(Graduate Texts in Physics) Henri Alloul (Auth.) - Introduction To The Physics of Electrons in Solids-Springer-Verlag Berlin Heidelberg (2011) PDF

GRADUATE TEXTS IN PHYSICS
Graduate Texts in Physics publishes core learning/teaching material for graduate- and
advanced-level undergraduate courses on topics of current and emerging fields within
physics, both pure and applied. These textbooks serve students at the MS- or PhD-level
and their instructors as comprehensive sources of principles, definitions, derivations, ex-
periments and applications (as relevant) for their mastery and teaching, respectively. Inter-
national in scope and relevance, the textbooks correspond to course syllabi sufficiently to
serve as required reading. Their didactic style, comprehensiveness and coverage of funda-
mental material also make them suitable as introductions or references for scientists enter-
ing, or requiring timely knowledge of, a research field.
Series Editors
Professor Richard Needs

Cavendish Laboratory
JJ Thomson Avenue
Cambridge CB3 0HE, UK
E-mail: rn11@cam.ac.uk
Professor William T. Rhodes

Florida Atlantic University
Imaging Technology Center
Department of Electrical Engineering
777 Glades Road SE, Room 456
Boca Raton, FL 33431, USA
E-mail: wrhodes@fau.edu
Professor H. Eugene Stanley

Boston University
Center for Polymer Studies
Department of Physics
590 Commonwealth Avenue, Room 204B
Boston, MA 02215, USA
E-mail: hes@bu.edu
The Ecole Polytechnique, one of France’s top academic institutions, has a long-
standing tradition of producing exceptional scientific textbooks for its students. The
original lecture notes, the Cours de l’Ecole Polytechnique, which were written by
Cauchy and Jordan in the nineteenth century, are considered to be landmarks in the
development of mathematics.
The present series of textbooks is remarkable in that the texts incorporate the
most recent scientific advances in courses designed to provide undergraduate stu-
dents with the foundations of a scientific discipline. An outstanding level of quality
is achieved in each of the seven scientific fields taught at the Ecole: pure and applied
mathematics, mechanics, physics, chemistry, biology, and economics. The uniform
level of excellence is the result of the unique selection of academic staff there which
includes, in addition to the best research in its own renowned laboratories, a large
number of world-famous scientists, appointed as part-time professors or associate
professors, who work in the most advanced research centers France has in each field.
Another distinctive characteristic of these courses is their overall consistency;
each course makes appropriate use of relevant concepts introduced in the other text-
books. This is because each student at the Ecole Polytechnique has to acquire basic
knowledge in the seven scientific fields taught there, so a substantial link between
departments is necessary. The distribution of these courses used to be restricted to
the 900 students at the Ecole. Some years ago we were very successful in making
these courses available to a larger French-reading audience. We now build on this
success by making these textbooks also available in English.
Henri Alloul
Introduction to the Physics

of Electrons in Solids
Translated by Stephen Lyle
123
Prof. Dr. Henri Alloul
Université Paris-Sud XI
CNRS
Bâtiment 510
Laboratoire de Physique des Solides
91405 Orsay CX
France
alloul@lps.u-psud.fr
Original French edition published by Édition de l’École Polytechnique, Palaiseau,

France, 2007
ISSN 1868-4513 e-ISSN 1868-4521

ISBN 978-3-642-13564-4 e-ISBN 978-3-642-13565-1
DOI 10.1007/978-3-642-13565-1
Springer Heidelberg Dordrecht London New York
Library of Congress Control Number: 2010932620
c Springer-Verlag Berlin Heidelberg 2011

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 microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations
are liable to prosecution under the German Copyright Law.
The use of general descriptive names, 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.
Cover design: WMXDesign GmbH, Heidelberg
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
The physics of condensed matter, that is liquid, solid, or amorphous materials, oc-
cupies something like 50% of physicists working in fundamental research, and the
industrial opportunities in the engineering sciences are very important. The best
known applications are in hi-tech, especially communications and information pro-
cessing (semiconductors, magnetic data storage), but also medical imaging (super-
conductivity and magnetism). However, this discipline has also led to spectacular
progress in the use of more commonplace materials employed in metallurgy (steels,
special alloys, composites) or the construction industry (concretes).
By its very nature, condensed matter physics has a broad interface with other
scientific disciplines relating to chemistry, biology, and mechanics. And this multi-
disciplinary aspect is growing steadily, through conceptual and instrumental devel-
opments which allow us to tackle ever more complex systems.
Before the advent of modern physics, it would have been difficult to understand
the physics of electrons in solids, and this is the key feature of condensed matter
physics. Since solids are collections of atoms, it is clear that the quantisation of
atomic electron states must play a major role in their properties. The basic concepts
of quantum mechanics and statistical physics are thus absolutely essential if we are
to understand the macroscopic behaviour of electrons in solids.
Although a whole range of different atoms can be built up from protons, neutrons,
and electrons, there is nevertheless a limit to what can be produced. On the other
hand, atoms can be associated in infinitely many combinations to make up a solid.
But from this apparent lawlessness, a range of original generic types of behaviour
emerge, whose properties would be difficult to imagine on the basis of the individual
atoms making up the solid. How is it that the electrons circulate freely in some
solids (metals), and induce significant magnetic forces in others (magnets)? The
great difficulty in predicting behaviour on an a priori basis is a characteristic of this
discipline, whence the predominant role of observation.
This is therefore a useful point to specify the spirit of our own teaching, and hence
of this textbook. In many countries, teaching traditions have always given pride of
place to a formal, and essentially deductive, presentation of the physics, i.e., starting
from formal hypotheses and leading up to observable consequences. This deductive
approach leaves a purely a posteriori verificational role to observation, and hides
the thinking that has gone into building up the models in the first place. Here we
v
vi Preface
shall adopt the opposite approach, which begins with the fact that in science in
general, and in solid state physics in particular, the qualitative understanding of a
phenomenon is an important step which precedes the formulation of any theoretical
development. We thus urge the reader to carry out a careful examination of the
deeper significance of experimental observations,1 in order to understand the need
for specific models and carry out realistic approximations.
In this way we can also present the main physical effects without necessarily
developing the whole theoretical formalism. This approach is indeed unavoidable,
since it is impossible today to explain the properties of all solids within a single the-
oretical framework. Quite the contrary, in fact, since the main themes discussed in
the book, viz., metals, superconductivity, and magnetism, are currently understood
through radically different approximations. For this reason, many issues which now
occupy researchers can be located precisely at the interface between these themes.
The best example is undoubtedly high-temperature superconductivity, discovered in
1986, which involves all three phenomena at the same time!
Another aim of this book is to demonstrate the high level of interplay between
fundamental scientific research and the development of modern technologies. In-
deed, these physical phenomena underlie many developments that are set to revolu-
tionise technology in the twenty-first century. We have thus decided not to restrict
here to the fundamental physical concepts, but to go ahead and introduce those ideas
that are essential for describing applications. Note that the nanotechnologies which
are so much in the news these days seek to exploit the properties of very small ob-
jects, with length scales in the nanometer range. However, it would be hopeless to
try to understand the characteristics of nanomaterials without a firm grasp of the
properties of larger physical systems and the methods used to study them.
From the earliest times, humans have exploited the properties of materials they
found in their natural surroundings, and many of these were solids. Of course, they
did not have to wait for the arrival of modern science in order to find uses for them.
The natural approach was always to take advantage of some observed behaviour,
e.g., the exposed edge of a broken flint for cutting purposes, the ductility of metals
for forging tools, or the good conduction of metals for making cooking pots, and so
on. The recognition of these properties, even if they may today appeal to elaborate
mathematical formulations, in no way requires an understanding of the fundamen-
tal underlying reasons. The mechanics of materials is almost always governed by
macroscopic constitutive laws based on observation, whose microscopic origins are
far from being fully understood. This in no way prevents their use. Magnetism is a
case in point. The natural occurrence of rocks able to attract iron has been known
since ancient times. The striking magnetic behaviour of iron has been ingeniously
exploited in the compass to help explorers find their way. But while observation
suffices for simple applications, it was only little by little that the chemical nature
of the constitutive elements, the regular structure of crystals, and other microscopic
features could eventually be identified.
1 To help things along, a number of questions have been interspersed throughout the book. The
answers to these questions can be found at the end of the chapter in which they have been raised.
Preface vii
The levitation of a magnet by a superconductor provides one of the most amazing manifestations
of superconductivity, a source of wonder to all that witness it for the first time, and especially to
students who have just themselves synthesised the superconducting ceramic YBa2 Cu3 O7 . How can
three insulating oxides Y2 O3 , CuO, and BaO react in the solid phase to generate a metallic mate-
rial which becomes superconducting in liquid nitrogen? This illustrates the fact that, in complex
systems, properties are very hard to predict from a straightforward understanding of the separate
constituents. The richness of condensed matter physics lies in the experimental revelation of spec-
tacular phenomena arising from complex systems, which leads us to seek rational explanations
viii Preface
Recent Trends in Condensed Matter Physics
The nineteenth and twentieth centuries witnessed a major change in condensed mat-
ter physics as developments in metallurgy, then in chemistry, produced more and
more artificial materials. For its part, physics tends to investigate natural materials
under artificial conditions, in order to characterise certain properties. Physicists try
to understand why some materials conduct electricity (Ohm’s law!), by studying
simple elementary metals. Experimentation can discover novel constitutive laws.
Superconductivity is a case in point. It was demonstrated for the first time in 1911
when investigating the properties of metals at very low temperatures. But the the-
ory needed to understand the microscopic origins of the behaviour was not yet
available.
The decisive step here was the advent of modern physics in the form of sta-
tistical physics and then quantum mechanics, which led to considerable progress.
It was found that many types of behaviour are macroscopic manifestations of
microscopic quantum phenomena. Examples are the quantum origins of magnetism
and the relation between the optical properties of solids and the organisation of
their electronic energy levels into energy bands, among many others. The process
of discovery picked up momentum after the 1950s, with the large-scale expansion of
scientific endeavour and the development of cross-disciplinary research: chemical
substances provided a host of new types of behaviour for physicists. The rate of fun-
damental discoveries has held up over the past thirty years, with the quantum Hall
effect, scanning tunneling microscopy, high-Tc superconductivity, fullerenes and
graphene, experimental observations that led in each case to Nobel prizes for those
involved.
In time, these new types of behaviour are put to use in applications. But here
scientists no longer limit themselves to understanding, nor engineers to application.
The two roles are moving closer and closer together. The discovery of semiconduc-
tors was the direct result of a hybrid approach, with applications (the diode and the
transistor) following very quickly after the discovery of the new type of behaviour
(the junction between two differently doped semiconductors). Likewise gas lasers,
and subsequently solid-state lasers, were purely artificial constructions which started
life as laboratory curiosities based on a sound understanding of the basic science.
The time scales between the discovery of new phenomena and concrete applica-
tions are shrinking fast, but still remain rather long (20 years for lasers, and 50 for
superconductors). Economic factors are clearly the key, but experience shows that a
deeply novel type of behaviour always leads in the end to some viable application.
This discussion might suggest that solid-state physics is a purely utilitarian sci-
ence, but this would be quite wrong. In order to understand a novel observation,
sophisticated theoretical models are often required, and these problems are natu-
rally more difficult to solve than those generated by simple systems that prove easier
to formulate. The difficulty here comes from trying to understand the macroscopic
properties of ever more complex systems. This still involves the use of sometimes
simplistic models, but which may be very hard to solve formally. More and more
sophisticated instruments are developed to probe the structure and properties of
Preface ix
materials on the microscopic scale. One has to deal with problems in which the basic
state of the physical system is far from being intuitive, such as superconductivity or
the Kondo effect. But novel theoretical ideas and methods are also devised. The the-
ory of the renormalisation group has revolutionised the study of phase transitions,
and the notion of disorder has become a subject of theoretical physics. Minimisa-
tion methods can determine the energy minimum of a disordered magnetic system,
such as a spin glass, but can be broadly generalised to study neural networks or
optimisation of the travelling salesman problem.
The style and choice of material in the following lectures on electrons in solids at-
tempts to reflect this kind of approach, which characterises the discipline and which
is not generally familiar to students. To achieve this, we have put the emphasis on
two themes, namely magnetism and superconductivity, which are manifestations on
the macroscopic scale of the quantum properties of solids, and which lead to many
current or potential hi-tech applications. Although these same themes have up to
now been the subject of detailed independent presentations, it seems judicious here,
for historical reasons, to present them in tandem.
Magnetism and Superconductivity
How do magnetism and superconductivity fit in to the general context of condensed

matter physics? The two disciplines are clearly distinct, in particular through their
history, and have different consequences for technology. Magnetic materials have
been known since ancient times, while superconductors were only discovered at
the beginning of the twentieth century. The former have well known applications
resulting from their behaviour, i.e., strong mutual attraction, and interaction with
fields produced by electric currents. The latter also display novel types of behaviour,
i.e., zero electrical resistance and diamagnetism, but which are only manifested un-
der currently unfavourable economic conditions, viz., low temperatures. Everything
would tend to distinguish the industrial impact of these two sciences.
But if one considers their most basic manifestations, the analogy between the
magnetic fields produced by currents or by magnets has been common knowledge
since the nineteenth century. Even before the end of the 1930s, quantum mechan-
ics had shown that the electronic moments of atoms, like the persistent currents in
superconductors, are quantum phenomena. It was soon understood that they result
indirectly from electrostatic interactions between electrons. The two scientific dis-
ciplines witnessed several novel developments, rewarded by Nobel prizes: Bloch–
Purcell–Pound, Mössbauer, Néel, Mott–Anderson–Van Vleck in one, and Bardeen–
Cooper–Schrieffer, Josephson–Giaver in the other. Common scientific applications
were developed, such as superconducting solenoids used in nuclear magnetic res-
onance (NMR) in chemistry, physics, and biology, but also instrumentation based
on superconductors for work on magnetism (SQUIDs). But there was still no com-
mon scientific textbook discussing both disciplines, and lecture courses remained
quite separate, reserved for a few hundred specialists. Magnetism is interesting
x Preface
for scientists because it can be used to study more general problems through
specific high-performance experimental tools (neutrons, NMR, Mössbauer), while
remaining susceptible to simplified and well tested theoretical formulation. In
this way, many problems of statistical thermodynamics, such as phase transi-
tions, low-dimensional systems, disorder, spin glasses, and so on, found elegant
solutions.
At the beginning of the 1980s, industry had little time for superconductivity,
but the applications of magnetism were many and varied. The electrical industrial
revolution was made possible thanks to the magnetic materials used in electrical
generators and motors. The magnetic recording industry alone represents an annual
turnover that can be measured in multiples of 100 million US dollars. But supercon-
ductivity has come into the limelight through its association with NMR in magnetic
resonanance imaging (MRI), bringing together the two disciplines in the first ap-
plication to affect the general public. But these results were not enough to lead
to generalised applications. (At the time, P.G. de Gennes, an original contributor
to both fields and future Nobel prizewinner, claimed that these were sciences of
the past!)
However, the discovery of high-temperature superconductors in 1986, with the
attribution of the Nobel prize to Bednorz and Müller, gave hope for applications in
the twenty-first century. Until then, theoreticians working on superconductivity had
thought that the critical temperature Tc of superconductivity would not be able to go
above 30 K. And yet here was an example of superconductivity with Tc ≈ 150 K, in
systems expected rather to exhibit novel magnetic properties. Understanding this
kind of superconductivity has proved to be a tremendous challenge for science.
These new superconductors incidentally allow us now to exhibit and popularize the
basic superconducting phenomena in simple demonstrations as that depicted in the
image of page vii, or using toy levitating trains.
Regarding magnetism, new high magnetisation ferromagnetic materials have
been discovered. Magnetic data storage gained little benefit from scientific devel-
opments in the 1970s, since increases in data storage density were achieved rather
through improved mechanics and accuracy of read heads. But more recently, a new
read method has been devised, based on a magnetic phenomenon known as giant
magnetoresistance, which has allowed a considerable step forward in this domain.
This too was rewarded by the Nobel prize, attributed to A. Fert and P. Grünberg
in 2007. This has initiated the novel field known as spintronics, which exploits the
electron spin in electronic devices, and has also given hope of new applications.
It involves new materials and is currently the subject of considerable interest in
research centers.
Magnetism and superconductivity thus raise novel problems of a fundamental
nature, but also in relation to materials and their applications, and they are now
commonly encountered together in the most recent developments. The same mate-
rials can be made magnetic or superconducting with the help of minimal chemical
adjustments. The instruments used to investigate them are similar, and often even
the same. This coexistence can only be strengthened by the advent of nanotechnol-
ogy, and the trend has reached a point at which it seems essential to bring the two
Preface xi
subjects together into a single lecture course for students of science, whether they
intend to go into research or engineering. The basic level of knowledge required by
the engineer is clearly going to increase in the decades to come!
A Brief Guide to the Course
Why are electrons free to move in some solids, namely metals, but localised in oth-
ers, namely insulators? To tackle this problem, we shall focus mainly on crystalline
solids. The existence of a three-dimensional periodicity in these solids allows one
to establish simple rules for the electron energy levels. We shall determine these
basic rules using the independent electron approximation, i.e., considering only the
Coulomb potential of the ions and accounting for the Coulomb repulsion of the
electrons through only an average value. In this way, we find that the electron en-
ergy levels are distributed over allowed energy bands separated by forbidden bands,
or bandgaps. This is shown in Chap. 1 for the simple case of a linear chain of atoms.
To deal with 3D solids, some notions of crystallography are introduced in Chap. 2.
We can then calculate the energy band structure using some simple approximations.
Several experimental methods for determining the characteristics of the band struc-
ture are described in Chap. 3. The data displayed in page xiii exemplifies the most
efficient technique, constantly improved since the 1990’s, with the development of
new synchrotrons. This energy level structure is the key to understanding the dif-
ference between metal and insulator, and introducing the notion of semiconductor.
The physical parameters constraining the conductivity of metals are discussed in
Chap. 4. These first chapters of the book already allow some simple applications
of methods developed in quantum mechanics and statistical physics. They provide
the basis for a subsequent understanding of semiconductors. The specificities of
the band structure and transport properties of graphene, for which A. Geim and
K. Novoselov have been awarded the 2010 Nobel prize, are highlighted as well.
The results of the first four chapters must be reconsidered if interactions be-
tween the electrons can no longer be neglected. But this is exactly what happens
in superconductors and magnetic solids. Understanding the consequences of these
interactions between electrons for the physical properties of the resulting solids is a
central problem in our field. It goes well beyond the framework of an undergraduate
course, which is the level aimed at here, but we feel it important to provide some
kind of overview in the case of simple magnetic and superconducting materials.
Students have already acquired some understanding of phenomena related to
magnetism. We have thus chosen to begin with superconductivity as the less fa-
miliar of the two disciplines. This should allow a gradual assimilation of the ideas,
for we shall then refer back to them on many occasions in the rest of the course.
A full discussion of the superconductivity of simple metals, which is now well un-
derstood, must be left to the doctoral level. For the present purposes, we shall stick
to a limited description based on a historical and experimental presentation. This
deliberate choice is used to place the student in the same conditions as those which
led to our present understanding of this phenomenon. This is all the more important
xii Preface
in that we shall then be able to illustrate an approach that is not altogether natu-
ral for students coming from a school education system. In any case, in the spe-
cific case of superconductivity, it would be optimistic to carry out a full theoretical
presentation without first putting across the subtleties of the physical behaviour of
superconductors.
Chapter 5 is thus devoted to the existence of persistent currents, the exclusion
of magnetic induction in a superconductor, and the resulting quantisation of the
magnetic flux. Chapter 6 studies the thermodynamic properties of superconductors,
and the notions of coherence length and mixed phase that follow. Finally, in Chap. 7,
we describe the experiments that help us to understand the microscopic origins of
superconductivity, and analyse the Cooper calculation which lays the foundation for
a full understanding.
In Chap. 8, we briefly discuss the origins of atomic magnetism and the exchange
interactions responsible for ferromagnetism and antiferromagnetism. The idea of
magnetic anisotropy which explains the existence of magnetic domains and domain
walls in ferromagnetic materials is essential for understanding applications of mag-
netism (Chap. 9). In Chap. 10, we present experimental methods used to characterise
the magnetic properties of materials. These reveal a methodological link between
magnetism and superconductivity. In both cases, the need to understand physical
properties and meet technological requirements has encouraged the development
of methods for observing magnetism on microscopic, and even nanoscopic scales.
Magnetic resonance detection is introduced in Chap. 11, and leads to a discussion
of nuclear magnetic resonance and the importance of hyperfine methods for study-
ing magnetism and superconductivity on the atomic scale. Finally, in Chap. 12, we
discuss the elementary excitations (spin waves) in a ferromagnetic material, and also
the basic theory of thermodynamic properties close to a phase transition (Landau
and Ginzburg–Landau theories).
Preface xiii
Experimental observation of the electronic states of Sr2 RuO4 obtained by Angular Resolved
Photoemision Spectroscopy (ARPES), a technique that takes benefit from the intense monochro-
matic light beams produced by synchrotrons. This technique has been intensively developed since
the 1990s in particular for the study of high temperature superconductors.
Sr2 RuO4 is a layered compound whose metallic behaviour is due to RuO2 planes interleaved
with insulating SrO planes. The Ru atoms form a square lattice with oxygen inserted between
them. The single crystal sample has been cleaved and exhibits a flat RuO2 plane. Irradiation by a
monochromatic light of sufficient energy emits photoelectrons (this is the photoelectric effect for
which A. Einstein has been awarded a Nobel prize). The study of the photoelectron intensity as
a function of the emission angle allows one to map out the energy distribution of the electronic
levels as a function of their in plane wave vector (kx , ky ). The projected curves represent the Fermi
surface (see question 3.6 in Chapter 3). Image courtesy of A. Damascelli, from experimental results
by Damascelli, A., Lu, D.H., Shen, K.M., Armitage, N.P., Ronning, F., Feng, D.L., Kim, C., Shen,
Z.-X., Kimura, T., Tokura, Y., Mao, Z.Q., Maeno, Y., Phys. Rev. Lett. 85, 5194 (2000)
xiv Preface
A Guided Choice of Problems
Over the years this lecture course on the physics of electrons in solids has been
taught at the Ecole polytechnique (Palaiseau, France), its organisers have produced
a significant number of written tests. We have often been led to propose themes
focussing on experimental observations, which encourage students to reflect on their
physical meaning.
In Chapter 13, we have included those problems which illustrate novel physical
effects that are not discussed in detail in the lectures. Indeed, it seemed important to
complement the basic ideas discussed in the book with a presentation of the more
important aspects of solid-state physics which could not be taught during the year
in the limited time available for lectures and supervisions. The aim here is not there-
fore a mere test of understanding acquired by studying the main course. Students
who have independently acquired an understanding of the basic ideas of solid-state
physics will be able to use the subjects here to glimpse some of the very active
ongoing research themes, illustrated by modern experimental methods.
Direct manifestations of electronic band structure in the optical response of solids
are examined in Problem 2: Reflectance of Aluminium and Problem 5: Optical
Response of Monovalent Metals. The total energy of the electronic band states
can affect the crystal structure of certain alloys (see the Hume–Rothery rules of
Problem 4: Electronic Energy and Stability of Alloys) or induce a Peierls transition,
which corresponds to a doubling of the unit cell and a metal–insulator transition in
1D compounds (Problem 6: One-Dimensional TTF-TCNQ Compounds).
We have illustrated the specific band structures of certain metals that are impor-
tant for their superconducting properties, such as the high-Tc cuprate YBa2 Cu3 O7
(Problem 3: Band Structure of YBa2 Cu3 O7 ), the so-called A15 compounds, such as
V3 Si in which the vanadium atoms are arranged in chains (Problem 14: Electronic
Structure and Superconductivity of V3 Si), or the new superconducting compound
MgB2 , whose highly 2D electronic structure is similar to that of graphene and
graphite (Problem 16: Magnesium Diboride: A New Superconductor?).
Since semiconductor technology is often the subject of specific lecture courses,
as is the case at the Ecole polytechnique, the discussion here simply reviews the
relevance of the band structure for their electronic transport properties. The classic
cyclotron resonance experiments are used to determine the effective masses of elec-
trons and holes (Problem 8: Cyclotron Resonance). The notion of transition from
a Mott insulator to a metal is introduced in such systems by examining the struc-
ture of donor electron levels introduced by substituting P in Si. This transition re-
sults from competition between Coulomb repulsion at an atomic site, which favours
electron localisation, and the tendency of electrons to delocalise due to hopping inte-
grals between neighbouring sites. This competition can explain the insulator–metal
transition observed in Si when it is strongly doped with phosphorus (Problem 7:
Insulator–Metal Transition).
The notion of atomic vibrations and their quantisation in terms of phonons is
barely touched upon in the main course chapters. Further considerations can be
found in Problem 9: Phonons in Solids, where their effects on the heat capacity
Preface xv
of solids and the resistivity of pure metals are exemplified. Their consequences for
X-ray diffraction diagrams are also studied in Problem 1: Debye–Waller Factor.
Many problems here are devoted to a deeper investigation of the novel properties
associated with superconductivity. For example, one can study the effect of flux
quantisation on the thermodynamic properties of superconductors (Little–Parks
experiment discussed in Problem 10: Thermodynamics of a Thin Superconducting
Cylinder) and the direct or alternating Josephson effect in zero field (Problem 11:
Direct and Alternating Josephson Effects in Zero Magnetic Field). The existence of
mixed states in a Josephson junction (Problem 12: Josephson Junction in a Mag-
netic Field), and the irreversibility of magnetisation curves induced by interactions
between vortices and the surface of a type II superconductor (Problem 13: Magneti-
sation of a Type II Superconductor) provide a better understanding of the physics of
vortices. The observation of these mixed states by scanning tunnelling microscopy
(Problem 15: Superconductivity of NbSe2 ) provides a window on the excited states
of superconductors and the fine structure of the superconducting gap in compounds
like NbSe2 or MgB2 (Problem 16: Magnesium Diboride: A New Superconductor?).
The aspects of magnetism dealt with here are the antiferromagnetism of undoped
cuprates (Problem 17: Electronic Properties of La2 CuO4 ) and the magnetic proper-
ties of antiferromagnetic materials in the molecular field approximation (Problem 18:
Properties of an Antiferromagnetic Solid). Finally, the importance of the mag-
netism of thin films is illustrated by the use of magneto-optical methods to show
how they decompose into domains (Problem 19: Magnetism of Thin Films and
Magneto-Optic Applications), and an investigation of surface effects for the mag-
netic anisotropy of thin films (Problem 20: Magnetism of a Thin Film).
It is hoped that the reader’s curiosity will be stimulated by the problem approach
to these exciting physical phenomena and the considerable reflection required to
explain the experimental observations discussed here. This problem set should allow
the reader to glimpse the rich rewards of the scientific method, which relentlessly
confronts observation with theoretical models.
Orsay, October 2010 Henri Alloul
Contents
1 The Quantum Mechanics of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 General Hamiltonian for a Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 The Need for Approximations . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Wave Functions and Pauli Exclusion Principle . . . . . . . . . . . . 7
1.2 Energy Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Tight-Binding Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Linear Atomic Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Energy Bands Associated with Tightly Bound
Atomic Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 Crystalline Solids: Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.1 Crystal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Crystal Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2 Two-Dimensional Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.3 Three-Dimensional Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1.4 Beyond the Perfect Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.1 General Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.2 Reciprocal Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Determination of Crystal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 On the Bragg Diffraction Condition . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 Diffraction Intensity and Basis of the Primitive Cell . . . . . . . 41
2.3.3 X-Ray Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Answers to Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 Electronic Structure of Solids: Metals and Insulators . . . . . . . . . . . . . . . . 51

3.1 Electrons in a Periodic Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1.1 Bloch’s Theorem in Three Dimensions . . . . . . . . . . . . . . . . . . 53
3.1.2 General Properties of Bloch States . . . . . . . . . . . . . . . . . . . . . . 55
xvii
xviii Contents
3.1.3 Electrons in a Weak Periodic Potential . . . . . . . . . . . . . . . . . . 56

3.1.4 First Brillouin Zone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Band Structure of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.1 Linear Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.2 Band Structure of Two-Dimensional Solids . . . . . . . . . . . . . . 64
3.2.3 Three-Dimensional Band Structures . . . . . . . . . . . . . . . . . . . . 68
3.3 Metals, Insulators, and Semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.1 Statistical Physics of Fermions . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.2 Real Solids: Conductor or Insulator? . . . . . . . . . . . . . . . . . . . . 74
3.3.3 Introduction to the Semiconductor Case . . . . . . . . . . . . . . . . . 77
3.3.4 The Special Case of Graphene . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.5 Thermodynamic Properties of Metals . . . . . . . . . . . . . . . . . . . 80
3.4 Experimental Determination of Band Structures . . . . . . . . . . . . . . . . . 84
3.4.1 Optical Absorption in Insulators . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.2 Tunneling Effect and Scanning Tunneling Microscopy . . . . . 85
3.4.3 Angle-Resolved Photoemission Spectroscopy . . . . . . . . . . . . 88
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4 Electron Transport in Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.1 Drude Model for Transport in an Electron Gas:
Relaxation Time and Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.1.1 Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.1.2 Thermal Conductivity and the Wiedemann–Franz Law . . . . . 105
4.2 Electron Transport in a Fermion Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2.1 Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2.2 Thermal Conductivity and the Wiedemann–Franz Law . . . . . 110
4.3 Electrons in a Lattice: Dynamics of Bloch Electrons . . . . . . . . . . . . . . 111
4.3.1 Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.2 Acceleration in Reciprocal Space and Real Space . . . . . . . . . 113
4.3.3 Electronic Conductivity in a Crystal . . . . . . . . . . . . . . . . . . . . 115
4.4 Origin of Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.1 Experimental Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.2 Scattering by Lattice Vibrations . . . . . . . . . . . . . . . . . . . . . . . . 118
4.4.3 Collisions with Impurities and Defects . . . . . . . . . . . . . . . . . . 119
4.5 Electrons, Holes, and Dopants in Semiconductors . . . . . . . . . . . . . . . . 121
4.6 Electrons, Holes, and Transport in Graphene . . . . . . . . . . . . . . . . . . . . 126
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5 Introduction to Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

5.1 Conditions for Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.1.1 Persistent Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.1.2 Critical Field and Critical Current . . . . . . . . . . . . . . . . . . . . . . 150
Contents xix
5.2 Difference Between a Perfect Conductor and a Superconductor:

The Meissner Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.1 Effect of a Magnetic Field on a Perfect Conductor . . . . . . . . 152
5.2.2 Meissner Effect: Exclusion of the Magnetic Field . . . . . . . . . 154
5.2.3 London Equations and Penetration Depth . . . . . . . . . . . . . . . . 155
5.3 A Macroscopic Quantum Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
5.3.1 Macroscopic Wave Function and Current . . . . . . . . . . . . . . . . 158
5.3.2 Quantisation of Magnetic Flux . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3.3 Measuring the Flux Quantum . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.3.4 Josephson Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.3.5 SQUID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6 Thermodynamics of Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

6.1 Thermodynamics of Bulk Superconductors . . . . . . . . . . . . . . . . . . . . . 177
6.1.1 Free Energy of the Superconducting State . . . . . . . . . . . . . . . 178
6.1.2 Entropy and Specific Heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2 Thin Films and Coherence Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.2.1 Thermodynamics of Thin Films . . . . . . . . . . . . . . . . . . . . . . . . 181
6.2.2 Coherence Length ξ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.2.3 Mixed State: A Simple Model . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.3 Two Types of Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.3.1 Type I Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.3.2 Simple London Model for a Vortex . . . . . . . . . . . . . . . . . . . . . 189
6.3.3 Type II Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.3.4 Applications of Type II Superconductors . . . . . . . . . . . . . . . . 194
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7 Microscopic Origins of Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . 201

7.1 Conventional Metal Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.1.1 Superconducting Metals and Alloys . . . . . . . . . . . . . . . . . . . . 203
7.1.2 Attractive Interaction Due to Phonons . . . . . . . . . . . . . . . . . . . 205
7.1.3 The Band Gap in the Superconducting State . . . . . . . . . . . . . . 208
7.2 Cooper Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.2.1 Electrons in the Conduction Band of a Semiconductor . . . . . 212
7.2.2 Extra Electrons in a Metal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
7.3 BCS Theory: Experimental Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.3.1 Ground State and Band Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.3.2 Excited States in the BCS Theory . . . . . . . . . . . . . . . . . . . . . . 216
7.3.3 Electron Tunneling Determination of the Gap . . . . . . . . . . . . 217
7.3.4 Critical Temperature and BCS Theory . . . . . . . . . . . . . . . . . . . 220
7.3.5 Coherence Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
xx Contents
7.4 High-Tc Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

7.4.1 Cuprates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.4.2 Other Families of Superconductors . . . . . . . . . . . . . . . . . . . . . 225
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
8 Magnetism of Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

8.1 Magnetic Behaviour of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
8.2 Magnetism of Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.2.1 Hydrogen Atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.2.2 Multielectron Atoms with Filled Shells . . . . . . . . . . . . . . . . . . 237
8.2.3 An Atom with a Partially Filled Shell: Carbon . . . . . . . . . . . . 237
8.2.4 Atoms with Partially Filled Shells: Hund Rules . . . . . . . . . . . 240
8.3 Paramagnetism of an Ensemble of Isolated Ions . . . . . . . . . . . . . . . . . . 243
8.4 Ordered Magnetic States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.4.1 Interatomic Exchange Interaction . . . . . . . . . . . . . . . . . . . . . . . 245
8.4.2 Heisenberg Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
8.4.3 Ferromagnetism: Molecular Field Approximation . . . . . . . . . 249
8.5 Antiferromagnetism and Ferrimagnetism . . . . . . . . . . . . . . . . . . . . . . . 252
8.6 From Insulator Magnetism to Metallic Magnetism . . . . . . . . . . . . . . . 255
8.6.1 Mott–Hubbard Insulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
8.6.2 Mott Transition and Doped Mott–Hubbard Insulators . . . . . . 258
8.6.3 Magnetism and Superconductivity . . . . . . . . . . . . . . . . . . . . . . 259
8.6.4 Metallicity and Magnetism in a Band Approach . . . . . . . . . . 261
8.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
9 Magnetic Anisotropy, Domains, and Walls . . . . . . . . . . . . . . . . . . . . . . . . . 269

9.1 Magnetic Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
9.1.1 Magnetocrystalline Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . 272
9.1.2 Effect of Anisotropy: Irreversibility . . . . . . . . . . . . . . . . . . . . . 273
9.2 Dipole Interactions, Demagnetising Fields, and Domains . . . . . . . . . . 277
9.2.1 Demagnetising Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.2.2 Demagnetisation Energy and Magnetic Domains . . . . . . . . . . 282
9.3 Bloch Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
9.4 Magnetic Hysteresis Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
9.4.1 Bitter Method for Observing Domains . . . . . . . . . . . . . . . . . . 288
9.4.2 Magnetisation Process and Hysteresis . . . . . . . . . . . . . . . . . . . 289
9.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
10 Measurements in Magnetism: From the Macroscopic

to the Microscopic Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
10.1 Macroscopic Magnetic Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 297
10.1.1 Susceptibility or Magnetisation . . . . . . . . . . . . . . . . . . . . . . . . 298
10.1.2 Examples of Measurements on Superconductors . . . . . . . . . . 301
Contents xxi
10.1.3 Hysteresimeters: Toroidal Geometry . . . . . . . . . . . . . . . . . . . . 303

10.2 Magnetic Surface Measurements and Magnetic Imaging . . . . . . . . . . 304
10.2.1 Microscale Measurements and Surface Imaging . . . . . . . . . . 305
10.2.2 Towards Submicron Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
10.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
11 Spin Dynamics and Magnetic Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

11.1 Dynamics in Magnetism: General Considerations . . . . . . . . . . . . . . . . 319
11.1.1 Linear Response and Dissipation . . . . . . . . . . . . . . . . . . . . . . . 320
11.1.2 Pulse Response and Frequency Response . . . . . . . . . . . . . . . . 321
11.2 Dynamics in Ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
11.2.1 Low Frequency Losses in Ferromagnets . . . . . . . . . . . . . . . . . 322
11.2.2 Ferromagnetic Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
11.3 Resonance in the Paramagnetic Regime . . . . . . . . . . . . . . . . . . . . . . . . 331
11.3.1 Resonance for a Thermodynamic Ensemble of Spins . . . . . . 331
11.3.2 Nuclear Magnetic Resonance (NMR) . . . . . . . . . . . . . . . . . . . 333
11.3.3 Spin Echoes: Transverse Relaxation . . . . . . . . . . . . . . . . . . . . 336
11.3.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
11.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
12 The Thermodynamics of Ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

12.1 Excited States and Low Temperature Properties . . . . . . . . . . . . . . . . . . 347
12.1.1 Magnons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
12.1.2 Low Temperature Thermodynamic Properties . . . . . . . . . . . . 349
12.1.3 Experimental Detection of Magnons . . . . . . . . . . . . . . . . . . . . 351
12.2 The Magnetic Phase Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
12.2.1 Mean Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
12.2.2 Landau and Ginzburg–Landau Theories . . . . . . . . . . . . . . . . . 354
12.2.3 Critical Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
12.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
13 Problem Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

Problem 1 Debye–Waller Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
Problem 2 Reflectance of Aluminium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Problem 3 Band Structure of YBa2 Cu3 O7 . . . . . . . . . . . . . . . . . . . . . . . . . 377
3.1 Isolated Copper–Oxygen Chain . . . . . . . . . . . . . . . . . . . 378
3.2 Isolated Copper–Oxygen Plane . . . . . . . . . . . . . . . . . . . . 379
3.3 Chain and Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
3.4 Realistic Models of YBa2 Cu3 O7 . . . . . . . . . . . . . . . . . . 381
Problem 4 Electronic Energy and Stability of Alloys . . . . . . . . . . . . . . . . 391
Problem 5 Optical Response of Monovalent Metals . . . . . . . . . . . . . . . . . 399
Problem 6 One-Dimensional TTF-TCNQ Compounds . . . . . . . . . . . . . . 405
6.1 Isolated Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
xxii Contents
6.2 Experimental Observations . . . . . . . . . . . . . . . . . . . . . . . 406

6.3 Dimerised Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
6.4 Peierls Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Problem 7 Insulator–Metal Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7.1 Tight-Binding Method for Hydrogen-Like Orbitals . . . 419
7.2 Interactions Between Electrons . . . . . . . . . . . . . . . . . . . . 420
7.3 Alkali Elements and Hydrogen . . . . . . . . . . . . . . . . . . . . 423
7.4 Insulator–Metal Transition in Si–P . . . . . . . . . . . . . . . . . 423
Problem 8 Cyclotron Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
8.1 Real and Reciprocal Space Paths of an Electron State . 441
8.2 A Semiconductor: Silicon . . . . . . . . . . . . . . . . . . . . . . . . 442
8.3 Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
Problem 9 Phonons in Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.1 Einstein Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.2 Debye Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.3 Experimental Detection of Phonons . . . . . . . . . . . . . . . . 455
9.4 Thermodynamic Properties . . . . . . . . . . . . . . . . . . . . . . . 456
9.5 Resistivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
Problem 10 Thermodynamics of a Thin Superconducting Cylinder:
Little–Parks Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
Problem 11 Direct and Alternating Josephson Effects
in Zero Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
11.1 Model Josephson Junction . . . . . . . . . . . . . . . . . . . . . . . . 481
11.2 Realistic Josephson Junction . . . . . . . . . . . . . . . . . . . . . . 481
11.3 Josephson Junction in a Microwave Field . . . . . . . . . . . 483
Problem 12 Josephson Junction in a Magnetic Field . . . . . . . . . . . . . . . . . . 491
12.1 Current Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
12.2 Screening of the Magnetic Field . . . . . . . . . . . . . . . . . . . 493
12.3 Josephson Plasma Resonance . . . . . . . . . . . . . . . . . . . . . 495
Problem 13 Magnetisation of a Type II Superconductor . . . . . . . . . . . . . . . 511
Problem 14 Electronic Structure and Superconductivity of V3 Si . . . . . . . 523
Problem 15 Superconductivity of NbSe2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
Problem 16 Magnesium Diboride: A New Superconductor? . . . . . . . . . . . 541
16.1 Atomic and Electronic Structure of MgB2 . . . . . . . . . . . 541
16.2 Superconductivity of MgB2 . . . . . . . . . . . . . . . . . . . . . . . 543
Problem 17 Electronic Properties of La2 CuO4 . . . . . . . . . . . . . . . . . . . . . . 557
Problem 18 Properties of an Antiferromagnetic Solid . . . . . . . . . . . . . . . . 563
18.1 Preliminaries: The Ferromagnetic Case . . . . . . . . . . . . . 563
18.2 Antiferromagnetic Transition . . . . . . . . . . . . . . . . . . . . . 564
18.3 Susceptibility in the Antiferromagnetic State . . . . . . . . 565
Problem 19 Magnetism of Thin Films and Magneto-Optic Applications . 573
Problem 20 Magnetism of a Thin Film . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
20.1 Uniform Magnetisation . . . . . . . . . . . . . . . . . . . . . . . . . . 580
20.2 Non-uniform Situations . . . . . . . . . . . . . . . . . . . . . . . . . . 581
20.3 Detailed Investigation of Non-uniform Situations . . . . 583
Contents xxiii
Appendix A Physical Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593
Appendix B Some Useful Functions and Relations . . . . . . . . . . . . . . . . . . . . 595
Appendix C Standard Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
Appendix D Specific Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613
Acknowledgements
This is a good point to acknowledge that the very spirit and style of these lectures
results from a long collaboration with my colleague Heinz Schulz. While teaching
a range of different topics to small groups of students, we gradually refined this
approach, which looks for a good balance between experimental observations, the-
oretical developments, and applications. Heinz Schulz actually wrote some parts of
the original manuscript for this book, but unfortunately his untimely death brought
this joint project to a premature end. Since then, the initial structure of the book has
been radically changed as a result of further interaction with students, and a subse-
quent harmonisation of the overall layout. I have nevertheless retained some trace of
his altogether original contributions, especially in Chaps. 1, 8, and 12, in homage to
this sadly departed colleague. To a lesser extent, I have also retained some elements
of the relevant contribution by C. Hermann, who helped us to set up this lecture
series in the first year. I extend my warmest thanks to her.
I would also like to thank Florence Albenque, Silke Biermann, Pierre Ledoussal,
and Gilles Montambaux for their critical readings of and comments on certain parts
of this manuscript. The French version of this book was prepared in the publishing
department of the Ecole Polytechnique at Palaiseau in France. Over the years it has
been a great pleasure to work with Mmes M. Digot, M. Maguer, V. Pellouin, and D.
Toustou. I thank them for their efficient, patient, and friendly assistance.
The translation of the French manuscript has been ensured by Stephen Lyle. It
has been a great pleasure for me to have this opportunity to learn many subtleties
of the English language through numerous e-mail exchanges. I am pleased to thank
him warmly here for his extremely efficient and friendly collaboration.
Orsay, October 2010 Henri Alloul
xxv
Heinz Jürgen Schulz (1954–1998)
H. Schulz was a brilliant theoretician in the field of solid state physics, who always
paid careful attention to the dialogue between theory and experiment. His scien-
tific contributions are distinguished by the use of elegant and powerful methods
of statistical physics. His main interest was in low-dimensional electron systems,
in highly competitive areas, hotly debated by the international scientific commu-
nity, such as one-dimensional organic conductors, high-temperature superconduc-
tors, and strongly correlated fermions. Heinz was highly esteemed, especially by
the students at the Ecole Polytechnique, for his broad scientific interests and his
depth of understanding, but also his ability to listen and his discrete kindness. He
was taken from us in November 1998 by a sudden illness which carried him away
in less than six months.
Chapter 1
The Quantum Mechanics of Solids
Contents
1.1 General Hamiltonian for a Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1.1 The Need for Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Wave Functions and Pauli Exclusion Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Energy Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Tight-Binding Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Linear Atomic Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Energy Bands Associated with Tightly Bound
Atomic Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
H. Alloul, Introduction to the Physics of Electrons in Solids, Graduate Texts in Physics, 1

DOI 10.1007/978-3-642-13565-1 1, c Springer-Verlag Berlin Heidelberg 2011
2 1 The Quantum Mechanics of Solids
Image obtained by scanning tunneling microscopy of the surface of a single-crystal TTF-TCNQ

(scale 150 Å×150 Å). The vertical chains formed by stacking up organic molecules are only very
weakly coupled to one another and hence constitute molecular wires with an electrical conductivity
about 103 times greater in the vertical direction than in the horizontal directions. See Problem 6:
One-Dimensional TTF-TCNQ Compounds for more details. Image courtesy of Wang, Z.Z. (LPN,
Marcoussis)
1.1 General Hamiltonian for a Solid 3
In this chapter, we discuss the quantum description of a solid. As we shall see in

the next section, the problem is much too complex to be treated in its most general
form. We must therefore introduce approximations motivated by physical consid-
erations. The first will be to neglect the dynamics of the atomic nuclei, treating
them as fixed in space, and in fact arranged in a regular crystal lattice. The second
approximation will be to treat the Coulomb interaction between the electrons only
through an average. We shall see that, using these approximations, the basic ideas
can be formulated and important physical phenomena can be explained, in particu-
lar, the difference between metals and insulators. Our aim here will be not so much
a detailed quantitative understanding of the properties of a given material as a qual-
itative approach to the phenomena that can be used to classify the various types of
behaviour. With a more detailed knowledge of a great many solids, these basic ideas
can be developed to produce quantitative results in many cases, provided that we
have recourse to a sufficient capacity for numerical computation. However, even a
brief outline of the arguments and methods involved in this kind of approach goes
well beyond the scope of the book.
In later chapters, we shall consider physical phenomena involving highly novel
behaviour that cannot be explained within the framework of these approximations.
We shall see in particular that the atomic motions must be taken into account in
order to explain the resistivity of metals, or the appearance of superconductivity
in certain metals. Likewise, magnetism arises in materials as a direct result of the
strong local Coulomb interactions between the electrons.
1.1 General Hamiltonian for a Solid
Let us construct the general Hamiltonian for a solid, and hence observe the impossi-
bility of solving the corresponding Schrödinger equation. We thus discover the need
to make approximations.
An arbitrary solid can be treated as made up of a very large number Nn of atomic
nuclei and Ne of electrons of mass m0 . To simplify the notation, we assume that
the nuclei are all identical, with mass M and charge Ze. Since the solid as a whole
is electrically neutral, it follows that Ne = ZNn . In principle, the arguments in this
section can be generalised to the case where there are several species of atomic nu-
clei without great difficulty. The Hamiltonian for this system is thus a sum of five
terms: the kinetic energies of the nuclei and the electrons, and the Coulomb interac-
tions between the nuclei, between electrons and nuclei, and between the electrons.
It therefore has the form:
1 1 Z2
Nn Ne Nn
H= P2i + p2j + Vc (Ri − Rj )
2M 2m0 2
i=1 j=1 i,j=1,i=j

Nn
Ne
1
Ne
−Z Vc (rj − Ri ) + Vc (ri − rj ) . (1.1)
2
i=1 j=1 i,j=1,i=j
Here the Ri are the positions of the nuclei, the rj are the positions of the electrons,
and the momentum operators are given by P = −ih̄∇R and p = −ih̄∇r for the nuclei
and electrons, respectively. Finally, the Coulomb potential has the well known form
e2
Vc (r) = . (1.2)
4π ε0 |r|
1.1.1 The Need for Approximations
If we could solve the Schrödinger equation with the Hamiltonian (1.1), we could in
principle predict all the properties of the system, e.g., its structure, including the ar-
rangement of all the atomic nuclei, its thermodynamic behaviour, and its electrical or
thermal transport properties. However, such a full solution is clearly impossible. In a
macroscopic solid, there are something like 1023 nuclei, and hence a similar number
of electrons. Due to the Coulomb interactions, the Hamiltonian cannot be separated
into distinct terms only affecting a limited number of independent particles. We
must therefore solve an equation in something of the order of 1023 variables, which
is just not feasible. As a comparison, recall that the problem can be solved for the
hydrogen atom because the corresponding Schrödinger equation expressed in polar
coordinates (r, θ , φ) separates into two differential equations. Even better, they each
have analytic solutions, so it is possible in that case to obtain analytic expressions
for the wave function [2, Chap. 11]. No such miracle is to be expected here. Indeed,
there is not even an analytic solution for the simplest problem after the hydrogen
atom, namely the helium atom (with one nucleus and two electrons).
To make any progress here, we are thus compelled to make approximations, mo-
tivated if possible by physical considerations. The first thing to note is that the nuclei
are generally 104 –105 times more massive than the electrons, except in the case of
the lightest elements. For this reason, they will have much slower dynamics than the
electrons, and their kinetic energy can thus be neglected to a first approximation.
Experience confirms this idea. Indeed, crystallographic studies show that, in most
solids, the atoms make up a static, i.e., time-independent, arrangement, with at most
very small oscillations about their equilibrium positions.
In monatomic solids, the only exception to this rule is helium. Owing to its relatively small
atomic mass and the weakness of the interatomic forces, the zero point quantum motion of the
atoms is large enough to prevent any solidification at atmospheric pressure, and helium remains
liquid even right down to the lowest temperatures. However, a moderate pressure can induce
solidification, but there remain significant effects due to zero point motion even in this solid
state. Quite remarkable differences between the properties of the isotopes 3 He (a fermion of
spin 1/2) and 4 He (a boson of spin 0) are then observed.
In molecular solids, certain motions persist and play a role in the physical properties. This
happens in particular in C60 compounds, for example. This molecule has the shape of a football
(see Chap. 2) and can rotate or distort, stretching into the shape of a rugby ball, for example.
In this chapter, we shall therefore omit the kinetic energy of the nuclei and assume
that they are held fixed at points Ri , determined experimentally. In this case, the
Coulomb interaction between the nuclei gives a constant contribution to the energy
and can thus be dropped. The Hamiltonian (1.1) then simplifies to
1
Nn
1
Ne Ne Ne
He = p2j − Z Vc (rj − Ri ) + Vc (ri − rj ) , (1.3)
2m0 2
j=1 i=1 j=1 i,j=1,i=j
and the only quantum operators remaining are the electron position and momentum
operators rj and pj , respectively (hence the subscript e on the Hamiltonian), since
the positions of the nuclei are now fixed parameters. In the following, we shall be
concerned only with crystalline solids, so the Rj will be arranged in a regular way
on the crystal lattice.
But the problems raised by the Hamiltonian (1.3) are almost as intractable as
those raised by the original problem. There are of the order of 1023 electron vari-
ables, and owing to the Coulomb interaction between the electrons, He still cannot
be separated into a sum of terms involving only a small number of electrons.
We could clearly separate the electron variables by simply neglecting the
Coulomb interaction term between the electrons. Under this assumption, consider
the situation in which an extra electron is added to the globally neutral solid con-
taining Ne electrons. By neglecting the interaction term between the electrons and
keeping the term in −ZVc in (1.3), this extra electron will be subjected to the poten-
tial of the enormous total positive charge ZNn e, a situation which obviously bears
no resemblance to the actual physical situation.
We can only separate the electron variables and (partially) solve this problem by
making a more careful approximation, wherein the electron–electron interactions
are treated by means of an average. A given electron is no longer subjected to a po-
tential depending on the instantaneous positions of all the other electrons, but rather
to a potential corresponding to the average distribution of the electrons. Such an
approximation raises no major difficulties if the electrons are not too close together,
but it becomes critical if they may approach each other. It is well justified when the
kinetic energies of the electrons are greater than their average potential energy, but
very often this criterion can only be accepted with hindsight, when the results of
simulations or experiment provide some a posteriori justification.
It is interesting to note here that exact solutions of the N-body problem of the kind posed by
the Hamiltonian (1.3) with N = Ne do in fact exist for one-dimensional systems. Restriction to
one dimension may seem to make the discussion somewhat academic. However, it should be
borne in mind that there are in fact many experimental systems that come rather close to this
situation: quasi-1D conductors or magnetic compounds, i.e., with highly anistropic structures,
1D semiconducting nanostructures called quantum wires, carbon nanotubes, and many others.
The exact solutions to the N-body problem in the 1D model often provide valuable information
about the physics of these materials. Note, however, that they involve difficulties of their own.
The underlying idea of the method for obtaining almost all the exact solutions currently known
was developed by Hans Bethe in 1931, along with the first applications. The first results that
could be directly compared with certain experimental observations were obtained in the 1960s,
and this subject is still one of the main themes of theoretical physics today.
For our present purposes, averaging the electron–electron interactions means replac-
ing (1.3) by the separable Hamiltonian
1 n
e N e N N
He = p2j + Vat (rj − Ri )
2m0
j=1 i=1 j=1

Ne
p2j
Nn
Ne
= + Vat (rj − Ri ) = Hj , (1.4)
2m0
j=1 i=1 j=1
where we have replaced the Coulomb electron–nucleus and electron–electron inter-

action potentials affecting a given electron by the single term Vat , or atomic poten-
tial, so that it becomes possible to separate He into a sum of Ne independent terms.
In this case, an extra electron added to the solid will ‘feel’ a total charge of zero,
since the positive nuclear charges and negative electron charges now balance one
another.
What can be said about the form of Vat ? If the extra electron is indeed to feel
a globally zero charge, this potential must correspond, via Poisson’s equation, to a
zero total charge. Using Fourier transforms, it can be shown that any potential falling
off faster than 1/r at large distances satisfies this constraint. To be more precise, note
that Vat represents the joint effect of the nuclei and the average distribution of the
Z electrons associated with it. This gives rise to an exponential decrease of Vat at
large distances. On the other hand, when an electron is located very close to the
nucleus, it will essentially feel the full nuclear charge. At short distances we will
thus find Vat (r) = −Ze2 /4π ε0 r. The full form for Vat is thus strongly dependent
on the relevant element, but also on the solid in which it finds itself . We shall not
attempt here to discuss the exact form, since it will have little importance in the
following. The only relevant point is that it is attractive at short range and thus
allows bound states. We shall also assume that it is invariant under rotations, i.e.,
Vat (r) = Vat (|r|).
To illustrate how the precise form of Vat might be determined in a specific example, consider
the case of a single atom in the context of the simplest approximation, where we assume that
a given electron interacts with the other electrons only through their average density. We thus
have

e2 Z ρ(r )
Vat (r) = − + d3 r , (1.5)
4π ε0 r |r − r |
where the first term represents the interaction with the nucleus and the second the interaction
with the average electron density ρ(r ) at the point r . We must now solve the Schrödinger
equation with the potential (1.5). This clearly brings out the complexity of the problem. The
density ρ is given by

ρ(r) = |ψk (r)|2 , (1.6)
k
where the wave functions ψk are themselves determined by the Schrödinger equation, and the
sum is over the occupied states of the atom. We thus do not know the form of the potential a
priori. It must be determined in some self-consistent way. For example, we can use an iterative
method. We begin with a first approximation for Vat , by calculating the ψk0 when there is no
electron–electron interaction, then solve the Schrödinger equation with ρ given by (1.6) for the
wave functions ψk0 . This leads to another set of eigenfunctions ψk1 , and these can be used with
(1.5) and (1.6) to obtain a new Schrödinger equation. This process continues until it converges
on a solution to the required accuracy. A procedure of this kind clearly involves a considerable
degree of numerical computation, but one that is perfectly accessible to today’s computers, at
least in the case of a single isolated atom. These calculations nevertheless become much more
time consuming if we have to treat large solids.
The approximation (1.5) is called the Hartree approximation. It does not take into account
the electron spins, and does not satisfy the Pauli principle (see Sect. 1.1.2). The analogous
approximation taking into account the Pauli principle, known as the Hartree–Fock approxima-
tion, requires even greater computation time. In solid state physics today, the most efficient
method, and hence the one most commonly used, is based on the density functional theory
(DFT). All such methods involve large amounts of computation time, and we shall not try to
describe them here.
1.1.2 Wave Functions and Pauli Exclusion Principle
In the second line of (1.4), we obtained a separable form of He , written as a sum

of identical terms, each one of which refers only to the coordinates and momen-
tum of one electron. The Hamiltonian (1.4) can thus be treated quite simply if we
can determine the eigenstates ψk (r) and energy eigenvalues Ek of the one-electron
Schrödinger equation corresponding to the general term
p2 n N
H1 = + Vat (r − Rl ) . (1.7)
2m0
l=1
A general solution of the Schrödinger equation (1.4) for the Ne electrons has the
form (see, for example, any textbook on statistical physics; this is the form of the
wave function used in the Hartree approximation)
Ψ (r1 , r2 , . . . , rNe ) = ψk1 (r1 )ψk2 (r2 ) . . . ψkNe (rNe ) , (1.8)
with energy eigenvalue

Ne
E= Eki . (1.9)
i=1
However, the function in (1.8) does not satisfy the Pauli exclusion principle. Accord-
ing to this, since electrons are fermions, their wave functions must be antisymmetric
when the positions of any two electrons are exchanged. Although this is not the case
in (1.8), the problem can be corrected by replacing (1.8) by the Slater determinant
(the wave function used in the Hartree–Fock approximation [2]):

ψk (r1 ) ψk (r2 ) . .
1 1

1 ψk2 (r1 ) . . .

Ψ (r1 , r2 , . . . , rNe ) = √ . (1.10)
Ne ! . . . ψkNe −1 (rNe )

ψkN (r1 ) . ψkN (rNe −1 ) ψkNe (rNe )
e e
This function corresponds to the same energy eigenvalue (1.9) as the not yet anti-
symmetrised function in (1.8). Note that the determinant structure of this function is
what ensures the antisymmetry of the wave functions and hence satisfaction of the
Pauli exclusion principle. Indeed, if two of the quantum numbers kj were the same,
the corresponding rows of the determinant would be identical, hence (1.10) would
not be an admissible wave function.
1.2 Energy Bands
The next step in the solution of this problem, to determine the electron states of
the macroscopic solid, is therefore to find the one-particle wave functions ψk (r). Of
course, it would be impossible to solve the Schrödinger equation for the Hamiltonian
(1.7) without knowing the exact form of the potential Vat . However, what interests
us here is to understand what happens to the electron states of the atoms when they
are put together to form a solid. The situation will be radically different depending
on whether we consider the states corresponding to electrons that are strongly bound
to the nucleus or electrons that are only weakly bound. These two cases correspond
respectively to strong and weak potentials Vat , respectively. In this book, we shall
consider these two simple limits, dealing with the first in the remainder of this chap-
ter and the second in Chap. 3. When the electrons are strongly bound, we speak of
the tight-binding approximation. Note that this refers to the binding of the electrons
to the nucleus, not to chemical bonds between the atoms.
1.2.1 Tight-Binding Approximation
In this context it is useful to consider the solid as a chemist might view it, as an
ensemble of rather isolated atoms, interacting only weakly together. A quantum
treatment begins by solving the Schrödinger equation for a single atom:
1.2 Energy Bands 9

p2
+ Vat (r) χn (r) = En χn (r) . (1.11)
2m0
Consider the case where Vat corresponds to the potential of a neutral ‘atom’ whose
lowest eigenstates with energies E1 , E2 , . . . , En , . . . , Ev are occupied by the Z elec-
trons (see Fig. 1.1a). To fix ideas, we consider only those electron states of the solid
resulting from the last occupied electron state of the neutral atom (the valence state),
with energy Ev and wave function χv (see Fig. 1.1b).
Let χv (r − Rn ) be the eigenfunctions of degenerate energy Ev corresponding to
the atomic states centered on the various sites Rn . The tight-binding hypothesis
assumes that the eigenfunctions ψk (r) of the system of Nn atomic potentials centered
on Rn have energy eigenvalues Ek that differ only slightly from Ev , but are well
removed from the energies Ev−1 and Ee of the closest atomic states of the neutral
atom.
In this case, it seems reasonable to assume (and it can be checked after the event)
that the functions ψk (r) should be expressible as linear combinations of the atomic
orbitals χv (r − Rn ) (or LCAOs):
1 n N
ψk (r) = √ ak,l χv (r − Rl ) ,
Nn l=1
corresponding to the quantum states
1 n N
|ψk = √ ak,l |Rl . (1.12)
Nn l=1
Ee
Ev
χv (r)
Ev−1
r
O
E2
Vat(r)
E1
(a) (b)
Fig. 1.1 (a) The eigenstates of the Hamiltonian (1.11) with energies E1 , E2 , . . . , Ev are occupied
by the Z electrons of the atom. The level Ee is the energy of the first excited state. (b) Attractive
atomic potential Vat centered at r = 0, and schematic representation of the wave function χv (r)
corresponding to the energy Ev
Here, the k are the quantum numbers of the eigenstates of the solid, while the ak,l
are the coefficients of the linear combination, which remain to be determined from
the Schrödinger equation. In (1.12), we have introduced the ket |Rl associated with
the wave function χv (r − Rl ).
In the rest of this book, the exact form of χv (r) will be of little importance.
Our aim will be only to describe the method used to determine eigenfunctions and
eigenvalues for simple situations to begin with, then for more realistic cases. We
will study the case of a non-degenerate state, for which χv will be real and even,
i.e., χv (r) = χv (−r).
Schrödinger Equation in the LCAO Approximation
We now seek the eigenfunctions and eigenvalues of the Hamiltonian (1.7) in the
context of the tight-binding approximation. To do this, we first transform the
Schrödinger equation, inserting the form (1.12) of the wave function in H1 |ψk =
Ek |ψk to obtain
Nn
Nn
ak,l H1 |Rl = ak,l Ek |Rl . (1.13)
l=1 l=1
We now separate the potential into two parts, one due to the atom at site l and the
other due to all the other sites. Combining (1.7) and (1.11), this leads to
H1 |Rl = Ev |Rl + Vl (r) |Rl , (1.14)
where we have defined

Nn
Vl (r) = Vat (r − Rm ) . (1.15)
m=1,m=l
Using (1.14), we may rewrite (1.13) in the form

Nn
Nn
ak,l Ev |Rl + Vl (r) |Rl = ak,l Ek |Rl . (1.16)
l=1 l=1
In order to simplify the calculation, we begin by assuming that the |Rl form an
orthonormal set, i.e.,
Rn |Rl = δn,l , (1.17)
where δ is the Kronecker symbol (δl,m = 1 if l = m, δl,m = 0 otherwise). Multiplying

(1.16) on the left by Rn | and introducing the transfer (or hopping) integrals
−tn,l = Rn | Vl (r) |Rl , (1.18)

1.2 Energy Bands 11
we obtain

Nn
− tn,l ak,l = (Ek − Ev ) ak,n . (1.19)
l=1
Note that the choice of sign in (1.18) gives tn,l > 0 in most cases. The Nn equations
(1.19) (the value of n can vary from 1 to Nn ) constitute a homogeneous linear system
in the Nn variables ak,n . Such a system has a solution only if Ek − Ev is an eigenvalue
of the matrix with entries −tn,l . We have thus transformed the initial Schrödinger
equation into a matrix eigenvalue problem.
1.2.2 Linear Atomic Chain
The above argument applies in practice to any type of solid. However, many prop-
erties of electron states in a solid can already be understood through the simplest
possible example, namely a linear chain of atoms. This example may seem rather
academic, but the generalisation to more realistic cases will be greatly simplified by
a prior discussion of this case, which will allow us to simplify the notation. The 1D
situation is clearly also of some practical interest for understanding the properties
of the quasi-1D conductors mentioned earlier.
In the linear chain, the atomic positions are specified by the vectors Rm = Rm x,
where x is a fixed unit vector in the x direction and the Rm are scalars given by
Rm = ma . (1.20)
The distance a between nearest neighbour atoms (see Fig. 1.2) is usually called the
lattice constant, for reasons to be discussed in more detail in Chap. 2.
In the special case of the linear chain, it is easy to give a graphical representation
of the total potential in (1.7), which is the sum of the atomic potentials centered
on the sites of the chain (see Fig. 1.3a). The potential Vl (r) deduced by simply
suppressing the term Vat (r − Rl ) is shown in Fig. 1.3b. The wave functions centered
on n and l, and the potential Vl (r) used to calculate the transfer integral tn,l in (1.18),
are shown in Fig. 1.3c.
a. Periodic Boundary Conditions and Bloch’s Theorem
A general analysis of (1.19) is neither straightforward nor useful without knowing

the coefficients tn,l . Even if all these coefficients are known, the solutions remain
1 2 a Nn
Fig. 1.2 Linear chain of atoms (dots) with lattice constant a

1 l Nn
(a)
1 l Nn
(b)
| Rn |Rl
1 Nn
(c)
n l
Fig. 1.3 (a) Periodic potential of (1.7). (b) Potential Vl (r). (c) Elements used to calculate tn,l with
the help of (1.18)
complicated for a finite chain of Nn sites, due to effects arising from the endpoints of
the chain. We shall therefore slightly modify the geometry of our chain by forming
a ring, as shown in Fig. 1.4a, in such a way that the site Nn ends up at distance a
from site 1. In general, one would expect this different geometry to lead to different
properties as compared with the finite chain. However, in the thermodynamic limit
where Nn is very large, which is practically the only one we shall be concerned
with, the angle between consecutive bonds tend to zero and the two systems become
equivalent. The chain is then invariant under translation through whole multiples of
the constant a.
This approach is equivalent to imposing periodic boundary conditions on the
finite chain (see Fig. 1.4b). These Born–von Kárman conditions [3, Chap. 7] are
used to make the system periodic by extending the wave functions in such a way
that ψk (x + Nn a) ≡ ψk (x), or
ak,n+Nn = ak,n . (1.21)
The translation invariance of the chain under shifts through integer multiples of a,
and the symmetry properties of Vat and χv chosen above imply that the coefficients
tn,l depend only on the distance between the sites n and l :
tn,l = t|n−l| . (1.22)
Nn 1
1 1 2 Nn Nn + 1
2
3
(a) (b)
Fig. 1.4 (a) Ring of Nn atoms. (b) Periodic boundary conditions for the chain
1.2 Energy Bands 13
Formally, these coefficients are only nonzero for sites that are no further apart than
(Nn − 1)a, but we shall see hereafter that they decrease quickly with distance, so that
only a few terms will be retained anyway. In the case of the ring, or the chain with
periodic boundary conditions, we can rewrite (1.19) by singling out the site n and
collecting together terms associated with equidistant neighbours:
/2−1
Nn
−t0 ak,n − tm (ak,n+m + ak,n−m ) = (Ek − Ev )ak,n ,
m=1
which can be written, after replacing the dummy index m by l

/2−1
Nn
− tl (ak,n+l + ak,n−l ) = (Ek − Ev + t0 )ak,n . (1.23)
l=1
Before trying to solve (1.23), note that (for the ring, or for the chain with peri-
odic boundary conditions) electrons separated by a distance a feel the same crystal
potential, so the Hamiltonian Ĥ1 in (1.7) is invariant under translation by a. For
this kind of periodic potential, Bloch’s theorem (see the box below) shows that the
eigenfunctions of Ĥ1 , known as Bloch functions, have the general form
ψk (x) = eikx uk (x) , with uk (x + a) = uk (x) , (1.24)
whence the function uk (x) is itself periodic.

If the form (1.12) for ψk (x) is to satisfy (1.24), we therefore require
1 n N
uk (x) = √ ak,l e−ikx χv (x − Rl )
Nn l=1
to be periodic, i.e.,
1 n N
uk (x + a) = √ ak,l e−ika e−ikx χv (x − Rl−1 ) ≡ uk (x) ,
Nn l=1
whence
ak,l e−ika = ak,l−1 ,
which implies
ak,l = ak,0 eikla .
The periodic boundary condition (1.21) then implies that eikNn a = 1, which leads to
the following quantisation condition for k :
2π m
k= , m integer . (1.25)
Nn a
Further, normalisation of the wave function ψk leads to ak,0 = 1, and the ak,n can
thus be expressed in the general form
ak,n = eikRn = eikna . (1.26)
Bloch’s Theorem
The potential felt by the electron is periodic with V(x + a) = V(x). Let T̂ be the elementary
translation operator through a distance a, such that
T̂ f (x) = f (x + a) ,
for any function f (x). This operator commutes with the Hamiltonian Ĥ1 , so there is a common
basis of eigenstates. The eigenstates of Ĥ1 can thus be sought amongst those of T̂. Since T̂ is
a unitary operator, the eigenvalue associated with any eigenstate of T̂ has unit modulus. The
eigenstates of T̂ can thus be characterised by an index k which determines their eigenvalue
T̂ψk = eika ψk .
For a state ψk to be an eigenstate of Ĥ1 , it must satisfy the periodic boundary conditions. We
must therefore have
ψk (x + Nn a) = (T̂)Nn ψk (x) = eikNn a ψk (x) = ψk (x) .
This leads to the quantisation condition (1.26) for k. Setting uk (x) = e−ikx ψk (x), we find that
uk (x + a) = e−ik(x+a) ψk (x + a) = e−ikx ψk (x) = uk (x) .
Hence the function ψk (x) does indeed have the form (1.24) of a Bloch function.
b. Solutions Under the Nearest-Neighbour Approximation
Equation (1.23) will receive a general solution below. Owing to the fast exponential
decline of the wave functions χv with distance, it generally suffices to consider only
the nearest-neighbour hopping integrals (see Fig. 1.5):
tn = 0 if |n| > 1 . (1.27)
t0 Rn t1 Rn Rn+1
n n+1 n n+1
Fig. 1.5 Potentials and wave functions used to calculate the hopping integrals t0 and t1 in the
nearest-neighbour approximation
1.2 Energy Bands 15
Equation (1.23) then reduces to
−t1 (ak,n+1 + ak,n−1 ) = (Ek − Ev + t0 )ak,n . (1.28)
The condition (1.26) on the ak,n can be used to determine the energy eigenvalue
associated with the states |ψk from (1.28):
Ek = Ev − t0 − 2t1 cos ka . (1.29)
The energy eigenvalues (1.29) are shown graphically in Fig. 1.6 as a function of
the quantisation index k. There are several points to note about this result which
will also apply in more complex situations, in particular, in the case of dimensions
higher than one:
1. The energy levels (1.29) are shifted by t0 from the discrete atomic level Ev , and
also broadened into an allowed energy band, with lower bound at Ev − t0 − 2t1
and maximum at Ev − t0 + 2t1 , so that the total width of this band is 4t1 . The
function Ek of k describes the structure of the energy band (hence the term ‘band
structure’). The LCAO approximation is only valid if all the states in the band
lie below the first excited state Ee of the atomic potential, i.e., we must have
2t1 Ee − Ev .
2. The state of energy Ek is given by
1 n N
|ψk = √ eikla |Rl . (1.30)
Nn l=1
We now note that, if we replace

k +2π/a, we obtain the same wave function,
k by
i.e., the states |ψk and ψk+2π/a are in fact the same, and k is only defined
E
Ev − t0 + 2t1
0 k
Ev k
E v − t0 4t 1
Ev − t0 − 2t1
− π /a π /a
Fig. 1.6 Tight-binding approximation. Energy band arising from the atomic valence level for the
linear atomic chain. To represent all the energy states, it suffices to restrict to the first Brillouin
zone (−π/a, π/a)
modulo 2π/a. In practice, to represent this periodic function, it is convenient to

choose k in the interval (for Nn even)
π π
− <k≤ . (1.31)
a a
This interval is known as the first Brillouin zone. Clearly, other choices are also
possible. The motivation for this particular choice will be spelt out in more detail
in Chap. 3. A graphical representation of the band structure in the first Brillouin
zone is shown in Fig. 1.6.
3. The form (1.26) for the coefficients ak,l looks like a plane wave, and is reminis-
cent of the wave function of a free particle, subject to periodic boundary condi-
tions. This is illustrated in Fig. 1.7, which shows the general form of the real part
of ψk , and the corresponding limiting cases at k = 0 and k = π/a. The first case
corresponds to a situation in chemistry where the orbitals are bonding, while the
second corresponds to antibonding orbitals. We may therefore wonder whether,
as in the case of free particles, there is a direct relation (p = h̄k) between the
wave vector k and the momentum p. The same question is raised for any Bloch
function. We shall see later that this relation is no longer valid, but that some
features of the behaviour of plane waves in vacuum will survive. For this reason,
the quantity h̄k is called the crystal momentum or quasi-momentum.
4. We should ask whether the states |ψk form a complete set. Owing to the
orthonormalisation relation (1.17), the |ψk form an orthonormal set:
ψk |ψk = δk ,k ,
where δ is taken modulo 2π/a. Furthermore, we have one orbital per atom, so
the Hilbert space has dimension Nn . Finally, the quantisation rule (1.25) shows
that, in the interval (1.31), there are also Nn states, so we find that the |ψk do
indeed constitute a complete orthonormal basis for the Hilbert space.
π /k
k
(a)
k=0 Bonding
orbitals
(b)
k = π /a Antibonding
orbitals
(c)
Fig. 1.7 Real part of the wave function ψk . (a) General case. (b) For k = 0. (c) For k = π/a
1.2 Energy Bands 17
c. Density of States
When calculating thermodynamic properties, transport properties such as electri-

cal and thermal conductivity, and many other physical quantities, one often has to
evaluate sums of the form

Nn a
G= g(Ek ) = g(Ek )dk , (1.32)
2π BZ1
BZ1
where g is an arbitrary function, and the sum is generally over the whole of the
first Brillouin zone. To obtain the second equality, we used the fact that, in the
thermodynamic limit, the allowed values of k, with spacing 2π/Nn a, almost form a
continuum, which justifies replacing the sum by an integral over k. The important
point in (1.32) is that the integrand only depends on k through the energy Ek . It is
thus convenient to be able to rewrite (1.32) in the form

G = Nn dE g(E)D(E) , (1.33)
where the density of states D(E) is such that Nn D(E)dE represents the number of
states with energies between E and E + dE for the solid comprising Nn atoms. While
it is already convenient for the 1D case, this ploy will become even more useful in
the 2D and 3D situations.
Note that there are various ki values for which E(ki ) = E. For each ki value the in-
terval of energy dE corresponds to an interval of k values such that
dE = (dE/dki )dki . The number of corresponding states close to ki is therefore
Nn a Nn a dE
|dki | = , (1.34)
2π 2π |dE/dki |
so the total density of states of energy E per atom is

a 1
D(E) = . (1.35)
2π |Ek i |
ki
Let us now consider the linear chain with Ek given by (1.29). If |E − Ev + t0 | > 2|t1 |,
there is no eigenstate and the density of states is zero:
D(E) = 0 if |E − Ev + t0 | > 2|t1 | . (1.36)
If this is not the case, there are two symmetric solutions for ki such that
E − Ev + t 0
cos ki a = − .
2t1
Ev−t0+2t1
k |t1|D(E)
Ev−t0
−π /a π/a 0.2 0.4 0.6
Ev−t0−2t1
Fig. 1.8 Dispersion relation (left) and density of states (right) of the linear atomic chain in the
LCAO approximation

As |Ek i | = 2a|t1 sin ki a| = 2a|t1 | 1 − cos2 ki a, it follows from (1.35) that
1 1
D(E) = if |E − Ev + t0 | < 2|t1 | . (1.37)
2π |t1 | 1 − (E − E + t )2 /4t2
v 0 1
The density of states is plotted versus the energy in Fig. 1.8. Note the square root
divergences at the band edges. These divergences arise for the same reason as those
in the free electron gas model where D(E) ∝ E d/2−1 , with d the space dimension
(see Chap. 3).
Generalisations
In the last section, we made two approximations. The first was the nearest-neighbour approxi-
mation (1.27), in which we neglect all hopping integrals linking the given site to those beyond
its immediate neighbours. The second assumed orthogonality of the orbitals at different sites,
expressed by (1.17). If instead we retain only the orthogonality of the orbitals but drop the
nearest-neighbour approximation, using the full equation (1.23) rather than (1.28), it is easy to
see that the wave functions are unchanged, but that their eigenvalues become
/2−1
Nn
Ek = Ev − t0 − 2 tl cos kla − tNn /2 eikNn a/2 . (1.38)
l=1
Here we have considered the case where Nn is even. Regarding the summation range, we ought
to make the following comment. Owing to the periodicity of the ring arrangement, we have
tl+Nn = tl , so we may write tNn −1 = t1 , for example. Furthermore, the maximal distance be-
tween any two sites is Nn a/2, but this distance appears only once in the sum, whereas all the
other distances appear twice, once when starting from any point and moving to the right, and
once when moving to the left. In the sum over l, the upper bound is thus correct. Note that the
quantisation condition for the k does indeed lead to a real value for the last term. When Nn is
odd, the upper bound of the sum is replaced by (Nn − 1)/2 and the last term in (1.38) does not
arise. Note finally that, at large distances, the hopping integrals tl decrease exponentially with
distance, and the sum (1.38) converges rapidly.
1.2 Energy Bands 19
In the thermodynamic limit Nn → ∞, the sum in (1.38) can be replaced by
∞

Ek = Ev − t0 − 2 tl cos kla . (1.39)
l=1
In the usual case where the hopping integrals decrease quickly with the distance l, the general
form of Ek does not change from what was obtained with the nearest-neighbour approximation,
and in particular retains a global maximum and a global minimum, without intervening local
extrema. This gives rise to a similar density of states curve to the one shown in Fig. 1.8, with
just two singularities at the band edges. However, we can envisage some (rather exceptional)
cases in which the tl are such that Ek has other, local extrema. In this case, the density of states
would be changed more radically by the existence of further singularities, also involving square
root divergences, in the middle of the band, and more precisely, at the locations of the local
extrema.
The situation becomes somewhat more involved if we drop the orthonormalisation condi-
tion (1.17). It is replaced by the general expression
Rn |Rl = αl−n = αn−l , (1.40)
where the fact that the coefficients αl−n (called overlap integrals) depend only on the distance
between the two sites results from the same symmetry properties as for the tl−n . We still con-
sider normalised wave functions, which corresponds to α0 = 1.
To calculate eigenfunctions and eigenvalues, we return to (1.16), which has changed slightly
to become
Nn
Nn
ak,l Vl (r) |Rl = (Ek − Ev ) ak,l |Rl . (1.41)
l=1 l=1
We multiply on the left by Rn | to obtain, after making the change of variable l → l + n,

Nn
Nn
− ak,l+n tl = (Ek − Ev ) ak,l+n αl . (1.42)
l=1 l=1
Thanks to the periodic boundary conditions on the ak,l and the periodicity properties of the tl
and αl , there is no need to change the summation limits. The coefficients ak,l have the same
form as before, viz., ak,l = exp(ikla), and after cancelling a factor of exp(ikna) on each side of
(1.42), we finally obtain the eigenvalues:

t0 + 2 ∞
l=1 tl cos kla
Ek = E v − ∞ , (1.43)
1 + 2 l=1 αl cos kla
where we have already taken the thermodynamic limit.
Note that the result (1.43) is a periodic function of k of period 2π/a, as in the result (1.39)
for the orthogonal case. Expanding in a Fourier series, (1.43) can thus be written in the same
form as (1.39), modifying only the coefficients by tl → tl . The wave functions take the same
form as in the simpler case of (1.30). A similar calculation to the one leading to (1.43) gives
the orthogonality relation
∞

