0)•. xyXY - 8H(1l1)•• xyXYZ
(slx2 - r) .;3 H(00l) •• 3"-r'(X - Y) + 2.;3 H(1l0)•• 3"-r'(Y - X)Z
(s/3z 2 - r 2) H(001) •. 3z'_r.(2Z - X - Y) - 2H(110)•• 3z'_r'( - 2XY + XZ + YZ)
(xIx) H(OOO)x.x + 2H(100)x.xX + 2H(100)y.y(Y + Z)
+ 4H(1l0)x.xX(Y + Z) + 4H(01l)x.xYZ + 8H(lll)x.xXYZ
(x/y) -4H(lIO)x.yXY - 8H(l I l)x.yXYZ
(xlxy) 2iH(01O)x.xy Y + 4iH(1l0)x.xyX Y + 4iH(01l) •.xy YZ
+ 8iH(11l)x.xy YXZ
(xlyz) -8iH(lll)x.y.XYZ
.J3iH(001)z.3z'-r'X + 2.J3iH(011)•• 3z'-r'(XY + XZ)
+2iH(01l)z.x'_y'X(Y - Z) + 8iH(1ll)x.x'_y'XYZ
-iH(001)z.3z'-r'X - 2iH(Oll)z.3z'_r'X(Y + Z)
+2.J3iH(Oll)z.x'_y'X(Y - Z) - (8/.J3)H(ll1)x.x'_Y'XYZ
2iH(00l)z.3z'-r'Z + 4iH(0ll)z.3z'-r'Z (X + Y)
+(161.J3)iH(IIl)x.x'_y'XYZ
(xy/xy) H(ooo)xy.XY + 2H(lOO)xy.XY(X + Y) + 2H(00l)xy.XY Z
+4H(110)xy.xyXY + 4H(Oll)xy.xy(X + Y)Z + 8H(1ll)xy.x..XYZ
(xy/xz) -4H(011)xy.xz YZ - 8H(11l)x...... XYZ
(xy/r _ y2) Zero
(xy/3z 2 - r 2) -4H(IIO}.y.3z'_r'XY - 8H(lll)xy.3z'-r'XYZ
(xz/x 2 - r) 2.J3 H(1 lO)xy.3z'-r'XZ + 4.J3 H(1 II)xy.3z'-r' XYZ
(xzl3z 2 - r 2) 2H(IIO)..... 3z'_r'XZ + 4H(1 I I)Xy.3z.-r'XZY
(x2 - y2/xl - r) H(ooo)" + lH(OOI)" (X + Y) + 2H(001)6 UX + ! Y + Z)
+3H(1IO)" (X + Y)Z + 4H(llW (XY + !XZ + ! YZ)
+8H(I11)" XYZ
(3z 2 - r 2/3z 2 - r 2 ) H(ooo)" + 2H(001)" UX + !Y + Z) + jH(001)6 (X + Y)
+4H(IIO)" (XY + !XZ + !YZ) + 3H(l1W (XZ + YZ)
+8H(lll)" XYZ
(xl - r/3z 2 - r 2) !.J3 H(OOI)" (-X + Y) -!.J3 H(001)6 (-X + Y)
+.J3 H(110)" (X - Y)Z -.J3 H(1l0)6 (X - Y)Z

• H(LMN)3"-r •• 3z2-r2.
b H(LMN)x2_y2.x2_y2.

Note: Assuming nearest-neighbor interactions only, in the simple cubic structure retain
only terms in (100), (010), and (001). For face-centered cubic retain only (110), (Oil), and (101).
For bodycentered cubic retain only (Ill).
Key: X = cos k ..a X = sin k ..a
Y = cos k ..a Y = sin k..a
Z == cos k.a Z = sin k.a
The band parameter constants are the integrals H(LMN)", •• = J'I':[r", + a(L, M, N)]2'
'I'.(r)d 3r
Source: Based on Table II in [6.3].
 6.2 Tight-Binding 213

The three p bands are related to one another by those rotations in k-space,
which permute the Cartesian components k",y,z' In each band, the contours of
constant energy have less than cubic symmetry, and even at small k, these con-
tours are not spherical but are ellipsoids of revolution with principal non-
equivalent axes along the k",y,z directions.
Had we chosen the axes of quantization of the three p functions along other
than a crystal axis, the interband matrix elements H(O, 0, a)n,m for n =F m would
not have vanished so conveniently. The eigenvalue equation, (6.6), which held
for the exact Wannier functions, must be replaced in the general case by the r-
dimensional equation

det IIH(k)n,m - Ec>n,mii = 0, (6.10)

where r = number of interacting bands; n, m = 1,2, ... , r. Here we have taken

advantage of translational invariance to form the Fourier transforms, and to
define

(n/m) == H(k) 1J,m = lN " ~

1,1
eik •R" f 1fI* .:YC'1fI
n,' m,} d 3r. (6.11)

In Table 6.1 are reproduced results by Slater and Koster, who have calculated
these matrix elements out to third nearest neighbor in the simple cubic structure.
This is sufficient also for obtaining nearest-neighbor interactions in face- and
body-centered cubic structures (and with some manipulations, next nearest
neighbors also, but we shall not be interested in these). Note that H(k)n,m is
abbreviated (n/m) in Table 6.1 for typographical simplicity. The matrix elements
H(k)n,m are given, abbreviated as (n/m), among s-like functions (s); p-like func-
tions (x, y, and z); and d-like functions (xy, xz, yz, 3z2 - r2, and x 2 _ y2).
It is sensible to consider these matrix elements as empirical constants; they
can be estimated by performing the appropriate two-center integrals.
Let us consider five d bands in the simple cubic structure, ignoring sand p
bands as well as non-nearest-neighbor interactions. Looking up Table 6.1, we
extract the following special case of the eigenvalue quation, (6.10):

(xy) (xz) (yz) x2 _ y2 3z 2 - r2

(xy) Fl(k) - E 0 0 0 0
(xz) 0 F2(k) - E 0 0 0
det (yz) 0 0 F3(k) - E i 0 0
.................................................................................................................
=0
x2 _ y2 0 0 0 Fik) - E V(k)
3z2 - r2 0 0 0 V(k) Fs(k) - E (6.12)
 214 6. Magnetism in Metals

The definitions of the F"s and V(k), and some features of the solutions, are dis-
cussed in Problem 6.2.

Problem 6.2. (a) In the example of (6.12) in the text, find F (k) and V(k) by re-
ferring to Table 6.1. Show that the solutions of the eigenvalue equation describe
three degenerate d bands with ellipsoidal contours of constant energy much like
the p bands of (6.9) and two non degenerate s-like bands. Obtain the contours of
energy of the latter near k = O. Plot the energy as a function of k along the three
principal directions, (100), (110), and (111).
(b) Calculate the first-order effect of an infinitesimal next-nearest-neighbor
interaction on the band structures calculated in part (a).

Making some assumptions about the relative and absolute magnitudes of the
band-structure parameters, Slater and Koster calculated the histogram density
of states curve (number of eigenvalues per unit energy) which includes all five d
bands in the body-centered cubic structure, such as in Fe. This is reproduced in
Fig. 6.1. The lower peak belongs to the bonding orbitals in the chemical termi-
nology, and, according to Slater and Koster, explains the anomalously great
binding energies of some metals in the first half of the iron transition series.

-
GJ
Z
Fig. 6.1. Original tight-binding density-of-states
histogram for d bands in bee structure. N(E) is
plotted E (Rydbergs), and is equal to the number
of eigenvalues E.(k) within ± 1.025 Ry of the
0_.08 o .08 E energy E, normalized such that the total area under
curve equals 5. [6.3]

What the histogram could not show are the so-called Van Hove singularities.
Whenever an E(k) curve has a minimum, maximum, or simply a saddle point,
its contribution to the density of states becomes excessive. [For example, a
totally constant E(k) = Eo contributes a delta-function singularity to the density
of states function.] Some years after the Slater-Koster work, Wohlfarth and
Conwell published the curve reproduced in Fig. 6.2, giving the density of states

N(E) ex: L:
_,k
J[E - En(k)] (6.13)

as the output of a computer calculation in which the Van Hove singularities

 6.3 Weak Magnetic Properties 215

were scrupulously preserved. The peaky nature of this new curve is evident, with
the principal maxima occurring whenever an energy band touches the Brillouin
zone. Note also the revision upwards of the estimated effective width of the
bands over the earlier work, by a factor of approximately 3.
As we shall see subsequently, it is of crucial importance in the theory of
magnetism in metals 3 whether the density of states is high or low, particularly in
the vicinity of the Fermi level/L

3
Fig. 6.2. More accurate
density-of-states curve N(E)
.... 2 ... vs E (Rydbergs) for bcc

-
l&.I
Z
iron. Dashed line is average
free-electron approximation
N(E) oc .JF;; and note how at

,,
several points (for example,
E = 0 or -0.4) the computed
I density of states can exceed the
average curve by a large fac-
0 E tor [6.4]
-0.7 -05 - 0.3 -0.1 00 0.1

6.3 Weak Magnetic Properties

All metals share some weak magnetic characteristics. The study of these has
enjoyed great vogue lately, because ofthe very detailed information on the band
structure and Fermi surface parameters which it divulges. As some examples of
important dynamic properties, we list the Hall effect, magnetoresistance, and
cyclotron resonance. Some important static properties of the electrons include
Pauli spin paramagnetism, Landau diamagnetism, and the De Haas-Van Alphen
effect. All these are well known and abundantly discussed in standard texts on
solid-state or metal physics 4 • Therefore, here we shall be content with a qualita-

3 The band structure of some magnetic metals is now known fairly well, such as nickel. See
[6.5].
• Less known, but also interesting, is the subject of high fields in metals, treated by Fawcett
[6.6].
216 6. Magnetism in Metals

tive discussion of the physical basis of the static phenomena listed, without em-
phasizing the mathematics which can become rather complicated. The principal
purpose is to display the important role of the density of states function N(E) in
the magnetic properties of electrons, and to introduce the concepts of Fermi
energy and Fermi distribution.
In the ground state of a normal metal (T = 0 K) all the one-electron states of
energy less than J1. are occupied, all those above are empty. J1. is the chemical
potential or Fermi energy. Moreover, in the absence of magnetic fields or spin-
orbit coupling to lift the Kramers' degeneracy, every state of given (n, k) below
the Fermi level is doubly occupied, by an electron with spin up and one with spin
down.
At finite temperature, states within ±kT of the Fermi level are partly oc-
cupied, as may be seen from the Fermi distribution function

1
feE,,) = e(Ek-f.l)lkT + 1 (6.14)

which gives the thermal-average probability that the state of energy Ek is oc-
cupied (absorbing band and spin indices into k). In a weak magnetic field, the
spins of the electrons within the ±kT neighborhood of the Fermi energy will
be free to orient themselves parallel to the field; and according to the laws of
Langevin and Curie, each will contribute a magnetization proportional to the
applied field, to Curie's constant C, and to the inverse temperature, viz.,

C
t5Jf ~ H T.

The number of participating electrons is ~2kT N(J1.), and therefore the total
paramagnetic spin susceptibility is

Jf/H == XP = 2CkN(J1.). (6.15)

This agrees with more rigorous derivations of Pauli's spin paramagnetism of

free electrons and is correct to O[(kT/J1.)2] at finite temperatures. (kT/J1. ~ 10-2
at room temperature for most nonmagnetic metals). Note the dependence on
N(J1.), the density of states at the Fermi energy. This susceptibility is smaller by a
factor 2kTN(J1.)/Jf/than that of Jf/free spins.
The De Haas-Van Alphen effect is not so easy to explain nor to understand;
nevertheless it also reflects the dependence of the thermodynamic properties of
the metal on the density of states. In this case, the density of states is affected by
a magnetic field, and therefore the thermodynamic functions will depend on the
field. For illustrative purposes, consider the wavefunctions in the free-electron
approximation,
p2 (hV)2
7C'=2~=
m - -m2 *.· (6.16)
 6.3 Weak Magnetic Properties 217

The effective mass m* may differ from the free-electron mass mo = 9.1 X 10-28
gram by one or more orders of magnitude (greater or smaller). The effective
mass approximation used here may be quite successful for describing s bands,
but it does not lead to a realistic density of states for the dbands, as seen in Fig.
6.2; so the following derivation is merely illustrative.
In a weak electromagnetic field, described by the vector potential A(r, t),
the electron momenta p become p - eA/c, and the time-dependent Schrodinger
equation becomes

[p - (e/c)A]2 . a
2m* 'Y' 'Yat
,,",,---'-...!.-.1.~IJI(r t) = lil- "'(r t)
, •
(6.17)

For a static magnetic field, A(r) = (0, Hx, 0) does not depend on the time and
satisfies the two equations

1 aA
vx A = (0, 0, H) and --~=
c ut
E(r,t) = O. (6.18)

Therefore, writing ",(r, t) = exp[i(kyY + kzz)] ~(x) exp( -iEt/li), we find that
~(x)obeys a harmonic-oscillator equation, so that the total energy is given by

n = 0,1,2, ... (6.19)

with the "cyclotron frequency" We defined by

eH
W" = m*c' (6.20)

Problem 6.3. Derive (6.19, 20) of the text, by solving Schrodinger's equation in
the manner described.

The expression for the energy may be interpreted as the result of quantization on
the classical circular motion of a charge in a magnetic field.
The density of states is obtained by differentiating the function which gives
the total number of states lying below energy E. Thus, with a factor 2 for the two
values of spin
d M(E)
f
(m+ 112M"",
L
"E
N(E) oc 2 - Iitikz (2m*)-1/2
dE m-I 0

M(E) 1
(6.21)
oc ~I -JE - (m + 1/2)liw e

where M(E) = largest positive integer for which the radicand is positive. A plot
of this function is given in Fig. 6.3a; it is similar to the free-electron function
218 6. Magnetism in Metals

N(E)......, EI/2 except for narrow, integrable, square-root singularities at half-odd-

integer multiples of the cyclotron energy IIwe.
The most interesting behavior occurs in the neighborhood of the Fermi
energy p. as the magnitude of H is increased. Whenever (p. - illwe) becomes an
integer multiple of IIwe, the above sums acquire a new integer M(p.), and the
density of states N(p.) itself acquires a square-root singularity. This is shown
in Fig. 6.3b. Clearly, this must lead to fluctuating, quasiperiodic behavior of all
the thermodynamic properties of the metal: the specific heat, magnetic suscep-
tibility, electrical resistance, etc., must all be oscillatory functions of the maxi-
mum integer, M(p.) oc IIH.