ψk |ψk = δk,k 1 + 2 αl cos kla . (1.44)
l=1
The |ψk are thus orthogonal but not normalised (although they could easily be normalised,
of course). Note that the norm of any wave function must be positive, which guarantees that
the denominator in (1.43) is always positive.
1.2.3 Energy Bands Associated with Tightly Bound Atomic Levels
So far we have been considering the case of the atomic valence level Ev of Fig. 1.1a.
Clearly, the same approach can be used for the deeper levels E1 , E2 , . . . , En , and will
lead to an analogous situation, i.e., each of these atomic levels will lead to an energy
band with width determined by an integral like t1 . Note that this hopping integral is
given by (see Fig. 1.5)

t1 = χm∗ (r − R0 )Vat (r − R0 )χm (r − R1 )d3 r , (1.45)
where we have taken here into account the realistic three dimensional atomic wave
functions χm∗ (r−R0 ) and χm (r−R1 ) for the given level, centered on two neighbour-
ing atoms in the chain. Now the atomic orbital wave functions have the property that
their spatial extents are narrower for more tightly bound electron levels. This can be
seen in Fig. 1.9a, which shows the functions rχm (r) corresponding to the various
states 1s, 2s, 2p, etc., of sodium, on two sites separated by a distance corresponding
to the interatomic distance in metallic sodium.
(a) (b)
Sodium
Na 3s
1s 1s22s22p63s1 1s
2p
2s
2p 2p
3s 3s
1s
2s 2s
r0 r
Fig. 1.9 (a) The product |r − Ri |χ(|r − Ri |) is shown for the wave functions of two nearest-
neighbour atoms in sodium (Na). This choice of representations magnifies the degree of overlap
between the orbitals. It increases continuously for the states 1s, 2s, 2p, and 3s (Adapted from [4],
p. 177). (b) Energy bands arising from the atomic levels 1s, 2s, 2p, and 3s for Na as a function of
the interatomic distance r
1.3 Summary 21
The figure shows clearly that, in the relation (1.45), the weight of the product
χm∗ (r − R0 )χm (r − R1 ) is such that the values of t1 are smaller for deeper energy
levels. It thus follows that the tight-binding approximation leads to energy bands in
the solid that get broader as the energy of the relevant atomic level gets higher.
This is depicted schematically in Fig. 1.9b, which shows the energy bands re-
sulting from the atomic levels 1s, 2s, 2p, and 3s for a hypothetical solid made up of
sodium atoms arranged with a spacing of r. As expected from the behaviour of t1
as a function of r, the bands broaden when the distance is reduced. The deep levels
give rise to extremely narrow bands (a few micro electronvolts or meV), while the
valence levels give rise to broad bands (in the electronvolt range). In the case of nat-
ural Na, with spacing r0 in Fig. 1.9b, the width of the band arising from the 3s level
is of the order of the distance between the 2p and 3s levels, and the tight-binding
approximation is no longer applicable. We shall see in Chap. 3 how the structure
of this band can then be determined using a different approximation, wherein it is
assumed that these electron levels correspond to nearly free electrons.
1.3 Summary
Two subjects have been discussed in this chapter. To begin with, we described the
quantum mechanics of a solid comprising a number of particles of the order of the
Avogadro number. It is easy to write down a Hamiltonian which takes into account
the kinetic energies of the atomic nuclei and the electrons, together with their mutual
Coulomb interactions. However, we soon realised that this problem would prove
insoluble in its full generality. A more practical and realistic approach here is to
begin by making two approximations:
1. We assumed that the atomic nuclei were static, which amounts to saying that the
nuclei sit at fixed positions in space. We also restricted to the case of crystalline
solids, characterised by a regular and periodic arrangement of the nuclei. This
is valid for a great many solids, but nevertheless rules out the interesting but
complex category of disordered solids.
2. The problem resulting from this approximation, which concerns only the dy-
namics of the electrons, is still too difficult to handle. We would have to account
for the interactions between some 1023 electrons, which would be strictly im-
possible. We thus further restricted to an average description of the Coulomb
interactions between the electrons, which is often quite adequate. In this case,
the problem involving a large number of electrons reduces to a large number of
problems involving just one electron, which can be dealt with by well known
methods of quantum mechanics and statistical physics.
We then studied the problem of a single electron in a crystalline solid, where it feels
a periodic potential. Instead of trying to investigate the general situation, for which
there are in any case no explicit solutions, we considered the so-called tight-binding
approximation, in which the electrons are tightly bound to the atoms. The wave
functions of an electron are then written as a linear combination of atomic orbitals

(LCAO). Having introduced this approximation for an arbitrary solid, we focused
on the specific case of a linear chain of atoms, for which we were able to carry out
the calculation exactly. The main result is that the influence of neighbouring atoms
in the solid leads to a broadening of the discrete atomic levels into a band of allowed
energies, characterised by a function E(k) which gives the energy of a quantum state
in terms of its crystal momentum k. For the linear chain of atoms, the function E(k)
is given by
Ek = Ev − t0 − 2t1 cos ka ,
where t0 and t1 are hopping integrals, characterising the possibility of transfer of
an electron from one atom to the neighbouring site. The spectrum arising from a
given atomic level is thus bounded but quasi-continuous. All the eigenstates can be
obtained for values of k lying in the interval (−π/a, π/a), which is called the first
Brillouin zone.
Chapter 2
Crystalline Solids: Diffraction
Contents
2.1 Crystal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1.1 Crystal Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2 Two-Dimensional Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.3 Three-Dimensional Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1.4 Beyond the Perfect Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.1 General Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.2 Reciprocal Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Determination of Crystal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 On the Bragg Diffraction Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 Diffraction Intensity and Basis of the Primitive Cell . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.3 X-Ray Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Answers to Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

24 2 Crystalline Solids: Diffraction
X-ray diffraction pattern for a C60 single crystal obtained with the experimental setup known as
a precession chamber. This directly visualises the Bragg spots corresponding to a plane of the
reciprocal lattice of the crystal (to solve Question 2.6 one has to take into account that the actual
diameter of the film in 12 cm). Image courtesy of Launois, P., Moret, R.: Laboratoire de Physique
des Solides. Orsay, France
2.1 Crystal Structures 25
Our main concern in this book is to describe the electronic properties of crystalline
solids. The existence of translation symmetries associated with such ordered crystal
structures leads to specific features in the electronic structure and to a specific rep-
resentation of the energy states in wave vector space. In Sect. 2.1, we explain how
a crystal structure can be described formally by defining a crystal lattice of non-
material points together with a repeated material motif, known as the basis. This
periodic structure of matter causes diffraction of electromagnetic waves, or equiv-
alently, of quantum particles. We shall see in Sect. 2.2 that these diffraction phe-
nomena lead to the notion of reciprocal lattice in wave vector space, and this will be
important later for characterising the electronic states of these solids. In Sect. 2.3, we
outline the experimental methods used to determine the crystal structures of solids
using diffraction methods.
2.1 Crystal Structures
It is usual to consider a crystal as a natural object with regular external geometric

features, as found for example in rock salt, diamonds, and so on. By the end of the
nineteenth century, the systematic study of the external shapes of such natural crys-
tals led scientists to conclude that this regularity of the outer faces must be due to
structural regularities on the microscopic scale. The molecules or atoms had to be
assembled in a periodic manner to make a crystal. In this chapter, we shall see how
to specify the arrangement of a crystal structure. This structure can be ascertained
experimentally either by direct observation, or by light diffraction (X-ray crystallog-
raphy). These experimental methods show that many solids actually have a crystal
structure, even when their outer surfaces do not give this impression. They are in fact
polycrystalline, i.e., made up of a host of small crystals, which may themselves be of
micrometric dimensions, but which are nevertheless of macroscopic size. We begin
in Sect. 2.1.1 by describing the notions of crystal lattice and unit cell. In Sect. 2.1.2,
these ideas will be exemplified in two dimensions, i.e., for planar crystal structures,
where the geometric representations are simpler. We then illustrate some simple 3D
systems in Sect. 2.1.3.
2.1.1 Crystal Lattice
A crystal is an arrangement of atoms or molecules that is invariant under transla-

tions in three space directions constituting a triad (a1 , a2 , a3 ). Many human con-
structions have ordered structures exhibiting such characteristics, especially in two
dimensions: wall paper and floor tiling often have periodic structures that can be
considered as 2D crystals. In these cases, the primitive material basis is repeated
through two translations. A molecular example, observed by a modern microscopic
method, is shown in Fig. 2.1. These are alkane molecules with chemical formula
Fig. 2.1 A monolayer of alkane C33 H68 deposited on graphite arranges itself into a 2D crystal. The
right-hand figure was obtained with twice the resolution to reveal details of the atomic structure.
The distance between molecules is 4.5 Å, and their length is 45 Å. Image courtesy of Cousty, J.:
SPCSI/CEA. Saclay, France
C33 H68 , deposited on a graphite surface. Note that, if we choose an origin O in

space, the crystal can be reconstructed in two steps:
• We produce a lattice of points obtained from O by all the translations
Rl = l1 a1 + l2 a2 + l3 a3 , (2.1)
where l1 , l2 , l3 are integers. This set of points (the lattice points or nodes) consti-
tutes the crystal lattice or Bravais lattice.
• In the second step, we arrange the material basis relative to these nodes in such a
way as to completely tile the space.
The Bravais lattice is thus a set of non-material points, whereas the atoms and
molecules make up the material basis, which we shall call the ‘basis’ from now on,
when no confusion is possible. For elementary solids, containing only one atomic
species, the atoms may coincide with the lattice nodes, since the basis then often
comprises a single atom. But as soon as the solid contains several atomic species,
such a situation is no longer possible.
Note that the lattice of points provides a way of defining a primitive unit cell for
the material basis, namely, as the smallest volume that can tile the space by applying
the translations Rl . If np is the density of lattice points, the volume of the primitive
cell is v = 1/np . The primitive cell and the triad (a1 , a2 , a3 ) are not unambiguously
defined, as can be seen from Fig. 2.2. The triad (a1 , a2 , a3 ) is often taken to be the
set of vectors that best reveals the symmetries of the lattice, e.g., (a1 , a2 ) rather than
(b1 , b2 ) for the square and rectangular planar lattices in Fig. 2.2. A primitive cell
b2
b2
b1
b1 a2 O a2
Wigner–Seitz a1
a1
(a) (b)
Fig. 2.2 (a) Square lattice. (b) Rectangular lattice. These can be specified by the two vectors
(a1 , a2 ) or the two vectors (b1 , b2 ). The vectors (a1 , a2 ) better reveal the symmetries of the lattice.
Four primitive cells are shown for the rectangular lattice, those specified by (a1 , a2 ) and (b1 , b2 ),
an arbitrary cell, and the Wigner–Seitz cell containing the lattice point O. The latter is obtained by
constructing the orthogonally bisecting planes of the four vectors (Oa1 , Oa2 , −Oa1 , −Oa2 )
tiling the space may be chosen with an arbitrary shape as displayed in Fig. 2.2(b),
but we often opt for the rhombohedral primitive cell specified by (a1 , a2 , a3 ).
Of particular importance is a primitive unit cell known as the Wigner–Seitz cell.
This is constructed in such a way that every point of the cell is closer to one lattice
point (for example, O) than to any other lattice point (see Fig. 2.2). It is bounded by
the orthogonally bisecting planes of the vectors Rl with origin the chosen node. For
an elementary solid, this volume constructed with one atom at the center represents
in some sense the region of influence of this atom.1
Note that the lattice points can be grouped together in parallel planes in infinitely
many different ways. These are called lattice planes. In two dimensions, they con-
stitute parallel rows (see Fig. 2.3). The lattice planes group together points that can
be obtained from one another by two of the translations Rl .
Fig. 2.3 Two families of lattice rows in the same Bravais lattice (oblique lattice in a 2D space)
2.1.2 Two-Dimensional Crystals
In the above, we illustrated the idea of a crystal by 2D representations. Since

real crystals are three-dimensional, these 2D representations may appear rather
1 We shall see that, in the reciprocal lattice to be defined hereafter, the corresponding unit cell will
specify the first Brillouin zone.
academic. However, important cases of 2D physics are becoming more and more
common today:
• The surface of any 3D crystal is obviously a 2D structure. One might think that
this would be that of the lattice plane corresponding to the infinite crystal. How-
ever, in many cases, the translational symmetry breaking associated with the ex-
istence of a surface leads to a significant modification of the surface structure.
We refer to this as surface reconstruction.
• Many 3D crystals occur in a layered form, with widely spaced molecular or
atomic layers. An example is graphite, or the high-Tc cuprate superconductors.
The structural and electronic properties of these materials are strongly affected
by the 2D nature of the material.
• Finally, novel fabrication methods devised in nanotechnology are now used to
deposit monomolecular layers (see Fig. 2.1), or even monatomic layers, on the
surfaces of crystalline substrates. In 2004, it became possible to peel off graphite
sheets and hence study isolated layers of graphene, which is an almost ideal 2D
crystal form of carbon with highly original electronic properties.
Quite generally, a given (Bravais) crystal lattice is characterised by the symmetry
operations that preserve its structure, and which include, apart from the lattice trans-
lations, axes of symmetry under rotation through some angle θ (or n-fold symmetry
axes, where θ = 2π/n), and planes of symmetry (mirror planes). In two dimen-
sions, there are only five types of Bravais lattice. These are, in order of increasing
symmetry:
• The oblique lattice (Fig. 2.3), which has the minimal 2D symmetry, i.e., only one
two-fold symmetry axis perpendicular to the plane.
• The rectangular lattice (Fig. 2.2b), which also has mirror planes parallel to the
shortest translation axes of the lattice.
• The centered rectangular lattice (see Fig. 2.4a), with mirror planes distinct from
the shortest translation axes of the lattice.
• The hexagonal lattice (Fig. 2.4b), with a 3-fold (and hence a 6-fold) symmetry
axis.
• The square lattice (Fig. 2.2a), with a 4-fold symmetry axis.
In the centered rectangular lattice of Fig. 2.4a, the primitive cell constructed from
(a1 , a2 ) does not help us to visualise the lattice symmetries as clearly as the con-
ventional unit cell (a, b), which is not a primitive unit cell, since it has double the
a2 a2
(a) b (b)
a1
a a1
Fig. 2.4 Plane lattices. (a) Centered rectangular, (b) hexagonal. In the first case, we observe the
conventional centered rectangular cell specified by (a, b) and the primitive cell (a1 , a2 ). In the
plane hexagonal lattice, all the primitive triangles are equilateral
area. In fact the latter contains two lattice points, viz., the point at the origin and the
center of the rectangle. These two points are indeed equivalent, as required by the
notion of a Bravais lattice, since each one is the center of the rectangle formed by its
four nearest neighbours. The conventional cell must be considered as a cell with one
basis (the two lattice points), and the crystal will be obtained by introducing twice
the material basis of the primitive cell around these points.
Question 2.1. 1. Determine the Bravais lattice and the primitive cell for the alkane crystal
observed by scanning tunneling microscopy in Fig. 2.1.
2. Determine the Wigner–Seitz cells associated with the centered rectangular and hexago-
nal lattices of Fig. 2.4.
The 2D structures attracting most attention since 2005 are the graphene honeycomb
structure of Fig. 2.5a and the so-called Kagomé structure. Some compounds like
hebersmithite ZnCu3 (OH)6 Cl2 are built up from alternating planes of ZnCl2 with a
simple 2D structure and Kagomé planes of Cu3 (OH)6 . The Cu2+ ions are arranged
as in Fig. 2.5b, while the (OH)− serve to bind the Cu2+ . The latter carry total elec-
tron spin 1/2, making this a very interesting material in quantum magnetism.
Question 2.2. Determine the Bravais lattice and a primitive cell for each of the two
structures of Fig. 2.5.
a) b)
Fig. 2.5 (a) Crystal structure of C atoms in graphene. (b) Atomic arragement known as ‘Kagomé’
(after a typical pattern of straw baskets made in Japan)
2.1.3 Three-Dimensional Crystals
In three dimensions, the simplest lattice to visualise is the cubic lattice. The three
primitive translation vectors (a1 , a2 , a3 ) form an orthogonal triad, and each of them
has length equal to the side a of the cube constituting a primitive cell (see Fig. 2.6a).
Chemical elements crystallise scarcely into such a simple cubic lattice, but it is
encountered in many polyatomic crystals (we shall discuss the example of CsCl
below).
However, many chemical elements crystallise into body-centered cubic (bcc)
crystal lattices (see Table 2.1). This Bravais lattice shown in Fig. 2.6b is the 3D
z a z a
a3 a3
a2 a2
y y
0 0
a1 a1
x x
a
(a) (b) (c)
Fig. 2.6 (a) Simple cubic lattice. (b) Body-centered cubic lattice. (c) Equivalence of nodes in the
body-centered cubic lattice
Table 2.1 Lattice constants a of the conventional cell for several elementary solids which form a
body-centered cubic structure with one atom per primitive unit cell
Element a [Å] Element a [Å] Element a [Å]
Ba 5.02 Li 3.49 Ta 3.31
Cr 2.88 Mo 3.15 Tl 3.88
Cs 6.05 Na 4.23 V 3.02
Fe 2.87 Nb 3.30 W 3.16
K 5.23 Rb 5.59
analog of the 2D centered rectangular lattice. In the bcc lattice, the conventional
cell contains two lattice points, one at the origin and the other at the center of the
cube. These two lattice nodes are indeed equivalent as each is the center of a cube
formed by its eight nearest neighbours (see Fig. 2.6c).
The most common lattice for elementary solids is the face-centered cubic lat-
tice (fcc). This is what is usually obtained when we try to stack hard spherical
balls. It is a common structure for many metals (see Table 2.2). The face-centered
Table 2.2 Lattice constants a of the conventional cell in several elementary solids which form
face-centered cubic structures
Element a [Å] Element a [Å] Element a [Å]
Ar 5.26 Ir 3.84 Pt 3.92
Ag 4.09 Kr 5.72 Pu 4.64
Al 4.05 La 5.30 Rh 3.80
Au 4.08 Ne 4.43 Sc 4.54
Ca 5.58 Ni 3.52 Sr 6.08
Ce 5.16 Pb 4.95 Th 5.08
Co 3.55 Pd 3.89 Xe 6.20
Cu 3.61 Pr 5.16 Yb 5.49
z
z a
a3 a2
y y
0 a1
x x
(a) (b)
Fig. 2.7 (a) Face-centered cubic lattice. The primitive cell is the rhombohedron specified by the
vectors a1 = (a/2)(x + y), a2 = (a/2)(y + z), and a3 = (a/2)(x + z). The conventional cell contains
four lattice nodes. (b) Cubic crystal structure of diamond. The atoms are all chemically identical
in diamond, Si, and Ge. In the case of InP or GaAs, the two species differ, and are represented by
empty spheres and full spheres
cubic lattice also has a conventional cubic unit cell containing a four-node
basis comprising one vertex of the cube and the three centers of the adjacent faces
(see Fig. 2.7a). Note that the primitive translations of the Bravais lattice are the
three vectors joining the cube vertex to the centers of the adjacent faces, so a
rhombohedral primitive cell can be constructed (see Fig. 2.7a). Among the systems
crystallising in this structure are diamond and many semiconductors such as Si, Ge,
GaAs, and InP. All atoms within the conventional cell are shown in Fig. 2.7b, where
the full and empty spheres correspond to the two atomic species in the case of binary
compounds, but are identical in the case of Si or Ge.
Question 2.3. Check that the crystals in Fig. 2.7b do indeed correspond to a face-centered
cubic Bravais lattice, and determine the atomic basis.
When we consider polyatomic crystals, the primitive cell necessarily contains sev-
eral atoms. A simple illustration is given in Fig. 2.8 for the alkali halides CsCl and
NaCl. Although Cs is at the center of a Cl cube, the associated Bravais lattice is the
(a) (b)
Fig. 2.8 (a) Crystal of CsCl, with simple cubic primitive cell. The basis comprises one atom of Cl
at the vertex of the cube and one atom of Cs at the center of the cube. (b) Crystal of NaCl with fcc
primitive cell comprising one atom of Cl and one atom of Na per primitive cell of Fig. 2.7a
simple cubic lattice of side a, with a basis comprising one atom of Cl at the vertex
of the cube and one atom of Cs at the center of the cube.
Many mixed oxides of transition metals crystallise into a cubic structure called
the perovskite structure, with primitive cell ABO3 , in which A and B are cations
with different size and valence. The small cation, generally A2+ , is surrounded by
an octahedron of oxygen atoms, while the large cation Bn+ (in general n = 3 or 4)
is surrounded by 12 oxygen atoms, as can be seen in Fig. 2.9a. These oxides can
exhibit a wide range of physical properties, from ferromagnetism in the manganites
LaMnO3 and cobaltites LaCoO3 to antiferromagnetism in iron-based perovskites
like LaFeO3 . Below 120◦ C, slight structural distortions with respect to the ideal
structure of Fig. 2.9a induce ferroelectric properties in the case of BaTiO3 .
Families of metal oxides with a highly 2D structure can sometimes be obtained
by combining planes with perovskite structure with square MO planes, where M
is a third metal cation. A classic example of such hybrid structures is provided by
the high-Tc cuprate superconductors, whose discoverers, Müller and Bednorz, were
rewarded by the Nobel prize. An example of such a structure is found in HgBa2 CuO5 .
This is shown in Fig. 2.9b. It is generated by intercalating sheets of Ba2 CuO4 , which
has the perovskite structure, with planes of HgO, or can alternatively be considered
simply as alternating planes of CuO2 /BaO/HgO/BaO/CuO2 , and so on. The prim-
itive cell of this structure is a right-angled parallelepiped whose sides a and b are
equal (tetragonal structure). In other cuprates with a = b, the structure is said to be
orthorhombic.
Ba
O
Cu
Hg
(a) (b)
Fig. 2.9 (a) ABO3 cubic perovskite structure. Below the 3D representation designed to show the
octahedra surrounding the A cations are two primitive cells centered respectively on the A and B
cations. (b) Quadratic structure of a high-Tc cuprate superconductor made up of planes of HgO
inserted between sheets of perovskite Ba2 CuO4 . This structure confers 2D physical properties on
the material
Naturally, there are many other 3D crystal systems with even fewer symmetries
than the simple lattices considered above. There is no question here of undertaking
an exhaustive study: there are 14 Bravais lattices in three dimensions!
2.1.4 Beyond the Perfect Crystal
Not all natural or artificial solids are crystalline. In many cases, there is no long
range order in the atomic arrangement. In particular, when a liquid is suddenly
cooled down below its solidification temperature, we can obtain a solid state which
simply freezes in the arrangement of atoms as it occurred in the liquid state. A glass
is obtained in this way by quenching the liquid, whereupon the atoms arrange them-
selves in a way that suffers only one constraint, namely that atoms are not allowed to
interpenetrate. If the atoms are thought of as hard spheres, the resulting glass struc-
ture looks like what would be obtained by putting beads in a container and shaking
them up. This glassy state is generally metastable, in the sense that the system can
have a lower free energy when the atoms are arranged into an fcc or hcp crystal
structure, which correspond to the closest packing of the beads. Crystallisation can
then be obtained by heat treatment, which amounts to shaking the box of beads in
our analogy.
But many other situations can be observed, with varying degrees of order. Con-
sider for example what happens for some alloys of two metals. A structure close to
a crystal structure can often be seen in these materials. The atoms distribute them-
selves randomly at the lattice points of a perfect crystal structure. This is called
a solid solution. This happens for example for the alloys Au1−x Cux , which can be
made with an arbitrary concentration x of Cu. The Cu and Au atoms distribute them-
selves randomly over the fcc lattice sites of pure Au, and the lattice spacing varies
slightly depending on the Cu concentration. This structure is not strictly speaking a
perfect crystal structure, although it can be treated as such in many respects.
Question 2.4. For some values of x, a heat treatment allows the atoms to arrange themselves
into a perfect crystal structure on the lattice. In the case of the Au1−x Cux alloys, indicate
one or more values x0 of x for which a perfect arrangement could in principle be obtained.
What would then be the associated primitive cell and Bravais lattice? In your opinion, if the
Au1−x Cux alloy forms a perfect crystal arrangement for x = x0 , is it likely from a physical
point of view that the same arrangement will be obtained for x = 1 − x0 ?
Imperfect crystal arrangements are observed in many other cases, in particular for
complex molecular structures. For example, the real crystals of cuprate supercon-
ductors shown in Fig. 2.9a are such that the HgO plane is highly deficient in oxygen.
The chemical formula is then HgBa2 CuO4+δ , and the oxygen vacancies are impor-
tant in determining the physical properties.
A novel illustration of disorder in crystals is shown below in the case of the
fullerene C60 , a molecule discovered in 1985, which has a football shape. Its
face-centered cubic structure, with large empty spaces between the C60 molecules,
Tetrahedral
Octahedral
Fig. 2.10 Left: C60 crystal. Right: Rb3 C60 crystal. These lattices are face-centered cubic. In
Rb3 C60 the rhombohedral primitive cell contains a C60 molecule and three Rb atoms
is shown in Fig. 2.10. It is easy to insert cations between the C60 molecules and
thereby create compounds of the form An C60 . The compound Rb3 C60 has attracted
considerable attention as it happens to be a metal that becomes a superconductor be-
low 27 K. Its fcc structure contains 3 rubidium atoms and one molecule of C60 per
primitive cell. The rubidium atoms have two different types of position: one, located
in the middle of an edge of the cube, has an octahedral C60 environment, while the
other two have a tetrahedral C60 environment (one vertex of the cube and three face
centers). Note that the C60 molecule has symmetries that are not compatible with the
face-centered cubic lattice. Indeed, there is no way of orientating the C60 molecule
so that it can map onto itself under all the symmetries of the lattice. There are not
really any 3D crystal structures whose primitive cells are given by those shown in
Fig. 2.10. These structures can nevertheless be considered as crystalline, but with
orientational disorder of the C60 molecules.
Crystals and Molecular Motions
At high temperatures, the C60 molecules are not immobile, but have rotational motions. These
rapid rotational movements are such that, on average, the C60 molecules behave like spheres,
and one can consider that the symmetry of the C60 molecule is no longer relevant. The average
structure is as shown in Fig. 2.10, treating the C60 as simple spheres. Since the molecules are
not fixed, we do not strictly have a crystal. Such systems in which molecular motions occur are
called plastic crystals.
For pure C60 , a phase transition takes place at 260 K from the face-centered cubic high
temperature structure of the plastic crystal to a body-centered cubic plastic crystal, as the rota-
tional motions of the C60 molecules occur in a correlated manner about particular axes relative
to the crystal axes.
At low temperatures, these rotational motions freeze, but the relative orientations of the C60
molecules in low temperature phases are not yet perfectly understood. Although one can speak
of an average face-centered cubic structure in Rb3 C60 , the state of relative disorder or order of
the C60 molecules has not yet been completely characterised.
2.2 Diffraction 35
2.2 Diffraction
In Sect. 2.2.1, we describe the general principles governing the diffraction of waves
by a periodic pattern, and show that, for a crystal, the directions in which diffraction
can occur are of course associated with the crystal structure. These diffraction con-
ditions are used in Sect. 2.2.2 to define a lattice of points in the wave vector space,
which is known as the reciprocal lattice of the crystal lattice.
2.2.1 General Principles
This effect was originally demonstrated by von Laue and the Braggs (father and
son) in 1912–1913. The electromagnetic waves (photons) are generally X rays, with
wavelength given in angstrom units (Å) and corresponding photon energy ε given
in keV. The latter quantities are related by
λÅ = 12.4/εkeV . (2.2)
The X rays used typically have energies in the range 10 < ε < 50 keV. In specific
cases, the radiation may also be in the form of neutrons or electrons. The wavelength
is then given by the de Broglie relation λ = h/p.
To understand how diffraction works, consider first the scattering of an arbitrary
plane (electromagnetic or matter) wave by some obstacle, usually an atom located
at the origin (see Fig. 2.11). The amplitude of the incident wave will then have the
form
ain (r, t) = a0 ei(k0 ·r−ωt) , (2.3)
where ω = c|k0 | for an electromagnetic wave, with c the speed of light, and
ω = h̄k20 /2m for a matter wave, with m the mass of the incident particles. Here,
a0 is a vector-valued amplitude in the case of electric or magnetic fields and a
complex-valued amplitude in the case of quantum matter waves. This amplitude de-
pends only on the intensity of the incident radiation. In general, the scattered wave
will have the form
Scattered
amplitude ascat
k1
Fig. 2.11 Scattering of an in-

cident plane wave with wave k1 = k0
Incident u
vector k0 . Waves scattered amplitude a in k0
in the direction k1 reveal
the phase difference between O
beams scattered by two ob-
jects at 0 and u

ascat (r, t) = a0 αk0 ,k ei(k·r−ωt) , (2.4)
k
where conservation of energy implies that |k| = |k0 |. The determination of the co-
efficients αk0 ,k is a (rather complex) problem of quantum mechanics which depends
on the type of wave and the quantum properties of the scattering object. However,
the exact expression for these coefficients is not needed to understand the underlying
principle of the diffraction methods described below.
Let us now ask what happens when the object is displaced through u from the
origin. Clearly, the scattered amplitude must be taken from the point u rather than
from the origin. We must therefore replace u by r − u in the exponential of the
expression in (2.4). But in addition, the incident wave will have a phase offset of
k0 ·u (see Fig. 2.11), and we therefore obtain

ascat (r, t) = a0 eik0 ·u αk0 ,k ei[k·(r−u)−ωt] . (2.5)
k
Finally, we use a detector that only detects waves scattered in a specific direction
k1 . The amplitude in this specific direction will then be
ascat (k1 ; r, t) = a0 αk0 ,k1 ei(k1 ·r−ωt) e−iK·u , (2.6)
with
K = k 1 − k0 . (2.7)
There is therefore a phase difference between the waves scattered in the direction
k1 by the objects located at O and u. In the general case of a crystal where the
identical scatterers are the material bases of the primitive cell repeated at the differ-
ent lattice points, these scattered amplitudes will generally undergo different phase
shifts exp(iK · u), leading to a low value of the total scattered intensity (sum of
the squared amplitudes). However, as the Braggs and von Laue observed, if all the
phase factors are the same in certain directions k1 , the resulting phase coherence
between the amplitudes scattered in these directions will lead to a high diffracted
intensity.
a. The Bragg Formulation
We consider the crystal lattice as an ensemble of lattice planes as shown in Fig. 2.12,
with O and u two lattice points belonging to one of the lattice planes, and we take
the scatterers to be the material bases of the primitive cell centered at O and u. If the
wave vector k0 of the incident wave makes an angle of incidence θ with the lattice
plane, it is easy to see that the phase difference K · u vanishes if k1 corresponds
to a reflection of k0 in the lattice plane. So in this direction, there will be phase
coherence between the intensities scattered by all the bases associated with this
2.2 Diffraction 37
Fig. 2.12 Reflection of in- k0

k1
cident radiation on lattice
θ θ
planes. The Bragg condition
O u
obtains when the path dif-
ference is a multiple of the d
wavelength
particular lattice plane. If in addition we wish to observe phase coherence between

the amplitudes scattered by the different parallel lattice planes, examination of
Fig. 2.12 shows that the condition that must be satisfied by θ is
2d sin θ = nλ , (2.8)
where d is the distance between the lattice planes and n is an integer. This diffraction
condition, known as the Bragg condition, is given here in a form that corresponds
to the representation of the crystal lattice in lattice planes.
b. General von Laue Formulation
We now consider the scattering by a very large number of atomic bases of the prim-
itive cell arranged in positions Rn . The total scattered wave will simply be the su-
perposition of many terms of the form (2.6):
aTscat (k1 ; r, t) = a0 ei(k1 ·r−ωt) A (K) , (2.9)
with

A (K) = αk0 ,k1 e−iK·Rn . (2.10)
n
In this equation, αk0 ,k1 characterises the radiation and the arrangement of atoms in
the primitive cell, while the sum over n is only associated with the spatial positions
of the Bravais lattice points. If K is chosen such that
eiK·Rn = 1 , for all translation vectors Rn in the crystal lattice , (2.11)
we obtain A (K) = Nαk0 ,k1 , where N is the number of primitive cells in the crystal.
Under these conditions, the diffracted intensity in the direction k1 is
Idiff ∝ |A (K)|2 = N 2 |αk0 ,k1 |2 . (2.12)
Equations (2.11) and (2.7) for k1 and k0 provide another expression of the Bragg
condition.
2.2.2 Reciprocal Lattice
Here we interpret (2.11) in the special case of a linear chain. In one dimension, we
have seen that the Bravais lattice is specified by Rm = ma1 = max, where a is the
period and x a (dimensionless) unit vector along the Ox axis. The relation (2.11)
implies that K is an integer multiple of a∗ = (2π/a)x.
In three dimensions, if K is specified relative to a frame (a∗1 , a∗2 , a∗3 ) by
K = x1 a∗1 + x2 a∗2 + x3 a∗3 , (2.13)

then (2.11) takes the form
(x1 a∗1 + x2 a∗2 + x3 a∗3 )(l1 a1 + l2 a2 + l3 a3 ) = 2π nl , (2.14)
with nl an integer, and this for all Rl , i.e., for all integer values of (l1 , l2 , l3 ). We thus
see that, by choosing the basis (a∗1 , a∗2 , a∗3 ) such that
ai ·a∗j = 2π δij , (2.15)
the relation (2.14) reduces to
x1 l1 + x2 l2 + x3 l3 = nl integer , (2.16)
and this for all (l1 , l2 , l3 ), implying that x1 , x2 , and x3 are also integers. The com-
ponents of K relative to the frame (a∗1 , a∗2 , a∗3 ) are called the Miller indices, and are
usually denoted by (h, k, l).
As a consequence, the vectors K satisfying (2.11) generate a Bravais lattice in
the k space. This is the reciprocal lattice associated with the Bravais lattice in po-
sition space, called hereafter real or direct. The reference frame (a∗1 , a∗2 , a∗3 ) of the
reciprocal space is defined in terms of the real space frame (a1 , a2 , a3 ) by (2.15). It
is easy to check that the a∗i are given by
a2 ∧ a3
a∗1 = 2π , (2.17)
a1 · (a2 ∧ a3 )
and cyclic permutations. Here the denominator is precisely the volume of the prim-
itive cell of the direct lattice.
Note that, according to (2.11), the direct and reciprocal lattices play symmetric
roles. In particular, the reciprocal lattice of the reciprocal lattice is just the direct
lattice. However, the real crystal is a lattice of atoms or molecules, or more generally,
a lattice of what we have called material bases, whereas the reciprocal lattice is a
lattice of points that are independent of the bases of the real crystal.
For example, the reciprocal lattice of a simple cubic lattice with lattice constant
a is a simple cubic lattice with lattice constant 2π/a. The reciprocal lattice of an
fcc lattice with lattice constant a is a body-centered cubic lattice of lattice constant
4π/a, e.g., Al, Si, GaAs. Conversely, the reciprocal lattice of a body-centered cubic
lattice is an fcc lattice, e.g., Fe.
2.3 Determination of Crystal Structures 39
Fig. 2.13 Determination of

the Bragg plane
0
K θ Bragg
k0
k1
θ k0
The notion of reciprocal lattice can be used to relate the Bragg and von Laue rep-
resentations of the diffraction conditions. Indeed, since |k0 | = |k1 |, the diffraction
condition k1 − k0 = K implies that
2k1 ·K = K 2 . (2.18)
As illustrated in Fig. 2.13, this means that k0 and k1 are obtained from one another
by a reflection in the orthogonally bisecting plane of the vector K. The direction
of diffraction k1 therefore corresponds to a Bragg diffraction on the lattice planes
of the crystal parallel to the orthogonally bisecting plane of the vector K. There
is thus a one–one correspondence between the lattice planes of the crystal and the
orthogonally bisecting planes of the reciprocal lattice vectors, which are known con-
ventionally as Bragg planes.
2.3 Determination of Crystal Structures
In Sect. 2.3.1, we show to begin with that the directions satisfying the Bragg con-
dition are well defined experimentally, but that they are not always easy to detect.
Once they have been found, the crystal lattice can be determined. In Sect. 2.3.2, we
describe how the basis of the primitive cell can be ascertained by determining the
intensities of the different Bragg diffraction spots. This will be exemplified for the
case of X-ray diffraction in Sect. 2.3.3.
2.3.1 On the Bragg Diffraction Condition
We have seen that the diffracted intensity is high when the factor

A0 (K) = e−iK·Rn
n
is large, and this occurs when the Bragg condition holds:
k1 − k0 = K , K ∈ reciprocal lattice . (2.19)

Consider now the scattered intensity if the Bragg condition is not satisfied. To sim-
plify the notation, we discuss the case of a simple cubic structure with lattice con-
stant a and length La in each of the three space directions. The sum in (2.10) then
becomes the product of three geometric sums, and direct calculation gives
xL
2 sin2
|A (K)| = αk0 ,k1 f (Kx a)f (Ky a)f (Kz a) , with f (x) =
2 2 , (2.20)
2x
sin
2
where K is not a reciprocal lattice vector here. Figure 2.14 shows f (x) for two val-
ues of L. We observe that, apart from the Bragg diffraction points 2nπ , where K is a
reciprocal lattice vector, this function remains small everywhere. If the Bragg condi-
tion is not satisfied, the scattered intensity will therefore remain very low. However,
the maxima at points x = 2nπ become sharper and sharper as L increases, which
supports the result (2.12). We thus conclude that the diffraction conditions are very
precisely specified, and this will only be limited experimentally by the size of the
diffracting crystal and the wavelength dispersion of the incident radiation.
Question 2.5. For the alloy Cu1−x Aux of Question 2.4, how does the diffraction pattern
change when the Cu and Au atoms arrange themselves for the concentration x0 ?
To determine the reciprocal lattice, and hence the Bravais lattice, it remains only
to determine all the space directions in which Bragg diffraction occurs. Experi-
mentally, it is not totally obvious how to determine the directions in which Bragg
diffraction will occur. This is illustrated in Fig. 2.15a, which shows the Ewald con-
struction, used to determine the diffraction directions for an incident wave k0 and
a given position of the crystal. The points of the reciprocal lattice corresponding
to this position of the crystal are shown in the left-hand figure, taking the origin
of the reciprocal lattice to be the point corresponding to the end of the vector k0 .
The diffraction directions are obtained by determining the points of the reciprocal
lattice on a sphere, known as the Ewald sphere, with radius |k0 | and centered at the
100 100
80 80
60 60
40 40
20 20
–2 0 2 4 6 8 –2 0 2 4 6 8
Fig. 2.14 The function f (x) for x between −π and 3π . Left: L = 10. Right: L = 20. Note that, on
the right-hand graph, the peak is at the ordinate value 400
k1
k1 k1 K
kd
O k0 Γ O k0 Γ O k0 Γ
(a) (b) (c)
Fig. 2.15 Geometry of Bragg diffraction in reciprocal space. For an incident vector k0 , diffraction
directions such as kd are obtained using the Ewald construction as shown in (a). The reciprocal
lattice with origin is depicted. For an incident vector k0 and a detector in the direction k1 ,
no Bragg diffraction will generally be detected (b), except for certain specific orientations of the
crystal, such that the associated reciprocal lattice is oriented as shown in (c), for example
origin O of k0 . Diffraction directions such as kd which join O to these points of the

reciprocal lattice are few and far between.
If we have only one detector set in the direction k1 , for example, there will gen-
erally be little diffraction in this direction. If the crystal is then rotated in space,
according to (2.17), the reciprocal space will also rotate about its origin . As can
be seen from Fig. 2.15, Bragg diffraction will only be detected in the given direc-
tion k1 for very specific orientations of the crystal. Given that we are working in
3D space, it is clear that, for a randomly chosen orientation of the crystal relative to
the incident radiation, we will only very rarely observe a Bragg diffraction peak (or
spot).
There are several experimental methods to get round this difficulty, e.g., using
powders rather than single crystals, but we shall not discuss the issue further here.
Knowing the orientations at which diffraction occurs, we may then characterise the
reciprocal lattice, and hence also the crystal symmetries and the size of the primitive
cell in the crystal lattice itself.
2.3.2 Diffraction Intensity and Basis of the Primitive Cell
So far we have not considered the structure of the diffracting object in any detail.
Suppose now that we observe the diffraction by a crystal whose primitive cell com-
prises Na atoms, possibly of different chemical nature, at positions rl inside the unit
cell. In this case, a coefficient αk(l)0 ,k1 is associated with each atom. The sum in (2.9)
over all sites l leads to a diffracted amplitude in the direction k1 given by
N

Na a
−iK·(Rn +rl )
αk0 ,k1 e−iK·rl −iK·Rn
(l) (l)
A (K) = αk0 ,k1 e = e . (2.21)
l=1 n l=1 n
There are thus two multiplicative factors in the expression for the diffracted inten-
sity. One of these,
A0 (K) = e−iK·Rn ,
n
depends only on the Bravais lattice of the crystal and is only nonzero when the
Bragg condition (2.19) is satisfied. The second,

Na
αk0 ,k1 e−iK·rl ,
(l)
S(K) = (2.22)
l=1
called the structure factor, depends only on the basis of the primitive cell and the
(l)
nature of the diffracted radiation through the coefficients αk0 ,k1 . If we measure the
intensities of many Bragg diffraction spots, we can then deduce the structure of the
primitive cell, provided we have a good understanding of the physical processes
involved in the scattering of radiation by the atoms, since it is these processes that
(l)
determine the values of the coefficients αk0 ,k1 . Note that these coefficients will be
different for X-ray diffraction, the X rays being scattered by electrons, and neutron
diffraction, since neutrons are scattered by the atomic nuclei.
2.3.3 X-Ray Diffraction
In the specific case of X-ray diffraction, the coefficients αk0 ,k1 can be expressed very
simply by considering the underlying physical process for the scattering of X rays by
atoms. Indeed, at the high frequencies associated with X rays, the electric field of the
electromagnetic wave couples predominantly with the electrons in the atom. Each
electron vibrates in phase with the incident wave at its frequency ω. The electron
charge thus has an oscillatory motion with angular frequency ω, and consequently
emits an electromagnetic wave at this same angular frequency, in phase with the
incident wave.
Consider an atom centered at the origin O and with electronic density distribution
ρ(u), where u indicates the position of an electron relative to O. The amplitude of
the wave scattered in the direction k1 by the electronic density at O is proportional
to the electronic density at this point:
a0 ρ(0)ei(k1 ·r−ωt) . (2.23)
Consider now an arbitrary point at position u. In an analogous way to Fig. 2.11, the
amplitude scattered in the direction k1 can thus be written
a0 ρ(u)ei[k1 ·(r−u)−ωt] eik0 ·u . (2.24)
The total wave diffracted in the direction k1 by all the electrons in the atom is
therefore

adiff (k1 , r) = a0 ei(k1 ·r−ωt) ρ(u)e−i(k1 −k0 )·u d3 u , (2.25)
where |k1 | = |k0 |. This expression, valid for any molecular ensemble of electronic
density ρ(u), shows that the diffraction amplitude of the X rays is directly related to
the Fourier transform of the electronic density. The diffraction pattern produced by
an arbitrary object thus contains information about the structure of the diffracting
object.
Going back to the case of a single atom and comparing (2.25) and (2.6), we define
the atomic form factor

αk0 ,k1 = ρ(u)e−i(k1 −k0 )·u d3 u = f (k1 − k0 ) . (2.26)
For a crystal whose primitive cell contains Na atoms at position rl , the diffraction
amplitude is the Fourier transform of the total electronic density, which is the peri-
odic reproduction of the electronic density of the primitive cell. The latter is given
by (2.21), where the structure factor is simply

Na
S(K) = fl (K)e−iK·rl . (2.27)
l=1
It can be determined once we know the form factors and the positions of the different
atoms in the primitive cell. We thus find that the experimental determination of
the intensities of the Bragg diffraction peaks or spots will be extremely useful for
ascertaining the arrangement of atoms in the primitive cell of the Bravais lattice.
Let us consider a specific example to illustrate how the basis of the primitive cell
affects the intensity of the Bragg peaks. Consider X-ray diffraction by a crystal of
C60 whose fcc primitive cell is shown in Fig. 2.10. A suitable experimental setup
records on a photographic film all the diffraction spots corresponding to the vectors
K in one plane of reciprocal space (see the image on p. 24). The pattern observed
can thus be used to directly visualise a plane of the reciprocal lattice and its symme-
tries. In the image on p. 24, the spots are such that the vectors K separating them
correspond to the vectors K in one plane of the reciprocal lattice up to a scale factor
given by |K| = 2π |K |/λL. Here λ is the wavelength of the X rays and L is a length
depending on the experimental setup. (Here λ = 1.542 Å and L = 0.06 m.)
Question 2.6. 1. Identify the body-centered cubic reciprocal lattice plane visualised in the
image of p. 24. Deduce the dimensions of the primitive cell for C60 .
2. Some of the diffraction spots are very faint. Which vectors of the reciprocal lattice do
these correspond to? To understand this, we take into account the fact that the X rays
are scattered by electrons. We may consider the C60 molecule as a uniform charge dis-
tribution p(r) over the surface of a hollow sphere of radius R0 , viz., p(r) = Aδ(r − R0 ).
Calculate the structure factor for the C60 crystal. Deduce the radius R0 of C60 .
This aspect of X-ray diffraction was very important in determining the spatial struc-
ture of complex molecules, such as biological molecules, which are commonly
conserved in solution. By crystallising N molecules, one then benefits from the fact
that, in a crystal, the Bragg diffraction spots have intensities that increase as N 2 ,
whereas the intensity only increases as N when the molecules do not have a crys-
talline arrangement. This method was used to determine the structure of DNA and
many other biologically important molecules.
Note that disorder or atomic and molecular motions modify the intensities of the
Bragg diffraction spots. The effect of lattice vibrations is discussed in Problem 1:
Debye–Waller factor.
2.4 Summary
A crystal lattice can be described as a combination of two entities: the Bravais lat-
tice, which is a periodic arrangement of lattice points in space, and hence an abstract
construction, and a material basis which is the actual physical entity associated with
each node of the Bravais lattice. These are given by
R = 1 a1 + 2 a2 + 3 a3 ,
where (a1 , a2 , a3 ) is a vector basis specifying a primitive cell of the lattice. In the
simplest cases, the material basis is a single atom, but it may be a much more com-
plex physical entity, such as an arrangement of atoms, one or more molecules, and
so on.
The structures of arbitrary molecular entities can be determined using diffraction
methods. An incident wave of wave vector k0 is elastically scattered by the various
objects making up the molecular entity. Interference between the scattered waves
leads to a diffraction pattern. In the case of a crystal, the diffracted intensity is only
significant in specific directions k1 satisfying
k 1 − k0 = K ,
where K is a vector belonging to another lattice called the reciprocal lattice for the
actual lattice in real space. This relation is the Bragg diffraction condition.
The vectors K of the reciprocal lattice are defined by the condition
eiK·R = 1 ,
for any R belonging to the Bravais lattice of the crystal. The reference triad
(a∗1 , a∗2 , a∗3 ) of the reciprocal lattice is given by
a2 ∧ a 3
a∗1 = 2π ,
(a1 ·a2 ∧a3 )
2.5 Answers to Questions 45
and cyclic permutations. The vectors k1 satisfying the Bragg diffraction condition
can be used to determine the reciprocal lattice. The amplitude of the diffraction is
related to the basis of the primitive cell of the crystal. It is given by

A(K) = S(K) e−iK·Rn ,
n
where S(K) is the structure factor of the basis of the crystal.

For X rays, scattering is due to electrons in the atomic orbitals, and each atom l
at rl is characterised by its atomic structure factor fl (k1 − k0 ). The structure factor
of the primitive cell is then

Na
S(K) = fl (K)e−iK·rl .
l=1
2.5 Answers to Questions
Question 2.1
1. The alkane molecules are arranged in parallel rows. Note that two consecutive
rows are staggered in a quincuncial arrangement. This is checked by looking at
Fig. 2.1 at grazing incidence with respect to the axis of the molecules. The lattice
is therefore centered rectangular. Figure 2.16 shows the centered rectangular con-
ventional unit cell (a, b) and the primitive unit cell (a1 , a2 ) containing one alkane
molecule. This is shown on part of the image and also on a molecular model.
2. It is an irregular polyhedron for the centered rectangular lattice and a hexagon
for the hexagonal lattice (see Fig. 2.17).
a
a2 b
a1
a
a2 b
a1
Fig. 2.16 Centered rectangular cell and primitive unit cell for the 2D alkane crystal of Fig. 2.1.
These are shown on an enlarged portion of the image of Fig. 2.1 (upper) and on a molecular model
(lower)
a2 a2
b
a1
a a1
Fig. 2.17 First Brillouin zones of the centered rectangular and hexagonal 2D lattices: construction
displayed on the right for each reciprocal lattice
Question 2.2
The honeycomb structure of graphene corresponds to a hexagonal lattice in which

one in three sites have been removed. The primitive cell (a1√ ,a2 ) is hexagonal with
three times the area of the initial hexagonal cell, i.e., 3a2 3/2. It contains two
carbon atoms placed at the nodes of two hexagonal sublattices of types A and B,
which differ in the opposite orientations of their nearest neighbours. Each atom A
has three nearest neighbours B and vice versa.
The Kagomé plane structure corresponds to a hexagonal lattice in which one
site in four has been removed (one site out of two and in one line out of two). Its
Bravais lattice is hexagonal with twice the lattice constant of the initial structure.
The primitive cell contains 3 atoms of Cu2+ and 6 (OH)− (see Fig. 2.18).
(a) (b)
(OH)−
a2 a a2
Cu2+
a1
a1
A B
Fig. 2.18 Unit cell and bases of the Bravais lattices for the graphene sheet and for the Kagomé 2D
planes of the hebersmithite crystal
Question 2.3
The crystal in Fig. 2.7b corresponds to an fcc lattice with full spheres at the lat-
tice points and empty spheres obtained from the full ones by translation through
(a/4, a/4, a/4). There are thus two atoms per polyhedral primitive cell in Fig. 2.7a,
shown by two empty spheres in Fig. 2.19, one at the corner of the cube and the other
at (a/4, a/4, a/4). There are 8 atoms per fcc unit cell: the vertex, the three centers of
adjacent faces, and the 4 empty spheres in Fig. 2.7b.
Fig. 2.19 Primitive unit cell z

of the diamond structure. The
atoms located on the two
empty spheres represent the
basis for this crystal
y
d
x
Question 2.4
When the atoms Au1−x Cux are randomly distributed over the sites of the fcc Bravais
lattice, the crystal structure is not perfect. The atoms must be placed at the sites
of the fcc lattice in a periodic manner to obtain a perfect arrangement. With two
different atoms, viz., Cu and Au, the primitive cell of the perfect crystal will have at
least one two-atom basis and hence a Bravais lattice with unit cell at least doubled
in volume.
It should be fairly clear that it is not possible to associate the sites of the fcc struc-
ture two by two to define a new Bravais lattice. However, if we take 4 atoms with 3
the same, one of the atoms can be placed at the vertex of the fcc conventional unit
cell and the three others at the centers of the adjacent faces (empty in Fig. 2.20a).
We thereby construct a crystal with simple cubic Bravais lattice of side a and a basis
of 4 atoms per cell.
z a
kz 2π a
ky
y O
O
x
kx
(a) (b)
Fig. 2.20 (a) Simple cubic crystal structure of Cu3 Au. Full spheres represent Au and empty
spheres Cu. (b) The reciprocal lattice of the fcc structure is the bcc lattice shown by full spheres.
Empty spheres represent the extra reciprocal space nodes which appear for the ordered cubic
structure of Cu3 Au
There are therefore two simple possibilities, those corresponding to Au3 Cu or

Cu3 Au, i.e., x = 0.25 or x = 0.75. Naturally, we can imagine ordered solutions for
lower concentrations of one of the metals, but there is no particular physical reason
why they should ever occur.
Note that, in the ordered Cu3 Au structure, the gold atoms have no Au nearest
neighbour. This structure is found to be stable on thermodynamical grounds, which
indicates that Au atoms strongly repel one another. However, the Au3 Cu structure is
not obtained experimentally, probably because the repulsion between copper atoms
is not strong enough. The ordered AuCu alloy exists, but crystallises into a cubic
lattice which bears no relation to the fcc structure of the pure metals.
Question 2.5
The ordered arrangement of the Au and Cu atoms can be detected by a diffraction

method. The fcc structure of the disordered solid solution has a primitive cell of vol-
ume a3 /4 and a reciprocal lattice with a body-centered cubic conventional cell of
side 4π/a. The ordered lattice of Cu3 Au shown in Fig. 2.20a has a cubic primitive
cell of volume a3 , and its reciprocal lattice is cubic with side 2π/a. There are there-
fore many diffraction spots for the ordered alloy. In Fig. 2.20b, the lattice points of
the body-centered cubic reciprocal lattice are represented by full spheres, while the
extra nodes of the simple cubic lattice are represented by empty spheres. The pres-
ence of order can thus be revealed by the appearance of diffraction spots associated
with these new vectors in reciprocal space.
Question 2.6
1. The C60 lattice is face-centered cubic with conventional unit cell of side a. Its
reciprocal lattice is body-centered cubic with conventional cell of side 4π/a (see
Fig. 2.20b). Spots are therefore expected for
K = h a∗ + kb∗ + c∗ , (2.28)
with |a∗ | = |b∗ | = |c∗ | = 2π/a and h, k, of the same parity. In the observed
plane of the reciprocal lattice, the spots form a square lattice. The only planes
of the body-centered cubic reciprocal lattice with this property are the planes
passing through one of the faces of the cube, i.e., (a∗ , b∗ ), or (b∗ , c∗ ), or (c∗ , a∗ ).
Assuming that this is the plane (a∗ , b∗ ), the a∗ and b∗ axes are shown in Fig. 2.21.
The spots correspond to the vectors K = (h, k, 0) with h and k even. If K0 is
the spacing of the observed square lattice which corresponds to K0 = 4π/a, it
follows that
4π 2π K0 L
= , and then a = 2λ . (2.29)
a λ L K0
ky
kx
)
20
(2
b∗
a∗
)
00
(2
)
20
(0
0
K
8
Fig. 2.21 Indexation of the reciprocal lattice plane of the C60 crystal detected by X-ray diffraction
(image on p. 24). See answer to Question 2.6
By measuring over various rows corresponding to 8 times the lattice spacing, and
knowing that the film diameter is 12 cm, we obtain 8K0 = 10.45 ± 0.05 cm in
Fig. 2.21, so finally,
a = 14.17 ± 0.07 Å . (2.30)
2. Note that the spots (h, 0, 0) and (0, k, 0) are rather faint. In fact their intensity is
found to be at least 103 times lower than the intensity of the spots (2 2 0). Set
ρ(r) = Aδ(r − R). The total number of electrons is 360 per C60 . Hence,
∞
360e = ρ(r)d r =
3
4π r2 δ(r − R)Adr = A4π R2 ,
0
which yields
360e
A= .
4π R2
Then

S(K) = ρ(r)eiK·r d3 r
∞ 2π π
360e
= r δ(r − R)dr
2
dφ eiKr cos θ sin θ dθ
4π R2 0 0 0
∞
360e sin Kr 2
= 2 r δ(r − R)dr
2R2 0 Kr
sin KR
= 360e ,
KR
and the spots have intensity
sin2 KR
I∝ . (2.31)
(KR)2
The intensity vanishes for KR = nπ with n integer. The fact that the spots corre-
sponding to K = n(4π/a) are almost extinguished thus shows that 4π R/a π ,
and then R ∼ a/4 or R 3.54 Å. With a relative intensity of 10−3 for the spot
(2 0 0) compared with the spot (2 2 0), the error obtained for R is found to be
R = 3.54 ± 0.04 Å . (2.32)
In fact, using (2.31) and taking into account the measured intensities for all the
spots, the error in R is much lower, namely R = 3.52 ± 0.01 Å.
Chapter 3
Electronic Structure of Solids:
Metals and Insulators
Contents
3.1 Electrons in a Periodic Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1.1 Bloch’s Theorem in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.1.2 General Properties of Bloch States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.3 Electrons in a Weak Periodic Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.1.4 First Brillouin Zone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Band Structure of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.1 Linear Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.2 Band Structure of Two-Dimensional Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.3 Three-Dimensional Band Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3 Metals, Insulators, and Semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.1 Statistical Physics of Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.2 Real Solids: Conductor or Insulator? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.3 Introduction to the Semiconductor Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.4 The Special Case of Graphene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.5 Thermodynamic Properties of Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4 Experimental Determination of Band Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.1 Optical Absorption in Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.2 Tunneling Effect and Scanning Tunneling Microscopy . . . . . . . . . . . . . . . . . . . . . . 85
3.4.3 Angle-Resolved Photoemission Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

52 3 Electronic Structure of Solids: Metals and Insulators
As illustrated on the cover, the Fermi surface of a metallic material can be represented in reciprocal
space using angle-resolved photoemission experiments. In this image, the Fermi surface of the
CuO2 plane of the cuprate superconductor BiSr2 CaCu2 O8 is reconstructed experimentally in the
reciprocal space plane kx , ky . Question 3.4 examines this experimental result in more detail. Image
courtesy of S. Borisenko, constructed from experimental results in Kordyuk, A.A., Borisenko, S.V.,
Golden, M.S., Legner, S., Nenkov, K.A., Knupfer, M., Fink, J., Berger, H., Forró, L., Follath, R.:
Phys. Rev. B 66, 014502 (2002)
3.1 Electrons in a Periodic Potential 53
In this chapter, we introduce the generalisation needed to discuss the electronic

properties of solids in more detail. We restrict the discussion to features that can be
described using the approximations presented in Chap. 1. Considering fixed atomic
positions, and averaging electron–electron interactions, the task boils down to solv-
ing a one-electron problem. We also restrict to the case of crystalline solids, having
introduced the basic ideas in Chap. 2.
In Sect. 3.1, we discuss the general form of the Bloch theorem and its implica-
tions for the band structure of periodic solids. While the tight-binding approximation
provided a first handle on the notion of energy band, we shall see that this notion can
also be introduced in the framework of quite the opposite approximation, in which
the periodic potential is weak. This will reveal the importance of the reciprocal lat-
tice and lead to a general definition of the Brillouin zones outlined in our discussion
of the linear atomic chain in Chap. 1. We also establish a link between the limiting
cases of weak and strong atomic potentials. We can then establish certain general
rules about the band structures for the chain, but also for two- and three-dimensional
solids (see Sect. 3.2). In Sect. 3.3, we use the band model to discuss some physi-
cal properties of solids and in particular their thermodynamic properties. We find
that this model implies only two possible states for a solid, viz., either conducting
(metallic) and paramagnetic, or insulating and non-magnetic. In addition, without
carrying out any detailed calculation, we can make several quantitative predictions
that can be compared with experiment. A discussion of these points will bring out
the strengths and weaknesses of the band model. Finally, in Sect. 3.4, we describe
some experimental methods for determining the electron density of states or the
band structure of solids.
3.1 Electrons in a Periodic Potential
We saw in Chap. 1 that, under the tight-binding approximation, the atomic electron
states broaden to give rise to energy bands. We shall see here that this organisation
of the structure of the electronic states into energy bands is in fact a general property
of the electronic energy spectrum in the presence of a periodic potential. We begin
in Sect. 3.1.1 by generalising the Bloch theorem to three dimensions. We shall then
see how the energy band structure can be obtained in a simple way from that of the
free electrons in the weak limit of the periodic potential (see Sect. 3.1.3). The notion
of Brillouin zone can then be extended in a very general way (see Sect. 3.1.4).
3.1.1 Bloch’s Theorem in Three Dimensions
In an infinite crystal, or more precisely, in a crystal subject to periodic boundary

conditions, electrons located at positions separated by a direct lattice translation Rl
[of the form (2.1)] will feel the same crystal potential, which means that the one-
electron Hamiltonian H1 will be invariant under the lattice translations associated
with the operators T̂ defined by T̂Rl f (r) = f (r + Rl ), for any function f (r). It is
straightforward to generalise Bloch’s theorem to a 3D space.1
The solutions of the Schrödinger equation

p2
+ V(r) ψ(r) = Eψ(r) , (3.1)
2m
where the crystal potential V(r) has the periodicity of the crystal, are called Bloch
functions. They have the form
ψn,k (r) = eik·r un,k (r) , (3.2)
where the function un,k (r) is periodic for the translations of the direct lattice:
un,k (r + Rl ) = un,k (r) . (3.3)
A state is thus specified by four quantum numbers, namely the three components kx ,
ky , kz of the vector k and an integer n, because (3.1) can have several solutions for a
given k. One consequence of (3.3) is that
ψn,k (r + Rl ) = eik·Rl ψn,k (r) . (3.4)
This wave function is such that the resulting physical properties at points r and
r + Rl are the same, because the wave function differs only by a phase factor that is
independent of r. In particular, the electronic density |ψn,k (r)|2 is periodic in space.
Furthermore, the periodic boundary conditions for a rhombohedral solid contain-
ing many primitive cells (see Fig. 3.1a), with edges N1 a1 , N2 a2 , N3 a3 , can be written
in the form
ψ(r + Ni ai ) = ψ(r) , i = 1, 2, 3 . (3.5)
Comparing (3.5) with the form of the Bloch functions implies that the values of k
must be quantised, with k·(Ni ai ) = 2π ni , where the ni are integers, which leads to
ni
k= a∗ .
Ni i
i=1,2,3
For a macroscopic solid, these values of k will be extremely close together on the
scale of the primitive unit cell of the reciprocal lattice (see Fig. 3.1b), since the Ni are
large. Note that, in the primitive cell (a∗1 , a∗2 , a∗3 ) of the reciprocal lattice, there will
be N1 N2 N3 quantised values of k, exactly corresponding to the number of primitive
cells in the crystal.
1 The generalisation to several dimensions is quite simple. We introduce three primitive translation
operators T̂ai , for i = 1, 2, 3, which are easily shown to commute with one another and with the
Hamiltonian. We can then apply the argument in the box in Sect. 1.2.2a almost word for word, to
arrive at (3.2) and (3.3).
(a) −a1 a2
(b)
−a3 a3∗ /N3

a3∗
k
a∗2
N3 a3 a2∗ /N2
N2 a2 0
a1∗ /N1 a1∗
r
r + N 1 a1
0
N1 a1
Fig. 3.1 (a) Periodic boundary conditions applied to a solid containing N1 N2 N3 unit cells.
(b) Quantised values of k in the primitive unit cell (a∗1 , a∗2 , a∗3 ) of the reciprocal lattice
The vectors k play a key role when describing the properties of crystalline solids.
They belong to the reciprocal space attached to the crystal, generated by the basis
vectors (a∗1 , a∗2 , a∗3 ) of the reciprocal lattice.
3.1.2 General Properties of Bloch States
Consider the Bloch function with wave vector k, such that k = k0 + K, where K is
a vector of the reciprocal lattice. We shall show that this Bloch function ψn,k (r) is
also a Bloch function for the wave vector k0 :
ψn,k (r) = eik·r un,k (r) = eik0 ·r eiK·r un,k (r) . (3.6)
The function exp(iK · r)un,k (r) is periodic as expressed by (3.3). Indeed, using the
definition (2.11) of the reciprocal lattice, it follows that
eiK·(r+Rl ) un,k (r + Rl ) = eiK·r un,k (r) . (3.7)
The function (3.6) is thus a solution of the Schrödinger equation with the same
energy for both k and k0 = k − K. The reciprocal space is therefore much bigger
than it need be to specify the Bloch functions, because it contains all the points k
and k−K, with K an arbitrary vector in the reciprocal lattice, for which the quantum
states are identical. Since the quantum states are only defined up to a vector of the
reciprocal lattice, they can all be specified by restricting to a particular primitive
unit cell of the reciprocal lattice.
The form (3.2) and (3.3) for the wave functions has one very important conse-
quence for any physical measurement involving external radiation (photons,
neutrons, electrons), generally represented by a plane wave exp(iq · r). The wave
function of an electron in a Bloch state ψn,k (r), where k is a vector belonging to
the primitive
cell, can be written as a superposition of plane waves with the general
form exp i(k + K)·r , where K is a vector of the reciprocal lattice. In an interac-
tion between incident radiation qin and a Bloch state kin , scattered to qout and kout ,
the transition probability can only be nonzero for partial waves conserving the total
momentum:
qin + kin + Kin = qout + kout + Kout ,
that is
qin + kin = qout + kout + ΔK. (3.8)
This means that, in a crystal, the quasi-momentum (or crystal momentum) k plays
the same role of a conserved quantity as the true momentum p in vacuum. The only
precaution required here concerns the existence of the term ΔK in (3.8), which is a
vector of the reciprocal lattice: if this term vanishes, we speak of normal scattering,
whereas for ΔK = 0, we speak of umklapp scattering. For reasons of energy conser-
vation, umklapp effects can often be neglected. It should be noted that, in a crystal,
the average of the true momentum p can never be treated as a conserved quantity. In
fact, p is not conserved, due to the presence of the potential in the Hamiltonian.
3.1.3 Electrons in a Weak Periodic Potential
Band structure calculations based on the tight-binding approximation, as discussed

in Chap. 1, apply in particular to the case where there is a wide separation between
atomic levels compared with the hopping integrals. In other cases, for example, the
case of valence electrons of certain metals, which may have a high kinetic energy,
the periodic potential can be treated as weak. For these nearly free electrons, the
crystal potential may even be treated as a perturbation. We demonstrate below that
the wave function of the electron is then close to a plane wave, and that the energy
varies to a first approximation as it would for free electrons. However, even though
the perturbation due to the periodic potential may be weak, it may become important
for specific values related to the crystal periodicity. This idea will lead to a more
general definition of the first Brillouin zone.
For a free electron gas of volume , the Hamiltonian
H0 = p2 /2m0
has plane wave eigenfunctions
ψk (r) = −1/2 eik·r , (3.9)
with associated energy

εk = h̄2 k2 /2m0 . (3.10)
In one dimension, the band structure thus reduces to a simple parabola. In a crystal,
the one-electron Hamiltonian also includes the potential due to the ions given in
(1.7):

Nn
V(r) = Vat (r − Rl ) . (3.11)
l=1
We assume here that the potential is very weak, and can thus be treated as a per-
turbation in comparison with H0 . We know that, if we consider the non-degenerate
states of the unperturbed Hamiltonian, the effect of the perturbation is to alter the
energy levels and wave functions in the following way [2, Chap. 9]:
| k|V|k |2
Ek = εk + k|V|k + +··· , (3.12)

εk − εk
k =k
k|V|k
|k̃ = |k + |k + · · · . (3.13)
ε
k
− ε k
k=k
These results are obtained by treating the perturbation to first and second order and
assuming that | k|V|k | |εk − εk |, i.e., that the perturbation only couples states
with widely separated energies.
To first order in perturbation theory, the energy shift is

1 1
k|V|k = e−ik·r V(r)eik·r d3 r = V(r)d3 r . (3.14)

The energies of the plane waves |k, which are particular Bloch functions with
uk (r) ≡ 1, are simply shifted rigidly by an energy that is independent of the state
under consideration, this energy being the average of the periodic potential. This
effect can be neglected, since it does not change the parabolic shape of the variation
of Ek as a function of |k|.
Let us consider the case of a linear chain, where it will be easier to represent
Ek graphically. The existence of the periodic potential means that periodic energy
eigenvalues must be obtained in reciprocal space. To first order in perturbation the-
ory, by continuity with the case where the potential is identically zero, the electron
states of the free electron parabola must be translated by all lattice vectors p(2π/a)x
(with p an integer) of the reciprocal lattice. This generates curves Ek that are peri-
odic in reciprocal space. We thus obtain the extended zone band structure in k space
(see Fig. 3.2a). We can also focus on the primitive unit cell [−π/a, π/a] of recipro-
cal space which we called the first Brillouin zone, thereby obtaining the restricted
zone band structure, in which the parabola appears to be folded up (see Fig. 3.2b),
to generate all the values of the energy for k ∈ [−π/a, π/a].
According to (3.12), the potential only affects the energy of a state |k to second
order if it couples it to other states |k with similar energies by a nonzero matrix
element k|V|k . To determine this matrix element, we use the fact that the poten-
tial V(r) is periodic. In one dimension, V(x) is thus expanded in a Fourier series
containing all the harmonics of (2π/a)x :
+∞

V(x) = Vp(2π/a) eip(2π/a)x . (3.15)
p=−∞
E(k) E(k)
(a) (b)
−4 π −2 π − π 0 π 2π 4π ka − π 0 π ka
Fig. 3.2 In the presence of a very weak periodic potential, the energies of the free electrons are not
modified. However, in the quasi-momentum space, (a) the 1D parabolic band is translated in such a
way as to reveal the periodicity associated with the potential (extended zone band structure). (b) If
we restrict to the first Brillouin zone, we obtain an E(k) diagram in which the parabola is folded up
(restricted zone band structure)
The terms in this expansion are nothing other than the products K · x, where K is
a vector of the reciprocal lattice. This Fourier series generalises for 3D periodic
functions, and the potential is then described by a Fourier series involving all the
vectors K of the reciprocal lattice:

V(r) = VK eiK·r , (3.16)
K
with V−K = VK ∗ , since V(r) is real. We thus have

1
k|V|k = VK ei(k −k+K)r d3 r = VK δ(k − k + K) . (3.17)

K K
This matrix element is only nonzero for k = k − K. We may thus write (3.12) and
(3.13) in the form
|VK |2
Ek = εk + , (3.18)
εk − εk−K
K=0
VK
|k̃ = |k + |k − K . (3.19)
εk − εk−K
K=0
As indicated above, these expansions are only valid if |VK | |εk − εk−K |. In par-
ticular, if εk ∼ εk−K , the states are quasi-degenerate and the above perturbation
expansion is no longer applicable. Note that the condition εk = εk−K on the states
k, viz., k2 = (k − K)2 , is obtained for
2k · K = K2 , (3.20)
which is satisfied if k belongs to the orthogonally bisecting plane of K, i.e., if it

belongs to a Bragg plane. In this case, even a very weak periodic potential will
have a significant effect on the electron energy levels. This is not surprising, be-
cause we know from the discussion in Chap. 2 that a plane wave with a vector
k belonging to a Bragg plane will be diffracted. We shall see more precisely in
Sect. 3.2 how this situation arises by carrying out a detailed calculation in the
case of the linear chain. In one dimension, the Bragg condition is satisfied when
k = mπ/a, and it is clear from Fig. 3.2a that, in these cases, the levels corre-
sponding to two translated branches of the initial free electron parabola are indeed
degenerate.
3.1.4 First Brillouin Zone
The Bragg planes are thus of singular importance for understanding the electronic
structure of a solid. These planes, which orthogonally bisect the reciprocal lattice
vectors taken from the origin, bound a particular primitive cell of the reciprocal
lattice, namely the Wigner–Seitz cell constructed around the origin, as defined in
Chap. 2. It is a primitive unit cell of the reciprocal lattice which specifies the first
Brillouin zone. Recall that it contains all points of reciprocal space that are closer
to the origin k = 0 than to any other reciprocal lattice node. It contains all points k
such that the line segment Ok does not intersect any orthogonally bisecting plane of
the vectors of the reciprocal lattice leaving O. The volume of the first Brillouin zone
is the same as that of any primitive unit cell of the reciprocal lattice, i.e., (2π )3 /v,
where v is the volume of the primitive cell in the direct lattice. In one dimension, it
corresponds to the interval [−π/a, π/a].
Question 3.1. Find the expression for the vectors of the reciprocal lattice in two dimen-
sions. Determine the reciprocal lattices for the hexagonal, centered rectangular, and oblique
lattices, together with their first Brillouin zones.
The first Brillouin zone of the cubic lattice, e.g., CsCl (see Fig. 2.8a), is a cube of
side 2π/a with center at the origin k = 0 (see Fig. 3.3a). The first Brillouin zone for
a face-centered cubic Bravais lattice, e.g., Si, Ge, GaAs, is shown in Fig. 3.3b. If a
is the edge of the cube in the actual crystal and the origin of the reciprocal lattice,
the points of symmetry of the reciprocal lattice shown in the figure are respectively
(π/a, π/a, π/a) for L and (2π/a, 0, 0) for X. The coordinates of equivalent points
under the symmetries of the cube are easily deduced.
kz kz
2π / a
L
ky Γ
0 kx
X ky
K
kx 4π /a
(a) (b)
Fig. 3.3 (a) Reciprocal lattice of the cubic lattice with side a. The first Brillouin zone, a cube
of side 2π/a, is illustrated. (b) The body-centered cubic lattice is the reciprocal lattice of
the face-centered cubic lattice. The first Brillouin zone is shown. This is the region bounded by
the orthogonally bisecting planes of the vectors joining the centre Γ of the cube to its vertices, and
by the faces of the cube, which are the orthogonally bisecting planes of the vectors joining Γ to
the centers of the neighbouring cubes
3.2 Band Structure of Solids
3.2.1 Linear Chain
To obtain a better understanding of the energy level distribution of the Bloch states
for a linear chain, in Sect. 3.2.1a, we continue the perturbation calculation for k
satisfying the Bragg condition, which is a Bragg point for the 1D case. (In 2D, this
would be a Bragg line, and in 3D, a Bragg plane, as we have already referred to
it.) We will then be able to show in Sect. 3.2.1b that a range of forbidden energies
appears in the distribution of energy levels. Finally, in Sect. 3.2.1c, we shall see how
to make a smooth transition from a situation in which the atomic potential is weak
to one in which it is strong.
a. Perturbation Calculation for a Bragg Point
Consider the degenerate eigenstates corresponding to the Bragg points k = ±π/a,

with energy ε0 = π 2 h̄2 /2m0 a2 and wave functions
ψπ/a = L−1/2 eiπ x/a , ψ−π/a = L−1/2 e−iπ x/a ,
where L is the length of the chain. To find out how these eigenstates are modified, we
keep only those terms corresponding to the Fourier series expansion of the potential,
viz.,
3.2 Band Structure of Solids 61
V(x)
(a) a x
(b) V1(x) x
ρ
| ψ − |2
(c) x
| ψ + |2
(d) x
Fig. 3.4 (a) Position dependence of the electrostatic potential energy of an electron subject to
the Coulomb field of the ions in a linear chain of atoms. (b) Potential limited to its Fourier
coefficients obtained for k = 2π/a and k = −2π/a, viz., V1 (x) given in (3.21). (c) and (d)
Probability of finding the electron in different positions for the states ψ− and ψ+
V1 (x) = V2π/a exp(i2π x/a) + exp(−i2π x/a) = 2V2π/a cos(2π x/a), (3.21)
where V2π/a is negative because the potential has minima at the positions of the
ions (see Fig. 3.4b). In the basis of states ψπ/a and ψ−π/a , the Hamiltonian takes
the matrix form
ε0 V2π/a
H1 = . (3.22)
V2π/a ε0
This has eigenstates

1 2 π
ψ− = √ ψπ/a + ψ−π/a = cos x , with E− = ε0 − |V2π/a |, (3.23)
2 L a

1 2 π
ψ+ = √ ψπ/a − ψ−π/a = i sin x , with E+ = ε0 + |V2π/a |. (3.24)
2 L a
The potential V1 (x) thus removes the degeneracy of the levels ε0 , and the wave func-
tions shown in Fig. 3.4 correspond to stationary waves. The probability of finding
the electron near an ion is increased for the state ψ− with lowest energy.
b. Band Structure of Nearly Free Electrons
The above perturbation calculations show that the only free electron states to be
modified significantly by the weak periodic potential are those close to a Bragg
plane, i.e., with the property that k ∼ pπ/a for the linear chain. The perturbing
potential only couples states for which the vectors k differ from a vector of the
reciprocal lattice, i.e., from p2π/a for the linear chain. In Fig. 3.5a, double-headed
arrows indicate states coupled by the term V1 (x) in the potential. For k π/a, (3.18)
gives the change in Ek :
|V2π/a |2 |V2π/a |2
Ek = εk + + . (3.25)
εk − ε(k+2π/a) εk − ε(k−2π/a)
Here the first correction term can be neglected as its energy denominator corre-
sponds to free electron states far apart in energy. The second correction term is
negative for k π/a and positive for k π/a, which results in a repulsion between
the states in the first two bands relative to the free electron parabola, as shown in
Fig. 3.5. For k → π/a, Ek tends to the values E− and E+ determined in the case
where the states are degenerate.
In the energy interval
ε0 − |V2π/a | < E < ε0 + |V2π/a | , (3.26)
there is no stationary state of the energy. The presence of the periodic potential
V1 (x) forces open a band gap near k = ±π/a. Figure 3.5 shows how the degeneracy
is removed in the unfolded and folded band structures.
Note that by applying the same argument to the other degenerate cases corre-
sponding to the states k = ±pπ/a, and taking into account the order p Fourier com-
ponents of the potential
(a) (b)
E(k) E(k)
2|V6π /a |
2π 2|V4π /a |
2 1 2|V2π /a |
−2π −π 0 π 2π ka −π 0 π ka
Fig. 3.5 Energies of the electron states for a linear chain in the approximation where the electrons
are assumed to be nearly free: (a) extended zone and (b) restricted zone band structure. The free
electron parabolas in Fig. 3.2 are shown as dashed curves, while the continuous curves take into
account the effect of the crystal potential as a perturbation. In (a), the states coupled by the Fourier
component V1 (x) of the potential are joined by a double-headed arrow: the first (1) couples states
that are non-degenerate in the absence of the periodic potential, while the second (2) couples states
that are degenerate in the absence of V1 (x). Band gaps appear. Their half-widths are the Fourier
components = 1, 2, 3 of the periodic potential
2π
Vp (x) = 2Vp2π/a cos px , p integer , (3.27)
a
we obtain band gaps for the energies
p2 ε0 − |Vp2π/a | < E < p2 ε0 + |Vp2π/a | . (3.28)
This calculation, carried out only in the 1D case for didactic reasons, shows that the
periodicity modifies the parabolic dispersion relation of the free electron, causing
band gaps to appear at the points which correspond in the reciprocal space to the
midpoints of vectors in the reciprocal lattice.
These marked singularities in the band structure, obtained for states correspond-
ing to the vectors k close to a Bragg plane, arise because the corresponding plane
wave states suffer a Bragg diffraction on the periodic potential of the ions, even if
this potential is very weak. The eigenstates for the wave vectors (3.23) or (3.24) are
no longer plane waves but stationary waves, obtained by diffraction.
c. General One-Dimensional Case
It thus remains to understand how the above approach based on the idea of nearly
free electrons can be related to the tight-binding approximation discussed in Chap. 1.
Naturally, the energy eigenvalues for the model in the last section can be obtained
numerically to any required accuracy, whatever the amplitude of the periodic po-
tential. The results of such a calculation, in which only the term V1 (x) has been
retained, are shown in Fig. 3.6.
We observe that, for a weak potential, the results are indeed those obtained us-
ing the nearly-free electron approximation. However, for a stronger potential, the
shape of the lower band approaches the cosine shape obtained in the tight-binding
approximation. For even stronger potentials, this trend becomes more pronounced
and extends to higher bands, provided we take into account the Fourier components
Vp (x) of the potential. We thus move smoothly from the case of nearly free electrons,
corresponding to a weak atomic potential, to the LCAO approximation, which cor-
responds to a strong atomic potential.
Fig. 3.6 Folded band struc- 3

ture according to the model
Ek /ε0
2
(3.21) in the first Brillouin
zone for V2π/a = 0.2ε0 (left) 1
and for V2π/a = 0.8ε0 (right).
The free electron parabola is 0
shown by a dashed curve. Ek −π 0 π ka
is even
3.2.2 Band Structure of Two-Dimensional Solids
We now extend the tight-binding and nearly-free electron methods (Sects. 3.2.2a
and b, respectively) to the case of more realistic solids than the linear atomic chain.
The notion of reciprocal space becomes essential here for representing the structure
of the energy states in two dimensions. The fact that the quantum numbers char-
acterising the Bloch state have two components apart from the band index leads to
specific spatial representations. We also show in Sect. 3.2.2c that the existence of
several directions in k space can lead to band overlap, something that did not arise
in the 1D case.
a. Tight-Binding Approximation
We consider a square plane lattice with one atom per primitive cell. Its reciprocal
lattice is also square with side 2π/a. The first Brillouin zone is in this case simply
a square of side 2π/a centered at O. It is easy to reproduce the LCAO type of
calculation in this case, taking into account the fact that the Bloch wave functions
are specified in the 2D k space. The wave function is therefore
1
|ψk = √ ak,l |Rl . (3.29)
Nn l
If ψk is required to be a Bloch function (3.2), the coefficients ak,n can be written in

the form
ak,n = eik·Rn . (3.30)
The general eigenvalue equation for the Ek obtained in (1.19) is clearly still valid,
i.e.,
− tn,l ak,l = (Ek − Ev )ak,n , (3.31)
l
with the usual general expression for the hopping integrals:
−tn,l = Rn |Vat (r)|Rl . (3.32)
1. As in the 1D case, we neglect all the hopping integrals beyond nearest neigh-
bours, i.e., for which |Rn − Rl | > a.
2. We take into account the symmetry of the crystal. In particular, the periodic
boundary conditions are such that the crystal is perfectly invariant under transla-
tion. The tn,l then only depend on the vector Rn − Rl .
3. In addition, the potential Vat , and often the wave functions too, are invariant
under rotation, something we shall assume here.2 The tn,l then depend only on
2 Note that for orbitals with nonzero angular momentum, one cannot assume that the wave function
is invariant under rotations. From the point of view of computation time, the changes this brings to
our analysis can be significant, but the results are not fundamentally different.
the length of the vector Rn − Rl . In our example, there is then only one nonzero
hopping integral, apart from t0 , which we denote by t1 .
Equation (3.31) now becomes

−t1 ak,n+d = (Ek − Ev + t0 )ak,n , (3.33)
d
where the sum over d concerns only the nearest neighbours of the site n :
d = ±ax , ±ay . (3.34)
Putting the expression (3.30) for the ak,n into (3.33), we immediately obtain the
eigenvalues as a generalisation of the 1D case:
Ek = Ev − t0 − 2t1 (cos kx a + cos ky a) . (3.35)
The surface representing this function E(kx , ky ) is shown in Fig. 3.7. The con-
stant energy curves, found by intersecting this surface by horizontal planes, are
projected onto the base plane. Their general shape in the first Brillouin zone is
shown in Fig. 3.7. For small values of |k|, the cosine functions can be expanded
to second order in a Taylor series, and the function E(kx , ky ) approximated by a sim-
ple paraboloid. The constant energy curves are then circles. Likewise, the constant
energy curves close to the energy maximum of the band are circles centered on the
points (±π/a, ±π a). Note that the band determined like this using the tight-binding
approximation has width 8t1 . Such 2D systems are encountered in high-Tc cuprate
superconductors.3
2
E(k)
1
ky a
3 −1
2
1 −2
0
ky a
–3
–2 –1
–1 −π
0 1
–2 −π −2 −1 0 1 2 π
2 –3
kx a 3 kx a
Fig. 3.7 Energy band for the square lattice in the tight-binding approximation. 3D representation
of E(k) (left) and constant energy curves obtained as projections on the (kx a, ky a) plane (right)
3 See Problem 3: Band Structure of YBa2 Cu3 O7 .

b. Nearly Free Electrons in Two Dimensions
We now consider the 2D generalisation of the band structure calculation, using a

periodic potential that is simplified to its first Fourier components in the x and y
directions:

p2 2π x 2π y
H= + 2V2π/a cos + cos . (3.36)
2m0 a a
When the potential is zero, the surface ε0 (kx , ky ) = h̄2 k2 /2m0 is a simple paraboloid
of revolution, and the constant energy curves are therefore circles. If the potential is
weak enough to use a perturbation method, the eigenvalues E(kx , ky ) only differ from
ε0 (kx , ky ) for values of k close to a Bragg plane. We thus see that the circles with
constant energy Ei are unchanged unless they intersect a Bragg plane, i.e., unless
they hit the first Brillouin zone. However, when the free electron circle of energy Ei
cuts a Bragg plane for k = ki , a forbidden energy zone (band gap) appears around
Ei in the direction ki . This gives rise to a distortion of the constant energy curve,
which deviates from the circle as shown in Fig. 3.8, with ever greater deviation as
the matrix element of the potential increases. These perturbations will occur for all
intersections with the Bragg planes, as happened in the 1D case.
ky
ki
kx
Fig. 3.8 Constant energy curves for a square lattice when the electrons are assumed to be nearly
free. Close to ki , the free electron circle is distorted
This suggests identifying the orthogonal bisecting planes of all the vectors of the reciprocal
lattice taken from the origin and generalising the idea of the Brillouin zone. While the first
Brillouin zone corresponds to the region of k space such that, moving out from k = 0, one
crosses no orthogonal bisecting plane of reciprocal lattice vectors, the second, third, and higher
order Brillouin zones correspond to the region of reciprocal space whose points are obtained
by crossing only one, two, or more orthogonal bisecting planes of reciprocal lattice vectors, re-
spectively. These different Brillouin zones are all rather special primitive cells of the reciprocal
lattice. Any of them can be used to represent the band structure. This can be illustrated in the
1D case, where the first Brillouin zone is (−π/a, π/a), the second comprises two segments,
viz., (π/a, 2π/a) and its mirror image (−2π/a, −π/a), and so on. We may represent the folded
band structure in the first Brillouin zone, but we may also represent the first energy band in the
first Brillouin zone, the second in the second Brillouin zone, and so on. In such a representa-
tion, which corresponds to the bold lines in Fig. 3.5a, it is easier to recognise the free electron
parabola. This unfolded representation in several Brillouin zones is thus more useful for repre-
senting a band structure obtained using the nearly-free electron approximation. The geometric
construction of the Brillouin zones for the square plane lattice is shown in Fig. 3.9.
Fig. 3.9 First (center square), 1

second (light grey), and third 2
(dark grey) Brillouin zones
for the square plane lattice.
The Bragg lines correspond-
ing to the first, second, and
third neighbours are indicated
by 1, 2, and 3 3
c. Forbidden Energy Band?
If E(k) is represented in a particular direction Ok of the (kx , ky ) plane, we obtain

something like the result shown in Fig. 3.6 for the chain, i.e., a band gap appears at
the point belonging to the first Brillouin zone. However, the energy position of this
band gap is not the same in all directions of the (kx , ky ) plane. This can be understood
from Fig. 3.10, where we have represented the results of a numerical calculation
carried out for different values of the perturbing potential V2π/a . Note that, as in
the 1D case, for a given vector ky , a band gap appears at kx = π/a. However, if
we look at the band structure as a whole, we see that, for a weak potential V1 (see
Fig. 3.10a), the maximum of the lower band is higher than the minimum of the
upper band. In this case there are therefore some states at all energies, and hence
no band gap, something that could not happen in the 1D case. A true band gap
can nevertheless be obtained by considering stronger potentials, as can be seen in
Fig. 3.10b.
E E
3 3
2 2
1
1
0
0
π ky a 0 kx a π π ky a 0 kx a π
(a) (b)
Fig. 3.10 The folded band structure Ek of the 2D model (3.36), for weak (a) and strong (b) periodic
potentials. The energy E is given in units of ε0 = (h̄2 /2m0 )(π/a)2
The absence of a band gap for weak potentials can also be illustrated in the representation
of Fig. 3.8, which shows the different Brillouin zones. When the circle of constant energy Ei
cuts the first Brillouin zone, some of its circular arcs will be in the second Brillouin zone. This
means that there are states of energy higher than Ei in the first Brillouin zone and some states of
energy lower than Ei in the second Brillouin zone. It thus shows that the two energy bands rep-
resented in the first and second Brillouin zone actually overlap. There is therefore no band gap
between these two bands. However, note that, if the perturbing potential is strong, the removal
of degeneracy at the edge of the Brillouin zone may be sufficient for the constant energy curves
to remain entirely within the first Brillouin zone, as happens in the LCAO approximation (see
Fig. 3.7). In this case, a band gap will appear.
Note that this result, obtained here by comparing the energies of the first two bands,
can be repeated for the other bands corresponding to higher energy levels. The result
depends in each case on the amplitude of the corresponding Fourier component of
the potential. The main conclusion here is that, in two or three dimensions, only a
strong enough periodic potential will be able to generate a band gap, in contrast to
the 1D case, where an arbitrarily weak potential will be able to do that. This obser-
vation will be important when we come to discuss the metal or insulating nature of
certain 2D or 3D compounds.4
3.2.3 Three-Dimensional Band Structures
There is no great difficulty in going from two to three dimensions, except that a
description of the E(k) surfaces would now require four dimensions. However, the
constant energy surfaces can easily be represented in a 3D space.
a. Constant Energy Surfaces
It is straightforward to generalise the tight-binding approximation to the case of a

cubic monatomic solid. By analogy with the 2D square lattice, each atom has six
nearest neighbours located at d = ±ax, ±ay, ±az, and the energy band is given by
Ek = Ev − t0 − 2t1 (cos kx a + cos ky a + cos kz a) . (3.37)
4 When there is no band gap, the effects of the band structure can nonetheless be observed through
the optical response (see, e.g., Problem 2: Reflectance of Aluminium and Problem 5: Optical Re-
sponse of Monovalent Metals). The change in the total energy of the electronic states pertaining to
a band can in some cases influence the atomic structure of a metal (see, e.g., Problem 4: Electronic
Energy and Stability of Alloys and Problem 6: One-Dimensional TTF-TCNQ Compounds).
Fig. 3.11 Constant energy surfaces for the simple cubic lattice, in the first Brillouin zone:
E = Ev − t0 − 2t1 (left) and E = Ev − t0 (right)
Figure 3.11 shows several constant energy surfaces. We see that, by increasing the
energy, we move from a simple sphere to more complex shapes.
As only few chemical elements form a simple cubic lattice, we consider instead
the body- or face-centered cubic structures. In the
√body-centered cubic lattice, each
atom has eight nearest neighbours at a distance a 3/2 and the displacement vectors
to these eight neighbours are given by
a
d = (±x ± y ± z) . (3.38)
2
In√the face-centered cubic structure, the distance between nearest neighbours is
a 2/2, and each atom of the conventional unit cell has twelve nearest neighbours,
given by the displacement vectors
a a a
d = (±x ± y) , (±x ± z) , (±y ± z) . (3.39)
2 2 2
The band structures are now determined by a similar argument to the one used in
the last section. The only difference concerns the sum over the vectors d in (3.33),
which is now specified by the different possibilities given by (3.38) or (3.39). The
coefficients of the wave functions are still given by (3.30), and we then obtain
kx a ky a kz a
Ek = Ev − t0 − 8t1 cos cos cos , (3.40)
2 2 2
for the body-centered cubic structure, and

kx a kya kx a kz a ky a kz a
Ek = Ev − t0 − 4t1 cos cos + cos cos + cos cos ,
2 2 2 2 2 2
(3.41)
for the face-centered cubic structure.
In most of the cases studied here (square lattice, simple cubic lattice, body-
centered cubic lattice), the total band width, that is, the difference between the
maximum and minimum of Ek , is equal to 2zt1 , where z is the number of nearest
neighbours of an atom. There follows an empirical rule which is generally satisfied,
with a few exceptions: the more neighbours there are, the broader the band will be.
Any remarks on the nearly-free electron approximation go over from two to three
dimensions.
b. Representing the 3D Band Structure
For a 3D solid, the relations E(k) cannot be fully represented in a 3D space. How-
ever, an incomplete representation which allows a full appreciation of the properties
of the band structure can be obtained by graphing En (k) for values of k along par-
ticular axes of the reciprocal lattice. The preferred axes for such a depiction are
the axes of symmetry of the first Brillouin zone. Consider for example the case
of germanium (Ge) with structure (Ar) 3d10 4s2 4p2 , which crystallises into the di-
amond cubic structure with lattice constant a = 5.66 Å, shown in Fig. 2.7b. The
Bravais lattice has the face-centered cubic structure. The first Brillouin zone of its
4
Ge
2
kz
0
−2
Energy (eV)
L −4
Γ
kx
ky −6
X K
−8
4π /a −10
−12
L Γ X
2π 1 1 1 2π
, , (0, 0, 0) (1, 0, 0)
a 2 2 2 a
(b)
Fig. 3.12 (a) First Brillouin zone of germanium. (b) Energy bands of the valence electrons of Ge
represented in the first Brillouin zone for the two directions L and X of (a)
body-centered cubic reciprocal lattice is illustrated in Fig. 3.12a. The curves E(k) in
the directions of the axes L and X are shown in Fig. 3.12b for the energy bands
corresponding to the electrons in the 4s2 4p6 orbitals.
c. Density of States
The density of states serves the same purpose in the 2D and 3D cases as it did for
the linear chain. The aim is to be able to transform any sum of an arbitrary function
g over values of k into an integral over the energies:

Nn
F= g(Ek ) = dd k g(Ek ) = Nn dE g(E) D(E) . (3.42)
ΩBZ1 BZ1
k
Here the vectors k range over the first Brillouin zone, of volume ΩBZ1 . The re-
placement of the sum by an integral in the second equality is valid in the thermo-
dynamic limit Nn → ∞ for which the allowed values of k become continuous, and
the third equality defines the density of states per atom D(E). Finally, the volume
element in k space is the one corresponding to the appropriate dimension d = 2
or d = 3.
The density of states D(E) can in principle be calculated as in Chap. 1. However,
in most cases, and in particular for the lattices we have been considering, an analytic
evaluation is not possible. Figure 3.13 shows the density of states for the square
and simple cubic lattices. Note the step-shaped singularity at the edge of the band
for the square structure, and a square root singularity for the cubic lattice. These
singularities are of the same type as those obtained for a free electron gas close to
k = 0. The singularity in the middle of the band, in the 2D case, is a logarithmic
divergence which corresponds to the saddle points of the constant energy surfaces
in Fig. 3.7. In the 3D case, D(E) has gradient discontinuities. These singularities in
the band structure are called Van Hove singularities.
0.25
0.20
0.15
|t1|D(E)
0.10
Fig. 3.13 Density of states 0.05

per atom for the square lattice
(continuous curve) and the 0.00
simple cubic lattice (dashed –5 0 5
curve) in the LCAO approxi- (E−Ev+t0)/|t1|
mation
3.3 Metals, Insulators, and Semiconductors
In the last section, we discussed the eigenstates of an electron subject to a periodic

potential in a crystal. In real crystals, there is a huge number of electrons, of the
order of the Avogadro number, and this must clearly be taken into account for any
question involving the electronic properties of the solid. With the approximation
made here, which considers the electrons as independent, we know all the electron
states, and the methods of statistical physics for fermions outlined in Sect. 3.3.1
can be used to determine the thermodynamic properties of the crystal. Note that,
when the periodic potential is zero, we recover the situation for a free electron gas,
as treated in statistical physics, presented here as a model for the metallic state.
Applying Fermi–Dirac statistics to the Bloch states of the crystal, we will be able to
identify what fundamentally distinguishes insulating solids from metals and show
that, to a first approximation, the properties of metals generally arise from the prop-
erties of free electrons (Sect. 3.3.2). Finally, in Sect. 3.3.3, we will be in a position
to discuss insulating solids with a small band gap, namely, semiconductors.
3.3.1 Statistical Physics of Fermions
The thermodynamic properties at temperature T of a system of independent fermions

with quantised energy levels En (k) can be determined from the grand canonical par-
tition function
2
−β En (k)−μ
ZG (T, μ) = 1+e , (3.43)
n,k
where we use the standard notation β = 1/kB T, and μ is the chemical potential
which determines the average number of electrons. Note that we have taken into
account the spin degeneracy of the states En (k) by including the square of each
factor in the product. We now consider the grand canonical potential

−β En (k)−μ

A(T, μ) = −kB T ln ZG = −2kB T ln 1 + e . (3.44)
n,k
The various thermodynamic quantities are obtained by taking derivatives of A. For

example, the average number of electrons is
∂A(T, μ)
N(T, μ) = − . (3.45)
∂μ
At thermodynamic equilibrium, the average occupation number of an energy level
E is given by the Fermi function (see Fig. 3.14a)
1
f (E) ≡ . (3.46)
1 + eβ(E−μ)
3.3 Metals, Insulators, and Semiconductors 73
1.5 kz
1.0 kF
f(E) ky
0.5
0.0 kx
0.0 0.5 1.0 1.5
E/μ
(a) (b)
Fig. 3.14 (a) Fermi occupation function in the zero temperature limit (continuous curve), for βμ =
40 (dashed curve), and for βμ = 10 (dot-dashed curve). (b) Fermi sphere bounding the set of
occupied states in wave vector space for a free electron gas
Note that, at zero temperature, this function is equal to 1 for E < μ and zero for
E > μ. This means that, at T = 0, all states are occupied up to the value of the
chemical potential μ, which is called the Fermi energy EF . The thermodynamic
equilibrium state is simply the ground state in which all electron states of lower
energy are occupied, taking into account the Pauli exclusion principle.
This has a simple illustration in the case of free electrons. Their eigenfunctions
are plane waves (3.9), characterised by a wave vector k, momentum p = h̄k, and
purely kinetic energy ε(k) = h̄2 k2 /2m0 . Note that, in the space of wave vectors k,
the surface of constant energy EF which bounds the set of all occupied states is
in this case a sphere of radius kF = (2m0 EF /)1/2 /h̄. This so-called Fermi sphere is
shown in Fig. 3.14b.
Considering a box of volume Ω = L3 with the usual periodic boundary condi-
tions, the quantisation condition leads to one state per unit of volume (2π )3 /Ω in
wave vector space. If Ne is the total number of electrons in volume Ω, we obtain
kF = (3π 2 Ne /Ω)1/3 , viz.,
2/3
h̄2 3π 2 Ne
EF = lim μ = . (3.47)
T→0 2m0 Ω
Note that the quantities EF and kF are completely determined by the density of
electrons per unit volume.
Before considering the band structure of a real solid, we observe that, since the
expression for A(T, μ) contains a sum over the energies of the eigenstates, it can
be replaced by an integral in a representation involving DΩ (E), the total electron
density of states of the band structure per unit volume and per spin direction:
∞
A(T, μ) = −2ΩkB T ln 1 + e−β(E−μ) DΩ (E)dE . (3.48)

0
In particular, note that the density of states for free electrons has a simple expres-
sion if we observe that the states with energies between E and E + dE have wave
vectors in the volume 4π k2 dk, where dE = h̄2 kdk/m0 . The density of states for free
electrons in three dimensions thus has the form

1 2m0 3/2 1/2
D3 (E) = E . (3.49)
4π 2 h̄2
The analogous argument for free electrons in one or two dimensions leads to a den-
sity of states with the simple expression
d/2
2m0
Dd (E) = Cd E(d−2)/2 , (3.50)
h̄2
where Cd is equal to 1/2π , 1/4π and 1/4π 2 , respectively, for d = 1, 2, and 3.
3.3.2 Real Solids: Conductor or Insulator?
We now consider a real solid with a periodic potential, described by band theory.
Fermi statistics applies for the electron states, and in the ground state, at zero tem-
perature, the electrons occupy levels up to some maximum. To determine this max-
imum, consider for simplicity the case of a crystal with monatomic primitive cell.
In a solid with Nn primitive cells, one band contains Nn values of k, and each of
these states can hold two electrons, owing to the two possible spin orientations. As
a consequence, one band can contain up to 2Nn electrons.
Consider a monovalent element such as sodium (Na) whose band structure was
described in Chap. 1. In this case, bands due to deep electronic levels (see Fig. 1.9b)
are all occupied, whereas the energy band resulting from the valence level is only oc-
cupied by Nn electrons, and so is only half filled. For example, if we consider a band
obtained using the LCAO approximation for a linear chain, a representation of the
ground state which minimises the total energy is shown in Fig. 3.15. The states are
filled up to the Fermi energy EF which corresponds to the middle of the LCAO band.
E E
Ev−t0+2t1
k D(E)
EF
Ev−t0+2t1
π −kF +kF π
−a +
a
Fig. 3.15 Occupation of quantum states for a monovalent solid in its ground state, in a 1D repre-
sentation. Occupied states are represented by bold lines. Here the Fermi energy is EF = Ev − t0
It can be seen from Fig. 3.15 that, for electrons close to EF , unoccupied states
are available at arbitrarily low energy differences, as for the free electron gas. These
electrons can thus absorb electromagnetic energy at arbitrarily low frequencies. This
is indeed what would be expected in a metallic conductor.
Now consider a divalent chemical element. We then have 2Nn electrons to
distribute over the same number of quantum states, and the band is therefore
completely filled. Hence in this case the only accessible electron states lie in a higher
band whose states will still be vacant. The electrons cannot be excited with a low
frequency electromagnetic wave, exactly as would be expected in the case of an
insulator. We thus arrive at the following two propositions:
Partly filled band ⇐⇒ metallic conductor.
All bands completely filled (or empty) ⇐⇒ insulator.
Question 3.2. Consider the band structure of Ge described in Sect. 3.2.3b and illustrated in
Fig. 3.12. Check that the ground state of Ge at zero temperature does indeed correspond to
total occupation of a certain number of bands, and conclude that Ge is an insulator.
This analysis suggests that monovalent elements should all be metals, and this is
indeed the case, while all divalent elements should be insulators, which is clearly
false. In fact, all the alkaline earths, such as Be, Mg, and Ca, are metals. The flaw in
the argument comes from the assumption that the bands arising from the different
atomic levels have distinct energies. But we saw above that this will not always be
the case in two or three dimensions, since the high energy bands can overlap when
the atomic potential is weak enough.
We now examine the general situation with the help of Fig. 3.16. Imagine first a
crystal with very large lattice constant, so that the hopping integrals are negligible.
The electron energy spectrum of the solid will then simply be the atomic spectrum,
as shown on the left of the figure. If the lattice constant is reduced, hopping integrals
EF
2
Fig. 3.16 Expected change in EF1
the electron levels in a solid
when the hopping integrals t
are gradually increased, i.e.,
when the atoms are moved
closer together. This evolution
can be obtained in the oppo- E
site direction if the strength of
the atomic potential is gradu-
ally increased. The Fermi en-
ergies EF1 and EF2 correspond
to monovalent and divalent
solids, respectively t V
(abbreviated to t in the figure) will become significant, and each atomic level will
broaden into a band. For a monovalent solid, the last occupied band will be partially
occupied, with Fermi energy EF1 in the middle of the band, whereas for a divalent
solid, the last occupied band will be fully occupied. In this case, the Fermi energy
EF2 lies in a region where there is no quantum state, i.e., in a band gap.5
We thus recover the distinction between a metal and an insulator discussed ear-
lier. However, another situation may arise if the lattice constant is reduced still fur-
ther, thereby increasing t even more. In this case, the bands begin to overlap, as we
have seen in Fig. 3.10a, and the corresponding band gaps will disappear, as shown
on the right in Fig. 3.16. It is then energetically favourable to remove electrons from
states close to the maximum of the lower band and place them close to the mini-
mum of the higher band. Instead of a full band and an empty band, which would
give an insulator, we find two partially filled bands, i.e., a metal. It is this situa-
tion with overlapping bands that arises in the alkaline earth elements, which are all
metals.
Conversely, we may imagine gradually increasing the periodic potential V from
an initially low value. As we saw in Sect. 3.2, a weak potential does not create a
band gap in two or three dimensions, so we find ourselves in the situation corre-
sponding to the right-hand side of Fig. 3.16. The weak potential situation typically
prevails in dense polyvalent metals like Pb and In. When V increases, band gaps will
eventually appear, as in the middle of the figure. This is what happens in diamond,
silicon, germanium, and many semiconducting compounds, such as GaAs, GaP, InP,
and so on. It should be noted that an increase in the periodic potential corresponds
qualitatively to a decrease in the hopping integrals, and vice versa.
We may now formulate several conclusions regarding conduction properties. To
begin with, odd valence elements should always be metals, whereas even valence
elements can be either metals or insulators, depending on whether their bands over-
lap or not. Looking at the periodic table, we find that this conclusion is indeed cor-
rect in the vast majority of cases, generally with a metallic behaviour. The few ex-
ceptions (insulators with odd valence) concern substances like nitrogen or chlorine
in the solid state at low temperatures. These exceptions are easy enough to explain:
the basis of the crystalline solid is not in this case the isolated atom, but rather a
molecule, viz., N2 or Cl2 . The basis of the crystal thus carries an even number of
electrons, and is therefore an insulator.
This conclusion can be extended without difficulty to non-elementary crystals,
i.e., ones with a more complicated basis than a single atom. In the band theory
framework, if the basis contains an odd number of electrons, the result should be a
metallic state, while if it contains an even number, the result may be either a metal
or an insulator.
5 In an insulator, the Fermi energy is not defined by band occupation at T = 0. It must then be
defined, as for semiconductors, from the chemical potential EF = limT→0 μ. We shall see below
that EF lies within the band gap.
3.3.3 Introduction to the Semiconductor Case
Note that according to the above definition a solid is only strictly an insulator at
absolute zero temperature. Indeed, if all the bands are either filled or empty at zero
temperature, a band gap will separate the last occupied state from the first empty
state at zero temperature. However, for nonzero T, the Fermi factor implies that the
last states E Ev of the lower band, usually called the valence band, are now only
partially occupied, whereas the first states of the upper band E Ec , known as the
conduction band, are partially occupied. The bands will thus be partially filled, and
the material may be slightly conducting. It is when such conditions are fulfilled that
we may define a semiconductor to be a material that can conduct significantly at
room temperature.
Question 3.3. In the case of Ge, whose band structure is depicted in Fig. 3.12, indicate the
points in reciprocal space corresponding to Ev and Ec and the energy of the band gap.
For a pure (or intrinsic) semiconductor at T = 0, the number ne of electrons

occupying states in the conduction band must correspond exactly to the number
nh of missing electrons (or holes) in the valence band, in order to ensure electrical
neutrality. This condition determines the position of the chemical potential for all
T. Given the densities of states Dc (E − Ec ) and Dv (Ev − E) of these two bands, we
have
∞ Ev

ne = 2 f (E)Dc (E − Ec )dE , nh = 2 1 − f (E) Dv (Ev − E)dE , (3.51)
Ec −∞
where the factor of 2 accounts for the spin. It is easy to check that, provided that
kB T is much less than the width Eg = Ec − Ev of the band gap, electrical neutrality
requires the Fermi population factor to be very small for E > Ec and very close to
unity for E < Ev , whence the following approximations:
∞

ne = 2 exp − β(E − μ) Dc (E − Ec )dE = Nc (T) exp[β(μ − Ec )] ,
Ec
(3.52)
Ev
nh = 2 exp − β(μ − E) Dv (Ev − E)dE = Nv (T) exp[β(Ev − μ)] ,
−∞
where we have set

∞ 0
Nc (T) = 2Dc (E) exp(−βE)dE , Nv (T) = 2Dv (E) exp(βE)dE . (3.53)
0 −∞
Eliminating μ between relations (3.52), we thus obtain the general result
ne nh = Nc (T)Nv (T) exp(−βEg ) = n2i . (3.54)
In an intrinsic semiconductor, ni is the electron density in the conduction band,

equal to the missing density in the valence band. Note that, in this case, the chemical
potential is then obtained by equating the right-hand sides of the two relations in
(3.52), which yields
1 1
μ = (Ec + Ev ) + kB T ln Nv (T)/Nc (T) . (3.55)
2 2
When the densities of states of the conduction and valence bands are equal, called
the symmetric case, the chemical potential and hence also the Fermi energy lie ex-
actly in the middle of the band gap (see Fig. 3.17). In this case, for bands in three
dimensions with free electron density of states given by (3.49), we obtain
ne = Nc (T) exp(−βEg /2) , (3.56)
where
3/2 ∞
1 2m0 kB T
Nc (T) = exp(−u)u1/2 du . (3.57)
2π 2 h̄2 0
The number of electrons transferred from the valence band to the conduction
band varies exponentially with temperature, and is therefore very low as long as
kB T Eg , as might have been expected. For example, when Eg = 1 eV at room
temperature, i.e., kB T = 25 meV, ne ≈ 10−12 per primitive unit cell.
We thus discover that the ideal situation described above is difficult to obtain
in practice, because it can be totally changed in real materials by the presence of
sometimes uncontrollable impurities in comparable concentrations. We shall see in
Chap. 4 how this situation has in fact been exploited by purifying certain semicon-
ductors and then introducing selected impurities at controlled levels.
EF
Fig. 3.17 Densities of states
of the valence and conduction
bands in a pure semiconduc-
tor. According to (3.55), EF Ev Ec
lies close to the middle of the
band gap
3.3.4 The Special Case of Graphene
Graphene is an allotrope of carbon obtained from graphite by detaching a sin-

gle crystal plane, in which the carbon atoms are positioned at the vertices of a
honeycomb structure. As shown in Question 2.2, its Bravais lattice is hexagonal
and the primitive cell contains two atoms in positions corresponding to two sublat-
tices A and B. The latter only differ through the orientation of their three nearest
neighbours, which belong to the complementary sublattice (see Fig. 3.18).
Fig. 3.18 Atomic structure,

Bravais lattice (a1 ,a2 ), and a
primitive cell of the graphene a2
sheet, showing the two sublat-
tices A and B of carbon atoms a1
A B
The structure is stabilised by the fact that the 2s2 2p2 orbitals of a given carbon
atom form three hybridised sp2 covalent bonds oriented at an angle of 120◦ with the
neighbouring carbon atoms. The corresponding electron energy levels constitute the
saturated inner levels of the electronic structure of graphene. This is reminiscent of
what happens in the benzene molecule, but replacing the hydrogen atoms by carbon
atoms, whence the whole structure can be extended indefinitely.
As in benzene, the last electron of the carbon atoms is positioned in the 2pz or-
bital, perpendicular to the plane of the graphene sheet. It is the delocalisation of this
orbital that leads, by analogy with the π orbital of benzene, to the outer electron
bands in the electronic structure of graphene. They can thus be described using the
LCAO tight-binding approximation by considering one electron per carbon atom in
the 2pz orbital. The only complication with regard to the electronic structures so far
investigated is that the primitive cell contains two atoms in the sublattices A and B,
which leads to two energy bands.

Setting ϕjA = ϕ(r − Rj ) and ϕjB = ϕ(r − Rj − d) for the eigenstates of energy
E0 associated with the ‘atomic’ Hamiltonians of the two atoms in the primitive cell,
we seek the eigenstates of the total Hamiltonian in the form

ψk = √1 λA ϕjA + λB ϕjB eik·Rj . (3.58)
N j

Assuming that the orbitals ϕjA and ϕjB are normalised and orthogonal, and con-
sidering only hopping integrals between nearest neighbours (j, A) and (j , B),
A B
ϕj H ϕj = −t , (3.59)

we determine the energies of the eigenstates ψk of the Hamiltonian:
1/2
Ek± = E0 ± t 3 + 2 cos k · a1 + cos k · a2 + cos k·(a1 − a2 ) . (3.60)
Question 3.4.
Projecting
the Schrödinger equation with periodic potential onto the two ato-
mic states ϕlA and ϕlB , show that we obtain two coupled equations, the first of which is

λA E0 − λB t 1 + e−ik·a1 + e−ik·a2 = λA E . (3.61)
Determine the second equation, which leads to Eq. (3.60).

Fig. 3.19 Dispersion relation E(k) of the π and π ∗ electronic bands of graphene, constructed
from the pz electronic orbitals of carbon. Note the six points at the vertices of the hexagonal first
Brillouin zone where the valence and conduction bands touch one another.
Near these points the
dispersion relations have a conical shape for energies such that E − E0 < t. Image courtesy of
Goerbig, M.O.: Laboratoire de Physique des Solides, Orsay, France
The two energy bands of (3.60) are shown in Fig. 3.19 for vectors in the hexago-
nal reciprocal lattice with lattice constant 4π/3a, whose first Brillouin zone is the
hexagon in the plane E = E0 . Note that these two binding π and antibinding π ∗
bands do not overlap, but touch one another solely at six points of energy E = E0 .
These points of contact are obtained for vectors k such that the term in curly brack-
ets in (3.60) vanishes. It can be checked that they correspond to the six vertices of
the Brillouin zone. Since the primitive cell contains two electrons, the lower band is
fully occupied at zero temperature and constitutes the valence band, while the upper
band is completely unoccupied. The Fermi level is thus EF = E0 , and graphene can
be considered to be a semiconductor with strictly zero band gap.
Graphite, consisting of a stack of graphene sheets, has a similar electronic struc-
ture, with touching bands. However, in this case, the hopping integrals between
graphene planes lead to overlaps between 3D orbitals, and graphite is a conductor at
T = 0. It is thus a semi-metal. Note that graphene is also close to a semi-metal, since
many electrons are thermally excited as soon as T = 0. As we shall see in Chap. 4,
the rather special electronic structure of graphene leads to extremely unusual elec-
tronic transport properties.
3.3.5 Thermodynamic Properties of Metals
The thermodynamic properties of solids are determined by the quantum states pop-
ulated at nonzero temperature. They thus provide valuable information about the
excited states. Some such states correspond to collective vibrational modes of the
atoms, known as phonons. They are present in all solids, whatever their electronic
band structure. On the other hand, the excited states of electronic origin are radi-
cally different in metals and insulators, and this leads to a significant difference in
the thermodynamic properties. One of the quantities that reflects these properties is
the specific heat at constant volume Cv = dU/dT, which is the amount of energy
dU required to cause a temperature rise of dT in the solid. It is thus determined by
the change in internal energy associated with thermal population of excited states.
For example, for a semiconductor, the electron energy at a temperature T is in-
creased above the ground state energy by the energy of the states excited from the
valence band into the conduction band, viz., U(T) − U(0) ≈ ne Eg . In the case of a
symmetric semiconductor, as considered in the last section,

dne Nc Eg2 Eg
Cv ≈ Eg = exp − . (3.62)
dT 2kB T 2 2kB T
Since very few electrons are excited, due to the exponential factor, this contribution
to the specific heat is negligible compared with the phonon contribution6 which
goes as T 3 . Consequently, in semiconductors and insulators, the specific heat tells
us little about electronic properties. We shall see in Sect. 3.4 that optical methods
are better suited to measuring the width of the band gap. In contrast, in metals, many
energy states close to EF can be thermally excited, whence electron levels make a
large contribution to the specific heat.
a. Specific Heat of Metals
The Fermi–Dirac distribution implies that the electron states over a width ≈ kB T at
the Fermi level are excited by an energy ≈ kB T. For a density of states DΩ (E), the
increase in the internal energy is therefore

ΔU ≈ kB T kB TDΩ (EF ) , (3.63)
which corresponds to a term linear in the temperature, i.e., Cv ≈ 2kB2 TDΩ (EF ). The
exact calculation is straightforward using the grand canonical potential. To first or-
der in temperature, it leads to an electronic contribution to the specific heat that we
shall denote by Ce , given by
∂U 2π 2 DΩ (EF )kB2
Ce (T) = = T . (3.64)
∂T 3
This term must be corrected by higher order terms expressed as powers of kB T/EF .
This yields the Sommerfeld expansion, in which the first correction to the electronic
density of states goes as (kB T/EF )2 [3]. It is generally negligible for metals with
no particular singularities in the density of states at the Fermi level. A metal is thus
characterised by an electronic contribution to the specific heat that is linear in T at
low temperatures, viz., Ce = γ T, where, according to (3.64), the coefficient γ can
be used to determine the density of states at the Fermi level.
6 See Problem 9: Phonons in Solids.
b. Pauli Susceptibility of Metals
Another quantity that clearly distinguishes between metals and insulators is the
magnetisation induced by a magnetic flux density B = μ0 H applied in some di-
rection z. This removes the spin degeneracy of the Bloch states. The Zeeman term
HZeeman = −μB = 2μB Bŝz , (3.65)
shifts the energy of the spin-up state sz = 1/2 =↑ relative to the energy of the spin-
down state sz = −1/2 =↓. Consequently, the populations of the two spin states are
no longer equal, and at equilibrium, the solid can acquire a magnetisation.
Note that at T = 0, for a fully occupied band, the energy balance is zero and the
resulting magnetisation vanishes. In the band theory framework, the electron spins
do not induce magnetic behaviour: an insulator is essentially non-magnetic. How-
ever, in a metal, some bands are not fully occupied. For these bands, the energies
of all the spin-up states are increased in the presence of the field, whereas those
of the spin-down states are lowered by the presence of the field. It follows that, at
the Fermi level, the excess of spin down states constitutes a set spanning an energy
width 2μB B, and hence 2μB BDΩ (EF ) in number. The drop in energy induced by
applying the magnetic field is therefore
ΔU = −(μB B)2 DΩ (EF ) .
The corresponding magnetisation is M = −∂ΔU/∂B = 2μ2B BDΩ (EF ), and the mag-
netic susceptibility is χP = ∂M/∂B, given by
χP = 2μ2B DΩ (EF ) . (3.66)
In the corresponding Sommerfeld expansion, the first correction as a function of

temperature is also of order (kB T/EF )2 . It is generally negligible, because EF > 1 eV
for most metals. We may conclude that band theory predicts a paramagnetic be-
haviour for the susceptibility of a metal which is essentially independent of T. This
is commonly referred to as Pauli paramagnetism. Note that a measurement of the
Pauli susceptibility can be used to estimate the density of states at the Fermi level.
c. Experimental Corroboration
The above discussion leads to several simple conclusions about the thermodynamic
behaviour of solids in the framework of a band theory assuming independent elec-
trons. There are only two possible situations:
• In an insulator, EF is located in a band gap which separates totally occupied va-
lence bands from the totally empty conduction bands. The density of states is
zero at the Fermi level, and the electron contribution to the specific heat is neg-
ligible compared with the phonon contribution at any temperature. In addition,
insulators exhibit no spin magnetism.
• In a metal, one or more energy bands cross the Fermi level, thereby defining a
Fermi surface comprising one or more sheets and representing all the last energy
states occupied in k space. These electron states are easily excited thermally and
contribute a term linear in T to the specific heat of the metal. Moreover, they are
responsible for a Pauli spin paramagnetism that is independent of temperature.
The specific heat and the susceptibility are determined, for T TF , by the density of
states at the Fermi level, using the simple relations (3.64) and (3.66). Note that a free
electron model leads to similar relations for these two quantities Ce and χP , if we
simply replace DΩ (EF ) in these relations by D3 (EF ), as calculated in (3.49) for free
electrons.
For the specific heat, the T-linear electron contribution only dominates at low
values of T, so measurements must be made at temperatures below a few kelvin for
the phonon contribution to be negligible. The Pauli susceptibility, on the other hand,
can be measured over a broad temperature range.
To make quantitative predictions of the thermodynamic properties of a solid, we
must compute the density of states appearing in (3.64) and (3.66) to a certain level
of accuracy, which generally proves to be a rather difficult task. However, one can
use an experimental measurement of one or other of these properties in order to
determine DΩ (EF ). Measuring both provides a way of checking that band theory
itself is applicable. Indeed, these two relations predict a ratio
TχP (T) 3μ2

lim = 2 B2 , (3.67)
T→0 Ce (T) π kB
which depends only on universal constants and not on any specific property of the
solid under investigation. Any deviation from this constant necessarily indicates that
one or other of the approximations made in Chap. 1 is no longer valid.
Let us consider the monovalent alkali metals. The electron density in the highest
occupied energy band is clearly one per atom. The data in Table 2.1 can be used to
calculate the density per unit volume. Using (3.47) and (3.49), we deduce the density
of states at the Fermi level D3 (EF ), to be used when there is no crystal potential, i.e.,
for the case of a free electron gas. Furthermore, measurements of Ce and χP can be
used to find two other values for DΩ (EF ), which we shall denote by DC and Dχ ,
respectively. Table 3.1 gives the values of the reduced quantities dC = DC /D3 and
dχ = Dχ /D3 , together with the Wilson ratio RW , which quantifies deviation from
the prediction in (3.67):

TχP (T) 3μ2B
lim = RW ⇐⇒ RW = dχ /dC . (3.68)
T→0 Ce (T) exp π 2 kB2
We find that the deviations of the density of states from the predictions of the free
electron gas model are significant in most cases, although not excessive (there is no
number of order 10 in the table). In every case, the Wilson ratio is close to unity.
In a metal perfectly described by band theory, this ratio should be exactly unity.
Table 3.1 Reduced densities of states dC and dχ , together with the Wilson ratio RW for the alkali
metals
Metal dC dχ RW
Li 2.3 2.5 1.1
Na 1.3 1.7 1.3
K 1.2 1.5 1.3
Rb 1.3 1.6 1.2
Cs 1.5 1.7 1.1
Deviations are mainly due to interactions between electrons, which are neglected in
this model. Note, however, that for more complicated metals, in particular those in
the transition series (partially occupied 3d or 4d orbitals), the Wilson ratios can be
much greater than unity, revealing the greater importance of interactions between
electrons in these cases.
The classification into paramagnetic metal or non-magnetic insulator often proves
accurate. However, there are many very interesting exceptions from the point of
view of this book, in particular, ferromagnetic substances which may be either in-
sulators or conductors, but which exhibit a spontaneous magnetic moment that is
not predicted by band theory, or superconductors which are perfect conductors but
which are not paramagnetic and whose specific heat is not linear in temperature. To
understand these phenomena, we must abandon one or other of the approximations
discussed in Chap. 1.
3.4 Experimental Determination of Band Structures
As we have seen above, thermodynamic methods and magnetic measurements can

be used to estimate the density of states at the Fermi level in metals. But beyond this
rather limited information, what experimental methods can be used to obtain more
details regarding the density of states as a function of energy or the band structure
of a solid? Here we shall discuss some experimental techniques for determining the
band gaps in an insulator (optical absorption), the energy dependence of the density
of states at the surface of conducting materials (tunneling effect and microscopy),
and the dispersion curves E(k) (angle-resolved photoemission spectroscopy).
3.4.1 Optical Absorption in Insulators
If we are only concerned with the value of the band gap in an insulator, spectro-
scopic methods are relatively simple and effective. Indeed, atomic energy levels are
usually determined by absorption or emission spectroscopy. Similarly, if we shine a
light beam at frequency ν on an insulating solid, this beam will pass right through
the solid without absorption if the photon energy hν is lower than the band gap
3.4 Experimental Determination of Band Structures 85
energy Eg . Otherwise there will be absorption of photons from the beam. The ratio
of the outgoing to incident light intensity depends on the thickness x of the sample
according to
I(x)/I0 = e−αx ,
whence the optical absorption coefficient α may be determined. This absorption
becomes significant when hν = Eg . The optical absorption threshold thus gives the
width of the band gap in various semiconductors (see Fig. 3.20). The change in α(ν)
for different values of ν also gives information about the band structure. Indeed,
the existence of many interband transitions at certain energies, notably close to the
saddle points of E(k), leads to singularities in the absorption curve. This is illustrated
in Fig. 3.20.
Fig. 3.20 Optical absorption 107

coefficient α for various well
known semiconductors, mea- 106
sured at 300 and 77 K as a
Absorption coefficient (cm−1)
105 GaAs
function of the energy of the
incident photons Ge
104
Si
103
300 K
102
77 K
101
100
0.6 1 2 5 10
hv (eV)
3.4.2 Tunneling Effect and Scanning Tunneling Microscopy
Another very effective method for probing the density of states of a conducting ma-
terial exploits the electronic tunneling effect. The basic idea is to determine the elec-
tron current transmitted through a thin insulating barrier between two conducting
materials. This technique was initially used to study the electronic structure of su-
perconducting materials (see Chap. 7).
a. Tunneling Effect
It is well known that incident electrons can tunnel through a potential barrier when
the width d of the barrier is not too great. The transmission coefficient T of the
barrier is

2d
T ∼ exp − 2m0 (U0 − E) , (3.69)
h̄
where U0 is the height of the potential barrier and E is the energy of the incident
electron [2]. The tunneling current is only significant if the width of the barrier is
less than a few tens of Å.
If the barrier separates two normal metals, the chemical potentials are equal on
either side at thermodynamic equilibrium, and there is no flow of electrons. How-
ever, if a bias V is applied across the metals, the potential energy −eV lowers the
energy levels on the + side relative to those on the − side, and there will be electron
transport (see Fig. 3.21). We see that the electrons in metal 1 with energies between
EF and EF − eV can tunnel into states of the same energy which have become ac-
cessible in metal 2. If D1 (E) and D2 (E) are the densities of states of metals 1 and 2,
and if we take into account the occupation probabilities at nonzero temperature, the
tunneling current density from 2 to 1 is given by
+∞

j21 = AT D1 (E)f (E)D2 (E + eV) 1 − f (E + eV) dE , (3.70)
−∞
where A is a constant and f (E) is the Fermi–Dirac function. At nonzero temperature,

there is also a smaller current in the opposite direction (see Fig. 3.21) given by
+∞

j12 = AT D1 (E) 1 − f (E) D2 (E + eV)f (E + eV)dE . (3.71)
−∞
The total current density is therefore

+∞

j = AT D1 (E)D2 (E + eV) f (E) − f (E + eV) dE . (3.72)
−∞
We thus see that applying a bias simply lowers the energy levels in metal 2, whence
a group of electrons in metal 1 corresponding to occupied states with energies lying
E E
EF EF
–eV –eV
1 2
D1 (E) f (E) D 2 (E + eV) f (E + eV)
Fig. 3.21 Left: Energy levels at zero temperature in normal metals 1 and 2 with a positive bias V
applied between 2 and 1. Electrons can only get from 1 to 2 by tunnelling. Right: Occupied states
on either side of the barrier at nonzero temperature
in the range EF to EF − eV are able to tunnel through to metal 2. If the latter is the
material whose density of states we aim to determine, metal 1 is chosen to be a metal
with a density of states that hardly varies with E at the Fermi level, so that D1 (E) can
be replaced by D1 (EF ) to a good approximation. However, if D2 (E) varies rapidly
with E, the dependence of j on V will be nonlinear. Replacing D1 (E) by D1 (EF ) and
setting E = E + eV in (3.72), it is straightforward to check that
+∞
dj ∂f (E − eV)
= AT D1 (EF ) D2 (E ) dE . (3.73)
dV −∞ ∂V
At zero temperature, the derivative of the Fermi function is a δ function, and then
dj
= AT eD1 (EF )D2 (EF + eV) . (3.74)
dV
By measuring the dependence of dj/dV on V, we may thus deduce the density of
states D2 (EF + eV).
b. Application to Scanning Tunneling Microscopy and Spectroscopy
A modern setup uses a very finely tapered metal tip for metal 1, with nanoscale
dimensions. This is displaced in a controlled way just above the surface under
investigation. The lateral displacement of the sample is achieved by means of a
piezoelectric stage, whose x, y, z position is controlled by three applied voltages (see
Fig. 3.22). The vertical position determines the tunnel current between the tip and
sample for a given bias V between them. Using a feedback loop to fix the value of
the tunnel current, one can then measure the change in z as the tip moves across
the surface, thereby constructing a topographic image of the surface. This method
has been used, for example, to produce images of alkane molecules deposited on
graphite (see Fig. 2.1).
By measuring the tunnel current as a function of the bias at a given point on the
surface with fixed height z, we obtain the density of states at this point as a function
Piezoelectric
Feedback crystal
Fig. 3.22 Basic setup of a
scanning tunneling micro-
z
scope. The diagram only
shows how the displacement Computer
of the piezoelectric crystal is x
controlled. The tunnel current
V
between tip and sample can be y
used to control the z position.
Tip
For fixed z, the dependence of
Feedback I
the tunnel current on V can be
generator
used to find the local density
of states Sample
Fig. 3.23 (a) Carbon nan-

otube of diameter of about
10 Å. Image courtesy of (a)
Y. Ihara (Hokkaido Uni-
versity, Japan) (b) Upper:
Density of states deduced
from the tunnel characteristic
dj/dV observed with a scan- D(E) dj/dV
ning tunneling microscope.
Lower: LCAO calculation of (b)
D(E) for a model nanotube.
From Kim, P., et al.: Phys.
Rev. Lett. 82, 1225 (1999)
American
c Physical Society.
http://link.aps.org/doi/10.1103
/physRevLett.82.1225
–2 –1 0 1 2
V (volts)
of the energy. An example is shown in Fig. 3.23, which displays the electronic den-
sity of states at a point of a carbon nanotube. The filamentary structure of this system
induces Van Hove singularities in the density of states, similar to those calculated
in Chap. 1. The experimental observations are compared with a calculation of the
electronic structure using the LCAO approximation.
3.4.3 Angle-Resolved Photoemission Spectroscopy
In a photoemission experiment, one exploits the photoelectric effect. A photon with

enough energy hν removes an electron from an occupied state below the Fermi
level, i.e., a conduction electron for a metal or a valence electron for an insulator
or a semiconductor (see Fig. 3.24). This electron leaves the solid. Its initial state
is a Bloch state of the crystal, while its final state is that of a free electron. By
conservation of energy and momentum in the interaction process, the dispersion
relation ω(k) of the Bloch states can be reconstructed.
UV source Energy analyser

Fig. 3.24 Setup for a photoe-
mission experiment (ARPES).
UV photons excite electrons,
which are then emitted with h̄ω
e−
energy E in a direction α. By α
measuring E and α for the
emitted electrons, the curves Sample
E(k) can be deduced Detector
(C)
Fig. 3.25 Experimental determination by ARPES of the Fermi surface (C) of a CuO2 plane, rep-
resented in several primitive cells of the reciprocal lattice
This technique, known as angle-resolved photoemission spectroscopy (ARPES),

is widely used for 2D systems such as cuprate superconductors, whose structure was
presented in Fig. 2.9b. In these compounds, the electronic structure of a CuO2 plane
at the surface of a single crystal can be investigated. The Fermi surface can thus
be determined experimentally in several primitive cells of the reciprocal lattice, as
shown for Bi2 Sr2 CaCu2 O8 in the figure on p. 52. Part of the negative of this image
is shown magnified in Fig. 3.25, in which the point is taken as the origin of the
reciprocal lattice.
Question 3.5. 1. Specify the Bravais lattice of the CuO2 planes, together with the primitive
cell and its reciprocal lattice.
2. From the shape of the Fermi ‘curve’, is it possible to say whether the electronic structure
of the CuO2 plane can be described by the nearly-free electron approximation or rather
by a tight-binding approximation? Explain why.
3. Determine the number of charges per CuO2 plane on the basis of the experimental Fermi
surface.
4. Is this the expected number, if we assume that the atoms are in ionic form: Bi3+ , Sr2+ ,
Ca2+ , Cu2+ , and O2− ? If not, can you explain why there should be a difference?
Question 3.6. The second illustration in the preface shows a similar ARPES determination
of the Fermi surface of another 2D system, viz., Sr2 RuO4 , in which the RuO2 planes have
the same structure as the CuO2 planes in the cuprates. What can be concluded about the
band structure of this compound?
Fermi level E (eV)
D(E) k
Fig. 3.26 Band structure of valence and conduction electrons of copper, determined by photo-
emission. The results of numerical simulations are shown by continuous curves. The corresponding
density of states is shown on the left. From [6, p. 145] Springer
c 1995. Experimental results from
Courths, R., Hüfner, S.: Phys. Rep. 112, 55 (1984). Band calculation from Eckardt, H., Fritsche,
L., Noffke, J.: J. Phys. F 14, 97 (1984)
The band structure of a 3D system can also be very accurately determined using
ARPES. Figure 3.26 shows the experimental results for copper. The bands result-
ing from the 3d and 4s electrons of copper are shown for various orientations of
the vector k in the reciprocal lattice. Since copper is fcc, , X, K, L are the points
of the first Brillouin zone given in Fig. 3.3b. The experimental results (points)
agree very well with a numerical calculation of the band structure, which shows
the quality of the methods currently available for calculating band structures. As
we shall see later (see Chap. 8, and in particular Sect. 8.6), in magnetic systems for
which local electron correlations become relevant, the approximations of Chap. 1
will no longer be applicable, and more elaborate methods are currently being
developed.
3.5 Summary
Bloch’s theorem in three dimensions gives the general form of the wave function of
an electron subject to a periodic potential (Bloch function):
ψn,k (r) = eik·r un,k (r) , where un,k (r + Rl ) = un,k .(r) .

3.5 Summary 91
Each eigenstate is characterised by a vector k specified in the reciprocal lattice, and

it is periodic under any translation along a vector K of the reciprocal lattice. This
redundancy leads us to define the eigenstates in a primitive cell of the reciprocal
lattice. An important consequence of this theorem is that it is the vector k, called
the crystal momentum or quasi-momentum, which plays the role of a conserved
quantity in collisions with other particles.
It is interesting to consider the quantum states of an electron in a solid from the
opposite standpoint to the one adopted in the tight-binding approximation, i.e., con-
sidering a free electron gas perturbed by a weak periodic potential. This approach
applies for instance to the valence electrons which are the least tightly bound. The
electron states most perturbed by the periodic potential are those whose crystal mo-
mentum k lies close to a Bragg plane:
2k · K = K2 .
In this case, the degeneracy of the electronic levels is removed by the perturbing
potential. The difference between the levels is 2VK , twice the Fourier component of
the perturbing potential at the wave vector K.
This analysis brings out the role of the Bragg planes in the band structure of the
solid, and leads us to specify the electronic states in the Brillouin zones, primitive
cells of the reciprocal lattice bounded by the Bragg planes. In particular, the first
Brillouin zone corresponds to those points of the reciprocal space closer to k = 0
than to any other point of the reciprocal lattice.
It is then observed that, in one dimension, an arbitrarily weak potential introduces
a band gap into the spectrum. On the other hand, in two or three dimensions, the
periodic potential must have a certain minimal amplitude in order to generate a
band gap. If the strength of the potential is increased, we recover a very similar
band structure to the one obtained in the tight-binding approximation.
The full quantum state of a solid is obtained by filling the one-electron states
with all the electrons of the given solid, while observing the rules of quantum sta-
tistical mechanics for fermions. We thus obtain a rather simple result: either the last
occupied band (the one of highest energy) is only partially occupied, in which case
we are dealing with a metal, or the last band is fully occupied, in which case the
material is an insulator at T = 0.
In insulators, the last occupied levels of the valence band are separated from the
first empty levels of the conduction band by a region of forbidden energy called the
band gap. When this band gap is narrow enough ( 1 eV), a significant number of
electronic states in the conduction band are populated by thermal excitation, leading
to a weakly conducting state at room temperature. The material is then called a
semiconductor.
Graphene is a 2D layer of carbon in which the outer pz electrons form two elec-
tronic bands which touch one another at two points of the Brillouin zone. At T = 0,
one is filled and the other empty, which corresponds to a very special electronic state
of a semiconductor in which the band gap is strictly zero (or of a metal in which the
Fermi surface reduces to just two points).
In the band theory framework, insulators are non-magnetic, with negligible elec-
tronic contribution Ce to the specific heat, whereas in metals, the electrons contribute
a significant term linear in T which dominates at low temperatures. In addition, met-
als exhibit temperature-independent Pauli paramagnetism. These properties of the
metal are determined by the highest energy occupied electronic states, whose energy
EF defines the Fermi energy. This means that the Fermi surface, which represents
these states in reciprocal space, assumes a particularly important role, as does the
density DΩ (EF ) of these states. In particular,
2π 2 DΩ (EF )kB2
Ce (T) = T, χP = 2μ2B DΩ (EF ) ,
3
and band theory predicts a universal constant for the ratio between these two quan-
tities, which is indeed confirmed experimentally in many simple metals.
Several experimental methods can be used to study the electronic density of states
or the band structure of solids. Optical absorption is a good way of determining
band gaps, particularly for semiconductors. The electron tunneling effect between
a nanoscale metal tip and a conducting surface gives the energy dependence of the
electronic density of states at the location of the tip over the metal surface. These
near-field microscopy methods can be used to study the topography of a surface at
subnanometric scales, and hence to image atomic or molecular arrangements. One
of the most powerful methods for studying the band structure of a solid is angle-
resolved photoemission spectroscopy, which can directly determine the dispersion
curves E(k), and in particular the Fermi surface in the case of a metal.
Question 3.1
Let a1 and a2 be the vectors specifying the real lattice, and let us define their angle
as (a1 , a2 ) = α. The 2D reciprocal lattice is given by
2π a2 ∧ k 2π a1 ∧ k
a∗1 = , a∗2 = ,
S S
where S = |a1 ∧ a2 | = a1 a2 sin α is the area of the primitive unit cell, and k the
normal to the (a1 , a2 ) plane (see Fig. 3.27). We thus have (a∗2 , a∗1 ) = (a1 , a2 ) = α,
and the reciprocal lattice has the same symmetry as the real lattice.
The rectangular lattice (ax, by) has a rectangular reciprocal lattice with
2π 2π
a∗1 = x, a∗2 = y.
a b
The reciprocal
√ lattice of the hexagonal lattice is hexagonal with lattice constant
4π/a 3, while the reciprocal lattice of the centered rectangular lattice is centered
rectangular with lattice constant 2π/a sin α. The first Brillouin zones are the Wigner–
Seitz cells determined in Question 2.1.
Fig. 3.27 Reciprocal lattice a2

vectors for a planar lattice k
α
a1
α
a∗2
a∗1
Question 3.2
The fcc primitive cell is shown in Fig. 3.28a. The basis of the diamond lat-
tice contains two atoms of Ge at (0, 0, 0) and a(1/4, 1/4, 1/4), shown as white
spheres in Fig. 3.28a. The reciprocal lattice is body-centered cubic with side 4π/a.
The first Brillouin zone is shown in Fig. 3.28b. There are two atoms per unit
cell which donate each their four (4s2 4p2 ) electrons to the bands issued from
the atomic (4s2 4p6 ) levels. These eight electrons therefore completely fill the
first 4 bands (see Fig. 3.29). The others are empty at T = 0 and Ge is thus an
insulator.
Question 3.3
The band gap separates the highest energy occupied levels at k = 0 and the lowest
energy empty level at the point L with coordinates (2π/a)(1/2, 1/2, 1/2). The energy
difference between these levels is Eg = 0.7 eV. The energy width of this band gap
kz
z
L
Γ
kx X ky
K
a 4π / a
x
(a) (b)
Fig. 3.28 Diamond structure. (a) Real space fcc unit cell with its two-atom basis. (b) Bcc recipro-
cal lattice and Brillouin zone
Fig. 3.29 Band structure of

Ge. Filled valence bands are 4
shaded and the band gap is
hatched. The k space loca- Ge
Electrons 2
tions of the hole and electron
states that can be populated by Ec
thermal excitation or doping Ev 0 Band gap
are indicated Holes
Energy (eV)
−2
−4
−6
−8
−10
−12
L Γ X
2π 1 1 1 2π
a 2, 2, 2 (0, 0, 0) a (1, 0, 0)
is therefore very small. At room temperature, the electronic states close to Ec with
ke (2π/a)(1/2, 1/2, 1/2) will be partly occupied and the states of energy Ev close
to kh 0 will be partly unoccupied.
Question 3.4
Write
! ! ! !

ϕlA H λA ϕjA + λB ϕjB eik·Rj = E ϕlA λA ϕjA + λB ϕjB eik·Rj .
j j
Given the orthonormality of the functions |ϕjA and |ϕjB , the right-hand term sim-
plifies immediately to
! !

λA ϕlA H ϕjA eik·Rj + λB ϕlA H ϕjB eik·Rj = λA Eeik·Rl .
j j
The first sum reduces to the term j = l, while the second includes only hopping
integrals from the site (l, A) to its three nearest neighbour B atoms, i.e., (l, B) and the
two B atoms translated by −a1 and −a2 . Cancelling the common factor eik·Rl then
yields

λA E0 − λB t 1 + e−ik·a1 + e−ik·a2 = λA E .

In the same way, projecting onto ϕlB , we obtain

λB E0 − λA t 1 + eik·a1 + eik·a2 = λB E .
These two equations have solution

(E0 − E)2 = t2 1 + e−ik·a1 + e−ik·a2 1 + eik·a1 + eik·a2 ,
which leads to the dispersion relations of the energy bands (3.60):

1/2
Ek± = E0 ± t 3 + 2 cos k · a1 + cos k · a2 + cos k·(a1 − a2 ) .
Question 3.5
1. The 2D Bravais lattice is square with side a, and the primitive cell contains a Cu
and 2O. The reciprocal lattice is plane square with lattice constant 2π/a. The first
Brillouin zone is square, centered on the origin and with vertices (±π/a, ±π/a).
The axes are easily identified using the symmetry of the experimental Fermi
surface (see Fig. 3.30).
ky
π /a
ne /4 π /a
γ
kx
nh
Fig. 3.30 Fermi ‘curve’ for the cuprate lattice. The square first Brillouin zone is displayed (see
text). Shaded areas allow determination of the carrier content
2. The Fermi ‘curve’ (C) has large discontinuities at its intersections with the first
Brillouin zone. The situation cannot therefore correspond here to one in which
the electrons are nearly free. It is more like one with tight-binding, because the
ionic potential gives large contributions at the edge of the zone. However, the
simple LCAO model with hopping integrals between nearest neighbour Cu can-
not explain this Fermi surface (see those plotted in Fig. 3.7). We must also intro-
duce hopping integrals between next nearest neighbours.
3. The surface bounded in the plane by the Fermi curve (C) contains the occupied
electronic states k. As the first Brillouin zone can accommodate 2 electrons per
primitive cell, including spin, the number of electrons per primitive cell in the
Brillouin zone is
Area within (C)
ne = 2 × .
Area of the Brillouin zone
This ratio can be obtained from Fig. 3.30 by comparing the green area upper
right with the area of one quarter of the Brillouin zone. By comparing the areas
on either side of the diagonal of the square, it is clear that ne /4 < 2(1/8), that is
ne < 1. Greater accuracy is obtained by determining the area of the Fermi surface
of holes, which is the green region bottom left. We find nh = 1.15±0.05 = 2−ne .
There are therefore 0.85 electrons per primitive cell in the CuO2 plane in the
partially occupied band. It is in fact this reduction of ne below unity that leads to
a metallic state. Indeed, for ne = 1, the CuO2 plane is an antiferromagnetic Mott
insulator owing to the strong electron correlations at the Cu site.7
4. This means that there are fewer electrons per primitive cell in the CuO2 plane
than would be expected from the chemical formula Bi2 Sr2 CaCu2 O8 . Electrical
neutrality thus implies an excess of negative charges outside the CuO2 plane.
This comes from the fact that, in the materials studied, the stoichiometry is
not perfect. For example, there are interstitial oxygen atoms with concentration
x = (1 − ne )/2, and this modifies the doping in the CuO2 planes. The chemical
formula should be Bi2 Sr2 CaCu2 O8+x . The possibility of changing the oxygen
content x provides a controlled way of modifying the number of carriers in the
CuO2 planes.
Question 3.6
The figure on the cover of the book shows at the top the intensity of emitted photo-
electrons as a function of their wave vectors kx for different values of ky . Peaks in
the emitted intensity correspond to the last occupied levels and provide a direct way
of visualising the Fermi surface of this compound in the first Brillouin zone via a
projection onto the plane (kx , ky ). This first Brillouin zone is a square in this plane,
as for the cuprates. The Fermi surface is much more complicated than that of the
cuprate in Fig. 3.25. Note that it is made up of three curves which can be generated
7 This situation is described for La2 CuO4 in Problem 17: Electronic properties of La2 CuO4 .
from any one of them by the four-fold symmetry of the square. This indicates that
the band structure of this compound contains three bands which intersect the Fermi
level and are therefore partly occupied.
For completeness, note that the crystal structure of Sr2 RuO4 is obtained from
a perovskite structure (see Chap. 2) and it is the same as that of La2 CuO4 .8 If we
consider the charges of the Sr2+ and O2− ions, it is easy to deduce that the Ru are in
a 4d4 configuration. However, in cuprates such as La2 CuO4 , where La is ionised to
La3+ , the Cu are in a 3d9 configuration. In the ruthenate, it is not therefore surprising
that several bands arising from the various d orbitals with similar energies should
overlap near the Fermi level, in contrast to what happens with the cuprates.
8 This structure is displayed in Problem 17: Electronic Properties of La2 CuO4 .

Chapter 4
Electron Transport in Solids
Contents
4.1 Drude Model for Transport in an Electron Gas: Relaxation Time and Collisions . . . . . . . 101
4.1.1 Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.1.2 Thermal Conductivity and the Wiedemann–Franz Law . . . . . . . . . . . . . . . . . . . . . . 105
4.2 Electron Transport in a Fermion Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2.1 Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2.2 Thermal Conductivity and the Wiedemann–Franz Law . . . . . . . . . . . . . . . . . . . . . . 110
4.3 Electrons in a Lattice: Dynamics of Bloch Electrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.1 Group Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.2 Acceleration in Reciprocal Space and Real Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3.3 Electronic Conductivity in a Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.4 Origin of Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.1 Experimental Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.2 Scattering by Lattice Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.4.3 Collisions with Impurities and Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.5 Electrons, Holes, and Dopants in Semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.6 Electrons, Holes, and Transport in Graphene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

100 4 Electron Transport in Solids
All but one of the everyday objects illustrated here exploit physical processes with a common
feature. As you read this chapter, you should be able to spot the odd one out, and explain why it
differs, indicating for each object the key physical process underlying its use (answer in Sect. 4.8)
4.1 Drude Model for Transport in an Electron Gas: Relaxation Time and Collisions 101
Our forebears were interested in metals for their malleability (at the forge) and good
thermal conductivity (for making cooking vessels), and they have been used for
centuries as a raw material for forging tools and weapons of various kinds. But
their most characteristic feature, namely their good electrical conductivity (about 20
orders of magnitude greater than insulators), has only been recognised for a few cen-
turies, since Coulomb and Ampère discovered the existence of electrical charge. In
the preceding chapters, our discussion of allowed or forbidden energy bands for the
electron levels in crystalline solids explains the origin of this qualitative difference
between metals and insulators. However, we have not yet established a physical
understanding for the limitation of conductivity in metals as described by Ohm’s
law. Indeed, we have seen that an electron in a metal is in a quantum state charac-
terised by a Bloch wave function, which is a wave function extending throughout
the space occupied by the metal. Since this state is stationary, one would expect an
electron to be able to move freely throughout the volume of the metal without dis-
sipation of energy, just as any electron in an atom remains around the nucleus in a
stationary state, without loss of energy. The idea of electrical resistivity is therefore
quite foreign to the quantum description we have made of the electrons in a perfect
metallic crystal.
Note also that the descriptions used so far in the quantum mechanics or statisti-
cal physics textbooks concern stationary states or thermodynamic equilibrium, not
directly relevant to charge transport in a material. The phenomenon of electrical con-
ductivity is a problem of non-equilibrium thermodynamics. However, the departure
from equilibrium here is not dramatic, provided that we consider only weak elec-
tric fields and currents. This phenomenon is thus described as a quasi-equilibrium
transport regime.
In Sect. 4.1, we begin by describing the approach used by Drude in the 1900s to
explain conductivity by treating the electrons as classical particles. We thus intro-
duce the concepts of relaxation time and collisions. These concepts are developed in
the more realistic context of a free fermion gas (see Sect. 4.2), then in the presence
of a crystal potential (see Sect. 4.3). Section 4.4 discusses experimental observations
that help us to determine the origin of collisions in real metals. Then in Sect. 4.5, we
introduce the notion of hole which plays an important role in the case of semicon-
ductors. Finally, the peculiar case of graphene, which has recently triggered intense
research activity, is considered in Sect. 4.6.
4.1 Drude Model for Transport in an Electron Gas:

Relaxation Time and Collisions
At thermodynamic equilibrium, there is no transfer of matter, heat, or charge within

a continuous medium. However, in everyday life, such transfers are commonplace.
While the underlying microscopic processes are generally very difficult to describe,
there is nevertheless a simple situation where they are controlled by a slight depar-
ture from thermodynamic equilibrium. Indeed, if a macroscopic gradient is imposed
on one of the intensive parameters characterising thermodynamic equilibrium, this

will lead to a flow of the extensive quantity conjugate to it. For example, a tem-
perature gradient ∇T will cause a heat flow (and hence a flow of energy jen ), an
electrical potential gradient (an electric field E ) will cause a flow of charge (and
hence a current j), and a chemical potential gradient ∇μ will cause a flow of matter.
In the limit of a very weak gradient, the resulting flow will be proportional to the
gradient producing it. In the first two examples above, this corresponds to Fourier’s
law (jen = −κ∇T) and Ohm’s law (j = σ E ), which define the thermal conductivity
κ and the electrical conductivity σ . In the rest of this chapter, we shall examine the
microscopic origins of these transport processes in metals.
4.1.1 Electrical Conductivity
Following Drude’s example at the beginning of the 20th century, we do not have
to take into account quantum phenomena to obtain a qualitative understanding of
the physical origins of electrical resistivity. Indeed, Drude treated the electron as a
classical particle. When there is no electric field, the electrons have velocities vi ,
with randomly distributed directions in space, like the molecules in a gas, so their
average velocity is zero, i.e., vi = 0. Under the action of an electric field E applied
at time t = 0, we know that an electron of velocity vi , mass m0 , and charge −e will
be accelerated, with
dvi
m0 = −eE . (4.1)
dt
All the electrons in the gas will therefore acquire a velocity component vd called
the drift velocity, corresponding to a uniformly accelerated motion in the direction
opposite to the applied field:
et
vi = vi (0) + vd = vi (0) − E. (4.2)
m0
This would imply that vd tend to infinity over a long period of time. However, we
know by Ohm’s law that a static field produces a constant current. The current
density j is the electron flow per unit time and per unit area. It is given by

j = −e vi = −ne vi . (4.3)
i
Here the sum over i concerns all electrons crossing the unit area per unit time, n is
the number of electrons per unit volume, and vi denotes their average velocity. The
only way to obtain a stationary solution with constant flow rate is to introduce a fric-
tional effect for the electron motion, whose role is to limit their drift velocity. To ob-
tain a stationary solution with constant average velocity vi , the friction term must
be proportional to vi , i.e., −(m0 /τe ) vi , where τe has the physical dimensions of
time:
d vi vi
m0 = −eE − m0 . (4.4)
dt τe
This leads to the stationary solution vi = −(eτe /m0 )E = −μe E, where μe is the
electron mobility, and finally,
ne2 τe
j= E. (4.5)
m0
We thus recover Ohm’s law:
j = σE . (4.6)
The electrical conductivity and resistivity are given by
ne2 τe 1
σ= = neμe = . (4.7)
m0 ρ
Note that the time τe introduced along with the friction hypothesis has a simple in-
terpretation. Suppose that the electric field is cut off at time t = 0. The dynamical
equation for the average velocity shows that this quantity decreases exponentially
to 0, with time constant τe . This is thus the lapse of time required for the velocity
distribution to return to equilibrium, and it is called the relaxation time. It therefore
quantifies the rate at which the electron gas can reach thermodynamic equilibrium
from a non-equilibrium state.
The relaxation time clearly originates from microscopic causes, which Drude
attributed to the fact that the electrons, like the molecules of a gas, undergo col-
lisions. The latter change the orientations of the electron velocities, but also their
magnitudes. Indeed, collisions that only change the direction of the electron veloc-
ity without changing its kinetic energy should not contribute to dissipation. Drude
thus assumed that the probability p(v) that an electron have velocity v after a colli-
sion would be independent of its velocity before the collision, and hence that these
collisions would destroy all memory of the increased drift velocity of the electrons.
All the electrons will therefore undergo collisions that completely change their
velocities. This is illustrated in Fig. 4.1, which shows the contribution of one elec-
tron to the drift velocity as time goes by.
vd
t
Δt1 Δt2 Δt3
Fig. 4.1 Time dependence of the component of the velocity of a free electron induced by the
presence of an electric field, according to the Drude model. The induced component is projected
in the field direction (the acquired velocity is antiparallel to the field). The drift velocity of the
electron is vd
Let dt/τ be the probability that an electron will undergo a collision during the
time interval dt, and v(t) the average value of the velocity at time t. Then at time
t + dt, the electrons fall into two groups:
1. The fraction dt/τ of the electrons which underwent a collision between t and
t + dt. These make no contribution (to order dt) to the average velocity at time
t + dt.
2. The fraction (1 − dt/τ ) of the electrons which did not undergo a collision
between t and t + dt. The velocities of these electrons increase to vi (t + dt) =
vi (t) − eEdt/m0 . Their contribution to the average velocity is therefore

dt eEdt
vi (t + dt) = 1 − vi (t) − , (4.8)
τ m0
or to order dt,
eEdt dt
vi (t + dt) − vi (t) = − − vi (t) , (4.9)
m0 τ
which leads back to (4.4).
We may thus identify τ with τe . In the box below it is shown that, in this model, τe
is also the average time elapsed between two collisions of a given electron.
Let us find the value of τe for metallic copper, for which σ = 6.54×107 Ω−1 m−1
(or ρ = 1.55 μΩ cm). Given that copper crystallises into a face-centered cubic lat-
tice with conventional unit cell of side a = 3.61 Å, and that the conduction band
contains one electron per copper atom, it follows that n = 4a−3 = 8.5 × 1028 m−3 .
Equation (4.7) yields τe = 3 × 10−14 s, which indicates that an electron undergoes
an enormous number of collisions per second.
Collision Probabilities
Let p(t) be the probability that an electron undergoes no collision between times t = 0 and t.
The probability of continuing until t + dt without collision, viz., p(t + dt), is the probability
that the electron has not had a collision until t, and in addition that it has not had a collision
between t and t + dt, whence
p(t + dt) = p(t)(1 − dt/τe ) . (4.10)
It follows that
dp = −pdt/τe , (4.11)
p(t) = exp(−t/τe ) . (4.12)

The probability P(t)dt that an electron undergoes no collision between 0 and t but undergoes
a collision between t and t + dt is then
P(t)dt = exp(−t/τe )dt/τe . (4.13)

The probability density P(t), described by an exponential distribution, can be used to calculate
the average time between two collisions:
∞
t = tP(t)dt = τe . (4.14)
0
Question 4.1. The Drude model can be extended to describe the response of a metal to an
alternating electric field E(t) = E(ω) Re {eiωt }. The associated current density can be used
to define a complex conductivity
σ (ω) = σ1 (ω) − iσ2 (ω) .
Making Drude’s hypothesis that there is a relaxation time τe , determine σ (ω). Express σ1 (ω)
as a function of σ (0), and represent it graphically as a function of ω.
4.1.2 Thermal Conductivity and the Wiedemann–Franz Law
While an electric field E, i.e., a potential gradient, produces a charge flow and hence
an electron current j, a temperature gradient ∇T induces an energy flow which can
be characterised by an energy current density
jen = −κ∇T . (4.15)
The constant of proportionality κ is called the thermal conductivity.
a. Kinetic Model
We consider the energy transfer associated with the electronic system.1 We assume
that the same collisions underlie energy transfers in the electronic system and limit
the drift velocity of the electrons when an external electric field is applied. If an
electron undergoes a collision at a point, it either acquires or gives up an amount of
energy that is determined by the temperature at the given point. It is this temperature
that defines the average velocity acquired by the electrons that undergo a collision
at this point. Take the temperature gradient in the metal to be along the x axis. As
above, we assume that the electrons thermalise at each collision. Let n(v) be the
number of electrons per unit volume with velocity v. Considering an area dS of
the metal perpendicular to the x axis, the electrons with velocity v crossing this area
between times t and t + dt lie in a volume vx dt dS, so that their number is n(v)vx dt dS
1 In solids, the vibrational modes of the atoms (phonons) also contribute to thermal conductivity.
However, at low temperatures, very few phonons are excited and, in metals, the thermal conduc-
tivity is dominated by electrons.
Fig. 4.2 Volume containing

all electrons of velocity v
crossing the surface element jen
dS during time dt. If dt = τe , v dt
then because of the tempera- dS
ture gradient, these electrons v x
undergo a change in energy as
given by (4.16)
T (x)
(see Fig. 4.2). If we take the time dt to be the time τe between two collisions, each
of these electrons of velocity v undergoes an energy change of
du du dT dT
u = vx τe = vx τe = vx τe c , (4.16)
dx dT dx dx
where u(x) is the electron energy at x and c is the specific heat per electron.
The energy current along the x axis, which is the rate of flow of energy per unit
area of the metal and per unit time τe , is thus obtained by integrating over v, which
leads to

dT 1 dT
jen = − d3 vn(v)vx vx τe c x = − d3 vn(v)v2 τe c x ,
dx 3 dx
or
1 dT
jen = − nv2 τe c x , (4.17)
3 dx
where v = v2 1/2 is the root mean squared value of the velocity. This can be used
to specify the average distance travelled between two collisions, known as the mean
free path of the electron, viz., e = vτe . We thus arrive at an expression for the
thermal conductivity:
1
κ = Ce ve , (4.18)
3
where Ce = nc is the electron specific heat per unit volume.
b. Wiedemann–Franz Law and Maxwell–Boltzmann Statistics
With his classical approach, Drude used Maxwell–Boltzmann statistics to interpret

the empirical law due to Wiedemann and Franz, according to which κ/σ is propor-
tional to T and independent of the metal (see the results in Table 4.1).
The energy equipartition theorem can be used to estimate the mean squared
velocity at temperature T from v2 = 3kB T/m0 , where kB is Boltzmann’s constant.
4.2 Electron Transport in a Fermion Gas 107
Table 4.1 Electrical and thermal conductivities at 0◦ C, and Lorenz constant L for various
common metals. From [8]
−1
Metal σ [107 Ω m−1 ] κ [102 watt m−1 K−1 ] L [10−8 watt Ω K−2 ]
Na 2.34 1.35 2.10

Cu 6.45 3.85 2.18
Ag 6.6 4.18 2.31
Be 3.6 2.3 2.34
Mg 2.54 1.5 2.16
Al 4.0 2.38 2.18
Pb 0.52 0.35 2.46
Bi 0.093 0.085 3.30
Pt 1.02 0.69 2.47
In addition, for a classical perfect gas of pointlike particles, the heat capacity per
particle is c = (3/2)kB , and C = (3/2)nkB . As a consequence, κ is given by
3 kB T 3 n
κ = nkB τe = τe k2 T . (4.19)
2 m0 2 m0 B
If we assume that the same particles, subject to the same collisions, determine σ and
κ, (4.7) and (4.19) can be used to eliminate n and τe . This leads to a simple relation
between κ and σ :