z
z
~~~=-------------E L--_ _--:-:_ _ _ _ H
(a)
Fig. 6.3. (8) Density-of-states N(E) vs E in constant magnetic field. Averaging the singu-
larities would lower curve below its zero-field value ,.fE. Therefore the energy of a constant
number of electrons in weak. magnetic fields is higher than in zero field, and the Landau motional
susceptibility is diamagnetic. (b) Density of states at the Fermi energy, N(P), as function of
strong applied magnetic field. The oscillatory behavior results in De Haas-Van Alphen effect.
Note that Jl is itself a function of H, determined by the requirement .Y = fP dEN(E) =
constant

The weak-field limit occurs either when IIwe «

kT or when scattering results
in a mean free path smaller than the radius of the cyclotron orbit. In either case
it is permissible to expand the free energy and other thermodynamic quantities
in powers of H.
A useful method due to Peierls [6.7] is based on the Poisson summation formula,

and some partial integrations to evaluate the leading terms in the free energy.'
As an example, the internal energy to leading order is

, See also [6. 8,9].

 6.4 Exchange in Solids: Construction of a Model Hamiltonian 219

with Xd a positive quantity. Because the increase in energy results in a force tend-
ing to repel the material from an applied field, this is a diamagnetic susceptibility.
Finally, the total susceptibility, combining the Pauli spin paramagnetism with
the Landau orbital diamagnetism can be shown to have the value

x = 2CkN(p) [ 1 - ~ (:~) 2J (6.23)

which explains the weak net paramagnetism of most metals (where m* '" mo)
and, on the other hand, the strong diamagnetism of bismuth, in which m* =
O(O.Olmo), (mo/m*)2 = 0(104 ).
In common with other magnetic phenomena studied in this book, the Landau
diamagnetism is a purely quantum mechanical effect, which disappears in the
correspondence limit by virtue of the oft-invoked Bohr-Van Leeuwen theorem.
Also the diamagnetic increase in energy, (6.22), is extensive, i.e., every unit
volume of the material contributes equally to the diamagnetic current density.
Note: It is possible to view the nonvanishing diamagnetism as a direct con-
sequence of the uncertainty principle; for if the electrons have perfectly sharp
momenta p, the vector potential A = (0, Hx, 0) cannot be simultaneously speci-
fied nor removed by a gauge transformation and vice versa. Therefore for small
H, the energy is raised above the ground state value it had in the absence of the
field. This point of view has been carried through by an expansion of the free
energy in powers of Ii in a review by Van Vleck [6.10] of the weak magnetic pro-
perties of metals and of the effects of exchange and correlation. Many-body effects
have been thoroughly explored by Isihara and Kojima [6.10].
An inaccurate but frequently heard explanation of the Landau diamagnetism
is that it is caused by the inability of surface currents to cancel volume currents,
due to quantum mechanical effects. But this is only a half-truth; for it obscures
the physically significant fact that Xd depends only on the bulk properties, and is
independent of the surface geometry, boundary conditions, scattering, etc.

6.4 Exchange in Solids: Construction of a Model Hamiltonian

In solids as in atoms, the really strong magnetic phenomena are electrostatic in

origin, the powerful Coulomb forces being "triggered" by the spins of the
electrons under the regulation of the Pauli principle. This is well demonstrated
in second quantization, introduced in Chap. 4. Specific applications to insula-
tors and metals will be the subject of the remainder of the chapter. In second
quantization, it is possible for a unique Hamiltonian to apply to all the various
sorts of solids, with only numerical parameters and the occupation of the various
220 6. Magnetism in Metals

bands remaining to be specified. Thus there is no need to deal differently with

insulators or metals at the present stage.
First, let us reformulate the theory of non interacting electrons, starting with
the operator c"n,m which which destroys an electron at the jth Wannier site, in
the nth band, with spin index m( = t or ,j,). The operator which creates an elec-
tron in precisely the same state is the Hermitean conjugate operator Cj~n.m' The
band Hamiltonian of(6.1) can be written in terms of these operators as

(6.24)

This represents quite graphically the "hopping" or transfer of an electron from

sitej to site i, with the matrix element previously calculated in (6.5), and displays
the conservation laws obeyed by true Wannier functions in the present (one-
electron) approximation. These are the conservation of the spin index, and of
the band index (which may also be considered as an "isotopic" spin), ensuring
that the one-electron bands are well defined.
The Fermion operators above obey the usual anticommutation relations

therefore
(6.25)

and the occupation number operator n, = c:c, has eigenvalues 0, only; here,
r or s stand for any set of quantum numbers, including the electron's spin, m.
The band Hamiltonian is diagonal in the Bloch representation. We show this
by means of a canonical transformation, which in turn is equivalent to choosing
the plane wave combination of Wannier operators, as follows:

1 N .
C - - - ' " e-1k'R,c
k,n,m - / N ...::.... I,n,m
'" 1=1

and (6.26)
1 N
c* - -/ -
k.n,m -
'" e+1k'R,c*
N ~ i,n,m .
,y l=l

The reader can easily verify that the ck's and ct's also satisfy anticommutation
relations, (6.25). The inverse linear combinations are simply

cI,n,m -- _1_
/ N 4'" elk'R,ck,n,m
AI km
first BZ

and (6.26a)
c* - _1_
I,n,m - ~ N
'" e-Ik'R,c*
fr.: k,n,m'
first BZ
 6.4 Exchange in Solids: Construction of a Model Hamiltonian 221

Therefore let us substitute these expressions into ~o, and obtain

~o = lN ~I eiCk-k').R, ~
i
H(R)
Ii 11,11
eik-RI/ ~ c"i c
lI,m k ,n,m i,n,m

(6.27)

using the definition of the energy of a Bloch electron En(k) given previously.
Because ~o is diagonal in the Bloch operator representation (nk,n,m = 0 or 1),
as eigenstate of this Hamiltonian can be specified merely by stating which
k, n, m are occupied, and which are not. For example, the no-particle vacuum
state 10) is annihilated by every Ck: Ck I0) = 0 therefore nk I0) = 0; and ~o
must also have zero eigenvalue in this state. A more important eigenfunction is
the Fermi sea, defined to be the state of lowest energy among all the eigenstates
containing precisely ,AI" electrons. In terms of the Fermi energy JI. (below which
there are precisely,Al" one-electron states k,n,m) the Fermi sea can be written as

IF) == n c:'n,mI O) (6.28)

where the product extends over all k, n, m for which En(k) < JI..
The eigenvalue of ~o in this state will be the "unperturbed" ground state
energy W o,
JJ JJ
Wo = ~ Eik) = J dEN(E)E where ,AI" == J dEN(E) (6.29)

and where the sum over k, n, m again extends only over the states in the Fermi
sea. It is the principal object of the theory of magnetism in metals to explain
precisely how the electronic interactions modify the Fermi sea, and perturb the
ground state energy.
One possible result of the interactions, and of thermal excitations as well, is
to create any number of elementary excitations. These are constructed by remov-
ing a single electron from the Fermi sea and placing it above in one of the
unoccupied states. For example, letting b stand for a set of labels k, n,m within
IF), and a for a set of such labels outside IF), the eigenfunction and eigenvalue
of a single elementary excitation are

and (6.30)

As an alternative to the above description, we may conceive the elementary ex-

citation of (6.30) as the creation of two quasiparticles: both a quasi electron of
energy E" - JI., and a quasihole of energy JI. - Eb are added to the ground state.
The energy of each quasiparticle, and of the elementary excitation as well, must
be positive (by definition of the ground state).
The elementary excitations occupy a continuum (in the limit L - 00, natural-
ly) of energy levels even when restricted to a specific total momentum. In the free
222 6. Magnetism in Metals

<l Fig. 6.4. Fermi sphere of radius k p with an elementary

excitation indicated: electron taken to (k + q), leaving hole
at k
k

X I

Fig. 6.5. Continuum

of elementary excita-
tions in nonmagn.etic
Fermi sea of noninter-
q acting electrons

electron approximation E(k) = k 2 in some appropriate units, with the Fermi

level at p = k:, the Fermi sea is represented in k space as a sphere of radius
k p (Fig. 6.4), with total momentum, total current, total spin all zero. The ele-
mentary excitations are c:+h IF), with k < k p and Ik +
q I > k p , omitting spin
indices. Even if q is fixed, there is a continuum of elementary excitations cor-
responding to the possible angles between k and q, and the energy of these is
bounded by two parabolas and the horizontal axis, as shown in Fig. 6.5. Bril-
louin zone and magnetic field effects on the spectrum of elementary excitations
are discussed in Problem 6.4.

Problem 6.4. (a) Discuss the double spectrum of elementary excitations + spin
flip,

for free electrons whose spins, only, are interacting with a magnetic field (e.g.,
an exchange field); that is, for which Em(k) = k 2 +
mpBH, where m = ± 1, and
IFIT) is the Fermi sea appropriate to this situation. Plot the continua in the
 6.4 Exchange in Solids: Construction of a Model Hamiltonian 223

manner of Fig. 6.5, and notice that since 1FK) is the ground state, all excitation
energies are required to be positive. Consider H both small, and large.
(b) Neglecting spin (as in the text above), show the effects of the Brillouin
zone by plotting qualitatively the elementary excitation spectrum of a half-filled
s band in the simple cubic structure. Pay special attention to the effects of
umklapp (k is necessarily in first B.z. but k + q is not), and to the maximum
energy cutoff in the spectrum.

Besides all one-body potentials, the band-theoretic Hamiltonian %0 can

include the averaged effects of two-body forces. Let us see why this is the case;
the most general two-body matrix element is, in the Wannier representation,

(6.31)

where

' . n,
V( I,J, , n,..I ,., , t ) -"'2
,J, t, f f
* (r ') "",J'(r ') Ir _e2
- 1 d 3r d 3r , ""'I' * (r )!JIn,j (r.)
r'l!JIn"l

(6.32)

These matrix elements will connect states which differ only by a change in the
quantum numbers of two electrons, fromjnm andj'tm' to in'm and i't'm', and so
long as we use orthogonal Wannier functions and two-body potentials (such as
the physically important Coulomb repulsion), there are no further matrix ele-
ments, and the total Hamiltonian consists of ~ot = %0 + %'. As stated above,
some of the terms in %' can be incorporated in %0; consider as one possible
example the terms with i' = j' and t' = t,

'" Vi('"
[ ..:::.... I,J, n , n; I.,., *
,J , t, t ) nj'tm, ] CjnlmCjnm • (6.33)

Although the factor which multiplies C~'mCjnm is an operator, its average value
in the Fermi sea serves as a useful estimate. Thus, we should incorporate into
%0 the terms,

with
(6.34)

and subtract them from %'. This is the sort of procedure which was already
anticipated in (6.1) and those following, when it was stated that VCr - R I ) is the
averaged potential due to the nucleus, and all other electrons except the one
under consideration. Thus the transfer of all the averaged effects of %' into %0
has the result that what remains of the former has vanishing expectation value
in the Fermi sea
224 6. Magnetism in Metals

(FI7C" - (FI7C" IF)F) == O.

This procedure "renormalizes" the band structure En{k) in a self-consistent

way. Note that the self-consistent one-electron band picture must be altered
somewhat if we consider a different state, such as the ferromagnetic state instead
of IF); but the resultant changes should be small if charge neutrality, the prime
consideration in the energy balance, is maintained. We may therefore imagine
7C'o to have constant parameters, to include a priori all the important electron-
electron interactions (on the average), as well as all the interactions of the
electrons with nuclei and their kinetic energy. From 7C" we shall extract for
present consideration only a significant subset of two-body terms. Using as a
guide, that we should consider only those interactions which matter most when
the atoms are very far apart (the intraatomic terms) or when they are very
close (the ubiquitous Coulomb interaction), we can extract from ~ot = 7C'o
+ 7C" the features of greatest physical significance and provide a starting point
for later more ambitious investigations.
The first of the interactions to be retained is the direct two-particle Cou-
lomb repulsion obtained from 7C" by setting in' = jn and i't' = j't. After a change
of dummy indices, it is

(6.35)

where the simplified notation is used,

VCR,,) == V(i, i, n, n;j,j, t, t)

= ~ f d 3r f d3r'IfI~,(r')IfI,.ir') Ir ~ r'llfI:'i(r)lfIn./{r) (6.36)

and the factor 1/2 prevents double-counting. At large distances the leading term
in a multi pole expansion of this integral is

(6.37)

modified to be accurate even as close as nearest-neighbor separation. The terms

subtracted in (6.35) ensure that we have electrical neutrality,

(FI~'IF) = 0
as required.
The next important class of terms are the exchange interactions, which are the
subset of terms in (6.31) for which i = j', n' = t and i' = j, t' = n. Because ct,'m
and c,nm anticommute, the interchange of these two operators to properly as-
sociate the pairs in the manner indicated, results in a minus sign. Except for the
 6.4 Exchange in Solids: Construction of a Model Hamiltonian 225

operator formalism, this then is the ordinary exchange interaction discussed

repeatedly in this volume. It acts as a correction to the COlJ.lomb repulsion of
two electrons in a relative triplet state, and lowers the energy of this state as
compared to the singlet configuration. By grouping pairs of Fermions into spin
operators, we can include in our Hamiltonian the natural and obvious generaliza-
tion to the exchange operator i(1 + 4S"Sj), including the variable occupation-
number feature of the second quantization. Associating the indices as prescribed,
one extracts from (6.31) the exchange Hamiltonian,

~x = - ,,~
1,1
J(R'j)",,,, {s",j ,S"',I + ! n",j[n"'1 - (FI n"',11 F)]} (6.38)
("'()¢("',Jl

where

n",j == n",I,t + n...j,1 (6.39)

and the spin operators,

and s; = S; - iS~ = c~Cpt (6.40)

are recognized as one of the Fermion representations of the spin one-half

operators, with Ii = I, andp == (n,j).
The exchange constant, in the simplified notation, is the positive-definite
integral

J(Rlj)".", = 2V(i,j, n', n;j, i, n, n')

2
= f d 3r f d 3 r'",!lr')",,,,ir ') Ir ~ r' I ",:'"(r)",,,,j(r) . (6.41)

The largest such exchange integral is the familiar intra-atomic Hund's rule inte-
gral to which we assign a special symbol,

J!"", == J(O) ... ",

= f d 3r f d 3r''''!i(r')'''",.i(r') Ir ~ r'I",:,,lr),,,,,,/(r) (6.42)

and in most cases it will not be necessary to consider any exchange contribution
other than J(O). The reason for this is that J(R'j) decreases much faster than the
Coulomb integral V(Rij), so that unless (6.42) vanishes by symmetry or for
any other reason, even the nearest-neighbor exchange will be small in com-
parison, and the non-nearest-neighbor contributions completely negligible. (An
exception: for only one band, two electrons can be in a triplet state only if they
226 6. Magnetism in Metals

are on different Wannier sites; and the leading exchange integral is (6.41) at
nearest-neighbor distances, with n = n').
The designation of the exchange forces has a certain degree of arbitrariness,
so it is useful to conclude this section with a review of the choice which was
exercised. One traditional approach which was implicitly discarded, was the one-
band, free electron approximation, with Bloch electrons interacting solely via the
long-range Coulomb repulsion, (6.35). While this idea might have appeared
plausible at the time Bloch [6.11] first proposed it, it had to be discarded in the
face of strong theoretical and experimental evidence. " ... those causes of the
magnetical motions ... we relinquish to the moths and the worms" (see [Ref.
1.2, p. 64]) Obviously, experiment cannot disprove a correct theory; the fact that
metals with simple nondegenerate bands are never found to possess strong mag-
netic properties should merely have been a spur to the experimentalists to look
harder, had it not been for wiser counsel, notably Wigner's calculation [6.12]
showing that correlations among the charged particles keep electrons both of
parallel spin and of anti parallel spin apart (see also Chap. 1). In fact when the
Coulomb interaction is strong enough, the ground state has to be a singlet as in
the hydrogen molecule. The key to magnetic behavior must therefore be the ex-
istence of degenerate bands permitting a large number of electrons to be on the
same or neighboring atoms. In such cases they cannot escape the Coulomb
repulsion even by elaborate correlations. The fact is that the same forces which
are present in the individual atoms to produce Hund's rule "atomic magnetism"
cannot suddenly be nullified in the solid state. 6 Compelling as these arguments
may be, they are not rigorous. A mathematical proof is available only in one
dimension, where we have shown ([6.14], see also Chap. 4) that the ground state
of %0 + any interaction Hamiltonian %e, is always a nonmagnetic singlet state,
regardless of the number of electrons or the nature of the potential energy. Our
proof breaks down, of course, in face of the degeneracy of three-dimensional
atoms, as symbolized by the exchange Hamiltonian of (6.38). This adds plausi-
bility to the choice of a single effective Hamiltonian for use in all solids which
includes (6.38), viz.,

~ff = %0 + ~ + ~x (6.43)

defined in (6.27, 35,38). The retention of that part of:Jt" which is explicitly spin-
dependent, namely ~x, enhances the possibility of magnetic behavior.
Unfortunately, the eigenfunctions and eigenvalues of :Jt'eff cannot be found
exactly because the three constituent terms do not commute with one another
and therefore cannot be simultaneously diagonalized. ~ff itself is only part of
the total Hamiltonian,

(6.44)

6 The credit for this idea belongs to Slater [6.13].

 6.5 Perturbation-Theoretic Derivation of Heisenberg Hamiltonian 227

defined in (6. 27,31), which contains all manners of complicated interactions,

some of which have spin-dependent consequences and some not. Although we
cannot assess their importance in any specific metal, we discard them because
they are not required in the high-density limit for which ~o suffices, nor in
the low-density limit where ~ + ~x will do. Thus, (6.43) which includes all
these essential terms, will suffice.

6.5 Perturbation-Theoretic Derivation of Heisenberg Hamiltonian

To illustrate the uses of ~rr, we shall obtain the Heisenberg Hamiltonian valid
in an insulating magnetic material. 7 This derivation is an extension to the many-
body problem of the method used in Problem I, Chap. 2; it suggests what is re-
quired in a general theory of magnetism in insulators, a complicated and rather
specialized subject when investigated in appropriate detail. According to the
theory, the larger contributions will mostly be the antiferromagnetic ones, in
accordance with experiment; although there exist exceptions to this prevalent
antiferromagnetism in insulators:
... a europium oxide of the formula EuO becomes truly ferromagnetic at 77 K with a
saturation moment of close to 7 Bohr magnetons. This is thus the first rare earth oxide to be
found to become ferromagnetic, and with the exception of Cr0 2 the only oxide to our know-
ledge that has true ferromagnetic coupling [6.15].

In the magnetic insulator, ~ + ~x is the big part of ~rr, and the "hopp-
ing" part, ~o is the perturbation. However, ~x may be further decomposed;
the Hund's rule part is retained in lowest order, whereas the nearest-neighbor
exchange is treated by first-order perturbation theory. The starting Hamiltonian
is, therefore,

(6.45)

describing the internal dynamics of noninteracting Wannier "atoms," each of

net spin SI of magnitude SI = i times the number of electrons in unfilled shell.
(It is assumed that a crystal field quenches the angular momentum, or else the
appropriate quantities are J I andjl)' The ground-state eigenfunctions of ~are
highly degenerate, in fact they number No,

No ;: n (2s + 1)
I
1

and it is the perturbation ~o which will lift this degeneracy. The perturbation
~o can be seen from (6.24) to transfer an electron from one site to a neighboring
site, this virtual transition occurring in second-order perturbation theory with
matrix element H(R IJ )",,, and an increase of energy U > 0 in the intermediate

7 The nature of the "insulating state" is the subject of [6.16] by Kohn

228 6. Magnetism in Metals

state due to the creation oftwo ionized sites: a positively ionized ion at R, and a
negatively ionized one at R J • Let one of the neutral ground states be Ia) and
one ofthe excited ionized states be IP); then U == (PI~IP) - (al~la) and the
second-order perturbation theoretic change in the energy of Ia) equals

JEa = _~ I(PI~ola)12
p<I<a U
_ (al~ola)2 - (al~5Ia) (6.46)
- U

with the second line obtained by closure [~p IP)(P I == 1]. ~o has vanishing
ground state expectation value in the insulator (as opposed to a metal) and
therefore (a I~o Ia) vanishes. In writing out the nonvanishing second term, one
collects such terms as

(6.47)

and others corresponding to S~Sj, as well as terms which do not depend on the
relative orientation of the two spins. The latter have the same magnitude for all
states Ia) and contribute a constant shift in their energy JEa. The degeneracy of
the ground state and relative ordering of the energy levels is given entirely by
the subset of spin-dependent terms we lump together into an effective Hamil-
tonian

(6.48)

It is important to note that the vectors S, are the total atomic spin operators and
not individual electron components such as S/,n, which cannot be specified in
any of the No ground states. Therefore the quantities H2(R/J ) and J(R,J ) are ap-
propriate averages of the corresponding quantities in the various bands. Note
that the non-Hun d's rule nearest-neighbor exchange J(R,J) has been reintroduced
at this point by means of first-order perturbation theory. The Hund's rule para-
meter J(O) is absent from (6.48). Its role has been to create the atomic spins St.
The total Heisenberg Hamiltonian in nonconducting media must be considered
as the sum of two effects: an invariably antiferromagnetic interaction due to the
virtual "hopping" of an electron from R, to RJ (and back), which is a "one-
body" or "kinetic" exchange mechanism. It is antiferromagnetic because the
hopping is enhanced when the spins are anti parallel, the exclusion principle
prohibiting certain hops when the spins are parallel. The second, ferromagnetic,
contribution arises from the exchange of two electrons not on the same atom; its
matrix elements among the degenerate ground states may be described by first-
order perturbation theory; it is always ferromagnetic because J> O.
The same arguments hold for the interactions between two distinct partly
occupied shells on the same atom, e.g., the valence and magnetic shells on a
 6.6 Heisenberg Hamiltonian in Metals 229