κ 3 kB 2
= T =LT . (4.20)
σ 2 e
This linear dependence of the ratio κ/σ on T is independent of the metal. The con-
stant of proportionality is called the Lorenz number:
L = (3/2)(kB /e)2 = 1.12 × 10−8 W Ω/K2 . (4.21)
This law due to Wiedemann and Franz is confirmed experimentally, with a larger
value of the constant L than the theoretical value but of the same order of magnitude
(see Table 4.1). This result was long regarded as a proof that Drude’s hypothesis as
formulated in the context of classical mechanics was actually correct.
4.2 Electron Transport in a Fermion Gas
At the time, Drude could legitimately consider the electron gas as a gas of classical
particles, but with the advent of quantum mechanics, this approach clearly become
obsolete insofar as the average velocity of the electrons cannot be given by the
kinetic theory of gases. We know that the Pauli exclusion principle and Fermi–Dirac
statistics lead to an electron kinetic energy which is determined by the Fermi energy
rather than by kB T. Moreover, we have seen that the classical approach suggests that
all the electrons contribute to the electrical conductivity, whereas we know that only
those electrons close to the Fermi level are likely to take part in energy exchanges
since they are close in energy to unoccupied states.
4.2.1 Electrical Conductivity
We treat the free electron gas as an ensemble of fermions whose eigenstates are
plane waves. All the occupied states in wave vector space, which correspond to n
electrons per unit volume, lie within the Fermi sphere of radius kF such that
4π 3 2
k =n, (4.22)
3 F (2π )3
which leads to the expression (3.47) for the Fermi energy:
EF = (h̄2 /2m0 )(3π 2 n)2/3 .
Plane waves are electron states with infinite spatial extent and precisely defined
momentum. To describe electron transport, we must adopt a spatially confined rep-
resentation of the electron, which can be done by considering the plane wave packet
with wave vectors around some k value. Such a wave packet is spatially localised,
at the expense of some uncertainty in the value of the wave vector. It has a group
velocity vg = p /m0 such that
1 h̄k
vg = ∇ k E(k) = . (4.23)
h̄ m0
When there is no applied force, for each occupied state with wave vector k there is
an occupied state with wave vector −k, and the average velocity of the electron gas
is strictly equal to zero, even if the electrons are moving at high speed.
In a quantum picture, the dynamics of an electron subjected to an applied force
that varies slowly in time and space compared with the spatial characteristics of the
wave packet can be described by a semi-classical equation which specifies the time
evolution of the wave packet [2]:
dvg h̄dk
m0 = =F . (4.24)
dt dt
For an applied electric field, (4.24) becomes
d p h̄dk
= = −eE . (4.25)
dt dt
4.2 Electron Transport in a Fermion Gas 109
The effect of the electric field is thus to increase the momentum of the electron
linearly with time, as happened within the classical description. This therefore shifts
the points representing the quantum states in k space in the opposite direction to
the electric field vector. And this shift corresponds to the appearance of a flow of
electrons, i.e., a current, in real space. Indeed, the states k on the shifted Fermi
sphere are no longer paired as described above. Note that, as for a classical gas, the
indefinitely accelerated electron motion once again contradicts Ohm’s law. Here,
too, we must introduce collision processes if we are to reach a stable situation in the
presence of an electric field, where the Fermi sphere is shifted by
h̄δk = −eEτe . (4.26)
Note that the number of occupied states within the Fermi sphere corresponds ex-
actly to the n electrons per unit volume of the metal. With the displacement of the
Fermi surface shown in Fig. 4.3, each is displaced by the above amount δk, which
corresponds to a drift velocity v = δvg = −eEτe /m0 , and hence a total current
given by
ne2 τe
j = −ne vi = E = σE . (4.27)
m0
We thus recover a transport equation that is practically the same as the one derived
by Drude. However, the quantity τe in this expression has a completely different
meaning. Indeed, for a classical gas, the electrons can all undergo collisions what-
ever their kinetic energy, and τe is a collision time which affects all the electrons
equally. Here, even though the effect of the field can be represented by a uniform
shift of all the states, the equilibrium can only be maintained by collisions which
ky
E
B kx
δk
Fig. 4.3 Fermi surface at equilibrium and in the presence of an applied electric field. In the steady-
state regime, the sphere is shifted by h̄δk = −eEτe from the equilibrium position. This stationary
state is obtained by collision processes which cause the electrons corresponding to the states A
(right-hand Fermi surface) to come back to B (left-hand Fermi surface)
redistribute the electrons between these states and prevent the distribution from con-
tinuing its progression towards the right of Fig. 4.3. According to the Pauli exclusion
principle, the collisions can only transfer electrons from occupied states to unoccu-
pied states. If kF is the radius of the Fermi surface when there is no field, these
collisions can therefore only bring states with |k| > kF on the right back to states
with |k| < kF on the left of the figure. Only collision processes involving electrons
with energy close to EF can therefore contribute to establishing the stationary state
and the corresponding redistribution of the velocities. Pauli’s principle thus restricts
the kind of collision that can occur, and the collision time τe only concerns electrons
at the Fermi level, with speed vF = h̄kF /m0 . Likewise, the mean free path concerns
only these electrons, which can undergo collisions. The mean free path is given by
e = vF τe (EF ) . (4.28)
Consider again the example of copper, where τe = 3 × 10−14 s. The drift velocity is
ve = −(eτe /m0 )E 5 × 10−3 E. Even for an electric field of 104 V/m, this implies
that ve ∼ 50 m/s. It can be compared with the Fermi velocity vF = (3π 2 n)1/3 h̄/m0 ,
according to (4.22), or vF 1.5 × 106 m/s, with the value of n determined in
Sect. 1.1.1. The drift velocity is thus very small compared with the Fermi veloc-
ity. This indicates that the displacement of the Fermi surface is at most 10−4 kF . The
representation in Fig. 4.3 is therefore not at all to scale. Finally, we may estimate
the mean free path e 5 × 10−8 m 500 Å, according to (4.28). In good metals,
at room temperature, an electron only moves a few hundred times the interatomic
separation before suffering a collision.
4.2.2 Thermal Conductivity and the Wiedemann–Franz Law
The kinetic arguments in Sect. 4.1.2a for expressing κ as a function of microscopic

quantities are still valid, provided we use the parameters associated with the Fermi
energy and the value of Ce obtained for a free electron gas (see Chap. 3):
π2 kB T
Ce = nkB . (4.29)
2 EF
Using (4.18), this implies that
1 π 2 kB2 T
κ = Ce v2F τe = n τe . (4.30)
3 3 m0
In the context of Fermi–Dirac statistics, the ratio κ/σ is thus equal to
2
κ π2 kB
= T =LT , (4.31)
σ 3 e
4.3 Electrons in a Lattice: Dynamics of Bloch Electrons 111
where the Lorenz number is exactly twice that obtained using classical statis-
tics. This theoretical value L = 2.45 × 10−8 watt Ω/K2 is much closer to the
experimental values of L given in Table 4.1. It is remarkable that the classical
and quantum calculations give such close results. It is an amazing coincidence if we
recall that the expressions for κ and σ are very different. Indeed, Cv is proportional
to kB T for a fermion gas, whereas it is independent of T for a classical gas. Con-
versely, v2 is equal to v2F and is independent of temperature for fermions, whereas
v2 is proportional to T for a classical gas. These two terms compensate one another
quite accidentally, up to a factor of 2 in the expression for the Lorenz number!
4.3 Electrons in a Lattice: Dynamics of Bloch Electrons
Is this description of electron transport still valid in a crystal, in the presence of a

periodic crystal potential? As in the case of free electrons, an electron in a crystal
will have to be described by a wave packet, but this time by a packet of Bloch waves
centered on a wave vector k, involving states k + k belonging to the same electron
energy band with index n. The wave packet thereby constructed will have spatial
extent r related to its extent k in reciprocal space by r ∼ 1/k.
4.3.1 Group Velocity
The wave vector k which specifies the Bloch state of energy En (k) is not simply
related to the momentum operator p = −ih̄∇, in contrast to the case of plane waves.
Indeed, for the Bloch function
ψnk = unk (r)eik·r , (4.32)
the momentum is given by
pnk = ψnk |p|ψnk = m0 vnk (4.33)

= h̄k − ih̄ unk (r)∇unk (r)d3 r = h̄k . (4.34)
However, the crystal momentum h̄k plays a very special role, as we shall see. A
simple calculation carried out in the box in next page relates the average velocity
vnk of the electron to the k dependence of the dispersion relations En (k). In fact,
we find that
1
vnk = ∇ k En (k) , (4.35)
h̄
which naturally also applies to free electrons [see (4.23)]. In the 1D case, this
expression reduces to v = h̄−1 dEn (k)/dk, and the velocity of an electron in a state
kz
v
E
M 0
M v kx ky
k
− π /a 0 π /a
(a) (b)
Fig. 4.4 Velocity of an electron in a Bloch state k, with energy E(k), represented by the point M.
(a) For the 1D band calculated using the LCAO approximation. (b) For a cubic crystal. A constant
energy surface is shown
(n, k) is nothing other than the derivative of the dispersion curve with respect to
k. This is illustrated in Fig. 4.4, for the simple band calculated using the LCAO
approximation. The velocity is zero at the bottom of the band and at the point
k = π/a of the first Brillouin zone.
In two or three dimensions, the geometric interpretation of (4.35) is a little
more complicated. It shows that the velocity vnk lies along the normal at the
point OM = k to the constant energy curve or surface constructed from En (k) (see
Fig. 4.4b).
Note that, in a Bloch state which is a stationary eigenstate of the Hamilto-
nian, the electronic velocity is well defined and time independent, agreeing with
the idea that the electron undergoes no collision on the crystal potential, the latter
being entirely accounted for by the Hamiltonian. The periodic crystal potential does
not scatter the electrons, but simply determines their velocity as given by (4.35).
This quantum description invalidates Drude’s hypotheses, according to which the
collision time τe is determined by collisions of the electrons on the atomic nuclei.
Relation Between vnk and En (k)
If En (k) is the energy eigenvalue associated with the Bloch state (4.32), we may write

p2
+ V(r) unk eik·r = En (k)unk eik·r , (4.36)
2m
which reduces to
h̄2
(k − i∇)2 + V(r) unk = En (k)unk . (4.37)
2m
This corresponds to an effective Hamiltonian given by
h̄2
Hk = (k − i∇)2 + V(r) , (4.38)
2m
which acts on the space of unk (r) functions and has eigenvalue En (k). The effective Hamilto-
nian corresponding to k + q can be written
h̄2
Hk+q = (k + q − i∇)2 + V(r) (4.39)
2m0
h̄2 h̄2 2
= Hk + (k − i∇)q + q . (4.40)
m0 2m0
For small q, a perturbation treatment to first order in q determines the energy of the corre-
sponding eigenstate to be
" #
h̄2 h̄2 2

En (k + q) = En (k) + unk (k − i∇)·q + q unk + · · · (4.41)
m0 2m0
h̄
= En (k) + unk eik·r |p|unk eik·r q + O(q2 ) + · · · (4.42)
m0
= En (k) + h̄ vnk ·q + · · · . (4.43)
Identifying with the Taylor expansion
En (k + q) = En (k) + q·∇ k En (k) + · · · , (4.44)
we do then recover (4.35).
4.3.2 Acceleration in Reciprocal Space and Real Space
Under the effect of an applied force F, the electron energy is modified. Assume that
F varies slowly in space on the scale of the unit cell, and slowly in time on the
scale of the frequencies associated with transitions between allowed energy bands.
The work dW done on an electron of velocity v and charge −e over a time inter-
val dt modifies its energy En (k) by changing the value of k and hence the crystal
momentum, whence
dW = F · vdt = dEn (k) (4.45)

1
= F· ∇ k En (k)dt = ∇ k En (k)·dk , (4.46)
h̄
or
dk
h̄ =F . (4.47)
dt
This is the acceleration theorem in reciprocal space. In response to an external force
that varies slowly in space and time, it is the derivative of the crystal momentum
(rather than the derivative of the momentum) that equals the applied force. In the
presence of a magnetic field B, it can be shown that (4.47) can be rewritten as
h̄dk
= −e(E + v ∧ B) = F . (4.48)
dt
Differentiating v, given by (4.35), with respect to time and using (4.47), we obtain
dv dk 1
= ∇ k v· = 2 ∇ k ∇k En (k) ·F , (4.49)
dt dt h̄
or
dvα 1
= Fβ , (4.50)
dt (me )αβ
β
with
1 1 ∂ 2 En (k)
= 2 . (4.51)
(me )αβ h̄ ∂kα ∂kβ
Equations (4.50) and (4.51) constitute the acceleration theorem in real space. Being
consequences of (4.47), they are subject to the same restrictions regarding the slow
spatiotemporal variation of F. Note that the tensorial relation between the accelera-
tion and the force lead to the definition of a tensor 1/(me )αβ whose components have
the dimensions of a reciprocal mass. This defines the effective mass tensor (me )αβ
at a point k of a given band n. By definition, the effective mass tensor is symmetric.
The idea of effective mass is of particular interest near an extremum k0 of an
energy band. In this case, close to k0 , if we set k = k0 + k, the energy is given to
lowest order by
h̄2
En (k) − En (k0 ) kα kβ . (4.52)
(me )αβ
αβ
In the particular case of a cubic crystal, at the center of the zone, if the energy is not
degenerate, the constant energy surfaces are spheres and the effective mass is thus
isotropic, given by me . The acceleration is then
dv
me =F, (4.53)
dt
and the derivative m0 dv/dt of the ordinary momentum is not equal to the external
force. In this same special case, the crystal momentum and velocity are related by
(4.35):
h̄k
v= . (4.54)
me
In general, the effective mass is positive near a band minimum and negative near a
maximum. According to (4.50), a negative effective mass expresses the fact that the
velocity acquired under the effect of Fβ is in the opposite direction to the velocity
that would be acquired by an electron subjected to Fβ in vacuum. This somewhat
paradoxical behaviour should come as no surprise, however. In the crystal, the elec-
tron feels both the applied electric potential and the crystal potential V(r). This
shows that, in a crystal, the response of an electron is determined to a large extent
by the potential associated with the ions of the crystal. Even when the effective mass
has the same sign as the mass m0 of the free electron, the value of me /m0 may differ
significantly from unity. While in metals me /m0 ∼ 1, the same is not always true
in semiconductors. For example, the effective mass at the bottom of the conduction
band is +0.067m0 for GaAs and +0.014m0 for InSb.
These effective masses can be determined by studying the response of the elec-
trons to an alternating electric field in the presence of a static magnetic field. The
eigenmodes of the electron states give rise to so-called cyclotron resonances.2
4.3.3 Electronic Conductivity in a Crystal
The results obtained for the Bloch states can help us to determine the conductivity
of a band. When there is no electric field, the total current density for a band is by
definition
e
j = −e v(k) = − ∇ k En (k) . (4.55)
h̄
k occupied k occupied
Since the functions En (k) are even in k, the total current vanishes. This is true for
the case of a fully occupied band, for example. Let us now apply an electric field in
the latter case. By the reciprocal space acceleration theorem, over a time interval dt,
each wave vector k will change by k = −eEt/h̄, and all wave vectors will thus
be shifted by the same amount.
Given the definition of the first Brillouin zone and the periodicity in k space, the
states leaving this zone are equivalent up to a vector K of the reciprocal lattice to
those that become empty, and the band remains fully occupied (see Fig. 4.5). Overall
E (k)
K = −2π/a
E
1 1
Δk
−π/a 0 π/a k
Fig. 4.5 Effect of an electric field E on the electron states of a filled band. The states in the ab-
sence of E (empty dots) are shifted by k = −eEt/h̄ (full dots). The states shifted out of the first
Brillouin zone are equivalent to those that are depopulated (1 is equivalent to 1). The band there-
fore remains entirely full and the total current is identically zero. The magnitude of k and the
quantisation of the levels with respect to k have been exaggerated in the figure for better visibility
2 See Problem 8: Cyclotron Resonance.

the electronic states of the band are conserved and j remains identically zero:
e
j=− ∇ k En (k) ≡ 0 , (4.56)
h̄
k full band
even when an external electric field is applied. We have thus shown what was
anticipated in Chap. 3, namely that a fully occupied band does not contribute to
the current. Metals thus correspond to the case in which bands are only partially
filled.
The conductivity of metals can then be understood in the same framework as free
electrons. The Fermi surface, which is no longer spherical in an arbitrary crystal, is
shifted when an electric field is applied, and the magnitude of the displacement is
determined by the number of collisions involving electrons at the Fermi level.
4.4 Origin of Collisions
So far we have seen how to understand why the conductivity of metals should be
finite. However, we still have no microscopic description of what causes collisions,
even though we have some intuitive notions. In particular, we have seen that a crys-
tal potential, resulting from the sum of the Coulomb potentials of the ions in the
crystal lattice will not contribute to any collision process if it is strictly periodic. On
the other hand, defects such as impurities will be able to scatter electrons and can
therefore be a source of collisions. As always, it is experiment that leads the way
to understanding and classifying the various collision processes. A key point here
is the temperature dependence of the resistivity of metals. We shall describe briefly
how the resistivity is measured and discuss the main results.
4.4.1 Experimental Observation
It is in principle an experimentally simple matter to measure the resistivity of a

metal. The idea is just to pass a constant current through the metal and measure
the voltage across the sample. However, when the conductivity is very high, some
precautions are necessary, otherwise the measurement wires and the resistance of
their contacts with the sample can distort the result. The method generally used is
shown in Fig. 4.6. The idea is to decouple the wires supplying the current from those
measuring the potential drop.
To investigate what causes conductivity, we may begin by considering a mono-
valent metal such as silver, for which the Fermi surface is almost spherical and
the conduction band is similar to a band of free electrons. The results obtained on
various silver samples with different levels of purity (see Fig. 4.7) show that the
resistivity is affected by the presence of impurities. They also show a marked tem-
perature dependence, with an almost linear variation at high temperatures and strong
4.4 Origin of Collisions 117
Silver lacquer
Current
Indium solder
Voltage Sample
Fig. 4.6 Measuring the resistivity of a metal. Four contacts are made on the sample, which is
placed on an insulating substrate. The two outer contacts are connected to the current supply. The
voltage measured across the two others is used to determine the resistivity ρ. Slightly resistive
contacts will not significantly affect the measurement
Fig. 4.7 Temperature depen-

dence of the resistivity of ‘Pure’ Ag
silver samples with varying
0.5 Ag+0.02% Sn
degrees of purity. Adapted
from result published in [5] Ag+0.05% Au
0.4
ρ (10−8 Ω m)
0.3
0.2
0.1
0
20 40 60 80 100
Temperature (K)
curvature at lower temperatures. These observations suggest that the resistivity takes
the form
ρ(T) = ρ0 + ρ1 (T) . (4.57)
The first term is largely determined by the purity of the metal, while the second
is practically independent of the sample. Compare these results with the expected
expression for the resistivity of a metal:
1 me 1
ρ= = 2 , (4.58)
σ ne τe
where we have introduced the effective mass me to account for the band structure.
The latter is in principle almost independent of temperature. The only modifications
arise from thermal expansion of the metal, which does not exceed about 1%. More-
over, the number of valence electrons introduced by impurities with concentrations
lower than 1% will barely alter the number of conduction electrons. It follows that
the only quantity that varies with the purity and the temperature T is the collision
rate 1/τe . The experimental results shown in Fig. 4.7 imply that this has approxi-
mately the form
1 1 1
= + . (4.59)
τe τ1 τ2 (T)
It thus looks as though two totally different collision processes are operating in
parallel. One of these, described by τ1 , is independent of temperature, but depends
on the purity of the material, while the other, described by τ2 , is exactly the opposite.
Experiment thus leads to a simple general law: if different independent microscopic
processes underlie the collisions of electrons at the Fermi level, the probabilities of
the collisions are additive. The additivity of the collision rates contributing to the
resistivity is called Matthiessen’s law. Below, we consider one by one the various
collision processes that can contribute to the resistivity.
4.4.2 Scattering by Lattice Vibrations
The temperature dependent term is insensitive to the presence of impurities and thus
seems to be a characteristic of the metal without defects. It dominates the resistiv-
ity at high temperatures when the metal is pure. It thus seems that the underlying
process is more effective at higher temperatures. We have seen that the Bloch states
which take into account the periodic potential of the crystal cannot be scattered by
the ions. However, any departure from periodicity, due to thermal agitation of the
ions from their equilibrium position, for example, is likely to cause electron scat-
tering. As the amplitude of these oscillations increases with temperature, such a
collision term would indeed be expected to increase at high T, as shown experi-
mentally. For this reason, the temperature dependent term is attributed to electron
scattering by atomic vibration modes.
These lattice vibrations are collective harmonic displacements of the ions from
their equilibrium positions. They propagate as waves and are called phonons. They
are characterised by a wave vector q, and by their energy h̄ω. The function ω(q) is
the phonon dispersion relation.3 An electron can thus be scattered from an initial
state k to a final state k , while the energy difference between the states is balanced
by absorption or emission of a phonon. This process can therefore modify the en-
ergy state of the electron system and lead to a state of thermodynamic equilibrium.
The associated dissipation corresponds to the Joule effect. The slowing down of the
electrons proceeds through excitation of vibrational modes which cause the metal to
heat up. In principle, the energy associated with these vibrations increases as kB T,
leading to a term linear in T for 1/τe and for the resistivity at high temperatures.
At low temperatures, only the low energy oscillatory modes of the atomic lat-
tice are excited. They correspond to low frequency, hence long wavelength modes,

4.4 Origin of Collisions 119
which are much less efficient. Given the vibrational eigenmodes of the atoms in
the crystal and their interaction with the electrons, we can calculate the resistivity
associated with the electron–phonon collision process. Phonon modes are charac-
terised by their maximal energy kB θD , and this defines a characteristic tempera-
ture θD called the Debye temperature. The resistivity due to the electron–phonon
interactions in different metals has a temperature dependence which is approxi-
mately a characteristic function of T/θD . This has been modelled by Grüneisen (see
Fig. 4.8).
Note that, since this temperature dependence varies little with the purity of the
material, measurements of the resistivity of a metal wire are often used as a sec-
ondary thermometer, after calibrating by one or two fixed temperature points. In
fact most precision electronic thermometers use platinum, a noble metal that barely
oxidises and is easily purified.
0.3
θD
Au 175 K
Na 202 K
0.2 Cu 333 K
Al 395 K
ρ Ni 472 K
ρ (θ D)
Fig. 4.8 Resistivities of dif- 0.1

ferent metals as a function
of T/θD , where θD is the
Debye temperature character-
istic of the phonons in the
given metal. Adapted from [7,
p. 221], 4th edn., Wiley c 0
0.1 0.2 0.3 0.4 T/θ D
1996
4.4.3 Collisions with Impurities and Defects
Electrons will obviously be scattered by crystal defects such as impurities. The

results in Fig. 4.9 do indeed show that the resistivity increases as a function of the
concentration of impurities, and that it depends on the kind of impurity. An electron
passing close to an impurity will be deflected by the interaction with the Coulomb
potential of the impurity. In contrast to scattering by phonons, the electron energy
is conserved along its trajectory. Once it has moved a long distance away, it will
recover the same kinetic energy. Such a collision is said to be elastic, and only
alters the direction of the electron velocity, not its magnitude. The trajectory and the
Fig. 4.9 Resistivity of copper

1.01 As
and various alloys of the form
Cu-M, where M is a chem- 8 1.13 Sb
ical element with concentra-
tion given in atomic percent-
age on the right of the figure. 0.87 Fe
Note that the residual re-
sistivity ρ0 associated with 3.32 Ni
6
scattering by impurities in-
0.61 Fe
creases with their concen-
m)
tration and depends on the
chemical nature of the ele- 2.16 Ni
ρ (10−8
0.89 Sn
ment inserted into the alloy.
4 0.40 Sb
Adapted from Linde, J.: Ann.
Phys. (Leipzig) 5, 15 (1932) 1.12 Ni
1.03 In
2 Pure Cu
0
80 160 240 320
T (K)
probability of scattering are independent of temperature, which explains why the

resistivity associated with this mechanism is independent of T. Note that impurities
are not the only defects occurring in a crystal that can play the role of scattering
center. The absence of an ion at a given crystal site, known as a vacancy, or an ion at
some interstitial position in the crystal lattice both constitute structural defects that
are extremely effective electron scatterers.
At very low temperatures, the resistivity of a metal is thus determined to a large
extent by its purity. However, even for the purest metals that can be obtained, the
resistivity is at best only 106 times lower than the resistivity at room temperature.
The electron mean free path can then reach the millimeter range for copper, or a few
centimeters for aluminium.
Question 4.2. 1. In Fig. 4.9, how does the contribution to ρ of the impurity M vary for
different values of the concentration c ?
2. Make a quantitative comparison of the effects of In, Sn, Sb, and As impurities on the
resistivity.
3. Atomic copper has the configuration 3d10 4s. In copper metal, only the 4s electron con-
tributes to the conduction band, which has a quasi-spherical Fermi surface. In, Sn, Sb,
and As give 3, 4, 5, and 5 electrons to the conduction band, respectively. The ionic cores
thus have valences 3+ , 4+ , 5+ , and 5+ , respectively. Express the result of part 2 as a
function of the valence of the impurity. Can you explain this experimental result quali-
tatively? What kind of calculation would be needed to give a quantitative explanation of
these results?
4.5 Electrons, Holes, and Dopants in Semiconductors 121
4.5 Electrons, Holes, and Dopants in Semiconductors
Consider a situation in which a band is almost completely occupied, except for a

few states at the top of the band. This may be the case, for example, in the va-
lence band of a semiconductor, out of which an electron has been excited, or in
the conduction band of a metal, when the Brillouin zone contains ‘pockets’ near
the Fermi level. Under such conditions, in order to investigate electron transport
problems, we shall see in Sect. 4.5a that, rather than considering all the electrons
in the band, it is simpler to restrict attention to the unoccupied states of the band,
called holes, which behave like positive charges. We shall then see in Sect. 4.5b
that, in semiconductors, the electron or hole nature and the concentration of the
charge carriers can be chosen by doping the semiconductor with the appropriate
impurity.
a. Holes
When an electric field is applied to the electrons in an electronic band that is almost
completely occupied, all the Bloch wave vectors representing the occupied states are
displaced in the opposite direction to the applied field. The same is therefore true
of the unoccupied states (see Fig. 4.10). The behaviour of the unoccupied states can
be used to describe the global response of the band. Consider for instance a valence
band containing just one empty state with quasi-momentum k = ke . It is tempting
to define a quasi-particle [valence band filled except for a single empty state], which
we shall call a hole, whose behaviour will describe the behaviour of the band. The
total current associated with the electron states in this band will define the current
jh associated with the hole:

jh = −e v(k) = −e v(k) + ev(ke ) = ev(ke ) , (4.60)
k occupied filled band
Ee (ke)
E
0
ke
Fig. 4.10 Hole in a filled electron band. The effect of an electric field E is to shift all the electron
states in the opposite direction (dark states shift into light states, while the empty state moves to the
empty dotted state). The current j associated with the band is negative according to (4.61), because
v(ke ) = h̄−1 dEe /dke < 0
with zero total current for the filled band (see Sect. 4.3.3). We thus have
e
jh = ev(ke ) = ∇ke Ee (ke ) , (4.61)
h̄
where Ee (ke ) is the dispersion relation of the valence electrons at the point ke .
Likewise, the total momentum of the filled band is zero because the electrons can
be associated in pairs with opposite momenta. The total momentum of the band in
the absence of the electron, that is, the wave vector of the hole, is therefore
kh = −ke . (4.62)
Finally, consider the total energy of the band. It is given by ET − Ee (ke ), where ET is
the total energy of the filled band. By choosing ET as the zero energy, we can define
the energy of the hole as
Eh = −Ee (ke ) . (4.63)
This energy increases as the energy of the missing electron gets smaller. Note in
particular that, under the action of an electric field, the change in energy of the band
is in the opposite direction to that of the missing electron. The hole thus has an
energy that varies in the opposite way to that of an electron under the effect of a
field. A positive electrical charge must therefore be associated with the hole:
eh = +e . (4.64)
With the above definitions, the current associated with the hole, as given by (4.61), is
e e
jh = ev(ke ) = ∇ ke Ee (ke ) = ∇ kh Eh (kh ) = evh (kh ) . (4.65)
h̄ h̄
The velocity of the hole is therefore simply the velocity of the missing electron at
ke . Finally, since kh = −ke , the dynamical equation describing the time dependence
of kh and vh in the presence of applied electric and magnetic fields is
dkh dke
h̄ = −h̄ = e(E + vh ∧ B) . (4.66)
dt dt
This is indeed the equation of motion of a positive charge of velocity vh in a state
with wave vector kh . The effective mass tensor for the hole can then be defined as
dvtα 1 1 1 ∂ 2 Eh
= Fβ , where = 2 . (4.67)
dt (mh )αβ
β
(mh )αβ h̄ ∂ktα ∂ktβ
It has the opposite sign to the effective mass tensor of the missing electron
(mh )αβ = −(me )αβ . (4.68)
Equations (4.62) to (4.68) thus provide the rules for moving from the electron
representation to the hole representation (see Fig. 4.11).
Eh(kh)
j
E
0 kh
Fig. 4.11 Energy, wave vector, and velocity associated with the band in Fig. 4.10 as represented
using the hole language. We do indeed have j < 0, because according to (4.65), jh = evh (kh ) =
(e/h̄)dEh /dkh < 0
The notion of hole is particularly important when treating semiconductors with

a fully occupied valence band at T = 0. Electron states in the conduction band and
holes in the valence band are populated when T = 0.
Question 4.3. In the case of Ge, which has the band structure shown in Fig. 3.12, what are
the vectors k and energies corresponding to excited electrons and holes at nonzero temper-
ature. Estimate their effective masses using Fig. 3.12.
A light wave of higher energy than the width of the band gap can also be used to
excite an electron of the valence band by an optical transition from a state ke to a
state of the same wave vector in the conduction band. An electron–hole pair can
thus be created even at low temperature.
Question 4.4. An electron–hole pair is created in the conduction and valence bands of a
semiconductor. What will be the total current associated with these carriers in the presence
of an applied electric field?
b. Doped Semiconductors
As we saw in Chap. 3, in a pure or intrinsic semiconductor, the concentration of

electrons and holes excited at room temperature is rather low. However, this situa-
tion can be altered by introducing impurities into the semiconductor in a controlled
way. For example, consider the case of silicon, a tetravalent element, into which
pentavalent phosphorus atoms are introduced. This amounts to breaking the trans-
lation symmetry of the crystal, by placing a P at an Si site. At this particular site,
the Coulomb potential will thus be higher, because there is an extra charge in the
atomic nucleus. For the first four valence electrons of the P, the situation is similar
to that of silicon. They are involved in sp3 bonds with the Si and fill the electron
states corresponding to those at the top of the valence band, thereby taking the place
of the missing silicon electrons.
However, the last electron donated by the P should be transferred to a state in the
conduction band of silicon, in an energy level Ec . But it is subjected to the Coulomb
potential due to the excess charge +e of the phosphorus nucleus, which will generate
bound electron levels, as in a hydrogen atom.
The main difference from the well known case of the hydrogen atom is that this
electron is not moving around in the vacuum, but rather in solid silicon. Its eigen-
states must be constructed from unoccupied electron states of the conduction band
of silicon, which are Bloch waves with a rather low effective mass me ( 0.3m0 for
silicon). In addition, this long range Coulomb potential produces an electric field
which acts on all the nuclei and electrons in the silicon. Hence, in this material,
which is basically an insulator, an applied electric field produces an electrostatic
dipole polarisation of the electronic orbitals around the Si nuclei. This polarisation
will in turn create a depolarising field, i.e., a field opposing the applied electric field.
This is the effect that leads to the dielectric constant ε of the insulator. It follows that
the Coulomb electric field produced by the extra charge of the P is reduced at large
distances by this factor ε. The net result of all this is that we have an excess electron
with effective mass me , subjected to a similar potential to that of the hydrogen atom
but reduced by the factor ε :
e2
Vc (|r − ri |) = − . (4.69)
4π εε0 |r − ri |
The ground state of this hydrogen-like atom is thus a bound state whose binding
energy measured from the bottom of the silicon conduction band is obtained by
replacing e2 by e2 /ε and m0 by me , whence
E1 me
EP = − , (4.70)
ε 2 m0
where E1 = 13.6 eV is the binding energy of the hydrogen atom. Likewise, the
associated wave function is a 1s orbital with Bohr radius equal to
aP = a1 εm0 /me . (4.71)
With ε 10 and me 0.3m0 for silicon, we obtain EP −40 meV and aP 16 Å.
The fact that aP is greater than the lattice constant of the silicon lattice justifies with
hindsight the approximation wherein silicon is treated as an effective medium for
this extra electron.
Furthermore, at zero temperature, the binding energy EP is high enough for the
extra phosphorus electron to occupy this electron state of energy
Ed = E c + E P , (4.72)
for which the wave function is a highly dilated Bohr orbit. On the other hand, the
binding energy is low enough to ensure that kB T |EP | at room temperature. The
electron is thermally excited into the states of the conduction band and thus con-
tributes to electron transport.
(a)
EF
Ev Ed Ec
(b) EF
Ev Ea Ec
Fig. 4.12 Band structures of (a) n-doped and (b) p-doped semiconductors. At zero temperature,
the Fermi level lies between the donor level Ed and the conduction band in the n-doped case, and
between the valence band and the acceptor level Ea in the p-doped case
All this can be formalised by extending the statistical physics arguments of

Chap. 3 to the band structure shown schematically in Fig. 4.12a. Here, apart from
the electron levels of the silicon valence and conduction bands, we have intro-
duced the level Ed of the extra phosphorus electron, which we call the donor level
Ed = Ec + EP . This level is not spin degenerate, because it is forbidden to put two
electrons on the corresponding Bohr orbit, the Coulomb interaction energy between
the two electrons on this orbit being greater than EP .4
At T = 0, all the electron states of the valence band and also the level Ed of the
phosphorus atoms are occupied, and the chemical potential μ = EF lies between Ed
and Ec . When T = 0, the electrical neutrality condition becomes
ne = nh + nd0 , (4.73)
where nd0 is the concentration of donors that have lost their electron (and are hence
ionised to P+ ). The latter is given as a function of the concentration nd of donor
phosphorus atoms by

nd0 = nd 1 − f (E) .
The relation (4.73) can be used to determine the exact position of μ at an arbitrary
temperature.
For temperatures such that kB T > EP = Ec − Ed , it can be checked that μ goes
below Ed and that this ends up exciting practically all the electrons of the donor
level into the conduction band, with
nh ne ≈ nd .
4 The consequences of this Coulomb interaction are examined in Problem 7: Insulator–Metal

Transition.
Substituting P atoms for Si atoms therefore creates a semiconducting compound

with majority electron doping, which is said to be n-doped. Note that the electron
and hole densities are always related by (3.54).
In the same way, some Si atoms could be replaced by atoms of a trivalent element
like B, which is an electron acceptor, introducing an acceptor level Ea above the top
of the valence band. One can thus obtain a p-doped compound, i.e., doped with
holes, with the electronic structure illustrated in Fig. 4.12b.
Note that the presence of positive charges can be detected directly by electron
transport experiments with a magnetic field B = μ0 Ha perpendicular to the current.
In the plane perpendicular to the field, the Laplace force deflects the carrier tra-
jectories in the same direction whether they be positively or negatively charged. It
follows that charges accumulate in the direction perpendicular to the current, and in
the steady-state regime, this produces an electric field transverse to the direction of
the current. The associated potential difference has the sign of the majority carri-
ers in the material. This effect, called the Hall effect, can be used to determine the
density and sign of the carriers in a semiconductor.
The industrial applications of semiconductors stem from this ability to control
the density and n or p type of the carriers by doping. This in turn led to the develop-
ment of microelectronics, where diodes and transistors were constructed by making
junctions between n- and p-doped semiconductors. Another important step was to
make transistors by applying an electric field to a semiconducting layer in order to
control the level of doping, as we shall see for the specific example of graphene.
4.6 Electrons, Holes, and Transport in Graphene
As we saw in Chap. 3, graphene has the band structure of a zero gap semiconductor.
We shall see here how the particular shape of the dispersion curves E(k) shown in
Fig. 3.19 thus leads to rather novel properties.
a. Consequences of the Graphene Band Structure
To begin with, consider the dispersion relations (3.60), taking E0 as the zero
energy. The experimental value t 3 eV is high, and we may restrict to eigenstates
of energy |E| t if we are interested in low energy properties at experimentally
accessible temperatures. Indeed, for neutral graphene, the conduction electrons and
holes appear for energies close to EF = 0. The corresponding momenta lie near the
six vertices of the Brillouin zone, to be reduced to two points K + and K − with
wave vectors Kε with ε = ±, called Dirac points (see box below). The four others
are obtained from K + and K − by translations along a vector of the reciprocal lattice
(a∗1 , a∗2 ), as shown in Fig. 4.13a. Setting
k = Kε + q ,
4.6 Electrons, Holes, and Transport in Graphene 127
E δk
θ
q v
ky ky
a∗2 a∗1 a∗2 K+
a∗1
K− K+ kx
(a) kx (b)
Fig. 4.13 (a) Hexagonal reciprocal lattice of graphene and its first Brillouin zone. The primitive
cell (a∗1 , a∗2 ) is shown, as are the two Dirac points K + and K − of this cell. (b) The conical dispersion
relation E(k) near the Dirac points is shown inside the same primitive cell. This is a zoom on
Fig. 3.19 in the vicinity of the Dirac points. The real space velocity v of an electron with wave
vector K+ + q is indicated
we may consider an expansion of (3.60) to first order in q :

3
Ek = ± at |q| . (4.74)
2
Figure 4.13b shows these dispersion relations, which are conical close to the Dirac
points. The velocity of an electron in a state Ek is given by
1 3at
vk = ∇ k Ek = ± n = ±vF n , (4.75)
h̄ 2h̄
where n is the unit vector normal to the constant energy circle at Kε + q, which
thus lies along q (see Fig. 4.13b). The electron velocity is aligned with its quasi-
momentum q and has magnitude vF = 3at/2h̄, which is independent of the electron
state. Note that the zero curvature of the constant energy surface implies that the
effective mass vanishes. Moreover, the dispersion relation (4.74) can be rewritten in
the form
EKε +q = ±h̄vF |q| . (4.76)
This dispersion relation provides a simple way of calculating the density of energy
states DS (E) per unit area and spin direction, when |E| t. The number DS (E)dE
of states with energies lying between E and E + dE is the number of electron states
contained in the area 2π qdq of reciprocal space, viz.,
DS (E)dE = 2π qdq × (2π )−2 × 2 ,
taking into account the existence of the two branches at points K + and K − . We
obtain
|E|
DS (E) = . (4.77)
π h̄2 v2F
This density of states, which vanishes at E = 0, increases linearly with energy, a

distinctly unusual behaviour compared with standard metals or semiconductors. A
detailed calculation using the general form of the dispersion curves (3.60) gives
Van Hove singularities for DS (E) when E = ±t, energies at which the dispersion
curves have saddle points, and a constant limiting value for the extrema of the energy
E = ±3t, where a parabolic dispersion relation and finite effective mass are obtained
(see Fig. 4.14).
Fig. 4.14 Density of states a2tDS(E)

per unit area for graphene, ob-
tained numerically from the
dispersion curves given in
(3.60). For |E| < t, we re- 0.3
cover the linear relation (4.77)
close to the Dirac points. Fig-
ure courtesy of Montambaux, 0.2
G.: Laboratoire de Physique
des Solides, Orsay, France
0.1
0
−3t −2t t 0 t 2t E
Dirac Points and Relativistic Physics
The similarity between (4.76) and the dispersion relation for photons, together with the vanish-
ing effective mass, has inspired an analogy between the equations of motion of the electrons
in graphene around the points K + and K − and those of massless quantum particles. The speed
vF given in (4.75), which is independent of the electron energy, plays the role of the speed of
light. For this reason, the points K + and K − are commonly referred to as Dirac points today.
For the same reason, the surfaces representing the energy dispersion relation are called Dirac
cones. The states in the conduction band then represent particles, while those of the valence
band are identified with antiparticles.
Note that the degeneracy between the two conical valleys introduces a complication which
we have hidden in (4.76), where the sign ± for the energy is obviously not related to the one
used to characterise the Dirac points Kε .
This analogy with 2D relativistic quantum physics can be taken a long way. As in this last
case, where the particle spin is represented by 2 × 2 Pauli matrices, in the case of graphene,
these matrices take into account the sublattices A and B. In this representation, the Hamiltonian
yielding the conical dispersion relations can then be written
Ĥ = h̄vF |q| ξ̂ .
The operator ξ̂ , with eigenvalues ±1, describes the projection of the sublattice ‘spin’ in the
direction of propagation of the electron in the graphene. It is analogous to helicity in particle
physics. Helicity is the projection of the real spin of the particle onto its direction of propa-
gation. This analogy leads us to consider massless charged particles, something unknown in
particle physics, since the only known massless particles, photons or neutrinos, are electrically
neutral.
b. Electron Transport
At T = 0, there are in principle no carriers, and one therefore expects zero conduc-
tivity for graphene. However, for T = 0, electrons with surface density ne are excited
in the conduction band, with a resulting hole surface density nh in the valence band,
and graphene becomes a conductor. Given the electron–hole symmetry of the band
structure, charge conservation implies that ne = nh and that the chemical potential
remains fixed at μ = EF = 0.
Question 4.5. Show that the number of carriers thermally excited varies as T 2 and that the
same is true of the electronic specific heat.
Insofar as the dispersion relation is not quadratic, and since the effective mass is
zero, the Drude relation is not applicable, in contrast to the case of standard semi-
conductors. We must therefore consider a more general expression for the electrical
conductivity, taking into account the total current of the occupied electron states, in
the form
j = −2 ev(q)f E(q) , (4.78)
q
where f (E) is the Fermi function and the factor of 2 is due to spin degeneracy. This
current is clearly zero when there is no electric field. If we assume, as discussed in
this chapter, that the electrons are subjected to viscous friction, in the presence of a
field E, the stationary solution implies a shift δk of the electronic states given by
h̄δk = −τe eE = h̄δq , (4.79)
where τe is the collision time, assumed to be the same for both electrons and holes,
since the problem is symmetric. Taking into account the displacement δq of the
Fermi surface in (4.78), we obtain

j = −2e v(q + δq)f E(q) = −2e v(q)f E(q − δq)
q q

df E(q) df E(q)
= 2e v(q) (∇ q E)·δq = 2eh̄δq 2
vx (q) .
q
dE q
dE
Equation (4.79) was used in the last equality, taking x to be a unit vector in the
direction j E δq. For temperatures such that kB T t, we can restrict the disper-
sion relations to their conical part given by (4.76). Setting θ = (q, δq) as shown in
Fig. 4.13b, this leads to

2 2 2 df E(q)
j = −2e τe vF E cos θ . (4.80)
q
dE
The sum can be replaced by an integral over $energy, after integrating over the
π
angular orientation of q at given energy E, viz., −π cos2 θ (dθ/2π ) = 1/2. We thus
obtain ∞
df
j = e τe vF E
2 2
− DS (E)dE . (4.81)
−∞ dE
The 2D conductivity can then be written in the form
∞ ∞
2 2 df 2 2 df
σ= e τe − EdE = e τ k
e B T − xdx , (4.82)
π h̄ 2
0 dE π h̄2
0 dx
where x = E/kB T. The integral has the form

∞
xdx
= ln 2 ,
0 cosh2 x
so
ln 2 e2
σ =2 τe kB T . (4.83)
π h̄2
We thus expect zero conductivity at T = 0, and growing linearly with temperature.

Note that, in two dimensions, the resistivity ρ = 1/σ corresponds to a current den-
sity j per unit width of material (not per unit area), so ρ has the dimensions of a
resistance (and the symbol Ω/ is sometimes used to express the units of ρ).
c. Doping by Field Effect
It is certainly an interesting fact that graphene is metallic at room temperature be-

cause carriers are thermally excited. However, for applications in electronics, it is
essential to be able to construct active components producing some specific response
to an applied voltage. One of the most widely used approaches with semiconduc-
tors is to make a metal oxide semiconductor capacitive structure, called a MOSFET
(MOS field effect transistor), to control the doping of the semiconductor through an
electric field obtained by applying a voltage to the metal gate.
This effect can be illustrated by the equivalent structure used for graphene, shown
schematically in Fig. 4.15a. In this case, the gate is a doped Si substrate, oxi-
dised at the surface to produce a film of SiO2 with a thickness d of a few hun-
dred nanometers. The graphene layer constitutes the second plate of the capacitor.
When a voltage VG is applied across the gate and the graphene, the charges
appearing on the capacitor plates dope the graphene. If ε is the dielectric constant
of the oxide layer, the capacitance C = εε0 /d per unit area of the junction induces a
charge
εε0
n= VG . (4.84)
ed
(a) 6 (b)
50 K
/ )
Gold
contacts 4 ρ
Resistivity (k
1/ρ
I
V
2
VG Doped Si SiO 2
0
−50 −25 0 25 50
VG (V)
Fig. 4.15 (a) Semiconductor–oxide–graphene structure used to dope the graphene sheet by varying
the voltage VG applied to the semiconducting gate. (b) Dependence of the resistance and conduc-
tivity of graphene on VG . Figure courtesy of A.K. Geim, from results published by Morozov, S.V.,
et al., Phys. Rev. Lett. 100, 016602 (2008)
The surface density n of the carriers induced in graphene by an electric field

thus varies linearly with the gate voltage VG . We therefore observe, e.g., in
Fig. 4.15b, that the resistivity of graphene falls off rapidly as a function of the gate
voltage.
Question 4.6. 1. Which experimental observations in Fig. 4.15b can be explained by the
approach discussed in Sects. 4.6b and c? Which cannot be explained? Compare the
density of carriers induced experimentally by the field effect with the one obtained in
Question 4.5 at thermodynamic equilibrium at 50 K (ε = 4, d = 300 nm).
2. What becomes of the chemical potential μ of the carriers in this experiment?
Note that the observations made with the atomic layer of graphene are similar to
those carried out on silicon MOSFET transistors, which are made by depositing a
metal film on a much thicker semiconducting layer whose carrier density is con-
trolled by the voltage across the metal gate. In the case of graphene, the change in
the resistance induced by a voltage VG = 10 V is of the order of a factor of 10,
which is a significant transistor effect. This type of structure has been used over the
past few years to demonstrate many novel physical properties in graphene, suggest-
ing applications for graphene or graphene multilayers in nanotechnology in the near
future.
Question 4.7. What experiment could determine the sign of the injected carriers? How
could one check experimentally whether the graphene energy spectrum does indeed cor-
respond to that of the Dirac cones?
4.7 Summary
Electrical conductivity is a non-equilibrium electron transport phenomenon. An

electric potential gradient induces charge transfer which, in a gas of non-interacting
classical or quantum particles, should lead to an accelerated, hence non-linear
motion of the electrons as a function of the applied electric field E. Ohm’s law,
which says that j = σe E, can thus only be explained if there is a frictional force due
to the collisions suffered by the electrons, whereby the electron velocities can reach
a steady state equilibrium distribution. The relaxation time τe of the velocity dis-
tribution is the average time between collisions for a given electron in the classical
Drude model. We thus define the electron conductivity σe and the mobility μe by
ne2 τe
σ= = neμe .
m0
In the more realistic case of a free fermion gas, the effect of an electric field is to
translate the points representing the electron states in the space of quasi-momenta
k by an amount h̄δk = −eEτe , where τe is now the collision time of the electrons
whose representative states lie on the Fermi surface, i.e., have an energy close to EF .
For these electrons, the electron mean free path is thus e = vF τe .
In a metal, the electrical and thermal conductivities of electronic origin, σ and κ,
respectively, are generally limited by the same collision processes, and this leads to
the Wiedemann–Franz law:

κ π 2 kB 2
= T =LT .
σ 3 e
The group velocity of an electron wave packet built up from Bloch states of quasi-
momentum k is given directly by the dispersion curves En (k) of the associated elec-
tronic band. In three dimensions, it is normal to the corresponding constant energy
surface in k space, and is given by
1
vnk = ∇ k En (k) .
h̄
The effect of a force on a Bloch state of quasi-momentum k is to shift the point
representing the state by an amount h̄dk/dt = F.
In real space, the electron acceleration is related to the force by
dvα 1
= Fβ ,
dt (me )αβ
β
where (me )αβ is the effective mass tensor, related to the curvature tensor of the
constant energy surfaces in reciprocal space.
Experimental observations show that various collision processes contribute to the
resistivity: collisions with impurities for which τe does not depend on temperature,
and interactions with vibrational modes of the crystal lattice, for which τe falls off
at high temperatures.
In semiconductors, two types of carrier contribute to the resistivity, namely, the

electron states at the bottom of the conduction band and the unoccupied states at the
top of the valence band, which are called holes. The latter can be treated as positively
charged carriers, with the velocity of the missing electron, but with energy, quasi-
momentum, and effective mass opposite to those of the missing electron.
The density of the carriers, electrons or holes, can be controlled in a semiconduc-
tor by doping it with heterovalent substitution impurities. For example, phosphorus
is an electron donor in silicon, whereas boron is an acceptor. The excess electron
of the donor is trapped at low temperatures by the Coulomb potential of the donor,
on a hydrogen-like atomic orbital constructed from Bloch states of the conduction
band. The effective Coulomb potential is reduced by the dielectric constant ε of the
material. As a consequence, this electron state has a lower binding energy than the
states of the conduction band, being reduced to
E1 me
EP = − ,
ε 2 m0
while its dilated Bohr orbit has radius
aP = a1 εm0 /me .
The Fermi level lies between the donor level and the conduction band, but at high
temperatures, this donor state is thermally excited toward the electronic states of the
conduction band. In this way, n-doped semiconductors can be obtained, in which the
majority carriers are electrons. In a similar manner, an acceptor produces a bound
hole state at low temperatures, with slightly higher energy than the top of the va-
lence band. This hole is excited into the valence band at high temperature, thereby
producing a p-doped semiconductor.
Junctions between n- and p-doped semiconductors can be exploited to make di-
odes and transistors, which have spurred the massive expansion of microelectronics.
The novel band structure of graphene, with its two Dirac points and an electronic
density of states that varies linearly with energy, leads to original electron trans-
port properties for this 2D system. In particular, a significant change in the carrier
density can be induced by applying an electric field perpendicular to the graphene
sheet. This structure is analogous to MOSFET structures obtained with standard
semiconductors, and is expected to give rise to specific applications.
Odd Man Out on Page 100
Most of the everyday objects illustrated in the figure make use of electrical or
thermal conductivity of electrons in a metal:
• The thermometer uses both effects: thermal conductivity of the metal tip which
thermalises at the temperature of the tested substance, and electron transport in a
metal or semiconductor, exploiting the temperature dependence of a resistance,
for example, to indicate the temperature. Electrical conductivity is clearly as cru-
cial in the underlying measurement and display circuitry.
• The iron exploits both effects, one to heat a resistance by electrical conductivity
and the Joule effect, the other to heat the iron shoe which is itself metallic.
• The lamp and fuses use electrical conductivity and the Joule effect to heat the
filament or melt the metal in the fuse to cut the current.
• The credit card stores data in magnetic form. This data is read by an electronic
chip which uses electronic circuits and hence exploits electron transfer for read
and write operations. Recorded data is only useful if it can be read.
• The kettle heats a liquid by thermal conductivity of electronic origin in the metal
element at the bottom of the container.
• The outdoor heater uses the infrared radiation produced by a chemical reaction,
namely oxidation of butane or propane, to heat objects receiving this radiation.
The metal surface of the reflector reflects infrared radiation back down. How-
ever, there is no electron transport here. Moreover, this reflection serves only to
avoid energy loss in the upward direction. The heater works, but in a less energy
efficient way, without the reflector.
We may conclude that the outdoor heater is the odd man out in this set of everyday
objects.
Question 4.1
With the Drude hypothesis, the equation of motion of an electron is

dve mve
m = eE − ,
dt τe
where ve is the drift velocity and τe the collision time. With E = E(ω)eiωt and ve =
ve (ω)eiωt , we easily obtain
mve (ω)
imωve (ω) = eE(ω) − ,
τe
and hence,
ne2 E(ω) 1
j(ω) = neve (ω) = = σ (ω)E(ω) ,
m iω + 1/τe
where
σ (0) ne2 τe
σ (ω) = , σ (0) = ,
1 + iωτe m
so that
σ (0) σ (0)ωτ
σ1 (ω) = , σ2 (ω) = .
1 + ω2 τe2 1 + ω2 τe2
The function σ1 (ω) is shown graphically in Fig. 4.16.
Fig. 4.16 Graph of the σ 1 (ω )

function σ1 (ω)
σ (0)
σ (0)/ 2
1/τ e ω
Question 4.2
1. The contribution of impurities to the resistivity depends linearly on their concen-

tration c. In the case of Ni or Fe, this can be checked by plotting the difference
ρ between the resistivities of the alloy and the pure metal as a function of c (see
Fig. 4.17).
2. It is easy to determine ρ for In, Sn, Sb, and As impurities, and normalise it by
the concentration c, to give ρi = ρ/c.
3. ρi is plotted as a function of Z in Fig. 4.18a. We observe that the resistivity
due to impurities grows rapidly with Z. This happens because the relaxation
rate τe−1 is determined by the probability of Bloch electronic states being scat-
tered by the potential associated with the impurity. As the Cu ions are in a 1+
state, the extra Coulomb potential introduced by an impurity is determined by the
charge difference between the ion associated with the impurity and Cu+ , that is,
Z = Z − 1. The probability of an electron being scattered by the extra Coulomb
6
5 Fe
m)
4 Ni
Δ ρ (10−7
3
2
1
Fig. 4.17 Difference ρ be-
tween the resistivities of an al- 0
loy and the pure metal as a 0 1 2 3
function of the concentration c c (%)
(a) (b)
Fig. 4.18 (a) ρi as a function of Z. (b) ρi as a function of (Z)2
potential Ze2 /4π ε0 r introduces the square of a matrix element of this potential.
The scattering probability must therefore go as (Z)2 . It can be checked that the
experimental results do indeed obey such a law (see Fig. 4.18b). A quantitative
calculation of the resistivity due to impurities could be done using the Born ap-
proximation, as for collision cross-sections between two particles [21, Chap. 6].
Note that Fe and Ni impurities behave anomalously. These are magnetic impuri-
ties, for which this approach is not appropriate.
Question 4.3
The effective masses are obtained from the curvatures of the bands. For free elec-
trons, an√energy E = 1 eV = 1.6 × 10−19 J would correspond to a wave vector
k = h̄−1 2m0 E = 5.4 × 109 m−1 = 0.49(2π/a). It can be checked from Fig. 3.29
that the curvature of the conduction band in the (111) direction is close to what is
found for free electrons. There is a corresponding effective mass me ∼ m0 . For the
valence bands, the last band of holes (doubly degenerate) has higher curvature: on
the right, 1 eV corresponds to 0.25(2π/a), so that mh ∼ 0.25m0 . The second band
of holes, with a width of about 8 eV, has a much greater curvature, and thus corre-
sponds to ‘lighter’ holes, with mh ∼ 0.08m0 .
Question 4.4
The optical excitation of an electron by a direct transition from a state ke in the

valence band (VB) to a state with the same wave vector in the conduction band
(CB) leaves a hole of wave vector kh = −ke in the valence band. If we apply an
electric field E to this system, we show that the electric currents due to electron and
hole drift are additive, for a standard band diagram like the one in Fig. 4.19 where
the effective masses me and mh of the electrons and holes are positive and isotropic.
According to the real space acceleration theorem, the motion of an electron in
the presence of a field E is given by
Fig. 4.19 Excitation of an electron–hole pair by direct optical transition. CB conduction band. VB
valence band (left). Electron and hole drift currents je and jh in the presence of an applied
electric field E (right)
dve
me = −eE ,
dt
and the velocity and current acquired over a time lapse t are
et e2 t
ve = − E, je = −eve = + E.
me me
For the hole,
dvh et e2 t
mh = eE , vh = E, jh = E.
dt mh mh
The electric currents due to electron and hole drift thus lie in the same direction as
E, and are therefore additive. This amounts to saying that the conduction band and
the valence band each contribute to the total current, one by an amount jh , and one
by an amount je . Figure 4.19 shows this result schematically.
This result is quite general, i.e., the total current is the sum of the contributions
from the different bands, which are described in ‘electron language’ or ‘hole lan-
guage’, depending on what is most convenient.
Question 4.5
We have
3t ∞
ne = nh = f (E)DS (E)dE 2(π h̄2 v2F )−1 Ef (E)dE ,
0 0
since, for kB T t, we can use (4.76) for the density of states, and sum to infinity:
2 ∞
2 2kB T
ne = nh = f (x)xdx .
π 3at 0
The number of carriers does indeed 2

$ ∞go as T , and we may estimate its order of
magnitude per unit area, given that 0 f (x)xdx = R(2) = π 2 /12:
2
2π kB T
ne = nh = . (4.85)
27 at
For graphene, a = 1.42 Å and t = 3 eV, which yields
ne ≈ 8 × 1010 cm−2 at T = 300 K . (4.86)
The specific heat can be evaluated approximately by noting that the energy of this
set of electrons which scales with T 2 is increased by kB T, and the internal energy
increases then as T 3 , and the specific heat as T 2 . More precisely,
∞ ∞
U=2 Ef (E)DS (E)dE = 2(π h̄2 v2F )−1 E2 f (E)dE ,
0 0
so ∞
U = 2(kB T)3 (π h̄2 v2F )−1 f (x)x2 dx ,
0
$∞
and with I(x) = 0 f (x)x2 dx = 4R(3)/3 ≈ 1.8, we obtain
dU
C= = 10.8kB (kB T)2 (π h̄2 v2F )−1 . (4.87)
dT
Note that these quadratic dependences expected for the number of carriers and the
specific heat bear no relation to what happens in standard metals or semiconductors.
These are direct consequences of the singular electronic structure of graphene. As
we shall see later, they have never been observed so far.
Question 4.6
1. Experiment shows that the resistivity decreases when either a negative or a posi-
tive voltage is applied, i.e., when either holes or electrons are created. This is to
be expected if the band structure is symmetric. Note that the field effect is very
efficient experimentally, because the conductivity σ increases very quickly and
almost linearly as a function of VG .
Using ε = 4 and d = 300 nm, we obtain the density of carriers created:
n = 7 × 1010 |VG | (cm−2 /V) . (4.88)

For comparison, according to (4.85), the expected content of thermally excited

carriers at 50 K is n = 2.2 × 109 (cm−2 ), so an applied voltage of |VG | 32 meV
would suffice to create an equivalent number of carriers. Such a gate voltage
should double the conductivity, but this is only observed in Fig. 4.15b for a volt-
age two orders of magnitude greater. It thus looks as though (4.85) does not apply,
since for VG = 0, it corresponds to far fewer thermally excited carriers than are
actually observed experimentally.
This certainly means that there are carriers due to charged defects causing
extrinsic chemical doping. But this kind of doping, if uniform and hence of defi-
nite sign, would break the electron–hole symmetry. For example, electron doping
should be reduced to zero by applying a suitable negative voltage. One would
thus expect to see the minimum of σ (VG ) shift toward the right, quite the oppo-
site of what is observed! The fact that the conductivity curve does not break the
electron–hole symmetry can only be explained by an inhomogeneous situation
on the graphene layer, which must comprise some positively charged and some
negatively charged regions, with average close to zero. The spatial extension of
these regions induced by the distribution of defects at the interface with the SiO2
layer prevents the recombination of electrons and holes.
Figure 4.20 shows experimental curves for σ = 1/ρ at various temperatures,
similar to the one in Fig. 4.15b obtained at 50 K. It shows that these conduc-
tivities, and in particular the minimum of the conductivity at VG = 0, vary little
with T. Moreover, the carriers associated with defects and those injected by field
effect seem to behave independently. Indeed, we may suppose that
ρ = ρ1 (VG ) + ρ0 , (4.89)
and thereby deduce an experimental value of σ1 (VG ) = 1/ρ1 (VG ). The result is
shown in Fig. 4.20. It is found that σ1 (VG ) is proportional to VG and almost
independent of T. This is not obvious a priori, and is hence a highly relevant
experimental observation. We shall discuss the implications below.
2. The dependence of n on VG corresponds to injection of electrons (VG > 0) in
the conduction band or holes (VG < 0) in the valence band of graphene. It thus
corresponds to a shift in the chemical potential for different values of VG .
At T = 0, if the dispersion equation (4.76) applies, and if we set EF = h̄vF qF , the
number of electrons with energy E < EF per unit area is 2 × 2 × π q2F /(2π )2 = n.
The relation between EF and the doping is therefore
EF = h̄vF qF = h̄vF (π n)1/2 . (4.90)
This dependence of EF on VG , and the corresponding occupation of the Dirac

cones on the doping, are both illustrated in Fig. 4.20.
Note that we have taken into account the effect of the carrier number on the
conductivity and the Fermi level, but we have not yet considered the relevance of
the collision time τe , which is the key factor limiting conductivity. Of course it
comes into the calculation of Sect. 4.6c, which leads to the expression (4.84) for
1/ρ1
6 EF
20 K
100 K
EF 180 K
220 K
σ (1/kΩ)
4 260 K
1/ρ
2
0
−50 −25 0 25 50
VG (V)
Fig. 4.20 Experimental curves for σ = 1/ρ vs. gate voltage at different temperatures (see text).
Figure courtesy of A.K. Geim, from results published by Morozov, S.V., et al., Phys. Rev. Lett.
100, 016602 (2008)
σ when there is no field-effect doping. Note, however, that when there is doping,
(4.82) remains perfectly valid, provided that we take the Fermi function with the
chemical potential μ = EF given by the above result. In this case, the calculation
proceeds easily for kB T EF , because in this case, −df /dE can be replaced by
a Dirac delta δ(E − EF ) and we have
1 e2
σ = e2 τe v2F DS (EF ) = τe |EF | . (4.91)
π h̄2
The fact that the experimental conductivity is almost independent of T implies

that τe must also be temperature independent and hence determined rather by de-
fects than by phonons. In a conventional metal, the Drude relation for σ would
naturally lead to a conductivity proportional to n, and hence proportional to VG .
But according to the expression obtained above, the conical dispersion relation
1/2
of graphene leads rather to σ ∝ τe EF ∝ τe n1/2 ∝ τe VG . One would then ex-
1/2
pect σ ∝ VG , whereas√experiment
√ shows that σ ∝ n ∝ VG ! This can only be
explained if τe goes as n ∝ VG , and this is counter-intuitive, since it implies
that the more we dope, the less the defects will scatter electrons! (Are the defects
screened by the carrier density?)
These experimental results are still under intensive investigation to determine
which defects cause collisions. Could it be charged defects on the interface layer
of SiO2 which arise with doping? And we still need to understand what increases
the conductivity at VG = 0.
Even if defects limit conductivity, the latter remains high. We can define an
electron mobility μe by writing σ in the form σ = neμe , and compare with
μe for doped semiconductors. The experimental value obtained for graphene is
μe 200, 000 cm2 /V s, which is huge at room temperature when compared with
the cleanest semiconductors we can produce.
Question 4.7
As for semiconductors, the carrier sign can be ascertained by measuring the sign of
the Hall effect, i.e., the transverse voltage induced in the presence of a current in
graphene when a magnetic field is applied perpendicularly to the graphene plane.
The experimental results in Fig. 4.21a show that the Hall voltage changes sign for
VG = 0. The reciprocal of the Hall voltage, called the Hall constant (ρxy )−1 = nq/B,
can be used to determine nq, where q is the carrier charge, which is indeed found
experimentally to be proportional to VG , except for VG ∼ 0, as we saw above.
Cyclotron frequency measurements can be used to determine the form of the dis-
persion relations. Recall that cyclotron modes are oscillatory eigenmodes of elec-
trons in an applied magnetic field B, determined from the area of the electron orbits.
In semiconductors, the cyclotron frequency ωc = eB/mc is used to obtain a cyclotron
mass mc related to the effective mass of the carriers.5 It is given quite generally by

h̄2 dS
mc = , (4.92)
2π dE E=EF
where S is the area of the electron orbit on the surface of energy E in reciprocal
space. In the case of the Dirac cones,
2
E
S(E) = π [q(E)] = π
2
, (4.93)
h̄vF
(a) (b)
−1 (1/k Ω )
ρ xy
B = 2T 0.6
10
T = 10 K
mc /ms
0.4
0
0.2
−10
0
−50 0 50 −6 −3 0 3 6
VG (V) n (10 12 cm −2 )
Fig. 4.21 (a) Hall effect conductance versus gate voltage showing the linear increase in carrier
content of either sign except near the neutral point. (b) Cyclotron resonance mass versus carrier
content for graphene. Figures courtesy of A.K. Geim, from results published by Novoselov, K.S.,
et al., Nature 438, 197 (2005)
5 See Problem 8: Cyclotron Resonance.

and then
EF h̄(π n)1/2
mc = = . (4.94)
(vF )2 vF
The experimental results in Fig. 4.21b confirm the n1/2 dependence, i.e., going as
1/2
VG . The Dirac cone dispersion relations are thus perfectly corroborated. Note that
mc is not at all the effective mass, which is zero for graphene.
The point about these experiments is that they help us to distinguish the ef-
fects of collision time and carrier number on transport phenomena. With cyclotron
resonance, the magnetic field used must nevertheless be strong enough to ensure
the condition ωc τe 1. This generally requires low temperature experiments to in-
crease τe . For graphene, τe is independent of temperature, as noted from Fig. 4.20,
and the mobility is high enough to allow cyclotron frequency measurements at
T = 300 K, like the ones shown in Fig. 4.21b.
In order to determine the dispersion relations E(k) directly, another method
described in Chap. 3 is angle-resolved photoemission spectroscopy (ARPES). In
2D systems, E(k) can be determined directly for occupied states by analysing the
energies of electrons emitted in one direction by incident photons. For graphene, as
shown in Fig. 4.22, the linear form of E(k) is observed directly at the Dirac point
for the π electron band. This experiment requires the surface to be homogeneous
over the extent of the incident photon beam, viz., 40 μm, and could not be carried
out for graphene deposited on SiO2 owing to the disorder induced by the substrate.
As mentioned in the solution to Question 4.6, this disorder-induced doping varies
over an area comparable with the beam area. The results of Fig. 4.22 were obtained
on a graphene multilayer deposited by epitaxy on the face of an SiC single crystal
made up entirely of carbon atoms. The graphene layers formed here do not stack up
0
−0.2
E − EF (eV)
−0.4
0.1
−0.6
−0.8 0
−1 −0.1
−0.2 0 0.2 −0.2

ky (A−1)
−0.3
−0.4
0.05 0.1 0.15 0.2
Fig. 4.22 ARPES observation of the conic dispersion relation of graphene (see text). Image cour-
tesy of C. Berger, from results published by Sprinkle, M., et al., Phys. Rev. Lett. 103, 226803
(2009)
as in graphite and are only weakly coupled. The experiment shows that they behave
like isolated graphene sheets. It can be seen from Fig. 4.22 that the dispersion is
indeed linear, and on the zoom, that the point of maximal intensity corresponding to
the Fermi level lies less than 20 mV from the point E = 0 of the dispersion curve.
According to (4.88), this shows that the doping induced by the SiC substrate is less
than 1010 holes/cm2 over the beam area. Also on Fig. 4.22, note the existence of a
second, less intense Dirac cone. It corresponds to a graphene sublayer rotated with
respect to the surface layer.
Chapter 5
Introduction to Superconductivity
Contents
5.1 Conditions for Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

5.1.1 Persistent Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.1.2 Critical Field and Critical Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5.2 Difference Between a Perfect Conductor and a Superconductor:
The Meissner Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.1 Effect of a Magnetic Field on a Perfect Conductor . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.2 Meissner Effect: Exclusion of the Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2.3 London Equations and Penetration Depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.3 A Macroscopic Quantum Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
5.3.1 Macroscopic Wave Function and Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
5.3.2 Quantisation of Magnetic Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3.3 Measuring the Flux Quantum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.3.4 Josephson Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.3.5 SQUID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
146 5 Introduction to Superconductivity
The levitation effect raises many questions. Peter on the right is practically minded and imagines
how this phenomenon might be put to use, while Claire and David, who have read up to Chap. 7,
try to visualise the strange mechanism that governs the behaviour of the electrons in the supercon-
ducting compound
5.1 Conditions for Superconductivity 147
We have seen how the conductivity of a metal is limited by collisions, which deter-
mine the electron mean free path. After a collision, an electron completely loses all
memory of its quantum state as specified by its quasi-momentum k. It is thus im-
possible to follow a Bloch state over any distance much greater than the mean free
path. To understand the microscopic origin of these collisions, one had to measure
the conductivity of very pure metals at low temperatures. This was made possible by
the work of the physicist Kammerlingh Onnes, who specialised in the liquefaction
of gases and opened the way to the use of cryogenic fluids. In 1911, he succeeded in
liquefying 4 He, at a temperature of 4.2 K. He then suggested using the low temper-
atures created in this way to study the low temperature conductivity of pure metals.
Quite unexpectedly, he discovered that the conductivity of mercury increased by
several orders of magnitude at temperatures below 4.18 K. This discovery of super-
conductivity remained a mystery for more than 40 years. With hindsight, it is clear
that the prerequisites of quantum mechanics had not yet been established. Of course,
this did not prevent the physicists of the first half of the twentieth century from grad-
ually getting a hold on the fundamental manifestations of this phenomenon through
ideas based entirely on experimental observations.
As mentioned in the Preface, we shall present superconductivity here in a way
that closely follows historical progress, to show how scientific research actually pro-
ceeds in practice. In Sect. 5.1, we indicate how existence of a superconducting state
can only be established by demonstrating the occurrence of persistent currents. In
Sect. 5.2, we show that this phenomenon cannot simply be attributed to the conduc-
tivity of a perfect metal. We can then discuss an effect discovered by Meissner in
1933, namely the total expulsion of magnetic flux from a superconducting material.
We go on to describe the phenomenological explanation of this novel magnetic man-
ifestation of superconductivity, due to the London brothers in 1935. In Sect. 5.3, we
suggest how the basic ideas of quantum mechanics allow one to establish the quan-
tisation of the magnetic flux, on the basis of these two phenomena. Experimental
observations then justify the idea that electrons are paired in the superconducting
state.
5.1 Conditions for Superconductivity
5.1.1 Persistent Currents
It was by trying to understand what limits the conductivity of a pure metal at low
temperatures that Onnes discovered that this conductivity becomes almost infinite
at low T in some metals. In his first published experiments, Onnes observed that,
below a temperature of 4.18 K, a mercury wire has a resistance that plummets from
0.1 Ω to well below 10−4 Ω. This sudden change in the resistance at a critical tem-
perature Tc cannot of course be taken to show that the resistance has completely van-
ished, but it does raise the question as to how the acccuracy of electrical resistance
ρ (μ cm) ρ (μ cm)
100
60
10
40
1
20 0.1
0 0.01
90 92 94 90 92 94
T (K) T (K)
(a) (b)
Fig. 5.1 Temperature dependence of the resistivity of a sample of the cuprate YBa2 Cu3 O7 . (a)
The resistivity is observed to drop abruptly at 91.7 K in an experiment similar to the one shown
in Fig. 4.6. (b) The measured voltage is plotted on a logarithmic scale. A residual voltage of
experimental origin limits the accuracy of the measurement. Figures courtesy of Rullier-Albenque,
F.: SPEC, CEA. Saclay, France
measurements might be improved. In the typical setup of Fig. 4.6, the results are
clearly limited by the accuracy of the voltmeter (see Fig. 5.1). For a given sample
and voltmeter, the most naı̈ve idea would be to increase the current. It is then ob-
served that the resistance of a normal metal reappears once the current becomes too
large. It would thus seem that there is a critical current density jc above which the
superconducting state cannot exist.
How then can we determine whether the superconductor has nonzero resistance?
To this end, it seems more interesting to ask whether we can demonstrate the ex-
istence of dissipation in a superconductor in which there is a current less than jc .
Indeed, we may consider a closed loop of conducting material of inductance L and
set up a current in it. If the loop has resistance R, the current I satisfies
dI
L + RI = 0 ,
dt
which implies that
Rt
I = I(0) exp − , (5.1)
L
in the case where a current I(0) has been established in this loop at time t = 0.
Any resistance, no matter how small, will thus determine the time constant for the
current to fall off. But how could such an experiment be realised in practice? How
can a current be made to circulate in a closed loop, and how can we measure the
way it varies in time?
There is no question of using a current generator, since such a device introduces
normal metal conductors and hence dissipation, and it would have to be discon-
nected in order to close the loop! Fortunately, classical electromagnetism tells us
that a current can be created by induction by applying a magnetic field. We may
thus apply a magnetic field Ha = Ba /μ0 in a direction perpendicular to the plane of
the loop, for a very short time, by means of some external source. Lenz’s law tells
us that the current induced in the conductor will oppose the time variation of the
magnetic flux Φ through the loop due to Ba (t). We may thus write
dΦ dBa dI
− = −S = RI + L ,
dt dt dt
where S is the area spanned by the loop. If the variation of Ba is produced in a very
short time compared with the time constant L/R of the loop, R can be neglected
during the time over which the current is being set up, which implies that
d(Ba S + LI)
=0.
dt
We thus find that, if R is small, the total magnetic flux, which is the sum of the flux
due to the applied magnetic induction Ba and LI due to the induced current, will be
constant inside the loop. For a change Ba in the magnetic induction, the induced
current is
SBa
I=− .
L
It only remains to measure the time variation of I and we may determine R using
(5.1). If R is really zero, the current will persist indefinitely. We thus see that, if
the superconductor is indeed a perfect conductor, it will be able to accommodate a
persistent current.
In practice, there is little hope of measuring the time dependence of I. In fact, it
is preferable to observe B. To increase sensitivity, it is easy to understand that the
most effective procedure will be to apply an external field Ba when the conductor is
resistive, i.e., at a temperature T > Tc . In this case, the resistance in the normal state
is big enough for the current to drop to zero rather quickly. The loop is then cooled
below Tc in the presence of the applied field, which in principle should change
nothing.1 The applied field Ba is subsequently reduced to zero over a very short time
Φ = Ba S
Ba
(a) (b)
Fig. 5.2 (a) Cooled metal loop in the normal state in the presence of an applied magnetic field.
(b) The field is switched off when the metal has been cooled to the perfectly conducting state.
According to Lenz’s law, the current induced in the loop is such as to oppose the change in the
magnetic flux through the loop. The current persists indefinitely in zero field if the conductor is
perfect
1 We shall see in Sect. 5.2 that changes in B nevertheless occur within the superconducting loop.
lapse, thereby inducing a current I = SBa /L. The lines of the magnetic flux density
are illustrated in Fig. 5.2 before cooling and after switching off the magnetic field.
If the initial time t = 0 is taken to be the moment when Ba has dropped to zero, the
magnetic flux induced by the current I is equal to SBa , and decreases exponentially
in time with time constant L/R. We thus measure the time dependence of B at the
center of the loop in order to determine this time constant and hence R. A very accu-
rate experiment was carried out in 1961 with a superconducting loop, and the value
of B was determined by measuring the Larmor resonance frequency of the nuclear
spins of the protons in a sample placed at the center of the loop. This frequency can
be measured extremely precisely, and no change was observed over two years. This
led to an upper estimate for the resistivity of a superconducting wire of 10−26 Ω m, a
value at least 18 orders of magnitude lower than that of copper at room temperature.
Superconductivity therefore corresponds to a new and profoundly original physical
state. For comparison, note that the resistivity of the best conducting materials in
their normal state is barely 20 orders of magnitude lower than those of the most
perfect known insulators.
5.1.2 Critical Field and Critical Current
We saw above that the fact that the resistance of a superconductor is zero led natu-
rally to the idea of using magnetic fields to study the behaviour of superconductors,
and this right from the moment that superconductivity was first discovered. The
above experiment suggests applying the strongest possible fields Ha in order to in-
crease the sensitivity of the measurements. But in fact it was observed instead that
superconductivity disappears for strong fields: if a current j jc is driven through a
superconducting wire, it is observed that, beyond a certain value Hc of the applied
field, a potential difference appears between the measurement contacts. This shows
that a magnetic field stronger than Hc destroys the superconducting state.
The temperature dependence of Hc was thus investigated, thereby determining
the domain of existence of the superconducting phase. The experimental results
shown in Fig. 5.3 can be represented approximately by the empirical law
2
T T
Hc = Hc (0) 1 − . (5.2)
Tc Tc
This limitation by an applied field shows that superconductivity is essentially a mag-

netic phenomenon, other manifestations of which will be discussed below.
But what exactly is the relevance of this critical current for the existence of the
superconducting state? In other words, does the disappearance of the superconduct-
ing state in the presence of a high enough current point toward some novel physical
effect, or is it related to the field induced by this current? It is easy to see that, for a
Fig. 5.3 Curves representing 7

the field Hc destroying super-
conductivity as a function of 6
the temperature T for differ-
ent superconducting metals. 5
The curves are approximately
modelled by (5.2)
μ 0 Ha (tesla)
4
Pb
3
2 α -Hg
Sn
1 In
Al
0 1 2 3 4 5 6 7 8
T (K)
zero applied field, a current I in a superconducting wire of radius a (see Fig. 5.4a)
will induce a tengential field H = I/(2πa) at its surface. If H > Hc , the supercon-
ducting phase will be destroyed at the surface, and the superconducting region could
only survive for a radius smaller than a. All the current would then have to flow
within this superconducting core, and this would generate an even stronger field.
This implies that the superconductivity must disappear suddenly and completely
whenever I = 2πaHc . This observation, first made by Silsbee, suggests that the rele-
vant physical quantities limiting the existence of the superconducting phase are the
temperature Tc and the field Hc (0).
In the geometry of Fig. 5.4a, this defines the simple relation
Ic = 2πa(Hc2 − Ha2 )1/2
I
Critical current Ic
Ha 2π aHc
P I
H=
Ha 2π a
0 Hc Ha
(a) (b)
Fig. 5.4 (a) Setup for measuring the critical current in a cylindrical wire, as a function of an
external field Ha applied along the axis of the wire. (b) Observed dependence of Ic on Ha . Adapted
from [18, p. 84]
between the critical current and the critical field (see Fig. 5.4b). Paradoxically, this
result, which can be checked experimentally, would appear to show that the criti-
cal current density jc is inversely proportional to the radius rather than the cross-
sectional area of the superconductor.2
Note that, for the materials discovered in the first half of the twentieth century,
the critical temperature Tc never exceeded 20 K, while Hc was limited to 0.2 tesla.
It was this second limit that blocked the way to any possible application of super-
conductivity for creating permanent magnetic fields, because the latter were then
limited to values well below what could be achieved with soft iron magnets.
5.2 Difference Between a Perfect Conductor

and a Superconductor: The Meissner Effect
5.2.1 Effect of a Magnetic Field on a Perfect Conductor
As superconductors are perfect conductors, it is important to understand the expected

effect of a field on a perfect conductor. When there is no collision term, the
acceleration of the electrons in the presence of an electric field E is
dv
m0 = −eE ,
dt
and if there are n electrons per unit volume, the current density is given by j = −nev.
This implies the first of the London equations:
dj ne2
= E. (5.3)
dt m0
Maxwell’s equations can be used to evaluate the behaviour of this perfect conductor
in the presence of a magnetic induction B :
dB dE
curl E = − , curl B = μ0 j + ε0 μ0 .
dt dt
The first Maxwell equation can be combined with the first London equation (5.3) to
give
m0 dj dB
2
curl =− . (5.4)
ne dt dt
In the steady state E = 0, otherwise j would grow indefinitely, according to (5.3). In
addition, we consider only fields and currents that vary slowly in time, thus exclud-
ing for example the response to an electromagnetic wave. In this case, E and dj/dt
2 We shall see that in reality currents only flow at the surface of the wire, over some thickness λ
(see Sect. 5.2). The critical current density is thus in this surface layer jc ≈ Hc /λ.
5.2 Difference Between a Perfect Conductor and a Superconductor 153
are both small, and since dE/dt is second order compared with dj/dt, the displace-
ment currents are negligible in the second Maxwell equation. The latter therefore
simplifies to curl B = μ0 j. Substituting into (5.4), we then obtain

dB m0 dB
− = 2 curl curl ,
dt ne μ0 dt
and hence

d m0
B + 2 curl curlB = 0 . (5.5)
dt ne μ0
Since curl curlB = graddivB − B and divB = 0, and setting
m0
λ2L = , (5.6)
ne2 μ0
where λL has the dimensions of length, it follows that
d(B − λ2L B)

=0. (5.7)
dt
For a conductor filling a half-space bounded by a plane surface perpendicular to the
x axis and subjected to an applied magnetic induction Ba (t) parallel to its surface,
the above equation leads to

dB(x) dBa x
= exp − ,
dt dt λL
implying that, at a great distance from the surface as compared with λL , the mag-
netic induction B(x) is time independent. In other words, a change in Ba induces
surface currents which oppose any change in the magnetic induction inside the per-
fect conductor (see Fig. 5.5).
The value of B(x) inside the metal is entirely determined by the field there when
it was cooled below Tc , and it does not vary in time for a perfect conductor. This
is illustrated in Fig. 5.6a. This result tells us that, under an applied field Ha , the
magnetic induction within a perfect conductor will differ depending on whether
dB a dB(x)
dt dt
Fig. 5.5 Variation of dB(x)/dt

inside a perfect metal obtained
Vacuum Perfect
when a variation dBa /dt is
conductor
imposed on the applied field.
It goes to zero exponentially
over a distance λL which
corresponds to the thickness
of the layer containing the x
surface currents λL
Room Room
Ba = 0 Ba Ba = 0 Ba
temperature temperature
Temperature
drop
Ba Ba
Ba Low T Ba Low T
Ba → 0 Ba → 0
(a) Perfect conductor (b) Superconductor
Fig. 5.6 (a) Magnetic field lines for a perfect conductor cooled in zero field, with application of a
magnetic induction Ba for T < Tc , and reduction of Ba to zero (left). Application of Ba above Tc ,
followed by cooling in the presence of Ba and suppression of Ba (right). (b) The same sequences for
a superconductor. In this case the field is excluded below Tc , whatever the magnetothermal history
of the sample. The state obtained at low temperature is reversible. Adapted from [18, pp. 18–20]
the metal was cooled below Tc in the presence or the absence of this magnetic field.
Note that λL is entirely determined by n, the number of conduction electrons per unit
volume in the metal, which is of the same order of magnitude as for conventional
metals.
For copper, n = 8.5 × 1028 electrons/m3 , and according to (5.6), this implies that
λL 200 Å. The surface currents preventing any change in the magnetic induction
in the metal only exist over a distance containing a few hundred atomic layers.
5.2.2 Meissner Effect: Exclusion of the Magnetic Field
It was not until 1933, some 22 years after the discovery of superconductivity, that
Meissner and Ochsenfeld were able to demonstrate experimentally that a supercon-
ductor is not the same thing as a perfect conductor. Indeed they observed that, not
only is the magnetic induction constant, but it is in fact zero inside a macroscopic
superconductor, and this whatever magnetothermal path is taken (see Fig. 5.6), i.e.,
B=0 .
This observation, since called the Meissner effect, confirms that superconductivity
is a novel magnetic effect of the electron gas, but it also shows that it is a fully
reversible phenomenon. The electron gas undergoes a transition at Tc to a new and
more stable thermodynamic state.
This observation leads to a series of considerations that we shall develop further

in Chap. 6, and which will allow us to characterise the behaviour of the electronic
system on the basis of experimental observations.
Note that the magnetisation in the bulk of a superconductor is given by
B = μ0 (H + M) ,
and then
M = −H . (5.8)
The magnetic susceptibility χ = χ̃ /μ0 of a superconductor is diamagnetic, with