transition or rare earth atom or ion. The net coupling, Jeff given by the square
bracket in (6.48), gives the tendency of the two shells to have their spins parallel
or antiparallel, with important consequences in the Kondo effect studied later in
this chapter.
As in the Heitler-London theory, the net interaction is the difference between
two positive contributions and can have either sign. (Estimates of the various
integrals indicate that the result is usually antiferromagnetic). However the
present derivation is free from some of the undesirable features of the H-L
derivation discussed in an earlier chapter. The natural expansion parameter here
is U-1, a physically measurable quantity; whereas in the older theory, the para-
meter I is not, in fact, an observable. The use of orthogonalized functions
removes an element of arbitrariness from the theory, and the usual mathematical
analysis of Hamiltonian mechanics can now be applied systematically.
In real materials (e.g., magnetite) the hopping occurs via an intermediary
nonmagnetic ion, such as 0- 2 • This is called superexchange, a mechanism first
proposed by Kramers [6.17] 50 years ago. The detailed theory of magnetism in
insulators, including the theory of superexchange, is discussed by Anderson
[6.18], who has contributed much toward s the development of this field.

6.6 Heisenberg Hamiltonian in Metals

The "indirect exchange theory" of magnetism in metals is another example

where an effective interatomic Heisenberg Hamiltonian can be derived by
second-order perturbation theory. The relative importance of the various
energies is here completely reversed over the previous section, for in a metal ~o
is the principal Hamiltonian, and ~c and ~ex the perturbations.
The theory developed below was first invented in connection with nuclear
magnetic resonance by Ruderman and Kittel [6.19] and independently, by
Bloembergen and Rowland [6.20]; these authors studied the effective long-
ranged interaction between nuclear spins due to the hyperfine coupling with the
common sea of conduction electrons. The extension of their analysis to the s-d
or s:finteraction [6.21] permitted the explanation of some significant experiments
by Zimmerman on the long-ranged interaction between Mn atoms 8 dissolved
in Cu. Briefly, the Mn impurity atom retains part of its Hund's rule magnetiza-
tion in the solute state and by Jeff> (6.48), polarizes the spins of the conduction
electrons in its neighborhood. The conduction electrons, constrained by the
Pauli exclusion principle, respond with a characteristic wavelength

(6.49)

8 See review of theory and experiments in [6.22-24]. and discussions on the "spin glass" else-
where in this volume.
230 6. Magnetism in Metals

and the resultant spin polarization is not well localized in the vicinity of the
impurity but is oscillatory and long-ranged. A second manganese atom at an
arbitrary distance from the first, suffers a ferromagnetic or an antiferromagnetic
interaction with it, depending upon whether it is in the trough or on the crest of
the polarization wave. The strength of the interaction gradually decreases with
distance, in a manner we shall calculate.
Assume a pair of solute magnetic atoms at RI and R 2 , in an otherwise ideal
nonmagnetic metal characterized by an s-band Hamiltonian ~. Internal
Hund's rule coupling maintains the magnitudes of the solutes' spins fixed at Sl
and S2, respectively, but the relative orientation of the two spins will be governed
by the interaction which is derived below. The exchange coupling of the localized
-electrons with the conduction electrons is the perturbation,

(6.50)

with the conduction-band spin operators se(R/) given by (6.40). The substitution
of Bloch operators for the Wannier operators, given in (6.26a), results in the
following:

(6.51)

With this definition in mind, let us calculate the eigenvalues and eigenfunc-
tions of

J'IC' = J'lC'o + J'lC'ex = ~ E(k)n k • m

m~t.1
k

(6.52)

by ordinary perturbation theory. In particular, we wish to see how the pertur-

bation lifts the degeneracy of the r == (2s 1 + 1) X (2s z + 1) states of orienta-
tions of the two solute spins. The conduction electrons will be assumed to be in
their ground state, except for the resultant polarization effects.
The first-order correction to the energy,

(6.53)

is seen to vanish. We use the notation IF; t) to indicate the product state of
Fermi sea with the two solute spins, with the index t spanning the range t = 1,
... , r of degenerate orientations of the two spins.
 6.6 Heisenberg Hamiltonian in Metals 231

The conduction-band spin operators c*c create elementary excitations of

energy E(k + f) - E(k); their matrix elements are unity if k < kp and 1k + f 1
> k p , and zero otherwise. Therefore, by second-order perturbation theory

(1Is\(s\ + 1) + S2(S2 + 1) + 2S\oS2COSfoR12lt)

E(k + f) - E(k)
(6.54)

The second line is the result of using closure on the intermediate states 1t'). It is
convenient to separate this formula into two parts: a self-energy

aE(2) = -KLS,(S,
I
+ 1) (6.55)

with

J~rr L 1 (6.55a)
K = 2N2 Ik+,I<kP
k<k. E(k + f) - E(k)

and an interaction energy, which is the eigenvalue of the effective Hamiltonian

(6.56)

where the indirect exchange coupling constant is

(J orr) cos foR '1

2
+ f) - E(k) .
'"
J(R'/)IB = Ii k~. E(k (6. 56a)
Ik+,I>k.

Ifinstead of only two impurities there are N I , the sum in (6.55) runs over all NI
spins and the interaction, (6.56), over all iNI(NI - 1) distinct pairs.
For an estimate of the indirect exchange coupling, one uses the effective mass
approximation, E(k) = h2k2/2m* and f1. = h2kM2m*, and introduces an imagi-
nary part hifT: to the denominator to account for a finite electronic mean free
path.

J(R) - J2 o )6
'I IB -
(a
orr 21t [~Z)]
d 3k r rd 3k ' (h2/2m*)(k
cos(k -
'2 -
k')oR'J
+
k 2 i2m*/hT:)

= J2 (ao) m* (41t) rdkk kP

dk' k' sin kR'I sin k'R'I
+
6 2 OOr

orr 21t 2h2 R'I 0 _00 k'2 - k 2 i2m* /hT:

= -J~rr (k pao/2)6 [sin 2k pR'J - 2k pR'I cos 2k pRIJ]e-RII /J.
(6.56b)
f1. 1t3 (2k p R'/)4

where A. = mean free path = hkpT:/m* and a~ = volume of unit cell. This result
232 6. Magnetism in Metals

(without the mean free path factor) was first published by Ruderman and Kittel
[6.19] and is universally referred to as the Ruderman-Kittel interaction. The mean
free path factor was introduced in the first edition of the present book, and has
since been measured [6.25].

6.7 Ordered Magnetic Metals: Deriving the Ground State

Some of the most interesting applications of the indirect exchange theory are to
metals containing elements in the gadolinium rare-earth series (lanthanides).
Generally the f-shell radii of the rare-earth atoms are so small that even nearest
neighboring atoms do not have significant direct overlap and the interaction is
assumed to be principally the Ruderman-Kittel indirect exchange mechanism
derived in the preceding section (with crystal field anisotropy and hybridization
i.e., band-mixing, the principal correction) [6.26]. Ordered alloys containing
transition elements are also described by this theory if the magnetic atoms are
sufficiently far apart for the d-shell overlaps to be unimportant.
Therefore we are led to consider the Hamiltonian

(6.57)

and to calculate its eigenstates and eigenvalues in cases where the spins S/ occupy
points on a regular lattice. For convenience, we shall assume it to be one of the
Bravais lattices, and particularly one of the three principal cubic lattices. There
is no exact method known to obtain the ground state, but the following pro-
cedure avoids unnecessary complications, and appears reliable. First, we con-
struct the product wavefunction

(6.58)

in which the V>/ are, as yet, unspecified but normalized states of the spins St. The
variational energy in this configuration is

where (6.59)

(6.60)

In the remainder, assume all N I magnetic atoms to belong to the same species, so
that s/ = s. The method of Luttinger and Tisza, similar to the "spherical model"
discussed later, can then be used to find the lowest energy attainable with a trial
function of the type in (6.58). It consists mainly of relaxing the inequality above,
requiring only
 6.7 Ordered Magnetic Metals: Deriving the Ground State 233

which is a weaker constraint. The lowest energy (6.59) subject to the weaker con-
straint (6.61) must be lower than the lowest energy (6.59) subject to the more
rigorous constraint. (6.60). E is calculated by Fourier transforms, letting Sk be
defined by

Sk == Nl I: e-ikoRJ(~li Sli ~I) and (6.62)

I I

substituting it into (6.59) and obtaining,

(6.63)

where

(6.64)

The weak inequality (6.61), by substitution of the Fourier transform, is found to

be
(6.65)

and if we define fo to be the wave vector for which J(fo) attains its largest value
(or one of the wave vectors which have this property ifthere are more than one)
then let Sf. = s and all other Sk = 0, and

(6.66)

is certainly the lowest energy subject to the weak constraint. Now choose the
wavefunctions ~I sueh that

(6.67)

which is called a spiral configuration of pitch fo. Inserting this variational ansatz
into (6.59) leads to precisely the energy E(fo) calculated above; and being an
upper bound as well as a lower bound to the ground-state energy (in the Hartree
product wavefunction approximation), E(fo) must itself be the Hartree ground-
state energy.
When fo = 0, all spins are parallel and the ground state is ferromagnetic.
When fo is a wavevector on one of the points of symmetry ofthe Brillouin zone
boundary, for example, It/a(1 ±, ± I, ± 1) in the simple cubic structure, the
ground state is an antiferromagnetic configuration of some sort (the Neel state
in the example given). If fo is none of these special wave vectors, the ground state
234 6. Magnetism in Metals

is a "spiral spin configuration," as these screw structures are known in general.

The various possible configurations can be found from group theory or by matrix
methods, even in non-Bravais lattices, particularly in insulators when J i } is
limited to the Heisenberg nearest-neighbor forces [6.27, 28]. Note that in present
derivation the range of the interaction does not matter, therefore the proof holds
equally well for the short-ranged Heisenberg interaction in insulators. In the
case of the long-ranged Ruderman-Kittel interaction, numerical calculation is
required to obtain J(k), and to determine qo.
The numerical calculation of J(O) is particularly interesting, because accord-
ing to the theory developed in the next volume, it is proportional to the paramag-
netic Curie temperature (). A negative () necessarily precludes ferromagnetism; a
positive () is a likely indication of ferromagnetism, but it is still possible for
qo "* °
to be the ground state solution, and for a spiral configuration to be the
stable ground state at low temperatures. Therefore it is also necessary to study
the spin-wave spectrum, or magnon energy

hW(k) = 2s[J(0) - J(k)] (6.68)

to determine that all hw(k) are positive, if the ferromagnetic state is to be stable.
This is a necessary but not suffiCient condition for ferromagnetism. 9 The minimiz-
ing wavevector qo can also be found at the minimum of the function hw(qo).
As an example, we calculate hw(k) for the Ruderman-Kittel interaction, in
the continuum limit kpa -+- 0, where the lattice sums can be replaced by integrals.
Let us define (XI == 2k p R, a = separation of n.n. magnetic atoms):

'" 3 sin XI - XI cos XI -RdA(l

e(k) =-..::..... a 5 R3 e _
cos k.R)I (6.69)
XI i

which is (6.68), the magnon energy, with unimportant constant factors elimi-
nated, for tabular convenience. Also,

(6.70)

is the paramagnetic Curie temperature with similar constant factors removed.

The limiting values are elementary integrals

1· () 81t (1
l;!!o = 10 + 32k1pA - 1 _I 1 )
161tk pAtan 2k pA (6.71)

• Cf. [6.29], in which it is shown that in some exceptional instances the ground state may be
nonferromagnetic, due to quantum fluctuations even though the ferromagnetic state is stable
a gainst the emission of (any finite number of) spin waves. This analysis has been extended,
see [6.30], and ongoing work. As for the practical task of fitting the formulas to experiment,
we may cite [6.31] for the Heusler alloys.
 6.7 Ordered Magnetic Metals: Deriving the Ground State 235

and

These are the magnetic parameters for a "continuum" or "jellium" lattice or

for a low density electron gas. Note that they are both positive, indicating ferro-
magnetism.
Closely associated with k p is the dimensionless parameter nela == number of
conduction electrons per magnetic atom. (If there is one conduction electron per
unit cell, but a magnetic atom is present only in every other cell, nela = 2, etc.
The "jellium" limit is equivalent to nela - . 0.) As nela is raised, the paramagnetic
Curie temperature goes through zero at approximately nela = 1/4 for the three
principal cubic lattices. There is then an antiferromagnetic region as nela is in-
creased further, until it exceeds the value 3/2 for the bcc and fcc lattices, and 5/2
for the sc lattice. At this point the paramagnetic Curie temperature becomes
positive once more until nela is approximately doubled, whereupon a new anti-
ferromagnetic region is encountered, etc.

SPIN WAVE
STABLE sws SWS SWS SWS
I (SWS) I ~ I----l H H
15

+ ARE >.. 2.00

10 fl ARE >..4.00

<%l

..sc 5
..,'"
VI
C
0
U
0
"0
(jj
'+-

~ -5
u'"
OJ
'0
~
-10

KFa

0 2 4 6 8 10 12 14 16 18 20

Fig. 6.6. 8 vs kpa in the sc lattice (bee, fcc are similar). Values of electron density n = (8n/3)
(k p a/2n)3 = number of conduction electrons per magnetic atom, are indicated by arrows. Main
curve is for 1 = 3 (some points at 1 = 2 or 4 are shown to indicate the insensitivity of the re-
sults). Spinwave-stable regions are indicated, and are seen to correlate well with 8 > 0
236 6. Magnetism in Metals

A plot of () as a function of k p and ne / a , which displays the features just dis-

cussed, is given in Fig. 6.6 for the simple cubic lattice. In Fig. 6.6 are also shown
the spin-wave stable regions where liw(k) > 0; these are seen to be somewhat
smaller than the regions of positive paramagnetic Curie temperature. To scale
the results so that they correspond to the Ruderman-Kittel interaction of (6.56)
with constant J, it is necessary to multiply the plotted and tabulated values by
appropriate functions of k p , a, and ao = nonmagnetic metal lattice parameter.
In the calculations, the value a = 1 is taken for convenience Therefore, com-
paring (6.56b, 68, 69), we obtain the physical parameters 8 and liw(k)

8() -_ liw(k)
e(k) ex:
J2(k
pao
)(k
pa
)3 (aao)
6
(6.73)

in terms of () and e(k), the computed functions.

In Fig. 6.7 we reproduce experimental results of Methfessel and collaborators
[6.32], on the paramagnetic Curie temperature of certain ordered Eu-Gd-Se

50
x-x-x
'x

'\
1(1)
1 \
0 I----'-_I-.----'-_L----..l.._+--..l..-~---L-
.2 .4 .6
X

\
-50
\x X
GdSe
kF 0::::: 2
(estimated)
Fig. 6.7. Experimental values ofe (paramagnetic Curie temperature) vs. composition (roughly,
k~) at constant a, obtained by Methfessel et al. [6.32] in a series of ordered rare-earth alloys. An
order-of-magnitude estimate is that kpa varies in the range 0-2 over the range of compositions,
and these results are in qualitative agreement with Ruderman-Kittel theory (cf. Fig. 6.6 over
same range of kp )
 6.7 Ordered Magnetic Metals: Deriving the Ground State 237

alloys, in which the electronic concentration could be varied from insulator to

metallic, corresponding to the range 0 < kpa < 2 (the upper value is an order
of magnitude estimate). These results are apparently in good qualitative agree-
ment with the applicable portions of the theoretical curves.

1-_-O:::-:;;,·...oC--::::
.....
........ ,
q=O
a~"(1Y,O,O) 100
(a) (b)

III
1----~__=,,4-~ 100

110
(e) (d)
Fig. 6.8. Plot of -J(k) for

,,.-._._._._. ._.-.
ak F =2.827 akF=TT 100
. the Ruderman-Kittel inter-
·)o. .......
'\ ~

100
-,
~.

,
action versus k in the three
...... principal directions, sc lat-

" , ,, 110
\ ,, 110 tice. In the range 0 < kpa ~
1t/2 the ground state is ferro-
,, \ magnetic. However, as k p is
, \
\
increased in frames (b) to (f)
various antiferromagnetic
\
aqo=(11, 11,11) III
\
\ , states become stable, indi-
cated by nonzero values of
aqo=(1T,1T,TT) III
spiral pitch parameter fo.
(e) (f) J(k) is defined in (6.56, 64)

A plot of J(k) in the simple cubic lattice is given in Fig. 6.8. One observes the
ferromagnetic state characterized by flo = 0 being succeeded by antiferromag-
netic configurations as k p is increased from frame (a) to (b). The resulting config-
uration consists of alternating planes of parallel spins, the alternation being in
the (lOO) direction. As one proceeds to frames (d), (e), and (f), he sees the alterna-
tion going into the (110) direction, and finally the (111) direction, the last being
the Neel state.
Note that if the Ruderman-Kittel interaction is a valid one to use for small
k , then the statistical mechanics will be very well described by the molecular
238 6. Magnetism in Metals

field theory which we develop in the next volume. For then, the interaction is
long-ranged and practically nodeless (Ji) is ferromagnetic out to distances
,.....I/kp and is very small beyond) and the criteria of the molecular field theory are
met at all but the lowest temperatures, where spin-wave theory is applicable.
Because the indirect exchange theory is used in the description of the mag-
netic properties of the rare-earth metals and alloys [6.33], it is interesting to note
that in many cases the angular momentum of the f shell in these ions is not
quenched, and the total II angular momentum must be specified and not just
the total spin SI. That is, the magnetic degrees of freedom of each rare earth are
described by 2jl + I eigenfunctions, and not by 2s 1 + 1. But this is easily taken
into account by using the definition of the Lande g factor. In the subspace of
the 2jl + I eigenfunctions, the following equality defines the Lande factor gl,

(6.74)

where MI is the magnetic moment operator of the ion. Subtracting II from both
sides of the equation, we obtain the prescription useful in the present case:

replace SI by (gl - 1)/1 (6.75)

in ~IE for all rare earths except whenj = 0, as in the case of Eu in some states.
Because (gl - I) can be positive or negative, a sort of "charge" is introduced

Table 6.2. g-factor and angular momenta of rare earths

Number of
electrons
in I shell Symbol s j g-1

0 La 0 0 0
1 Ce I
l" 3 5
l" -t
2 Pr 1 5 4 I
3"
3 Nd 3
l" 6 9
l"
_1-
\I
4 Pm 2 6 4 2
3
5 Sm 5
l" 5 l"
5
-t
6 Eu 3 3 0
7 Gd 7
l" 0 7
l" +1
8 Tb 3 3 6 H
9 Dy 5
l" 5 15/2 H
10 Ho 2 6 8 +I-
II Er 3
l" 6 15/2 H
12 Tm 1 5 6 H
13 Yb I
l" 3 7
l" H
14 Lu 0 0 0

Note: From the following formula: g - 1 = UU + 1) + s(s + 1) - l(l + 1)1/2j(j + 1).

 6.8 Kondo Effect 239

into the indirect exchange theory: ions with opposite signs of (g/ - 1) will
interact antiferromagnetically for ferromagnetic J li , and vice versa, so that this
gives to mixed rare-earth alloys yet another degree offreedom.
The indirect exchange Hamiltonian is thus,

(6.76)

except whenjl = 0, when S/ is used. Table 6.2 lists the rare earths and their effec-
tive "spin charge" gl - 1.
In transition metal ions the angular momentum is quenched because of
strong crystal field effects on the relatively extensive d orbitals. This is reflected
in experimentally measured g factors close to 2, the theoretical spin-only value.
In those cases, the correct low-lying states are designated by m., and the correct
vector operator is just St.

6.8 Kondo Effect

In the preceding, a large number of spins arranged in a regular way in a metallic

matrix were found to correlate their orientations via the Ruderman-Kittel inter-
action mediated by the nonmagnetic conduction electrons. The symmetry of the
ground state depended on geometric factors, such as the ratio of distance between
magnetic atoms to the deBroglie wavelength of an electron at the Fermi surface.
In this section, we give the reader a brief introduction to the case of random,
dilute, magnetic alloys, in which the concentration of spins may be as few as
parts per million (ppm). In the extremely dilute limit where the self-energy of
each spin is the dominant factor, one finds the Kondo effect [6.34]: an anomaly
in the conduction electrons in the vicinity of the impurity spin. In the moderately
dilute case, when interactions among neighboring spins become more significant
than the self-energy of each, the spin-glass phase-a condensed phase in the
presence of a high degree of disorder-will yield the presumed ground state. In
either case, the internal degrees of freedom peculiar to spins are the cause of
highly unusual phenomena, which theorists, assisted by a variety of experi-
mental facts, have been seeking to understand for the past two decades and to
which we shall return more comprehensively in the companion volume.
One goes back to the 1930's for the first inklings of these phenomena. Meis-
sner and Voigt [6.35] first noted an anomaly in the resistivity of nominally pure
gold samples, in reality containing trace impurities of iron from the manu-
facturing process. These showed an increasing resistance as the temperature was
lowered below some 10 K. Such behavior is in diametric opposition to that of
ordinary metals and alloys, which have their minimal resistance at T = when °
all thermal vibrations have ceased. A recent confirmation of the Meissner-Voigt
discovery is shown in Fig. 6.9. The effects are highlighted in Fig. 6.10, from
which the resistivity contributed by lattice vibrations has been subtracted out. A
240 6. Magnetism in Metals

0 .63
\o
0.6 1
\, / 0.59
.2% Fe in Cu /'
,..-0 0.57

/0
000.0 010-0- 0 - 0---"
~ 0.5 1
"
J 0 .55 0"

~0.49 \g / t ;
0 .47 \\, 0 .33 "ii:
0.45 _ _ '~.1% Fe in c~. . .o""',o 0 .31
I .&1> ....0
0.43 0<>
~ 0 0.29
~, 0/ 0.27
°b 0 .05% Fe in Cu .,p /
p..-J> 0.25
~
o 20 40 60 8(p .23
T [K)
Fig. 6.9. Resistance as a function of temperature for dilute magnetic alloys; note various
scales [6.361

1.5
1~-9p-.-r""'''---'--'-''''---'--'-'rrT,..,....--.-r-r--.-.,....,..rrr---'r-r-r-r-=

E
...
a.
a.
~1.4
E
<..>

E 63 ppm
.s::
.2 ·22 ppm
o
Q.. 1.3 ·22 ppm
<1 0 22 ppm
· 560 ppm

0.04 0.1 0.2 0.4 0.6 1.0 2.0 4 .0 6.0 10 20 40

T (K)
Fig. 6.10. Low temperature Resistivity of CuFe, per iron atom; with phonon contribution
subtracted out [6.371 Note logarithmic scale
 6.8 Kondo Effect 241

few ppm iron in eu or Au are seen to be capable of substantially affecting the

resistivity and, as it turns out, a variety of other thermodynamic properties as
well. Between the low-temperature saturation of the resistivity (at lK in the
example of Fig. 6.10) and the high temperatures at which the magnetic contri-
butions become negligible, there exists a wide range over which the resistivity
follows an approximately logarithmic behavior. This is seen in Fig. 6.11 for
several samples, which satisfy a universal law, once the temperature is expressed
in terms of suitable units. The scaling temperature (the unit) is constant for a
given alloy but varies from one to the other. It is most appropriately denoted the
"Kondo temperature" T K, given as:

TK = (D/k B) exp(-D/2IJI) (6.77)

with D = bandwidth, J the exchange parameter (J > 0 for ferromagnetic

s - d coupling, < 0 for AF coupling), kB = Boltzmann's constant. Kondo first
gave an expression for the resistivity [6.34]:

R = Ro(T) + c(P/D)s(s + l)A [1 - ~ In(D/kBT)] (6.78)

in which Ro is the nonmagnetic contribution, A is a lumped constant, s = spin

of the impurity, and the other parameters have already been given; the Kondo
temperature TK signals the onset of peculiar behavior in (6.78) at T ~ T K. In
fact, (6.78), derived by perturbation theory, ceases to be valid at low tempera-
ture. Since the enunciation of the logarithmic law, it has been observed in
a number of laboratories and Fig. 6.11 is typical Kondo behavior.
It is also generally accepted that this behavior is the consequence of a large
"soft" cloud of conduction electrons' spin polarization resonating about the
impurity's local moment, so that the role oftemperature is to break up this cloud

1.0 ~-_--=="--

~0.8
£
0:::
::::. 0.6
I-

0:::0.4

0.2

0.1 10
T/TK
Fig. 6.11. The normalized resistivity R(T)IR(O) for a variety of samples, plotted as function
oflog (TIT,,). The results suggest a universal curve [6.38]
242 6. Magnetism in Metals

and reduce the strong scattering that such a resonance produces. It should then
be possible to predict at which concentration of magnetic impurities the overlap
of neighboring clouds becomes more significant than the energy binding each
to its local moment. At such a value of the concentration, the spin glass phase
must become important; at yet higher concentrations, the ordered magnetic
alloys studied in previous sections become relevant.
The one-spin interaction Hamiltonian appropriate to the very dilute Kondo
limit (ppm) is precisely half of what we considered in (6.50,51), viz.

with J = J err of (6.48). (6.79)

The isolated spin is taken to be located at the origin, without loss of generality.
Now, however weak the interaction parameter J may be compared with the
other parameters: Fermi energy ep , electron bandwidth D, etc., it cannot be
considered a small perturbation because of a singularity at J = o. Specifically,
one can show [6.39]:
In any large region surrounding the magnetic defect, the ground state spin
(impurity + conduction band) has the value Stot = s - ! for antiferromagnetic
interactions and Stot = s or s + ! for ferromagnetic interactions.
The proof of this follows the arguments in Chap. 4, by comparing the ground
state of the present system with that of a reference system for which the ground
state quantum numbers are known. It is a tricky proof, however, as the energy
levels of the conduction band form a continuum and they are so dense that their
fluctuations can overwhelm the impurity. Cragg and Lloyd [6.39] have investi-
gated this problem by numerical means and use of the renormalization group,
finding for J > 0 (ferromagnetic coupling) a dependence of the ground state
symmetry on the number of iterative steps in their procedure. For J < 0 (AF
coupling), however, the predicted Stot = s - ! is obtained without ambiguity.
In any event, the significant feature is the change in symmetry as the sign of J is
varied; this indicates that perturbation theory in powers of J has a vanishing
radius of convergence, as the point J = 0 is singular.
It is known from thermodynamics that the entropy 9"(T) relative to the value
at high temperature 9".. is given by

9"(T) = 9".. -
..SdT'c(T')/T'
T

with 9".. = kBln 2 for spin one-half. In Fig. 6.12 the specific heat curves of dilute
copper-iron alloys are shown; the area under the suitable integral yields kBln2,
implying s = 1/2 on the iron and confirming that the ground state entropy is
zero. Such experimental evidence of a singlet ground state is paralleled by the
variational estimates concerning the ground state of the system for antiferro-
magnetic coupling [6.40] and by the above-mentioned theorem [6.39] applied to
 6.8 Kondo Effect 243

s = 1/2. A variety of approximate methods lO have all shown the expected de-
crease in total moment as the temperature is lowered (J < 0) or the increase
(J> 0).

1.0

0.8
:.:::
~ 0.6
--
E

Fig. 6.12. The excess specific

heat for the CuFe alloys, corres-
ponding to a total entropy of
precisely kBln2 per spin [6.36]

In (6.55) we have already obtained the self-energy of a magnetic ion in a

metal, seen to be finite and symmetric in J. It is thus the higher-order terms that
must lead to the expected anomaly, third- or higher-order in the energy or
second- or higher-order in the Born expansion of the scattering amplitude. To
pinpoint the origin of the difficulties, consider what occurs when one attempts to
eliminate the perturbing Hamiltonian ~x by a unitary transformation, even to
lowest order in the "small parameter" J. First,

(6.80)

~x = (-J/2N)S. r:.
kk'
c'tmO'mmlCk1ml
mml

in which 0' = (ax, O'y, O'z) is a vector with 2 X 2 matrix components, the Pauli
spin matrices inserted into (6.79) ; cf (3.79). Following a common procedure, 11
pick an operator 0 such that

[0, ~o] = ~x . (6.81)

It will be O(J). A canonical transformation ~- e-o~ eO produces

10 See the various chapters of [6.41].

II A simplified version of time-ordered perturbation theory.
244 6. Magnetism in Metals

$f'o -+ $f'o - [D,~] + 1! [.0, [.0, $f'o]] + ... 1 (6.82)

~x-+~x - [.0, ~x] + ...
and

(6.83)

The operator linear in J has been transformed away (except for the small part
"on the energy shell" which leads to scattering and thus cannot be physically
removed). The .0 which achieves this and solves (6.81) is,

(6.84)

It is well defined except on the energy shell (e = e'). Writing S,se(O) as Sas~ and
S.le as SpAg using a summation convention on repeated indices, we find, to new
leading order in J, a Hamiltonian

(6.85)

in which we have used the fact that S commutes with all components Se of and le'
For classical spins, the components of S commute with each other and the
second commutator vanishes. This commutator is also absent in ordinary poten-
tial scattering. The first commutator is representative of conventional scattering.
le and Se are both quadratic forms in fermion operators; their commutator is,
again, quadratic. The contribution from this term to the scattering amplitude
(which, in first order, was J) is O(P/D). It is negligible for small J.
For quantum mechanical spins, the second commutator does not vanish. In
terms of components, the spin commutation relations of Chap. 3 are

(6.86)

(with h = 1, eapy = ± 1 for even/odd permutations of x, y, z and 0 otherwise.)

The operator s~ Ag is quartic in the fermion operators. Among the contributions
to electron scattering on the energy shell, we find such new terms as

(6.87)

in which fk' = <ct,ck,) is the Fermi function. The principal part of (6.87)
diverges logarithmically as ek -+ ep and T -+ 0; the Fermi function discontinuity
at ep is sharper at low temperature.
 6.9 Spin Glasses 245

Indeed, the calculation by Kondo showed that the scattering amplitude of an

electron (k) including constructive interference between first and second orders
became

(6.88)

at T = O. The scattering cross section should then vary as

(6.89)

to the stated order. A typical energy is Gk = Gp + kT; then, the resistivity is:
(6.90)

in which Ro contains the temperature-independent contributions, R J is a lumped

constant. The result is the Kondo formula quoted earlier, (6.78).
The magnitudes of TK estimated from resistivity measurements range from
less than lK (MgMn, CuMn, CdMn, etc.,) to greater than 1000 K (AuTi, CuNi)
[6.42] while typical conduction electron bandwidths D range from 103 to 104
X k B • From this we may assume that even at the largest TK'S the Kondo problem
remains within the realm of "weak coupling", however peculiar the ground
state might be. The analyses by which one extracts the leading divergences
from an expansion in a "small" parameter are beyond the scope of this volume,
but will be taken up again in the companion volume on finite temperature effects.
See also, chaps. 7-9 in Vol. 5 of the Magnetism series [6.41]

6.9 Spin Glasses

As the polarization cloud surrounding a magnetic ion which is antiferromag-
netically coupled to the conduction electrons effectively screens it from further
interactions with other magnetic ions, there must be a phase transition to an
interactive phase when the interaction energy between two neighboring spins
becomes more favorable than the Kondo binding energy of each. For spins
ferromagnetically coupled to the conduction sea, this question does not pose it-
self: the Ruderman-Kittel interaction will connect them at all concentrations c.
We can estimate c for the AF coupled ions by recalling that the interaction
energy is ~(J2ID) (aIR/J)3 cos(2kpRIj)' With c --- (aIR/J )3, this yields

Cerlt = A(DIJ)2exp(- DI2IJI) (6.91)

where A is a constant. For JID much less than 0.1, the critical concentration
drops exponentially to zero. Much of the current research in this topic is centered
246 6. Magnetism in Metals

on the distinction, and competition, between Kondo and spin glass phases. It
is clear that the spin glass (SG) phase is the more common.
In the SG phase, the spins interactiong via the Ruderman-Kittel interactions,
have either ferromagnetic or AF couplings depending on the phase of cos (2k p
Rtf)' an uncertain quantity at large distances Rtf" Later, we shall see that the
principal requirements for the existence of a SG phase concern the AF bonds:
the concentration of AF bonds, disposed at random, must exceed a critical value,
the magnitude of which depends on the topology of the lattice and the effective
range of the interaction. We shall return to these considerations in due course,
but first, point out the results of numerical experiments by Binder and Schroder

~
0 0 0 0 0 0 0
--,
0/0 0 0 0 0

0
0

0
0

0
0

0
t
0
0

0
0

0 0 0 0 0 0 0

0 0" 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0

tt
0 0 0

0 0 0 0 0

0 0 0 0 0 0 0

00
2000 I
I
I

1000

500

200

100

50

20
o

10 Fig. 6.13. Interactions leading to

Spin Glass phase in dilute magnetic
alloy. as computed in [6.43]. Top
5 figures represent Ruderman-Kittel
bonds among spins randomly distri-
2L-__ ~ ____- L_ _ _ _ ~~ _ _L-~ buted in dilute alloy. Lower figure
2.10- 4 -1.10- 4 0 1 .10- 4 - 2 .10- 4 shows the computed distribution of
J/Jo ~i which results
 6.9 Spin Glasses 247

on the statistical properties of some half a million bonds J ll , as summarized in

Fig. 6.13. Note the near symmetry in the distribution of positive and negative
bonds. [6.43].
Given a particular set of bonds, the spin orientations in the ground state have
to minimize the energy. While presenting the appearance of total disorder, the
spin-glass ground state is, in fact, highly correlated. It is parametrized by the
internal fields (also called "molecular fields") felt by each spin. The distribution
P(h) of molecular fields h characterizes the ground state uniquely. On the other
hand, the ground state itself is not unique, for there may be many different
configurations yielding the same energy; all, however, with presumably the same
P(h).

400
b
90

Fig. 6.14 a, b. Properties of spins, from a computer simulation spin-glass of Fig. 6.13. (a) Distribu-
tion of ground state internal fields and (b) Density of elementary excitations. Curves shown are
smoothed approximations to the histograms in [6.44]

A computer simulation of the ground state and elementary excitation

spectrum of the Ruderman-Kittel SG has been worked out by Walker and Wal-
stedt [6.44], with the SI taken to be classical vectors. The spectrum of elementary
excitations is found by quantizing each spin along the direction in which the
classical spin points, then diagonalizing leading terms in the Holstein-Primakoff
LSW approximation, the quadratic Hamiltonian discussed in Chap. 5. The re-
sults of their computations, confirmed by semi-empirical theoretical considera-
tions, lead to the curves of Fig. 6.14 obtained for 324 spins on a fcc lattice,
c = 0.003. It should be noted that the elementary excitations (magnons) are
extended states for the entire low-frequency spectrum, although the states in the
high-frequency "tail" are localized at the various strongly interacting closest
neighbor pairs. We return to this model in a later discussion of the finite temper-
ature properties of the SG phase, but now fix our attention on other popular
models.
The simplest "separable," model [6.45], takes the signs of the bonds to be
separately random
248 6. Magnetism in Metals

(6.92)

with the 8/ = ± I random variables, distributed independently according to a

specified probability distribution. The F(R/j ) can be chosen to mimic the Ruder-
man-Kittel long-range interaction, but two simpler choices can be made:
F(R/j) =1= 0 for R/j a nearest-neighbor vector only, or F(R/j) = constant, inde-
pendent of distance. The former leads straightaway to the ordered n.n. models
which have been extensively studied, and are found in the various chapters of
the present book. The long-range version [6.46] is soluble in closed form by
mean-field theory. The trick leading to the solution of the separable models is
the transformation from spins SI to new spin variables (1/, as follows: S/ == 81(1/
(= ±(I/). Only for a quantum Heisenberg model is this transformation in-
admissible, for it is forbidden to invert all three components of a spin vector
[recall the example following (3.6) of Chap. 3]. For classical spins, or quantum
XY or Ising models, the transformation brings the Hamiltonian into a simpler
form

(6.93)

The latter is the type Hamiltonian we have already considered in Chap. 5. Now,
obtaining correlated (l1°(lj in the ground state (e.g., parallel spins in the case
F> 0) does not mean the physical spins SI have an obvious correlation, for the
quantity S/oSj = 818j(l/o(lj will have the sign of 8tBj and the appearance of a
random variable. Nevertheless, it is obvious that the S's are as strongly cor-
related as the (I'S, for a given set of 8/. The long-range version of this model is
remarkably simple; We illustrate for the ground state, replacing Fby its average

~.pL.R. = - <~(~ (1/)2. (6.94)

For <F) < 0, we have ~(li = 0, a condition satisfied by almost any random
configuration of the (I's. The ground state is then totally disordered, just as for
N noninteracting spins, regardless of the magnitude of F! This implies also the
lack of an order-disorder phase transition at finite T, as we shall determine later.
In the absence of correlations, the SG phase does not exist, so these spins are
either perfectly free or are in their Kondo ground states. At the opposite extreme
is <F> >0, for which the ground state (It's are all parallel. Although the cor-
responding St's would give the appearance of being uncorrelated, we already
know this is not the case.
The physical problem studied numerically by Walker and Walstedt does not
have all bonds positive or negative, but rather a mixture of both. There have
been several attempts to introduce models that combine this element together
with some simplification, sufficient to render the analysis tractable. Notably,
Edwards and Anderson [6.47] and Sherrington and Kirkpatrick [6.48] have
 6.9 Spin Glasses 249

introduced models with such features, but these have also, to a great extent,
resisted analysis. Although such studies have attracted much attention lately,
the nature of the ground state is still not clear, not to mention the question of the
existence of a thermodynamic phase transition at finite temperature! Recent
studies indicate [6.49J that Tc = 0 in less than 4 dimensions (if the bonds are
random with zero mean, and restricted to nearest-neighbors). Nevertheless, the
study of ground state and low-lying states in such models is of the greatest
interest.
Take, for example, a model in which Ising spins S/ = ± 1 are regularly
arrayed at the vertices of a square lattice with nearest-neighbor bonds J/J = ± 1.
Let P be the probability of an AF bond, be in the range 0 ~ P ~ 1/2 (p > 1/2
can be obtained by symmetry). The ground state energy, Eo, is easily found nu-
merically; for P in the range 0 < P ~ Pc (Pc""" 1/4) the energy increases mono-
tonically from - 2N to approximately _21/2 N. From Pc on, Eo is independent of
P and maintains the value _2"2 N; this is suggestive of a distinct phase. And
indeed, in 3D, Eo starts at -3N atp = 0 and rises to approximately -3" 2N at
Pc, staying constant thereafter. In ID, we of course have the trivial result; -IN
and _11/2N, with Pc = O. It's not known whether in 4D the numbers are 4 and
4" 2 , respectively, but this is certainly suggestive; the typical behavior being
shown in Fig. 6.15. The ground state properties have been studied, in a variety
of models, by a variety of techniques too numerous to list. 12

p
o Pc 0.5

ferro magnetic
Fig. 6.15. Ground state energy per spin in
z n.n. model, bonds ± 1, withp = probability
of AF bond. Energy is independent of p in the
I--~~ SG----I-I SG phase, and takes the values - D'/2 for D =
1,2,3 (\. c., sq. and sC);Pc varies according to
D

A somewhat more tractable model SG is based on the "spherical model";

each spin interacts with every other spin, the N(N - 1)/2 bonds J/J = ± I/N"2
being randomly selected. If the spins are treated in the spherical model, i.e.,
~Sl = N is required (and there is no constraint on the individual S7) the
ground state and thermodynamic properties can be calculated [6.53] from the
eigenvalue spectrum of the J/J matrix. Fortunately, this spectrum is known
exactly [6.53, 54]. For example, the ground state energy is precisely Eo = - N.

12 Ground state properties, particularly the problems of "frustration" were pioneered in [6.50].
The quantum features are treated in [6.51] and also in [6.45]; the S-K model [6.48] in [6.52].
250 6. Magnetism in Metals

6.10 Magnetism Without Localized Spins: Preliminaries

The strongly magnetic properties of the iron transition series metals and of their
alloys must be explained by band theory. This would be required for the fol-
lowing two reasons alone: the number of Bohr magnetons per atom is generally
far from being an integer, and both calculated and observed band widths are of
the order of electron volts. Neither of these facts could be accorded with a
scheme based on localized spins, and together with such additional evidence as
the abnormally high specific heat (which can only be explained by a continuum
of states for the magnetic carriers) they point instead to the need for associating
the uncompensated spins with Bloch electrons rather than with their Wannier
counterparts in the first approximation.
The band theory of magnetism is currently a very active field of research, and
it is not yet possible to predict the degree of accuracy and sophistication it may
attain. At present, it is far less developed than the relatively straightforward
Heisenberg theory for insulators; but the physical mechanisms are established
and some of the consequences therefrom approximately understood. Let us start
by visualizing, inaccurately perhaps, localized spins in the indirect exchange
theory being gradually modified so as to allow some overlap and the formation
"magnetic" electron bands. The energy gap against the excitation of electrons
in the magnetic states disappears and a finite density of states appears at the
Fermi surface. This now allows charge fluctuaitons, hence the conduction of
electricity by magnetic electrons; it permits them to contribute to the electronic
specific heat, microwave absorption, etc. On the other hand, there is no parti-
cular reason why the inherently magnetic properties-such as the magnon
spectrum-should be affected to leading order. As with indirect exchange, there
is no universally valid reason that the long-range order in the ground state be
ferromagnetic, and spiral or other anti ferromagnetic configurations, such as are
actually observed in Mn and in Cr, are not difficult to reconcile with this ap-
proach.
One salient difference with the indirect-exchange theory, is that nonmagnetic
s-like bands can be ignored in the first approximation. Doubtless some indirect
exchange still goes on, but the dominant, primary interactions are now among
the magnetic bands themselves. An interesting point of view has it that the d
electrons (e.g., in Fe) effectively split into two sub-bands; one of which is nar-
row, so that the electrons it contains are well localized like the rare-earth/states,
and the other is more like an ordinary conduction band. It is quite likely, that
even were such a model describable by the indirect exchange Hamiltonian of
(6.52), the perturbation-theoretic solution on which the Ruderman-Kittel inter-
action is based, would not be valid. The most spectacular feature of the band
theory is that with neglect of the nonmagnetic electrons the remaining mag-
netic ones do not number any particular rational multiple of the number of
atoms. Therefore, the occupation number of the magnetic bands is a vital, and
often an adjustable, parameter in the theory.
 6.10 Magnetism Without Localized Spins: Preliminaries 251

For definiteness in this study, let us assume that whenever an ordered state
exists, it is ferromagnetic. We then calculate the magnon spectrum, and if some
magnon has negative energy, the assumption is then evidently false and the
ground state must be some species of AF. This procedure is simpler than cal-
culating the energy of every possible ordered state, although, as we saw in the
indirect exchange theory, it leads to almost identical results. The first important
physical fact we shall wish to demonstrate is the existence of a threshold magni-
tude for the interactions. That is, we shall show that below a certain magnitude
of the Coulomb repulsion and Hund's rule coupling parameters, there is no
metallic magnetism. To establish the possible orders of magnitude, we list in
Table 6.3 the relative strengths of various plausible physical forces.

Table 6.3. Scale of energies

Order of magnitude Mechanism

-lOeV (a) Atomic Coulomb integrals U

I
!
(b)
(c)
(d)
Hund's rule exchange energy, J
Energy of electronic excitations violating Hund's rule.
Electronic band widths, W (also denoted D)
- 1 eV (e) off -7- (density of states at Fermi Surface)
0.1 - 1.0 eV Crystal field splittings
10-2 - 10-1 e V Spin-orbit coupling
kTcorkTN
10-4 eV Magnetic spin-spin coupling
Interaction of a spin with external field 10 kG
10-6 - 10-5 eV Hyperfine electron-nuclear coupling.

In the creation of magnetic moments in the metal, the competition is between

the "kinetic energy" terms (d) and (e), and the potential energy terms (a)-(c),
representing forces internal to each incomplete shell in each atom of the solid.
For isolated atoms, the kinetic terms are zero and the magnetization of the
incomplete shell proceeds according to Hund's rules, detailed and proved in an
earlier chapter but what happens when kinetic energy is introduced?
And once the criteria for the existence of magnetic moments are satisfied,
what is their alignment? We have already observed in connection with the
indirect-exchange theory, that the geometric factor kpa plays an important role,
i.e., the filling of the band. For an almost full or almost empty band, k p ~ 0 and
long wavelengths predominate-Le., ferromagnetism. For nearly half-filled
bands, some species of AF are possible. The use of a variable axis of quantization
for local spin polarization as an adjunct in the study of electron interactions in
magnetic metals has become increasingly popular in the recent literature [6.55].
Much of the commonplace understanding about itinerant ferromagnets
comes from the example of nickel , a ferromagnetic metal having 0.6 Bohr magne-
252 6. Magnetism in Metals

tons per atom, with 0.6 holes per atom in the d band. Alloying with copper,
having a similar band structure but a higher Fermi level, "plugged up" the holes
and quenched the ferromagnetism in direct proportion to the added number of
electrons. Now, the important discovery of metallic ferromagnetic alloys made
out of ordinarily nonmagnetic constituents [6.56]-e.g., ZrZnz-has made the
need for a reliable theory of itinerant electron ferromagnetism more urgent.
Neutron diffraction has shown that some of the magnetic moment in such an
alloy resides between the atoms and not on them, i.e., is identified with the
bonding, mobile electrons [6.57].
The 1966 systematic review of itinerant electron magnetism theory by
Herring [6.2], did much to place the topic in perspective. In the following sec-
tions, we deal with easily understood aspects of this knotty many-body problem:
the exactly soluble low-density limit (appropriate to the few holes in Ni) and the
case of a nearly half-filled band. We proceed to an elementary study of the
magnon spectrum in magnetic metals, and suggest the type of calculation re-
quired to understand the magnon spectrum of Iron.

6.11 Low-Density Electron Gas

The low density limit of the electron gas can be calculated systematically. First,
one recognizes that the long-range part of the electron-electron Coulomb
interaction (the troublesome part) is of no interest in the magnetic problem:
the high cost in energy to create a plasmon guarantees that charge neutrality is
maintained throughout the metal. On the other hand, there is no requirement
for charge neutrality on an atomic scale. A Hamiltonian incorporating the
kinetic energy of electrons in a single nondegenerate band, with intra-atomic
Coulomb repulsion U between electrons of opposite spin (electrons of parallel
spin cannot simultaneously occupy the same atom) has been extensively studied
by Hubbard [6.58] and a number of other workers, in the form

% = %0 + U 2: nil nil
i
(6.95)

with %0 from (6.24) or (6.27). We shall solve it exactly for two particles. The
triplet eigenstates and eigenvalues are

and Ekfc1 == E(k) + E(k') (6.96)

for all k, k' in the first BZ, with E(k) the Bloch energy. Two projections of(6.96):

and (6.97)
 6.11 Low-Density Electron Gas 253

are also, trivially, eigenstates of ~belonging to the same eigenvalue. However,

for the same two electrons in a singlet state, the energy involves U, and is
obtained from scattering theory. For brevity define

(6.98)

having band ("kinetic") energy Ew [defined in (6.96)]. Then let the eigenstate
have energy Wkk' and be of the form

(6.99)

the usual form, an incoming wave and a scattered part. The Schrodinger equa-
tion consists of three parts:

(6.100a)

(6.100b)

(6.100c)

Equating the coefficients of bw yields an expression for the energy eigenvalue

Ww = Ew + ~ II + ~ f~ LfJ. (6.101)

The scattering amplitudes L f are obtained from the coefficients of the bHf,k'+f'
After minor algebra, the result is

L = 1 x U (6.102)
f Ww - EHr,k'+r 1 + UGo

where

1 1
Go == N~E Wkk' . (6.103)
r k+r,k'+r -

Consequently, the energy is

(6.104)
254 6. Magnetism in Metals

which also serves to define the scattering t matrix, which is the expression of the
interaction energy between the two particles with all multiple scattering taken into
account. For all practical purposes the Ww can be replaced by the unperturbed
Ekk' from which it differs by only 0(1/ N). (The exception is for bound states, i.e.,
the zeros of 1 + UG o = 0 which, if they exist, lie above the highest Ekk' and
thus will not be relevant to the subsequent discussion. (In passing, one notes
that for negative U, no matter how weak, such a pole in the t matrix could
develop, below the lowest Ekk'; the lowest of these bound states belongs to
q = 0, and is the zero-momentum Cooper pair, famous in the BCS theory of
superconductivity [6.59]. Our interest, however, is in fairly large, positive U).
The modification of 2-body scattering required for the low density electron
gas is straightforward: the range of q's in (100) et seq., especially Go, (6.103),
must be restricted to the unoccupied states: k + q, -k' -q must lie outside
the Fermi volume, i.e., E(k + q) and E( -k' -q) must exceed Ep • Denote the
suitably modified Go by Go and similarly for l(Ekk') to obtain the effective
Hamiltonian for the singlet pairs
_ U
(6.105)
t == 1 + UG o
To proceed we need evaluate iCE). Our approach is greatly simplified, yet retains
essential features. In a dilute electron gas, the Fermi level is near enough to the
bottom of the conduction band(s) (or, for holes, the top) that the effective mass
approximation E(k) = tPP/2m* is appropriate. With this, all the integrals that
go into the calculation of the ground state energy become simple.
First, we introduce the cutoff ko to retain, properly, the volume of the first BZ

l ~ I = I = 41t(k oa o)3. (6.106)

N k~BZ 3 21t

The density parameter p is defined: the number of electrons per atom per band
in a given spin direction i or ,l.. With a spherical Fermi surface at k p , p is given as

1~ k<kp
1 == p = (k p /k o)3 (by comparison with (6.106». (6.107)

Similarly the kinetic energy (KE) per band per spin component is

KE = E(k) =
N~ E(k)
= N
[f dkk 4 h2/2m*]
l dkk2
~ k<kp ,,-0- - - : - k - - -

k<kp ~1
o

(6.108)
 6.11 Low-Density Electron Gas 255

We start with the exact expression for Go:

and evaluate it, neglecting k,k' { ko, in the dilute limit. An approximate but
very convenient expression, independent of k and k', is what results:

- - ( /2)3
G
o- a 'It
f d3 1
q (h2/2m *)2q2
- 3m* (1 _
- h2k20 P
1/3) (6.109)
k.>,>k.

using (6.107) to eliminate k F • With this, we obtain a 1 matrix which is inde-

pendent of k,k' and depends explicitly on p. Kanamori [6.60] has made a
different simplification, with basically similar final results.
As the combination h2k5/3m* occurs in the interaction and in the KE, we
eliminate it. It is recognized as a measure of the bandwidth, in units appropriate
to the effective mass approximation. We therefore denote it Was usual

(6.110)

With Go and 1 suitably re-expressed in terms of W, and with the KE also

simplified, we now have for the ground state energy of (6.105) the perspicuous
expression

E NW[2 9 513 Up2 ]

o= °10 P + W + U(1 _ p 1/3 )

9 Upl/3 ]
5" + + U(l
- S13[
- NW p W _ p1/3) •
(6.111)

Both KE and interaction energies have been expressed in terms of the sole
physical parameters of interest: the bandwidth W, the coupling constant U, and
the electron density p. Writing a similar expression for the totally ferromagnetic
state at this stage, we do not bother with the added complications of partially
magnetized states, so PI = 2p, and PI = 0; 1 vanishes and we find

E ferro = N W (2 p) 5/3 ( 10.

9 ) (6.112)

A criterion for the occurrence of ferromagnetism can be obtained by com-

paring the two expressions. If (6.112) is lower, then the ground state has spin
..Y ih. After a few most elementary algebraic manipulations, this criterion is
seen to be
256 6. Magnetism in Metals

UpI/3
U + w> 0.5139, (6.113a)

There is no solution for p < (0.5139)3 = 0.136. Other wise, we find ferromagne-
tism when

U I
(6.113b)
w> pl/3 _ (0.136)1/3 for p> 0.136 .

The phase diagram is plotted in Fig. 6.16, where the (uncertain) application to
high-density (half-filled bands) is indicated by a dashed curve. Note the existence
of a critical density Pc = 0.136. For densities lower than this no magnetism is
possible, even at infinite interaction strengths U = 00. This general feature is
in excellent accord with a theorem proved in an earlier chapter, concerning the
non-existence offerromagnetism of 2 electrons, and extends it (for this particular
interaction) to an infinite number of particles as long as the density does not
exceed Pc' In Fig. 6.17 we anticipate the AF spatial ordering. As we have seen in
the indirect exchange theory, the p = 1/2 density is always AF [see point labelled
n = (2p) = I in Fig. 6.6], and we may combine this knowledge with (6.113)

1. 64 f.--- t -- - - - ' - -'--- --.y----;

NonmagnetIc Metal

Nonmag net i c M eta l

o 1- Qc 1.0

--a _b

Fig. 6.16. Phase diagram of I-band Hub- Fig. 6.17. Same as preceding, with magnetic
bard model at T = 0, based on low-density ordering taken into account. The half-filled
theory (see text). Region (a) is low density for band is most easily susceptible to AF ordering
electrons and (b) for holes. The highdensity (Examples: NiO, and also ID Hubbard mo-
results e-e-e- near 0.5 are unreliable, due del [6.62)) but at strong enough coupling in
to limitations of the theory. Magnetic 2D or 3D (not ID) gives way to the ferroma-
region is shaded, and Pc = 0.136 gnetic phase (cross-hatching). The equation
of curve C separating the magnetic phases is
derived in the next section
 6.12 Quasi-Particles 257

to produce a phase diagram of the type shown in Fig. 6.17. Similar diagrams
have been obtained from many other theories, e. g., [6.61] or Liu [6.55], (note
their cor n is twice our p). The exact solution by Deb and Wu [6.62] of the one-
dimensional Hubbard model for a half-filled band conclusively showed this to
be an AF insulator for any finite U and to have a spectrum of low-lying ex-
citations readily identifiable as antiferromagnons. This interesting work is the
generalization to the interacting electrons of the Bethe solution for interacting
spins, but involves mathematical techniques beyond the scope of the present
text. It should be noted that the replacement of Go by its average value as in
(6.109) is not permissible in ID (nor perhaps in 20), for the integral is very
sensitive to the precise value of k,k' in this case. Indeed, for k ,....., k' ,....., ± k p ,
Go""'" 00 in 10 and thus l ~ O. This has the interesting consequence that the
well-known instability of a ID "metal" against perturbations of wave vectors
2kp -is compensated by the vanishing, at precisely the same wave vectors, of the
interactions!

6.12 Quasi-Particles

Interactions affect the states of the system. At a given total momentum, only one
state is absolutely stable; interactions cause higher lying states to decay. This
is shown, by redefining the interacting Fermi sea, whether or not magnetic,
to be the new "vacuum". Excited states are then described in terms of quasi-
particles (fermions) or collective modes (bosons) in one-to-one correspondence
with the familiar electrons (above the Fermi surface, FS) or holes (below the
FS) of the noninteracting electron gas and the plasmons of tl->e charged gas,
as well as the magnons of the magnetic systems.
If one electron is added to those already present, the added energy is

E(k) + N1 k'<k.
2: l(Ew) == E(k) + A(k) . (6.114a)

The quasi-particle energy must be measured from the FS, thus we define it to be

e(k) = IE(k) - E(k p ) + A(k) - A(k p ) I

~ 1i21k2 - k~I/2m (6.l14b)

which serves to define the renormalized quasi-particle mass m in terms of the

original band-structure mass m*. This quantity has been exhaustively studied
for the low density electron gas [6.63] with the results at U = 00 being:

m = m* [1 + 1:1t2 (7 In 2 - 1)(kp a)2]

258 6. Magnetism in Metals

in which a= hard sphere radius. We identify 2a as the distance of closest ap-

proach, i.e., ao in the lattice analog. With (6.106) used to eliminate this, and
(6.107) to bring in the electron density, we obtain the simple result

m= m*(1 + 0.79 p2/3) , (6.115)

in which m is the quasi-particle effective mass and m* the unperturbed, band-

theoretic effective mass.
Along with a change in inertia, the excited particle above the FS can be
scattered to lower energy while an electron within the Fermi sea is excited out of
it. This process can conserve energy, and thus give an imaginary contribution to
Go. As the amount of phase space depends on how far above the FS we start, it
is found that the lifetime 1:k for scattering becomes infinite as (k - k p )-2.
More precisely [6.63], with the above substitution for a and U = 00,

(6.116)

This is not necessarily related to transport properties, as the interactions

leading to this scattering do conserve momentum. An identical formula holds
for hole quasi-particles.
At a density of 1/8, the increase in mass (6.115) is 20%, at a density of 1/2
(where the low-density theory is unreliable, however) the increase is of 50 %. The
lifetime effects of quasi-particles near the FS are, according to (6.116), negligible.

6.13 Nagaoka's Model

Figure 6.17 shows the possibility of ferromagnetism at extremely large U for an

approximately half-filled band. In view of the usual tendency of a half-filled band
toward antiferromagnetism, also shown in this figure but at lower values of the
interaction parameter, Nagaoka's mechanism [6.64] is worthy of special note.
It predicts what would happen if a sufficient number of electrons were removed
from the prototypical antiferromagnet NiO, viz., a phase transition to the ferro-
magnetic metal. The following calculation will show this is plausible in 3D,
whereas it can never occur in ID, where the ferromagnetic phase is rigorously
excluded.
Initially, we consider the U = 00 limit. In (6.46) we see that this effectively
sets the tendency to antiferromagnetic ordering ~ W 2/Uto zero. In this limit, N
electrons, one at each site in the crystal, now have perfectly arbitrary spin
configurations. The ground state is 2N_ fold degenerate. The introduction of even
a single hole changes this dramatically in 2D or 3D (but not for nearest-neighbor
 6.13 Nagaoka's Model 259

hopping in ID where n.n.n. hopping is required [6.65]). To see this, let us calcu-
late the eigenstates of the hole in a linear chain first. Assume a matrix element-B
for moving an electron from a site n to the two neighboring sites n ± 1 in the
linear chain. As we know from the tight-binding theory, (6.8), the eigenvalues
are then E(k) = - 2B cos ka, with W = 4B. The infinite potential energy
inhibits the motion of all the electrons except those in the vicinity of the hole.
Let the initial configuration be
Inn :
• -2• -1• ••2 •.-----
m-3 1'Tl..2 m_1 ml m2 m3 •••
O~-..-- ~ (I)
n : -3 0 1 3 ' ..

with the hole indicated by an open circle at n = 0 and the electrons by dark
circles elsewere. The spins mn = ±1/2 of the electrons are a given, arbitrary,
set. The initial configuration connects to a linear combination of two

00.
• • •0 •1 •2 •3
m·3 m-2 m_1 ml m2 m3
0
-3 -2 -1 .00
1 +
y'f (II)

•2 •3
m·3 m-2 m_1 ml m2 m3
• -2• -1• •0
-3
0
1

with matrix element ""2

B. This, in turn, connects to I and to a new set of con-
figurations denoted III (where the hole is either at -2 or at +2) with matrix
element B. This connects back to II, and to a new set, IV, with matrix element
B, and so forth. The resulting matrix has the appearance

""2""2
0 0
0 1 0
0 0 (6.117)
~=-B
0 0.'·.

0
bearing some similarity to the Hamiltonian we diagonalized in Chap. 5 for spin
waves near the surface of ferromagnets. The eigenvectors

(6. 118a)

are readily found

260 6. Magnetism in Metals

Vo = 2- 1/2 , Vn = cos(qn), Aq = (2/N)1I2 (6. 118b)

for N ~ 00 and E(q) = - 2B cos q. Thus, the local density of states at the
origin of the motion:

NotE) == L 1Aqvo 12 J[E - E(q)] (6. 119a)

q

I
= N ~ J(E + 2B cos q) (6. 119b)

has precisely the usual form for free-particle states. As all the results obtained
are for any, arbitrary set of the {m n } , we see that the extra hole does nothing to
promote ferromagnetism. We can proceed to 2 or more holes by use of a deter-
minantal function, and obtain that regardless of the number of holes, all the
eigenstates in ID are paramagnetic: their energy is independent of the spins in
the absence of an applied magnetic field even in this U - + 00 limit.
In two or higher dimensions the configurations which result from moving
the hole to some point R/ depend on the path taken, except in the ferromagnetic
case. Supposing for the ferromagnetic case a configuration ~IJ connects to a
configuration ~b through two different paths: the matrix element will be -2 (in
some units). Ifin the AF state the two final configurations are different, say ~IJI
and ~b'" then the matrix element connecting ~IJ to 2-1/2(~b' + ~b") in the same
units will be - ~2. This difference favors the ferromagnetic state over all other
configurations, as we now see with a specific example.
We illustrate with a sq. lattice, starting with the hole at some initial site

mlO

'1115 m, ms

mg m4 m2 mII (I)
me m3 m7

m12

with spin indices noted, and distinguishing n.n.n. according to whether they are
neighbors to a single n.n. (_) or to two (x).
The state to which (I) connects is the sum over 4 configurations: (II),
which, in turn, connect to a linear combination of configurations (III), in which
the hole has moved to the n.n.n. spot. It is the motion into the spots marked X
that creates a new situation. These spots are accessible two ways. For example,
the two configurations shown explicitly in (II) connect to the two (of 12 con-
figurations) making up (III). But they are not necessarily distinct; if ml = m4
= m" they are identical. If the m's are random, there is 1/4 probability that
 6.13 Nagaoka's Model 261

ml
1
+ ... +
"2 m, ml m2 m4 m2 (II)
m3 m3

ms ml
and I or (III)
m4 ml ms m4

they are, indeed, identical. If the state is ferromagnetic, they are necessarily
identical and, finally, if the state is AF all odd subscripted m's are + and even
subscripted-, so that they are never identical. It is easily seen that the fewer
the distinct configurations, the larger the matrix elements. In all cases, HI,II are
-B~4, regardless of the m's. But at the second step

HII,III = -B~5 (ferro) (6.120)

HII,III = -B~3 (AF).

AsymptotiCally, the matrix elements must be

HM,M+I = -2B (ferro)

and, approximately,

HM,M+I = -2(3/5)1/2B (AF) (6.121)

to yield a ground state eigenvalue

-4B (ferro)
-4(3/5)112B (AF) (6.122)

and something intermediate for the random case. The smoothness of the ferro-
magnetic state results in the lowest energy. The AF state splinters into many
extra nonidentical configurations such as shown in (III) above, and results in
262 6. Magnetism in Metals

smaller matrix elements, hence a ground state which is not nearly so favorable.
In ID where only a single path connects any two states regardless of spins, this
tendency is lacking. In 3D, where the number of paths connecting various con-
figurations is greater than in 2D, the tendency to ferromagnetism is substantially
increased.
For a small density nh of holes, the energy favoring ferromagnetic order is
thus Anh W, taking the bandwidth W to be proportional to B. On the other
hand, at finite U the energy favoring AF near a half-filled band is A'W2jU.
With nh = t - p, we set the opposing energies equal to obtain the phase boundary
curve C, Fig. 6.17. By symmetry the equation for a more-than-half-filled band
is similar, with the final result being

Ip - ~ I= a WI U , (6.123)

a being some lumped constant.

Richmond and Rickayzen extended Nagaoka's work in a very interesting
way [6.61], formulating a trial function in which all but one of the electron spins
are aligned. The odd electron is localized about a particular site, whilst the re-
maining electrons adjust their motion to the presence of the repUlsive potential
that it represents. This is exactly soluble by the scattering theory that we have
used in many connections throughout this book, and the bound state (if any!)
energy is a variational upper bound to the true energy. If a bound state is found,
the ferromagnetic state is unstable; thus their calculation also provides an
equation for curve C. The results which they obtain agree very well with
(6.123).

6.14 Degenerate Bands and Intra-Atomic Exchange Forces

To assess the role of intra-atomic exchange, we here turn to the special case of
two degenerate d-like bands, with various Coulomb and exchange integrals. We
then assess the situation in the iron series ferromagnetic metals, to see which
are the physical parameters to which theory must address itself.
The interaction of a singlet pair within each band is Uaa = Ubb and by a
t matrix, which we approximate by the constant

(6.124)

For a singlet pair of which one member occupies one band and one the other,
there is a similar expression with Uab replacing U aa • It is a simple mathematical
identity for Coulomb integrals given in (6.32) to prove Uaa > U ab > O. {Intui-
tively, electrons in two distinct orbitals overlap less, hence have a smaller Cou-
 6.14 Degenerate Bands and Intra-Atomic Exchange Forces 263

lomb repulsion, than two electron in identical orbitals). Moreover, for triplet
pairs (e.g., electrons of parallel spins) in two different bands, this Coulomb
integral is further reduced by an exchange correction, (6.41), and becomes
Uab - Jab = U~b' which we also write as Uab(l - j). The parameter j is thus the
fractional inter-band exchange parameter, 0 < j < 1. The ground state singlet
energy thus includes 3 distinct interactious:

(6.125)

It is to be compared to the ferromagnetic state, with half the particles in band

a and half in band b but all with spin "up", which has energy

(6.126)

and with the ferromagnetic state in which all the particles, of spin up, are also
in a single band, say a (this represents spin and orbital magnetism, and should
occur when the perturbations of the solid are too weak to quench the orbital
moments of the individual atoms, as in/shells of the rare earths). This state has
energy

(6.127)

The case of maximal interband exchange j = 1 is special; we compare the

energies (6.125-127) and conclude that (6.127), representing spin + orbital
magnetism, can never lie lowest. The phase diagram showing the region where
the spin-only magnetic moments form is remarkably similar to Fig. 6.16, with
only the numerical value of parameters Pc = 0.04 and Umini W = I being dif-
ferent. Doubling the number of bands allows magnetic moment formation at
much lower densities and interaction parameter U than previously.
In the physically more plausible cases of partial exchange j = JablUab < I
the situation may be quite different. We compared (6.125-127) by numerical
calculation, assuming Uab = iUaa == U, and found: spin + orbital magnetism
never occurs for UIW < 20 nor for densities far from 1/2, so this case may be
eliminated from practical considerations even though it is good to know it exists
(in principle) in the atomic limit W ~ O. Spin magnetism has itself a restricted
range of stability, which depends strongly on j. For j = 0.5 or greater, the
situation is qualitatively similar to j = 1, whereas for smaller values of j (0.2 or
0.1) the regions of stability shrink rapidly and become nonexistent at j = O.
Thus, regardkss of the strength of the interaction parameter U, the existence
of a magnetic moment ultimately depends on the Hund's rule exchange param-
eter, as a stabilizing factor. The situation is clearly summarized in Fig. 6.18.
264 6. Magnetism in Metals

6 Fig. 6.18. Moment form-

ation for 2 degenerate
bands in terms of U aa =
U bb = U, U ab = iU, W =
bandwidth, p = density
per atom per band (1 -
p = density of holes) and
exchange parameter j =
JabliU. For curve labeled
j = 0.8 the region lying
above the curve is mag-
netic, and similarly for j =
0.5. For j = 0.1 or 0.2, the
region enclosed by the re-
spective curves is magnetic.
There exists a region of spin