χ̃ = −1. A superconductor is thus a perfect diamagnetic system. This is illustrated
in Fig. 5.7, where we have plotted the measured magnetisation of a superconductor
for different values of the applied magnetic field. One of the easiest manifestations
of this diamagnetism to demonstrate experimentally is the levitation of a magnet
above a superconductor, as illustrated on p. 146. The exclusion of the magnetic field
by the superconductor corresponds to a repulsive interaction between the magnet
and the superconductor, leading to an equilibrium position for the magnet at a height
h such that the force due to the magnetic interaction exactly balances the weight of
the magnet.
B M
Ha
µ0
=
al B
et 0
m M=0 Ha
al
m
or Superconductor
N
0 B=0 Hc Ha M = − Ha
Fig. 5.7 Dependence of B and M in a superconductor on the field Ha
5.2.3 London Equations and Penetration Depth
Given the state of knowledge in the 1930s, it was just as difficult to understand the
vanishing of the resistivity as the disappearance of the magnetic induction in the
bulk of the superconductor. However, it was observed that the vanishing of B in the
bulk of the material could be explained by setting the argument of the derivative
equal to zero in Eq. (5.7). The phenomenological relation
B − λ2L B = 0 (5.9)
was thus used by the London brothers from 1935 to account for the Meissner effect.
Note that (5.9) results from the need to account for observation, and cannot be justi-
fied on a priori grounds. However, it implies that it is B rather than dB/dt that goes to
zero exponentially over the distance λL from the surface of the superconductor. The
field is excluded only from the heart of the bulk superconductor. The existence of a
penetration depth λL for the magnetic induction was demonstrated experimentally
by measurements made on low dimensional samples, i.e., metal films and powders.
This shows that persistent currents arise in a thickness λL at the surface and exactly
compensate for the applied field in the bulk of the superconductor.
Question 5.1. Consider a lead cylinder of inner radius ri = 10 μm and outer radius ro =
20 μm. In zero applied field, lead becomes superconducting at Tc = 7.23 K. Its critical field
is Hc = 0.08 tesla. The density of superconducting electrons at zero temperature in lead
is ns = 2 × 1028 /m3 . This cylinder is cooled in a magnetic induction field Ba = μ0 Ha =
3.3 mtesla applied parallel to its axis, starting when T > Tc and continuing to a temperature
T Tc . The source of the magnetic field is maintained after cooling. What is the magnetic
field between 0 and ri ? What is the magnetic field inside the cylinder between ri and re ?
If there are currents in the cylinder, locate them and find their direction. How could the
magnitude of these currents be estimated?
In fact the two London equations which best describe the experimental properties
of superconductors in this phenomenological way are
dj
λ2 μ0 =E , (5.10)
dt
and
λ2 μ0 curl j = −B . (5.11)
The first is just (5.3), and accounts for the absence of resistivity. The second London
equation replaces (5.4) here, and implies (5.9), i.e., the Meissner effect, with the help
of the second Maxwell equation.
These equations are barely more justifiable3 than (5.9). They replace Ohm’s law
for the superconductor, and do not help to establish the value of λ as λL in (5.6).
The penetration depth is thus defined experimentally. Measurements show that λ
increases with temperature, and diverges at Tc , leading to a certain continuity with
the normal metal state, in which the field penetration is perfect.
3 It is worth noting how important empirical models can be for scientific progress. A full formula-
tion is often only possible much later, when many conceptual obstacles have been removed. And
even though today the BCS theory provides an explicit understanding of the observed relation-
ships, the simplified representation due to the London brothers remains an extremely useful guide
to understanding many experimental observations.
Fig. 5.8 Temperature depen- λ (μm)

dence of the penetration depth
measured in tin. Contradict-
ing the simple London model
0.2
of (5.6), it varies with the
temperature and diverges
for T = Tc , as shown by
(5.12). From Schawlow, A.L., 0.1
Devlin, G.E.: Phys. Rev. 113,
120 (1959). With the kind λ0
permission of the American
Physical Society. http://link.
aps.org/doi/10.1103/PhysRev. 0 1 2 3 4
113.120 Temperature (K) Tc
The measured temperature dependence is often represented in an empirical form

(see Fig. 5.8):
4 −1/2
T
λ = λ0 1 − . (5.12)
Tc
For most metals, the value of λ0 measured at low temperatures is of the same or-
der of magnitude as λL , but generally larger. To account for these observations,
one commonly used approach is to use the two-fluid model for the electrons in a
superconductor. The superconducting electrons, whose density ns varies with the
temperature, coexist with normal electrons of density n − ns . Only the supercon-
ducting electrons contribute to the persistent currents, so by analogy with (5.6), we
write m0
λ2 = . (5.13)
ns e2 μ0
Comparing (5.12), (5.13), and (5.6), we thus obtain
m0
λ20 = , ns (T) = ns (0) 1 − (T/Tc )4 . (5.14)

ns (0)e2 μ0
In this model, the density of superconducting electrons thus decreases as the tem-
perature increases, vanishing at T = Tc , when the normal state is reached.
Note that this two-fluid model can be used to predict the response of a super-
conductor to an electric field at frequency ω. The complex alternating conductivity
σs (ω) of the superconducting electrons adds to σn (ω) for the normal electrons, de-
termined in Question 4.1.
Question 5.2. 1. Determine the imaginary part of σ2s (ω) using the London equations.
2. What response would you expect for σ1s (ω) at zero frequency? [It can be found as the
limit of σ1 (ω) for a metal when τe → ∞.]
3. At what frequency ωc are the amplitudes of the two currents jn and js equal? Assuming
that ns (0) is the total electron number n per unit volume, determine ωc at T = 0.9Tc .
(Take τe ∼ 10−12 s.)
4. Can this approach be applied for T → 0?
5.3 A Macroscopic Quantum Effect
How do we explain the lack of electrical resistivity? Clearly, defects are not going to
disappear at Tc , and superconductivity is observed in particularly impure alloys, e.g.,
lead–indium alloys, where the residual resistivity is high in the normal state. The fact
that currents persist over macroscopic lengths suggests that the electron flow is not
perturbed over such lengths. There are almost as many obstacles as electrons, and
yet the electrons go through undisturbed, as if these obstacles were just not there!
This is as striking a paradox as the discovery that electrons orbit around atomic
nuclei without radiating electromagnetic energy.
The solution is clearly of the same kind as the one proposed by Bohr to resolve
the latter puzzle. The superconducting state of the electron gas can only be a sta-
tionary quantum state of the electronic system. The main difference with the Bohr
atom is that this quantum state involves a macroscopic number of electrons, so if it
is to be characterised by a wave function, this must necessarily involve a collective
state of these electrons.
5.3.1 Macroscopic Wave Function and Current
The wave function of this macroscopic quantum state must be a multielectron wave
function, and hence exceedingly complex. It is not useful a priori to try to describe
the probabilities of finding the various electrons, but the macroscopic wave function
can be reduced to one denoted by ψ(r) that delivers the spatial variation of the
probability density of the superconducting carriers,4 viz., |ψ(r)|2 . By doing this,
we preserve the key feature of a wave function, which is to include a phase factor,
with

ψ(r) = φ(r) exp iθ (r) . (5.15)
As a general rule, we shall be concerned with situations in which the superconduct-

ing carrier density φ(r)2 = φ 2 is spatially uniform in the bulk of the superconductor.
However, we shall see in Chap. 6 that we must expect a spatial dependence of this
density in the vicinity of the interface between a superconductor and any other ma-
terial.
In the presence of a magnetic field with vector potential A, the momentum oper-
ator p = −ih̄∇ for a particle of mass m and charge q is [2, Chap. 15]
p = mv + qA , (5.16)
4 Can the quantities φ(r) and θ(r) be defined at the same point of space? In the case of a one-particle
wave function, the state and energy are defined, but the phase cannot be simultaneously measured.
However, in a macroscopic system, the density and phase can be defined simultaneously, because
the relevant uncertainty relation is Δns Δθ > 1. Since ns is macroscopic, the two quantities can be
defined accurately enough, and can thus be treated semi-classically.
5.3 A Macroscopic Quantum Effect 159
where v is the velocity operator at point r. The electric current density at point r is
the product of the charge and the probability current, viz.,
q
j(r) = qRe(ψ ∗ vψ) = Re ψ ∗ − ih̄∇ − qA(r) ψ ,
m
h̄q ∗ q2 A 2
j = −i (ψ ∇ψ − ψ∇ψ ∗ ) − |ψ| .
2m m
With the expression for the wave function given above, this implies
qφ 2
j= (h̄∇θ − qA) . (5.17)
m
Note that, unlike θ and A, the current j is a measurable physical quantity. It is there-
fore independent of the choice of gauge for the vector potential, whence a change
of gauge for A must lead to a change in θ (r). In particular, if B = 0 in the bulk of a
superconductor, the current j will be associated with a gradient in the phase of the
wave function obtained in the gauge for which A = 0.
5.3.2 Quantisation of Magnetic Flux
The simple fact that a superconductor constitutes a macroscopic quantum state has
important implications, because the phase θ (r) of the wave function cannot vary in
an arbitrary way in space. In particular, continuity of the wave function implies that
the change in θ (r) along any closed path within a superconductor must be a multiple
of 2π :
%
∇θ (r)·d = 2nπ . (5.18)
(C)
Substituting the value of ∇θ (r) given by (5.17) into (5.18), we obtain

%
m
2
j + qA ·d = nh . (5.19)
(C) qφ
&
Since Φ = A·d is the magnetic flux through the closed path (C), we may define
a quantised quantity Φ ∗ with the dimensions of a flux, called the fluxoid:
%
∗ m
Φ =Φ+ 2 2 j·d = nh/q . (5.20)
q φ (C)
We thus consider a contour (C) inside a connected superconductor (see Fig. 5.9a).
If we choose a contour that is everywhere distant from the surface of the supercon-
ductor, i.e., at a distance much greater than λ, then B = 0 and j = 0 imply that n = 0,
and the phase of the wave function is uniform.
(C)
(C)
λ
λ λ
Ha
(a) (b)
Fig. 5.9 (a) Contour (C) inside a connected superconductor. (b) Contour (C) around a hole in a
non-simply-connected superconductor. The flux Φ of the magnetic induction in (C) reduces to the
flux of the applied field μ0 Ha through the hole, provided the latter has a diameter much greater
than λ
For a non-simply-connected superconductor, consider a contour (C) around a

hole (see Fig. 5.9b). If (C) is taken a long distance away from the edges of the hole
compared with λ, then j is zero and, according to (5.20), we obtain
Φ = nΦ0 , Φ0 = h/q . (5.21)
If we consider a ring superconductor with a hole of dimensions much greater than λ,

the flux Φ is practically equal to the flux of the external field through the hole. The
magnetic flux inside the superconducting ring is quantised, and the flux quantum is
Φ0 . Note that this quantisation of the flux is only obtained because of the Meissner
effect, and would not hold for a perfectly conducting metal or for a superconductor
with thickness less than λ. In the latter case, there is still a quantised quantity for
any integration contour, but it is the fluxoid Φ ∗ , which is not the magnetic flux and
is more difficult to measure.5
5.3.3 Measuring the Flux Quantum
The quantisation of the flux can be demonstrated experimentally by measuring the

flux in a superconducting tin cylinder as a function of the magnetic field in which
the tin tube has been cooled. As can be seen from Fig. 5.10, the trapped flux varies
in steps, and is a multiple of about 2 × 10−15 weber. This is two times smaller than
the value h/e = 4.14 × 10−15 weber expected for an electron.
5 In an almost perfect metal with very low resistivity, the mean free path can be long enough to
ensure that the electron wave functions are coherent on a macroscopic scale. Effects associated
with this quantum coherence can then be detected, appearing as periodic oscillations in certain
macroscopic quantities, such as the magnetic susceptibility or the resistance of the ring. These
observations bear no relation to a quantisation of the magnetic flux.
Fig. 5.10 Flux trapped in a 6

superconducting tin cylinder,
Trapped flux Φ (10 −15 Wb)

measured by determining the
magnetic field on the cylinder 4
axis. The value of the flux
trapped in the cylinder is 2
plotted as a function of the
magnetic field in which the
tin tube was cooled. From
0
Goodman, W.L. et al.: Phys.
Rev. B 4, 1530 (1971). With
the kind permission of the −2
American Physical Society.
−4 0 4 8 12 16 20 24
/PhysRevB.4.1530 Magnetic field (10 −7 tesla)
This means that the charge q of the superconducting carriers is not −e but −2e.
Hence, the electrons are paired in the superconducting state. The flux quantum is
therefore
h
Φ0 = = 2.07 × 10−15 weber . (5.22)
2e
In (5.16), (5.17), (5.18), (5.19), and (5.20), we must therefore take
q = −2e . (5.23)
It is natural to take the mass m of these electron pairs to be
m = 2m0 . (5.24)
Finally, comparing the curl of (5.17), viz.,
4e2 φ 2
curl j = − B,
m
with (5.11), viz.,
ns e2
curl j = − B,
m0
it follows that the modulus φ 2 of the wave function must be equal to half the density
of electron pairs
φ 2 = ns /2 . (5.25)
So by measuring the flux quantum, it is established that, in a superconductor,

the carriers are pairs of electrons. This is a common feature of all the electronic
superconductivity phenomena so far observed. We shall examine the origin of these
electron pairs in more detail in Chap. 7.
Question 5.3. Reconsider the lead superconducting ring of Question 5.1. Suppose it is
cooled below Tc in an applied field Ba = μ0 Ha = 3.3 μtesla. Answer the same questions
as in Question 5.1.
Let us note that with (5.13) for λ2 , and (5.23), (5.24), and (5.25), the relation (5.17)
between the current and phase of the superconducting wave function can be rewrit-
ten as6
Φ0
λ2 μ0 j = − ∇θ − A . (5.26)
2π
5.3.4 Josephson Effect
A further spectacular manifestation of the macroscopic quantum nature of the su-

perconducting state is revealed by the characteristics of the tunneling effect between
two superconductors. We have already seen in Chap. 3 how electrons can transfer by
quantum tunneling between two metals separated by a thin vacuum layer. An anal-
ogous situation can be produced by introducing a thin insulating barrier between
two superconductors, as shown in Fig. 5.11a. However, there is nevertheless an im-
portant difference, as we shall see, between a tunnel junction between two normal
metals and a tunnel junction between two superconductors. As discussed in Chap. 3,
a tunneling current will only be set up if there is a potential difference between two
metals. However, in the case of two superconductors, the quantum coherence effects
I−
SC1 SC1
Thin insulating ψ 1 = φ1ei θ 1
V+
layer
V−
ψ2 = φ2 eiθ 2
SC2 I+
SC2
Substrate (a) (b)
Fig. 5.11 (a) Junction between two superconducting metals (called a Josephson junction). The
superconductor and insulator films can be deposited by sputtering and masking. In some cases, the
surface of the superconducting metal SC2 can simply be oxidised to form the insulating barrier.
(b) Schematic representation of the junction in (a). There is a step in the phase of the wave function
between SC1 and SC2
6 Note that, for a connected sample, Eq. (5.26) resumes into

λ2 μ0 j = −A
and we directly obtain then an elegant representation of the two London equations, which are
recovered by differentiating with respect to time or by taking the curl.
are such that a current of paired electrons can pass through an insulating barrier
even if the potential difference is zero.
Below we shall explain how the flow of superconducting pairs is related to the
phase of the macroscopic wave function of the superconducting electrons. This is
called the Josephson effect, named after the British physicist who predicted it in
1962 (Nobel prize in 1973). This will only occur in general for very thin insulating
barriers, much thinner than those leading to a tunneling current between normal
metals.
A schematic view of the junction is given in Fig. 5.11b, where we consider two
superconductors (1) and (2), assumed identical to simplify. In this case, each super-
conductor is described by a macroscopic wave function
n 1/2 n 1/2
s1 s2
ψ1 = eiθ1 , ψ2 = eiθ2 , (5.27)
2 2
where ns1 and ns2 are the superelectron densities in the two superconductors.
If the insulating barrier is thin enough for the electron pairs to cross it by tun-
nelling, there will be a coupling between the two superconductors. This tunneling
effect can be described by an effective coupling K between the two superconductors,
whose value will depend on the physical characteristics of the barrier. In particular,
it will decrease exponentially with the thickness. Suppose also that a constant poten-
tial difference V is applied across the two superconductors. The Schrödinger equa-
tions for the two wave functions ψ1 and ψ2 characterising the two superconductors
can be written in the form
⎫
∂ψ1 ⎪
⎪
ih̄ = E0 ψ1 + Kψ2 , ⎬
∂t
(5.28)
∂ψ2 ⎪
ih̄ = (E0 − 2eV)ψ2 + Kψ1 , ⎪ ⎭
∂t
where E0 is the energy per electron pair of the ground state of each of the super-
conductors when there is no coupling and no electric potential V. The energy of
superconductor (2) is E0 − 2eV, since the effective charge of the superconducting
carriers is −2e. For K = 0, we do indeed obtain the equations for the two isolated
superconductors. The terms in Kψ2 and Kψ1 describe the coupling between the two
superconductors due to the fact that the wave function associated with one super-
conductor does not suddenly drop to zero in the other.
Using (5.27), we have

∂ψ1 1 dns1 dθ1
= +i ψ1 . (5.29)
∂t 2ns1 dt dt
Substituting (5.29) and the analogous relation for ∂ψ2 /∂t into (5.28), we obtain
1/2
dθ1 1 dns1 ns2
ih̄ i + − E0 = K ei(θ2 −θ1 ) ,
dt 2ns1 dt ns1
1/2 (5.30)
dθ2 1 dns2 ns1 i(θ1 −θ2 )
ih̄ i + − E0 + 2eV = K e .
dt 2ns2 dt ns2
The imaginary parts of these equations can be written

dns1 2K dns2
= (ns1 ns2 )1/2 sin(θ2 − θ1 ) = − . (5.31)
dt h̄ dt
The current density
dns2 dns1
J = −e −
dt dt
in the junction can thus be written
J = Jc sin(θ2 − θ1 ) , (5.32)
setting
4eK 1/2
Jc = ns1 ns2 .
h̄
The real parts of the equations in (5.30) are
1/2
dθ1 ns2
−h̄ − E0 = K cos(θ2 − θ1 ) ,
dt ns1
1/2 (5.33)
dθ2 ns1
−h̄ − E0 + 2eV = K cos(θ2 − θ1 ) .
dt ns2
If the two superconductors of the junction are identical ns1 = ns2 , the right-hand
sides of (5.33) are equal, and equating the left-hand sides, we obtain
d(θ2 − θ1 ) 2eV
= . (5.34)
dt h̄
The two equations (5.32) and (5.34) relating the current and the voltage across the
junction are the two Josephson equations.
a. Current–Phase Relation
The first relation (5.32), known as the current–phase relation, tells us that the cur-
rent crossing the insulating barrier is directly related to the phase difference θ2 − θ1
between the wave functions of the two superconductors. It shows that a current can
cross the barrier when there is no applied voltage.
For V = 0, according to the second Josephson relation (5.34), the phase differ-
ence and hence the current J are time independent. The existence of a finite current
through the junction can be understood by analogy with (5.17), which shows that
a gradient ∇θ of the phase of the superconducting wave function always implies a
current of electron pairs. Here the phase does not vary continuously, but suddenly
in going from superconductor (1) to superconductor (2). In a certain sense, (5.32) is
like a generalisation of (5.17) (with zero magnetic field here).
In practice, a current is injected through the junction by means of a current

source. In this case, the phase difference θ2 − θ1 between the superconductors ad-
justs in such a way as to satisfy (5.32), which is only possible for an injected current
less than the maximal value Jc , called the critical current of the junction. It is pro-
portional to the coupling coefficient K and depends sensitively on the characteristics
of the junction.
b. Voltage–Phase Relation
When V is nonzero, the second Josephson relation (5.34), called the voltage–phase
relation, shows that if a constant, finite voltage V is applied to the junction, it will
induce a linear time dependence of the phase difference, which leads, according to
(5.32), to an alternating current. So a constant voltage will generate an alternating
current with frequency νJ proportional to V :
2e
νJ = V . (5.35)
h̄
The full set of physical consequences that follow from the two relations (5.27) and
(5.28) are commonly referred to as the Josephson effect.7 This effect, which is a
direct consequence of the existence of a macroscopic wave function, has impor-
tant applications. It has provided a very precise way of measuring the ratio e/h.
Conversely, using (5.35), a precision oscillator can be made by applying a well cal-
ibrated voltage, or a voltage can be determined with unequalled accuracy by mea-
suring the frequency νJ . Indeed, the Josephson junction is used today to specify the
voltage standard.
5.3.5 SQUID
The quantisation of flux inside a superconducting loop, as discussed in Sect. 5.3.3,

is a novel manifestation of superconductivity that can be put to use. Indeed, if a su-
perconducting ring is subjected to a magnetic field B, a very accurate measurement
of B can be made by determining the quantum number corresponding to the number
of flux quanta inside the ring. Recall that, for a ring of diameter 10 μm, the flux Φ0
is just the flux of the Earth’s magnetic field, i.e., about 0.2 gauss or 2 × 10−5 tesla.
However, it is difficult to change this number of quanta directly by changing the
field applied to a simple superconducting loop, and nor is it easy to measure the
number of flux quanta.
On the other hand, if two Josephson junctions are inserted into the superconduct-
ing ring, as shown in Fig. 5.12, the flux in the loop is no longer quantised. Indeed,
due to the insulating barriers, no continuous contour can now be defined in the
superconductor.
7Some features can be understood by attempting Problem 11: Direct and Alternating Josephson
Effects in Zero Magnetic Field.
Fig. 5.12 Interferometer with

ψ 1 = φ 1 e iθ1 Ia ψ 2 = φ 2 e iθ2
two Josephson junctions, as
used in the DC-SQUID,
which runs on a direct cur-
rent I
I I
(1) (2)
Ib
The Josephson junctions force a current of superconducting pairs to move around

the loop in the absence of any applied voltage. By applying a magnetic field through
this closed circuit, we may act on the phase difference θ2 − θ1 between the two
superconducting circuits on either side of the junctions. In this way, we set up an
interferometry experiment between the currents in the two branches of the loop,
whose phase relation depends on the applied magnetic field B.
Question 5.4. Assuming that the two Josephson junctions are the same, show that the max-
imum total current Imax that can go around the loop varies sinusoidally with the applied
magnetic field, where
Φ
Imax = I0 cos π .
Φ0
This is what is observed experimentally, as shown in Fig. 5.13. Can you explain why the
oscillations disappear beyond a certain value of the applied field?
This kind of interferometer can be used to measure the magnetic flux Φ through the
loop with a much higher resolution than the flux quantum. As the latter is very small,
experimental devices developed on the basis of Josephson junctions, called SQUIDs
(superconducting quantum interference device) were used to make highly sensitive
magnetometers in the 1980s. These were commercialised in the 1990s, resulting in
the large scale dissemination of these magnetic measurement techniques. Several
applications will be discussed in Chap. 10.
I
Fig. 5.13 Current through an
interferometer containing
two Josephson junctions as
a function of the applied
field. From Jaklevic, R.C.,
Lambe, J., Mercereau, J.E.,
0
Silver, A.H.: Phys. Rev. A
140, 1628 (1965). With the
kind permission of the Amer-
ican Physical Society (APS).
http://link.asp.org/doi/10.1103 −40 −20 0 20 40
/PhysRev.140.A1628 Ha (μtesla)
5.4 Summary 167
5.4 Summary
Superconductivity is a state of the electron gas of a metal with zero resistivity, man-
ifesting itself through the presence of persistent currents. This state occurs below a
critical temperature Tc , but only for an applied magnetic field smaller than a certain
critical field Hc . The temperature dependence of the critical field is approximately
2
T
Hc (T/Tc ) = Hc (0) 1 − .
Tc
The fact that there is a critical field Hc also implies a bound on the current that can
pass through the superconductor (the critical current).
A superconductor is not simply a perfect conductor. The difference is manifested
via the Meissner effect, which shows that the magnetic induction is zero in the bulk
of a superconducting material, which is therefore a perfect diamagnetic material.
In a perfect metal, the magnetic induction is uniform, but it is determined by the
magnetothermal history of the material. In a superconductor, on the other hand, the
zero magnetic induction is reversible, and corresponds to a specific thermodynamic
state of the electron gas.
The phenomenological London equations explain the Meissner effect, and show
that the state of zero magnetic induction results, in the presence of an applied field
Ha < Hc , from the occurrence of persistent currents at the surface of the supercon-
ductor. The latter circulate in a surface layer of thickness λ, called the penetration
depth. This is given by
m0
λ2 = .
ns e2 μ0
In this expression, the experimental quantity ns is the number of superconducting
electrons per unit volume. It vanishes at the superconductor–metal transition.
The superconducting state is a quantum state of the electron gas whose coherence
is preserved at the macroscopic scale. The probability density of the superconduct-
ing carriers is given by a wave function
ψ(r) = φ(r) exp iθ (r) ,
where the phase factor θ (r) plays an important role, because it determines the per-
sistent currents through
qφ 2
j= (h̄∇θ − qA) .
m
The macroscopic quantum nature of this system manifests itself through the quan-
tisation of the magnetic flux through a macroscopic superconducting loop. The flux
through a superconducting loop is quantised with
Φ = nΦ0 ,
where the measured flux quantum is
Φ0 = h/2e .
This shows that the current carriers in a superconductor are pairs of electrons and
allows one to rewrite the relation between the current and phase of the wave function
as
Φ0
λ2 μ0 j = − ∇θ − A .
2π
Another manifestation of the macroscopic quantum nature of the wave function
in the superconducting state is the Josephson effect, observed when we consider
the tunneling current of electron pairs through a thin insulating layer between two
superconductors. The Josephson equations
d(θ2 − θ1 ) 2eV
J = Jc sin(θ2 − θ1 ) , = ,
dt h̄
relate this current and the voltage across the junction to the phase difference between
the wave functions of the two superconductors on either side of the junction.
Josephson junctions are used to make SQUIDs, quantum interferometers that
can measure the magnetic flux through a macroscopic loop to much higher accuracy
than the flux quantum.
Question 5.1
The applied field is less than the critical field Hc , so the cylinder is superconducting.
The magnetic flux through the tube is held at its initial value (see Fig. 5.14a). The
value of ns is used to determine the London penetration depth from
m0
λ2 = .
ns e2 μ0
This leads to λ = 376 Å. Since the thickness of the tube is re − ri = 10 μm λ, the
Meissner effect cancels the magnetic induction in the region ri < r < re , apart from
the regions of thickness λ next to the surface (see Fig. 5.14b).
Currents are localised in the ring-shaped regions of thickness λ at the surface
of the tube. The current at the outer surface creates a field within the tube which
opposes the external field. On the inner surface of the tube, the currents are opposite,
and tend to maintain the flux inside the cylinder. To carry out a detailed calculation,
we must solve B − λ2 B = 0 in cylindrical coordinates and then determine the
currents using μ0 j = curl B
μ 0 Ha μ0 Ha
μ 0 Ha
ri
r
re λ
μ 0 Ha μ 0 Ha
ri
re
B=0
(a) (b)
Fig. 5.14 (a) Lead cylinder of inner radius ri = 10 μm and outer radius ro = 20 μm. (b) Cross-
section showing magnetic induction and currents when the ring is cooled in an applied field of
3.3 mtesla
Question 5.2
1. The London equation

dj
λ2 μ 0 =E
dt
implies
E (ω)
j(ω) = ,
iωλ2 μ0
and then
i
σs (ω) = − ,
ωλ2 μ 0
so
1 ns (T)e2
σ2s (ω) = = .
λ2 ωμ0 mω
2. For a superconductor, σ1s (ω) is infinite at zero frequency. For a perfect conductor,
the conductivity σ1 (ω) can be obtained by letting τe tend to ∞. The function in
Fig. 4.16 thus has a width that tends to zero. This implies that σ1s (ω) is a delta
function δ(ω). Note that the area under σ1 (ω) is independent of τe :
∞ ∞
ne2 τe π ne2
σ1 (ω)dω = dω = ,
0 m 0 1 + ω τe
2 2 2 m
so that
π ns (T)e2
σ1s (ω) = δ(ω) .
2 m
Therefore,

ns (T)e2 π i ne (T)e2 τe
σs (ω) = δ(ω) − , σn (ω) = (1 − iωτe ) .
m 2 ω m 1 + ω2 τe2
3. We have
ne e2 τe
|jn | = |E | ,
m 1 + ω2 τe2
and, for ω > 0,

ns e2
|js | = |E | .
mω
Hence,

jn ne
= τe ω .
j n
s s 1 + ω2 τe2
Letting X = ns (T)/ne (T), we will have |jn /js | = 1 for ωc2 τe2 = X 2 /(1 − X 2 ). From
the T dependence of the penetration depth λ given in the two fluid model by
(5.14), we obtain an estimate
X = ns /ne = 1 − (T/Tc )4 .
For a temperature T = 0.9Tc , this leads to ωc τe ∼ 0.36, which corresponds to

a frequency νc ∼ 0.6 1011 Hz for τe = 10−12 s. One really needs to reach such
high frequencies even close to Tc to feel the dissipation of the ‘normal’ electron
current.
4. For T → 0, there will be no dissipation whatever the frequency, according to
this simple two-fluid model. This is hardly possible, because the electronic sys-
tem has excited states. The two-fluid model does not take these excitations into
account. The BCS calculation predicts that there will be a gap in the excitations
and absorption will occur for h̄ω > 2Δ.
Question 5.3
The flux Ba = μ0 Ha = 3.3 μtesla inside the cylinder is
Φ = Ba (π ri2 ) = 1.03 × 10−15 weber/m2 .
This is close to half the flux quantum Φ0 . Consequently, since the flux inside the
tube is quantised, we will only be able to trap a flux Φ = 0 or Φ = Φ0 in the tube.
B B
6 6 μtesla
3 3 μtesla
B=0
(a) (b)
Fig. 5.15 Superconducting ring cooled with an applied axial field of 3.3 μtesla. (a) B ≡ 0 for
0 < r < re . (b) B = 6.6 μtesla for 0 < r < ri
The field between 0 and ri will thus be either 0 or 6.6 μtesla, with a probability close
to 1/2 depending on the experiment (see Fig. 5.15).
Inside the cylinder walls, the field obviously remains zero, except in the surface
regions of thickness λ, where it must match the value outside the superconducting
walls. The current at the outer surface produces a field inside the tube which always
opposes the external field. For the case shown in Fig. 5.15a, where B ≡ 0 for 0 < r <
re , there is no current on the inner face of the tube. For the case shown in Fig. 5.15b,
where B = 6.6 μtesla for 0 < r < ri , the current flows in the same direction as in
Fig. 5.14b.
Note that the current on the outer face of the tube is 1,000 times smaller than for
the case illustrated in Fig. 5.14b. However, the current on the inner face of the tube
for the case shown in Fig. 5.15b is 500 times smaller than for the case illustrated in
Fig. 5.14b.
Question 5.4
If θ1 , θ2 , and θ1 , θ2 are the phases of the wave functions of superconductors 1 and
2 on either side of the two junctions, as shown in Fig. 5.16, while I0 is the critical
current of the two junctions, assumed identical, we have
I = I0 sin(θ2 − θ1 ) + I0 sin(θ2 − θ1 ) . (5.36)
In the absence of an applied field, the phase of the superconducting wave functions
is uniform on both superconducting sections (1) and (2), so that θ1 = θ1 and θ2 = θ2
obviously yield I = 2I0 sin(θ2 − θ1 ).
For Ha = 0, the relation (5.26) between current and phase applies and can be
integrated within the superconductor along the contour C shown in Fig. 5.16 to
yield
Fig. 5.16 Interferometer with θ1 θ2

two Josephson junctions in a
DC-SQUID, which operates
with a direct current I. Here
the sides of the upper junc- 1 2
tion are labelled 1 and 2, and I I
those of the lower junction 1
and 2 . The contour C does not (1) (2)
include the junctions 1 2
C
θ1 θ2
% % 1 % %
2 Φ0
μ 0 λ2 j·dl + A·dl + A·dl = − ∇θ ·dl (5.37)
C 1 2 2π C
Φ0
= (θ1 − θ1 + θ2 − θ2 ) (+nΦ0 )
2π
The first integral vanishes because j is zero in the bulk of the superconductor. The
other two integrals on the left-hand side represent the flux Φ through the loop if we
neglect the circulation of A within the insulating barriers. We thus have
Φ0
Φ= (θ1 − θ1 + θ2 − θ2 ) (+nΦ0 ) , (5.38)
2π
or
Φ
θ2 − θ1 = θ2 − θ1 + 2π (+2π n) ,
Φ0
and, setting δ = θ2 − θ1 , (5.36) gives

Φ
I = I0 sin δ + I0 sin δ + 2π ,
Φ0

Φ Φ
I = 2I0 cos π sin δ + π . (5.39)
Φ0 Φ0
The total critical current Imax thus varies periodically with the flux in the loop:

Φ
Imax = 2I0 cos π .
Φ0
This current clearly results from interference between the currents through the two
junctions, which is why this DC-SQUID is called a two-junction interferometer.
Such a setup allows a precise determination of the number of flux quanta through
the loop.
The experimental results shown in Fig. 5.13 do indeed reveal such a periodicity.
Note, however, that the current in the interferometer is not strictly periodic. This
happens because we have considered ideal junctions here, with a spatially uniform
phase difference on the junction. This would only be realistic for a junction with
a very small cross-sectional area, or for a negligible flux of Ha through the junc-
tion. The geometry of the junctions is responsible for the decrease in Imax when Φ
increases.
For a single junction of finite cross-section, the non-uniformity of the phase dif-
ference across the surface of the junction produces interference between the currents
at the center and at the edges of the junction.8 The critical current then depends
sinusoidally on the flux in the junction.
Note: On the left-hand side of (5.37), we neglected the integrals over the part of the
circuit through the two junctions. In fact the expression (5.26) used for the current
versus the phase difference is not strictly accurate in the presence of a magnetic
field, as it is not gauge invariant. Strictly speaking, we must replace θ2 − θ1 = δ in
(5.36) by
%
2π 2
δ = θ2 − θ1 − A·dl ,
Φ0 1
to obtain a gauge invariant expression for the current.8
It can be checked that, on the right-hand side of (5.37), the sum of the phases
transforms to
% %
2π 2 2π 1
δ − A·dl − δ − A·dl (+2π n) .
Φ0 1 Φ0 2
We then recover precisely those terms neglected in the integral over the interfero-
meter loop, so that
Φ
δ = δ + 2π ,
Φ0
which is a rigorous proof of the result in (5.38).
8 See Problem 12: Josephson Junction in a Magnetic Field.

Chapter 6
Thermodynamics of Superconductors
Contents
6.1 Thermodynamics of Bulk Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

6.1.1 Free Energy of the Superconducting State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.1.2 Entropy and Specific Heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.2 Thin Films and Coherence Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.2.1 Thermodynamics of Thin Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.2.2 Coherence Length ξ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.2.3 Mixed State: A Simple Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.3 Two Types of Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6.3.1 Type I Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.3.2 Simple London Model for a Vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.3.3 Type II Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.3.4 Applications of Type II Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
176 6 Thermodynamics of Superconductors
Abrikosov vortex lattice. The Meissner effect is only perfect in some superconductors. In the so-
called type II superconductors, with a high applied field, the thermodynamic equilibrium state
is one in which normal metallic regions appear within the superconducting material. Tubes of
normal material are arranged in a hexagonal lattice, called an Abrikosov lattice, named after its
discoverer (Nobel prize in 2003). The exceptional image above was obtained by scanning tunneling
microscopy (STM). Image courtesy of Davis, J.C. and coworkers, Cornell University, Ithaca, NY.
Scale 3,000 Å×3,000 Å
6.1 Thermodynamics of Bulk Superconductors 177
While the historical discovery of superconductivity sprang from the observation

of a zero resistance state, we have seen that the most novel manifestation of this
phenomenon is the Meissner effect. The existence of a diamagnetic transition is
today the main experimental signature sought to establish the presence of super-
conductivity in new materials. In 1986, it was indeed the observation by Bednorz
and Müller of the Meissner effect in La2−x Bax CuO4 , after the original observa-
tion of a drop in the resistance at temperatures below 35 K, which was taken as
incontrovertible proof of the superconductivity of this compound. This observa-
tion, rewarded by the Nobel prize in 1988, initiated the study of a whole range
of similar copper oxides which make up the family of high-Tc cuprate superconduc-
tors.
The fact that a superconductor excludes a magnetic field has many physical con-
sequences. This fact alone implies the existence of persistent currents, and leads to
the understanding that superconductivity is essentially a magnetic phenomenon, at
thermodynamic equilibrium, due to the condensation of the electron gas in a novel
macroscopic quantum state at low temperatures. This state must therefore have a
lower free energy than the electron gas in a normal metal.
In Sect. 6.1, we discuss the first thermodynamic consequences of this observa-
tion, and in particular the existence of a critical field for superconductivity. The
experimental investigation of the free energy in the presence of an applied field will
help us to identify certain physical quantities which characterise the superconduct-
ing state. In Sect. 6.2, we show how the thermodynamic properties of thin supercon-
ducting films lead to the notion of a coherence length for superconducting pairs and
the so-called mixed state. A brief description of type I and type II superconductors
and vortices is undertaken in Sect. 6.3.
6.1 Thermodynamics of Bulk Superconductors
In the presence of a magnetic field, the free energy of a material is lowered if it is

paramagnetic and raised if it is diamagnetic: the exclusion of a magnetic field by
a superconductor corresponds to an increase in its free energy. It follows immedi-
ately that the Meissner effect cannot hold out indefinitely if the field is increased
because, for a given value of the field, the increase in free energy due to exclusion
of the field will compensate for the drop in free energy associated with condensa-
tion of the electron gas in the superconducting state. The free energy of the normal
state, which hardly depends on the magnetic field, thus becomes lower for a strong
enough applied field. The very existence of the Meissner effect therefore implies
that the electron gas must return to its normal state above some thermodynamically
critical value Hc . We thus see that, by investigating the temperature dependence of
Hc , we will obtain valuable insight into the condensation energy in the supercon-
ducting state.
6.1.1 Free Energy of the Superconducting State
In a magnetic system, the thermodynamic potential can be taken as the free enthalpy,
or Gibbs free energy [1]:
G = U − TS + PV − μ0 Ha · M , (6.1)
where U is the internal energy. The intensive thermodynamic quantities T, −P, and
μ0 Ha are the conjugate variables with respect to G of the extensive variables S, V,
and M, respectively. So for constant T and P, we have
dG = −μ0 M · dHa , (6.2)
in such a way that the dependence of the free energy Gs (T, Ha ) of the superconductor
on the applied field is given by the area under the magnetisation curve (see Fig. 6.1):
Ha Ha
Gs (T, Ha ) − Gs (T, 0) = −μ0 M · dHa = μ0 Ha dHa , (6.3)
0 0
μ0 2
Gs (T, Ha ) − Gs (T, 0) = H . (6.4)
2 a
Likewise, the change in free energy related to the Pauli paramagnetism of the elec-
trons in a metal (M = χ̃P Ha ) is given by Gn (T, Ha ) − Gn (T, 0) = −μ0 χ̃P Ha2 /2. When
the applied field is equal to the critical thermodynamic field Hc , the normal and su-
perconducting phases have the same free energy, viz.,
μ0 2 μ0
Gs (T, Hc ) = Gs (T, 0) + Hc = Gn (T, 0) − χ̃P Hc2 = Gn (T, Hc ) .
2 2
The difference in free energy between the superconducting and normal states is
therefore
μ0
Gs (T, 0) − Gn (T, 0) = − (1 + χ̃P )Hc2 .
2
M Hc G
Ha Gn
Free energy density
0
M=0
μ0 2
H
(a) 2 c (b)
M = −Ha Gs
0 Hc Ha
Fig. 6.1 (a) Magnetisation of a superconductor as a function of Ha , exhibiting the Meissner effect.
(b) Free energy density of a superconductor and the normal metal as a function of Ha
6.1 Thermodynamics of Bulk Superconductors 179
Since χ̃P 1 for the normal metal, the energy of condensation of the electron gas
in the superconducting state is given to a good approximation by
μ0 2
Gs (T, 0) − Gn (T, 0) = − H . (6.5)
2 c
Hence if we neglect the change in the free energy of the normal state with field, we
can express the free energy of the superconducting state in the form
μ0 2
Gs (T, Ha ) − Gn (T, 0) = − Hc − Ha2 . (6.6)
2
This quadratic dependence on the field is shown in Fig. 6.1.
6.1.2 Entropy and Specific Heat
Given the free energy and the temperature dependence of Hc , we may immediately
deduce the following expression for the entropy:
dGs dHc
Ss = − = Sn + μ0 Hc , (6.7)
dT dT
which implies that the entropy of the superconducting state is less than that of the
normal state, because Hc decreases with T. We do therefore find that the supercon-
ducting state is more ordered than the normal state. This entropy difference suggests
that there is a latent heat L = T(Sn − Ss ) at the normal/superconducting transition in
the presence of a magnetic field. This is a first order thermodynamic transition. On
the other hand, experiment shows that Hc goes to zero at Tc with a finite slope for
Hc (T) (see Fig. 5.3), and according to (6.7), this means that Ss = Sn at Tc . There is
no latent heat, and the transition is second order in zero field. However, there is a
discontinuity in dS/dT and hence also in the specific heat Cs = TdSs /dT :

Cs − Cn d(Hc dHc /dT) dHc 2 d 2 Hc
= μ0 = μ0 + Hc . (6.8)
T dT dT dT 2
For T = Tc , the critical field Hc vanishes, and it follows that the discontinuity in Cn
at Tc is given by

dHc 2
Cs (Tc ) − Cn (Tc ) = μ0 Tc .
dT T=Tc
It is worth pausing for a moment to consider the experimental measurements of

the specific heat, since they can be achieved to much higher accuracy than mea-
surements of Hc (T), and according to (6.8), provide a handle on the temperature
derivative of this quantity. Suppose in particular that we accept the quadratic form
of Hc (T) given by (5.2), viz.,
2
T
Hc (T) = Hc (0) 1 − . (6.9)
Tc
By (6.7), it is easy to show that

2 2
Hc (0) T
Ss − Sn = −2μ0 1− T,
Tc Tc
and hence,
2
C s − Cn Hc (0) 2 T
= −2μ0 1−3 .
T Tc Tc
Experiment shows that Cs /T vanishes when T → 0, which implies a relation be-
tween the characteristic quantities of the normal and superconducting states1 :

Cn Hc (0) 2
= 2μ0 =γ .
T Tc
The quadratic form (6.9) for Hc (T) can thus be used to deduce the following relation
between the specific heats of superconductor and normal metal:
2
Cs T
= 3γ . (6.10)
T Tc
The discontinuity in the specific heat at the transition would therefore be given by
(Cs − Cn )/Tc = 2γ .
The expected changes in S/γ Tc and Cs /γ Tc are shown in Fig. 6.2. Experiment
shows that these results are not quite right. For example, in Fig. 6.3, it is observed
that the discontinuity in C/T at the transition is slightly less than 2γ . Moreover,
the linear dependence on (T/Tc )2 is not corroborated. This suggests that the form
(6.9) for Hc (T) is not such a good approximation. Paradoxically, the most signifi-
cant physical difference regarding this discrepancy does not result from the observed
difference at Tc , but rather from the low temperature behaviour, which is exponen-
tial. We shall understand the full importance of this experimental result in Chap. 7.
Indeed, it led to the suggestion that there might be a band gap in the spectrum of
energy states of a superconductor.
Question 6.1. Explain the temperature dependence of the specific heat observed in the nor-
mal state.
1 As discussed in Chap. 3, in a normal metal, the entropy Sn and specific heat Cn due to the
electrons are given by Sn = γ T = Cn , where γ ∝ D (EF ).
6.2 Thin Films and Coherence Length 181
S
γ Tc Sn 3
1 ΔCn
2
C
γT
Ss 1 Cn
0
0 T / Tc 1 (T/Tc)2 1
Fig. 6.2 Entropy and specific heat as a function T/Tc for the normal metal and superconducting
state, obtained using (6.9) for Hc (T)
Fig. 6.3 Measurements of 1.5

C/T as a function of T 2 , Gallium
carried out on metallic gal- B = 0.02T
lium in the superconducting tor
B=0 duc Tc
state and in the normal state on
C/T (mJ mole−1 deg−2)
e rc
restored by applying a field 1.0
Sup
Ha > Hc . The dependence
on T 2 expected on the basis
Normal
of the approximate quadratic
form for Hc (T) is not corrob-
0.5
orated. From [7, p. 365], 8th
edn., Wiley c 2005
0
0 0.5 1.0 1.5
T 2 (K2)
6.2 Thin Films and Coherence Length
So far we have only discussed bulk superconductors, and we have focused on the
characteristic behaviour in the bulk of such materials, without worrying too much
about the surface region in which the persistent currents are to be found. But the
experimental understanding of superconductivity suggests that we should look more
closely at thin samples, in particular to investigate the penetration depth and its
temperature dependence. To get some idea of the kind of calculations involved, it
will be useful to consider thin films.
6.2.1 Thermodynamics of Thin Films
Consider for example a film of thickness 2a perpendicular to the x axis, sub-

jected to an applied field Ha in the z direction. By symmetry, the field within the
superconductor only has a component in the z direction, and the London equations
imply λ2 d2 B/dx2 = B(x), according to (5.8).
The general solution of this differential equation is
x x
B(x) = B1 exp − + B2 exp .
λ λ
With the boundary conditions B = μ0 Ha at +a and −a, we obtain
cosh(x/λ)
B(x) = μ0 Ha , (6.11)
cosh(a/λ)
and if a λ, this does indeed correspond to a profile in which the field decreases
exponentially over a length λ on the two faces of the film and vanishes at the center
(see Fig. 6.4).
Note that, in a film, the surface currents can be determined from μ0 j = curl B,
whence they must be oriented in the y direction, with
dB μ0 Ha sinh(x/λ)
μ0 jy = − =− . (6.12)
dx λ cosh(a/λ)
This profile is shown in Fig. 6.5.
In order to measure the penetration depth λ, it is useful to reduce a/λ as far as
possible. The important point here is that, if a ≈ λ, the diamagnetism is no longer
perfect, as it would have been in a bulk superconductor. Naturally, it would be nice
to be able to measure the spatial variation of B across the film directly. But this is
no easy matter, even with the techniques and instruments of modern physics.
As we shall see in Chap. 10, one quantity that can be simply measured is the
magnetisation of the film, which is the spatial average of (1/μ0 )B(x) − Ha , i.e.,
a
B(x) dx
M= − Ha ,
−a μ 0 2a
μ 0 Ha μ 0 Ha
−a 0 a −a 0 a −a a
x x
(a) (b) (c)
Fig. 6.4 Penetration of a field (a) in a thick layer a > λ, (b) in a thin film a ∼ λ, (c) and in an
ultrathin film a < λ
Fig. 6.5 Magnetic induction

and surface current profiles in B μ 0 Ha
the y direction for a thin film
jy
a
−a x
Ha a
− tanh
λ λ
and using (6.11), it is easy to check that

λ a
M = Ha tanh − 1 . (6.13)
a λ
If a λ, we do indeed recover M = −Ha . By measuring M as a function of the

film thickness a, we can thus determine λ. Knowing the material making up the
film, the free energy of condensation Gs (T, 0) in the superconducting state is the
same as the one measured for bulk samples of this metal, expressed by (6.5) as
a function of the critical field Hc . On the other hand, if a ∼ λ, the free energy
Gsf (T, Ha ) of the film increases less than it would in a bulk sample in the presence
of a field Ha , since the field is barely excluded now. As in (6.3), we clearly have
Ha
μ0 Ha2 λ a
Gsf (T, Ha ) − Gs (T, 0) = −μ0 M · dHa = 1 − tanh .
0 2 a λ
Hence, substituting in the expression (6.5) for Gs (T, 0),

μ0 Ha2 λ a μ0 Hc2
Gsf (T, Ha ) − Gn (T, 0) = 1 − tanh − . (6.14)
2 a λ 2
The free energy of the superconducting film will only equal that of the normal state
when the right-hand side of this equation is zero, i.e., when Hc satisfies

2 λ a
Hc 1 − tanh = Hc2 . (6.15)
a λ
Since Hc > Hc , this shows that, in a thin film, the superconducting state can be
maintained for larger fields than Hc . In the limit a λ, a truncated expansion of
(6.15) leads to √
λ 3
Hc = Hc .
a
The critical field Hc of the film can be made significantly larger than Hc if a λ,
as has been confirmed by measurements on thin films of cadmium or tin. This result
shows that the original limitations on the use of superconductivity due to the low
values of Hc can be overcome by using films or wires of thickness less than λ.
The thermodynamic properties of a thin superconducting tube are also altered in the
presence of a quantised magnetic flux.2
6.2.2 Coherence Length ξ
We shall see that the effect described above can be used to raise a very basic ques-
tion. We have seen that the free energy required to exclude the magnetic field can
be decreased by reducing the dimensions of the superconductor. As a consequence,
since the superconducting state is a thermodynamic equilibrium state, why should a
sample not divide itself up spontaneously into normal and superconducting regions
when a field is applied? By accepting to lose condensation energy in the normal
regions, this would allow the material to considerably reduce the increase in free
energy due to exclusion of the field in the superconducting regions. For example, for
a superconductor in the form of a layer of thickness 2d λ, it is easy to imagine a
structure built up from superconducting layers of thickness 2a ∼ λ alternating with
normal layers of thickness 2b such that, in an applied field, the free energy is less
than that of the bulk sample (see Fig. 6.6). The point is just to make b extremely
small in order to gain some free energy!
Experimentally, many situations have been observed where the Meissner effect
is not perfect, and much effort has been expended in the search for explanations,
although other possibilities related in particular to the inhomogeneity and shape
of the samples were initially favoured (see Chap. 10). However, some bulk super-
conducting metals exhibit a perfect Meissner effect which disappears suddenly for
H = Hc . In these cases, no lamellar structure was observed, suggesting that we may
have made some mistake in the above analysis, and that some crucial physical ele-
ment has escaped our notice. Since thin films of the same metals exhibit supercon-
ductivity above Hc , as mentioned above, there must be an energy cost that opposes
the formation of interfaces between superconductor and normal metal.
In the above argument, we assumed that there could be a sharp interface between
normal and superconducting states. The fact that we are considering a crystal with
2d
N N N N N N
S S S S S S S
Fig. 6.6 Structure built up
from alternating thin films of Ha
superconductor and normal
metal, with thicknesses 2a
and 2b, respectively, which
can in principle reduce the
free energy in an applied
magnetic field 2a 2b
2 See Problem 10: Thermodynamics of a Thin Superconducting Cylinder.

a lattice spacing of a few angstrom units will certainly impose a physical limit on
the thickness of such an interface. But this would not really be a sufficient limitation
because the penetration depth is macroscopic, and this would still leave room for
macroscopic thicknesses b of normal regions. Indeed, we can today make artificial
structures that retain the properties of a metal while comprising only two, or even
just one atomic monolayer!
However, it seems reasonable to ask whether electrons can transfer easily within
the same crystal from a collective quantum state to an independent electron state
over a very short distance. In other words, over what distance can the density of
superconducting electrons go from ns to zero? This suggests reexamining the very
hypotheses that lead to the London equation close to the surface of a superconductor.
Indeed, it was assumed there that ns would remain constant right up to the surface.
But the notion of electron pairs suggests that this could not be possible. An electron
close to a surface has a lower probability of finding a partner than one in the bulk of
the superconductor.
The superconducting state, a coherent macroscopic quantum state, can only be
set up over a spatial distance at least as long as the average distance between paired
electrons. We thus allow ns to vary from ns to zero over a finite distance ξ , called
the coherence length of the superconducting state (see Fig. 6.7). This means that the
interface has a minimal thickness and an energy cost, which explains the existence
of the Meissner effect.
Indeed, close to an interface, the free energy per unit area increases by approx-
imately ∼ ξ μ0 Hc2 /2 compared with the bulk superconductor, because the density
of superconducting electrons is lower. However, it is reduced by the penetration of
the magnetic field by an amount of the order of ∼ λμ0 Ha2 /2. While the first term
dominates, there is no energy advantage in forming a lamellar state. However, if the
field is increased, such a structure might appear before Ha reaches Hc . We shall now
make this explicit by developing the above model.
Fig. 6.7 Penetration depth λ Normal Superconductor

and coherence length ξ at a
metal–superconductor inter- B
face ns
ξ
6.2.3 Mixed State: A Simple Model
We can now investigate whether the structure specified above can have a lower
free energy than the bulk superconducting phase. Insofar as the normal metal–
superconductor interface requires a minimal thickness of the order of ξ , there is no
longer any real problem in imagining that the thickness b of the normal layers might
be infinitesimal (the existence of ξ is tantamount to imposing a minimal thickness of

ξ for b). The structure thus looks like a series of thin films, with period 2a. The free
energy of such a film is then given by (6.14) increased by the loss of condensation
energy in the interface of thickness ξ that is (ξ/a)μ0 Hc2 /2.
The reduction in free energy Gms (T, Ha ) per unit volume of this mixed state com-
pared with the normal state is therefore given by

μ0 Ha2 λ a μ0 Hc2 ξ
Gms (T, Ha ) − Gn (T, 0) = 1 − tanh − 1− , (6.16)
2 a λ 2 a
and this should be compared with the value obtained for a perfect Meissner effect,
as given in (6.6):
μ0 2
GM = Gs (T, Ha ) − Gn (T, 0) = − (H − Ha2 ) , (6.17)
2 c
μ0 Ha2 λ a μ H2 ξ
0 c
Gms (T, Ha ) − Gn (T, 0) = − tanh + + GM . (6.18)
2 a λ 2 a
It follows that the lamellar structure is only thermodynamically favoured over a
perfect Meissner effect if
tanh α ξ
−Ha2 + Hc2 <0,
α λα
where we have set α = a/λ. This can also be written
2
Ha λ
tanh α > 1 .
Hc ξ
Since tanh α < 1, this will be impossible when λ/ξ < 1, for any Ha ≤ Hc . When
λ < ξ , the Meissner effect therefore persists until the thermodynamic critical field is
reached.
On the other hand, a lamellar solution will be possible if λ > ξ . The width of
the layers giving the lowest free energy is obtained by minimising Gms with respect
to a, i.e., solving dGms /da = 0. Cancelling a factor of μHa2 λ/2 in (6.18), we obtain
2 a

a
−1
1 ξ Hc
− 2 − tanh − aλ cosh2 =0,
a λ Ha λ λ
which simplifies into

2
α ξ Hc
tanh α − = . (6.19)
cosh α
2 λ Ha
The function on the left-hand side is a monotonic increasing function of α which

tends asymptotically to 1. There is therefore no solution unless the right-hand side
is less than 1, i.e., for Ha < Hc , there will only be a solution if ξ < λ, agreeing with
2a ∼ ξ λ
B B
μ0 Ha
2a
μ0 Ha
x Ha Hc2 x
2λ Hc1 Ha
(a) (b)
Fig. 6.8 Change in the magnetic induction in lamellar structures obtained for different values of
the field Ha . The normal region is assumed to have infinitesimal thickness. (a) For Ha Hc1 , the
distance between layers is much greater than λ. The change in the magnetic induction in each layer
is given by (6.11) and shown in Fig. 6.4. (b) For Ha Hc2 , the layer thickness tends to ξ . For the
case depicted here, λ ξ and the field penetrates almost uniformly
what we have already seen above. Even in this case, a solution will only exist for
fields above a critical field Hc1 defined by
1/2
ξ
H a > Hc = Hc1 . (6.20)
λ
We may thus conclude that there is a total Meissner effect for 0 < H < Hc1 , and
that beyond Hc1 , a lamellar structure can appear. For Ha = Hc1 , α is infinite. On
the other hand, as soon as Ha > Hc1 , α drops off and interfaces form with very long
period to begin with. As H increases, the density of interfaces steadily increases (see
Fig. 6.8).
This structure will clearly remain in much larger fields than Hc , until the free
energy in (6.18) becomes equal to that of the normal state. This happens for a second
critical value Hc2 of the field such that
a μ H 2
μ0 Hc22 λ 0 c ξ
Gms (T, Hc2 ) − Gn (T, 0) = 1 − tanh − 1− = 0 , (6.21)
2 a λ 2 a
2
H c2 tanh α ξ
1− − 1− =0. (6.22)
Hc α λα
The period of the layers is given at Hc2 by (6.19), viz.,

2
α ξ Hc
tanh α − = . (6.23)
cosh α
2 λ Hc2
Equations (6.22) and (6.23) thus imply the simple relation

Hc
tanh α = .
Hc2
In the limit λ ξ , a truncated expansion of (6.23) then leads to

Hc 3ξ
= ,
Hc2 2λ
which corresponds to Hc2 Hc .
The main result here is that, when ξ < λ, a state in which superconducting re-
gions alternate with normal regions is energetically favoured over the perfect Meiss-
ner state for a range of magnetic field strengths between Hc1 and Hc2 .
Note that, for λ ξ , the period of the layers is of the same order of magnitude
as ξ when Ha comes close to Hc2 . Under such conditions the magnetic field pene-
trates almost everywhere in the superconductor, while the change in B can become
relatively small in the lamellar structure (see Fig. 6.8b).
The calculation exposed above brings out the main physical phenomena, but it
artificially favours a lamellar structure for the field penetration. There is no justifi-
cation for this choice, which is based on the form chosen for the sample, and takes
into account its surface geometry. One would be unlikely to choose the same form
for the interfaces in the case of a cylindrical sample! The thermodynamic argument
does indeed suggest that the state obtained has a lower free energy than the one cor-
responding to the Meissner effect, but it in no way guarantees that this will be the
state of minimal energy. If the sample is not assumed to have any particular geom-
etry, there is only one privileged direction, that of the applied field, and we should
expect rather to observe a penetration of the field in the form of ‘field tubes’. This is
what happens in reality, and we shall describe it in more detail in the next chapter.
6.3 Two Types of Superconductivity
The above analysis of experimental observations showed that we require another

characteristic length, the coherence length ξ , in order to describe the behaviour of
superconductors in an applied magnetic field. As indicated at the end of Sect. 6.2.2,
the formation of an interface between the superconducting state and the normal state
has an energy cost of the order of
μ0
(ξ Hc2 − λHa2 )
2
per unit interfacial area. It is this interface energy that determines the resulting state
of thermodynamic equilibrium. Two types of behaviour are then possible depending
on the value of the ratio λ/ξ , which is generally denoted by κ. For κ = λ/ξ 1,
a Meissner effect is observed up to Hc , and the normal state is restored above Hc .
We then have a type I superconductor. On the other hand, for ξ λ, the Meissner
effect is only observed up to Hc1 . Above Hc1 , a mixed state is established, and this
only disappears above a second critical field value Hc2 . We then speak of a type II
superconductor.
6.3 Two Types of Superconductivity 189
6.3.1 Type I Superconductors
Let us note to begin with that, for a type I superconductor, the interface energy
between a superconductor and a normal metal is always positive. This situation
forbids the creation of interfaces within the superconductor and leads to a perfect
Meissner effect. In this case, the density of superconducting pairs is low close to
the surface of the superconductor, in a region of thickness equal to the penetration
depth of the magnetic field (see Fig. 6.9).
This does not correspond to the London limit considered in Chap. 5, for which
the density of superconducting carriers was assumed constant right up to the surface.
The phenomenological London model does not therefore apply under the conditions
for which it was designed. This explains why the value λ0 of the penetration depth at
zero temperature is not λL . The fact that the density ns of superconducting carriers is
reduced over a much greater distance than λ clearly suggests that λ0 > λL . This was
first understood by Pippard on the basis of ultrahigh frequency absorption experi-
ments, which led him to introduce the notion of coherence length. As we shall see
later, the two lengths λ and ξ depend on the properties of both the superconducting
state and the normal state of the metal.
λ λ
B ns B ns
ξ k l
ξ
κ 1 κ 1
Fig. 6.9 Lengths λ and ξ at the interface between superconductor and normal metal, for type I
(left) and type II (right) superconductors
6.3.2 Simple London Model for a Vortex
As in the case of the layers discussed above, a simple model for describing a vortex
can be obtained by considering, in the infinite superconducting material, a cylindri-
cal core of normal material of radius ξ along the z axis of the magnetic field. This
amounts to assuming a discontinuous change in ns . Applying the London equations
in the superconducting material, we can thus determine the magnetic field distribu-
tion around the normal core. We have
λ2 B(r) − B(r) = 0 .
Taking into account the cylindrical symmetry of the problem, this can be written
d2 Bz (r)
λ2 − Bz (r) = 0 . (6.24)
dr2
As the field vanishes in the bulk of the superconductor, i.e., for r → ∞, this equation
has the solution
Bz (r) = B0 K0 (r/λ) , for r > ξ , (6.25)
where K0 (x) is the modified Bessel function of the second type, which is given
approximately by
π 1/2
K0 (x) ≈ exp(−x) , for x → ∞ ,
2x
and
K0 (x) ≈ − ln x , for x 1 .
$∞
Note that 0 K0 (x)xdx = 1.
In the normal core of the vortex, the constant field is, by continuity,
Bz (r) = B0 K0 (ξ/λ) , for r < ξ . (6.26)
This field is maintained by the persistent currents circulating tangentially around the
core of the vortex which, from μ0 j = curl B, can be written
B0
j⊥ (r) = − K (r/λ) . (6.27)
λ 0
The spatial dependencies of Bz and j⊥ , which both decrease exponentially, are
shown in Fig. 6.10.
Question 6.2. Show that the value of B0 can be determined generally by taking into account
flux quantisation inside a contour containing the core of the vortex. Show that, in the case
of an extreme type II superconductor for which λ ξ , if the flux through this contour is a
flux quantum Φ0 , the constant B0 is independent of ξ and given by
Φ0
B0 = . (6.28)
2π λ2
For ξ λ, the core of the vortex reduces to a singularity, because ξ → 0. A simple

way of describing the properties of the vortex mathematically is then to insert a δ(r)
source term for the field into the second London equation (5.11) which becomes
Bz j⊥
ns
0 0
ξ r ξ r
Fig. 6.10 Spatial dependencies of the superconducting carrier density, the magnetic field, and the
current circulating around the core of the vortex in the London model
λ2 μ0 curl j + B = Φ0 δ(r)z . (6.29)
It can be checked that this equation does indeed yield the solutions obtained in (6.25)
and (6.27) for the fields and currents, and also the flux quantisation condition, but
with an unphysical divergence of the field and current for r < ξ . To account for the
physical reality, we simply equate the field value in this case with that for r = ξ , i.e.,
Φ0
Bz = K0 (ξ/λ) , for r < ξ .
2π λ2
This simplified model of the vortex can be used to determine the energy of the vortex
per unit length, interactions between vortices, critical fields, and so on.3
6.3.3 Type II Superconductivity
As early as 1950, physicists working with Landau in Russia were developing similar
thermodynamic arguments to the ones we presented in Sect. 6.2, but more realistic.
In particular, in the expression for the free energy, Ginzburg and Landau allowed for
a spatial variation of the superconducting pair density, which defines the coherence
length. Their analysis does not lead to simple analytic solutions. This is why we
have chosen here to develop more approximate models which nevertheless include
the main physical effects. They show in a similar way to what was said above that
superconductivity is type I when
λ 1
κ= <√ ,
ξ 2
and that, when this does not hold, the energy required to create a vortex becomes
negative4 above a lower critical field

Φ0 λ ξ λ
Hc1 (T) = ln = √ ln Hc (T) . (6.30)
4π μ0 λ 2 ξ λ 2 ξ
3 This can be seen in Problem 13: Magnetisation of a Type II Superconductor.

4 This almost linear relation between Hc1 and ξ/λ may seem somewhat surprising, since the lamel-
lar model led to Hc1 = Hc (ξ/λ)1/2 . The difference comes about because, for the geometry of a
vortex, the energy of formation of the normal region per unit length is proportional to
μ0 2 2
ξ Hc − λ2 Ha2 ,
2
since the cross-sectional area of the normal core of the vortex goes as ξ 2 and the field penetrates a
cylinder of cross-sectional area proportional to λ2 .
For H < Hc1 , the Meissner effect occurs, but since λ > ξ , this situation is close to
that prevailing for the London model. Indeed, when κ 1, i.e., λ ξ , the density
of superconducting carriers barely changes over the penetration depth, and hence
λ0 λL .
As long as H Hc1 , the field penetrates via widely spaced vortices, as shown
in Fig. 6.11a. These vortices look very similar to those described in Sect. 6.3.2, but
in this more realistic case, the spatial dependencies of ns (r), Bz (r), and j⊥ (r) do not
exhibit discontinuities, as shown in Fig. 6.11b. Since the spacing between vortices
is much greater than λ, the current vanishes between them. The flux corresponding
to one vortex is quantised, and exactly equal to the flux quantum Φ0 .
Abrikosov succeeded in minimising the free energy in the presence of a field,
and thereby demonstrated that the lattice formed by these vortices is hexagonal, as
observed experimentally in the image on p. 176.
As Ha increases, the size of the primitive cell of this lattice decreases, and when
the magnetic field reaches the upper critical field value
√
Φ0 λ 2
Hc2 (T) = = Hc (T) , (6.31)
2π μ0 ξ 2 ξ
the field penetrates uniformly and the metal becomes normal again.
Question 6.3. In principle, the ratio λ/ξ can be increased indefinitely. Show using a ther-
modynamic argument that the value of Hc2 is nevertheless bounded.
ns
2ξ
r
B
(a) Ha (b)
λ
r
js
λ
r
2ξ
Fig. 6.11 Penetration of the field in a type II superconductor. (a) Penetration by tubes of field,
arranged in a hexagonal lattice. The flux is Φ0 per primitive cell. (b) Vortex structure close to Hc1 .
Radial dependencies of the superconducting electronic density ns , the magnetic induction B, and
the supercurrents j, indicating the lengths λ and ξ
When Ha Hc2 , as we have already seen for the lamellar model, the vortex spac-
ing becomes lower than λ, and the vortices can no longer be considered independent,
in contrast to those for fields close to Hc1 , as described in Fig. 6.11b. Close to Hc2 ,
the situation is similar to that described in Fig. 6.8b for the lamellar model. A map
of the magnetic field still reveals a hexagonal lattice, but with lattice constant ever
closer to ξ . The flux per primitive cell of this lattice remains at the same value,
i.e., Φ0 , but there is no more than a spatial undulation of the magnetic field, whose
amplitude is steadily attenuated until it disappears altogether for Ha = Hc2 . This
disappearance occurs more gradually than in the lamellar model. Indeed, the result
obtained in Sect. 6.2.3 gives a discontinuity in B at Hc2 , and this no longer exists in
Abrikosov’s exact solution, nor in the experimental observations (see Chap. 10).
Question 6.4. Can you determine the magnetic field that was used experimentally to obtain
the image of the vortex lattice in the figure on p. 176?
The phase diagram (H, T) of a type II superconductor is shown in Fig. 6.12.

Ginzburg and Abrikosov received the Nobel prize in 2003 for this work.
A non-uniform current distribution is also produced in Josephson junctions in the
presence of a magnetic field.5
Question 6.5. Is there a simple way to find out whether a material is type I or type II? Would
it be possible via a levitation experiment, for example, if you were given a superconducting
pellet with area S and three small magnets with much smaller dimensions than S and with
magnetisations μ0 M0 equal to 0.01, 0.03, and 0.1 tesla, together with a cryogenic fluid that
could be used to cool the pellet below Tc . How would you find out whether the material was
type I or type II?
H
Hc2 (T )
Normal
Mixed phase metal
Hc (T )
Hc (T )
1
Fig. 6.12 Phase diagram for

a type II superconductor 0 T Tc
5 See Problem 12: Josephson Junction in a Magnetic Field.

6.3.4 Applications of Type II Superconductors
The existence of type II superconductors provides a way of using superconductivity

to produce high magnetic fields. One obvious reason is that one can significantly
increase the magnetic field range over which the superconducting state persists. A
current can thus be circulated without losses in a superconducting coil to create
uniform fields Ha close to Hc2 . In the geometry of a solenoid comprising many
layers of wire, one inner turn of the coil is subjected to a total field Ha which is well
above the field created by the current passing through the turn itself, so the limiting
factor will indeed be Hc2 .
But the use of type II superconductors provides another advantage. We have seen
that, in a type I superconductor, whatever the geometry of the material, the current
will only circulate in a thickness λ, with a critical current density determined by
Hc /λ. In an extreme type II superconductor, close to Hc2 , the vortices have a spacing
of the order of ξ , and although the microscopic currents are modulated over this
distance, the transport current in the superconductor circulates throughout the bulk
of the material. So for a given critical current density, the current through a wire
of diameter d is larger than for a type I superconductor by a factor of the order of
d/λ. As λ ≈ 500 Å, the current gain is very high, viz., ≈ 2, 000 for a wire measuring
1 mm across.
Note on the other hand that these favourable conditions nevertheless lead to a
certain level of technical complexity due to the effect of the electric current and
magnetic field on the vortex lattice. In particular, the inner wires of the coil are
subjected to a field Ha that is perpendicular to the current circulating in the wire. The
associated Lorentz force tends to displace the vortices perpendicularly to these two
directions. As the vortices contain a core of normal material, this displacement leads
to dissipation, and the material is no longer strictly speaking superconducting! In
practice, these vortex displacements only occur easily if the material is exceedingly
clean, this allowing rapid attainment of thermodynamic equilibrium. On the other
hand, when there are defects that can trap the vortices and prevent them from being
displaced by the Lorentz force, a superconducting state without dissipation can be
maintained. Considering all the various requirements, it turns out that the technical
optimisation of the wires for superconducting coils is much more complex than one
might have imagined.
The transition from idea to actual application inspired a technical ingenuity that
would have been difficult to anticipate. The photograph in Fig. 6.13 is an elec-
tron microscope image of a wire made from an NbTi alloy of the kind used to
make superconducting coils that can produce high persistent fields of the order of
8 tesla. As can be seen, it comprises several thousand NbTi filaments a few mi-
crons in diameter inserted within a copper matrix. The arrangement of the NbTi
filaments is similar to that used in a thick rope. They are distributed in thin strings
that are twisted in the direction perpendicular to the plane of the image. These
strings are themselves twisted together to form the rope structure that constitutes the
wire.
6.4 Summary 195
Fig. 6.13 Cross-section of a

wire comprising 6,000 super-
conducting NbTi filaments in
a copper matrix. The NbTi
multifilament structure was
designed to optimise the ther-
modynamic stability of the
fields and currents in extreme
type II conditions when the
applied field is close to Hc2 .
With the kind permission
of D.R.
6.4 Summary
The superconducting state has a lower free energy than the normal state. The
exclusion of the magnetic field from a superconductor leads to an increase in the
free energy. The Meissner effect thus implies the existence of a thermodynamical
critical field Hc for which these two effects balance out. Knowing only the experi-
mental temperature dependence of Hc , the following thermodynamic quantities can
be determined: the free energy Gs , entropy Ss , and specific heat Cv , which charac-
terise the superconducting phase. In particular,
μ0 2
Gs (T, Ha ) − Gn (T, 0) = − Hc − Ha2 .
2
If we consider a thin superconducting film of thickness smaller than the penetration
depth λ, the Meissner effect is not perfect and the increase in free energy in the
presence of a field Ha is less than the value given above for a bulk superconductor.
Superconductivity therefore persists for fields above Hc .
This observation implies that, in order to minimise its free energy in the presence
of a magnetic field, a superconductor should divide up into normal and supercon-
ducting regions. The fact that the Meissner effect exists shows that this subdivision
has a nonzero energy cost, and that the normal metal–superconductor interface can-
not be very abrupt. We are compelled to introduce a new characteristic length ξ
called the coherence length, which characterises the average distance between two
paired electrons in the superconducting state.
Calculating the thermodynamic free energy (illustrated for the lamellar model),
it can be shown that there are two types of superconductor:
√
1. Type I superconductors, for which λ < ξ/ 2 and there is a perfect Meissner
effect up to Hc . √
2. Type II superconductors, for which λ > ξ/ 2 and there is a Meissner effect up
to a lower critical field

Φ0 λ ξ λ
Hc1 (T) = ln = √ ln Hc (T) .
4π μ0 λ 2 ξ λ 2 ξ
Beyond Hc1 the magnetic field penetrates in the form of flux tubes called vor-
tices. In this mixed phase, the vortices are arranged into a hexagonal lattice. The
density of the lattice increases with increasing field, until the superconductivity
is destroyed and the field penetrates uniformly for an upper critical field value
√
Φ0 λ 2
Hc2 (T) = = Hc (T) .
2π μ0 ξ 2 ξ
The existence of this mixed phase provides a way of making superconducting
wires that can be used to produce intense magnetic fields.
Question 6.1
With a larger magnetic field value than Hc (0), the normal state is restored. We ob-
serve that, for T > 0.3 K, the specific heat has the form
Cn /T = 0.6 + 0.057T 2 .
The first term gives the coefficient γ of the electronic specific heat, while the second
is a contribution to the measured specific heat going as T 3 . It is produced by the
vibrational degrees of freedom of the atoms in the metal, i.e., phonons.6
We also observe at low T an increase in the specific heat that does not exist
for zero field in the superconducting state. There are therefore excited states which
appear when a field is applied. These are thus of magnetic origin and not associated
with the metallic state. They would appear to be due to paramagnetic impurities
exhibiting a Curie susceptibility. It is easy to check that the internal energy of such
an ensemble of paramagnetic entities contributes in the following way to the specific
heat [1]:
dU −2
C= = NkB (μB B/kB T)2 cosh(μB B/kB T) .
dT
Since μB B kB T under the experimental conditions, the corresponding contribution

of such terms to Cn increases as T −2 at low T.
Question 6.2
The quantisation condition for the flux through a contour (C) around the vortex is
%
Φ0 = λ μ0 j · dl +
2
B · ndS .
(C) S

If we consider a circular contour of radius R → ∞, then |j| → exp(−R), and the first
integral vanishes for large R. Only the second contributes to the flux, which can be
determined by taking an integration surface extending over the whole plane. Then,
using (6.24) and (6.25), we can write
ξ ∞
Φ0 = B0 K0 (ξ/λ)2π rdr + B0 K0 (r/λ)2π rdr .
0 ξ
This can be used to determine B0 by straightforward integration:

∞
ξ 2
Φ0 /B0 = 2π λ2 K0 (ξ/λ) 2 + K0 (x)xdx .
2λ ξ/λ
For an extreme type II superconductor, ξ/λ 1, the first term corresponds to the
flux in the core of the vortex and tends to 0, while the integral is equal to 1. We thus
recover the proposed expression
Φ0
B0 = .
2π λ2
Question 6.3
In a field Ha , the drop in free energy of the normal state associated with the Pauli
susceptibility of the metal is
1
Gn (T, Ha ) − Gn (T, 0) = − μ0 χP Ha2 .
2
We neglected this term at the beginning of Chap. 6 when discussing the Meissner
effect. However, for high fields close to Hc2 , this term can no longer be ignored.
When
1 1
− μ0 χP Ha2 < Gs (0, 0) − Gn (0, 0) = − μ0 Hc2 ,
2 2
the normal state will necessarily be reestablished. This therefore leads to an upper
bound for any critical field, and in particular for Hc2 , called the Pauli limit:
√
Hcm2 < Hc / χP .
Question 6.4
Flux quantisation implies that, in a primitive cell of the vortex lattice of area S,
μ0 Ha S = Φ0 .
√
If a is the measured dimension of the hexagonal cell, we have S = a2 3/2, and
hence
2Φ0
μ0 Ha = √ .
a2 3
From the figure, we obtain a = 550 Å, so
μ0 Ha = 1.06 tesla .
Question 6.5
If the material is type I, the Meissner effect is total, provided that H < Hc . The
properties of the magnet–pellet system are the same whatever its thermomagnetic
history. The levitation will thus be the same whether the magnet is approached by
the cold pellet or the magnet is placed on the pellet and subsequently cooled. The
first case tests screening, the second the Meissner effect.
However, if one of the magnets has a magnetisation M0 > Hc , this reversibility
may not be observed. It might levitate when brought near to the pellet and remain
at a distance from the pellet such that the field on the pellet is less than Hc . On the
other hand, if the magnet is placed on the hot pellet and then cooled, part of the
pellet may remain in its non-superconducting state, and the force due to the residual
Meissner effect may not be sufficient to oppose gravity. In this case, the pellet would
not levitate.
If the material is type II, there is in principle no difference with a type I material
as long as M0 < Hc1 , but a significant difference with a type I material is to be
expected if M0 > Hc1 . When the magnet is placed on the pellet and the system
subsequently cooled, the field will penetrate in the form of vortices when T goes
below Tc . In this case, the diamagnetism is significantly reduced and the levitation
will be less high than when the pellet has been cooled in zero field. For this reason,
it is not always so easy to distinguish a type I material with a magnet such that
M0 > Hc and a type II material with M0 > Hc1 .
For a type II material and M0 < Hc1 , the identical behaviour to a type I material
and M0 < Hc only becomes effective at thermodynamic equilibrium. Indeed, for a
type II material, Hc1 varies with T and vanishes at Tc . When the magnet is placed
on the pellet and the latter is then cooled down to T0 Tc , the field continues to
penetrate for T Tc , and must be excluded in the form of vortices when T goes
down to T0 . If the superconductor is very pure, this exclusion occurs very quickly,
and at T0 the vortices are effectively excluded if M0 < Hc1 . There is reversibility
and the same situation occurs with a type I superconductor. On the other hand, if the
material is not pure, the exclusion does not occur, because the vortices are trapped
by defects (vortex pinning) and the levitation is reduced, even if M0 < Hc1 (T0 ).
The situation is thus generally irreversible for a type II superconductor, whatever
the value of M0 , while it is reversible for a type I material when M0 < Hc .
If the vortex pinning is effective, the magnetisation of the type II superconductor

may even be positive and lead to an attraction of the magnet by the superconducting
material (see Fig. 10.7, curve E). In this case one cannot strictly speak of levitation
of the magnet over the superconductor, but rather of attraction and trapping of the
magnet at a well defined distance. This is important for applications in which one
wants to levitate a superconducting material over magnets, while ensuring an attrac-
tion and stability of the object above the magnetic material. This is indeed the case
for a levitated train, where an on-board superconducting coil levitates over magnetic
rails.
Chapter 7
Microscopic Origins of Superconductivity
Contents
7.1 Conventional Metal Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

7.1.1 Superconducting Metals and Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.1.2 Attractive Interaction Due to Phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.1.3 The Band Gap in the Superconducting State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
7.2 Cooper Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.2.1 Electrons in the Conduction Band of a Semiconductor . . . . . . . . . . . . . . . . . . . . . . . 212
7.2.2 Extra Electrons in a Metal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
7.3 BCS Theory: Experimental Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.3.1 Ground State and Band Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.3.2 Excited States in the BCS Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
7.3.3 Electron Tunneling Determination of the Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
7.3.4 Critical Temperature and BCS Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
7.3.5 Coherence Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
7.4 High-Tc Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.4.1 Cuprates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.4.2 Other Families of Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
202 7 Microscopic Origins of Superconductivity
After solving this puzzle, you should understand its physical meaning with the help of the present
chapter. You should also be able to work out what Claire and David are thinking about in the picture
on p. 146
7.1 Conventional Metal Superconductors 203
By measuring the flux quantum, it was established that the carriers are electron
pairs in a superconductor. This is a key feature of all electron superconductivity
phenomena so far observed. Note, however, that the discussion of the model based
on Cooper pairs in Chap. 5 seriously distorts the actual chronology of these discov-
eries. London did indeed realise as early as 1950 that superconductivity would have
to be treated as a macroscopic quantum effect, and envisaged flux quantisation, but
he assumed that the superconducting carriers had charge −e. The point is that the
experiment testing the flux quantisation hypothesis was not carried out until 1961,
by which time the existence of Cooper pairs was well accepted.
In Sect. 7.1, we describe how the microscopic origin of superconductivity was
established for metals and conventional alloys. To begin with, we consider the prop-
erties of simple metal superconductors. The experimental observations we describe
led step by step to the essential elements of the puzzle that were then assembled in
1957 by Bardeen, Cooper, and Schrieffer (BCS) into a theoretical edifice that could
explain the origins of superconductivity in conventional metallic systems.
Section 7.2 examines the conditions required for an attractive interaction between
two electrons to produce a bound state in the conduction band of an insulator, and
in a free electron gas. We will then be able to describe the calculation with which
Cooper demonstrated in 1956 the instability of the electron gas in the presence of
an attractive interaction between electrons, no matter how weak it may be.
Section 7.3 presents without derivation the main consequences of the BCS theory,
which have led to a fairly complete explanation of the experimental properties of
superconductivity in metals and alloys.
The originality of the new superconductors with high critical temperature, dis-
covered in 1986 by Bednorz and Müller, is outlined in Sect. 7.4. We shall then be
able to understand the intense activity in this area, not only on the fundamental level
but also from the point of view of applications.
7.1 Conventional Metal Superconductors
We begin in Sect. 7.1.1 with a presentation of the known metal superconductors. The
thermodynamic considerations presented in Chap. 6 led us to seek a ground state for
the electron system that had lower energy than a free electron gas. We shall see in
Sect. 7.1.2 how the involvement of the atomic nuclei in the superconducting state
was established experimentally and led naturally to the idea of an attractive inter-
action between electrons, associated with the Coulomb interactions with the atomic
nuclei. Experiment also revealed other novel properties of metal superconductors,
and in particular the existence of a band gap in the excitations of the electronic
system (Sect. 7.1.3).
7.1.1 Superconducting Metals and Alloys
One of the first questions raised right after the discovery of superconductivity by
Kammerlingh Onnes was the issue of whether this phenomenon was particular to
certain specific systems, or whether it had a more general relevance. The conduc-
tivity of metals was thus systematically studied at low temperatures. New cryogenic
techniques had to be developed, in particular using 3 He to achieve temperatures
below 1 K, before it could be demonstrated that many metals were in fact super-
conducting for a wide range of different values of Tc . Examining the periodic table
shown in Fig. 7.1, we may make the following observations:
1. There are some rather surprising exceptions to superconductivity, including good
metals like the noble metals Cu, Ag, and Au, but also the alkaline earth metals
Li, Na, and K.
2. Metals with low conductivity at room temperature, such as the 4d or 5d transition
metals, are among the best superconductors.
3. Semiconductors and some semi-metals in groups III to VI become conducting
and superconducting under pressure.
4. The magnetic 3d and lanthanide metals are not superconducting.
Among the elementary metals, the one with the highest critical temperature is nio-
bium (Nb) with Tc = 9.2 K. On the other hand, some have very low critical tem-
peratures, e.g., tungsten (W), with Tc 0.01 K, and their superconductivity was
only discovered when they could be produced in a pure enough form. But does this
suggest that superconductivity is limited only to pure metals? Far from it! Specific
alloys and compounds even exhibit higher critical temperatures than pure metals
(see Table 7.1). Note, however, that alloys are generally type II superconductors,
whereas most pure metals are type I.
Fig. 7.1 Periodic table showing those elements that are superconducting (full colours) and those
that are magnetic (green hatching). Grey: Elements that become superconducting at their critical
temperature under standard pressure. Orange: Elements that become superconducting under pres-
sure, with observed value of Tc in brackets
Table 7.1 Superconductivity

Compound Tc (K) Compound Tc (K)
of some metal compounds and
alloys Nb3 Sn 18.05 V3 Ga 16.50
Nb3 Ge 23.0 V3 Si 17.10
Nb3 Al 17.5 UCo 1.70
Nb3 Au 11.5 Ti2 Co 3.44
NbN 16.0 La3 In 10.40
MoN 12.0 InSb 1.90
Without even mentioning more recent discoveries of new compounds exhibiting

superconductivity, made since the 1980s, it is quite clear that this phenomenon is
far from being the exception. It is therefore essential to understand not only its
microscopic origin, but also in the opposite case the fundamental explanation as to
why some metals are not superconducting.
7.1.2 Attractive Interaction Due to Phonons
We have seen that Tc characterises the condensation energy of the electron system
in the superconducting state, i.e., the drop in energy that occurs when the electron
gas becomes superconducting. Some effort has been made to determine whether the
superconducting transition is associated with a change in crystal structure or in the
lattice parameter. However, careful crystallographic studies using X rays or neutrons
have never revealed any evidence of such a structural change, and this suggests that
these processes are essentially electronic. In a system of independent electrons like
those considered in Chap. 1, no such effect could occur. We must therefore introduce
a hitherto neglected negative energy contribution into the Hamiltonian of such a
system. Experimentally, the obvious thing to do in order to find the origin of this
contribution is to look for microscopic factors affecting the critical temperature. But
to reach a clear conclusion, a given factor has to be modified experimentally while
keeping all other factors the same, in order to see whether any change in Tc occurs
as a consequence. For example, we may ask whether Tc varies with the density of
carriers in the normal metal. This can be altered by applying a pressure. It is easy
to see that the pressure will also change the hopping integrals between the orbitals
of nearest neighbouring atoms, and hence also the band structure (see, for example,
what happens for the semiconductors mentioned above, which become metallic). It
is thus difficult to change one physical parameter without changing others too.
a. The Isotope Effect
It turns out that the mass of the ions can be changed quite easily without changing
their charge, and therefore without changing the Coulomb potentials affecting the
electrons. This is done by exploiting the fact that some elements of the periodic
table exist in the form of several different stable isotopes. An example is mercury
(Hg), which has five stable isotopes, with masses between 198 and 204, and also tin
(Sn). One byproduct of the nuclear power industry has been the possibility of pro-
ducing stable isotopes artificially, and improving techniques for isotope separation.
Since the 1950s, it has thus been possible to prepare isotopically enriched samples
of Hg and Sn, which have been used to show that the critical temperature of su-
perconductivity decreases as the atomic mass M of the ions is increased. Accurate
measurements of the critical temperature for different values of M for a given metal
show that the dependence is (see Fig. 7.2)
Tc = kM −α , α 0.5 . (7.1)
This isotope effect constitutes an irrefutable demonstration that the ions are involved
in the stabilisation processes of the superconducting state.
Fig. 7.2 Dependence of the 0.585

critical temperature of tin on
Sn
the atomic mass (log –√log
0.580
graph). Dashed line: 1/ M
dependence. Different sym-
bols correspond to measure- 0.575
ments made by different M −1/2
logTc
groups. From [6, p. 245] 0.570

Springer c 1995
0.565
0.560
2.05 2.06 2.07 2.08 2.09 2.10
logM
b. Electron–Phonon Interaction
With the simplifying hypotheses of Chaps. 1 and 3, we assumed that the ions held
fixed positions in the crystal. However, we have already seen that their oscillations
about their equilibrium positions are responsible for a contribution to the electrical
and thermal conductivities (see Chap. 4). The motions of the ions can be decom-
posed into eigenmodes called phonons,1 whose energy dispersion relation defines
the Debye frequency ωD , or equivalently, the Debye temperature θD = h̄ωD /kB . It is
the interaction between electrons and phonons that leads to a temperature dependent
contribution to the electron relaxation time. The ions thus play an important role in
the physical properties of a metal. But can they help to create a different ground
state energy for the electron system?