+ orbital magnetism, not
shown in this figure, near
p = i and for U/W> 20.
Spatial ordering of any
moments (ferro. vs AF)
is not considered in this
calculation

O.S 1.0
P

The spatial ordering of the moments, once they are created, is a delicate com-
petition between several mechanisms already considered in this chapter, princi-
pally, the indirect exchange and Nagaoka mechanisms. But before proceeding,
it is prudent to consider to what extent electrons in real materials satisfy the
various simplifying assumptions we have made, for example, the effective-mass
approximation E(k) = fz 2 k 2 /2m with m = m* the band structure effective mass
before interactions and = in the total mass after the interactions (which are
necessarily strong and thus nontrivial) have been incorporated.
The hypothesis of two kinds of d electrons, has been given added credence
lately by Stearns. Her "95 %local" and" 5 %" itinerant model counters the pure
itinerant or pure Heisenberg models, against which substantial evidence has
been accumulating. In her view [6.66], most (95 %) of the d electrons lie in the
relatively flat portions of the band structure, where the high density of states
(or small W, large U/W) promotes magnetic moment formation, with the small
 6.14 Degenerate Bands and Intra-Atomic Exchange Forces 265

residual fraction occupying states well described in the effective mass approxi-
mation, with effective masses --- mel. To see this, let us examine the Hartree-
Fock band structure of iron, as shown in Figs. 6.19 and 20 reproduced from her
article, "Why is iron magnetic?" [6.66]. There are shown the flat parts of the d
bands, labeled dloc and the itinerant parts, labeled dit , drawn in heavy lines. The
latter have a curvature corresponding to approximately the electron free mass
mel' and Fermi wave vectors that have been given as [6.69]

ak p +/21t akp -/21t

Fe 0.50 0.19
Ni 0.65 0.53 (6.128)

-ky

MAJORITY SPINS MINORITY SPINS

MAJORITY SPIN Fig. 6.19. Fermi surface of iron, as

given in [6.67], here adapted from
[6.66]

-1.2

~-1.4F r P N r H
~
w MINORITY SPIN
~ 02

Fig. 6.20. Band structure of iron,

-1.0 for majority and minority electrons
-1.2
as calculated in [6.68] and reproduced
F r P N r H in [6.66]
k
266 6. Magnetism in Metals

in agreement with tunneling experiments. Various other experiments indicate

that the s electrons are only weakly spin polarized, and that the nodes in the
Ruderman-Kittel oscillation make this weak indirect exchange mechanism
favor antiferromagnetism. It follows that the ferromagnetic alignment arises
from the d electrons themselves; but the above wave vectors are all indicative
of AF (see Fig. 6.8 f, or Fig. 6.6 for the expected ground state symmetry at the
above values of k p ), except the minority one for Fe, k p _. There is thus a mystery
to be resolved. Could it be that nearest-neighbor ferromagnetic exchange is
sufficiently strong that Heisenberg-like ferromagnetism coexists with the iti-
nerant electron minority? The answer: nearest-neighbor exchange is too weak
by one order of magnitude [6.70], so this will not help.
A plausible explanation lies in a combination of physical effects. First, the
density of itinerant d electrons is small, and the interactions are large-placing
the parameters in the ferromagnetic region of the phase diagram, Fig. 6.17. Se-
cond, the indirect exchange mechanism is no longer given by the Ruderman-
Kittel formula of (6.56) with RIJ an interatomic distance, for the local moments
are not point-like but rather spread out, occupying a large fraction of the
volume in each unit cell. The paramagnetic Curie temperature measuring the
strength of the indirect exchange and the spin wave spectrum are then better
given by the jellium model, (6.71, 72), in which both are positive, guaranteeing
ferromagnetism. The only remaining question concerns the applicability of what
is basically second-order perturbation theory to the interactions between itin-
erant and localized d-electrons, of equal or greater strength than the relevant
energy denominators. The only practical procedure is to calculate the Hartree-
Fock energy first, as deduced from the band structure, Fig. 6.20, and add
thereto the interaction energies as computed from the i 's, as in the 2-band case
analyzed above. This total energy is then to be compared to the energy of
varying states of polarization, including the antiferromagnetic ordered magnetic
phases and the nonmagnetic state. The qualitative arguments given here indicate
that a proper calculation will favor the ferromagnetic state of iron. The magnon
spectrum, such as we calculate next, should confirm this.

6.15 Magnons in Metals

Once Coulomb and exchange forces have created a magnetic solid, they pro-
foundly affect the elementary excitations. To the quasi-particles and collective
modes which are common to all metals, one must add magnons for the magnetic
metals. These are the quantized spinwaves which are the property of an ordered
magnetic medium. This section is devoted to their properties, especially in
ferromagnetic metals.
With the exception of the strong-coupled one-band model, our knowledge
of the itinerant ground state is imperf~ct even for ferromagnets. We cannot
simply proceed by computing an excited state, subtracting the ground state
 6.15 Magnons in Metals 267

energy therefrom, as even small errors of O(N) will obscure effects 0(1). A more
direct method is required, that of equation of motion. In this procedure, the
scattering of a certain type excitation is treated exactly but other terms are
neglected ("random phase approximation"). The neglected terms vanish at low
density, and the procedure gives satisfactory answers at long wavelengths, there-
fore one can accept the results with the same confidence (or skepticism!) as for
magnons in 3D Heisenberg antiferromagnets-as a semiquantitative, systematic
approximation.
For simplicity, we treat the case of a single band in some detail, including
the cases-so far ignored-of partial magnetization. We then indicate the
extensions of the theory to multi-band cases, and to questions of stability of the
assumed ferromagnetic ground state, such as we already examined in connection
with the indirect exchange mechanism.
So, let U and p be in the range shown in Fig. 6.17 for which the ground state
is ferromagnetic. In the Heisenberg model, translational invariance was sufficient
to construct the one-magnon states uniquely. Now, it is no longer enough. The
degrees of freedom in a metal are sufficiently numerous, that if we only specify
that the spin angular momentum must decrease by one unit and the total mo-
mentum wave vector must be q, there will be a very large number of excita-
tions to satisfy this requirement. We therefore add to these requirements that of
a very long or infinite lifetime, and find then that the magnon is a bound state,
a linear combination of all the states in which a hole is created in the majority-
spin band (taken to be spin "down" henceforth) while a particle, with extra
momentum liq, is added to the minority spin ("up") sub-band.
The quasi-particle energies are,

Em(k) = E(k) + Nl 2: l(l{k')f-m(k')

i'

== E(k) + Am(k) (6.129)

where the Fermi functionfm(k) is 1 if Em(k) S Em(k Fm ) and zero otherwise, as

discussed previously. Thus the elementary unit of magnetic excitation must be
at+fta,tl IF), where IF) is the ground state, and aim destroys a particle (or
creates a hole) in spin sub-band m, and atm creates one. The one-magnon state
must be Iq) == D f IF), where D f is the operator

(6.130)

The energy of each unit must be,

(6.131)
268 6. Magnetism in Metals

in the effective mass approximation. Each scatters into the others with a strength
proportional to [(kk') and to the occupation-number factor. We determine this
from the equations of motion; that is, if Of is a raising operator for :J1t" by an
energy liw" then

[:J1t",O,] = liwp, . (6.132)

The kinetic energy contribution to this commutator yields

(6.133)

whereas the scattering part yields

(6.134)

omitting terms such as at,+rmaj;lmat+,-,'takt which are smaller by some power

of the density p and are neglected in the random-phase approximation. We have
identified at",akm = nkm with its average, fm(k) , the Fermi function.
Equating terms on both sides of the equation of motion (6.132) we obtain the
equations of the amplitudes Fk

(6.135)

This equation is exactly soluble at 'I = 0, by the choice Fk = const., and yields
,wo = O. This is an exact result, reflecting the rotational invariance of the magne-
tic ground state.
*"
For 'I 0 this represents a transcendental equation that must generally be
solved graphically or numerically. However, in the special case of l(kk') = l
= const, there is a very simple solution Fk = A(Ek' - liw,)-' which leads to the
secular equation for the eigenvalue

(6.136)

This yields an approximately parabolic magnon liw, = DIJ2 for 'I < IJmax; the
maximum q is not at the edge of the BZ but occurs when the denominator of
(6.136) becomes complex. This signifies the onset of the scattering regime, in
which the bound state does not exist; an attempt to excite a magnon at 'I > IJmax
will result only in a broad resonance. Rather than detail this calculation, we
examine the case of two degenerate bands, which we do calculate in detail. This
case has an interesting feature: the elementary units in (6.130) can be at+,.",t
ak,,,,1 ± a't.t"b,tak,b,l' where. a, b are the two bands in question. The (+) com-
 6.15 Magnons in Metals 269

bination yields S~t at q = 0, and thus commutes with the Hamiltonian. It is

identified as the zero energy mode which expresses the rotational invariance of
our approximation of the magnetic ground state; the (+) mode at q =1= 0 is
identical to that of a single band worked out above. These are denoted the
"acoustic magnons" by analogy with lattice vibrations, where the low-lying
spectrum are the acoustic modes. Correspondingly there exist the "optical
magnons", the (-) branch, which do not exist for a single band ferromagnet.
The optical spectrum start at 2(..1, - A,) at q = 0 in the two-band case, and in-
creases slightly at higher wave vectors. The equations (with 2m* = 1, Ii = 1,
and kF = 1) are given below, and the calculated results plotted in Figure 6.21.
[The acoustic branch is identical to the solution of the one-band case, (6.136)].
We replace the parameter A, - A, by the single parameter A, and separately
consider the cases when L1 < 1 (weak-coupling and partial magnetization), = 1
(intermediate), and> 1 (strong-coupling and saturation magnetization).
When A ~ 1, nk' = 0 and we find the following transcendental equation:

2 1
±-=-L(Q) where (6.137)
3..1 2q

2q
Q = A + q2 - liwq and (6.138a)

L(Q) = Q2
1[1"2 (Q 2 - 1) In I QQ _ 11 1+ + QJ. (6. 138b)

For A ~ 1, nil '*' 0, and we obtain a slightly more complicated equation:

± 1.[1 - (1 - A)3/2J = 2q
~ [L(Q) - (1 - A)L(Q')] where (6.139)
3 A

,_ 2q(1 - ..1)1/2
Q - A - q2 - liwq (6.140)

and (6.138) defines the other parameter and the function L( Q).
The optical magnon mode ( - ) starts at 2..1 for q = 0 and increases somewhat
before merging with the continuum. The acoustic (+ ) branch starts at liwo = 0
for q = 0 and increases approximately __ Dq2 + O(q4), with the parabolic
approximation liwq -- D q 2 improving in relative accuracy as A is increased.
Expansion of the equations leads to it formula for D

1 + (1 - ..1)3/2 - 4/5[1 - (1 - ..1)5/2]/..1

D = 1 - (1 _ ..1)3/2 for A ~ 1 (6.141a)

and
270 6. Magnetism in Metals

o
(e)
Fig. 6.21. Acoustic (A) and optical
(0) magnons in the band theory, for
various values of LI, the "Stoner gap
parameter. Continuum indicated
(shading) is for elementary excitations
with spin flip; continuum for ele-
I • q
1.0 1.5 mentary excitations without spin flip
remains the same as shown in Fig. 6.5

4
D= 1 - - for A ~ 1. (6.141b)
5A

The acoustic branch enters the continuum and "dies" at qmax = 0.75A for A ~ 1.
Antiferromagnetism occurs when spin-down electrons start to fill the Brillouin
zone, as may be seen in the following demonstration for strong coupling. Assume
that A ~ 1, and every state in the spin-down zone is filled, while every state in the
spin-up zone is empty. The eigenvalue equation can be expanded in power of
E(k + q) - E(k) == w(k, q), a procedure certainly valid for small q. We make
use of the assumption E( -k) = E(k) to prove,

p = 0,1,2, ... , all q, (6.l42)

which together with the identity LW2p ~ 0 readily establishes the desired result
 6.15 Magnons in Metals 271

liwo =0~ liw, all q =1= 0. (6.143)

Not only does a collective magnon mode exist at every q in the Brillouin zone,
but the q = 0 mode is a maximum, and the ferromagnetic state must be unstable
against the emission of any number and any type of magnons. The new ground
state must then be an antiferromagnetic, or a spiral spin configuration of the type
previously discussed.
The antiferromagnetic behavior sets in even before the spin-down Brillouin
zone is completely filled by electrons, in the neighborhood of a half-filled zone,
although the precise point at which it occurs must be calculated numerically. It
is akin to the anti ferromagnetism of insulators. Whereas the exclusion principle
has the effect in the ferromagnetic state of preventing the kinetic motion of
electrons from atom to atom, such motion is not prohibited in the antiferro-
magnetic configurations. The difference with insulators, is that the "hopping"
(band, or kinetic, energy) represents real transitions in the metallic state, but
only virtual transitions in the insulator.
Connection between the band theory and the indirect exchange theory is very
natural and easy to establish. For when one band is supposed narrow, the
Hund's rule splitting of Kramers' degeneracy usually will send the entire spin-up
band above the Fermi level, whereas the entire spin-down band will remain
below this energy. This explains why narrow bands most likely lead to integral
numbers of Bohr magnetons,jusi as in a localized electron theory. If this narrow
band interacts with a broader band, the broader band will be only slightly
polarized. Whenever perturbation theory is applicable, something like the
Ruderman-Kittel formula must result therefrom. If perturbation theory is not
applicable, then the formulas derived in this section and the previous one
provide a fair initial approximation to a strong-coupling theory, and should be
used instead. Instead of two degenerate bands, one should consider the two
dissimilar bands-one very narrow and the other delocalized, and the resultant
magnon spectrum. The magnons are constituted out of the pair excitations,

(6.144)

with r, r' labeling the respective bands; there are thus 4 such operators for each
k, q; but the linear combinations are no longer the obvious ± ones, due to lack
of degeneracy. Nevertheless, the coefficients can be readily obtained and the
eigenvalue problem solved if the relevant scattering amplitudes (r,,, are con-
stants, independent of k, k'. Such a calculation, with the correct band structure
such as in Fig. 6.20, is required for a semi-quantitative comparison of theory
with such experiments as have been performed on the iron series metals. This is
one of the presently unfinished tasks of the type being addressed by Hubbard
[6.55] and others. We now turn to questions ofthe stability of magnetic moments
in metals.
272 6. Magnetism in Metals

6.16 Marginal Magnetism of Impurities

A magnetic impurity atom does not necessarily retain its spin or any finite
fraction thereof when it is immersed in a nonmagnetic metal, although the
tendency to do so is greatest when the magnetic orbitals are isolated from the
host metal by virtue of small size (the case for f-shell magnetism) or disparity in
energies (lack of resonance with the conduction band). Consequently, there have
been studies to determine the properties of marginally magnetic solute atoms,
to establish the key physical parameters involved in the preservation of a
moment. This work, initiated by Friedel, Anderson, Wolff and taken up by
many others 13 has been evolving over two decades. The question to be answered
concerns the magnitude of the overlap between the magnetic shell and the
nonmagnetic conduction band, which so reduces the effective interaction t(U, p)
that the spin disappears.
This question is somewhat confused by the Kondo effect, for if a moment
were to appear, it would so couple to the conduction band electrons that the
resultant spin might remain zero. So it is not clear under what circumstances
one expects a discontinuity in the ground state parameters with a change in the
interaction parameters, and under what circumstances one does not. In this
section, we shall indicate some of the methods for calculating the criteria for the
appearance of a localized moment. The Kondo effect, principally interesting
because of the finite T effects, will be treated in the appropriate chapter of
the second volume.
The method we present below is illustrative of the simplifications that can be
made in the study of an impurity in a 3D solid. First, it is only required to study
s waves, for the higher angular momenta have little if any overlap with a point
impurity at the origin. Second, the conduction band can be treated in the leading
approximation, in which the correct density of states is preserved near the band
edges. This is achieved by assuming that the matrix element connecting one
quasi-spherical shell about the origin to the next distant shell is a constant,
- Wj4. For example, if the central atom is a regular host atom, the Hamiltonian
matrix for the s states is merely

~o=
-!W 0 -!W . ·0 (6.145)
1 1
o -4W 0 -4W

ol
13 The topic is reviewed from many points of view in Magnetism, see [6.41]. An earlier, still
valuable review was given by Heeger in Solid State Physics, see [6.38].
 6.16 Marginal Magnetism of Impurities 273

taking the center-of-gravity of the band to be at zero. It is, in fact, possible to

prove that ~o can always be put into tri-diagonal form such as shown above
for arbitrary band structure [6.71], but the matrix element connecting the central
n = 0 site to the first n = I shell is not necessarily equal to that which connects
the n = I to the n = 2, etc. However, for any band structure in 3D, the matrix
elements become - W/4 asymptotically. It is a great simplification to consider
them constant down to n = 0, so let us examine this case in detail. By inspection,
the eigenvectors of(6.145) have components Un = (2/N)1/2 sin ken + I), i.e.,

2)1/2..
(N .
II = (SIn k, SIn 2k, ... , SIn Nk) (6.146)

with n = 0 labeling the origin, and the sample assumed to be spherical in shape,
in shape, of radius Na. Thus, the k's are determined by the vanishing of
sin(N + I)k, and are

k = p1t/(N + I), p = 1,2, ... (6.147)

and the energies are,

(6.148)

with the bandwidth W, as desired. The local density of states, as measured at the
origin, is

N(E) = 1~ sin k J(E + ~ Wcos k)

2 (6. 149a)

= (2/1t) I dk sin 2k o( E + ~ W cos k) (6. 149b)

!
= (8/1tW2)( W 2 - E 2f'2. (6. 149c)

It is semicircular in shape, and has the proper square root dependence on the
energy near the band edges, characteristic of all energy bands in 3D; the ap-
proximation of constant matrix element sacrifices only the van Hove singu-
larities.

Problem 6.5. Derive the density of states (6.149c) from (6.149a), filling in the
indicated steps and integration.

It requires only a slight modification to incorporate the changes due to an

impurity at the origin. The n = 0 to 1 matrix element can be written -2W/4,
with 2 generally { 1 for a compact magnetic shell; this feature (small 2) is the
essence of Anderson's model magnetic impurity [6.72], whereas the Wolff model
274 6. Magnetism in Metals

[6.73, 74] takes A. = 1; both also include a two-body potential at the n = 0 site.
In the Hartree-Fock approximation, we can even take this into account by an
effective potential 8 m == V + U(no.- m ) acting only on the central site, where
V = one body potential (localizing the orbital with respect to the c.o.g. of the
conduction band), U the parameter characterizing the two-body forces, and
(no. m) , with m = t or J, (or ± 1/2), the occupation of the impurity orbital,
averaged in the ground state. This is to be determined self-consistently as we
shall see. The Hamiltonian now has the matrix representation

8 -A.41 W ........ .

o (6.150)

o
If, as a further idealization we assume a half-filled band, the energy 8 at the
impurity site has the effect of lifting an existing symmetry between holes and
electrons and "charging up" the impurity orbital relative to the host atoms. A
compensating charge must thus exist in the conduction band of the metal, in the
vicinity of the impurity. Similarly, if the impurity acquires a net spin magnetic
moment by the mechanism we shall shortly investigate, then a compensating
spin polarization of the conduction electrons in the neighborhood of the im-
purity should cancel it, resulting in a net singlet state. The calculation of these
compensations requires many-body techniques, whereas (6.150) does not.
The eigenstates of(6.150) can be specified by a phase shift for n ~ I but that
leaves the amplitude at n = 0 undetermined. Specifically, the equations for
n ~ 2 are solved by taking the eigenvalue to be -i Wcos k as before, together
with

n ~ 1 (6.151)

and Uo = (2/N)1/2Ak' with Ak and (}k to be determined by the equations at

n = 0 and 1. At N + 1, the requirement sin(kN + (}k) = 0 yields the slightly
shifted k's

(6.152)

referring to the evenly spaced k's previously found in (6.147) as k o• and defining
the conventional phase shift Ok = (}k - k.
The eigenvalue equation at n = 1 yields
 6.16 Marginal Magnetism of impurities 275

-),,! WA k - ! Wsin(2k + fh) = - ~ Wcosksin(k + Ok)

which is solved by inspection by setting

(6.153)

The equation at n = 0 is

(6.154)

These two equations in the two unknowns A and 0 are to be solved, with the
resulting 0 then used in (6.152) to yield the new k's.
It is more convenient to use eik == z than k as the independent variable, and
e than 0 as the dependent variable, for (6.153, 154) are simple binomials in the
i8

exponentiated functions. Thus, complex variables impose themselves in this

problem.

e2i8 (z) = 48Z + W(Z2 + I) - ),,2W

or (6.155)
48Z + W(Z2 + 1) - ),,2WZ2

_
Ok - tan
-1 [
48
),,2 W sin k
+ W(2 _ ),,2)COS k
J and

)"Wsin k
(6.156)
Ak = [(48 + W(2 - ),,2)coskY + (),,2Wsink)2],/2·.

There can be no bound state as long as [48 + W(2 - ),,2)COS k]does not vanish,
i.e., as long as the inequality

(6.157)

is satisfied.
The quantity of greatest interest is the occupation of each spin component at
the impurity, given by

()"Wsin k)2
f dk (48 + W(2 -
kp
= (2/lt)
o ),,2)cosk)2 + (),,2Wsmk)2
.
m
276 6. Magnetism in Metals

+kP
= (2Aht)-1 f dk(l - cos 28k ) (6.158)
-kP

in which em == V + U (no,-m) and we use the fact that 8 is odd in k. For the
special case of a 1/2-filled band this integral is exactly calculated by complex
variables. With) a contour of half the unit circle, we have

(6.159)

As there are no poles within the unit circle with the exception of z = 0, this
contour can be deformed into the line integral up the imaginary axis t with the
origin excluded, and evaluated in terms of logarithms. One finds

I I 1 dy [ I + y2 ]
<nom> = -2 - 2'(1
1tl -
A,z) f0 -Y Y 2 - CmlY
.
- b- C.C. (6.160)

where