Note that the existence of an attractive Coulomb interaction between the elec-
trons and ions is such that, when an electron propagates, it creates a distortion of
the crystal lattice. It attracts towards it those positive ions that sit close to its tra-
jectory. The distortion produced in the environment of the trajectory will persist for
a certain time because ionic displacements occur at low frequencies (see Fig. 7.3).
Another electron will thus gain free energy if it stays in the region distorted by the
first electron rather than in a non-distorted region. This corresponds to an attractive
interaction between the two electrons.
A simple picture that may help to understand this fact considers two heavy beads
lying on a horizontal elastic membrane. One bead distorts the membrane. The total
gain in potential energy and distortion energy of the membrane is greater for two
nearby beads than for two widely separated beads. The result is an effective attrac-
tive interaction between the beads, which thus prefer to be stuck one next to the
other.
The case under consideration here is dynamic and hence much more complicated
than the static model of the beads. Indeed, an electron moves with speed close to
the Fermi speed. When it passes close by the ions, the latter begin to move, but
since the collective modes of the ionic motions have a low Debye frequency ωD ,
the amplitude of their displacement will only reach its maximum after a time of the
order of 2π/ωD . We can thus understand that the maximum energy gain for another
electron will only be obtained roughly speaking when the distance between the two
electrons is of the order of 2π vF /ωD , and this specifies a characteristic distance
between the electrons in a pair. This in turn can be used as a physical basis for
estimating the coherence length ξ , at least to within an order of magnitude. This idea
was put forward by Fröhlich as early as 1950, and was supported by the discovery
of the isotope effect. Indeed, the gain in energy due to the distortion induced by the
electron is determined by the energy ∼ h̄ωD of the atomic vibrational modes. Since
these harmonic modes are determined by the repulsive interaction forces K between
the ions, their frequencies, and hence also ωD , are proportional to (K/M)1/2 , which
provides an explanation for the isotope effect.
νF
ν
Positive ions ξ
Electron
(a) (b)
Fig. 7.3 Ion displacements induced by an electron in a solid. (a) A slowly moving electron pro-
duces a distortion which follows its trajectory. (b) If the electron moves quickly, the distortion
lags behind. This is the situation when electrons move at the Fermi velocity in a metal. A second
electron can follow the distortion caused by the first
The discovery of the isotope effect thus suggests that there is some kind of at-
tractive effective interaction between electrons induced by the vibrational modes of
the ions, and that this interaction plays a key role in the mechanisms of supercon-
ductivity. We shall see in Sect. 7.2 how an attractive interaction between electrons
can induce a bound state between electrons.
7.1.3 The Band Gap in the Superconducting State
The thermodynamic measurements discussed in Chap. 6 brought out the difference

in energy between the superconducting ground state and the normal metal state.
However, it is just as important to understand the excited states of the supercon-
ductor. We can find out about the spectrum of excited states by investigating the
thermodynamic properties at low temperatures. More direct information can be ob-
tained by studying the absorption of photons. These experiments show that there is
a band gap between the ground state and excited states of the electron system.
a. Specific Heat at Low Temperature
We return here to the specific heat measurements already mentioned in Chap. 6.

They showed that the temperature dependence of the critical field was not perfectly
quadratic. Recall that CdT is by definition the energy dU that must be supplied to
increase the temperature of the electron gas by dT. At low temperatures, the spe-
cific heat can thus be used to test the energy distribution of the excited states of the
system. For a normal metal, all the states close to the Fermi level are accessible and
a great deal of energy must be supplied to significantly change the occupations of
the levels. In the superconducting state, the specific heat is found experimentally to
be much lower (see Fig. 6.3), indicating that few excited states are actually accessi-
ble. With the quadratic form of Hc (T) given in (6.9), we expect Cs /T to have a T 2
dependence, but a much slower variation is in fact found experimentally.
Plotting the experimental results of Fig. 6.3 as a function of 1/T, we find that
(see Fig. 7.4)
Cs = a exp(−/kB T) . (7.2)
This looks like what would be expected for the electron contribution to the specific
heat of an insulator. As discussed in Chap. 3, the population
factor of the levels at
the bottom of the conduction band is proportional to exp − (Ec − EF )/kB T , which
would lead to a specific heat going as exp(−Eg /2kB T), where Eg is the width of the
band gap [see (3.62)].
These specific heat measurements, carried out on superconductors in 1954, thus
suggest that there is a band gap of width in the electronic level structure of the
superconducting state (see Fig. 7.4). The energy is found experimentally to lie
close to 1.5kB Tc for various metals (V, Sn, Al, and others).
Fig. 7.4 Measurements of the

specific heat Cs of metallic
gallium. The experimental 1.0
values of Cs (experimental 7.46exp(−1.39Tc/T)

points shown in Fig. 6.3,
obtained for the lowest values
of T) are plotted on a loga-
Cs/ΓTc
0.1
rithmic scale as a function of
1/T, and are found to vary as
exp(−/kB T). This shows
that there is a band gap in
the elementary excitations 0.01
of the superconducting state.
Gallium
From [7, p. 365], 8th edn.,
Wiley 2005
c
0.001
1 2 3 4 5 6
Tc/T
b. Microwave Absorption
This similarity with insulators naturally encouraged the search for spectroscopic
methods that could determine the band gap. We know that a semiconductor is trans-
parent to photons with lower energy than the band gap energy Eg , and absorbs them
by creating electron–hole pairs if hν is greater than Eg . We saw in Chap. 3 that the
optical absorption or reflectivity can be used to determine the band gap of a semi-
conductor. Similar experiments were thus carried out on metal superconductors in
1955. From the values of suggested by the measurements of C, the band gap is
expected in the microwave frequency range (10 K ∼ 1 meV ∼ 10 cm−1 ).
The experiments shown in Fig. 7.5 demonstrate the existence of a band gap. The
microwave absorption is zero for low frequencies and increases suddenly above a
certain frequency that can be used to determine the width of the band gap. The
1.0
0.8
σ1 /σN
0.6
0.4 Indium
0.2
h̄ω/kBTc
0 2 4 6 8 10 12 14 16 18 20 28 30
Fig. 7.5 Frequency dependence of microwave absorption in superconducting indium for T Tc .

The absorption normalised by the value in the normal state is plotted as a function of the frequency
normalised at the critical temperature. An absorption threshold is observed for h̄ω ∼ 3.8kB Tc . From
Ginsberg, D.M., Tinkham, M.: Phys. Rev. 118, 990 (1960) Standard American Physical Society
Copyright. http://link.aps.org/doi/10.1103/PhysRev.118.990
energy of the microwave absorption gap is found to be twice the value obtained
via specific heat measurements. This reflects the fact that a photon can only be ab-
sorbed in the superconducting state if the photon energy is high enough to break a
superconducting electron pair, and hence excite two electrons by making them ‘nor-
mal’. The measured gap therefore corresponds to twice the energy per electron
needed to populate a thermally excited state. We also understand that this pair bind-
ing energy 2 is related to the condensation energy in the superconducting state,
and hence to kB Tc .
7.2 Cooper Instability
The electron–phonon interaction in a solid can lead to an indirect attractive interac-

tion between two electrons. This may therefore induce a state in which the electrons
are bound together in pairs. In quantum mechanics, we know that a potential well
does not always yield a bound state, i.e., a particle is not necessarily confined spa-
tially by a potential well. This leads us to ask under what conditions an attractive
interaction between two electrons in a solid will in fact yield a bound state. Let us
examine the cases where these two electrons are in the conduction band of a semi-
conductor or at the Fermi level of a metal. We consider the simple situation where,
in the absence of any interaction between the electrons, the eigenstates are plane
waves.
The Hamiltonian for two interacting electrons is thus
1
H= (p2 + p22 ) + V(r1 − r2 ) , (7.3)
2m0 1
where r1,2 are the positions of the two electrons, p1,2 are their momenta, and m0 is
their mass. The interaction potential V(r) tends to zero at infinity.
By the usual transformation where R is the position of the center of mass,
1
R = (r1 + r2 ) , r = r1 − r2 , (7.4)
2
the total wave function factorises to
ψ(R, r) = eiK·R φ(r) , (7.5)
where the relative wave function φ(r) satisfies

h̄2
− r + V(r) φ(r) = Eφ(r) , (7.6)
m0
with r the Laplacian with respect to r.

7.2 Cooper Instability 211
The wave function φ(r) can be expanded relative to the basis of plane waves
r|k = ()−1/2 eik·r ,
with energy eigenvalues h̄2 k2 /m0 = 2εk . Here is the volume over which periodic
boundary conditions are imposed. The wave function thus becomes

|φ = φk |k . (7.7)
k
We now expand V(r) in a Fourier series over the same states

1
V(r) = Vk eik ·r . (7.8)

k
Substituting (7.7) and (7.8) into the expression in (7.6), we obtain

1
(2εk − E)φk |k + Vk eik ·r φk |k = 0 .

k k k
If we now project this equation onto an eigenstate of the basis with wave vector k,
by multiplying on the left by the bra k|, only those terms such that k + k = k will
survive in the double sum. This will therefore lead to
1
(2εk − E)φk + Vk−k φk = 0 . (7.9)

k
The usual Schrödinger equation (7.6) is thus replaced by the set of integral equations
(7.9) written for all possible k. Such equations are often more difficult to handle
than differential equations, and the formulation (7.9) of the problem is rarely used
in general. However, in this case, it will prove to be very useful.
We are concerned here with the case where the interaction between electrons is
short range. When the interaction is a delta function δ(r) with zero range, Vk is
constant in momentum space, with
Vk = −V0 , (7.10)
where V0 is positive. In this particular case, (7.9) can be written

V0
(2εk − E)φk − A=0, (7.11)

where A = φk can be eliminated by summing φk over all the k, whereupon
k
V0 1
=1 . (7.12)
2εk − E
k
This equation can be used to determine whether the energy E of the electron pair
can be lowered in the presence of the interaction V0 , in which case the ground state
is a bound electron pair.
7.2.1 Electrons in the Conduction Band of a Semiconductor
Suppose to begin with that two extra electrons are injected into a semiconductor at
T = 0. In these cases, the available states |k are those in the conduction band, and
their energy levels are given by εk = h̄2 k2 /2m0 as measured from the bottom of the
conduction band. Replacing the sum over k in (7.12) by an integral over the kinetic
energy εk of the two particles, we obtain
∞
Dd (ε)
V0 dε = 1 , (7.13)
0 2ε − E
where the density of states is given by Dd (ε) = Cd εd/2−1 for electrons in dimension
d (d = 1, 2, 3), and Cd = Cd (2m0 /h2 )d/2 was determined in (3.50). The existence of
a bound electron pair corresponds to a solution of (7.13) with E < 0.
Note that the integral in (7.13) diverges at the upper bound, except for d = 1. This
reflects the fact that the form (7.10) for the interaction corresponds to an attraction
with zero range. To remove this anomaly, we could consider a potential with nonzero
range. In the present case, where the attraction between the electrons is transmitted
via phonons, the interaction is only significant if the energy difference between the
two electrons is lower than the typical energy h̄ωD of a phonon. We saw a qualitative
illustration of this in Sect. 7.1.2. This suggests limiting the interaction to states with
lower energy than h̄ωD , and hence replacing the upper bound at ∞ in the integral of
(7.13) by h̄ωD . We thereby obtain
h̄ωD ε d/2−1
V0 Cd dε =1. (7.14)
0 2ε − E
A considerable effort is required to treat the phonon dynamics in a more rigorous
way, while the main results can in fact already be obtained from (7.14).
The condition for the existence of a bound electron pair2 corresponds to a solu-
tion of (7.14) with E < 0. Note that the left-hand side of (7.14) is a function of E
which tends to zero as −1/E when E →√−∞. In dimension d = 3, it remains finite
when |E| → 0 and takes the value V0 Cd h̄ωD . Consequently, a solution E < 0 and
hence a bound state only
√ exist if the attractive potential is stronger than the crit-
ical value V0c = 1/(Cd h̄ωD ). In one or two dimensions, however, the integral in
2 Electrons can be injected into the conduction band of a semiconductor by illuminating it with
a laser at an energy above the band gap. The above calculation shows that bound pairs, called
bipolarons, can only exist under certain specific conditions.
7.2 Cooper Instability 213
(7.14) diverges when |E| → 0, so in these cases a bound state exists for an attractive
potential of any strength, even infinitely weak.
7.2.2 Extra Electrons in a Metal
Now consider the case of a metal, described as a free electron gas. All states with |k|
< kF are therefore occupied. We may ask once again whether two extra interacting
electrons added to the electron gas will form a bound pair. Compared with the pre-
vious situation, the Pauli exclusion principle stipulates that the eigenstates of the
electron pair can only be constructed from unoccupied plane wave states above the
Fermi level, and the φk are only nonzero if |k| > kF . By restricting the sum over k to
|k| > kF in (7.9), the condition (7.12) for the existence of a bound state changes to
V0 1
=1. (7.15)
2εk − E
|k|>kF
When there is no interaction between the extra electrons, they will have an
energy 2EF . The existence of a bound electron pair thus requires a solution of (7.15)
with E < 2EF (rather than with E < 0 as in the previous case). Going to an integral
over the energies, (7.15) is equivalent to
EF +h̄ωD EF +h̄ωD
D(ε) 1
V0 dε ≈ V0 D(EF ) dε =1. (7.16)
EF 2ε − E EF 2ε − E
In the second expression, we have neglected the energy dependence of the density
of states over the given energy range, because the typical values of h̄ωD are such
that h̄ωD EF .
Equation (7.16) is easy to understand. Note the resemblance between this equa-
tion and the case d = 2 of (7.14). We thus expect there to be a bound state whatever
the value of V0 . Indeed, calculating the integral in (7.16) and setting E = 2EF − δ,
we find a solution with binding energy δ given for small V0 , i.e., V0 D(EF ) 1, by

2
δ = 2h̄ωD exp − . (7.17)
V0 D(EF )
This is the Cooper effect: whatever the value of the attractive potential, in the pres-
ence of a gas of independent electrons, two electrons will always form a bound pair,
which fully justifies the name Cooper pair.
From (7.11), it is easy to see that φk = φ−k , so we also have φ(r) = φ(−r), or
φ(r2 −r1 ) = φ(r1 −r2 ). The spatial part of the wave function is even under exchange
of the two electrons. The Pauli principle thus requires the spin part of the total wave
function to be odd, which means that the bound state is a spin singlet. We conclude
that two extra electrons subjected to an attractive interaction in a free electron gas
will form a singlet bound state with energy below 2EF .
Fig. 7.6 Spatial representa-

tion of a Cooper pair. The S1
distortions shown by orange −k
and green shading are pro- r1
duced by the electrons at r2
and r1 , respectively
V
r2
k
S2
Since φk = (AV0 /)[2(εk − EF ) + δ]−1 decreases rapidly for εk − E > δ/2, the
pair formed in this way is built up from plane wave states k close to EF .
The spatial representation of the pair depicted in Fig. 7.6 shows that each electron
in the pair moves in association with the delayed distortion of the lattice produced
by the motion of the other electron.3 We shall see below that this wave function can
be used to find the spatial extent ξ of the pair.
Note that the binding energy δ of the pair given by (7.17) depends exponentially
on the reciprocal of the coupling V0 , and cannot by expanded as a power series
in V0 . This shows that there is no way of generating this paired electron state in
which the attractive interaction V0 between the electrons can be treated as a simple
perturbation of the non-interacting electron system.
7.3 BCS Theory: Experimental Evidence
The calculation carried out by Cooper in 1956 refers to an unrealistic situation in

which the electron gas is assumed to be free from interactions, in which only the
two ‘added’ electrons actually interact. In a real metal, a macroscopic number of
electrons with energies close to the Fermi level actually interact together, and the
above argument will apply to many electron pairs. We may thus anticipate a global
instability of the set of electrons on the Fermi surface, leading to the formation of an
infinite number of Cooper pairs. Naively, one might think of an ensemble of pairs
like the one in Fig. 7.6, which could be viewed as independent bosons. This would
correspond to a picture in which the gas of independent electrons is like a set of
3 The representation in Fig. 7.6, which provides a good picture of what an isolated Cooper pair
might look like, corresponds to a pair of electrons with orbital angular momentum, and thus breaks
rotation symmetry. When there is no interaction that can remove the orbital degeneracy, the ground
state of the pair is strictly speaking a superposition of such states with opposite rotations.
7.3 BCS Theory: Experimental Evidence 215
individualistic dancers on the dance floor who, at temperature Tc , gain internal en-
ergy by pairing off into couples dancing a tango. Note that this state would corre-
spond to independent bosons, of the kind arising in Bose condensation discussed in
the box below. However, the superconducting state is in fact very different, because
the density of superconducting pairs is so high that the average distance between
pairs is less than the coherence length ξ .
This is what Bardeen, Cooper, and Schrieffer realised in 1957, when they con-
structed a ground state comprising an infinite number of pairs, such that the pairs
actually lose their individual identities. At any particular moment, an electron is
paired with another electron, but the pairs change round all the time, in such a way
that the macroscopic ensemble is coherent in the quantum sense. An analogy along
the above lines would be a square dance, in which each male dancer must have a fe-
male partner at any moment of time, but change partner in a correlated way as time
goes by. In this way, the dancers constitute a coherent ensemble on the dance floor,
while the long range coherence of the electron pairs is induced by the interaction
between the electrons. (In the square dance, coherence results from the rules for the
dance learnt beforehand by the participants!)
It is beyond the scope of the present book to give a full demonstration of the
BCS theory, which was rewarded by the Nobel prize in 1972. However, Cooper’s
calculation provides the basic theoretical arguments required to understand the BCS
results and the way they account for experimental observations. One success of the
BCS theory was the prediction of a band gap, together with its value.
7.3.1 Ground State and Band Gap
The BCS theory can be used to show that the ground state of the electron system
is a quantum state involving a macroscopic number of electron pairs. It goes well
beyond Cooper’s calculation, because it can determine the energies of the excited
states, and it has the advantage of showing that that there is a band gap, since the
first thermally excited state of the macroscopic electron system lies at an energy
2 above the ground state. An energy of 2 must be supplied to excite an electron
pair, i.e., to break up a pair and transform the electrons into ‘non-superconducting’
carriers (see Fig. 7.7).
In the weak coupling limit V0 D(EF ) 1, is given by

1
= 2h̄ωD exp − . (7.18)
V0 D(EF )
As we saw in the Cooper calculation, the ground state is constructed from electron
states with energies close to the Fermi energy EF (in an energy band h̄ωD ). The
energy of condensation in the superconducting state, which is related to the thermo-
dynamic critical field by Ec = −μ0 Hc2 /2, is determined in the BCS theory to be
Fig. 7.7 Ground state and E

excited states of a supercon-
ductor Excited states
BCS ground state

1 2
Ec = − D(EF ) = −2(h̄ωD ) D(EF ) exp −
2 2
. (7.19)
2 V0 D(EF )
Note that Ec and have a similar form to the binding energy δ calculated by Cooper
for an interacting electron pair in a gas of independent electrons [see (7.17)]. In both
cases, we obtain an exponential function of the reciprocal of the coupling. The ab-
sence of resistivity and the existence of the Meissner effect can be understood in
this quantum ground state.
7.3.2 Excited States in the BCS Theory
At zero temperature, the BCS theory can determine the spectrum of excited states of
the system. For all practical purposes, these can be treated as independent ‘normal’
electron states. The density of excited states is given by
⎧
⎪ E − EF
⎨ D(EF ) 1/2 , for E > EF + ,
Ds (E) = (E − EF )2 − 2 (7.20)
⎪
⎩
0, for EF < E < EF + .
This function is shown in Fig. 7.8.

Note that, while the excited states can be considered as independent electron
states, the same cannot be said for the ground state. The latter is a collective state of
the electron system in which the electrons are strongly correlated, since it is made up
of electron pairs. In the schematic representation of Fig. 7.7, electrons can only be
thermally excited out of the ground state in pairs, so the gap is indeed equal to 2.
Regarding the excited states, we may consider that the gap appears by shifting the
set of accessible states between EF and EF + in the normal metal beyond EF + .
The BCS calculation can be carried out as a function of temperature. It transpires
that, not only is the population of excited states given by a Fermi factor, but the gap
closes as the temperature increases, and vanishes at Tc , guaranteeing continuity with
Fig. 7.8 Density of states of E

excited levels. The vertical
scale is much expanded, so Superconductor
the density of states D(E)
in the normal state appears
constant D(EF ). The states
that can be reached in the EF + Δ
normal metal (light shad-
ing) are shifted beyond the
gap (dark shading) in the Normal state
superconductor
EF
D(EF ) D(E )
the normal state. The critical temperature is found to be directly related to the value
of at zero temperature by
2(0) = 3.52kB Tc . (7.21)
To begin with we discuss experiments to check the prediction of the energy spectrum
of the superconductor, before considering the validity of the theoretical determina-
tion of Tc on the basis of (7.21).
Question 7.1. Can BCS theory explain the experimental values of the critical field in
Fig. 5.3? What can be deduced about the coefficient γ of the electronic specific heat for
these metals? What about the Pauli limit for Hc2 ?
7.3.3 Electron Tunneling Determination of the Gap
We have seen that, in order to observe the gap and the excited states in the super-
conducting state, we could use measurements of specific heat or reflectivity in the
microwave range. The former only inform us about an integral of the density of
thermally occupied states, while the latter are difficult to set up experimentally, be-
cause measurements must be made as a function of the microwave frequency. To
do a detailed study of the energy spectrum of the excited states, tunneling effect
measurements prove to be extremely effective.
We saw in Chap. 5 that tunneling of superconducting electron pairs through a
thin insulating barrier gives rise to the Josephson effect, exploited to make SQUIDs.
Here we consider the tunneling of electrons between a normal metal and a super-
conductor. This kind of junction can be made by depositing thin films as shown in
Fig. 5.11. In this case, one of the layers is a normal metal, while the other becomes
superconducting for T < Tc .
As we saw in Chap. 3 regarding scanning tunneling microscopy, the current–
voltage characteristic of a tunnel junction can inform about the density of states of
the two metals as a function of energy over an energy range eV determined by the
tunnel voltage V. In general, metals 1 and 2 are such that D1 (E) and D2 (E) vary
little with the energy in their normal state, and according to (3.74),
jNN = AT D1 (EF )D2 (EF )eV , (7.22)
which is an Ohmic characteristic, independent of temperature.

Consider what happens if metal 2 becomes superconducting for T Tc . Since
the excited levels are separated by from the ground state, no electron can be
transferred to the superconductor as long as eV < , because there is no accessi-
ble state. The tunneling current is therefore zero and rises suddenly when eV > ,
because the number of accessible states also goes up suddenly (see Fig. 7.9). The
current–voltage characteristic j(V) can thus be used to determine for T Tc . It
only became possible to carry out such experiments in 1960, when the techniques
for depositing thin films were finally developed.
The tunneling conductance dj/dV accurately gives the function D2s (EF + eV)
rounded off by the temperature dependence of the Fermi factor [see (3.73)]. This
was achieved with great accuracy by Giaver (Nobel prize with B. Josephson in
1973), who showed in 1962 that the tunneling characteristics agree extremely well
with those deduced from the BCS density of states (7.20) (see Fig. 7.10). The tem-
perature dependence of as measured by tunneling also agrees extremely well with
the predictions of the BCS theory (see Fig. 7.11).
Question 7.2. The Abrikosov vortex lattice imaged on p. 176 was observed by scanning
tunneling microscopy at the surface of a single crystal of the compound NbSe2 , for which
Tc = 7.2 K. The colour code represents the value of the tunneling conductance at each point
of the crystal surface for V = 1 mV. What characteristic length is represented by the diameter
of the spots in the image? What observations of this image might seem unexpected? The
detailed modifications of the tunneling effect in the mixed state4 will help to understand
some of these observations.
E
j
1 2 j NN
j NS
EF
−eV
Δ
D 2s (E + eV ) Δ /e V
D 1 (E )
Fig. 7.9 Tunneling between a normal metal (1) and a superconductor (2) with a positive applied
voltage V. An electron can be injected by tunneling into an excited state of the superconductor if
eV > . The current–voltage characteristic j(V) is used to determine
4 Examined in Problem 13: Magnetisation of a Type II Superconductor.

Fig. 7.10 Tunneling conduc-

tance as a function of V for 3
an Mg/MgO/Sn junction at
T = 0.3 K. The experimental T = 0.3 K
(dI/dV) SN /(dI/dV) NN
results (continuous curve) Δ = 0.6 meV
agree extremely well with the 2 Sn/MgO/Mg
theoretical prediction from the
BCS density of states (dashed
curve). From Giaver, I., Hart, 1
H.R., Megerle, K.: Phys. Rev.
126, 941 (1962). With the
kind agreement of the Amer-
ican Physical Society (APS).
http://link.aps.org/doi/10.1103 0 1 2 3
/PhysRev.126.941 eV/Δ
Fig. 7.11 Temperature de-

1 Pb
pendence of the gap energy
for lead, measured by tunnel-
ing. These results agree well 0.8
with the prediction of the BCS
Δ (T )/ Δ (0)
theory (continuous curve). 0.6

Adapted from Gasparovic,
R.E., Taylor, B.N., Eck, R.E.: 0.4
Solid State Com. 4, 59 (1966)
Elsevier copyright 0.2
0
0 0.2 0.4 0.6 0.8 1
T/Tc
The BCS theory shows that, when V < 0, the behaviour is totally symmetrical.
As soon as e|V| > , an electron can be transferred to the normal metal, leaving
an excitation equivalent to a hole in the superconductor, i.e., an electron which no
longer has a partner. This is identical to the excitation corresponding to a solitary
injected electron, so we obtain an analogous representation to the one obtained for a
Fig. 7.12 Electron tunneling. Δ

Description of the excited
states of a superconductor us- Δ
ing a semiconductor model −eV
(7.23). An electron is trans- EF
ferred to the normal metal
when e|V| > , releasing an
excited state in the supercon-
ductor that resembles a hole D 1 (E ) D 2s (E + eV )
semiconductor (see Fig. 7.12), for which Ds (E) is symmetrical with respect to the
Fermi level:
⎧
⎪ |E − EF |
⎨ D(EF ) 1/2 , for |E − EF | > ,
Ds (E) = (E − EF )2 − 2 (7.23)
⎪
⎩
=0, for |E − EF | < .
In this representation, the superconducting ground state is not described. Only the
excited states are considered.
The superconducting state cannot be described as a metallic state of independent electrons in
which a gap appears at Tc and leads to a new one-electron state described by the semiconductor-
type band structure of Fig. 7.12. The point is that the energy involved in the appearance of such
a band gap would correspond to a transfer below EF − of the normal electron states which
would lie between EF − and EF . It can be shown that the drop in energy associated with such
a transformation is given by
EF C
Ec∗ = E D(E) − D(EF ) dE = − D(EF )2 ,
−∞ 2
where C > 1. The drop in energy in the superconducting state given by (7.19) is smaller as it
involves the interaction energy between electrons in the BCS ground state, which is not taken
into account in this independent electron band gap.
A metal–insulator transition that can be described by the appearance of such a band gap
occurs in some metals when a lattice distortion leads to a doubling of the primitive cell. This
kind of transition, called a Peierls transition for a metal in one dimension, can only occur if the
gain in electron energy is greater than the cost in elastic energy needed to cause the distortion.5
7.3.4 Critical Temperature and BCS Theory
As we have just seen, the validity of the BCS solution is well attested by experimen-
tal determination of the gap and its temperature dependence. But can the theory ex-
plain the measured values of Tc ? Note that (7.21), which tells us that = 1.76kB Tc ,
is also perfectly satisfied for many metals. Moreover, the fact that Tc is propor-
tional to h̄ωD obviously leads back to the isotope effect, insofar as the Debye fre-
quency goes as M −1/2 . As we said earlier, it is this observation that suggests that
the electron–phonon interaction underpins the attractive interaction in conventional
metal superconductors. Combining (7.18) and (7.21), the BCS results thus imply

1
kB Tc = 1.14h̄ωD exp − , (7.24)
V0 D(EF )
in terms of the parameters ωD , V0 , and D(EF ) which characterise the normal metal.
The last two parameters appear inside the exponential in the expression for Tc ,
which explains the differences in order of magnitude of Tc between different metals.
5 Such a case is described in Problem 6: One-Dimensional TTF-TCNQ Compounds.

However, any error in these quantities, which must be determined by integrating

over the Fermi surface, will lead to a large error in the estimate for Tc . In fact there
is not a single metal for which it has been possible to determine Tc directly by cal-
culation before obtaining its value experimentally. It is generally easier to justify the
measured values a posteriori!
In order not to seem too negative about this point, we should nevertheless note
that the various factors appearing in the expression for and hence also in the
expression for Tc can provide a qualitative understanding of the criteria for the ex-
istence of superconductivity, and also the differences in the values of Tc across the
periodic table.
To begin with, the coupling between electrons is only attractive if the Coulomb
repulsion between them is weak compared with the electron–phonon coupling. The
latter will be stronger the more the electrons feel the crystal potential due to the ions.
Now we know that the potential of the crystal lattice can only significantly modify
the electronic band structure as compared with a free electron model in those metals
for which the Fermi surface intersects one or more Brillouin zones. In this case,
a modification of the energies of the electron states is induced close to vectors k
corresponding to these intersections (see Chap. 3).
Such an effect will not generally occur for monovalent metals, for which the
Fermi surface is included within the first Brillouin zone and is generally close to
spherical. In such metals, the electron–phonon interaction is weak and so the term
associated with phonons in the resistivity is rather small. The fact that superconduc-
tivity is not detected in such systems therefore suggests that the electron–phonon
coupling is not sufficient to dominate the Coulomb repulsion.
On the other hand, Tc is higher for polyvalent metals, for which the Fermi sur-
face intersects many Brillouin zones, and for which the electron–phonon coupling
is therefore strong, which corresponds to high resistivity in the metallic state. This
therefore perfectly explains the fact that good superconductors are often rather poor
conductors.
Finally, the maximal values of Tc are obtained for the transition metals in the
4d and 5d series, for which the two terms V0 and D(EF ) are large, which is indeed
expected on the basis of (7.24). Indeed, in these systems, the overlaps between the
d orbitals are relatively small, and this leads to narrow bands, and hence high elec-
tronic densities of states.
To end this section, note that the fact that the magnetic metals are not supercon-
ducting suggests some kind of incompatibility between magnetism and supercon-
ductivity.
7.3.5 Coherence Length
We have seen that the wave function of the Cooper pair is built up from plane wave
states taken over an energy range δ above EF . The corresponding plane wave packet
has wave vectors with magnitudes between kF and kF + k, where
(h̄2 k2 /2m0 ) = h̄2 kF k/m0 δ .
Corresponding to this extension k = δ/h̄vF in momentum space is an extension in

position space given by r = 1/k = h̄vF /δ. This spatial extent of the Cooper pair
therefore provides an estimate of the coherence length ξ . The more exact BCS cal-
culation leads to a similar form for ξ , in which δ is replaced by the superconducting
gap , that is
h̄vF
ξ= . (7.25)
π
This is valid for a pure metal system. Since decreases with temperature, it follows
that ξ increases and diverges at the critical temperature.
Dirty Superconductors
In reality, the two lengths λ and ξ depend on the properties of the normal state of the metal.
In particular, if the metal is not pure, i.e., if the mean free path is short in the normal state, the
coherence length gets shorter, while the penetration depth increases. These changes in λ and
ξ take place without affecting the microscopic properties of the superconducting state, i.e., the
condensation energy and gap. Introducing impurities in controlled amounts into a metal thus
leads to an increase in the ratio λ/ξ , and it is then possible to move gradually from type I to
type II superconductivity. We shall see an experimental example in Chap. 10. Note also that the
temperature dependences of λ and ξ are similar and that the ratio κ = λ/ξ does not therefore
depend on temperature.
Superconductivity and Bose Condensation
In the known superconductors, the fact that the electron pairs are in spin singlet states is some-
times said to mean that the pairs behave as bosons. In simple metals, superconductivity is then
likened to Bose condensation, a phenomenon predicted at low temperatures for systems of
non-interacting bosons (see any textbook on statistical physics). Superfluidity, that is, the ab-
sence of viscosity at low temperatures, which is observed in the boson 4 He, is also similar to
Bose condensation. However, strictly speaking, one cannot really speak of Bose condensation
in either case, because the bosons under consideration here interact rather strongly together.
This is already the case with 4 He, owing to the significant Van der Waals forces between the
atoms in the liquid phase. As far as superconductivity in metals is concerned, there is not re-
ally any transition between a state in which the bosons are non-interacting and a condensed
boson state. The electron pairs identified with bosons in this case are just not present in the
metal at temperatures above Tc , and the pairing only occurs at the superconducting transition.
In addition, in the BCS ground state, the electron pairs are intimately ‘mixed’, the average
distance between paired electrons being greater than the interatomic distance. The comparison
with Bose condensation is not therefore justified. Bose condensation in dilute gases, where the
atoms are only weakly interacting, has been observed (for the first time in 1995, with rubidium
gas). These gases are confined by laser beams, which permit to considerably reduce the kinetic
energy of the atoms and thereby cool them to very low temperatures (a few tens of nK).
7.4 High-Tc Superconductors 223
7.4 High-T c Superconductors
By the 1970s, BCS theory seemed to provide a satisfactory explanation of supercon-

ductivity, while theoretical estimates suggested that the maximal value of Tc would
be intrinsically limited to 25 K. The only effort in those days was therefore to try
to reach this upper bound in technically usable compounds, so that liquid hydrogen
could be used as cryogenic fluid, with a liquefaction temperature of T = 20 K at
standard pressure.
The discovery by Bednorz and Müller in 1986 of copper oxide-based materials,
the cuprates, with critical temperatures Tc ≈ 40 K, was a genuine shock to the sci-
entific community, and they were awarded the Nobel prize in 1988, an exceptionally
short time after their discovery. Since then, the quest for new superconductors has
snowballed and further compounds of the same family have been discovered, fol-
lowed by other families with values of Tc exceeding all expectations. In many cases,
the applicability of the BCS theory was put in question. Here we shall review some
aspects of these discoveries.
7.4.1 Cuprates
In the year following the discovery of the compound La2−x Bax CuO4 by Bendorz
and Müller, many other superconducting compounds of the same family were found,
the current record being 135 K at standard pressure and ≈150 K at high pressure.
Since Tc has gone above the liquefaction temperature of nitrogen at standard pres-
sure (77 K), the prospects for applications have increased significantly. But the fun-
damental questions raised by this discovery are themselves exceedingly important.
Indeed, these new high-Tc superconductors are lamellar compounds comprising
copper oxide sheets, separated by planes of transition ions or other oxides. We see
immediately that we are dealing with quite different materials from the conven-
tional elemental metals discussed so far. These are rather ionic crystals, like many
other transition metal oxides, which are generally insulators. The very fact that some
cuprates are metallic is not obvious!
One novel feature of these compounds is of chemical origins, because some of
these planes play the role of charge reservoirs. The oxygen content can often be
modified by simple thermal treatment. For example, the archetypal cuprate super-
conductor YBa2 Cu3 O7 , with Tc = 92 K, comprises two types of copper oxide planes
(see Fig. 7.13). The CuO2 planes at the center of the unit cell, separated by a layer
of Y, are chemically highly stable. In contrast, the oxygen content of the base plane
of the cell, made up of linear CuO chains, can be reduced by thermal treatment in
vacuum.
These CuO chains can be totally deoxidised to obtain a composition YBa2 Cu3 O6 .
This compound is insulating, its valence band being associated with the bands
formed from the Cu(3d)–O(2p) orbitals of the CuO2 planes. To increase the oxygen
Fig. 7.13 Crystal structure of

YBa2 Cu3 O7 (Tc = 92 K). The
conducting planes are the cen-
tral CuO2 planes, which cor-
respond to the Cu(2) sites. The Ba
base planes containing CuO CuO
chains play the role of charge chains
reservoirs. By partially sup-
pressing the O(4) oxygens of
the CuO chains, the number CuO 2
Y
of carriers (holes) is reduced planes
in the CuO2 planes. The com- Cu(2)
pound YBa2 Cu3 O6 is an anti-
ferromagnetic insulator O(2) O(3)
c O(1)
b Ba
a
Cu(1)
O(4)
content and thereby return to the compound YBa2 Cu3 O7 , the idea is to introduce
electronegative O2− ions into the base plane. These extra oxygen ions capture elec-
trons from the CuO2 planes and thereby produce holes in the Cu(3d)–O(2p) valence
band. This energy level system is valid for practically all the cuprate superconduc-
tors. The hole density nh can thus be continuously varied in the CuO2 planes of these
materials by changing their oxygen content, and it is observed that Tc is maximum
for nh ∼ 0.15 holes per Cu in the CuO2 planes.
The density of free carriers which determines the metallic behaviour of these
superconducting compounds is therefore much lower than in metal alloys. Their
Fermi energies, and hence also the value of vF , are both low. Consequently, the
fact that Tc and hence are high implies immediately, according to (7.25), that
the coherence length ξ of the material is very low. The superconductivity of the
cuprates is thus always type II. Note that the conductivity and superconductivity
are both highly anisotropic. Taken together, these properties lead to novel behaviour
of the vortex system, which has a different phase diagram to conventional type II
superconductors (see Chap. 10). The low value of ξ (∼ 20 Å) also leads to signficant
practical difficulties when making junctions between superconductors, which must
be structurally perfect on the scale of ξ .
Are These Superconductors Exotic?
Since the Debye temperature is of the same order of magnitude as one finds in metal alloys,
and since the density of states is lower, it seems hard to imagine on the basis of (7.24) that the
electron–phonon coupling can be the fundamental cause for such a high value of Tc . Various
approximations of the BCS calculation need to be reexamined, including the weak coupling
approximation V0 D(EF ) 1. Moreover, in a 2D metal, D(E) exhibits Van Hove singularities,
so that it cannot be treated as constant over the energy range h̄ωD , and the band structure must
be fully taken into consideration.
7.4 High-Tc Superconductors 225
Furthermore, the CuO2 planes have the property of being magnetic when not doped, the Cu
ions (3d 9 ) carrying spin 1/2. This is true for YBa2 Cu3 O6 . In this case, as we shall see in the
following chapters, it seems difficult to neglect the interactions between electrons at the copper
sites, which are responsible for the magnetic behaviour. Many experiments have shown that
the cuprates have strange magnetic properties in their metallic state. Various possibilities have
been invoked to explain their high-Tc superconductivity. In some models, there is more talk of
Bose condensation, because some authors consider that the pairs may actually exist above Tc .
One key experimental observation was the demonstration that, in the cuprates, the super-
conductor gap is not uniform in momentum space. It vanishes in the so-called nodal directions
(±π/a, ±π/a) of the wave vector in the Brillouin zone, and is maximum for the antinodal
directions (0, ±π/a). The gap thus has a similar symmetry to the d atomic wave functions,
and changes sign under a rotation through 90◦ . This special d-type symmetry of the super-
conductor gap is today the strongest evidence for the exotic origin of the superconductivity of
the cuprates. It leads to the novel properties of the Josephson junctions and also, in specific
configurations of the junctions, to the existence of a vortex with a flux 0 /2, as shown in the
experiment illustrated in the figure on p. 296.
For many physicists, this observation provides strong evidence that electron pairing has
magnetic origins. Despite intense effort since the cuprates were discovered, which has gen-
erated considerable progress on both the experimental and theoretical fronts, there is still no
general agreement within the scientific community. This highlights the difficulty of the prob-
lem, which remains one of the main challenges of solid state physics.
7.4.2 Other Families of Superconductors
Following the discovery of the cuprates, new families of compounds with values of
Tc above 25 K were identified, including alkaline earth compounds with fullerene
C60 , discussed in Chap. 2 (see Fig. 2.10), or the compound MgB2 .6 In the latter
two cases, electron–phonon coupling and the BCS theory seem applicable, but with
several special features to explain the high values of the critical temperature Tc .
On the other hand, since 1980, many other families of superconducting com-
pounds with low Tc have been investigated. Some 1D or 2D organic salts, called
Bechgaard salts,7 and compounds of metals in the lanthanide series, called heavy
fermion compounds, seem also to be exotic superconductors, like the cuprates.
More recently, lamellar compounds of the transition elements, such as the ruthenate
Sr2 RuO4 , with the electronic structure illustrated on the cover page of this book,
or the cobaltates Nax CoO2 , have been attracting much interest. Regarding the lat-
ter, the cobalt atoms in the CoO2 planes are arranged in a hexagonal structure, and
doping is obtained by electron transfer from intercalated Na atoms. In 2003, it was
found that these metallic compounds become superconducting, with Tc = 4.5 K,
only when the CoO2 planes are pushed apart by insertion of H2 O between the Na
and CoO2 planes.
6See Problem 16: Magnesium Diboride A New Superconductor?

7Those have similar molecular structure to that displayed in Problem 6: One-Dimensional
TTF-TCNQ Compounds.
Finally, in 2008, new families of lamellar compounds comprising square planes

of Fe, e.g., the Fe pnictides, where the FeAs2 films are separated by planes which
may provide electron or hole charge transfer, turn out to be superconducting with a
maximal Tc of 55 K! The physical chemistry of 2D transition metal compounds will
surely continue to deliver quite unexpected superconductivity properties.
Magnetism plays a key role in the electronic structure of these compounds, as for
the cuprates, as we shall see in Chap. 8 after introducing some elementary notions
of magnetism. In most of these compounds, a magnetic origin is often suspected for
the coupling between electrons that leads to pairing.
7.5 Summary
Many metallic elements are superconducting, with values of Tc ranging over sev-
eral orders of magnitude. The best conductors, in particular the noble metals, are
not. The discovery of the isotope effect, i.e., the dependence of Tc on the atomic
mass, shows that the ions are involved in the processes stabilising the supercon-
ducting state. The Coulomb interaction between electrons and ions, i.e., with their
vibrational degrees of freedom known as phonons, leads to an indirect attractive
interaction between electrons.
Specific heat and microwave absorption experiments reveal the existence of a
band gap above the ground state of the superconductor. The width of the band gap
corresponds to the energy 2 that must be provided to break up an electron pair and
make them ‘normal’.
Under what conditions can an attraction between two electrons lead to a bound
state? In the case of a 3D insulator, two electrons in the conduction band will only
form a bound state if the attractive potential exceeds a certain minimum value. On
the other hand, in a metal, two electrons in attractive interaction at the Fermi level
lead to a bound state whatever the strength V0 of the interaction. This result due to
Cooper suggests an instability of the electron ensemble near the Fermi level at low
temperatures. The BCS theory shows that this new ground state of the electron gas
comprises spin singlet superconducting pairs. This ground state is separated from
the excited states by an energy

1
= 2h̄ωD exp − = 1.76kB Tc .
V0 D(EF )
The excited states can be observed by tunnel effect in N-I-S junctions, where elec-
trons or holes are injected into the superconductor from the normal metal. The par-
ticles injected in this way propagate in states separated by more than from the
Fermi level. The density of states of these excitations can be described by a semi-
conductor model with
⎧
⎪ |E − EF |
⎨ D(EF ) 1/2 , for |E − EF | > ,
Ds (E) = (E − EF )2 − 2
⎪
⎩
0, for |E − EF | < .
The BCS theory can explain all the properties of the conventional metal supercon-
ductors described so far, and in particular the broad range of values of Tc and the
fact that the good metals are not superconducting. But having understood the su-
perconductivity of metals and simple alloys, new superconducting compounds were
discovered which apparently cannot be described as a form of superconductivity in
which phonons are responsible for electron pairing.
Among these, many compounds have a lamellar atomic or molecular structure
and electronic properties with distinct 2D features. Lamellar copper oxides, called
cuprates, are without doubt the materials that have done most to shake the scientific
community, since superconductivity occurs with the highest values of Tc yet ob-
tained (as high as 150 K). Beyond the applications that become feasible with such
high critical temperatures and the high associated critical fields, there is consider-
able debate over the origins of this kind of superconductivity.
Many new superconducting compounds have been discovered since the 1980s,
and in many cases the origin of the superconductivity remains poorly understood. It
will be a great scientific challenge to push our understanding beyond the BCS the-
ory. Industrial applications of this physical phenomenon, something nobody doubts
in the long term, will require a very tight control on the processes for synthesising
these materials on a large scale.
Puzzle on Page 202
The pictures should lead you to the following sentence:
Cooper pairs are not independent bosons ,
Cooper - pears - R - knot - in - D - pendent - bows - onze (French) .

In the superconducting state, the electron–phonon interaction is responsible for the
pairing of the electrons. In the macroscopic quantum state of the BCS theory, many
pairs condense out at the same time, and the characteristic size of a pair, as deter-
mined by the coherence length ξ , is much greater than the average distance between
electrons. In this sense, the pairs of dimension ξ are totally entangled with one an-
other. While an isolated Cooper pair might possibly be considered as a boson, the
Cooper pairs in the BCS state are in a highly correlated state, and therefore can-
not be considered as non-interacting bosons. Note that the pairing of the Cooper
pairs only occurs below Tc , and that there are no states resembling bosons above
the critical temperature Tc .
This situation should be contrasted with Bose–Einstein condensation, where non-
interacting bosons exist at all temperatures. In their ground state, these independent
bosons are in the same quantum state. The occupation number of this ground state
becomes an extensive quantity below the Bose condensation temperature TB . The
bosons are thus independent particles in the quantum sense at all temperatures, but
move in a quantum mechanically coherent way for T < TB .
Note: In the figure on the first page of Chap. 5, Claire and David are looking for
explanations of superconductivity. David is trying to understand how the elastic
deformation of the lattice can induce an attractive interaction between electrons,
responsible for the formation of Cooper pairs. Claire is more intrigued by the col-
lective phenomenon leading to quantum coherence over macroscopic distances. The
electrons seem to her to execute a coordinated danse, rather like square dancers.
Question 7.1
The condensation energy of the superconducting state at zero temperature is given

by the BCS theory as
1 1
Gs (0, 0) − Gn (0, 0) = − D(EF )2 = − μ0 Hc2 .
2 2
This suggests therefore that the temperature dependence of Hc may be related to the
temperature dependence (T) of the gap, these two quantities vanishing at T = Tc .
At T = 0, this leads to
μ0 Hc (0) 1/2
= μ0 D(EF ) .

Now according to the BCS theory,
= 1.76kB Tc .
Using the relation between the electronic density of states at the Fermi level and the
electronic specific heat, viz.,
2π 2
γ= kB D(EF ) ,
3
we obtain
2
μ0 Hc (0)
= 0.47μ0 γ .
Tc
The experimental curves are approximately homothetic, with an almost constant

ratio Hc (0)/Tc , which corresponds to an electronic specific heat and density of states
that barely depend on the metal. This is indeed the case for the metals Al, In, Sn, Hg,
and Pb considered in Fig. 5.3. In fact, the above relation applies perfectly with the
experimental values of γ . The 3d and 4d transition metals have a higher density of
states, leading to a higher ratio between the field Hc (0) and the critical temperature.
The Pauli limit for Hc2 determined in Question 6.3, viz.,

√
Hcm2 < Hc / χP ,
can be found by taking the value of Hc (0) given by the BCS theory:
1/2
μ0 Hc (0) = μ0 D(EF ) ,
and
χP = 2μ0 μ2B D(EF ) .
We immediately obtain
√
μ0 μB Hcm2 = / 2 ,
which yields, using = 1.76kB Tc given by BCS,
μ0 Hcm2 = 1.84Tc (tesla/K) .
For example, for V3 Ga, which is a superconducting compound8 at the relatively

high Tc = 14 K, this would give the Pauli limit μ0 Hcm2 = 25.7 tesla.
It is clear from Fig. 7.14 that the experimental curve for Hc2 (T) in V3 Ga is not
parabolic. For the quadratic shape, the tangent at T = Tc cuts the vertical axis at
2Hc (0). The tangent to the experimental curve would give μ0 Hc∗2 (0) = 28.5 tesla,
exp
which is higher than the experimental value μ0 Hc2 (0) ∼ 21 tesla. The parabola
Hc2 (tesla)
30 Hc∗2 Parabolic extrapolation
20
10
Fig. 7.14 Experimental curve

for Hc2 (T) in V3 Ga, show-
ing that it saturates at low
temperature with respect to a 0 2 4 6 8 10 12 14 16
parabolic extrapolation Temperature (K)
8 See Problem 14: Electronic Structure and Superconductivity of V3 Si.

fitted to the experimental results for T ∼ Tc is shown in Fig. 7.14. This saturation of
the experimental curve Hc2 (T) is undoubtedly related to the existence of the Pauli
limit.
Question 7.2
The image on the first page of Chap. 6 was obtained at the surface of NbSe2 by
scanning tunneling microscopy for V = 1 meV. The gap predicted by the BCS theory
is
= 1.76kB Tc ≈ 1.05 meV , for Tc = 7.2 K .
We thus see that the tunnel conductance must be close to zero in the superconducting
state since the characteristic is taken for V slightly less than the value of the gap. This
therefore corresponds to the blue continuum background. But in the core region of
the vortices, the tunnel conductance increases and tends to the value for the normal
metal. The white spots, of diameter about 7 mm, are thus close in size to 2ξ , and we
obtain
7 3, 000
ξ= × Å = 80 Å .
2 130
Chapter 8
Magnetism of Insulators
Contents
8.1 Magnetic Behaviour of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

8.2 Magnetism of Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.2.1 Hydrogen Atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.2.2 Multielectron Atoms with Filled Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.2.3 An Atom with a Partially Filled Shell: Carbon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.2.4 Atoms with Partially Filled Shells: Hund Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
8.3 Paramagnetism of an Ensemble of Isolated Ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
8.4 Ordered Magnetic States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.4.1 Interatomic Exchange Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8.4.2 Heisenberg Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
8.4.3 Ferromagnetism: Molecular Field Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 249
8.5 Antiferromagnetism and Ferrimagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
8.6 From Insulator Magnetism to Metallic Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
8.6.1 Mott–Hubbard Insulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
8.6.2 Mott Transition and Doped Mott–Hubbard Insulators . . . . . . . . . . . . . . . . . . . . . . . 258
8.6.3 Magnetism and Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
8.6.4 Metallicity and Magnetism in a Band Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
8.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
232 8 Magnetism of Insulators
An important means of information storage is to record data on magnetic media. The information
recorded on a digital audio tape (DAT) can be visualised by magnetic force microscopy (MFM, see
Chap. 10). On this tape, the magnetisation of the grains in a magnetic film deposited on a polymer
substrate has been imposed by a read head that moves much more quickly than the winding rate of
the tape, and which can write the successive tracks on the tape. Here a 1 or a 0 corresponding to an
upward magnetisation (black) or a downward magnetisation (white) is imposed on a region of area
0.25 × 10 μm2 . Data is written onto hard disks in a similar way, with a data storage density that
is currently 400 times higher (1 bit on 25 × 250 nm2 ). Image courtesy of Miltat, J., Thiaville, A.:
Laboratoire de Physique des Solides. Orsay, France
8 Magnetism of Insulators 233
Magnetic fields are understood by classical electromagnetism as resulting from

macroscopic currents circulating in conductors. However, this description does not
help us to understand the existence of magnetic fields spontaneously induced by
materials like iron or cobalt when there are no macroscopic currents imposed by
external sources. Only quantum mechanics can explain electronic states correspond-
ing to permanent currents at the microscopic scale. We have just seen an exam-
ple in the case of superconductors. Likewise, the electron orbital angular momen-
tum and spin are the purely quantum entities which underlie the magnetism of
materials.1
In this chapter we begin by explaining the quantum origins of spontaneous mag-
netic moments in materials, and in particular ferromagnetism, using a model which
applies to insulating materials for which the electron magnetic moments interact via
an exchange interaction between near neighbours. We shall see in Chap. 9 that the
presence of magnetic domains in the materials around us involves a very low energy
compared with the exchange interaction responsible for magnetism. These low en-
ergy phenomena nevertheless govern most magnetic material behaviours which are
of great importance for their technical uses. In Chap. 10, we shall discuss experi-
mental methods used for measurements and imaging of magnetic materials. Chap-
ter 11 deals with the response of magnetic materials to alternating fields, on the
basis of which we can discuss magnetic resonance methods. Finally, in Chap. 12,
we consider the thermodynamic properties of magnetic materials.
Note that the methods used in Chaps. 1 and 3 to describe electronic bands in
solids, and in particular the approximation which consists in only considering the
averaged interaction between electrons, led to the consideration of only two types
of solid: metals exhibiting Pauli magnetic susceptibility, and non-magnetic insula-
tors. There is thus no room for spontaneous magnetism in such a description. The
magnetic properties of superconductors stem from the existence of an attractive in-
teraction between electrons which is induced by the phonons of the atomic lattice,
and this attractive interaction leads to a ground state in which the electrons are no
longer independent, and where persistent currents can be established on a micro-
scopic scale. To be able to describe the magnetic states of matter, we must therefore
take into account the interactions between electrons.
After a brief review of the various kinds of magnetic behaviour observed in
materials (Sect. 8.1), we shall illustrate how interactions between electrons (direct
Coulomb interactions or indirect interactions induced by the Pauli principle) explain
atomic magnetism, i.e., the existence of atoms with an electronic magnetic moment
(Sect. 8.2). We then investigate how ions carrying independent electronic magnetic
moments lead to Curie magnetism (Sect. 8.3). In Sect. 8.4, we shall see how the
1 Note that atomic nuclei also carry spins, and have magnetic moments that give rise to Curie
magnetism. However, the nuclear magnetic moments are very small compared with that of the
electron and make an altogether negligible contribution to the magnetic properties of the solid,
which are thus entirely electronic in origin. The nuclear magnetism can only be directly detected
by a magnetometer in quite exceptional circumstances (materials with weak electronic magnetism
at very low temperatures), but is easier to observe using resonance methods (see Chap. 11).
overlap between electron atomic orbitals induces a coupling between the electronic
magnetic moments that is responsible for ordered magnetic states observed at suf-
ficiently low temperatures. A minimal approach for dealing with the interactions
between electrons will be outlined in Sect. 8.6 in order to understand how magnetic
and metallic behaviour might coexist.
8.1 Magnetic Behaviour of Solids
The vast majority of natural compounds exhibit very weak magnetism. A highly sen-
sitive magnetometer is needed to measure the magnetic susceptibility in this case.
Measurements show that these compounds are generally diamagnetic. This arises
because most materials are insulators made up of ions (or molecules) with saturated
electronic shells whose orbitals barely overlap with those of nearest neighbours.
In the models of electronic structure in Chaps. 1 and 3, the electronic states differ
little from those of isolated ions (or molecules) and give rise to similar magnetic
behaviour to that observed in the corresponding liquid or gaseous states. For ions
(or molecules) with saturated electronic shells, the electrons occupy electronic lev-
els whose spins and orbital angular momenta cancel. In this case, the only source of
magnetism lies in modifications to the electronic orbitals when an external field is
applied. The classical analogy between an orbital and an electronic current can be
used to understand that the current induced by applying an external field will oppose
this field, simply from Lenz’s law, and this situation corresponds to a diamagnetic
magnetisation. The diamagnetic susceptibility is small and temperature independent
(see Fig. 8.1), since it is a quantum mechanical property of the electronic ground
state. Indeed, the excited states are hardly populated thermally, because they are
generally several eV away. This diamagnetism (χ̃d −10−5 ) is negligible com-
pared with values for superconductors (χ̃ = −1). Note that χ̃d is directly related
to the size of the electronic orbital in the direction perpendicular to the applied
field, and will be particularly relevant for electrons involved in molecular states that
χ Ordered
magnetic
phases
Curie
paramagnetism
Metal
Pauli paramagnetism
Fig. 8.1 Schematic represen-
tation of the magnetic sus-
ceptibilities of various types
T
of substance as a function of
temperature Diamagnetic insulator
8.1 Magnetic Behaviour of Solids 235
extend over several atoms. This happens, for example, in the benzene rings of aro-
matic molecules.
In metals, the inner electron shells also overlap very slightly and lead to a similar
contribution. In addition to this diamagnetic susceptibility from the inner shells,
there is another due to the electrons in the partially filled bands which exhibit Pauli
spin paramagnetism [see (3.66)]:
χP = 2DΩ (EF )μ2B , (8.1)
which is also almost independent of the temperature insofar as the conduction

band is broad and has no fine structure on the scale kB T 25 meV. If the den-
sity of states DΩ (EF ) is small, the total resulting susceptibility χ = χd + χP re-
mains diamagnetic, as happens for example for copper. For metals in the transi-
tion series, e.g., the 4d series, the bands are narrow and the densities of states
higher than for the noble metals. The resulting magnetic susceptibility is then
paramagnetic.
The most significant magnetic effects arise in solids containing ions whose
atomic shells are not saturated, and which carry a magnetic moment. In this case,
strong paramagnetism is observed, increasing according to χ C/T as the tem-
perature decreases. This is Curie paramagnetism, which arises from the removal of
degeneracy of the electron spin states when a field is applied. This paramagnetism
of independent magnetic moments corresponds to a reduction in the statistical dis-
order of the orientations of the moments in the presence of a field. It is also of purely
atomic origin.
However, it is found experimentally that, in many cases, the observed param-
agnetism deviates significantly from this law, especially at low temperatures. In
particular, at low temperatures, spontaneous magnetisation is often observed, and
the solid retains a nonzero magnetic moment even if the applied magnetic field is
switched off:
lim M(Ha ) ≡ M0 = 0 . (8.2)

Ha →0
2
Fe
M (10 3 kA/m)
Ni
Fig. 8.2 Spontaneous mo-
ments of Ni and Fe as a
function of temperature. 0
0 500 1000
These are metals with high
Curie temperature TC Temperature (K)
The existence of a state with spontaneous magnetisation is due at least in part to

an ordering of the magnetic moments. We have here a collective phenomenon of
the system of electronic moments induced by interactions between the moments
of neighbouring atoms. This collective magnetism only occurs in the solid phase.
The best known case is no doubt ferromagnetism, which corresponds to a parallel
alignment of all the spins, at least on average. The typical temperature dependence
of the spontaneous magnetisation M0 (T) is shown in Fig. 8.2 for nickel and iron.
Spontaneous magnetisation appears at the Curie temperature TC .
8.2 Magnetism of Atoms
Before considering the origins of the ordered magnetism which arises in certain
solids, let us briefly review some features of the electronic structure of atoms.
8.2.1 Hydrogen Atom
Consider first the electronic energy levels in the simplest atom, namely hydrogen
(see Fig. 8.3a) [2, Chap. 11]. Note that there is a high level of degeneracy in the
levels. Each level with principal quantum number n = 1, 2, 3, . . . , corresponds to a
state of energy En = −E1 /n2 and is degenerate with respect to different values of
the orbital quantum number l (with l = 0, 1, . . . , n − 1). Each of these levels is itself
2l + 1 times degenerate. If we include the spin degeneracy, the total degeneracy of a
given energy level is 2n2 . Note that, since the hydrogen atom has only one electron,
4p
4s 4p 4d 4f 4s 3d
E E
3s 3p 3d 3p
3s
2s 2p 2p
2s
(a) (b)
1s
1s
Fig. 8.3 (a) Energy level diagram for the hydrogen atom. The angular momentum quantum num-
bers are given using the traditional notation, where s corresponds to l = 0, p to l = 1, and so on.
The energy is En = −E1 /n2 , where E1 = 13.6 eV. The ground state is 1s. (b) Energy level diagram
for a multielectron atom. The energy of an electron state depends on the quantum numbers n and l
8.2 Magnetism of Atoms 237
its ground state is clearly the 1s level, with angular momentum l = 0, while the spin
is S = 1/2. In this case, the magnetic moment of the atom is
μ = 2μB S = μB . (8.3)
8.2.2 Multielectron Atoms with Filled Shells
For a multielectron atom with Z electrons, the Coulomb potential of the nucleus of
charge Z is −ZVc (r). If we only take this potential into account, the corresponding
electron levels are similar to those of the hydrogen atom with En,Z = −Z 2 E1 /n2 .
However, the Coulomb interactions between electrons will modify these levels, and
in particular, they will remove the degeneracy of levels with the same principal
quantum number but different values of l. The energy of a level increases with its
orbital angular momentum (see Fig. 8.3b). Only the orbital degeneracies (due to
the spherical symmetry of the Coulomb potential) and spin degeneracies (due to
the rotational symmetry in spin space) will remain. The degeneracy of a level in
Fig. 8.3b is thus 2(2l + 1).
This partial lifting of degeneracy, revealed directly through the structure of the
periodic table, can be understood in a highly qualitative way by the following ar-
gument. Electrons in s levels have wave functions concentrated close to the nucleus
and therefore feel the repulsive effect of other electrons only rather weakly. In con-
trast, an electron in a p level (and even more so in d or f ) with the same principal
quantum number is on average further from the nucleus and thus feels the repulsive
effect of other electrons to a greater extent, which increases its energy compared
with the s level. Note that, in this level diagram, the Z electrons occupy the lowest
levels allowed by the Pauli exclusion principle in the ground state. If the number of
electrons is such that all the low energy states can be filled, the atomic shells are
saturated and the atom has no spin or orbital magnetic moment.
8.2.3 An Atom with a Partially Filled Shell: Carbon
To understand the origin of the magnetic moments of atoms with partially filled
shells, consider the case of carbon, which has 6 electrons. Recall that, if Vat (r) is the
atomic potential including the Coulomb potential of the nucleus and the mean value
of the repulsion between electrons, the energy levels of this multielectron atom are
solutions of
p2
H0 = + Vat (r) .
2m0
The eigenstates correspond to levels 1s, 2s, 2p, and so on. In the ground state of
the atom, two electrons occupy the 1s level and two the 2s level, while the last
Fig. 8.4 px , py , pz orbitals z z z
x y x y
x y
two are in the 2p state. The pairs in the 1s and 2s levels must be in configurations
with total spin zero, owing to the Pauli principle. It remains to determine the exact
configuration of the two p electrons which occupy the six-fold degenerate energy
level E2p . A basis for these eigenstates is formed by the three radial wave functions
px , py , and pz oriented along the x, y, and z axes, shown in Fig. 8.4. These states have
the same energy, as they can be obtained from one another by rotations in space.
To construct the two-electron eigenstates, a first possibility is to put the two
electrons in the same orbital, so the spatial part of the wave function would be
|px px = |px 1 |px 2 , where α and β in the notation |αβ indicate the states of the first
and second electrons, respectively. According to the Pauli exclusion principle, the
total wave function of these two electrons must be antisymmetric under exchange
of the two particles (r1 ↔ r2 ). As this spatial wave function is symmetric, the spin
wave function must be antisymmetric, giving a total wave function
1
|xx, S = 0 = √ | ↑↓ − | ↓↑ |px px . (8.4)
2
Here, in a ket like | ↑↓, the first (second) arrow gives the z component of the spin
of the first (second) electron. In the spin part of (8.4), we recognise the spin function
of a singlet state, where two spins 1/2 are combined to give a total spin S = 0.
Alternatively, the two electrons can be put in two different orbitals, e.g., |px py .
This product clearly has no well defined symmetry properties. We thus consider √
instead the symmetric√ and antisymmetric linear combinations (|px py + |py px )/ 2
and (|px py − |py px )/ 2, respectively. The first also gives rise to an antisymmetric
wave function (S = 0)
1 1
|x, y, S = 0 = √ | ↑↓ − | ↓↑ √ |px py + |py px . (8.5)
2 2
The total wave function associated with the antisymmetric spatial wave function is
obtained by multiplying by a symmetric spin wave function, e.g.,
1
|xy, S = 1, Mz = 1 = | ↑↑ √ |px py − |py px . (8.6)
2
Note that the spin part is the Mz = 1 component of the triplet wave function, with
total spin S = 1, of a pair of spin 1/2 particles, which explains the notation used on
the left-hand side of (8.6). For

the symmetric
√spin wave function required here, we
could also take the functions | ↑↓ + | ↓↑ / 2 or | ↓↓, which are the Mz = 0 and
Mz = −1 components of the spin triplet.
The singlet states (8.4) and (8.5) and the triplet state (8.6) constructed here are
eigenstates of the two-electron Hamiltonian
p21 p2
H1 = + Vat (r1 ) + 2 + Vat (r2 ) .
2m0 2m0
They are degenerate and have energy 2E2p . However, in order to treat the repulsive
Coulomb interaction between the two p electrons in a fully rigorous way, we must
use the Hamiltonian
p21 p2
H= + Vat (r1 ) + 2 + Vat (r2 ) + Vc (r2 − r1 ) . (8.7)
2m0 2m0
Given the eigenstates of H1 , the effects of the Coulomb interaction term denoted by
V12 = Vc (r2 − r1 ) can be treated as a perturbation. In this case, the energies of the
different eigenstates are given to first order in perturbation theory by
Exx,S=0 = 2E2p + px px |V12 |px px = 2E2p + Kxx , (8.8)

1
Exy,S=0 = 2E2p + px py | + py px | V12 |px py + |py px
2
= 2E2p + px py |V12 |px py + py px |V12 |px py
= 2E2p + Kxy + J0 , (8.9)
Exy,S=1 = 2E2p + Kxy − J0 . (8.10)
In these expressions, the terms Kxx , Kxy , and J0 are all positive. Note that, in the
integral Kxx , the electrons are on the same orbital px and have higher probability
of being at the same place than in Kxy , where they occupy two different orbitals.
The Coulomb repulsion is therefore stronger in the first case than in the second,
so Kxx > Kxy . The Coulomb repulsion between electrons thus tends to make the two
electrons occupy different orbitals. The energy level diagram is as shown in Fig. 8.5.
Note that this level diagram highlights the importance of the integral
!
J0 = py (r1 )px (r2 )Vc (r2 − r1 )px (r1 )py (r2 ) . (8.11)
|xx,S = 0 (3)
+Kxx
Fig. 8.5 Level diagram taking |xy,S = 0 (3)

into account Coulomb repul- 2J0
sion between the two p elec-
2E2p +Kxy
trons. The degeneracies of the |xy,S = 1 (9)
levels are given in brackets
This integral is called the intra-atomic exchange integral because it involves states in
which the electrons are exchanged from one side to the other. It requires the ground
state to be the spin triplet state S = 1 rather than the singlet state S = 0 constructed
with the same radial wave functions. This effect can be understood qualitatively by
examining (8.5) and (8.6). In the first spin singlet case, the symmetry of the spatial
wave function allows the electrons to be at the same position, whereas in the case of
the triplet, the antisymmetry of the spatial wave function means that the probability
of the two electrons being at the same position is zero. The divergence of Vc at short
range does not contribute to the corresponding matrix element, and this implies a
weaker Coulomb repulsion for the triplet state. The maximal spin state S = 1 is thus
the state with lowest energy.
This is a first example of one of the key ideas underpinning almost every mag-
netic phenomenon: the Coulomb interaction between electrons and the Pauli princi-
ple constitute the microscopic explanation for atomic magnetism. Note that no direct
spin-dependent interaction is needed to remove the degeneracy between singlet and
triplet states.
8.2.4 Atoms with Partially Filled Shells: Hund Rules
The above considerations for two electrons can be generalised to any situation in
which an atomic shell is partially filled, and this leads to the Hund rules, which
determine the orbital and spin angular momenta of the atomic ground state:
1. Choose the wave function with maximal total spin S.
2. Among the remaining possibilities, choose the wave function with maximal or-
bital angular momentum L, taking care to respect the Pauli exclusion principle,
i.e., to build a totally antisymmetric wave function.
3. The total angular momentum J is given by |L − S| if the shell is less than half-
filled, and by L + S if it is more than half-filled (an exactly half-filled shell always
gives L = 0).
Rules 1 and 2 are explained by a generalisation of the previous argument, appealing
to the Coulomb interaction between the electrons and the Pauli principle, while
rule 3 is due to a relativistic correction to quantum mechanics, namely the spin–
orbit interaction. Finally, an atom of total angular momentum J carries a magnetic
moment
μ = −gJ μB J , (8.12)
where the Landé factor gJ is given by

3 1 S(S + 1) − L(L + 1)
gJ = + . (8.13)
2 2 J(J + 1)
Note that, in the case of carbon, rule 3 gives a total angular momentum J = 0,
and hence zero magnetic moment, but there are clearly going to be many other
situations, e.g., three or more electrons in a p shell, two or more electrons in a d
shell, etc., where J will not vanish. The most important cases (when the d or f
shells are partially filled) are given in Tables 8.1 and 8.2, which specify the order
in which the electronic shell must be filled in order to satisfy the first two Hund
rules.
For example, for the d shell, we have l = 2 and lz = 2, 1, 0, −1, −2, and the cor-
responding electron can have spin ↑ or ↓, so there are a maximum of 10 electrons
in the shell. To satisfy the first Hund rule, we have to put in a maximum of spin ↑
electrons before filling states with spin ↓. Choosing the maximum orbital quantum
numbers for each electron then ensures the correct application of the second Hund
rule.
Consider, for instance, the case of Mn (3d5 4s2 ). The Mn3+ ion will thus have a
3d configuration, i.e., n = 4 electrons in the d shell. The occupation of electronic
4
states for n = 4 is specified in the fourth line of Table 8.1. The electrons must all have
spin ↑ with l decreasing from 2 to −1. The resulting value for the spin is therefore
S = 4(1/2) = 2, and for the orbital angular momentum L = 2 + 1 + 0 − 1 = 2. The
value of J results from the third Hund rule and corresponds to J = |L − S| = 0
since the shell is less than half-filled. Naturally, the Hund rules imply that there is
no magnetic moment for filled shells with L = S = J = 0, whereas for a half-filled
shell, L = 0 and the magnetism is entirely due to the spin, as is the case for the Mn2+
ion, for example.
In a similar way, Table 8.2 specifies the occupation of the f shell, which corre-
sponds to l = 3 and the rare earth series.
Table 8.1 Minimum energy configurations of the partially filled d shell. n is the total number of
electrons and lz the z component of the orbital angular momentum of an electron. L, S, and J satisfy
the Hund rules
n lz = 2 1 0 −1 −2 S L J
1 ↑ 1/2 2 3/2
2 ↑ ↑ 1 3 2
3 ↑ ↑ ↑ 3/2 3 3/2
4 ↑ ↑ ↑ ↑ 2 2 0
5 ↑ ↑ ↑ ↑ ↑ 5/2 0 5/2
6 ↑↓ ↑ ↑ ↑ ↑ 2 2 4
7 ↑↓ ↑↓ ↑ ↑ ↑ 3/2 3 9/2
8 ↑↓ ↑↓ ↑↓ ↑ ↑ 1 3 4
9 ↑↓ ↑↓ ↑↓ ↑↓ ↑ 1/2 2 5/2
10 ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ 0 0 0
Table 8.2 Minimum energy configurations of the partially filled f shell. n is the total number of
electrons and lz the z component of the orbital angular momentum of an electron. L, S, and J satisfy
the Hund rules
n lz = 3 2 1 0 −1 −2 −3 S L J
1 ↑ 1/2 3 5/2
2 ↑ ↑ 1 5 4
3 ↑ ↑ ↑ 3/2 6 9/2
4 ↑ ↑ ↑ ↑ 2 6 4
5 ↑ ↑ ↑ ↑ ↑ 5/2 5 5/2
6 ↑ ↑ ↑ ↑ ↑ ↑ 3 3 0
7 ↑ ↑ ↑ ↑ ↑ ↑ ↑ 7/2 0 7/2
8 ↑↓ ↑ ↑ ↑ ↑ ↑ ↑ 3 3 6
9 ↑↓ ↑↓ ↑ ↑ ↑ ↑ ↑ 5/2 5 15/2
10 ↑↓ ↑↓ ↑↓ ↑ ↑ ↑ ↑ 2 6 8
11 ↑↓ ↑↓ ↑↓ ↑↓ ↑ ↑ ↑ 3/2 6 15/2
12 ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ ↑ ↑ 1 5 6
13 ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ ↑ 1/2 3 7/2
14 ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ ↑↓ 0 0 0
Spin–Orbit Interaction
The fact that the nucleus is charged and moving relative to the electron implies that it will
create a magnetic field that couples to the electron spin. This is called the spin–orbit coupling
because it involves the electron spin and its orbital degrees of freedom, and it takes the form
Eso = λl · s. It leads to a coupling ESO = L · S between the total spin and the total orbital
angular momentum of the multielectron atom. When there is no spin–orbit coupling, the mu-
tually commuting operators L2 , Lz , S2 , and Sz each commute with the Hamiltonian, whence
one can find a common basis of eigenfunctions which are also eigenfunctions of the Hamilto-
nian. In this case, (L, mL , S, mS ) are good quantum numbers for characterising an eigenstate. On
the other hand, the spin–orbit coupling term does not commute with these operators. It can be
shown that the operator J = L+S is such that Jz and J2 = (L + S)2 = L2 +S2 +2L · S commute
with L · S. It then follows that Jz , J2 , and the Hamiltonian including the spin–orbit coupling
term have a common basis of eigenstates. The good quantum numbers for the Hamiltonian
with spin–orbit coupling are therefore J and mJ . The value of J corresponding to the ground
state is the one giving the lowest value to the spin–orbit coupling term, which is proportional
to 2L · S = J2 − S2 − L2 .
The expectation value of the total magnetic moment of the ground state, viz.,
μ = μB (L + 2S) ,
is μ = − J, mJ | L + 2S | J, mJ μB . This can be determined using the Wigner–Eckart theorem,
which tells us that the expectation value of μ/μB in the subspace | J, mJ is proportional to the
expectation value of J, with multiplicative coefficient [21]

J | μ · J | J /μB
gJ = − .
J(J + 1)
It is easy to check, using the expressions for J · S and J · L in terms of J2 , L2 , and S2 , that
−2μ · J/μB = 3J2 − L2 + S2 , which yields the expression (8.13) for gJ .
8.3 Paramagnetism of an Ensemble of Isolated Ions 243
8.3 Paramagnetism of an Ensemble of Isolated Ions
Consider an ionic solid containing isolated atoms with magnetic moments, like the
one discussed in the last section. In the presence of a magnetic field Ha , the spin
degeneracy of the electron wave function is removed, the spin levels being
EJz = +gJ μ0 μB Jz Ha , Jz = −J, −J + 1, . . . , J − 1, J .
If the ions are independent, statistical thermodynamics can be used to determine

the statistical average of the component μz of the magnetic moment operator
μ̂ = −gJ μB Ĵ from
+J
Jz =−J gJ μB Jz exp −βEJz
μ̂z = +J ,
Jz =−J exp −βEJz
where β = 1/kB T. If the number of paramagnetic ions per unit volume is N, a sim-
ple but tedious modification of this expression leads to the following result for the
magnetisation Mz = N μ̂z along the z axis:
α
2J + 1 2J + 1 1
Mz = NgJ μB J coth α − coth
2J 2J 2J 2J
= NgJ μB JBJ (α) , (8.14)
7
Gd 3 +
5
Fig. 8.6 Measurements of the Fe 3 +
magnetisation per paramag-
netic ion for different salts 4
containing the ions indicated M
in the figure, as a function
N μB
of the magnetic field and for 3
different temperatures. For a
given salt, the magnetisation Cr 3 +
is indeed a function of the ra- 2
tio Ha /T. Continuous curves 1.3 K
are the Brillouin functions for 2.02 K
the ions under consideration. 3.0 K
1
From Henry, W.E.: Phys. Rev. 4.2 K
88, 559 (1952). With the kind
permission of the Ameri-
can Physical Society (APS). 0 1 2 3 4
/PhysRev.88.559 μ0 Ha /T (tesla/K)
where we have set α = gJ μ0 μB JHa /kB T. The function BJ (α) is called the Brillouin
function. It reduces to B1/2 (α) = tanh α in the special case where J = 1/2. At high
temperatures, we have α 1 and the expansion coth u = u−1 + u/3 for small u gives
BJ (α) = α(J + 1)/3J, whence the magnetic susceptibility obeys Curie’s law:
1 ∂M (gJ μB )2 J(J + 1)
χ (T) = lim =N . (8.15)
Ha →0 μ0 ∂Ha 3 kB T
This law is found to apply well to many paramagnetic solids, as can be seen from
Fig. 8.6.
Question 8.1. From Fig. 8.6, can you determine the values of gJ and J for the various ions
considered there? Recall that the configurations of the neutral atoms are Cr (3d4 4s2 ), Fe
(3d6 4s2 ), and Gd (4f 7 5s2 5p6 5d1 6s2 ). Do the values of gJ and J correspond to what
would be expected from Tables 8.1 and 8.2? What conclusion can you draw?
8.4 Ordered Magnetic States
We saw in Sect. 8.1 that in many cases the Curie law will only work at high tem-
peratures, and that an ordered magnetic state appears at low temperatures, e.g., a
parallel alignment of the spins in the ferromagnetic state below the Curie temper-
ature TC . Clearly, such an alignment in zero field at thermodynamic equilibrium is
only possible if there are interactions between the spins which favour that. (If there
were no interactions, the only magnetic effect would be Curie paramagnetism.)
The first type of interaction between spins we can think of would be a dipole
interaction. Since each spin carries a magnetic moment m, there is a potential energy
of interaction between two spins of the form

μ0 3
V(r) = m 1 ·m2 − (m1 ·r) (m 2 ·r) , (8.16)
4π r3 r2
where r is the displacement vector between the two spins. The order of magnitude
of this interaction is easily estimated. The magnetic moment of an atom is of the
order of a few times μB , so taking a value of μB , we have
μ0 μ2B
V(r) ≈ . (8.17)
4π r3
With an interatomic distance r ∼ 2 Å, we then obtain V(r) ≈ (1/8) × 10−23 joule,
which corresponds to
V(r)/kB ∼ 0.1 K . (8.18)
This should be compared with typical Curie temperatures, which can be as high as
1,000 K. It thus seems inconceivable that an interaction of the order of 0.1 K should
give rise to ferromagnetism at temperatures a thousand times higher. The dipole
interactions cannot be the cause of ferromagnetism!
8.4 Ordered Magnetic States 245
8.4.1 Interatomic Exchange Interaction
It is in fact now well established that the origin of magnetic ordering is actually a
purely quantum interaction called the exchange interaction, which arises as a con-
sequence of the Coulomb interactions between electrons in combination with the
Pauli exclusion principle. We have already seen in the last section how these two
effects can give rise to atomic magnetic moments.
In order to understand this interaction, let us consider the simplest case, viz., the
hydrogen molecule, comprising two electrons and two protons. The Hamiltonian
will then be
h̄2 2 e2 1
H= (p1 + p22 ) + (8.19)
2m0 4π ε0 |r1 − r2 |

e2 1 1 1 1
− + + + ,
4π ε0 |r1 − R1 | |r1 − R2 | |r2 − R1 | |r2 − R2 |
where ri and pi are the positions and momenta of the two electrons, while the Ri are
the positions of the two protons. The first line of (8.19) represents the kinetic energy
of the two electrons and their mutual Coulomb repulsion, and the second line is
the Coulomb attraction between electrons and protons. The protons are assumed
fixed, so their is no term for their kinetic energy here. Without explicitly solving
the Hamiltonian (8.19), we may nevertheless make the following observations: H
is invariant under exchange of the two electrons, r1 ↔ r2 , and as a consequence
the spatial parts of the wave functions are either even or odd with respect to this
transformation:
ψ± (r1 , r2 ) = ±ψ± (r2 , r1 ) . (8.20)
The situation is similar to the one encountered in Sect. 8.2.3 for the two p electrons
of the carbon atom. The energies E± of these states will generally differ. The Pauli
principle requires the total wave function, including both spatial and spin factors, to
be antisymmetric under exchange of the two electrons. We thus have either a spin
singlet wave function (S = 0), viz.,
1
ψtot,+ = √ | ↑↓ − | ↓↑ ψ+ (r1 , r2 ) , (8.21)
2
or a triplet wave function (S = 1) with three possible components (M = −1, 0, 1):

⎧
⎪
⎪ | ↓↓ψ− (r1 , r2 ) ,
⎪
⎨ 1
ψtot,− = √ | ↑↓ + | ↓↑ ψ− (r1 , r2 ) , (8.22)
⎪
⎪ 2
⎪
⎩
| ↑↑ψ− (r1 , r2 ) .
A classic picture of these two situations is one in which the two spins are antiparallel
(S = 0) or parallel (S = 1).
For the hydrogen molecule, the ground state is the + type, i.e., S = 0, whereas
the first excited state is of − type, i.e., S = 1. (This is where the Coulomb interaction
between the electrons comes in. Without it, these different states would be degen-
erate.). If we are interested in relatively low energy phenomena, where only the
ground state and lowest excited state are involved, we can then specify these states
by their spin, with S = 0 for the ground state and S = 1 for the first excited state.
In order to represent the system, we can take an effective Hamiltonian that is
diagonal in this S = 0, S = 1 representation. We note that the Hamiltonian Ŝ1 ·Ŝ2
2
satisfies this condition. Indeed Ŝ = (Ŝ1 + Ŝ2 )2 implies that
1 2 2 2

Ŝ1 ·Ŝ2 = Ŝ − Ŝ1 − Ŝ2 , (8.23)
2
so for two spin 1/2 particles,
1 2 3
Ŝ1 · Ŝ2 = Ŝ − , (8.24)
2 4
and then
3
Ŝ1 ·Ŝ2 |S = 0 = − |S = 0 , (8.25)
4
1
Ŝ1 ·Ŝ2 |S = 1 = |S = 1 . (8.26)
4
The Hamiltonian can thus be represented in the subspace of the first singlet and
triplet states by
Ĥ = −J Ŝ1 ·Ŝ2 , (8.27)
up to an additive constant, with
J = E+ − E− , (8.28)
where J is called the exchange constant. This exchange Hamiltonian (8.27) deter-
mines the spin of the ground state. For J < 0 (this is satisfied for hydrogen), the
ground state is a singlet (with ‘antiparallel’ spins), while for J > 0, it is a triplet
(with ‘parallel’ spins).
It is important to understand that this is an effective Hamiltonian describing the
combined effects of the Coulomb interaction and the Pauli exclusion principle. The
original Hamiltonian (8.19) contains no terms depending on the spins of the elec-
trons. The process which governs the appearance of magnetism is then similar to the
one that yielded atomic magnetism. There, the repulsive interaction between elec-
trons favoured spatially antisymmetric states for the carbon atom, which must be
triplet states due to the Pauli principle. We could indeed have treated the subset of
the two states (8.5) and (8.6) of the carbon atom using a Hamiltonian Ĥ = −2J0 s1 ·s2
whose ground state is the triplet state, with 2J0 representing the intra-atomic ex-
change coupling.
Note that the energies involved here are either kinetic or Coulomb and lead to
much higher values of J than the dipole interaction energies. This kind of coupling
can therefore explain the high values found for Curie temperatures. Note that, as a
general rule, the values of J are very difficult to determine theoretically, and are
usually obtained experimentally.
Exchange Hamiltonian
The term ‘exchange Hamiltonian’ demands a little more explanation. Consider the effect of the
operator Ŝ1 ·Ŝ2 on the quantum states of the two spins. Noting that
1 + −
Ŝ1 ·Ŝ2 = Ŝ1z Ŝ2z + Ŝ Ŝ + Ŝ1− Ŝ2+ ,
2 1 2
we have
1 1
Ŝ1 ·Ŝ2 | ↑↑ =| ↑↑ , Ŝ1 ·Ŝ2 | ↓↓ = | ↓↓ ,
4 4 (8.29)
1 1 1 1
Ŝ1 ·Ŝ2 | ↑↓ = − | ↑↓ + | ↓↑ , Ŝ1 ·Ŝ2 | ↓↑ = − | ↓↑ + | ↑↓ .
4 2 4 2
This can be rewritten in the more compact form

1 1
Ŝ1 ·Ŝ2 + |αβ = |βα , (8.30)
4 2
where α and β represent one of the two orientations ↑ or ↓. The operator Ŝ1 ·Ŝ2 + 1/4 thus
exchanges the two spins. Up to the additive constant 1/4, it is precisely this operator that
appears in the Hamiltonian (8.27), which justifies the given name “Exchange Hamiltonian”.
8.4.2 Heisenberg Model
In a magnetic insulating solid, there are a great many spins, one for each mag-
netic atom. These spins interact through exchange forces like those discussed pre-
viously, and the exchange constant J between two atomic moments depends on
the overlap between the electron wave functions of the two atoms. These functions
decrease exponentially with distance, so the exchange constant also decreases very
fast as the distance between atoms increases. In the vast majority of cases, it is then
enough to take into account only the exchange interactions between nearest neigh-
bours. This leads to the Heisenberg model for a magnetic insulator, described by the
Hamiltonian

Ĥ = −J SR ·SR + 2μB B· SR , (8.31)
RR R
where the R are the sites occupied by the magnetic atoms (the lattice points of a
Bravais lattice in the simplest cases), and SR is the spin operator of the atom at the
site with position R. The notation RR on the sum indicates that it is to be taken
over all pairs of nearest neighbours. The first term of Ĥ thus describes the exchange
interactions. The second term represents the interaction between the spins and an
external magnetic field B = μ0 Ha .
Consider the case where all the atoms have spin 1/2 (the generalisation to the
case of arbitrary spins, rather important in practice, is quite straightforward), and
also J > 0, which is usually called the ferromagnetic case for reasons that will soon
become clear. Suppose also that the external field is oriented along the z axis and
positive. Under these conditions, the ground state of the Heisenberg model (8.31) is
simply a state in which all the spins are parallel:

|0 = | ↓R , (8.32)
R
where | ↓R is the spin-down state of the atom at site R. It is easy to check that |0
is effectively an eigenstate of Ĥ. The exchange operator acting on a state in which
the two spins are parallel simply reproduces this state up to a prefactor of 1/4 [see
(8.29)]. We then obtain

NzJ
Ĥ|0 = − − NμB B |0 , (8.33)
8
where N is the number of atoms per unit volume in the solid and z is the number
of nearest neighbours. Since |0 minimises each term of Ĥ individually for J > 0,
this state is indeed the ground state. This is still true even if B = 0. The spins are all
aligned, and this is indeed a ferromagnetic state.
Consider in particular the case where the applied field is zero. Note that the mag-
netic moment
M = −2μB S (8.34)

of the state (8.32), where S = R SR is the total spin, is oriented along the positive z
axis. However, there are other states with the same energy −NzJ /8. For example,
the state
|0 = | ↑R (8.35)
R
has all spins pointing in the same direction and opposite magnetisation to |0. More
generally, any state obtained by repeated application of the operator

+
S+ =
y
SR = (SRx
+ iSR ) (8.36)
R R
to the state |0 is an eigenstate of Ĥ with the same energy. (This is easily shown by
noting that Ĥ and S+ commute and that S+ increases the z component of Ŝ by 1.)
Note in particular that |0 = (S+ )N |0 and that there is a total of N + 1 states of the
same energy.
Which state is actually chosen by the system from among these N + 1 states
will depend on how the system is prepared, i.e., the experimental circumstances.
However, each of these states is characterised by a nonzero magnetisation oriented
in some specific direction in space. This is an example of an important phenomenon
known as symmetry breaking. The Hamiltonian itself is invariant under simultane-
ous rotation of all the spins, since it depends only on their scalar products. However,
each of its ground states, and more generally, as we shall see in Chap. 10, each of its
thermodynamic equilibrium states at low enough temperatures, is characterised by

a preferred direction in space. The ground state thus has lower symmetry than the
Hamiltonian.2
8.4.3 Ferromagnetism: Molecular Field Approximation
The explanation for ferromagnetism arose from a simple idea put forward at the
beginning of the 1900s by Weiss and Curie, based on the properties of an ensem-
ble of paramagnetic spins. An applied magnetic field partially aligns the magnetic
moments. In the magnetic state, a field thus seems to exist even when there is no
applied field. They called this the molecular field. Their idea was to treat this as
the field created at a vacant site by all the moments. It therefore seemed natural to
assume that it would be proportional to the magnetisation itself. If we consider the
Heisenberg model, we can understand the basis for this approximation. Indeed, we
need only consider all the terms in this Hamiltonian involving a spin at a given site
R, viz.,

HR = SR · −J SR + 2μB μ0 Ha . (8.37)
R
Here the sum over R refers to nearest neighbours of the site R. The molecular field
approximation consists in replacing the spin operators at sites R in this Hamiltonian
by their mean value, the underlying assumption being that fluctuations about this
mean value are small. If we carry out the transformation
M
SR −→ SR = − = SR , (8.38)
2μB N
the Hamiltonian of site R then takes the simple form of the Hamiltonian of a spin in
an effective magnetic field Beff :
HR = 2μB Beff ·SR . (8.39)
The effective magnetic field is given by

zJ
Beff = μ0 Ha + λ∗ M , λ∗ = , (8.40)
4μ2B μ0 N
where z is the number of nearest neighbours of the site R. It remains only to de-
termine M. In this approach, all the spins have the same expectation value, as
2 In practice, these considerations need to be treated cautiously. In a solid, there are usually mag-
netic anisotropies, to be discussed in the next chapter, stemming from the dipole interaction be-
tween spins and/or the spin–orbit interaction. These terms tend to orient M with respect to the
crystal axes. However, the operation S → −S remains a symmetry of the Hamiltonian even in this
case. A ferromagnetic state will thus always break at least one symmetry.
Fig. 8.7 Graphical solution of M

T > TC T = TC
(8.42). There is only a solu- Nμ B T < TC
tion with M = 0 if the gra-
dient T/NμB of the straight
line representing M/NμB as a 1
function of M/T is less than
the gradient at the origin of
the function tanh. The value
(8.43) is then obtained for TC
0 M/T
one would expect in a ferromagnetic material. We thus have M = −2NμB SR .

However, the thermodynamic average for an ensemble of spins subjected to a field
B is simply given by (8.14). Assuming that M is parallel to z, and for J = S = 1/2,
this gives
μB Beff
M = NμB tanh , (8.41)
kB T
which becomes, for zero applied field,
λ∗ μ0 μB M
M = NμB tanh . (8.42)
kB T
This equation thus determines M in a self-consistent way. Apart from the obvious
solution M = 0 which yields the paramagnetic state, there is only a solution with
M = 0 if λ∗ Nμ0 μ2B > kB T, as can be seen from the graphical solution shown in
Fig. 8.7.
It is easy to check that the solution M = 0 has lower free energy than the para-
magnetic solution M = 0, and hence that the ferromagnetic solution is stable for
T < TC , where
zJ
TC = . (8.43)
4kB
1.0
1
J=
0.8 2
J=1
Fig. 8.8 The magnetisations
of Ni and Fe in Fig. 8.2 0.6
M/Ms
are plotted in reduced co-

ordinates. Results are com- 0.4
pared with the solutions ob-
tained using the molecular 0.2
field approximation for differ-
ent values of the total angu-
0
lar momentum J. From [12], 0 0.2 0.4 0.6 0.8 1
Springer 1939
c T/TC
We can then identify TC with the experimentally measured Curie temperature. The
results in Fig. 8.2 for Fe and Ni are plotted in Fig. 8.8 in reduced coordinates, and it
can be checked that the agreement is qualitatively quite good.
Question 8.2. Apply the mean field theory to the general case of the Brillouin function for
an arbitrary total angular momentum J. Can you deduce the value of the effective field for
Fe and Ni? Take g = 2 and J = 1 and J = 1/2, respectively, for Fe and Ni.
Let us point out here that Fe and Ni are ferromagnets which display good metallic
properties. They cannot therefore be described correctly by the Heisenberg model
developed so far, which is restricted to localised electrons in an insulator. We shall
sketch in Sect. 8.6 the fundamental aspects which differentiate the electronic struc-
ture of such magnetic insulators from the band insulators considered so far, as well
as some model approaches which allow one to describe the coexistence of mag-
netism and metallicity.
Note, however, that M(T) is observed to deviate from the predictions of the
molecular field theory both at low and at high temperatures. In Chap. 12, we
shall see in more detail that the mean field approximation is in fact just one of
the more elementary approximations for describing magnetic phase transitions,
but that it often provides an excellent qualitative description of the phenomena
coming into play in a phase transition. Table 8.3 gives some values for TC and the
spontaneous magnetisation at low temperatures in various ferromagnetic solids.
Table 8.3 Curie temperature

TC and spontaneous mag- TC (K) M0 (kA/m)
netisation at low temperature Fe 1043 1752
(T → 0) for various solids. El-
ements or compounds that are Co 1388 1446
ferromagnetic at room tem- Ni 627 510
perature are all metallic Gd 293 1980
Dy 85 3000
CrBr3 37 270
Au2 MnAl 200 323
Cu2 MnAl 630 726
Cu2 MnIn 500 613
EuO 77 1910
EuS 16.5 1184
MnAs 318 870
MnBi 670 675
GdCl3 2.2 550
8.5 Antiferromagnetism and Ferrimagnetism
In nature, the exchange integral J is negative in many cases. This situation is of-
ten encountered in oxides, sulfides, and fluorides, for which the magnetic ions are
separated by ions with fully occupied shells like O2− , S2− , or F− . The interaction
J < 0 between magnetic ions then occurs via the filled shells of the anion. This is
known as a superexchange interaction.
If the spins are considered as classical vectors, it is natural to assume that they
may arrange themselves so as to be antiparallel between nearest neighbours. Clearly,
such an arrangement is possible for crystal structures which separate simply into
equivalent sublattices A and B, as is the case for the square lattice, for instance (see
Fig. 8.9), the simple cubic lattice, or the body-centered cubic lattice. (But note that
this is not possible for the triangular or fcc lattices.) It seems natural to introduce the
Néel state (named after the French physicist Louis Néel, who won the Nobel prize
in 1970 for demonstrating the existence of antiferromagnetism):
|Néel = | ↑↓↑↓ . . . . (8.44)
In this state, all the spins of sublattice A are up, while all those of sublattice B are
down. This clearly minimises the energy if the spins are treated as classical vec-
tors. This state has zero magnetisation, since the partial magnetisations MA and MB
cancel one another. However, we may still consider the alternating magnetisation
Malt = MA − MB . (8.45)
This quantity is nonzero in the Néel state, and since this vector introduces a pre-
ferred direction in space, as in the ferromagnetic case, the symmetry is broken once
again. Compounds with nonzero alternating magnetisation are said to be antiferro-
magnetic, and the temperature below which the alternating magnetisation appears is
called the Néel temperature. This temperature is given in Table 8.4 for some com-
pounds.
(a) (b)
Fig. 8.9 (a) Division of the square lattice into sublattices A (dark disks) and B (clear disks). (b)
Two-dimensional antiferromagnetic square lattice
8.5 Antiferromagnetism and Ferrimagnetism 253
Table 8.4 Some antiferro-

TN (K) TN (K)
magnetic compounds and
their Néel temperatures MnO 122 MnF2 67.3
FeO 198 FeF2 78.4
CoO 291 CoF2 37.7
NiO 600 MnCl2 2
RbMnF3 54.5
KFeF3 115
KMnF3 88.3 Sr2 CuO3 5
KCoF3 125 VS 1040
KCuF3 39 Cr 311
(a) (b)
Fig. 8.10 (a) Crystallographic and magnetic structure of MnO. The Mn2+ ions occupy the lattice
points of a face-centered cubic lattice. The O2− ions are in the middle of the sides and at the center
of the cube. In the antiferromagnetic phase, neighbouring moments have alternating orientations
pointing along the diagonals of the faces. (b) Primitive cell of the antiferromagnetic lattice of MnO.
Only the Mn are shown
The case of MnO is illustrated in Fig. 8.10. The existence of an antiferromag-

netic state is not so easy to demonstrate as the existence of a ferromagnetic state,
since there is no measurable spontaneous magnetisation.3 In fact the most direct ev-
idence comes from crystallographic measurements. Owing to the fact that the spins
alternate, the unit cell of an antiferromagnetic solid is twice that of the same crystal
without antiferromagnetism. The magnetic unit cell of MnO is shown in Fig. 8.10b.
The doubled cell then gives rise to new Bragg reflections. These can be observed
by neutron diffraction (a neutron has a magnetic moment and thus interacts with the
spins of the crystal). In Fig. 8.11, new Bragg reflections are clearly visible for MnO
below the Néel temperature. In principle, X-ray diffraction should yield the same
result. However, the coupling of light with the atomic magnetic moment is weaker,
and it is only just becoming possible to exploit this method with the advent of very
intense sources of synchrotron radiation.
3 The magnetic susceptibility of an antiferromagnetic compound has specific features that often
allow one to establish the existence of the antiferromagnetic state (see Problem 18: Properties of
an Antiferromagnetic Solid).
(111) (311) 80 K
100 MnO
80
a 0 = 8.85 Å
60
Intensity (neutrons/min)
40 (331) (511)
20
0 (111)
293 K
100
80 a 0 = 4.43 Å
(100) (110) (200) (311)
60
40
20
0
10 20 30 40 50
Angle of diffraction (degrees)
Fig. 8.11 Intensity of neutron diffraction by an MnO powder above (T = 293 K) and below (T =
80 K) the Néel temperature. Additional Bragg peaks are clearly detected in the antiferromagnetic
state. From Shull, C.G., Strauser, W.A., Wollan, E.O.: Phys. Rev. 83, 393 (1951). With the kind per-
mission of the American Physical Society (APS). http://link.aps.org/doi/10.1103/PhysRev.83.333
Question 8.3. Given the primitive cell of MnO shown in Fig. 8.10, can you see whether the
Bragg reflections expected for MnO are indeed observed in the paramagnetic state? What
reflections are expected in the antiferromagnetic state if the magnetic structure is indeed
the one given? Confirm this using the indexation of Bragg peaks and the lattice parameters
given in Fig. 8.11. Are any reflections missing? Why is this?
The Néel state should be treated with some care. Due to the relation (8.29), this state
is not an eigenstate of the Heisenberg Hamiltonian. Instead, we have
1 1
S1 ·S2 | ↑↓↑↓↑↓ . . . = − | ↑↓↑↓↑↓ . . . + | ↓↑↑↓↑↓ . . . , (8.46)
4 2
and similar relations for the other pairs of nearest neighbours. The second term on
the right of this relation shows that |Néel cannot be an eigenstate. However, sys-
tematically taking into account the extra terms generated in (8.46), it can be shown
that these effectively reduce the alternating magnetisation, without completely de-
stroying it. The very existence of antiferromagnetism is not therefore in doubt, but
rather the exact form of the ground state is more complicated than the simple Néel
state.
Due to their zero magnetisation, antiferromagnetic compounds are not of great
technological interest. However, compounds exhibiting a spontaneous magnetisa-
tion can be obtained from antiferromagnetic exchange interactions (more common
in nature than ferromagnetic interactions). It suffices to place atoms of different total
8.6 From Insulator Magnetism to Metallic Magnetism 255
Fig. 8.12 Spin arrangement ferro

in ferromagnetic, antiferro-
magnetic, and ferrimagnetic
solids
antiferro
ferri
Table 8.5 Curie temperature TC (K) M0 (kA/m)

(TC ) and spontaneous mag-
netisation at low temperature Fe3 O4 858 510
(T → 0) for several ferrimag- CoFe2 O4 793 475
netic solids
NiFe2 O4 858 300
CuFe2 O4 728 160
MnFe2 O4 573 560
Y3 Fe5 O12 560 195
spin on the two sublattices. Even an antiparallel alignment will then give rise to a
nonzero magnetisation. Such compounds, called ferrimagnetic compounds, are ex-
tremely important because they often have high saturation magnetisations, even at
room temperature, while being insulators, which is important for alternating current
applications, especially at high frequency. (Note that most ferromagnetic solids at
room temperature are metals, hence good conductors.) For comparison, Fig. 8.12
shows schematically the spin arrangement in ferromagnetic, antiferromagnetic, and
ferrimagnetic solids. The characteristics of some ferrimagnetic solids are given in
Table 8.5.
8.6 From Insulator Magnetism to Metallic Magnetism
We have examined so far two completely different limiting descriptions of elec-

tronic states in a solid. In the band structure approach used in the four first chapters
of this course we have described the case of electrons considered as independent,
their interactions being restricted to an averaged potential. The delocalisation of
these electrons between the atomic sites driven by the transfer integrals may yield
metallic states. In contrast, in the present chapter we are considering the specific
situation for which electrons localised on ionic states lead to local atomic mag-
netic moments. Those arise when the Pauli principle and on site inter-electronic
Coulomb repulsion are taken into account properly. We have assumed implicitly
that the transfer integrals between electrons on neighbouring ions are small enough
in such solids, so that these electrons do not delocalise. This then corresponds to an
insulating magnetic state quite different from the band insulating states considered
so far in the independant electron band approach.
The actual situation in real materials does indeed sometimes correspond to these
limiting cases, but a wide variety of solids correspond to intermediate situations,
like the one already mentioned for ferromagnetic metals such as Fe or Ni. These
intermediate cases are quite important both for the fundamental questions raised
and for the applications of the novel physical effects which come into play.
In this chapter we shall demonstrate first in Sect. 8.6.1 how the local Coulomb
interaction on atomic sites and the transfer integrals between atoms compete, which
explains the occurence of these intermediate situations. We recover then a compre-
hensive definition of the magnetic insulator cases named after the approach pro-
posed by Mott and Hubbard. In Sect. 8.6.2 we shall understand under which con-
ditions a Mott–Hubbard insulator can exhibit a transition into a correlated metallic
state either by application of a pressure, or by chemical doping. The latter case has
been emphasized by the experimental discovery of the strange metallic properties of
High Temperature superconducting cuprates, which will be sketched in Sect. 8.6.3.
Finally in Sect. 8.6.4 we shall show that magnetism in metals can also be described
in a band structure approach, provided that the local Coulomb repulsion is taken
into account at least approximately.
8.6.1 Mott–Hubbard Insulator
The very existence of magnetic insulators raises a fundamental problem for the
band theory. According to the discussion in Chap. 3, all insulators should be non-
magnetic (or more precisely, slightly diamagnetic). In order to explain the magnetic
behaviour discussed in this chapter, we must therefore reassess the underlying ap-
proximations that led to the band theory, and especially the averaging approach to
the Coulomb interactions between electrons.
Let us begin by considering the case of an isolated atom (on the left in Fig. 8.13),
which was the starting point for the tight-binding theory in Chap. 1. In this con-
text, it was assumed that the energy brought to the system by an extra electron
would be ε0 , and that a second electron on the same atom would also bring ε0 , so
that the total energy would be 2ε0 for a doubly negatively charged ion. But this is
obviously not very realistic, owing to Coulomb repulsion. Apart from its ‘orbital’
energy ε0 , the second electron will also be subject to the Coulomb repulsion of the
first electron, and its energy will thus be higher than ε0 by an amount usually de-
noted by U, which represents the Coulomb repulsion between the first and second
electrons added to an initially neutral atom. The total energy of the doubly negative
ion is thus 2ε0 + U. Note that U can vary considerably depending on the atom (from
about 1 eV to more than 10 eV). This very simple approach goes by the name of
the Hubbard model.
2nd electron
ε 0 +U
E 1st electron
ε0
Fig. 8.13 Level diagram for the Mott–Hubbard model. Left: Isolated atom. Center: Mott–Hubbard
insulator obtained for a small hopping integral t < U. Right: Metallic situation corresponding to
t>U
If we now consider this ion in a crystal, the hopping integrals between nearest
neighbours will broaden the discrete atomic levels into bands. If to begin with we
consider the limiting case of small hopping integral compared with U, we find our-
selves in a situation corresponding to the middle of Fig. 8.13. There are two allowed
energy bands called the upper and lower Hubbard bands, separated by a band gap.
This gives the impression that we have a typical insulator (or semiconductor). But
this is not in fact correct. There is one additional one-electron state per atom, so that,
in a solid comprising Nn atoms, the lower band of the middle column can contain
up to Nn electrons, rather than up to 2Nn electrons, as was the case in the context of
the band theory developed in Chap. 3. In particular, if there is now one electron per
atom (or more generally an odd number of electrons per primitive cell), the lower
band will be completely filled and the upper band completely empty. We will thus
have an insulator with an odd number of electrons per primitive cell, in contrast to
the predictions of Chap. 3. This is of course a consequence of the interactions U
between electrons. The very existence of such an insulator (usually called a Mott–
Hubbard insulator in recognition of the two British scientists who first studied them
in the 1960s) is thus a consequence of the Coulomb interaction between electrons.
Important examples of Mott–Hubbard insulators are undoped cuprates in which the
Cu2+ ions are in a 3d9 state.4
So what are the magnetic properties of a Mott–Hubbard insulator? If we begin
by considering the limiting case of very small hopping integrals, we end up with
isolated atoms. The electron in the level ε0 can then have spin ↑ or spin ↓, behaving
like an isolated spin 1/2. In the solid, these spins taken together will give rise to
Curie paramagnetism χ ∼ 1/T, and we then have a paramagnetic insulator, once
again contradicting band theory, which predicted that any insulator must be weakly
diamagnetic. If we also take into account the finite value of the hopping integrals,
we find that, at low enough temperatures, the spins will arrange themselves antifer-
romagnetically.
4 See Problem 17: Electronic Properties of La2 CuO4 .

We have seen in this chapter that there are not only magnetic insulators of spin
1/2 (in fact these are in the minority), but that in most cases the spin per atom is much
higher. This is due to the fact that, in almost all cases, the atomic orbitals in question
are not s levels (hence non-degenerate), but d- or f -type (hence five- or seven-fold
degenerate). In this situation, Hund’s rules determine the angular momentum of each
atom in the Mott–Hubbard insulating state.
The Hubbard model, which replaces the true Coulomb potential V(r) ∼ 1/r by
a repulsion which only acts if the two electrons are located on the same atom, is
clearly a drastic simplification of the actual physical situation. However, it is rather
naturally justified in the context of the theory of magnetic phenomena. We have
seen in this chapter that in this context the Pauli exclusion principle is of funda-
mental importance. Pauli exclusion is essentially important for two electrons on the
same atom, since two electrons separated by some multiple of the lattice constant
will hardly be affected by the Pauli principle, so the interaction between them will
depend very little on the relative orientation of their spins, and will have very little
influence on their magnetic properties.
The main conclusion to this section is that, going beyond the possibilities offered
by band theory (paramagnetic metals and diamagnetic insulators), the presence of
Coulomb interactions between electrons, if they are strong enough, can give rise to
an insulating state with a variety of magnetic properties, such as Curie paramag-
netism, ferromagnetism, antiferromagnetism, and so on.
8.6.2 Mott Transition and Doped Mott–Hubbard Insulators
In a Mott–Hubbard insulator, if we increase t (or if we consider compounds with

lower values of U), for a certain critical value of t/U, the upper and lower Hubbard
bands begin to overlap (see Fig. 8.13 right), causing the band gap to disappear and
leading to a metallic state. Such an increase in t can be produced by bringing the
atoms closer together.
This was first achieved in the case of doped semiconductors by increasing the
donor concentration, e.g., by increasing the concentration of phosphorus in sili-
con. This causes the hydrogen-like orbitals of P to move much closer together (see
Sect. 4.5b) and increases the hopping integrals, while remaining in a configuration
corresponding to one electron per donor atom.5 A simpler way to achieve this sit-
uation directly without changing the number of electrons in a material is to apply
an external pressure. This increases the hopping integrals t by bringing the atoms
closer together, provided that the material is compressible.
In the metallic state thereby induced, one then observes a Pauli paramagnetism
and a specific heat that is linear in temperature, but the Wilson ratio discussed in
Chap. 3 is generally greatly increased.
5 See Problem 7: Insulator–Metal Transition.

Note that a Mott–Hubbard insulator looks at first glance like a band insulator, the
only difference being that here each Hubbard band contains only Nn states rather
than 2Nn states in the case of the band theory described in Chap. 3. Chemical treat-
ment may be envisaged to change the number of electrons in a Mott insulator. For
example, it can be doped with holes, reducing the number of electrons in the lower
Hubbard band by a number Ne smaller than Nn .
This is exemplified by the high-temperature superconducting compounds, where
one typically has Ne = 0.75–0.9Nn . It is interesting to ask how many states there
are in the upper Hubbard band. The answer is simple. A state in the upper band
corresponds to a doubly occupied atom. But in order for an atom to be doubly oc-
cupied, a first electron must clearly be present. Since there are Ne electrons in the
lower Hubbard band, the number of states in the upper Hubbard band will be Ne ,
i.e., it will depend on the number of electrons present. The situation is very different
in a band insulator.
Furthermore, the total number of states in the upper and lower Hubbard bands
must be 2Nn . For the lower band, we thus find 2Nn − Ne available states. For the
so-called doped case, i.e., Ne < Nn , there are thus more states than electrons, and a
metallic situation is expected, with nonzero conductivity and Pauli paramagnetism.
A very similar analysis can be applied to the case Ne > Nn , with analogous results.
The ‘doped’ Mott–Hubbard insulator is therefore a metal.
8.6.3 Magnetism and Superconductivity
According to the above discussion, cuprates such as YBa2 Cu3 O6 or La2 CuO4
are antiferromagnetic Mott insulators. At low doping levels, their metallic prop-
erties arise from the situation described in the last section, and their metallic be-
haviour is far from being like that of conventional metals described in Chaps. 3
and 4. Experimental investigations carried out on the cuprates, and also on certain
other classes of doped Mott insulators, have shown that doping gradually reduces
the Néel temperature of the antiferromagnetic state. This AF state is completely
suppressed for a low level of doping, of the order of nh ≈ 0.05, as can be seen
in the phase diagram for the cuprates in Fig. 8.14. The static magnetism gives
way to a metallic and superconducting state for nh > 0.05. However, most of the
relevant physical quantities turn out to be very different from those of more ‘or-
dinary’ metals. Up to about nh ≈ 0.15, the existence of magnetic correlations be-
tween the electron moments of Cu tends to suppress low temperature electronic
excitations, which justifies the name pseudogap phase for the corresponding elec-
tronic state. The magnetic susceptibility does not exhibit a Pauli behaviour in
this regime, and the existence of the pseudogap can be seen in all the thermody-
namic quantities, as in optical absorption, photoemission (ARPES), or tunnel effect
experiments.
The temperature T ∗ at which this pseudogap appears decreases with increasing
nh and eventually meets the superconducting critical temperature when the latter
0 0.05 0.10 0.15 0.20 nt

T(K)
Y Ba 2 Cu 3O 6 + x
T*
400
TN
300 PG
200 AF
100 Tc
SC
0
0.2 0.4 0.6 0.8 1
x
Fig. 8.14 Cuprate phase diagram versus hole doping nh of the CuO2 planes. The latter can be
modified in the YBa2 Cu3 O6+x compound by changing the oxygen content of the Cu intermediate
planes (and formation of CuO chains). The approximate correspondence between nh and x is given
on the upper scale. The insulating Mott antiferromagnetic (AF) state is destroyed for small dopings
and opens the way to the ‘pseudogap’ (PG) phase in the ‘underdoped’ range. In this compound
the ‘optimal’ doping for which Tc reaches its 93 K maximum value is reached for x ≈ 0.95 and
nh ≈ 0.15. For this hole content, the metallic state remains strange with a T linear resistivity. The
‘overdoped’ regime nh > 0.17 is observed in other cuprate families and exhibits a fast decrease of
Tc , with a progressive evolution towards a more classical metallic behaviour. This figure has been
composed from experimental results reported by Alloul, H., Bobroff, J., Gabay, M., Hirschfeld, P.:
Rev. Mod. Phys. 81, 45 (2009)
becomes maximal. The pseudogap is said to appear in the underdoped region of the
phase diagram. When the doping becomes optimal for superconductivity, it is nev-
ertheless observed that the electronic properties of the metallic state still differ from
those of normal metals. The pseudogap disappears, but the resistivity varies linearly
with temperature, and the measured electron collision rate is given by τ −1 ≈ kB T/h̄,
and is much larger than in normal metals. This high level of electron scattering is
attributed to the persistence of very short range magnetic correlations, so that this
region of the phase diagram is referred to as the strange metal region. When the
doping is raised above nh ≈ 0.20, the physical properties gradually begin to re-
semble those of a ‘normal’ metal, or Fermi fluid, whereas the superconductivity
disappears rapidly. This region of the phase diagram is called the overdoped region.
While this experimental situation gives an impression of consistency, little is re-
ally understood on the theoretical level at the present time (apart from the undoped
case). The unusual metallic behaviour, the pseudogap, and the origins of supercon-
ductivity in the cuprates remain front line research topics.
Note that many families of compounds exhibiting superconductivity suspected of
being exotic have remarkably similar phase diagrams to the cuprates, with regard to
the proximity of the magnetic and superconducting phases, the system moving from
one state to the other for different levels of doping. In many cases, e.g., organic
compounds, heavy fermion compounds, and more recently iron pnictides and the
fullerene compound Cs3 C60 , the magnetic behaviour can be reduced by applying a
pressure, leading to a metallic, and often superconducting state.
8.6.4 Metallicity and Magnetism in a Band Approach
The notion of Mott–Hubbard transition has revealed that it should be possible to

treat the electronic structure of a Mott insulator and of a metallic state within a
unique framework. Although major steps have been taken in this direction, this is
far from being achieved. The developments currently under way are at the forefront
of the research on the physics of correlated electron systems and are well beyond the
scope of this course. However, in view of the importance of the physical properies
of the corresponding materials for applications we shall present here some limiting
situations where one can at least reach a significant qualitative description of the
relevant physics.
a. Distinct Electronic Bands for Magnetism and Metallicity
In Sects. 8.6.1–8.6.3 we have considered magnetic Mott–Hubbard states for a single

orbital model. In many actual compounds with polyatomic unit cells, Mott localised
states on some specific atoms of the unit cell may coexist with delocalised electrons
on distinct atomic sites. This happens for instance in some compounds involving
rare earth atoms and lighter elements. The external f electrons of the rare earths
have a limited spatial extension and for particular atomic structures these orbitals
can display very little overlap with the other atomic orbitals. An excellent example
of this situation can be found again in the YBa2 Cu3 O6+x family of cuprates.
Indeed such compounds can be synthetized with nearly any 4f rare earth instead
of Y. Let us consider the case of GdBa2 Cu3 O6+x . There the Gd atoms, which are
(like Y) embedded between two CuO2 planes, are in a 3+ state but, while Y3+ has
no f electron, Gd3+ has 7 electrons on its unfilled 4f orbitals. The Gd3+ ion is then
in its well defined 4f 7 state with a total spin S = 7/2, as expected for an isolated
Gd3+ ion (see Table 8.2 ). Adding an electron on Gd to bring it into the 4f 8 elec-
tronic state would require a prohibitive energy U which cannot be provided by the
weak transfer integrals from the orbitals of the CuO2 planes. Consequently Gd3+ re-
mains a well isolated ion with a local magnetic moment of 7μB very weakly coupled
to the electronic states of the CuO2 plane which are responsible for the 2D metal-
lic behaviour and superconductivity. The coupling is so small that the metallic and
superconducting properties of the CuO2 planes are absolutely identical to those of
YBa2 Cu3 O6+x with Tc values which differ by at most 1 K. However the small trans-
fer integrals are sufficient to induce a weak coupling between Gd ionic spin states
with an antiferromagnetic ordering of the Gd moments below a Néel temperature
TN = 2.25 K.
This simple example just demonstrates that magnetism can be sustained by some
orbitals (those of Gd3+ ) which remain partly filled in a Mott–Hubbard state, while
a metallic character can occur in independent bands (those of the CuO2 planes). In
cuprates, from what we have seen in Sect. 8.6.3, the problem of strange metallicity
in a correlated band, which occurs in the CuO2 planes themselves, nevertheless
remains the same there for the Y and Gd compounds.
Other systems are such that the metallic bands display weak correlations and
might be treated in an independent electron picture. This is for instance the case
for the family of comounds discovered by Chevrel, an archetype being HoMo6 S8 .
Here the Mo6 S8 units give rise to 4d metallic bands while the rare earth Ho3+
order ferromagnetically below TC = 0.69 K, so that magnetism and metallicity can
be considered as independent. In that case a classic BCS superconductivity with
Tc = 1.2 K is sustained by the 4d bands. This Tc being small, superconductiv-
ity is nevertheless suppressed below TC = 0.69 K by the large internal field in-
duced on the Mo sites in the magnetic ordered state of Ho3+, contrary to the case
of GdBa2 Cu3 O6+x for which the high temperature superconductivity is quite ro-
bust to the internal field developed below the AF ordering temperature of the Gd3+
moments.
Such specific situations of coexistence between magnetism and metallicity are
not so difficult to understand and to model. In those cases we just need to remember
that the strong Coulomb interaction pins the occupancy of the rare earth orbitals at a
fixed number (7 for Gd), and does not allow electrons to transfer from the other sites
to complete the filling of the rare earth levels. In the band theory of Chaps. 1, 2, 3
and 4 we often considered the case of atomic ions with filled electronic shells. Here
the difference is that the partial occupancy of some atomic levels which sit below
the Fermi level leads to local moment and magnetism somewhat disconnected from
the properties of the metallic bands.
b. Metallicity and Magnetism in the Same Bands
If we now consider cases such as that of Fe or Co we address a quite different prob-

lem, since in these elemental metals, although s electrons contribute to the metallic
transport, magnetism and metallicity also occur for the electrons of the 3d bands.
So how can we describe this possibility to get ordered magnetism for delocalised
electronic states ? We immediately see that, as in the Mott–Hubbard case, we have
to consider that the strong Coulomb interaction U on the Fe (or Co) sites forces us
(i) to relax the rule that every 3d orbital should be doubly populated
(ii) to break the symmetry between the possible orientations of the electron spins.
One way to implement that in the band structure theory was suggested initially
by Stoner who considered that in a magnetic state one should differentiate the en-
ergy bands for up and down spin orientations, the simplest approach being then to
shift the energy band of the down spin orientation with respect to that of the up spin
orientation of an energy 2δE which will be controlled by the on site Coulomb inter-
action U, as displayed in Fig 8.15. This reduces the electronic correlation energy at
the expense of an increased kinetic energy.
One can note that the total correlation energy for the 3d band can be written
Un↑ n↓ where n↑ and n↓ are the total occupancies of the two sub-bands, which are
initially equal, so that the variation of correlation energy due to the shift 2δE of the
↑ spin energy band with respect to the ↓ spin band is given by
δE
EF
D(E )
Fig. 8.15 In a purely paramagnetic state the densities of states D(E) of the up and down spin
energy bands are degenerate in the absence of any applied field (broad s band and dashed lines for
the narrow 3d band). Strong enough electronic correlations in the 3d band might favor energetically
a state where the degeneracy of the two spin directions is lifted (full lines). Such a splitting 2δE
between the two 3d electronic bands yields non integer moments per site and differentiates the
density of carriers of the two spin states which also acquire distinct mobilities

U n↑ − D(EF )δE n↓ + D(EF )δE − Un↑ n↓ = −U [D(EF ) δE]2
where D(E) is the density of states per spin diection, and we have assumed for sim-
plicity that δE is smaller than the bandwidth and that the bands are rigidly shifted.
The total kinetic energy of the band increases by
D(EF )(δE)2 ,
so that the total electronic energy variation
[1 − UD(EF )] D(EF )(dE)2
can only be negative if

UD(EF ) > 1.
The latter equation is then the condition for which magnetic moments can appear in
the metallic bands and may give rise to magnetic order at low T.
As D(EF ) ≈ 1/W (with W ≈ zt, where z is the number of neighbours), we natu-
rally recover a condition analogous to the Mott–Hubbard condition U/W > 1 for the
metal to insulator transition in Sect. 8.6.1. This condition can be met for high values
of the density of states D(EF ), as found in transition metals. This simple qualita-
tive argument explains furthermore why elements of the 3d series like Fe, Co and
Ni are magnetic contrary to the 4d elements, for which the d orbitals have larger
spatial extension, so that t and W are therefore larger. A further important merit of
this approach is that, contrary to the Heisenberg model, it explains the non integer
measured values of the magnetic moment per atom in the transition metals.
This approach initiated by Stoner has been taken as giving a criterion for a fer-
romagnetic instability in a metallic band. But as the spin quantization axis is not
defined in this approach and could vary from site to site, it only if fact tells us
that moments may appear. The ground state of the electronic system will only be
ferromagnetic if the effective exchange coupling between electrons mediating this
instability is indeed ferromagnetic.
One may notice that the Pauli principle prevents electrons with the same wave
vector k and the same spin from sitting at the same point r, so that even in a free
electron band the wave function of a k ↑ electron at r feels a lower density of
k ↑ than of k ↓ electrons. This reduces the total Coulomb repulsion energy for
electrons with the same spin direction, and leads then to a reduction of the ground
state energy for electrons with the same spin orientation with respect to that of
electrons with antiparallel spin orientation, which does indeed correspond to an ef-
fective ferromagnetic interaction. So the Pauli principle for delocalised electrons
seems to favor ferromagnetism while the short range Coulomb repulsion (like the
on site local U) rather favours antiparallel ordering, which yields AF if the lattice is
bipartite.
We have therefore sketched here that both ferromagnetism and antiferromag-
netism may be sustained in metallic bands if the Pauli principle and Coulomb re-
pulsions are taken into account. Subsequent efforts are being made to generalize the
band structure calculations (similar to those introduced in Chaps 1, 2, 3 and 4) by in-
troducing U explicitly. Such spin polarized band structures, computed with different
approximation schemes are under intense scrutiny in comparison with experimental
data, such as ARPES.
Now such a differentiation of ↑ and ↓ spin bands also has implications for the
transport properties of ferromagnetic metals since the numbers of carriers differ
for the two spin bands (see Fig. 8.15). Then the contribution of the majority and
minority spin carriers to the conductivity will be different, as the scattering times of
these two types of carriers on magnetic defects will also differ.
This spin dependent transport is at the origin of the so-called GMR (Giant
Magneto-Resistance) effect which has led to the attribution in 2007 of the Nobel
prize to A. Fert and P. Grümberg. These authors had noticed that the resistance of
multilayers of metallic films can be dependent on the relative magnetic ordering
of these layers. For instance Fe/Cr multilayers can be synthetized with alternating
orientations of the magnetization of the ferromagnetic Fe layers, while the inter-
mediate Cr layers are non magnetic. A small applied field can suffice to align the
magnetism of the Fe layers in the same direction so that electrons are more scat-
tered in the antiparallel case than in the parallel case. This variation of resistance
is used to detect the magnetic field induced by the magnetic bits recorded in a disk
memory, so that the GMR is nowadays efficiently used in read heads for hard disk
drives.
More subtle effects induced by spin dependent scattering are being investigated
in order to attempt to detect and/or reverse the magnetization of domains in nanos-
tructured materials. The global effort in this field, known as “spintronics”, associates
fundamental and applied research with many potential applications.
8.7 Summary 265
8.7 Summary
The magnetism of multielectron atoms is related to the removal of orbital degener-

acy in the electron levels resulting from the Coulomb repulsion between electrons
in these orbitals and the Pauli exclusion principle. The existence of atomic magnetic
moments then results indirectly from interactions between electrons.
An ensemble of independent magnetic ions with magnetic moment μ = gμB J
has Curie magnetic susceptibility
(gJ μB )2 J(J + 1)
χ (T) = N .
3 kB T
The magnetisation of ferromagnetic materials results from the macroscopic align-

ment of the atomic moments. The magnetic dipole interaction between atomic mag-
netic moments is too weak to be the cause of ferromagnetism. The Coulomb inter-
action between electrons on neighbouring atoms removes the degeneracy of their
electronic levels. Together with the Pauli exclusion principle, this leads to an effec-
tive interaction between the spins of near-neigbouring atoms, as expressed by the
exchange Hamiltonian
Ĥ = −J Ŝ1 ·Ŝ2 ,
where J depends on the distance between neighbouring atoms and their orbital
configurations.
The Heisenberg model describes the magnetic properties of an insulator, taking
into account only the exchange interactions between nearest neighbours:

Ĥ = −J SR ·SR + 2μB B· SR .
RR R
A ferromagnetic ground state is obtained for J > 0. A molecular field theory, in

which each moment is subjected to an effective field due to its nearest neighbours,
leads for a moment J = 1/2 to a magnetisation satisfying
λμ0 μB M
M = NμB tanh ,
kB T
with a Curie temperature

zJ
TC = .
4kB
When J < 0, neighbouring spins prefer to be antiparallel. When the crystallo-
graphic structure allows it, a configuration in which each spin is antiparallel to
its neighbours is obtained. This antiferromagnetic (Néel) structure has zero macro-
scopic magnetisation in systems involving only one type of magnetic ion. With an
ordered structure of different magnetic ions, a macroscopic magnetic moment can

survive (ferrimagnetism).
The existence of such magnetic insulators is shown to result naturally when the
energy cost associated with a large on site atomic Coulomb ineraction U prevents
double occupancy of the corresponding atomic orbitals. When U overcomes the
effect of the transfer integrals t which favours the electron delocalisation, the result-
ing Mott–Hubbard insulator displays only single occupancy of the electronic states.
However if t and U become of comparable magnitude a transition to a metallic state
with unusual magnetic properties may occur.
In undoped cuprates the CuO2 planes are in a Mott–Hubbard insulating state with
one hole per copper site. The superconductivity of cuprates is obtained by chemical
doping of the CuO2 planes which induces a metallic state in which strong signatures
of the large electronic correlations remain.
In many compounds Mott localised states on some specific atoms of the unit
cell may coexist with delocalised electrons on distinct atomic sites. In such situa-
tions a conventional non interacting band approach might apply for the delocalised
electrons and the metallic and magnetic properties are somewhat independent.
However in other limiting cases for which transfer integrals are comparable to
U for the same band, a magnetic metallic state might be described in a band ap-
proach as long as one relaxes the rule that each state should be doubly occupied and
one then differentiates states with up and down spin electrons. In such a situation
the spin polarized bands correspond to two distinct types of carriers and the result-
ing spin dependent transport properties are prominent and might be quite useful in
a variety of “spintronics” devices.
Question 8.1
We obtain the values gJ J = 7, 5, and 3 for Gd3+ , Fe3+ , and Cr3+ , respectively.
The slope at the origin is g2J J(J + 1)μB /3kB , i.e., 0.22g2J J(J + 1) in SI units. The
experiment can thus be used to determine (J + 1)/J and hence also J. However, the
accuracy of the experiment is rather poor. The most accurate value obtained is for
Cr3+ , for which we find 1 + 1/J = 1.66, so that J = 3/2 and hence gJ ∼ 2.
According to Table 8.2, we expect J = S = 7/2 for Gd3+ , which corresponds to
a 4f 7 ion, and therefore gJ = 2. This would appear to be confirmed experimentally.
Likewise, from Table 8.1, we expect J = S = 5/2 for Fe3+ , which is a 3d5 ion, and
hence gJ = 2. This is also confirmed experimentally.
However, Cr3+ is a 3d3 ion, for which we expect J = 3/2, S = 3/2, and L = 3,
and hence gJ = 0.4. We should therefore find gJ J = 0.4(3/2) = 0.6. This is not
confirmed experimentally. We have instead J = 3/2 and gJ = 2, as though only
the total spin were contributing to the total angular momentum, i.e., S = 3/2 and
L = 0. This is effectively the case. The reason is that the Hund rules apply to free
ions, but must be modified for solids. In an ionic crystal, the sum of the Coulomb
potentials due to the charges of the ions surrounding a given ion must be added to
the Coulomb potential due to the atomic nucleus. This term, called the crystal field
potential, cannot be neglected in the case of 3d ions and changes the order of the
atomic electronic levels. For ions in the 3d series, this means that we obtain L = 0
for the lowest level, and the orbital angular momentum is said to be quenched by
the crystal field.
Question 8.2
The value of M at saturation is Msat = gJ μB J. The relation between TC and J

becomes
J(J + 1) Jz
kB TC = zJ , λ∗ = ,
3 Nμ0 g2J μ2B
which implies, at low temperatures, when the magnetisation is saturated, that
3kB TC
Beff = .
gJ μB (J + 1)
For Ni (J = 1/2), this gives
kB TC
Beff = ,
μB
whereas for Fe, we obtain
3 kB TC
Beff = .
4 μB
With the values of TC in Fig. 8.2, TC = 630 K for Ni and 1,050 K for Fe, we ob-
tain values Beff ∼ 1, 000 tesla, well above anything that could be achieved in the
laboratory.
Question 8.3
The MnO lattice is face-centered cubic with lattice constant a, with one Mn and one
O per primitive cell of the Bravais lattice. The reciprocal lattice is body-centered
cubic (see Chap. 2). If we consider the reciprocal lattice of the cubic lattice of side
a∗ = 2π/a, the lattice points correspond to (2h, 2k, 2l) and (1 + 2h, 1 + 2k, 1 + 2l).
We thus expect reflections for (111), (200), and (311), which are indeed observed,
although (200) is weak, whereas there is no reflection for (100).
The magnetic lattice is obtained by doubling the lattice spacing in the three space
directions, and is also fcc with a basis. We find the same reflections with half the
lattice constant a∗ /2 in the reciprocal lattice. The corresponding reflections have the
same indices (111), (200), (311), (331), (511), and so on, with half the spacing.
Note that the Bragg reflection angles given by sin θ = nλ/2dhkl are very small
and expressed in arc minutes, because the wavelength of the neutrons used is very
short compared with the size of the primitive cell. In this case, sin θ θ and the re-
flection angles are divided by 2 for the magnetic lattice as compared with the atomic
lattice, as can be checked by comparing the top and bottom figures in Fig. 8.11. (Re-
flections with the same indices occur at half the angles.) Note, however, that there
is no magnetic reflection at (200) or (220). This corresponds to the fact that there is
extinction here. The structure or form factor (2.22) vanishes due to the high level of
symmetry of the magnetic basis.
Chapter 9
Magnetic Anisotropy, Domains, and Walls
Contents
9.1 Magnetic Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

9.1.1 Magnetocrystalline Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
9.1.2 Effect of Anisotropy: Irreversibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
9.2 Dipole Interactions, Demagnetising Fields, and Domains . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.2.1 Demagnetising Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.2.2 Demagnetisation Energy and Magnetic Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
9.3 Bloch Walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
9.4 Magnetic Hysteresis Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
9.4.1 Bitter Method for Observing Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
9.4.2 Magnetisation Process and Hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
9.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
270 9 Magnetic Anisotropy, Domains, and Walls
Ferromagnetic materials do not necessarily have a detectable magnetisation in zero magnetic field.
This is well known for iron: dressmaker’s pins are only magnetised when subjected to the magnetic
field of a magnet! When the applied magnetic field is zero, domains of opposite magnetisation ap-
pear spontaneously and cancel each other out in the macroscopic magnetisation. These domains
have been imaged here by magnetic force microscopy (MFM) measurements of the magnetisa-
tion at the surface of an Fe single crystal cleaved in the direction perpendicular to the (111) axis
(see Chap. 10 for more on MFM). The domains appearing naturally here display a structure that
many artists would have been pleased to produce. Scale 40 × 40 μm2 . Image courtesy of Miltat, J.,
Thiaville, A.: Laboratoire de Physique des Solides. Orsay, France
9 Magnetic Anisotropy, Domains, and Walls 271
We have seen that the exchange couplings are responsible for the appearance of
ordered magnetic states, determining for example the Curie temperature TC of the
ferromagnetic state. But everyday experience shows that a piece of iron at room
temperature does not generally exhibit a magnetic moment, even though it is well
below its Curie temperature TC ≈ 1, 000 K. In fact, in order to give it a macro-
scopic magnetic moment, it must be magnetised by placing it in a relatively weak
magnetic field, of the order of a few 10−3 tesla. If this field is then removed after
the application, the piece of iron will retain a significant magnetisation, called
remanent magnetisation. But how can such a weak applied field, corresponding
to such a small magnetic energy compared with kB TC , produce such a marked
effect? The explanation for this phenomenon is to be found in various interactions
that we have neglected in the Heisenberg model, to be discussed in the present
chapter.
In Sect. 9.1 we shall show that several sources of anisotropy related to the crys-
tal structure introduce energy terms that tend to stabilise the magnetisation along
one crystal axis. We can thus understand the existence of magnetic hysteresis, but
not the vanishing magnetisation when a material has never been in the presence of a
magnetic field. This effect can be attributed to the existence of magnetic dipole inter-
actions between the microscopic moments, which we shall investigate in Sect. 9.2.
Since these interactions are long ranged, we shall show that they induce effects
that are very sensitive to the geometric shape of the material. This observation is
quite specific to the magnetic properties of highly magnetised materials. As a rule,
physical properties of materials such as the resistivity, specific heat, and so on, are
bulk quantities that are insensitive to surface effects, insofar as the material has
macroscopic dimensions. They can be measured without regard for the shape of
the sample. Magnetisation is also a bulk quantity, but since there can be no mate-
rial that is not bounded in space, any measurement of M will depend critically on
the shape of the outer surfaces of the material. The effects of demagnetising fields
due to the dipole interaction, and the existence of a demagnetisation energy, will
allow us to understand why, in zero applied field, a magnetic material will spon-
taneously decompose into magnetic domains. The magnetic configuration of the
interfaces between magnetic domains, known as Bloch walls, will be discussed in
Sect. 9.3, where we will determine their formation energy. Taken together, these
phenomena will enable us to understand in Sect. 9.4 the various contributions to
magnetic hysteresis, which is very important for technical applications of magnetic
materials.
In the last chapter, we stressed the quantum origins of magnetism. But in this
one we shall appeal to many semi-classical descriptions, which are adequate for the
task of accounting for the main physical effects. A fully quantum description would
barely change the result, and would increase the complexity to little advantage given
that the relevant quantities are generally hard to predict and must often be deduced
from experiment. However, the reader should bear in mind that these semi-classical
arguments can only be justified if we accept the existence of magnetic moments
and interactions of purely quantum origins. There is no hope here of constructing a
classical model of magnetism by letting Planck’s constant h tend to zero.
9.1 Magnetic Anisotropy
It is observed experimentally that, for single crystals in which M = 0 in the ab-

sence of an applied field, the magnetisation is more easily saturated if a field is
applied in certain crystallographic directions. For example, for nickel, which has a
face-centered cubic structure, M grows most rapidly for a field applied along the
[111] axis, whereas for iron, which has a body-centered cubic structure, M is most
easily established in the [100] direction (see Fig. 9.1). These directions are said to
correspond to easy magnetisation axes. For Co (hexagonal close-packed), the easy
magnetisation axis is the c axis of the hexagonal structure (see Fig. 9.1). This tells us
that certain axes are favoured by the crystal structure, and hence that the rotational
symmetry of the Heisenberg Hamiltonian is not preserved. This Hamiltonian must
therefore be supplemented by terms that take into account the loss of symmetry.
M (kA/m) M (kA/m)
[0001] 1600 [100] [111]

1200 [0001] [110]
1200 [111]
800 [1010]
800 [110]
400
400 [100]
Co [1010] Fe
0 0.2 0.4 0.6 0 0.02 0.04 0.06

μ0 Ha (tesla) μ0 Ha (tesla)
Fig. 9.1 Magnetisation as a function of the applied field for single-crystal samples of cobalt (left)
and iron (right) and for various orientations of the applied field with respect to the crystallographic
axes (T ≈ 300 K). Adapted from Monda, K., Kaya, S.: Sci. Rep. Tohoku Univ. 15, 721 (1926); 17,
1157 (1928)
9.1.1 Magnetocrystalline Anisotropy
Magnetic anisotropy is mainly due to the spin–orbit coupling and the crystal field
(see Chap. 8) which introduce correction terms into the energy levels of an ion in the
crystal. These levels depend on the orientation of the magnetic moment with respect
to the crystal axes. In the simplest case, this leads to the addition of an extra term in
the Heisenberg Hamiltonian (8.31):

z 2
HA = D (SR ) . (9.1)
R
Such a term is called the single-ion magnetocrystalline anisotropy. It clearly breaks

the rotational symmetry of the Heisenberg Hamiltonian. More precisely, if D < 0,
the energy is minimised when the spins and hence the magnetisation are aligned
9.1 Magnetic Anisotropy 273
with the ±z axis, which is the easy axis of magnetisation. Otherwise, when D > 0,
the spins prefer the (x, y) plane, which is then the easy plane of magnetisation.
In real solids, the magnetic anisotropy terms can take various forms, but gener-
ally more complicated than the one indicated above. Note in particular that (9.1)
attributes a special role to the z axis, which can be the case for a crystal with tetrag-
onal or hexagonal structure (e.g., cobalt), but which is incompatible with the cubic
structure of many magnetic solids. In these structures, which do not distinguish the
x, y, and z axes, one might try to remedy the problem by writing
y
x 2 z 2
HA = D (SR ) + (SR )2 + (SR ) .
R
y
x )2 + (S )2 + (Sz )2 = S(S + 1) is constant and this term is isotropic. In
However, (SR R R
a cubic solid, the simplest form for an anisotropy term is rather given by
y
x 4 z 4
HA = D (SR ) + (SR )4 + (SR ) . (9.2)
R
Question 9.1. What will be the easy magnetisation axes or planes for this form of
anisotropy?
A further consequence of the spin–orbit interaction is the existence of exchange

anisotropy: the exchange constant in the Heisenberg model is then replaced by
a rank 2 tensor. However, this effect is generally less important than single-ion
anisotropy. The above discussion shows that the exact form of the anisotropy terms
depends heavily on the structure of the solid under examination (and also on the spin
of each atom). It is nevertheless clear that one can expect to find easy magnetisation
axes (or even planes).
9.1.2 Effect of Anisotropy: Irreversibility
If the material is magnetised along the easy magnetisation axis z, an applied field
Ha in a direction other than z will tend to rotate the magnetisation, whereas the
anisotropy term will tend to oppose this rotation. If the applied field lies along the
easy magnetisation axis, it will stabilise this magnetisation if it points the same way
along the axis, but if it is then decreased and reversed to point the other way along
the easy magnetisation axis, it will then favour the opposite direction. There will
therefore be a value of the field for which the magnetisation reverses. We shall see
that the existence of the anisotropy term brings about magnetic hysteresis.
The behaviour of the magnetisation can be calculated as a function of the applied
field by making several simple assumptions. Suppose the exchange interaction is
strong enough to ensure that the spins remain parallel and that the macroscopic
Fig. 9.2 Influence of a mag- z M

netic field Ha applied along
α
the OZ axis on the magneti-
sation M of a single-domain
magnetic crystal. The direc- Z
tion of the magnetisation M is
specified relative to the easy Ha
axis of magnetisation Oz
φ
magnetisation behaves like a vector of constant magnitude M0 = NgμB . In this case,

the effect of the field will simply be a rotation of the macroscopic magnetisation.
One can then carry out a classical treatment of the anisotropy terms and interaction
with the field.
Suppose the easy magnetisation axis is aligned with the Oz axis and α = (Oz, M)
as shown in Fig. 9.2. If the anisotropy term is given by (9.1), the associated energy is
Ea = NDS2 cos2 α . (9.3)
As D is negative, we can define a macroscopic anisotropy energy per unit volume

by K = −NDS2 = −DM02 /Ng2 μ2B . Up to a shift in the energy origin, the energy
associated with the anisotropy term can be taken as
Ean = K sin2 α , (9.4)
the zero minimum value of Ean being obtained when M lies along the easy axis of
magnetisation.
Now apply the magnetic field in a direction OZ, with signed magnitude Ha . The
energy of the coupling between the magnetisation and the applied field is
EZ = −μ0 Ha M0 cos(φ − α) , (9.5)
where φ = (Oz, OZ), and the total energy is therefore
ET = K sin2 α − μ0 Ha M0 cos(φ − α) . (9.6)
The equilibrium positions of the magnetisation are thus given by

dE
= K sin 2α − μ0 Ha M0 sin(φ − α) = 0 , (9.7)
dα
with d2 E/dα 2 > 0.
9.1 Magnetic Anisotropy 275
There are generally two minima. One corresponds to a metastable magnetisation

state. If we begin at a position where the magnetisation is in stable equilibrium,
e.g., Ha positive and infinite, hence M Ha , and the magnitude of Ha is decreased,
the magnetisation remains in the direction corresponding to the same minimum of
dE/dα until d2 E/dα 2 = 0, the condition for this minimum to become unstable. The
magnetisation then flips over to the second minimum for
d2 E
= 2K cos 2α + μ0 Ha M0 cos(φ − α) = 0 . (9.8)
dα 2
If we set μ0 HK = 2K/M0 , and h = Ha /HK , the simultaneous solution of (9.7) and
(9.8) is obtained for
tan α
tan(φ − α) = − , (9.9)
1 − tan2 α
which means that the magnetisation reverses for α = αrev and h = hrev given by
tan3 αrev = − tan φ , (9.10)

and
3
h2rev = 1 − sin2 2αrev . (9.11)
4
In the simple case (φ = 0) where the field is applied along the z direction, α = 0
for all h > 0. If the field is reversed, the reversal occurs for αrev = 0 and hrev = −1.
If the field is then increased from h < −1 with α = π , the magnetisation is stable up
to h = 0. After reversing the field, the reversal occurs for hrev = 1. We thus have a
rectangular hysteresis cycle, with reversal of the magnetisation for Ha = ±HK (see
Fig. 9.3).
−1 0 1
m
◦
φ =0
1 1
20◦
45◦
70◦
90◦
0 0
−1 −1
−1 0 1
h
Fig. 9.3 Hysteresis cycles of the magnetisation component of a single-domain magnetic particle in
the direction of the applied field as a function of the angle φ between the field and the easy axis of
magnetisation. Here m = Mφ /M0 is plotted as a function of h = Ha /HK . Adapted from [14, p. 336]
Wiley 1972
c
Quite generally, the magnetisation has a component
Mφ = M0 cos(φ − α)
along the axis OZ Ha . This follows a hysteresis cycle that can be calculated from
the above equations. The results are graphed in Fig. 9.3 as a function of the magni-
tude Ha of this field for different values of the angle φ between the direction of Ha
and the easy axis of magnetisation.
In the special case where the field is applied perpendicularly to the easy axis of
magnetisation, i.e., for φ = π/2, we find αrev = −π/2 and hrev = −1, but this does
not correspond to M reversing, because m has then already reached the value −1, the
solution of (9.7) being simply sin α = h, i.e., Mx = M0 Ha /HK . The magnetisation
then rotates reversibly in the (Oz, Ox) plane between Mx = M0 and Mx = −M0 as
the applied field varies from HK to −HK (see Fig. 9.3).
Apart from this case, it is clear that the introduction of an anisotropy energy
term explains why there are hysteresis cycles. The reversal of the magnetisation is
an irreversible and sudden process that corresponds to sudden switch to a state of
minimum energy. The total energy dissipated during one cycle is given by the area
of the cycle, which is 4M0 HK , for example, when φ = 0.
However, the above analysis does not explain why a ferromagnetic sample that
has not been subjected to a magnetic field generally has zero magnetisation. One
would expect it to magnetise spontaneously along one of its easy magnetisation
axes. Moreover, it is often found that a much smaller field than HK will produce an
appreciable magnetisation of a ferromagnetic material.
Question 9.2. Can the results of Fig. 9.1 be explained with the model developed above?
Consider instead the case of Co, which has uniaxial anisotropy.
Anisotropies and Broken Symmetries
Note first that all the anisotropic interactions (single ion or exchange anisotropy, dipole in-
teractions) break the perfect rotational symmetry of the initial Heisenberg model. However,
the operation S → −S (which implies M → −M) is still a symmetry of the full Hamiltonian,
and as a consequence, a ferromagnetic state always breaks (at least) this symmetry. In addi-
tion, the anisotropy energies per spin are generally very small compared with the exchange
energy and therefore play a minor role as long as there is no magnetic order (the spins be-
having basically as though they were independent in the paramagnetic phase). On the other
hand, in the ordered state, all the spins are aligned, and thus have common dynamics, which
multiplies the anisotropy energies by the (considerable) number of spins present and thereby
gives rise to quite significant anisotropy effects. More generally, the appearance of a quantity
(the spontaneous magnetisation here) which characterises collective behaviour in a reduced
symmetry phase at low temperature can bestow importance on parameters that would be neg-
ligible in the high symmetry phase at high temperature, e.g., the transition from a liquid to a
solid.
9.2 Dipole Interactions, Demagnetising Fields, and Domains 277
9.2 Dipole Interactions, Demagnetising Fields, and Domains
The anisotropies mentioned so far (single-ion or exchange anisotropies) are short-

range interactions, between nearest neighbours at the very most. However, we have
ignored the dipole interactions between spins given by (8.16). We have seen that
the dipole interaction between nearest neighbours is negligible compared with the
exchange coupling, and quite often even with the magnetocrystalline anisotropy. But
the dipole interaction is long ranged, decreasing as 1/r3 , and it concerns a number
of spins that increases in proportion to the volume. The detailed effects of these
dipole interactions are often hard to calculate owing to the very large number of
spins involved, and the fact that they depend to a large extent on the shape of the
sample.
9.2.1 Demagnetising Fields
Up to now we have decoupled the field sources and the materials. We have consid-
ered currents in external coils as the sources of the field Ha . But this is no longer
valid when we consider a material that is spontaneously magnetic. It is then itself
a source of field. It is well known that a magnetised material and a solenoid with a
current through it are equivalent, creating fields with similar behaviour at long range
(see Fig. 9.4a and b). In the first case, the field is the sum of the fields induced by the
different turns in the coil, while in the second it is the sum of the dipole fields pro-
duced by the different magnetic dipoles. The identity of the resulting fields at long
range arises because the fields produced by a magnetic dipole or by a current loop
are equivalent far from the source. When we study the magnetism of a material, the
field produced by the material itself, both outside and inside, plays an important role
in determining its properties. But this field depends to a large extent on the shape of
the outer surfaces of the material.
To begin with, we consider a uniformly magnetised material, and from there we
shall be able to understand why the thermodynamic equilibrium of a given sam-
ple might lead to situations where the magnetisation is non-uniform. For the little
Fig. 9.4 Equivalence of a

solenoid (a) and a cylindrical
magnet (b). (c) The field
produced by the magnet can
be determined by considering S
the + and − magnetic charges
at the north and south ends of
the bar (a) (b) (c)
magnet considered above, the field results from the existence of north and south
poles at its ends (see Fig. 9.4c). This is obtained by analogy with the electric field
created by the + and − electric charges at the surface of an electric dipole.
Field Produced by a Magnetisation Distribution
Let us apply Maxwell’s equations to a magnetised material when there are no macroscopic
currents, which means that we consider only the fields due to the magnetic moments. (Fields
due to macroscopic currents are superposed additively, so this restriction does not lead to a loss
of generality.) In this case, curl H = 0 and H is then associated with a magnetic “potential”
U(r)
H = −grad U(r) . (9.12)
Moreover, div B = μ0 div (H + M) = 0, which implies that
U(r) = div M . (9.13)
Since M is zero outside the material, the magnetic potential thus satisfies the above equation
inside the material and U(r) = 0 outside it. The continuity conditions for H and B⊥ at the
surface of the material require U(r) to be continuous at the surface, and

∂U ∂U
− = M·n . (9.14)
∂n int ∂n ext
It can be shown that the solution to the problem summarized by equations (9.13) and (9.14) for
U(r) is

1 div M(r ) 3 n · M(r ) 2
U(r) = −
d r +
d S . (9.15)
4π V |r − r | S |r − r |
We shall not prove this here, but this is similar to the problem for the electrostatic potential
produced by a charge distribution which satisfies analogous equations. M results from the
magnetic dipole moments, which can be considered as made up of fictitious magnetic charges
of opposite signs. The magnetic potential goes as 1/r, like the electrostatic potential.
The H field produced by the magnetic dipoles is analogous to the electric field produced
by a distribution of electric charges. The first term in (9.15) is the potential associated with
the bulk distribution of these magnetic dipoles. It is zero if the magnetisation is uniform, as
for the electric field created by a uniform distribution of electric dipoles. The second term is
the potential associated with the surface density of magnetic poles σ = M · n, which arises
from the discontinuity in the magnetisation at the surface. Its equivalent in electrostatics is a
distribution of electrical charges at the surface. The charges are positive (north pole N) when
the magnetisation points out of the surface, and negative (south pole S) when it points inwards.
Equation (9.15) can be used to find U(r) and hence H(r) = −grad U(r), given M(r ) at all
space points r . Although these expressions are in principle rather simple, it is quite a different
matter to implement them. We shall not go into great detail with such calculations, but instead
aim for a qualitative understanding of the physical effects that follow from these dipole fields.
For example, for the little uniformly magnetised cylindrical rod, the magnetic potential is sim-
ply due to the second term associated with the discontinuities in M(r ) at the ends of the rod,
i.e., the surface densities +M and −M corresponding to the north and south poles N and S of
the permanent magnet. Outside the rod, we obtain the standard field lines of the dipole which
leave the north pole N and curve round to enter the south pole S.
a. Influence of Sample Shape on Magnetisation
In order to understand the importance of these magnetostatic effects, consider the

small cylindrical rod of ferromagnetic material magnetised along its axis in the ab-
sence of any applied field. The field Ĥd produced inside the rod by the north and
south poles N and S is of opposite sign to the field produced outside the rod and
along its axis. It opposes the magnetisation of the rod (see Fig. 9.5), which is why it
is called the demagnetising field.
The magnetisation of the rod is not therefore the one that characterises the ma-
terial when the applied magnetic field is zero, but rather the one corresponding to
the material when it is subjected to its own demagnetising field Hd with opposite
sign to M. To minimise this effect due to the shape of the sample, we must take an
infinitely long rod, so that the magnetic poles are pushed out to infinity. In this case,
this effect due to the demagnetising field will disappear, and the material will really
be subjected to zero field.
The argument is the same if we place a magnetic material in an applied magnetic
field Ha parallel to its axis. The field produced by the dipole adds to the external field
and modifies the field lines so that they are no longer parallel (see Fig. 9.6a, b). The
magnetisation of the material induces a demagnetising field Hd (M) inside the rod
which opposes the applied field.1 The field seen by the material is the effective field
Hi = Ha + Hd (M) ,
rather than Ha . If M, whilst being uniform, depends on the applied field, as we saw
for example in Sect. 9.1.2 when there is anisotropy, then the magnetisation measured
with a magnetometer is not M(Ha ), but M(Ha + Hd ). To make a direct measurement
H B
H
(a) (b) (c)
Fig. 9.5 Demagnetising fields of a short cylindrical sample, assumed uniformly magnetised.
(a) The magnetic poles appearing at the surface induce the magnetic field distribution illustrated in
(b). Inside the sample, the demagnetising fields oppose magnetisation. (c) Spatial distribution of
the induction field
1 Note also that, in this example of the short cylindrical sample, (9.15) is such that H (r) depends
d
on the position r inside the rod, and then the magnetic material is not subjected to a uniform field.
Apart from the physical effects which may lead to non-uniformity of B inside the material, such as
vortices in a superconductor, the sample shape can itself lead to a non-uniform distribution of the
effective field Ha + Hd as seen by the sample.
(a) (b) (c)
Fig. 9.6 Effect of the demagnetising field in the presence of an applied field. Field distribution
outside samples for which the demagnetising field is strong (a) and (b) or weak (c)
of M(Ha ), one must use a long cylindrical sample in order to make the edge effects
negligible (see Fig. 9.6c).
b. Demagnetising Fields for Ellipsoidal Samples
To be able to assess the effects of demagnetising fields, we shall consider the spe-
cial case of an ellipsoidally shaped material, for which it can be shown that the
demagnetising field inside the material is uniform if the magnetisation is uniform
(see Fig. 9.7). In this case,
Hd = −NM , (9.16)
where N is a tensor of unit trace whose eigenvectors are the three principal axes
x, y, z of the ellipsoid. In this case, the effective field Hi seen by the sample is
Hi = Ha + Hd = Ha − NM . (9.17)
c. Expression for the Tensor N in Several Special Cases
For an ellipsoid of revolution about the z axis, we have by symmetry

1
Nx = Ny = (1 − Nz ) . (9.18)
2
In the special case of an elongated ellipsoid, for which the ratio r of the diameters of
the ellipsoid in the directions z and (x, y) is greater than unity, using the formalism
developed in (9.15), we find

1 r
Nz = 2 ln r + (r − 1)
2 1/2
−1 . (9.19)
r − 1 (r2 − 1)1/2
Fig. 9.7 Distribution of the

magnetic induction in an el-
lipsoidal sample. B, H, and M
are uniform inside the sample
N
In particular, for an infinitely long cylinder, obtained as the limit of an ellipsoid

whose elongation in the z direction tends to infinity, i.e., r infinite, Hd vanishes for
Ha Oz, as indicated intuitively above [we do indeed obtain Nz = 0 from (9.19)],
so that
Nx = Ny = 1/2 .
Likewise, an infinite plate parallel to the (x, y) plane corresponds to the limit of an
ellipsoid of revolution whose diameter perpendicular to the axis of revolution tends
to infinity. The magnetic poles are pushed out to infinity for a field applied parallel
to the plane of the plate, implying that
Nx = Ny = 0 , then Nz = 1 .
Finally, note that, in the case of a sphere, we clearly have
Nx = Ny = Nz = 1/3 .
In order to identify the properties of the material in the absence of demagnetis-

ing field effects, we must use the value of M to determine Hi given the known
value of N. The magnetisation curve in the absence of demagnetising field effects
is given by the function M = f (Hi ). These effects due to the geometric shape of the
sample can seriously affect the value of M, especially in the case of superconduc-
tors or ferromagnetic materials, with significant consequences, as we shall see in
Chap. 10.
Question 9.3. Consider a spherical sample of a material with magnetisation proportional to

the field, i.e., with a susceptibility χ̃ = M/H. Determine the apparent susceptibility, taking
the demagnetising field into account. What would be measured for an insulating sample
containing no paramagnetic ions and for which χ̃ = −10−5 ? Or again, for a superconductor
exhibiting a perfect Meissner effect?
9.2.2 Demagnetisation Energy and Magnetic Domains
We saw above that the magnetic fields produced by the elementary magnetic dipoles
induce demagnetising fields that can influence the magnetisation of a uniformly
magnetised sample. But more generally, the energy associated with these interaction
terms between spins must be taken into account in order to describe the ground state
of the spin system. The problem we have already highlighted above is that these
terms depend on the shape of the free surfaces of the magnetic material. We begin
by determining the energy associated with these dipole interactions, which we shall
call the demagnetisation energy.
a. Demagnetisation Energy
If hij is the dipole field created by the moment μj at the site i, the energy associated
with the dipole couplings is
1
Ed = − μ0 μi · hij , (9.20)
2
ij
where the factor of 1/2 is introduced to ensure that we only count the coupling
energy between μi and μj once. Seperating the sums over i and j
1 1
Ed = − μ0 μi hij = − μ0 μi Hd (i) , (9.21)
2 2
i j i
where Hd (i) is nothing other than the value of the demagnetising field at site i. Going
to an integral over the volume,2 the demagnetising energy is thus

1
Ed = − μ0 M(r)Hd (r)d3 r . (9.22)
2 V
For a uniformly magnetised ellipsoidal sample, this term can be written in the form
μ0
Ed = M·N·M , (9.23)
2
using (9.16). At thermodynamic equilibrium, a magnetic configuration can be

established that minimises the demagnetisation energy.
2 The replacement of the sum by a volume integral is not rigorous in the vicinity of site i. In this
case, one must retain a discrete sum in a small region around this site. The correction to (9.22) is
found to correspond to a crystalline anisotropy term of dipole origin. This can be included with the
other magnetic anisotropy terms that are difficult to obtain in any other way than by experiment.
Question 9.4. Consider a sample with ellipsoidal external shape and uniform magneti-
sation. What will be the orientation of its magnetisation when the magnetocrystalline
anisotropy energy is zero? What about an extensive thin film? Consider the case of an
ellipsoid of revolution whose major axis c lies along the axis of revolution (like a rugby
ball). Give the form of the magnetostatic energy as a function of the angle between the
magnetisation and the axis c. What conclusion can be drawn?
b. Magnetic Domains
We see immediately that, for zero applied field, the configuration corresponding to
uniform magnetisation is not the most favourable. The demagnetisation energy is
only zero for certain geometrical shapes of the sample. To minimise Ed , it is in-
tuitively clear that it will be favourable to divide the sample up into ever thinner
layers in such a way as to minimise the demagnetising field in each individual layer.
Examples of configurations gradually reducing Ed for a parallelepiped of material
are shown in Fig. 9.8. This subdivision could be continued indefinitely, but the num-
ber of magnetic domains is bounded owing to the energy cost associated with cre-
ating interfaces. Indeed, these spatial regions, shaded in Fig. 9.8 and referred to as
Bloch walls, separate regions with different magnetisation orientations. This prob-
lem is reminiscent of the one encountered with type II superconductivity. However,
while in the superconductivity case the physical origin of the subdivision was to
be sought in the microscopic physical properties, the subdivision into domains in
the present case is related to the external shape of the material. In the case of an
infinitely long cylindrical wire, there would be only one magnetic domain along the
axis of the wire, and zero demagnetisation energy, since the magnetic poles would
be at infinity.
Note that the expression (9.22) cannot inform us about all the factors that de-
termine the configurations minimising the demagnetisation energy. The expression
for Ed can be expressed in a form implying that Ed will be minimal if both the de-
magnetising fields and the fields produced outside the material are minimised (see
Fig. 9.8 Different arrange-

ments of the magnetisation
of a ferromagnetic solid. The
demagnetisation energy of the
layers decreases as they are
made thinner, leading to a
simultaneous reduction in the
field produced outside the
material (a) (b) (c)
Fig. 9.9 Closure domains allow the magnetisation to be parallel to the surface of the material and
cancel the field outside
below). This therefore leads to a subdivision into magnetic domains with M parallel
to the surface insofar as this is possible. This is indeed achieved by the subdivision
into layers as shown in Fig. 9.8. However, the field produced at the surface is not
zero, even though it is reduced by these alternating magnetisation configurations. It
may be just as interesting to reduce the surface charges to zero by forming domains
known as closure domains such that the magnetisation is always parallel to the outer
surfaces, as shown in Fig. 9.9.
Question 9.5. Do you think closure domains will be equally favourable for single crystals
of Co and Fe?
Another Expression for the Demagnetisation Energy
In the general case, we integrate H · B over a volume V enclosing the whole magnetic material.
Using the expression (9.12) for H as a function of the magnetic potential U,

H · Bd3 r = − B · grad Ud3 r = − ∇(UB)d3 r = − UB · ndS . (9.24)
The second equality is obtained using the identity ∇ · (UB) = B · grad U + Udiv B, in which
the second term is zero. If the integration volume is extended over the whole of space, we note
that U goes as 1/r at infinity, and hence B = μ0 H goes as 1/r2 , so the surface integral vanishes.
The integral of H · B over the whole of space is therefore zero, and

H · Bd3 r = μ0 H · (H + M)d3 r = μ0 H2 d3 r + μ0 Hd · Md3 r = 0 . (9.25)
space space space V
In the second equality, we have decomposed the integral into two terms and used the fact that M
vanishes outside the magnetic material of volume V, within which the field is none other than
the demagnetising field Hd . This implies another expression for the demagnetisation energy:

μ0
Ed = H2 d 3 r , (9.26)
2 space
where the integral is carried out over the whole of space, both inside and outside the sample.
This second expression shows that Ed is always positive. In order to minimise Ed , the field
must be minimised everywhere, both inside and outside the sample. In particular, this means
avoiding surface charges which create a magnetic field outside the sample.
9.3 Bloch Walls 285
9.3 Bloch Walls
The formation of domains leads to a gain in magnetostatic energy, but it requires the
formation of interfaces. As we have already seen in the case of superconductivity,
the formation of an interface generally involves an energy cost. It is the balance be-
tween these two energy terms that will eventually specify the minimum energy state.
So what is the energy of formation of an interface? It might be thought a priori
that there is a sharp boundary on the atomic scale, and that the moments to the right
of this wall will all be oriented downwards, while those to the left will all be oriented
upwards. In a simple cubic crystal, such a spin arrangement would cost an exchange
energy equal to 2S2 |J | per primitive cell of the wall. It is easy to see that this is not
the lowest energy arrangement. It is better to rotate the spins progressively through
a small angle for each lattice spacing, as shown in Fig. 9.10, because the angle
between two neighbouring spins is small, so the cost in exchange energy will be
reduced, even if the number of spins for which the exchange energy is not minimal
is actually increased.
Exchange tends to favour a ‘spreading’ of the wall, in that it makes the angle
between neighbouring spins as small as possible while extending the region over
which the transition between the ↑ and ↓ orientations takes place.
On the other hand, the existence of magnetocrystalline anisotropy has the oppo-
site effect. If the wall is sharp between two domains for which the magnetisation lies
along an easy axis of magnetisation, the anisotropy term is minimised on either side
of the wall, and the cost in anisotropy energy is zero. But if the magnetisation rotates
gradually through the wall, there will on the contrary be an anisotropy energy cost
which increases as the width of the wall (indeed the number of spins deviating from
the easy magnetisation direction increases as the width of the wall). Any magnetic
anisotropy thus tends to make the wall ‘thinner’. We thus see that the configuration
that minimises the wall formation energy will depend on the relative values of the
exchange energy and the magnetic anisotropy term in the Hamiltonian of the spin
system.
To be more precise, consider a specific case in which the anisotropy term takes
the simple form (9.1) with D < 0, giving an easy axis of magnetisation parallel
z Wall
1
ζ y
2
Fig. 9.10 Spin configuration θ Domain
in a Bloch wall. The magneti-
x
sation rotates steadily about
the x axis in the (y, z) plane.
The angle θ = (z, M) varies
from 0 to π for increasing x.
Adapted from [14, p. 289] x
Wiley 1972
c
to the z axis. To obtain the structure of a wall, the spins are treated as classical
vectors by writing S = S(sin θ cos φ, sin θ sin φ, cos θ ). A wall separates regions with
S = (0, 0, +S) and S = (0, 0, −S). If we assume that the wall is parallel to the (y, z)
plane, the two values of S are reached for |x| → ∞, and the orientation of a spin SR
depends only on the x component of its position (see Fig. 9.10).
To simplify, consider also that we are dealing with a simple cubic lattice with
cubic unit cell of side a, and assume that the x component of the spin is everywhere
zero (φ ≡ π/2). Under these conditions the total exchange and anisotropy energy is
given by

E= −J Sl ·Sl+1 +DSz2 = −J S2 cos(θl+1 −θl )+DS2 cos2 θl , (9.27)

l l
for a row of spins along the x axis, with coordinates x = la indexed by l.