The Wolf model [6.73] corresponds to ,1= 1, Anderson's model [6.72] to

the limit ,1--+-0. Let us discuss Wolff's model first. The physics is particularly
simple. One atom differs from the others only in the energy of its atomic orbital
(V) relative to the others, and in the magnitude of the intra-atomic Coulomb
potential U (assumed zero on the rest of the lattice). The bond connecting it to
its neighbors is unexceptional, hence A = 1. In the absence of the potentials, the
occupation of the "impurity" is <nom> = 1/2 for both values of m. So now let
us define the "electron occupation number defect" i5n m :

(6.161)

put ,12 = 1 in (6.160) and obtain a somewhat simpler relation

(6.162)

Performing the integrals, one obtains two equations in the unknowns i5n t and
i5n j

(6.163)
 6.16 Marginal Magnetism of Impurities 277

with

(6.164)

defining a new parameter Jl.

Although derived under the assumption that no bound state exists, these formu-
las are now serendipitously valid for the entire range of parameters because of
analytical continuation. Thus the appearance or disappearance of bound states
turns out not to be a significant consideration in the appearance or disappear-
ance of a local moment. We have introduced the parameter Jl = (V + fU)/U
in (6.164), not to be confused with the Fermi energy which is always zero in this
section and in the next. As for the magnetic moment: nonmagnetic solutions
have (jn, = (jn j , magnetic solutions have m = (jn, - (jn, =1= o. At Jl = 0 sym-
metry dictates (jn, = - (jn j = ±m/2 for the magnetic solution; insert this
into (6.163) to obtain

(Jl = 0) x = 1t W [X-I + (I
2 (2U) - x- 2) tan-Ix] (6.165)

writing x = 2Um/W. This equation always has the t~ivial solution x = 0, but in
addition it may have a nontrivial solution if the slope of the right-hand-side
exceeds 1 at the origin. This yields

4 4U
(Jl = 0) - .- > 1 or 4U/W> 31t/4 = 2.3562 (6. 166a)
31t W

as the necessary condition for a local moment to form. At Jl =1= 0 one requires a
larger ratio 4U/Wfor the moment to persist. Expanding (6.163) to third order
in (jn m yields a parabolic relation valid for small Jl

4 1 + -5 -Jl
(IJlI ~ 1) 4U
-W > -31t [ 6 (31t ) 2 + ... ] (6. 166b)
8 .

Numerically, one may solve for the simultaneous solutions of (6.163) or one
may converge to them by relaxation: insert an initial value for (jn j , calculate
(jn, by (6.163), reinsert the result into (6.163) to obtain a new (jn j , etc. This pro-
cedure is found to always converge to the magnetic solution whenever it exists.
When there is no magnetic solution, it quickly converges to the nonmagnetic
solution (jn, = (jn j • This type of behavior characterizes the magnetic solution
as a "stable fixed point" of the coupled nonlinear equations, and the trivial
solution as an "unstable fixed point". An independent confirmation can be
found in a calculation of the energy associated with each solution. Where a
278 6. Magnetism in Metals

magnetic solution exists, the energy is a minimum at the proper value of m

(and -m), and a local maximum at m = O.
The results of numerical calculation of (6.163) can then be put in graphical
form, as shown in Figure 6.22. In the figure we also indicate the regions where
bound states form: above the band for /1 > 0 and below it for /1 < O. Such
bound states give the magnetic impurity the aspect of an isolated one-electron
atom. But there is ample parameter space for a magnetic solution without any
bound state detaching itself from the continuum, and this is of course the new
feature.

Fig. 6.22. Parameter space in the Wolff model. Mag-

netic solutions occupy the approximately semicircular
region. Shading indicates regions free of one-body
bound states. Above this, one level will detach itself
and lie above! W, the maximum permissible energy in
the continuum; below this, it will lie below -!W, the
lowest energy of the continuum

Anderson's model is characterized by two parameters, one of which is the

same /1 and the other is denoted L1

L1 == ~ W).?/U. (6.167)

We solve it first, then provide explanations. In the limit ).? --+ 0, the amplitudes
(6.156) are seen to yield a Lorentzian probability centered at the impurity level
e; only that part lying below the Fermi level will contribute to the occupation of
the impurity orbital. The result:

~ m
un =.17t t an -1 (/1 - L1t5n_ m ) . (6.168)

We calculate some of the physical properties from the numerical solution of these
coupled equations, much as we did for (6.163). Some of the quantities of interest
are: the net electron occupation-number defect t5n = t5n, + t5n j in the magnetic
 6.16 Marginal Magnetism ofImpurities 279

and nonmagnetic states, shown in Fig. 6.23; the magnetization as a function of

J.l for several values of L1 shown in Figure 6.24; and noting that m is a maximum
at J.l = 0, the dependence of m on L1 at J.l = 0 in Fig. 6.25. The symmetry of the
solutions with respect to J.l is maintained throughout, of course, as it was in the
Wolff model previously. J.l = 0 is called "symmetric Anderson model".
The electron probability distribution at the impurity site, the Lorentzian
distribution characterizing Anderson's model, is shown in Fig. 6.26. The asym-
metric magnetic configuration is degenerate: one can interchange spins up and

1.0 b
Ii = 0.01 Fig. 6.23. Electron deficit (or
charge accumulation) on = on! +
/"
onj on the impurity's orbitals as a
/""""" function of the net effective poten-
tial /l defined in the text. The mag-
/~",,//
c: 0.5
00 netic solution (curve a) is stable
Q until approximately /l = !; at
,"/ larger /l the nonmagnetic solutions
/ yield curve b. The nonmagnetic

0 " 025 II
solution at smaller potentials,
curve c, is unstable. The magnetic
solution shows a great stability
against charge accumulation com-
pared to the (unstable) nonmag-
netic solution. This rather typical
curve was calculated for Lf = 0.01.
For /l-+ -/l, on -+ - on

1.0r-----..:1i:..;;.,=0:;;,;.0;.;1~

III

-=- 115
E

At = lI.l1'=0318

0.5 o 025
Ii

Fig. 6.24. Magnetization (in Bohr Fig. 6.25. The curve m(Lf) at the
magnetons as function of effec- most favorable potential, /l = 0, i.e.
tive potential parameter /l defined V +t
u = 0, U> 0, the so-called
in the text. For /l ~ ! there is no "symmetric Anderson model".
magnetic solution, at smaller values
the curves m(/l) depend on the kinet-
ic energy parameter Lf, the general
tendency being shown by the three
curves for increasing Lf
280 6. Magnetism in Metals

down to obtain a distinct, yet degenerate state; a magnetic field applied to the
impurity site only, would lift this degeneracy by an amount linear in the applied
field. Thus, we know that a magnetic moment is associated with the state.

E Ikl Elkl
lw
b
2

• • and +

Fig. 6.26a,b. (Amplitude)2 of waves at impurity in cases (a) moment exists and (b) no magnetic
moment for the same Anderson model impurity. The approximately Lorentzian shaped curves
<
apply for the case A I. Magnetic case (a) is degenerate: a solution with i and! inter-
changed is equally valid. Note: V < 0, V + U > 0 are assumed

The original physical basis for Anderson's model was the following: a magne-
tic d-orbital (shell index n = 0 in our nomenclature) is weakly connected (!AW)
to an s-orbital (shell index n = I) which then connects to the rest of the lattice
in the usual way. The 2-body forces and the "weak link" characterize the magne-
tic shell. After the exact Hartree-Fock solution given here, we seek some
insight into 2-body correlations in the following Section.

6.17 Correlations and Equivalence to s-d Model

By solving the Hartree-Fock equations, we determined the range of parameters

for which an impurity atom, imbedded in a nonmagnetic metal, acquires a
localized moment. Now we consider the correlations required to re-establish the
rotational invariance of the model, and to understand the concomitant polariza-
tion of the conduction electrons near the localized moment. For the symmetry-
breaking Hartree-Fock solutions, which offer us a choice of two ground states
(Fig. 6.26a with majority spins "up" or "down"), are obviously just the first step
toward the correct singlet ground state.
De Gennes [6.75], and Schriejjer and Woiff[6.76] have established the cor-
respondence between the magnetic solution of Anderson's model and the sod
exchange model, have shown that the effective exchange parameter is AF in sign
and determined its magnitude from the microscopic parameters such as A, W,
 6.17 Correlations and Equivalence to s-d Model 281

U, etc. Their principal motivation: the s-d model being explicitly rotationally
invariant, no further demonstrations are required. Moreover, the Kondo effect
and other thermodynamic and dynamic phenomena are fairly well understood
for the s-d model but have not yet been thoroughly studied for the models of the
preceding section for which one must go to much higher orders in the perturba-
tion theory. In fact, a direct calculation of the Kondo effect starting with
Anderson's model requires an accuracy better than O(A,6) and is reasonably
complicated [6.77].
For maximum simplicity, we treat only the symmetrical model f.J, = 0,
(V + jU) = 0, but the reader can proceed to the more general case with no
essential modifications. We assume U is large so that m -- 1, and follow the
Schrieffer-Wolff type analysis. We then show how to extract the effective ex-
change parameter, not just in the Anderson model, but for arbitrary A, including
the Wolff limit A, = 1.
To start, suppose there is precisely one electron of spin "up" on the impurity
site (the zeroth shell, denoted r = 0) with the rest of the electrons in the normal
Fermi sea on the shells r = 1, 2, ... Then the matrix element for the scattering
of an electron k,J.., of spin "down" from near the Fermi level, into the localized
r = 0 shell, is A, W(2/ N)l/2, at the cost of an additional excitation energy j U. In
second order, this process restores the electron to the conduction band at k'.
Thus, a scattering from k,J.. to k',J.. has occurred, with amplitude -(A,W)2(2/N)/
(U/2). Conduction electrons of spin "up" do not benefit from this scattering
mechanism, which is forbidden to them by the exclusion principle. Thus, we
have found a spin-dependent scattering mechanism that can be written as the
sum of a potential and a spin term

(6.169)

with 2/N instead of the usual 1/N to take care of the different amplitudes in our
shell formalism, and the k, k' restricted to the neighborhood of the· Fermi
k p = j1t. Sz is a counter: + 1/2 for an electron with spin "up" at r = 0, and
-1/2 for spin "down". For no kt scattering to occur, we require

and for the k,J.. scattering to yield the correct result, we need

This leads to the following value for the exchange constant and effective potential

and VeCf = -4WA . (6.170)

282 6. Magnetism in Metals

A similar reasoning yields the spin-flip scattering terms. In the intermediate

state, 2 electrons of opposite spin occupy the r = 0 shell. If instead of the spin
"down" electron returning to the Fermi sea, the spin "up" electron did so, we
would have a scattering from kJ, to k't. This is described by

with S- indicating that the counter has gone from + 1/2 to -1/2, and the
negative sign from the Pauli principle (the ordering of two electrons has been
interchanged). The magnitude of J 1. is again given by second-order perturbation
theory: -16 WLf, and is identical to Jz. The inverse process provides us with the
Hermitean conjugate of the above term, and completes our effective s-d exchange
Hamiltonian

(6.171)

where Ie and n are the local spin-density and particle-density operators, re-
stricted to the neighborhood of kF (more or less) and carrying the additional
factor 2 if characterized by the 1D set of k's.
For a more quantitative study at finite A" we can start by the Hartree-Fock
solution of the preceding section and then add in higher-order terms if desired.
The proper of an AF exchange parameter is that the conduction sea close to the
local moment is polarized antiparallel to it. Then the Friedel (i.e., Ruderman-
Kittel) oscillations periodically reverse the polarization as one proceeds further.
We shall verify these features explicitly.
At the rth shell, r = 1,2, ... the magnetization m(r) is

(6.172)

With z = exp(ik) and lJ(z) given in (6.155) we follow the procedure of (6.159)
et seq. to evaluate the above as the following integral:

A,2( -1)'
mer) = 21ti(I _ A,2) I dyy y2r(l + y2)
I

X HCr - c,iy - b)-I - (y2 - c,iy - b)-I] - c.c.} (6.173)

with Cm, b defined in (6.160). The main difference with the r = 0 calculation is
the newly appeared factor A,2 and the (-yZ)'. The results, plotted in Fig.
6.27, show that the first shell is polarized anti parallel to the r = 0 impurity shell,
therefore imply the effective AF coupling. The change of sign at each successive
shell reflects the half-filled band, kF = 1t/2. To obtain Jerr by this procedure,
 6.17 Correlations and Equivalence to sod Model 283

1.0 Fig. 6.27. Schematic Hartree-Fock

result: magnetization m(r) of the r'h
shell relative to magnetic shell at r = 0,
according to (6.173)

-0.5

evaluate the spin oscillations that would be caused by an equivalent sod inter-
action [restricted to JzSZs~(O) for fair comparison with the Hartree-Fock]. The
latter is also a point-scatterer which is positive for spins up and negative for
spins down, and can be analyzed by the same set of equations above. Thus, we
re-evaluate mer) which is once more given in (6.173), with however a different set
of parameters: A = 1 and Ct = 0, C j = -V.. corresponding to the appropriate
exchange and potential scattering. Denote the result, m.ir). The best fit to the
equation

(6.174)

will yield the value of J z (alias J err ) in terms of the more "microscopic" para-
meters V, U, A, W. Such is the object of active, on-going research by some of the
most sophisticated techniques developed in the many-body problem [6.77]. The
goal is to explain the figure and Table on p. 36, starting from first principles. Is
it possible that we physicists are now struggling more and more to obtain less
and less? Only History will judge, while our present task-the study of thermal
properties continues in Volume 2.
Bibliography

Taken as a whole, the hundreds of references annotating the textual material

in this book aready constitute a broad bibliography. We have consistently re-
ferred to review articles and compendia wherever possible, to lighten the search
into the literature by our reader. However, this had to be balanced against a de-
sire to cite the original research for this too is important, to credit the innovators.
Among the continuing series of publications in magnetism, we cite here a
few culled from out of many:
G.T. Rado and H. Suhl have edited a series of volumes under the title Magne-
tism: a Treatise on Modern Theory and Materials (Academic New York) in
which review articles by various authors outline the main developments in the
field. A look at the latest volume of this on-going series, and at the latest pub-
lished proceedings of the bi-annual International Magnetism Conference will
quickly bring a technically minded reader up-to-date in the field. More classical
references include vols. 18/1 and 18/2 of the Handbuch der Physik, ed. by S.
Flugge (Springer, Berlin, Heidelberg, New York 1968) which are entirely dedi-'
cated to the topic of magnetism, and a textbook;.
A. Herpin: Theorie du Magnetisme (Presses Univ. de France, Paris 1968).
The microscopic details of the interactions between the magnetic ion and the
lattice in which it is imbedded are thoroughly explored in
H. J. Zeiger, G. W. Pratt: Magnetic Interactions in Solids (Clarendon, Oxford
1973) containing appendices on symmetry properties and group theory.
Many books and articles have appeared recently on the topic of phase transi-
tions, particularly those of magnetic systems. A selected bibliography for this
rapidly evolving field will be included in the second volume of this book. For the
present, we include only readings which are designed to supplement or illuminate
the list of references already given in the first chapter of the present volume,
devoted to the history of the theory of magnetism. It is a topic not covered
often or well in the usual scientific literature, and many of the references we
have cited will be, perforce, unfamiliar to the reader. We therefore follow with
brief descriptions and annotations on some of the more interesting books we have
read, with suggestions for further reading.
J. B. Porta: Natural Magick, Naples, 1589, 1st English Edition 1658, with the
second edition of 1669 being "even more scarce than the first, whose coiled and
damaged examples testify to the fact that it was used and worn out by the prac-
286 Bibliography

tical man and in the laboratory." From reprint of 1st ed., Basic Books, Inc.,
New York, 1957 preface. It is a colorful treatise on "the causes of wonderful
things" such as "counterfeiting gold", the "wonders of the loadstone", "beauti-
fying women", "cookery", etc.
W. Gilbert: On the Magnet (in Latin De Magnete), London, 1600 (transl. for
tercentenary of Gilbert club, London, of which Lord Kelvin was president).
This 1900 edition edition was a private one limited to 250 copies, but a faithful,
indeed luxurious reproduction was made available by Basic Books, Inc., New
York, 1958. The reader should shun an inferior translation by P. F Mottelay,
1893, a bibliographer whose main contribution was the thorough but uncritical
Bibliographical History of Electricity and Magnetism, London, 1922, covering
the period up to and including Faraday.
E. Whittaker: A History of the Theories of Aether and Electricity (Harper and
Row, New York 1960) in 2 volumes, provides a most complete scientific and
bibliographical history of electricity and magnetism from earliest times up to and
including the advent of quantum mechanics. Free use of equations is delightful
for the scientific reader but renders this work inaccessible to the layman, who
may turn to
F. Cajori: A History of Physics, (Dover, New York 1962) or the interesting and
diversified Histoire de la Science, Maurice Daumas, Ed., Encyclop. de la Pleiade,
Paris, 1957.
J. C. Maxwell: A Treatise on Electricity and Magnetism, the famous and com-
plete account of the theory (before the discovery of the electron). Published in
1874, it is available as a reprint by Dover, New York, 1954 as are Maxwell's
collected works. These should be read, for if we are to follow Maxwell's own
expressed advice: it "is of great advantage to the student of any subject to read
in the original ... for science is always most completely assimilated when it is
found in its nascent state. Every student of science should, in fact, be an antiqu-
ary in his subject." Some polishing and pruning transformed the voluminous
Treatise into J. H. Jeans' The Mathematical Theory of Electricity and Magne-
tism, Cambridge Univ. Press, Cambridge, 1907. One follows subsequent history
with the various editions of "Jeans": the second, revised to include "motion of
the electrons", 1911, the fourth, revised to include the theory ofrelativity, 1919,
and finally the fifth, of 1925, incorporating the quantum theory of Bohr.
J. A. Ewing: Magnetic Induction in Iron and other Metals (Electrician Publ. Co,
London 1900) covers the more practical, technical aspects of magnetism. It was
Ewing who, for example, coined the word "hysteresis". The modern successor to
this work is
R. M. Bozorth's Ferromagnetism (Van Nostrand, Princeton 1951) one thousand
pages of collected facts and figures about magnetic materials and including a
chronological bibliography covering the period 1850-1950.
The influence of modern quantum theory is first seen in the proceedings of
the 6th Solvay Congress, Brussels, October 1930 on the topic of Magnetism,
 Bibliography 287

Gauthier-Villars, Paris, 1932. The contributions of W. Pauli: "L'Electron Mag-

netique", and of P. Weiss and W. Heisenberg (all in French) are a milestone. The
next milepost is
J. H. Van Vleck: A Survey of the Theory of Ferromagnetism, Rev. Mod. Phys.
17, 27 (1945). A post-war explosion in numbers of journals and publications
made this simple review the last one capable of being written by one person
within a reasonably small number of pages. The student of recent history of
magnetism must now turn to conference proceedings or compendia.
Finally:
J. C. Slater: Quantum Theory of Molecules and Solids, Vol. 1. McGraw-Hill,
New York 1963). Critical analysis and thorough textbook for exchange within
the molecular theory. An earlier, related work is
P. o. Lowdin: Quantum Theory of Cohesive Properties of Solids, Adv. Phys.
S, 1 (1956) discussed the matrix analysis of the nonorthogonality problem.
Zeiger and Pratt, op. cit. afford a detailed examination of the exchange processes
as does
J. Goodenough: Magnetism and the Chemical Bond (Interscience Publ. Co., New
York 1963).
References

Chapter 1
1.1 Lucretius Carus: De Rerum Natura, 1st century B.C. References are to vv. 906 If., in the
translation by Th. Creech, London, 1714
1.2 Pliny, quoted in W. Gilbert: De Magnete, trans., Gilbert Club, London, 1900, rev. ed.,
Basic Books, New York, 1958, p. 8
1.3 John Baptista Porta: Natural Magick (Naples, 1589), reprint of 1st English ed. (Basic
Books, New York 1957) p. 212
1.4 P. A. Schilp (ed.): Albert Einstein: Philosopher-Scientist, Vol. I (Harper & Row, New
York 1959) p. 9 .
1.5 W. Gilbert: De Magnete, trans., Gilbert Club, London 1900, rev. ed. (Basic Books, New
York 1958) p. 3
1.6 J. Dryden: From "epistle to Doctor Walter Charleton, physician in ordinary to King
Charles I"
1.7 Nathaniel Carpenter, Dean of Ireland, quoted in P. F. Mottelay: A Bibliographical
History of Electricity and Magnetism (London 1922) p. 107
1.8 Spoken by Sagredus in Galileo's: Dialogo sopra i due massimi sistemi del mondo Tolemaico
e Copernicano, 1632
1.9 Opera di Galileo Galilei, Ed. Nazionale, Firenze, 1890-1909, Vol. V, p. 232
1.10 Marin Mersenne: "Cogitata physico-mathematica" (1644), the part entitled "Tractatus de
Magnetis proprietatibus"
1.11 J. Keill: lntroductio ad Veram Physicam, 1705 (transl., 1776)
1.12 Although this work has never been translated from the Latin, a good account of it is
given by Pere Rene Just Haiiy in his "Exposition raisonnee de la tMorie de l'electricite et
du magnetisme", 1787.
1.13 Almost all of Coulomb's memoirs are collected in a single work published by the Societe
francaise de physique in 1884, Collection des Memoires Relatifs a la Physique, Vol. 1
1.14 Memoire lu a /'Institute Ie 26 prairial, an 7, par Ie citoyen Coulomb, Memoires de I'Institut,
Vol. III, p. 176
1.15 S.D. Poisson: Memoire sur la Theorie du Magnetisme, Memoires de I'Academie, Vol. V, p.
247
1.16 F. Cajori: A History of Physics (Dover, New York 1962) p. 102
1.17 J. B. Priestley: History of Electricity (London, 1775) p. 86
1.18 Collection des Memoires Relatifs a la Physique, Soc. Franc. de Phys., 1884, Vol. II, pp.
141, 144
1.19 A. Einstein: The Method of Theoretical Physics (Oxford, 1933)
1.20 J. C. Maxwell: A Treatise on Electricity and Magnetism, 1873 (reprinted by Dover, New
York 1954)
1.21 H. Adams: The Education of Henry Adams (Random House, New York 1931)
1.22 G. Johnstone Stoney: Trans. Roy. Dub. Soc. 4, 583 (1891)
1.23 For a more complete account of the birth of the electron, see D. L. Anderson: The Dis-
covery of the Electron (Van Nostrand, Princeton, N. J. 1964)
 References 289

1.24 "Mathematical and Physical Papers of W. Thomson, Lord Kelvin", Vol. V (Cambridge,
1911)
1.25 A. Sommerfeld: Electrodynamics (Academic, New York 1952) p. 236
1.26 P. Curie: Ann. Chim. Phys. 5 (7), 289 (1895), and Oeuvres (Paris 1908)
1.27 P. Langevin: Ann. Chim. Phys. 5 (8),70 (1905), and J. Phys. 4 (4),678 (1905)
1.28 P. Weiss: J. de Phys. 6 (4), 661 (1907)
1.29 J. H. Van Vleck: The Theory of Electric and Magnetic Susceptibilities (Oxford 1932) p. 104
1.30 H. Goldstein: Classical Mechanics (Addison-Wesley, Reading, Mass. 1951)
1.31 W. Gerlach, o. Stern: Z. Phys. 9, 349 (1922)
1.32 A. H. Compton: J. Franklin Inst. 192, 144 (1921)
1.33 G. Uhlenbeck, S. Goudsmit: Naturwiss.13, 953 (1925)
1.34 See Compt. Rend., September, 1923, and also De Broglie's thesis. For this work, he was
awarded the 1929 Nobel prize in physics
1.35 W. Heisenberg, M. Born, P. Jordan: Z. Phys. 35, 557 (1926)
1.36 M. Born, N. Wiener: J. Math. Phys. (M.l.T.) 5,84 (1926)
1.37 E. SchrOdinger: Ann. Phys. 79 (4), 734 (1926)
C. Eckart: Phys. Rev. 28, 711 (1926)
In these references, and in much of the narrative in the text, we follow a truly fascinating
account given by
Sir E. Whittaker: A History of the Theories of Aether and Electricity, Vol. II (Harper &
Row, New York 1960)
1.38 P. Jordan, E. Wigner: Z. Phys. 47, 631 (1928)
1.39 P.A.M. Dirac: Proc. Roy. Soc. 117A, 610 (1928)
1.40 P.A.M. Dirac: Proc. Roy. Soc. 126A, 360 (1930)
1.41 J.e. Slater: Phys. Rev. 34, 1293 (1929)
1.42 Quoted by J. H. Van Vleck in a talk, "American Physics Becomes of Age", Phys. Today
17, 21 (1964)
1.43 We thank Ch. Perelman for this insightful observation.
1.44 E. Ising: Z. Phys. 31, 253 (1925)
1.45 E. C. Stoner: Magnetism and Matter (Methuen, London 1934) p. 100
1.46 F. Bloch: Zeit. Phys. 61, 206 (1930)
1.47 J. C. Slater: Phys. Rev. 35, 509 (1930)
1.48 W. Pauli: In Le Magnetisme, 6th Solvay Conf. (Gauthier-Villars, Paris 1932) p. 212
1.49 L. Neel: Ann. Phys. (Paris) 17, 64 (1932); J. Phys. Radium 3,160 (1932)
1.50 G. H. Wannier: Phys. Rev. 52,191 (1937)
1.51 J. e. Slater: Phys. Rev. 52, 198 (1937)
1.52 J. C. Slater: Phys. Rev. 49, 537, 931 (1936)
R. Bozorth: Bell Syst. Tech. J. 19, 1 (1940)
1.53a E. C. Stoner: Proc. Roy. Soc. 169A 339 (1939)
1.53b A. Hubert: Theorie der Domlinewlinde in geordneten Medien, Lecture Notes in Physics,
Vol 26 (Springer, Berlin, Heidelbesg, New York 1974) is a recent review. Quantum aspects
are first treated in
A. Antoulas, R. Schilling, W. Baltensperger: Solid State Commun. 18, 1435 (1976); and
R. Schilling: Phys. Rev. B15, 2700 (1977)
1.54 C. Kooy, U. Enz: Philips Res. Rept. 15,7 (1960)
1.55 A. H. Bobeck: Bell Syst. Tech. J. 46,1901 (1967)
1.56 A. H. Eschenfelder: Magnetic Bubble Technology, Springer Series in Solid State Sciences,
Vol. 14 (Springer, Berlin, Heidelberg, New York 1980)
1.57 M. Prutton: Thin Ferromagnetic Films (Butterworth, Washington 1964)
R. Soohoo: Magnetic Thin Films (Harper & Row, New York 1965)
1.58 M. Blois: J. Appl. Phys. 26, 975 (1955)
1.59 R. Sherwood et al.: J. Appl. Phys. 30, 217 (1959)
290 References

1.60 A. Thiele: Bell Syst. Tech. J. 48, 3287 (1969)

A. Thiele et al.: Bell Syst. Tech. J. 50, 711, 725 (1971)
1.61 H. Callen, R. M. Josephs: J. Appl. Phys. 42,1977 (1971)
1.62 M. Pomerantz, F. Dacol, A. Segmuller: Phys. Rev. Lett. 40, 246 (1978), Physics Today
34,20 (1981)
1.63 L. Onsager: Phys. Rev. 65, 117 (1944)
1.64 L. Onsager: Nuovo Cimento Suppl. 6, 261 (1949)
1.65 Quoted from E. Montroll, R. Potts, J. Ward: "Correlations and Spontaneous Magneti-
zation of the Two-Dimensional Ising Model" in Onsager celebration issue, of J. Math.
Phys. 4, 308 (1963)
1.66 N. Mermin, H. Wagner: Phys. Rev. Lett. 17, 1133 (1966)
1.67 H. E. Stanley, T. A. Kaplan: Phys. Rev. Lett. 17, 913 (1966)
1.68 J. Kosterlitz, D. Thouless: J. Phys. C6, 1181 (1973)
1.69 J. Friedel: Can. J. Phys. 34, 1190 (1956); J. Phys. Rad. 19, 573 (1958)
1.70 P. W. Anderson: Phys. Rev. 124,41 (1961)
1.71 J. Kondo: Prog. Theor. Phys. 28, 846 (1962); 32, 37 (1964)
1.72 K. G. Wilson: The renormalization group: Critical phenomena and the Kondo prob-
lem. Rev. Mod. Phys. 47, 773 (1975)
1.73 D. Mattis: Phys. Rev. Lett. 19, 1478 (1967)
1.74 K. Hepp: Solid State Commun. 8, 2087 (1970)
1.75 N. Andrei: Phys. Rev. Lett. 45, 379 (1980)
1.76 P. B. Wiegmann et al.: Phys. Lett. 81A, 175, 179 (1981)
1.77 J. Kogut: Revs. Mod. Phys. 51, 659 (1979)
1.78 A. Heeger: Solid State Phys. 23, 283 (1969)
1.79 A. Clogston: Phys. Rev. 125,541 (1962)

Chapter 2
2.1 P. A. M. Dirac: Proc. Roy. Soc. 112A, 661 (1926)
2.2 W. Heisenberg: Z. Phys. 38, 441( 1926)
2.3 J. Blatt, V. Weisskopf: Theoretical Nuclear Physics (Wiley, New York 1952) p. 136
2.4 P. A. M. Dirac: Proc. Roy. Soc. 123A, 714 (1929)
2.5 J. H. Van Vleck: The Theory of Electric and Magnetic Susceptibilities (Oxford 1932)
2.6 E. Lieb, D. Mattis: Unpublished work (1962)
2.7 W. Heitler, F. London: Z. Phys. 44, 455 (1927)
2.8 R. S. Mulliken: Phys. Rev. 43, 279 (1933)
2.9 J. C. Slater: Quantum Theory of Molecules and Solids (McGraw-Hill, New York 1963)
2.10 C. Herring: Rev. Mod. Phys. 34, 631 (1962)
2.11 J. C. Slater: Phys. Rev. 35, 509 (1930)
2.12 D. R. Inglis: Phys. Rev. 46,135 (1934)
2.13 T. Arai: Phys. Rev. 134, A824 (1964)
2.14 E. Harris, J. Owen: Phys. Rev. Lett. 11,9 (1963)
2.15 D. S. Rodbell, I. S. Jacobs, J. Owen, E. A. Harris Phys. Rev. Lett. 11, 10 (1963)
2.16 W. J. Carr: Phys. Rev. 92, 28 (1953)
2.17 F. Takano: J. Phys. Soc. Jpn. 14, 348 (1959)
2.18 P. O. Lowdin, et al.: J. Math. Phys. 1,461 (1960)
2.19 T. L. Gilbert: J. Math. Phys. 3, 107 (1962)
2.20 J. Calais, K. Appel: J. Math. Phys. 5, 1001 (1964)
 References 291

Chapter 3
3.1 A. R. Edmonds: Angular Momentum in Quantum Mechanics (Princeton Univ. Press,
Princeton, N. J. 1957)
3.2 E. U. Condon, G. Shortley: The Theory of Atomic Spectra (Cambridge Univ. Press, New
York 1935)
3.3 E. Goto, H. Kolm, K. Ford: Phys. Rev. 132, 387 (1963)
3.4 P. A. M. Dirac: Proc. Roy. Soc. (London) A123, 60 (1931)
3.5 J. Schwinger: "On Angular Momentum", U.S. Atomic Energy Commission Rpt. NYO-
3071 (1952) reprinted in L. Biedenharn and H. Van Dam (eds.): Quantum Theory of
Angular Momentum (Academic, New York 1965)
3.6 T. Holstein, H. Primakoff: Phys. Rev. 58, 1048 (1940)
3.7 J. Villain: J. de Phys. 35, 27 (1974)
3.8 P. Jordan, E. Wigner: Z. Phys. 47, 631 (1928)
3.9 D. Mattis, S. Nam: J. Math. Phys. 13, 1185 (1972)

Chapter 4
4.1 W. Pauli: Z. Phys. 31, 765 (1925)
4.2 G. E. Uhlenbeck, S. Goudsmit: Naturwissenschaften 13, 953 (1925)
4.3 W. Pauli: Z. Phys. 43,601 (1927)
4.4 J. C. Slater: Phys. Rev. 34, 1293 (1929)
4.5 E. Lieb, D. Mattis: Phys. Rev. 125, 164 (1962)
4.6 E. Lieb: Phys. Rev. 130,2518 (1963)
4.7 F. Hund: Linienspektren und periodisches System der Elemente (Berlin 1927)
4.8 V. Fock: Z. Phys. 61, 126 (1930)
4.9 J. C. Slater: Phys. Rev. 35, 210 (1930)
4.10 E. U. Condon, G. Shortley: The Theory of Atomic Spectra (Cambridge Univ. Press, New
York 1935) pp. 174-179
4.11 American Institute of Physics Handbook, 2nd ed. (McGraw-Hill, New York 1963) pp. 7-21
4.12 J. S. Griffith: The Theory of Transition-Metal Ions (Cambridge Univ. Press, New York
1961) H. F. Schaefer (ed.): Methods of Electronic Structure Theory (Plenum, New York
1977)
4.13 E. A. Abbott: Flatland (Grant Dahlstrom, Pasadena, California, 1884)
4.14 E. Lieb, D. Mattis: Unpublished work (1962)
4.15 R. D. Mattuck: A Guide to Feynman Diagrams in the Many-Body Problem, 2nd ed. (Mc-
Graw-Hill, New York 1976)

Chapter 5
5.1 J. Van Kranendonk, J. H. Van Vleck: Rev. Mod. Phys. 30, 1 (1958)
5.2 F. Keffer, T. Oguchi: Phys. Rev. 117. 718 (1960)
5.3 M. H. Cohen, F. Keffer: Phys. Rev. 99,1128,1135 (1955)
5.4 R. Griffiths: Phys. Rev. 176, 655 (1968)
5.5 M. Wortis: Phys. Rev. 132, 85 (1963)
5.6 N. Fukuda, M. Wortis: J. Phys. Chern. Sol. 24, 1675 (1963)
5.7 M. Wortis: Phys. Rev. 138A, 1126 (1965)
5.8 F. J. Dyson: Phys. Rev. 102, 1217, 1230 (1956)
5.9 R. Boyd, J. Callaway: Phys. Rev. 138A, 1621 (1965)
K. Hepp: Phys. Rev. B5, 95 (1972)
292 References

5.10 E. W. Montroll: "Lattice Statistics", in Applied Combinatorial Mathematics, ed. by E. F.

Beckenback (Wiley, New York 1964); Also, see M. L. Glasser, I. J. Zucker: "Extended
Watson Integrals". Proc. Natl. Acad. Sci. USA 74, 1800 (1977)
5.11 E. Lieb, D. Mattis: J. Math. Phys. 3, 749 (1962)
5.12 E. Lieb, T. Schultz, D. Mattis: Ann. Phys. (N. Y.) 16, 407 (1961) Appendix B
5.13 D. Mattis: Phys. Rev. 130,76 (1963)
5.14 T. Nagamiya, K. Yosida, R. Kubo: Adv. Phys. (Philos. Mag. Suppl.) 4 (13), 1 (1955)
M. Steiner, J. Villain, C. Windsor: Adv. Phys. (Philos. Mag. Suppl.) 25,87 (1976)
5.15 S. Katsura: Phys. Rev. 127, 1508 (1962)
5.16 T. Niemeijer: Physica 36, 377 (1967)
5.17 M. Bevis, A. Sievers, J. Harrison, D. Taylor, D. Thouless: Phys. Rev. Lett. 41, 987 (1978)
5.18 D. R. Taylor: Phys. Rev. Lett. 42, 1302 (1979)
5.19 H. Bethe: Z. Phys. 71, 205 (1931)
5.20 L. Hulthen: Ark. Met. Astron. Fysik 26A, Na. 11 (1938)
5.21 R. Orbach: Phys. Rev. 112, 309 (1958)
5.22 L. R. Walker: Phys. Rev. 116, 1289 (1959)
5.23 J. Des Cloizeaux, J. J. Pearson: Phys. Rev. 128,2131 (1962)
M. Fowler: Phys. Rev. B17 2989 (1978); J. Phys. Cll, L977 (1978)
M. Fowler, M. Puga: Phys. Rev. B18, 421 (1978)
A. Ovchinnikov: Sov. Phys. JETP 29, 727 (1969)
5.24 R. B. Griffiths: Phys. Rev. 133, A768 (1964)
5.25 C. N. Yang, C. P. Yang: Phys. Rev. 150, 321 (1966)
5.26 R. J. Baxter: Ann. Phys. (N. Y.) 70, 323 (1972)
5.27 J. D. Johnson, B. McCoy: Phys. Rev. A6, 1613 (1972)
5.28 J. D. Johnson: Phys. Rev. A9, 1743 (1974)
5.29 T. Yamada: Prog. Theor. Phys. Jpn. 41, 880 (1969)
5.30 J. C. Bonner, B. Sutherland, P. Richards: AlP Conf. Proc. 24, 335 (1975)
5.31 P. Hohenberg, W. Brinkman: Phys. Rev. BI0, 128 (1974)
5.32 M. Steiner, J. Villain, C. Windsor: Adv. Phys. 25, 87 (1976)
G. Muller, H. Beck, J. Bonner: Phys. Rev. Lett. 43, 75 (1979)
M. Puga: Phys. Rev. Lett. 42, 405 (1979)
5.33 R. Birgeneau, G. Shirane: Phys. Today, Dec. (1978) p. 32
H. Mikeska, W. Pesch: J. Phys. C12, L37 (1979).
S. Satija, G. Shirane, Y. Yoshizawa, K. Hirakawa: Phys. Rev. Lett. 44,1548 (1980)
5.34 J. Johnson, S. Krinsky, B. McCoy: Phys. Rev. A8, 2526 (1973)
A. Luther, I. Peschel: Phys. Rev. B9, 2911 (1974); B12, 3908 (1975)
H. Fogedby: J. Phys. Cll, 4767 (1978)
5.35 P. W. Anderson: Phys. Rev. 86, 694 (1952)
5.36 H. L. Davis: Phys. Rev. 120, 789 (1960)
5.37 J. Oitmaa, D. D. Betts: Can. J. Phys. 56, 897 (1978)
R. Jullien et al. Phys. Rev. Lett. 44, 1551 (1980)
5.38 T. Oguchi: J. Phys. Chern. Sol. 24,1649 (1963)
5.39 R. Bartkowski: Phys. Rev. B5, 4536 (1972)
5.40 D. Mattis: Phys. Rev. Lett. 42, 1503 (1979)
5.41 J. Oitmaa, D. Betts: Phys. Lett. 68A, 450 (1978)
5.42 W. P. Wolf: Rep. Prog. Phys. 24, 212 (1961)
5.43 T. Nakamura, M. Bloch: Phys. Rev. 132, 2528 (1963)
5.44 R. Jellito: Z. Naturforsch. 19a, 1567, 1580 (1964);
R. F. Wallis, A. A. Maradudin, I. Ipatova, A. K. Klochikhin: Solid State Cornrnun.
5, 89 (1967):
D. L. Mills, A. A. Maradudin: J. Phys. Chern. Solids 28, 1855 (1967);
L. Dobrzyinski, D. L. Mills. Phys: Rev. 186, 538 (1969);
 References 293

E. Ilisca, E. Gallais: J. de Phys. 33, 811 (1972). See also

R. De Warnes, T. Wolfram: Phys. Rev. 185, 720 (1969) and for AF: ibid, 762, and
for case of additional n.n.n. couplings
I. Harada, O. Nagai. J. Phys. Soc. 42, 738 (1977); J. Appl. Phys. 49, 2144 (1978); Phys.
Rev. B19, 3622 (1979);
J. Banavar, F. Keffer: Phys. Rev. B17, 2974 (1978)
5.45 J. Kosterlitz, D. Thouless: J. Phys. C6, ))81 (1973)
5.46 J. Kosterlitz: J. Phys. C7, 1046 (1974)
5.47 J. Scott Russell: Proc. Roy. Soc. Edinburgh (1844) p. 319
5.48 J. Tjon, J. Wright: Phys. Rev. B15, 3470 (1977)
5.49 K. Nakamura, T. Sasada: Phys. Lett. A48, 321 (1974)
5.50 M. Lakshmanan, T. Ruijgrok, C. Thompson: Physica 84A, 577 (1976)
5.51 L. A. Takhtajan: Phys. Lett. 64A, 235 (1977)
5.52 H. Fogedby: "Theoretical Aspects of Mainly Low-Dimensional Magnetic Systems", notes
from the Inst. Laue-Langevin, Grenoble, France (March 1979); J. Phys. A 13, 1467 (1980)
5.53 H. Mikeska: J. Phys. Cll, L29 (1978), C13, 2913 (1980)
K. Leung, D. Hone, D. Mills, P. Riseborough, S. Trullinger: Phys. Rev. B21, 4017 (1980)
J. Jose, P. Sahni: Phys. Rev. Lett. 43, 78 (1978), erratum 43,1843 (1978)
5.54 D. C. Mattis: The Theory of Magnetism II, Springer Ser. Solid-State Sci., Vol. 55 (Springer,
Berlin, Heidelberg 1985)

Chapter 6
6.1 J. C. Slater: Rev. Mod. Phys. 25, 199 (1953)
6.2 C. Herring: "Exchange Interactions Among Intinerant Electrons", in Magnetism, ed. by
G. Rado, H. Suhl (Academic, New York 1966)
T. Moriya: J. Magn. Magn. Mat. 14, I (1979)
6.3 J. Slater, G. Koster: Phys. Rev. 94, 1498 (1954)
6.4 E. Wohlfarth, J. Cornwell: Phys. Rev. Lett. 7, 342 (1961)
6.5 H. Ehrenreich, H. Philipp, D. Olechna: Phys. Rev. 131,2469 (1963)
J. C. Phillips: Phys. Rev. 133, A1020 (1964)
L. F. Mattheiss: Phys. Rev. 134, A970 (1964)
C. Wang, J. Callaway: Phys. Rev. B15, 298 (1977)
D. R. Penn: Phys. Rev. Lett. 42, 921 (1979)
L. Kleinman et al.: Phys. Rev. 822, ))05 (1980)
6.6 E. Fawcett: Adv. Phys. (Philos. Mag. Suppl.) 13, 139 (1964)
6.7 R. E. Peierls: Quantum Theory of Solids (Oxford Univ. Press, Oxford 1955) p. 148
6.8 M. Glasser: J. Math. Phys. 5, ))50 (1964); erratum J. Math. Phys. 7, 1340 (1966)
6.9 A. W. Saenz, R. O'Rourke: Rev. Mod. Phys. 27, 381 (1955)
6.10 J. H. Van Vleck: Nuovo Cimento 6, (Ser. X, Suppl. 3) 857 (1957)
D. Kojima, A. Isihara: Phys. Rev. 820, 489 (1979)
6.11 F. Bloch: Zeit. Phys. 57, 545 (1929)
6.12 E. P. Wigner: Trans. Faraday Soc. 205, 678 (1938)
6.13 J. C. Slater: Phys. Rev. 49, 537, 931 (1936); 52,198 (1937)
6.14 E. Lieb, D. Mattis: Phys. Rev. 125, 164 (1962)
6.15 8. T. Matthias, R. Bozorth, J. H. Van Vleck: Phys. Rev. Lett. 7, 160 (1961)
6.16 W. Kohn: Phys. Rev. 133, A171 (1964)
6.17 H. A. Kramers: Physica 1,182 (1934)
6.18 P. W. Anderson: Solid State Phys.14, 99 (1963); in Magnetism, Vol. I, ed. by G. Rado, H.
Suhl (Academic, New York 1964) Chap. 2
6.19 M. A. Ruderman, C. Kittel: Phys. Rev. 96, 99 (1954)
294 References

6.20 N. Bloembergen, T. J. Rowland: Phys. Rev. 97, 1679 (1955)

6.21 K. Yosida: Phys. Rev. 106,893 (1957)
6.22 W. Marshall, T. Cranshaw, C. Johnson, M. Ridout: Rev. Mod. Phys. 36, 399 (1964)
6.23 F. Smith: Phys. Rev. Lett. 36, 1221 (1976)
6.24 G.S. Rushbrooke: J. Math. Phys. 5, 1106 (1964)
6.25 A. Heeger, A. Klein, P. Tu: Phys. Rev. Lett. 17,803 (1966)
6.26 R. J. Elliott: Phys. Rev. 124, 346 (1961)
H. Miwa, K. Yosida: Prog. Theor. Phys. (Kyoto) 26,693 (1961)
T. A. Kaplan: Phys. Rev. 124, 329 (1961)
6.27 D. Lyons, T. Kaplan: Phys. Rev. 120, 1580 (1960); 126, 540 (1962)
6.28 E. F. Bertaut: in Magnetism, Vol. 3, ed. by G. Rado, H. Suhl (Academic, New York
1963) Chap. 4
6.29 D. Mattis: Phys. Rev. 130, 76 (1963)
6.30 H. Bader, R. Schilling: Phys. Rev. BI9, 3556 (1979); B20, 1977 (1979)
6.31 J. Reitz, M. B. Stearns: J. Appl. Phys. 50 (3),2066 (1979)
6.32 S. Methfessel, private communication.
6.33 T. A. Kaplan, D. H. Lyons: Phys. Rev. 129,2072 (1963)
6.34 J. Kondo: Prog. Theor. Phys. (Kyoto) 32, 37 (1964)
6.35 W. Meissner, G. Voigt: Ann. Physik 7,761,892 (1930)
6.36 J. P. Franck, F. D. Manchester, D. L. Martin: Proc. Roy. Soc. (London) A263, 494 (1961)
6.37 M. D. Daybell, W. Steyert: Phys. Rev. Lett. 18, 398 (1967)
6.38 A. J. Heeger: "Localized Moments and Nonmoments in Metals: the Kondo Effect", in
Solid State Physics, Vol. 23, ed. by F. Seitz, D. Turnbull, H. Ehrenreich (Academic,
New York 1969) p. 284 and Fig. 28 on p. 380
6.39 D. Cragg, P. Lloyd: J. Phys. C12, L215 (1979)
D. Mattis: Phys. Rev. Lett. 19, 1478 (1967)
6.40 K. Yosida: Phys. Rev. 147,223 (1966)
6.41 K. Yosida, A. Yoshimori: in Magnetism, Vol. 5, ed. by G. Rado, H. Suhl (Academic,
New York 1973) p. 253
6.42 M. Daybell: "Thermal and Transport Properties", in Ref. 6.41
6.43 K. Binder, K. Schroder: Phys. Rev. B14, 2142 (1976)
6.44 L. R. Walker, R. E. Walstedt: Phys. Rev. Lett. 38, 514 (1977), Phys Rev. B22,3816(1980)
6.45 D. Mattis: Phys. Lett. 56A, 421 (1976); err. 60A, 492 (1977)
D. Sherrington: Phys. Rev. Lett. 41, 1321 (1978)
W. Y. Ching, D. L. Huber: Phys. Rev. B20, 4721 (1979)
6.46 J. M. Luttinger: Phys. Rev. Lett. 37, 778 (1976)
6.47 S. F. Edwards, P. W. Anderson: J. Phys. F5, 965 (1975), (nearest-neighbor model)
6.48 D. Sherrington, S. Kirkpatrick: Phys. Rev. Lett. 35, 1792 (1975); Phys. Rev. B17, 4384
(1978), (long-range interaction model)
6.49 P. Reed: J. Phys. CI2, L799 (1979) and references therein L. Morgenstern, K. Binder:
Phys. Rev. Lett. 43, 1615 (1979), (for 2D)
6.50 G. Toulouse: Commun. Phys. 2, 115 (1977)
S. Kirkpatrick: Phys. Rev. BI6, 4630 (1977)
E. Fradkin et al.: Phys. Rev. B 18, 4789 (1978)
6.51 L. Marland, D. Betts: Phys. Rev. Lett. 43,1618 (1979)
6.52 R. Palmer, C. Pond: J. Phys. F9, 1451 (1979)
6.53 J. M. Kosterlitz, D. J. Thouless, R. C. Jones: Phys. Rev. Lett. 36, 1217 (1976)
6.54 M. L. Mehta: Random Matrices (Academic, New York 1967)
D. C. Mattis, R. Raghavan: Phys. Lett. 75A, 313 (1980)
6.55 S. H. Liu: Phys. Rev. B17, 3629 (1978)
V. Korenman, J. Murray, R. Prange: Phys, Rev. BI6, 4032, 4048, 4058 (1977)
J. Hubbard: Phys. Rev. B19, 2626; B20, 4584 (1979)
 References 295

6.56 B. Matthias, R. Bozorth: Phys. Rev. 109604 (1958)

6.57 S. Pickart, H. Alperin, G. Shirane, R. Nathans Phys. Rev. LeU. 12, 444 (1964)
6.58 J. Hubbard: Proc. Roy. Soc. A266, 238 (1963); A277, 237 (1964); A281, 401 (1964);
also see [6.2]
6.59 J.R. Schrieffer: Theory 0/ Superconductivity (Benjamin, New York 1964)
6.60 V. M. Galitski: Sov. Phys. JETP 7, 104 (1958)
J. Kanamori: Prog. Theor. Phys. (Kyoto) 30, 275 (1963)
6.61 P. Richmond, G. Rickayzen: J. Phys. C2, 528 (1969)
6.62 E. Lieb, F. Wu: Phys. Rev. Lett. 20, 1445 (1968)
6.63 R. Mattuck: A Guide to Feynman Diagrams in the Many-Body Problem, 2nd ed. (McGraw-
Hill, New York 1976) p. 183 (p. 156 in the 1967 ed.)
6.64 Y. Nagaoka: Phys. Rev. 147, 392 (1966)
J. Sokoloff: Phys. Rev. B3, 3826 (1971)
6.65 D. Mattis, R. Pena: Phys. Rev. 10, 1006 (1974)
6.66 M. B. Stearns: Phys. Today (April 1978) pp. 34-39
6.67 A. V. Gold, L. Hodges, P. Panousis, R. Stone: Int. J. Magn. 2, 357 (1971)
6.68 K. Duff, T. Das: Phys. Rev. B3, 192,2294 (1971)
6.69 M. B. Stearns: J. Magnetism Magn. Mater. S, 167 (1977)
6.70 R. Stuart, W. Marshall: Phys. Rev. 120, 353 (1960)
6.71 R. Haydock, V. Heine, M. Kelly: J. Phys. CS, 2845 (1972)
6.72 P. W. Anderson: Phys. Rev. 124,41 (1961)
6.73 P. A. Wolff: Phys. Rev. 124, 1030 (1961)
6.74 J. Friedel: Can. J. Phys. 34,1190 (1956)
6.75 P. G. de Gennes: J. Phys. 23, 510 (1962)
6.76 J. R. Schrieffer, P. A. Wolff: Phys. Rev. 194,491 (1966)
6.77 D. Hamann: Phys. Rev. B2, 1373 (1970)
P. B. Wiegmann: Phys. Lett. 80A, 163 (1980)
H. Fukuyama, A. Sakurai: Progr. Theor. Phys. 62, 595 (1979)
H. Krishna-murthy, J. Wilkins,
K. G. Wilson: Phys. Rev. B21, 1003-1083 (1980)
Subject Index

Amperian loop 12, 13 Compound angular momentum

Anderson model of magnetic de- 79
fect atom 35, 278-280 Coupled-boson representation of
(connection to s-d model: angular momentum 82ff.
280ff.) Coupled-fermion representation of
Angular momentum 67ff. angular momentum 90,91
Angular momentum Curie law 19
classical 67 Curie-Weiss law 20
commutation relations 68- Cyclotron resonance 215
70
matrices 76 Defect atom in solids, eigenstates
Antiferromagnetism 94,169, of 272
175-189,256,271 Degeneracy, effects of 110, 113,
117, 120, 121, 226,262
Band magnetism 250ff., 265 Demagnetization factor 143
Bethe's ansatz l75ff. Density of states (electronic)
Binary interaction operator 149, 214,217,218
150 see also t-matrix Descartes 6, 7
Biot-Savart law 22 Diamagnetism 19, 23, 43, 215
Bloch function 208 Domains 30, 31, 206
Bloch T 3/2 law 28
Bohr magneton 24 Edwards-Anderson spin glass 248
Bohr-van Leeuwen theorem 21- Electron 17, 18
23 density of states 214,217,218
Boson commutation relations 82, density of states, semicircular
136 approximation 273
Bound state of magnons see effective mass 257, 258, 264,
Magnons, bound states 265
Brillouin zone 163 gas 252
Bubbles 30-33 scattering lifetime 258
scattering in Kondo effect
C1ebsch-Gordan coefficients 80, 37,240,241,245
81 (Table 3.2) Elementary excitation 172, 221,
Como ass 3 222
298 Subject Index

Equations of motion of angular Ising model of ferromagnetism

momentum 81,82,198 34,35
Exchange integral 45-47,54,
58,59,225 J (spin and angular momentum)
Exchange operator 27, 38ff., 99 111
J ordan-Wigner transformation
Faraday 14 90,91
Fermi surface (iron) 265
Fermion operators 90, 133,220 Kondo effect 37, 239ff., 272
Ferrimagnetism 189-192
Ferromagnetic resonance 146 LCAO 210,212 (Table 6.1)
Ferromagnetism 28,29, 139- Lie b-Ma ttis theorem
169,234,237,256, two-electron 106-109
of iron 265,266 many-electron 125 -129
Friedel oscillations see Ruder- Lifetime 149,258
man Kittel interaction Local moment formation 36,37,
272ff.
Gilbert 3,4 Lodestone 1-3, 5, 8
Green theorem 10 Lorentz force 18
Ground state
antiferromagnets 183-188 Magnetic bubble 30-33
ferro magnets 141, 146 Magnetic ground state 146, 172,
general method of obtaining 179,232,247-249,251
232 Magnetic impurity in nonmagnetic
Linear chain (Bethe solution) metal 35-37, 272ff.
197 Anderson's model 272,278-
XY model 172 280
Gyromagnetic ratio (g) 25, 26, s-d exchange model 280-283
238 (Table 6.2) Wolff's model 272, 276-279
Magnetic monopole 72-74
Hall effect 215 Magnetite 1
Heisenberg Hamiltonian 40, Magnetoresistance 215
138ff.,227-232 Magnons 138ff.
Heitler-London method 44,54, one-magnon states 146-160
160 one-magnon states in metals
Ho1stein-Primakoff operators 88, 267-270
139, 162 two-magnon states 147-
Hund's rules 109-112,226, 160,175
263 multimagnon states 165, 169
Hydrogen molecule 44-56 bound states 155ff.
Hyperbolic operators 87 in antiferromagnets 181
 Subject Index 299

nonlinear theory 162ff. Schwinger's coupled boson repre-

scattering lifetime 149 sentation 82
spectrum 179,181,183 s-d model 229, 280
spectrum in metals 267ff. Second quantization 91,133,
Many~lectron states 95ff. 220
Maxwell's equations 16 Slater determinants 96-98
Mean-free path factor 232 Solitons 198ff.
Molecular field 20 Spherical harmonics 70, 73
(Table 3.1)
Nagaoka model of ferromagnetism Spin 25
258ff. Spin glass 245ff.
Nodal surface 41,108,129 Edwards-Anderson model
Node1ess ground state 109 248
Nonorthogonal wavefunctions 45 separable model 247 -248
Nonorthogonality catastrophe Sherrington-Kirkpatrick model
56-66 248
spherical model 249
Oersted 11 Walker-Walstedt calculation
Orbital magnetism in metals 263 247
Overlap 41 Spin density oscillations 229,
230,246,283
Paramagnetic Curie temperature Spin waves 28, 199, 200 see
(0) in metals 234-236 Magnons
Paramagnetism 19,23,43,215 Spins one 92
Pauli spin matrices 78 Stoner's itinerant electron mag-
Peregrinus 3 netism 250ff.
Phase diagram Superconductivity 38
Hubbard model 256 Surface magnetism 192-195
degenerate Hubbard model Symmetry operations 42,49,99
264
magnetic impurity 278-280 t-matrix 254,262 see also
Poisson 10 Binary interaction operator
Term splittings 116,120
Quasiparticle 179, 221, 257 Terms for equivalent electrons
Quadratic form of spin coupling 121 (Table 4.3)
93,225 Thin films 33-35
Three~lectron states 49, 102,
Rare earth alloy 236, 238 112
Rotations 85 Tight-binding theory 209
Ruderman-Kittel interaction Two-dimensional magnetism 33-
229,231,237,246 35
300 Subject Index

Two-electron theorem 106-109 Weiss molecular field 20

Wolff model of magnetic defect in
Villain representation of angular metals 276 - 278 (connection
momentum 89 to s-d model 280ff.)
Vortices 196-198
XYmodel 169-174
Wannier functions 29,208,209
Weak magnetism of metals 215ff. Zeeman effect 17, 18
Weiss domain 30