The minimal energy configuration of the wall is obtained by minimising E with
respect to any change in the orientation of an arbitrary spin of the wall, that is, by
setting dE/dθl equal to zero, which yields the condition

J sin(θl+1 − θl ) + sin(θl−1 − θl ) + D sin 2θl = 0 . (9.28)
We may also assume that the angle between neighbouring spins is small. We may
then:
• replace the first two sine functions in (9.28) by their arguments,
• treat θl as a continuous function θ of the variable x = la.
Equation (9.28) then becomes the differential equation
d2 θ
J a2 + D sin 2θ = 0 . (9.29)
dx2
Setting x = x/ς with
J a2
ς2 = , (9.30)
2|D|
this can be written

d2 θ sin 2θ
− =0. (9.31)
dx2 2
This is an Euler equation whose solution must in this case satisfy the boundary
conditions θ → 0 for x → −∞ and θ → π for x → +∞. The solution satisfying
these conditions and for which θ = π/2 for x = 0 is

θ (x ) = 2 arctan ex . (9.32)
9.3 Bloch Walls 287
Fig. 9.11 Dependence of the π 2

angle θ between the spin and π
the z axis on the position in the
wall θ
π 2
−π 2
0
−4 −2 0 2 4
x
This function is plotted in Fig. 9.11. The region in which θ (x ) deviates significantly
from the limiting values 0 and π can be estimated by a linear approximation around
x = 0 (dotted line in Fig. 9.11).
Note that the function θ (x ) satisfies the simple relation
dθ
= − sin θ , (9.33)
dx
and that the linear extrapolation corresponds to a width x = π , or x = π ς for the
wall. The length ς given by (9.30) thus characterises the thickness of the wall.
The energy of formation of the wall can be estimated by substituting this solution
for θ (x ) into (9.27). The increase in energy compared with the value for a uniform
magnetisation for the row of spins defines the energy of the wall Ewall per primitive
cell of the wall surface. It is given by

J S2 a2 dθl 2
Ewall = − DS sin θl .
2 2
(9.34)
2 dx
l
Replacing the sum over l by an integral over x and hence over θ , and then using
dx = −dθ/ sin θ , it is found that the two contributions to the energy of the wall due
to exchange and anisotropy are actually equal, and that this wall energy per area a2
of the primitive cell is given by

Ewall = 2S2 2|D|J . (9.35)
It can be checked that, for fixed J , a zero value of D would correspond to an

infinitely thick wall, with zero energy cost for creation of the wall. On the other
hand, an increase in the anisotropy energy, i.e., an increase in |D|, raises the energy
required for the creation of the wall, but reduces its thickness ς . Similar calculations
can be carried out for other forms of the anisotropy energy, or for walls in which
the orientation of the magnetisation changes by an angle less than π . The results
only differ from those obtained here by numerical factors in the expressions for
the energy and thickness of the wall. However, it should be noted that this type of
calculation is only valid for small anisotropy energies, for which the thickness of
the wall remains greater than the lattice constant a. In this case the step from (9.28)
to (9.31) is legitimate. In the other case, |D| J , the wall corresponds to a sudden
reversal over one primitive cell and its energy is simply 2S2 J . In many materials,
the anisotropy energy is small enough for the energy cost of the walls to be too low
to prevent the formation of domains. The wall energy will nevertheless limit the
number of walls per unit volume and will thus determine the size of the magnetic
domains.
Note that this investigation of the Bloch wall shows that, in a magnetic material,
the balance between exchange forces, anisotropy, and demagnetising fields can lead
to complex spatial configurations for the magnetisation on the microscopic scale.3
9.4 Magnetic Hysteresis Cycles
We have set up the tools required to understand how a magnetic material divides up
into alternating magnetisation domains. The equilibrium configuration in zero field
for a sample of given shape results from minimisation of the sum of the exchange,
anisotropy, and demagnetisation energies. It is the minimisation of this demagneti-
sation term of dipole origins that imposes the absence of macroscopic magnetisation
of a material in zero field conditions.
9.4.1 Bitter Method for Observing Domains
It was obviously important to be able to determine the domain configuration of

magnetic materials by experimental means. Now we know that a magnetic mate-
rial subjected to a magnetic field gradient suffers a force. When it became possible
to make large amounts of small magnetic particles with well determined sizes, in
particular for the fabrication of magnetic recording tapes, it was interesting to see
what would happen when such particles were sprinkled on the surface of a mag-
netic material. The little particles are attracted by regions where the gradient of the
magnetic field is high. This is the idea behind the method used by Bitter in 1931
to visualise the Bloch walls in ferromagnetic materials. Consider, for example, two
domains with opposite magnetisations. The field induced is tangent to the surface
except in the region above a Bloch wall, where the moments in the wall induce a
field component perpendicular to the surface. The field gradient is then high at the
surface, in the region where the Bloch wall emerges. If a suspension of magnetic
particles in a fluid is deposited on the surface, the magnetic particles (usually ferrite
Fe3 O4 ) naturally tend to move into this region (see Fig. 9.12a). The walls can then
be visualised by optical microscope, as shown in Fig. 9.12b and c.
3 A relatively simple case is discussed in Problem 20: Magnetism of a Thin Film.

9.4 Magnetic Hysteresis Cycles 289
Fig. 9.12 (a) Bitter’s method

for observing domain walls.
The magnetic powder tends to Surface
accumulate vertically above
the Bloch wall. (b) The wall
appears lighter when observed Wall H=0
optically with transverse (a)
illumination. (c) Observation
of the walls at the surface of
an iron single crystal with
excellent crystal quality. The
displacement of the walls un- H
der the application of a mag-
netic field can be observed.
Adapted from [14, p. 293]
Wiley 1972
c
(b) (c) H
9.4.2 Magnetisation Process and Hysteresis
We may wonder what happens when a material is magnetised by increasing the

applied magnetic field Ha ? Consider the relatively simple case of a single-crystal
material. Initially, the domains oriented parallel to the field will grow, to the detri-
ment of those that are antiparallel to it, as can be seen in Fig. 9.12c.
In a weak applied magnetic field, the displacements of the walls are reversible,
leading to reversible changes in the magnetisation. This is called the Rayleigh
regime, shown schematically in Fig. 9.13 (H = 0 → H1 ). Above a certain field
strength Ha , sudden displacements of the walls can occur and domains can disappear
altogether (between H1 and H2 ).
Finally, for sufficiently strong Ha , the magnetisation increases by rotation of do-
mains (H2 → H3 ), until it reaches its saturation value, when all the moments are
parallel to Ha . If the applied field is now reduced, the magnetisation decreases more
slowly. Note in particular that, for Ha = 0, the magnetisation has a nonzero value Mr
called the remanent magnetisation, which can only be reduced by applying a field
in the opposite direction. The field −Hc needed to cancel the magnetisation is called
the coercive field.
If a sample is subjected to a magnetic field cycle of the form (0, H), (H, −H),
(−H, H), the change in M follows a hysteresis cycle, as shown in Fig. 9.14. The
H3
H2
H1
H=0
Fig. 9.13 Typical domain configurations in a ferromagnetic material under increasing field
Fig. 9.14 First magnetisation M Reversible rotation of

of a ferromagnetic mate- magnetisation
rial showing the reversible
Rayleigh regime. The hys- Ms
teresis cycle obtained after Mr Irreversible
applying a high field Ha displacements
shows the saturation mag- of walls
netisation Ms , the remanent O →χ
magnetisation Mr , and the co- −Hc
ercive field Hc Hc
Rayleigh
regime
−Ms
two quantities Mr and Hc characterise the hysteresis cycle qualitatively. This cycle
has a more or less rectangular shape, depending on how suddenly the magnetisation
reverses, which in turn depends on the system. The fact that a frictional force pre-
vents the Bloch walls from moving around freely leads to a dissipation of energy
given by the area enclosed by the cycle. Mr and Hc are key quantities for specifying
applications of magnetic materials.
In order to make permanent magnets, we seek materials with high remanence and
high coercive field (see Fig. 9.15), referred to as hard ferromagnetic materials. In
contrast, soft ferromagnetic materials, for which Hc is negligible, have small losses
and high susceptibility. They are used to build transformers, inductances, and read
heads for hard disks. For magnetic recording media, we use materials with square
hysteresis cycles and with a coercive field that is not too high, in order to facilitate
magnetic inscription.
300
Mr
Ms
180
200 K
60
300 K
μ0 M (tesla)
425 K
−60 Hc
−180
−300
−2 −1 0 1 2
μ0 H (tesla)
Fig. 9.15 Characteristic hysteresis cycle of a hard ferromagnetic material. Figure courtesy of
Pascard, H.: LSI. Ecole Polytechnique, France
9.5 Summary 291
9.5 Summary
z 2
Magnetic anisotropy terms, e.g., HA = D R (SR ) must be added to the Heisenberg
Hamiltonian, whereupon they determine the equilibrium orientations of the mag-
netic moments relative to the crystallographic axes, but not their directions. For this
form of anisotropy, the z axis is the easy axis of magnetisation of the material.
The fact that a reversal of all the spins will not modify the energy for the spin
system leads to the existence of magnetic hysteresis and an associated energy dis-
sipation. We can define an anisotropy energy K = −DM02 /Nμ2B and an anisotropy
field HK = 2K/μ0 M0 , which corresponds to the reversal of the uniform magnetisa-
tion when the magnetic field is applied along the easy axis of magnetisation.
Magnetic dipole interactions generate magnetic fields both outside and inside
magnetised materials, and these depend on the external shape of the material.
The fields produced can be attributed to fictitious magnetic charges that appear at
the surface of the material and whose surface density is given by the component
of the magnetisation normal to the surface. The fields Hd (r) created within the ma-
terial are demagnetising, i.e., opposed to M. For uniformly magnetised materials
with ellipsoidal shape,
Hd = −NM ,
where N is a tensor of unit trace with eigenvectors along the three principal axes
x, y, z of the ellipsoid. These demagnetising fields have a significant effect on the
magnetisation of a sample. This magnetisation M(Ha + Hd ) is indeed strongly af-
fected by the external shape of the sample.
The energy associated with the dipole couplings between spins can be written in
the form

1
Ed = − μ0 M(r)Hd (r)d3 r .
2 V
In zero field, the minimisation of the total energy of the spin system, sum of the
exchange energy, the anisotropy energy, and the demagnetisation energy, leads to the
formation of magnetic domains magnetised in different directions. This magnetic
domain structure depends on the external shape of the sample. The minimisation of
the demagnetisation energy implies a domain configuration minimising the external
fields, i.e., the surface charges.
Two domains differ through the direction of their magnetisation M. They are
separated by a Bloch wall in which M rotates steadily between its directions in the
two domains. In the case of a single-axis anisotropy, the thickness ς of the wall and
its energy per primitive cell are given respectively by
J a2
ς2 = , Ewall = 2S2 2|D|J .
2|D|
Images of the distribution of magnetic domains at the surface of a material can be
obtained by Bitter’s method, which consists in sprinkling a powder of submicron
magnetic particles on the material. These tend to accumulate where the Bloch walls
reach the surface.
An increasing applied magnetic field gradually magnetises a material through
the displacement of the Bloch walls, then by rotation of the magnetisation in the do-
mains. The fact that there are irreversible processes in which the walls are displaced
and domains disappear leads to a magnetic hysteresis cycle. It is characterised by
the remanent magnetisation Mr obtained in zero field and by the coercive field Hc ,
which is the field for which the magnetisation is reversed.
Question 9.1
We may write
2
Ea = D Sx2 + Sy2 + Sz2 − 2D Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 .
As the first term is isotropic, only the second contributes to anisotropy. It is easy
to check that, if D < 0, the second term is minimal and zero for (100), i.e., when
the magnetisation lies along one of the sides of the cube, as happens for the iron in
Fig. 9.1. On the other hand, for D > 0, the minimum of Ea is obtained for (111), i.e.,
when the magnetisation lies along the diagonals of the cube, as happens in nickel.
Question 9.2
We do not expect to obtain M = 0 for H = 0 as happens experimentally in Fig. 9.1,

and this because the magnetisation of the single crystal is made up of domains. The
curves shown are first magnetisation curves rather than hysteresis cycles.
However, when the applied field is perpendicular to the easy magnetisation axis,
a significant contribution to the magnetisation comes from rotation of the domains.
The situation is then close to the one considered for φ = π/2 in the case of a single
domain. This is valid in particular in the case of Co in Fig. 9.1. The gradient of the
curve in the [1010] direction corresponds to
M Ha
= ,
M0 HK
and then
M M2
χ= = 0 .
μ0 Ha 2K
The gradient at the origin gives χ = 1.4 × 106 /0.6, and this implies the estimate
K = 4.2 × 105 A/m3 .
Question 9.3
We have
M = χ̃ Ha − M/3 ,
implying an apparent susceptibility
χ̃
χ̃app = .
1 + χ̃ /3
For a weakly diamagnetic material, the correction is negligible because χ̃ 1. For a

superconductor exhibiting a perfect Meissner effect, χ̃ = −1, and the above relation
gives χ̃app = −3/2. Such a sample can be used to carry out an absolute calibration
of a magnetometer.
Question 9.4
For a particle in the form of an ellipsoid of revolution about the z axis (see Fig. 9.16),
1
Nx = Ny = (1 − Nz ) .
2
If Mz and Mp are the components of M parallel and perpendicular to the z axis,
respectively,
2Ed 1 − Nz 2
= Nz Mz2 + Mp .
μ0 2
Since Mz2 + Mp2 = Ms2 ,
2Ed 1 − Nz 2
= Nz (Ms2 − Mp2 ) + Mp
μ0 2
1 − 3Nz 2
= Nz Ms2 + Mp .
2
If Nz < 1/3, Ed is minimal for Mp = 0. If Nz > 1/3, Ed is minimal for Mp = Ms . So
for a sphere, the orientation of M is indeterminate, for a flattened ellipsoid, M lies
in the (x, y) plane, and for an elongated ellipsoid, M lies along the z axis.
Mz M
Mp
Fig. 9.16 Ellipsoid of revolu-
tion for Question 9.4
α
M
φ Ha
Magnetic
tape
N S
Ha
tape thickness
y
Fig. 9.17 Left: Elongated ellipsoid for Question 9.4. Right: Elongated particles with the long axis
aligned with the feed direction in a magnetic tape
Note that these cases correspond to situations in which the magnetic surface
charges are as far away as possible. In the case of an almost infinite plane, the
magnetisation must lie in the plane in order to push the surface charges to infinity.
In the case of an ellipsoid that is elongated along the z axis, with the notation in
Fig. 9.17 (left), the demagnetisation energy can be expressed in the form
2Ed 1 − Nz 2 1 − 3Nz 2
= Nz cos2 α + sin α = + sin α + constant .
μ0 Ms2 2 2
Setting K = (1 − 3Nz )/2 > 0, this has the same form as for the anisotropy energy in
(9.4). The magnetisation will therefore describe hysteresis cycles as in Fig. 9.3. We
thus see that even when there is no magnetocrystalline anisotropy energy, a small
particle will exhibit hysteresis with high coercive field if it has non-spherical shape.
A ferromagnetic material with weak magnetocrystalline anisotropy can therefore
be used to make a magnetic recording tape if the particles have elongated shape,
provided that they lie parallel to the tape feed axis on the polymer substrate (see
Fig. 9.17 right). Such an arrangement is convenient for inscribing a magnetisation
produced parallel to the substrate (see Fig. 10.10).
Question 9.5
In the example, the closure domains are oriented at an angle of π/2 to the magneti-
sation in the main domains. In the case of Co, as the anisotropy is single axis, this
corresponds to a difficult magnetisation axis, and a high cost in anisotropy energy.
On the other hand, in Fe, which is cubic with easy axes lying along the (001) di-
rections, the domains oriented at π/2, i.e., along (100), are also easy axes. These
domains at π/2 are often observed in Fe (see, for example, Fig. 9.12).
Chapter 10
Measurements in Magnetism:
From the Macroscopic to the Microscopic Scale
Contents
10.1 Macroscopic Magnetic Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297

10.1.1 Susceptibility or Magnetisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
10.1.2 Examples of Measurements on Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . 301
10.1.3 Hysteresimeters: Toroidal Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
10.2 Magnetic Surface Measurements and Magnetic Imaging . . . . . . . . . . . . . . . . . . . . . . . . . 304
10.2.1 Microscale Measurements and Surface Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . 305
10.2.2 Towards Submicron Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
10.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
10.4 Answers to Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
296 10 Measurements in Magnetism
The magnetic field induced by vortices at the surface of a superconductor can be measured by
displacing the loop of a micrometric SQUID across the surface. In this way, the penetration of the
magnetic flux in a film of the high-Tc superconductor BiSr2 CaCu2 O8 deposited on a single-crystal
substrate can be detected directly. In this system, the vortices are not arranged in a lattice, because
they are trapped by defects (vortex pinning). Here, the deliberate choice of a substrate comprising
three crystals allows one to generate a film which itself comprises three crystals. Note in particular
the alignments of vortices along the boundaries between the three crystals. The field pointing from
the vortex located at the meeting point of these three crystals has a smaller magnitude than the
fields pointing from the other vortices. This particular vortex corresponds to half a flux quantum
φ0 /2. This experiment establishes a special feature of the cuprate superconductors as compared
with metal superconductors: their band gap vanishes in some directions of the reciprocal lattice
(four-leaf clover, or d type symmetry). Image courtesy of Tsuei, C.C. (IBM Yorktown Heights)
and Raffy, H. (Physique des Solides, Orsay, France). The corresponding experiment is described
in Europhys. Lett. 36, 707 (1996)
10.1 Macroscopic Magnetic Measurements 297
An unadvertised observer usually gains information on the magnetic properties of a

material by testing its interaction with a known magnet, e.g. attraction strength for
a ferromagnet or levitation height for a superconductor. These observations involve
the field induced by one of the materials on the other, and are very difficult to exploit
as the fields depend sensitively on the shape of the materials.
To characterise the magnetic response of a material, it is obviously simpler to
subject it to a uniform magnetic field (Sect. 10.1). However, even in this case, we
have seen in the preceding chapters that the magnetic response of a material depends
strongly on its shape. Unlike many other physical properties of solids, manifesta-
tions of magnetism can be extremely non-uniform, both in superconductors and in
magnetic materials, even when the material is ideal in the sense that its other proper-
ties are uniform. In fact, quantities such as χ and M which are averaged quantities,
represent only the first level of magnetic information.
The ideal observation would be one that could visualise the currents and spins
in every detail. Such information is clearly inaccessible! However, in order to get a
better understanding of the fundamental phenomena, we must try to visualise vor-
tices and walls between magnetic domains, and devise local observational methods.
Some such attempts will be discussed briefly in Sect. 10.2. This is an area where sci-
entific perfectionism encounters practical considerations. The aim is to write data in
magnetic form on substrates with the highest possible density, hence the need to
devise methods for writing and reading data on microscopic or even nanoscopic
scales.
10.1 Macroscopic Magnetic Measurements
All methods for making magnetic measurements use magnetic fields. These fields
must therefore be known to great accuracy. The field produced by a current in a
coil can be accurately calculated if the geometry of the coil is sufficiently well con-
trolled. Moreover, since the field–current relationship is linear, one calibration for
a given current value is sometimes enough. However, if very high accuracy is re-
quired, there are many physical effects that can be used to establish measurement
standards for magnetic fields. Simple but important methods consist of measuring
the Hall voltage induced in a semiconductor sample or the resistance of a sample
with high magnetoresistance. A more sophisticated method to implement, but which
proves much more accurate, consists of measuring the Larmor frequency of the nu-
clear spins contained in a reference sample, e.g., protons in water (see Chap. 11).
Measurements of magnetisation or magnetic induction appeal to simple electro-
magnetic techniques. In Sect. 10.1.1, we shall discuss the main methods currently
used, with a geometry that minimises sample shape effects. Examples of measure-
ments are presented in Sect. 10.1.2, and these will illustrate the way the sample
shape can modify the spatial structure of the magnetism in unexpected ways. Finally,
in Sect. 10.1.3, we discuss the most propitious geometry for studying hysteresis
cycles in ferromagnetic materials with low coercivity.
10.1.1 Susceptibility or Magnetisation
The basic idea of a magnetometer is to detect the flux of the magnetic induction B in
a coil A in which the sample has been inserted, the whole thing being subjected to
an applied field Ha produced by a long external solenoid S0 (see Fig. 10.1a). In this
section, we take the sample and coils to have very long cylindrical shapes in order
to make the effects of the demagnetising field negligible.
The flux inside the coil is Φ = nBS, where S and n are the cross-sectional area and
the number of turns in the coil, respectively, and we assume that this coil perfectly
fits the shape of the sample. This method can in principle be used to make a direct
measurement of B if the field Ha is applied suddenly at time t = 0. The EMF induced
in the coil by the change in the flux Φ(t) is then
dΦ dB(t)
V(t) = = nS .
dt dt
$
If we use a device that integrates V(t), we thus measure V(t)dt = nSB. This method
can be used to measure B directly, but it requires a variation in Ha and hence in M
to carry out the measurement.
In order to get round this difficulty, it is more convenient to consider a system
comprising two opposite coils. In a fixed magnetic field, the sample is extracted
from coil A and inserted in coil B, as shown in Fig. 10.1b. In this case, the total flux
changes from
(nS)A B − (nS)B μ0 Ha to (nS)A μ0 Ha − (nS)B B .
V t V t
A S0 IM A
S S0
R
Supply B
voltage
(a) U (b)
Fig. 10.1 (a) Setup for measuring B by integrating the voltage induced in the coil when the applied
field is varied from 0 to Ha . (b) Setup for measuring M in a fixed field Ha by integrating the voltage
induced in two opposite coils when the sample is extracted from coil A and inserted in coil B.
Adapted from [18, pp. 49–51]
If the two coils are identical, (nS)A = (nS)B and the flux variation when the cylin-
drical sample is moved from A to B is therefore
Φ = −2nS(B − μ0 Ha ) = −2nSμ0 M . (10.1)
With this method, the sample can be kept in the field Ha all the time and the mag-
netisation μ0 M = B − μ0 Ha can be measured directly.
Note that this measurement technique shows why it was necessary to introduce
the notion of magnetisation. We do not know a priori how to measure the spatial
dependence of B, i.e., the spatial dependence of the currents in a superconductor,
or the domains in a ferromagnet. However, we are able to measure the total flux,
or rather the difference in the flux with and without the sample. It is thus natural
to consider that the sample is equivalent to a fictitious surface current density that
produces the field M which must be added to Ha . This is in fact perfectly accurate
in the case of a superconductor exhibiting a perfect Meissner effect. But in all other
cases, M only represents the current density that would produce the$same variation
as the one produced by the sample. The value of B in (10.1) is S−1 (S) B(r)d2 r, the
spatially averaged value of B(r) over the cross-section of the sample.
Question 10.1. Must the sample and coil have the same cross-sectional area in the setups
of Figs. 10.1a and b? What are the advantages and disadvantages of these two methods?
The variation of the flux can be obtained by electronic or digital integration of the
temporal variation of the voltage induced in the coil during extraction. In modern
magnetometers, exceptional sensitivity is obtained using a SQUID to make the flux
measurement. This is done by making the coils A and B with superconducting wire.
The free ends of the coils, joined in series, are reconnected on a superconducting
coil, which is itself inductively coupled to the loop of a SQUID. This forms a flux
transformer. The variation of the flux induced on the SQUID is proportional to Φ
(see Fig. 10.2). It is thus by counting the number of flux quanta in the SQUID loop
External Heating
compensation
SQUID
r.f. excitation
Transformer
Amplifier
A
Coils
and
sample
B
Fig. 10.2 Superconducting flux transformer coupled to a SQUID for measuring the flux variation
in the magnetometer of Fig. 10.1b. Adapted from the technical notes for the SQUID Quantum
Design, San Diego, CA (1990)
Pump
Helium transfer Sample holder
Nitrogen
fill
Double walled vacuum

Primary coils
(excitation
field)
Secondary Thermometer
coils Sample
(detection) Helium gas flow
Heater
Capillary
Fig. 10.3 Magnetic susceptometer allowing to measure the voltage induced by mutual inductance
between the primary winding and two opposite secondary windings. The detected signal in pres-
ence of the sample is proportional to vχ. The thermalised cell in which the sample is housed is
represented
that the magnetisation of the sample can be determined. These magnetometers are
highly versatile, and they are so sensitive that there is no rival method for measure-
ments on weakly magnetic materials or very small samples. It is commonplace to
measure the magnetisation of an atomic monolayer of Fe with an area of only a few
square millimeters.
The magnetic susceptibility χ of a sample can be measured directly using an
alternating exciting field. The latter is produced by an alternating current at fre-
quency ν in a primary coil (see Fig. 10.3). The signal induced across the terminals of
the two oppositely wound secondary coils is zero when there is no sample. The dif-
ference in the mutual inductance due to the sample gives a signal at the frequency ν,
with amplitude proportional to νχ . The signal induced in this kind of susceptometer
can also be measured using a SQUID, if very high sensitivity is required.
A setup known as the Faraday balance measures the force acting on a sample
when it is subjected to a constant magnetic field gradient ∇Ha . The force exerted
on the sample is
F = −μ0 ∇(M · Ha ) . (10.2)
Ha and M are horizontal in the setup of Fig. 10.4 (M is induced by the main compo-
nent of Ha which is horizontal), while a gradient of Ha is produced in the z direction
by appropriately shaped pole pieces on the magnet. In this case,
dHa
Fz = −μ0 M ,
dz
and M can be found by measuring Fz with an electronic balance. For ferromagnetic
materials, the sensitivity required is not very high and less sophisticated methods
Fig. 10.4 Faraday magne- Balance

tometer for measuring the
force acting on a sample
placed in a horizontal field Sample
with vertical gradient
dHa z
z S
dz
Ha z
are still very convenient. In particular, in Sect. 10.1.3, we shall discuss the usual
method for measuring the hysteresis cycles of soft ferromagnetic materials.
Question 10.2. If we needed to carry out magnetisation measurements at different temper-

atures, would the various measurement setups discussed above be equally well suited to the
task?
10.1.2 Examples of Measurements on Superconductors
a. Magnetisation of Type II Superconductors
If the magnetisation of a bulk type I superconductor, e.g., a long cylinder of diameter

d λ, is measured with a magnetometer, a perfect Meissner effect is found for
Ha < Hc . For a type II superconductor, the perfect Meissner effect occurs up to Hc1 ,
but when the field penetrates in the form of vortices, B gradually increases, and the
diamagnetic magnetisation decreases to zero when Ha reaches Hc2 (see Fig. 10.5).
Figure 10.6 shows the magnetisation curves for pure lead (A) and Pb–In alloys
with increasing concentrations of indium (B, C, D).
B M
Hc1 Hc Hc 2
0
Ha
Fig. 10.5 B and M in a type

II superconductor for different 0
values of the applied field Hc 1 Hc Hc 2 Ha
0.06 μ0 M (tesla) A: Pb
A B: Pb In 2.08%
C: Pb In 8.23%
0.04 B D: Pb In 20.4%
C
0.02
D
0.04 0.08 0.12 0.16 0.2 0.24 0.28 0.32 0.36
μ0 Ha (tesla)
Fig. 10.6 Curves of M(Ha ) for Pb–In alloys with different indium concentrations. From
Livingston, J.D.: Phys. Rev. 129, 1943 (1963). With the kind permission of the American Physical
Society (APS). http://link.aps.org/doi/10.1103/PhysRev.129.1943
Question 10.3. What type of superconductors do the magnetisation curves in Fig. 10.6
belong to? What does the area under the magnetisation curve represent? How do the thermo-
dynamic quantities characterising superconductivity vary with the indium concentration?
The results shown in Fig. 10.6 correspond to a system in thermodynamic equilib-

rium, where penetration occurs in the form of a regular vortex lattice. In many ex-
perimental cases, the observations are not as beautiful as those shown above. In
particular, the measured magnetisation often exhibits irreversibility, and can even
change sign (see Fig. 10.7). The existence of irreversibility suggests that it is dif-
ficult to establish a thermodynamic equilibrium situation. This should not happen
in an ideal material. On the other hand, in any real material, structural defects are
inevitable, and can introduce frictional effects which slow down or inhibit vortex
motion. For example, Fig. 10.7 shows the magnetisation curves of a Pb–In 8% alloy
after cold-working, then after longer and longer annealing at room temperature. In
each case, the sample was cooled in zero field and the magnetisation was measured
in increasing then decreasing field.
A
μ0 M (tesla) B
0.04 C
C D
E
0.02 D
A
E
B
0
−0.02
0.04 0.08 0.12 0.16 0.20 0.24
μ0 Ha (tesla)
Fig. 10.7 Magnetisation curves of a Pb–In 8% alloy. The sample, initially cold-worked to intro-
duce structural defects, was then annealed for longer and longer periods at room temperature to
gradually remove them. From Livingston, J.D.: Phys. Rev. 129, 1943 (1963). With the kind permis-
sion of the American Physical Society (APS). http://link.aps.org/doi/10.1103/PhysRev.129.1943
Question 10.4. Compare these results with those for the sample with the same concentra-
tion in Fig. 10.6. What differences do you observe? Explain these observations qualitatively.
Note that the simple fact that there is a free surface can prevent nucleation or elimi-
nation of vortices when the magnetic field is changing.1
b. Effect of Shape on the Magnetisation of a Type I Superconductor:

Intermediate State
We have seen that, as a rule, the existence of demagnetising fields can significantly
alter magnetisation measurements in magnetic materials. This will also be true for
superconducting materials whose diamagnetic magnetisation is high in the Meissner
state. Let us consider the special case of a type I superconducting sphere in an
applied field. It is observed (see Fig. 10.8) that its magnetisation curve does not
exhibit a perfect Meissner effect, because the diamagnetic magnetisation does not
vanish suddenly, but instead decreases gradually when the applied field exceeds a
value Hc .
B −M
μ 0 Hc Mc
0 Hc Hc Ha 0 Hc Hc Ha
Fig. 10.8 Average magnetic induction B and magnetisation M measured for a sphere of type I
superconducting material
Question 10.5. Using the demagnetising fields defined in Chap. 9, show that the absolute
value of the apparent susceptibility of this spherical superconductor in weak fields is greater
than unity. Determine the values of Hc , Hc , and Mc for the case shown in Fig. 10.8 as a
function of the thermodynamic critical field Hc . What happens in the applied field interval
Hc < Ha < Hc ?
10.1.3 Hysteresimeters: Toroidal Geometry
For strongly magnetised materials, it is important to reduce the effects of demag-

netising fields when carrying out a magnetic measurement. To do this, long cylinders
1 See Problem 13: Magnetisation of a Type II Superconductor.

Oscilloscope
(a) (b) a.c.

source
Fig. 10.9 (a) Magnetisation configuration in a toroidal sample. (b) Hysteresimeter used to observe
hysteresis cycles in insulating ferrites with weak coercivity
or thin films can be used, arranged in the direction of the measurement field. How-
ever, it is generally preferable to use a closed magnetic circuit, since this completely
avoids the appearance of magnetic poles. The geometry usually employed is that
of a torus, which is just a cylinder that closes up on itself. In this case, the demag-
netisation energy is minimised when the magnetisation is everywhere parallel to the
surface. There is then no demagnetising field and no field is created outside (see
Fig. 10.9a).
By winding an excitation coil and a detection coil on the torus, B can be measured
directly as a function of Ha . Indeed, the current injected into the excitation solenoid
determines the applied external field Ha , while the flux detected in the measurement
coil can be used to measure B. With an alternating excitation current, the curve
V(I) allows a direct determination of the hysteresis cycle in the material. In this
configuration the flux is uniform in the torus (see Fig. 10.9b).
This method is particularly well suited to measurements on soft magnetic ma-
terials, i.e., those with weak coercive field. In this case the fields Ha required are
also weak and can be produced by a small coil. With an alternating current I, one
can also find out how the hysteresis cycle varies with frequency. The losses asso-
ciated with the pinning of Bloch walls by defects may depend significantly on the
frequency and amplitude of the exciting field (see Chap. 11). These characteristics
are important when ferrites are used in high-frequency electronic devices.
10.2 Magnetic Surface Measurements and Magnetic Imaging
In combination with suitable models, macroscopic magnetisation measurements

help to understand certain features of the thermodynamics of magnetic materials.
However, the existence of domains, walls, and vortex lattices obviously encour-
age the development of methods for observing the distribution of B in greater de-
tail. Several techniques were devised in the second half of the twentieth century to
10.2 Magnetic Surface Measurements and Magnetic Imaging 305
achieve magnetic imaging. They all exploit the fact that the non-uniform distribu-
tion of magnetisation within the material induces a non-uniform distribution of the
magnetic field at its surface.
Most observations are thus made at the surface of magnetic or superconduct-
ing materials. This configuration is also used to record information on magnetic
substrates. The developments made to extend our fundamental understanding of
magnetism have produced methods and materials that will improve magnetic data
storage methods in the coming decades. To begin with, we present methods for car-
rying out microscale measurements. This is the current scale on which magnetic
materials are manipulated by read/write systems in the hard disks of our comput-
ers (see Sect. 10.2.1). Methods for attaining the submicron scale will be outlined in
Sect. 10.2.2, since they are already opening up new prospects.
10.2.1 Microscale Measurements and Surface Imaging
Here we shall distinguish methods that carry out local quantitative measurements
from those that only visualise the distribution of magnetism at the surface either
directly or indirectly.
a. Measurement by Microscopic Probes
The first method developed for the read/write heads of magnetic tapes and hard disks
used a microscale magnetic field sensor. This could be an air gap several microns
across in a closed magnetic circuit (see Fig. 10.10a). When the magnetic substrate
is moved across the air gap, the magnetisation component parallel to the surface
produces a flux variation in the circuit and this induces a voltage in the windings
(see Fig. 10.10b). This method requires a steady movement of the substrate with
respect to the measurement head.
To avoid any limitation of sensitivity due to irregularity in the mechanical dis-
placement of the tape, it is also possible to use sensors giving a signal that is directly
N S S N N S S N
1 0
(a) (b)
Fig. 10.10 (a) Magnetic writing in a tape recorder. Data is written onto the tape by a soft iron
magnet with a very narrow air gap. The signal injected into the coil induces a magnetisation of
the magnetic particles embedded in a polymer film. (b) Method for reading the data on a magnetic
tape. As the tape passes through, the signal induced by the magnetisation of the particles trapped
in the tape is detected and amplified. From [14, pp. 588– 589] Wiley 1972
c
sensitive to the magnetic field. This is what happens in magnetoresistive probes or

probes measuring the Hall effect in a semiconductor. Now that recent progress in
microelectronics can significantly reduce component sizes, sensors have been de-
veloped for determining the magnetic field on a piece of a surface of size less than
a few μm2 .
Sensors of this kind comprising a rod holding a dozen or so Hall effect probes
have been used to study the magnetic induction distribution in superconductors. Po-
sitioned on the surface of a single crystal, they can measure B at points only ten
microns apart. This method cannot be used to obtain either λ or the vortex structure.
In the mixed state, however, it can identify the spatial variation in the average mag-
netic induction associated with the geometric shape of the sample (see Fig. 10.11a
and b). Indeed, we have seen that, when the magnetic field penetrates the super-
conductor, an irreversibility is observed in the magnetisation curves. This may arise
because the vortices produced at the surface are subject to an attractive force that
holds them near the surface. The density of the vortex lattice, i.e., the average mag-
netic induction, thus varies with the distance from the surface.
The measurement made with a given probe is used to study the functional de-
pendence B(Ha ) locally in a region of the crystal where the magnetic induction is
uniform. In superconducting samples with high Tc , in the mixed phase, between Hc1
and Hc2 , it was thereby shown that there is a sudden change in M as a function of
Ha (see Fig. 10.11c). This measurement reveals a phase transition between a low
temperature region with an ordered vortex lattice and a liquid vortex phase at high
temperatures. Indeed, the high value of Tc allows thermal fluctuations to appear
before superconductivity is destroyed. The (H, T) phase diagram of these systems is
(a)
Ha
120 μ 0 Ha
B- μ 0 Ha (10 −4 tesla)
Bi2Sr2CaCu2O8
B z (x) (10 −4 tesla)
T = 40 K 2.6
solid liquid
80
2.8
40
T = 80 K
3.0
0 80 160 54 58 62
Distance (μm) μ 0 Ha (10 − 4 tesla)
Edge Center
(c)
(b)
Fig. 10.11 (a) Arrays of Hall effect probes used for local measurement of the magnetic induction
at the surface of a superconducting crystal. (b) The value of B depends on the distance to the edge
of the sample in the irreversible flux penetration regime. (c) Transition to a liquid vortex phase
between Hc1 and Hc2 in a sample of Bi2 Sr2 CaCu2 O8 . Illustrations courtesy of Konzcykowski, M.:
Laboratoire des Solides Irradiés. Ecole Polytechnique, France
H H
VL
Hc 2 VL Hc2
VS
VS
0
Tc T 2f Tc
Tf T
Meissner Meissner
(a) (b)
Fig. 10.12 Schematic phase diagrams determined for high-Tc cuprate superconductors. VS = vor-
tex solid, VL = vortex liquid. (a) YBa2 Cu3 O7 and (b) Bi2 Sr2 CaCu2 O8 which is more anisotropic
thus very different from those observed in conventional superconductors. The extent
of the liquid vortex phase increases with the anisotropy of these layered materials
(see Fig. 10.12).
b. Magneto-Optical Method: Kerr and Faraday Effects
The method described above makes local measurements, whereas the Bitter method
was able to visualise domains. An important method which can combine observation
and measurement uses the rotation of the polarisation of a light wave when it is
reflected (the Kerr effect, shown in Fig. 10.13a) or transmitted (the Faraday effect)
by a magnetic medium. These optical properties of materials should be distinguished
from birefringence, which is observed in crystallographically anisotropic materials.
1
ple
Sam
2
Image
Analyser
Polariser
Fig. 10.13 (a) Kerr effect. (b)
Magnetic dots obtained by
X-ray lithography on a 1
Au/Co/Au film. The Co layer
comprises 5 atomic planes. 2
The dots are squares of side
0.5 μm. The black squares
are images of dots with
M = +Ms , while the invis-
ible squares correspond to
dots with M = −Ms . Image
courtesy of Jamet, J.P.: Lab-
oratoire de Physique des
Solides. Orsay, France
Magneto-optical effects are generally weaker and tend to be used to study cubic
materials for which there is no birefringence, in the absence of mechanical stresses.
Some amorphous alloys of transition metals and rare earths, or cubic garnets, give
rise to significant magneto-optical effects. While these methods can only be used
for specific magnetic materials, they are nevertheless of great interest because they
allow observations at any temperature and for any magnetic field.2 Moreover, the
dynamics of domain wall displacements can be studied directly.
Future methods of digital magnetic data storage might consist in delimiting do-
mains, or engraving magnetic dots on thin magnetic films (see Fig. 10.13b) with
anisotropy perpendicular to the substrate. For such media, a read method using the
Kerr effect with a laser has long been envisaged. Writing might be achieved by
locally heating the relevant region with a laser beam in the presence of a mag-
netic field. Selective heating of a dot reduces its coercive field and thus alters its
magnetisation. With the development of semiconductor lasers and optical com-
pact disks (CD), such methods might become available in the form of magneto-
optical CDs.
10.2.2 Towards Submicron Scales
The methods discussed above can only be extended to nanometric scales with the
help of modern methods of microscopy. Electron microscopy was used quite early
on, e.g., to visualise vortices in superconductors. The discovery in 1985 of new
methods of microscopy called near-field microscopy, able to make observations on
the submicron scale, opened the way to a new field of investigation.
a. Bitter Method and Microscopy: Vortices in Superconductors
This method was used by Trauble and Essman in 1967 to visualise for the first
time the vortex lattice in a type II superconductor. Indeed, at the surface of a
superconducting sample, in the presence of an applied field Ha in the z direction,
perpendicular to the surface, the field component parallel to the z axis is only sig-
nificant at the points where the vortices reach the surface. Very small magnetic par-
ticles are thus attracted to points vertically above the vortex lines. It is not easy
to transpose the Bitter technique to this situation. Ultrasmall particles measuring
about 10 nm across are required, because the distance between vortices is less than
the micron. The particles deposited on the sample at low temperatures are fixed by
a non-magnetic deposition. A replica of the surface layer was then made and ob-
served by electron microscopy. Figure 10.14 shows quite clearly that the clusters of
ferromagnetic particles form a hexagonal lattice characteristic of the vortex lattice.
2 See for example Problem 19: Magnetism of thin films and Magneto-Optic Applications.
Fig. 10.14 Observation of a vortex lattice using the Bitter method. Images of clusters of very
small magnetic particles deposited below the superconducting critical temperature on the surface
of superconductors are observed by electron microscopy. (a) Note the high level of order in the
case of Nb, on a light background, at 1.2 K in a field of 82 mT. (b) For YBa2 Cu3 O7 , near 15 K
in a weak field (2 mT), the significant level of disorder is due to pinning of vortices by defects.
From [15, p. 151]
More recently, such observations were carried out on high-Tc superconductors. In

this case, one can observe positional irregularities due to defects in the sample.
Question 10.6. In Fig. 10.14, it is observed that, for a sample of niobium, the vortices form
an almost perfect hexagonal lattice. But significant defects are nevertheless visible in this
2D crystal lattice, in the top left corner of the figure. Can you locate them? To get a better
view of them, you should look at the figure at grazing incidence in various space directions.
What kind of defects can you locate?
b. Lorentz Microscopy
The Bitter method played an important role in the first observations of walls and vor-
tices. More recently, electron microscopy has been able to visualise these features
directly. Lorentz microscopy exploits the fact that an electron beam is deflected in a
direction that depends on the sign of the magnetic induction. This method is able to
visualise walls or vortices, and determine the direction of the magnetisation inside
each domain. Indeed, the walls show up as white or black lines, depending on the
relative orientations of the domains (see Fig. 10.15).
More detailed electron microscopy methods exploit the information contained in
changes in the phase of the electron wave function as it crosses the sample. Indeed,
for a beam of electrons with the same kinetic energy, the phase change is directly
related to the circulation of the vector potential in crossing the sample (see Chap. 5).
This difference can be used to determine the magnetisation profile across the sample.
For thin films, the method can give quantitative information.
Incident electrons
(a)
Sample
Wall (b)
Photographic film
(c)
Fig. 10.15 Lorentz microscopy. The electron beam crossing a thin sample (thickness 0.1 μm) is
deflected in a direction that depends on the direction of magnetisation of the domain encountered.
This contrast can be identified by analysing the electron beam (left). Observation by Lorentz mi-
croscopy of a sample of permalloy of dimensions 3 μm × 0.75 μm and thickness 60 nm. The
intensity of the image is related to the component of the magnetisation perpendicular (a) or par-
allel (b) to the long dimension of the sample. The orientation of the magnetisation can thus be
determined (c). From [15, p. 260] (right)
c. Near-Field Microscopy
It is natural to end this chapter on imaging by mentioning recent developments in

near-field microscopy. These techniques consist in scanning the surface of a material
with a very fine tip of nanometric dimensions. We saw in Chap. 2 how the tunneling
effect between a metal tip and a metallic sample could be used to image its surface
topography (Scanning Tunneling Microscopy, STM).
This method cannot be used to study insulating samples. However, a tip on a
very thin and flexible support called the cantilever, as shown in Fig. 10.16 (left), can
be very sensitive to extremely weak forces of attraction between tip and surface.
For example, a non-magnetic insulating tip is attracted toward the surface by the
Van der Waals forces. Any slight bending of the cantilever toward the surface is
then detected by a reflected laser beam, as shown in Fig. 10.16 (right), for different
positions of the sample. The height of the tip above the sample is controlled by Vz .
This method, known as atomic force microscopy (AFM), can determine the surface
topography of many different kinds of material, from insulators, to metals and soft
matter.
If the tip and sample are magnetic, a magnetic attractive force adds to the Van
der Waals force. Since these forces depend on the height z of the tip as zα with very
different values of the exponent α, the two forces can be distinguished by varying z.
As the magnetic forces decrease slowly with z, a magnetic image of the domains at
the surface of the sample can be made by scanning the tip across at a distance of
0.5 μm Flexure
Las
50 μm detector
re
Cantilever Feedback
loop
Tip
10 μm
Piezoelectric crystal
controlling
xyz displacements
Fig. 10.16 Cantilever and microtip made from silicon for an atomic force microscope. Image cour-
tesy of Miltat, J., Thiaville, A.: Laboratoire de Physique des Solides. Orsay, France (left). Setup
for an atomic force or magnetic force microscope (AFM or MFM). The feedback loop controls the
force acting on the cantilever (right)
a few hundred angstrom units from the surface. This method, known as magnetic
force microscopy (MFM), is currently used to resolve magnetic structures on scales
below 0.1 μm (see Fig. 10.17).
Such methods should benefit from future developments in nanolithography,
which will produce ever more precisely controlled tips. It should be possible by
nanolithography to make very small Hall effect and magnetoresistive probes on the
cantilever, allowing more detailed quantitative measurements of local magnetisation
values and eventually leading to imaging on scales of a few tens of nanometers.
Fig. 10.17 Image of magnetic

domains obtained in zero field
on a Pt/Co multilayer by
magnetic force microscopy
(MFM). The scale of the
image is 5 μm × 5 μm.
The resolution is 50 nm.
Image courtesy of Thiaville,
A.: Laboratoire de Physique
des Solides. Orsay, France
10.3 Summary
The most basic information about the magnetic response of a system is its mag-
netisation or macroscopic magnetic susceptibility. These quantities are obtained by
measuring the change in the magnetic flux inside a coil. The most sensitive meth-
ods use a SQUID to detect the flux. A SQUID is a superconducting loop containing
a superconductor–insulator–superconductor (SIS) junction. Sensitivities well below
the flux quantum Φ0 are thereby reached. Determination of the magnetisation of
type II superconductors and hysteresis cycles in ferromagnets illustrate the impor-
tance of such basic measurements.
The existence of demagnetising fields associated with the long range of the
magnetic dipole interaction mean that the sample shape can have important con-
sequences for macroscopic measurements. In type I superconductors, the sample
shape can induce a non-uniform magnetic penetration for H < Hc . In this inter-
mediate state, normal and superconducting regions alternate through the sample.
Detailed measurements of the spatial dependence of the magnetic induction are
made at the surface of a material by using microscopic sensors. Magnetic recording
and reading use magnets with air gaps that have gradually been reduced in size to
reach micron dimensions today in computer hard disk readers. Hall effect sensors
are used to study with great sensitivity the phase diagrams of high-Tc superconduc-
tors in the mixed state. The vortex liquid phase observed in these materials is due to
their anisotropy and the high value of Tc .
Images of the spatial distribution of magnetic induction at the sample surface
can be obtained using microscopy techniques. The Bitter method uses submi-
cronic magnetic powders which tend to aggregate where domain walls or vortices
(in a superconductor) emerge from the sample surface. Magneto-optical methods
exploit the rotation of polarisation of light induced by interaction with the sur-
face of a magnetic medium. Likewise, the interaction of the electron spin with
the magnetisation of a thin sample can produce an observable contrast under elec-
tron microscopy. Near-field methods using the interaction of a tip with a surface to
explore its topography are currently the subject of promising developments. Mak-
ing a magnetic deposit on the tip, the magnetic interaction force between tip and
surface can be ascertained. These methods can simultaneously reveal the surface
topography and the spatial dependence of its magnetisation with a resolution better
than 0.1 μm.
Question 10.1
With these methods, it is not necessary to use a sample with the same cross-section
as the coil. However, in the first case (Fig. 10.1a), the change in B outside the sample
due to the change in the applied field Ha will soon become much greater than the
change in B inside the sample, and this will quickly reduce the sensitivity of the
measurement.
The measurement by the method of Fig. 10.1b is less sensitive to the geometry
of the sample compared with that of the coil. A simple calibration with a sample of
known susceptibility will suffice, e.g., a ferromagnetic sphere with χapp = 3 (see the
solution to Question 9.3).
The first method can measure the instantaneous value of M induced by the change
in the field Ha , while the second will be better suited to a measurement of the ther-
modynamic equilibrium value of the magnetisation in the field Ha .
Question 10.2
We may cool or heat both the coil and sample, or just the sample. In the case of
Fig. 10.1a, the first solution is preferable, in order to maintain sensitivity (see the
solution to Question 10.1). Otherwise, a calorimetric container must be inserted into
the coil or coils, while ensuring that this container does not itself exhibit a magnetic
response.
This type of complication is minimal in the case of Fig. 10.1b, because only
the magnetisation of the moving part of the sample holder will contribute to the
observed signal. In this kind of setup, the coils can be held at fixed temperature,
which is absolutely essential if they are to be made from a superconducting material,
as in the case of a flux measurement made by a SQUID.
Note that these measurements with a SQUID nevertheless require a cryostat con-
taining liquid helium, as would be necessary to make low temperature measure-
ments. On the other hand, this complicates very high temperature measurements,
above 1,000 K, e.g., on materials with high Curie temperature. In such cases, the
magnetisation to be measured is generally high and the Faraday balance achieves a
good compromise.
By using SQUIDs made with high-Tc superconductors operating at the temper-
ature of liquid nitrogen, it should soon be possible to insert very high temperature
ovens inside detection coils, and thus increase the sensitivity of very high tempera-
ture measurements.
Question 10.3
We saw in Chap. 7 that pure metals like Pb are usually type I superconductors,
as shown by the magnetisation curve (A), which exhibits a perfect Meissner ef-
fect. Introducing indium, the material is transformed into a type II superconductor.
A gradual reduction of Hc1 and an increase in Hc2 are observed when the In concen-
tration is increased. $
According to Chap. 6, the area of the magnetisation curve, viz., μ0 MdHa , rep-
resents the condensation energy in zero field of the electronic system in the super-
conducting state. We observe in Fig. 10.6 that this varies very little as a function of
the indium concentration, suggesting that the condensation energy and hence Tc are
barely modified by introducing indium. The main effect of the indium is therefore
to increase the ratio λ/ξ , which explains the transition from type I to type II, then a
gradual reduction in Hc1 and an increase in Hc2 . The effect of impurities and defects
in BCS superconductors is effectively to reduce the coherence length ξ , and also to
increase the penetration depth λ.
Question 10.4
In every case the Meissner effect is perfect at low fields and even seems to lead to
ever higher values of Hc1 as the defect concentration increases. On the face of it, this
contradicts the observations made with different indium concentrations in Fig. 10.6.
However, above 0.24 tesla, the magnetisation disappears and the metal becomes
normal. The value of Hc2 (and hence of Hc1 ) does not seem to be modified by cold-
working and repeated annealing. This therefore suggests that the defects introduced
by cold-working do not alter the thermodynamic properties of the superconductor.
On the other hand, the achievement of thermodynamic equilibrium seems to be
strongly inhibited. Indeed, when an applied field Ha is increased above Hc1 , the
first vortices must appear and the density of the vortex lattice must increase through
the nucleation of vortices at the surface and the compression of the vortex lattice.
This process requires macroscopic displacements of vortices. The defects induced
by cold-working play the role of pinning centers, restricting or even preventing such
motions.
Likewise, after reaching the normal state for H > Hc2 , when Ha is decreased, as
soon as the superconducting state is reached, the magnetic field is screened and leads
to strong diamagnetism, because a high vortex density cannot suddenly penetrate the
medium.
Question 10.5
The initial magnetisation of a spherical sample is given by

M χ̃
M = χ̃ Ha − , which yields M = Ha .
3 1 + χ̃ /3
For a superconductor exhibiting a perfect Meissner effect, χ̃ = −1, so M = −3Ha /2,

and χ̃app = −3/2, which corresponds to a higher apparent diamagnetism than for a
perfect diamagnetic material.
The effective field Hi = 3Ha /2 is equal to the critical field Hc for Ha = 2Hc /3.
Therefore, beyond this value, the superconductivity should disappear. We would
then have a normal metal sphere for 2Hc /3 < Ha < Hc , which would be paradoxical!
It suggests that, for Ha > 2Hc /3, owing to the effects of the demagnetising field, the
behaviour of the material can no longer be uniform. Indeed, it is easy to understand
B
SC
M
SC
M
SC
M
SC
B
Fig. 10.18 For some values of an applied field, a spherical type I superconductor has an interme-
diate state comprising normal metal (M) and superconducting (SC) regions of macroscopic size
that in this case there must be some kind of compromise, in which superconducting
and normal regions are produced (see Fig. 10.18).
For 2Hc /3 < Ha < Hc , the associated magnetisation curve varies steadily from
−Hc to zero, and resembles the curve for a type II superconductor. In Fig. 10.8, we
thus have
Hc = 2Hc /3 , Mc = −Hc , Hc = Hc .
Note that this situation, referred to as an intermediate state, is in no way related to

the mixed state. In a type II superconductor, the penetration of the flux in vortex
lines corresponds to an intrinsic thermodynamic equilibrium state of the material,
whereas the state described here for a type I superconductor is simply induced by
surface effects. So for more than ten years, experimenters were tempted to attribute
the fact that the magnetisation in superconducting samples does not vanish suddenly
above a given field to sample shape effects rather than to the appearance of a mixed
state. This misunderstanding was all the more comprehensible in that it was well
known that demagnetising fields induce a domain decomposition in ferromagnets.
Fig. 10.19 Pair of disloca-

tions A and B observed in the 1 μm
hexagonal vortex lattice in
superconducting niobium
Question 10.6
Two defects are visible. Each corresponds to the addition (or suppression) of a half-
line of vortices which ends at point A or B, as shown in Fig. 10.19. To accommodate
this excess, the lattice is deformed locally. Such defects are called dislocations in
the crystal lattice. They often occur in pairs to minimise local stresses.
In a 3D crystal, a dislocation is obtained in the crystal lattice by inserting an extra
half-plane of atoms in a direction corresponding to an atomic plane. The dislocation
is usually represented by the line of dislocation, which is visualised by the row of
atoms along the edge of this extra half-plane. For example, two 3D dislocations
would be obtained by stacking up planes like the one in Fig. 10.19. The correspond-
ing dislocation lines would then be the vertical lines above A and B. Dislocations
play a key role in the mechanical properties of materials.
Chapter 11
Spin Dynamics and Magnetic Resonance
Contents
11.1 Dynamics in Magnetism: General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

11.1.1 Linear Response and Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
11.1.2 Pulse Response and Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
11.2 Dynamics in Ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
11.2.1 Low Frequency Losses in Ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
11.2.2 Ferromagnetic Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
11.3 Resonance in the Paramagnetic Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
11.3.1 Resonance for a Thermodynamic Ensemble of Spins . . . . . . . . . . . . . . . . . . . . . 331
11.3.2 Nuclear Magnetic Resonance (NMR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
11.3.3 Spin Echoes: Transverse Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
11.3.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
11.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
318 11 Spin Dynamics and Magnetic Resonance
Magnetic resonance imaging (MRI) is a technique developed in the 1980s to visualise the spatial
distribution of the nuclear magnetisation in the body of a patient submitted to a magnetic field
produced by a superconducting coil. The patient lies along the axis of the superconducting coil.
Subsidiary coils specify a straight line along which a uniform field gradient is produced. The Lar-
mor frequencies of the nuclear spins (usually protons in hydrogen) can then be coded as a function
of their position along the straight line. Fourier analysis of the precessing nuclear magnetisation
by the techniques described in this chapter is then able to determine the proton density as a func-
tion of the position on the straight line. A cross-sectional image of the organism is produced by
displacing this straight line across a plane. With this imaging technique, the operator can choose
the cross-section, and also specify various types of contrast between the protons of the organism,
depending on whether they belong to fluid or solid, moving or fixed tissues, and so on. Taken from
Les défis du CEA, 119, 17 (2006), edited by the Commissariat à l’Energie Atomique, France
11.1 Dynamics in Magnetism: General Considerations 319
So far we have considered cases where the response of a magnetic or supercon-

ducting system can be treated as static. However, it is obvious that in most practical
applications the dynamics of the response will be very important. The materials used
in transformers, motors operating at low frequencies, or fast electronic components
have very different physical characteristics.
Furthermore, we have noted that in many cases the response is not linear in the
magnetic field, and may become irreversible, in particular in the case where high
fields are applied. This hysteresis of the magnetisation curves reflects a dissipation
of energy which is determined by the area of the cycle. Materials must be optimised
to reduce such losses, in order to render them usable in setups involving high al-
ternating currents. On the other hand, in weak enough applied fields, the magnetic
response does become linear, and under such conditions the dissipation may be ex-
tremely weak.
In addition, as with many other physical systems, magnetic systems exhibit reso-
nant eigenmodes whose specific features can lead to novel physical phenomena and
a great many applications.
In this chapter, we begin with a simple and very general representation for treat-
ing the magnetic response of a physical system in the linear regime and discussing
dissipation (see Sect. 11.1). In Sect. 11.2, after illustrating this formalism in the case
of a ferromagnet in a weak field, we introduce the idea of magnetic resonance in the
case of a saturated ferromagnet. These general ideas are then used in a more de-
tailed way to discuss paramagnetic systems in Sect. 11.3, as exemplified by electron
paramagnetic resonance (EPR) and nuclear magnetic resonance (NMR). Finally, we
discuss the couplings between electronic and nuclear moments, known as hyperfine
couplings. It is through these interactions that the nuclear moments can be used as
atomic probes for local measurement of electronic magnetic properties.
11.1 Dynamics in Magnetism: General Considerations
In many physical situations, we wish to study the response of a system to some form
of excitation. In a solid, for example, we investigate the response to an electric or
magnetic field. If the excitation modifies the energy level spectrum, a thermody-
namic quantity conjugate to the excitation will be altered. In every case, for a weak
enough excitation, the response will be a linear function of the excitation. The mag-
netisation of an ensemble of paramagnetic ions is a linear function of the applied
field for gμB B kB T, while the magnetisation of a type II superconductor is lin-
ear for H < Hc1 , and so on. However, in these different cases, the response is never
instantaneous, because a finite time is required for the system to reach thermody-
namic equilibrium. If the excitation varies in time, the response will only follow the
excitation if the frequency is low compared with the reciprocal of the time required
to reach equilibrium. Moreover, the energy exchanges between the system and the
energy reservoir constituted by the thermostat will depend on the excitation fre-
quency. This suggests a relationship between the frequency response and the energy
dissipation. Here we shall describe the simple relations between quantities in the
case where the response is linear.
11.1.1 Linear Response and Dissipation
Consider an arbitrary system subjected to an alternating magnetic field. Under al-

ternating excitation conditions, a dissipative phenomenon is revealed by a phase
difference between response and excitation. This happens with the losses in a ca-
pacitor, or with the resistive loss associated with the resistance of an inductance L.
For a magnetic material, the response to an excitation by an alternating magnetic
field
μ0 Ha (t) = B1 cos ωt
will lead to a magnetisation offset by a phase difference δ:
M(t) = M0 cos(ωt − δ) = M0 (cos ωt cos δ + sin ωt sin δ) ,
or
M0 cos δ M0 sin δ
M(t) = B1 cos ωt + B1 sin ωt .
B1 B1
We may thus set
M(t) = χ (ω)B1 cos ωt + χ (ω)B1 sin ωt . (11.1)
Using the complex notation B(t) = B1 eiωt , we note that (11.1) corresponds to the
real part of

M(t) = Re χ (ω) − iχ (ω) B1 eiωt . (11.2)
We may thus define a complex magnetic susceptibility
χ (ω) = χ (ω) − iχ (ω) = |χ (ω)|(cos δ − i sin δ) .
Note that the average energy absorbed by the material per cycle T = 2π/ω and per
unit volume is
T
dB
W =− M dt ,
0 dt
or
T
W= ωB21 χ (ω) sin ωt cos ωt + χ (ω) sin2 ωt dt = π χ (ω)B21 ,

0
which corresponds to an absorbed power

ω
P= χ (ω)B21 . (11.3)
2
11.1 Dynamics in Magnetism: General Considerations 321
The component in quadrature χ (ω) thus corresponds to the energy absorption,

while χ (ω) is called the dispersion. The real part of the susceptibility χ (0) at zero
frequency is the static susceptibility.
11.1.2 Pulse Response and Frequency Response
Here we examine the relation between the frequency response and the temporal
response for a magnetic system. In a linear response regime, an exciting field B(t )
applied at time t for a time interval dt induces a magnetisation m(t − t )B(t )dt at
a later time t. The magnetisation induced at time t by a time-varying magnetic field
B(t ) is therefore
t
M(t) = m(t − t )B(t )dt . (11.4)
−∞
In this expression, the function m(t) can be identified as the temporal response func-
tion to a pulse excitation produced at time t = 0. [To see this, consider an excitation
bτ δ(t = 0), which leads to M(t) = bτ m(t).]
The response to a sinusoidal excitation B(t ) = B1 exp(iωt ) is
t t

M(t) = m(t − t )B1 exp(iωt )dt = B1 exp(iωt) m(t − t ) exp iω(t − t) dt
−∞ −∞
∞
= B1 exp(iωt) m(t) exp(−iωt)dt ,
0
and comparing with (11.1), we find

∞
χ (ω) = m(t) exp(−iωt)dt , (11.5)
0
and hence also

∞
1
m(t) = χ (ω) exp(iωt)dω . (11.6)
2π −∞
Quite generally, in a linear response regime, the frequency response and the pulse
response are related by Fourier transform.
Note that these relations are obtained by assuming that the system has a linear,
stationary, and causal response. The first two of these conditions require a response
function which depends only on t − t and also a monochromatic response for a
monochromatic excitation. The causality relation arising from the fact that the re-
sponse cannot precede the cause leads to a response function m(t) that is zero for
t < 0, and real for all t, because it corresponds to the physical response to a real
excitation δ(0). It can be shown completely generally that these conditions require
the functions χ (ω) and χ (ω) to be interdependent. Knowing one of them for all ω,
one can fully determine the other via the Kramers–Kronig relations [11]:
∞
−1 χ (ω )
χ (ω) − χ (∞) = π P
dω ,
−∞ ω − ω
(11.7)
∞
χ (ω ) − χ (∞)
−1
χ (ω) = −π P dω ,
−∞ ω − ω
where P denotes the principal part, i.e.,

ω−ε ∞
P = lim + .
ε→0 −∞ ω+ε
In these expressions, which apply to many other response functions, e.g., the dielec-
tric constant, the imaginary part χ of the response vanishes for ω → ∞, otherwise
the absorbed power would tend to infinity.1 Note also that χ vanishes by definition
for ω = 0. In the next section, we shall study some simple cases.
11.2 Dynamics in Ferromagnets
Many applications in magnetism concern materials with high magnetic permeability

in transformers, motors, and so on. At low frequencies, metallic magnetic materi-
als, in particular iron-based materials, can be used. In these materials, the resistive
losses due to Foucault currents are the main source of dissipation. This explains
why transformer plates are made from cold-worked metal structures, as a way of
limiting resistive losses. Metallic materials cannot be used in the radiofrequency
range ( 10 MHz) or ultrahigh frequency (UHF) range (GHz). For these applica-
tions ferrimagnetic insulating materials are essential. These ferrites are usually made
of agglomerates of grains (materials called ceramics), and their properties depend
to a large extent on the grain size and porosity, i.e., on their microstructure. The
frequency behaviour of such materials is quite different depending whether they are
non-magnetised, with high permeability, or fully magnetised (saturated magnetisa-
tion). We shall now examine these two situations in turn.
11.2.1 Low Frequency Losses in Ferromagnets
In ceramic materials, the magnetisation is zero when there is no applied field. This
is due to the presence of walls, generally scarce in each grain. The high permeabil-
ity is due to the high mobility of the walls in a structure without defects. While the
1 However, the real part of the response can generally have a finite value for ω → ∞. This is not
true for the magnetic susceptibility, because there is no physical system for which the magneti-
sation can keep up with an excitation at infinite frequency. On the other hand, if we consider the
magnetic permeability μ = 1 + χ as response function, this contains a term that corresponds to the
exciting magnetic field, and μ → 1 for ω → ∞. In (11.7), we have kept χ (∞) for the purposes
of generality.
11.2 Dynamics in Ferromagnets 323
applied field is still very weak, in the so-called Rayleigh regime described in Chap. 9
(see Fig. 9.14), the displacement of walls is reversible. Under such conditions, the
dissipation is due to friction between the walls and defects in the material. Let us
assume that the time required for a wall to reach equilibrium is an exponential func-
tion, with a characteristic time τ determined by the mobility of the walls. In this
case the magnetisation response function m(t) in (11.4) can be written

χ0 t
m(t) = exp − . (11.8)
τ τ
Then, from (11.5), we may deduce that

χ0 ∞ 1
χ (ω) = exp − + iω t dt ,
τ 0 τ
or −1
χ0 1 1 − iωτ
χ (ω) = + iω = χ0 , (11.9)
τ τ 1 + ω2 τ 2
χ0 χ0 ωτ
χ (ω) = , χ (ω) = . (11.10)
1 + ω2 τ 2 1 + ω2 τ 2
This kind of behaviour is often observed in insulating ferrites (see Fig. 11.1).
These ceramic materials are obtained by solid phase synthesis from fine powders.
The initial susceptibility and the relaxation time τ depend heavily on the microstruc-
ture of the material. To achieve minimum losses, very pure materials are clearly
needed, with a minimum of stoichiometric defects and without inclusions. In this
case, the wall mobility is limited by friction with intergrain defects. Detailed studies
60 Dm = 9.5 μm
μ ,μ μ
50
40
Fig. 11.1 Real and imaginary 30
parts, μ and μ , respectively, μ
of the magnetic permeability 20
μ = 1 + χ in two samples
10
of the ferrite NiFe2 O4 as a
function of the frequency ν = 0
ω/2π (logarithmic scale). A 30 μ Dm = 4 μm
relaxation behaviour close to
the one given by (11.10) is 20
observed. Figure courtesy of 10
Pascard, H.: Laboratoire des μ
Solides Irradiés. Ecole Poly- 0
technique, France 10 100 1000 ν (MHz)
for different mean grain sizes Dm in the ferrites NiFe2 O4 do in fact show that τ
and χ0 are proportional to D2m and Dm , respectively. The choice of microstructure
therefore determines the initial permeability and the range of frequencies for which
losses are acceptable in a given application. This exemplifies the key importance of
microstructure for the physical properties of materials. The microstructure is also a
determining factor for critical currents in superconductors, or the mechanical prop-
erties of metal alloys.
11.2.2 Ferromagnetic Resonance
Now consider a ferromagnet in which there are no walls. This would happen either
in the case of a high magnetic field B0 , or because the anisotropy is sufficient to
ensure that the remanent state corresponds to uniform magnetisation. The exchange
coupling maintains the coherence of the spin system, and here we shall only study
the response to a spatially uniform excitation, for which the magnitude Ms of the
macroscopic magnetisation is conserved. These modes correspond to a collective
rotation of all the magnetic moments. They can be obtained by arguments from
classical mechanics.
a. Larmor Precession
To begin with, we consider a situation in which the demagnetising field is zero

and neglect anisotropies. The total magnetic moment is associated with an angular
momentum M/γ , where γ = −gμB is the gyromagnetic ratio. The time derivative
is equal to the couple M ∧ B exerted on the magnetisation. We thus obtain
dM
= γM∧B . (11.11)
dt
This relation shows that, if we apply a fixed magnetic field B0 parallel to the z axis
at time t = 0, the magnetisation will precess at the Larmor frequency
ωL = −γ B0 , (11.12)
about the z axis. Indeed, dMz /dt = 0, and since |M| is constant, M moves with
constant angular frequency ωL on a cone with vertex angle θ = (Oz, M(0)), i.e., the
angle between the applied field and the magnetisation at time t = 0 (see Fig. 11.2).
The Larmor frequency is an eigenfrequency of the spin system. It is determined
by the gyromagnetic ratio, itself fixed by the Landé factor g (see Chap. 8). Note that,
according to (11.11), this motion is perpetual and does not bring M into line with
the z axis. However, this could not be otherwise, because the system we have been
considering is isolated, and the magnetic energy −B0 ·M is a constant of the motion.
In order for the magnetisation to come back to the minimum energy configuration
Fig. 11.2 Larmor precession z B0
dM/dt
M
M∧B
0 y
x
corresponding to thermodynamic equilibrium, there must be mechanisms allowing

energy exchanges with a reservoir. The time required to return to equilibrium will
be determined by the microscopic nature and magnitude of the couplings between
the magnetisation and the degrees of freedom characterising the reservoir.
b. Detection of Ferromagnetic Resonance
A term can be included to allow a return to equilibrium M(B) depending exponen-

tially on time, with characteristic time T1 called the spin lattice relaxation time:
dM M − M(B)
= γM∧B− . (11.13)
dt T1
Projecting this equation in the z direction for a field B = B0 z for which the equilib-
rium position is M(B0 ) = Ms z, we find that
dMz Mz − M s
=− ,
dt T1
and then
t
Ms − Mz = (Ms − Ms cos θ ) exp − ,
T1
for the initial condition considered here. This produces a spiralling return to equi-
librium, as shown in Fig. 11.3 (left).
Under these conditions, we can investigate the dynamic susceptibility of our sys-
tem at equilibrium M B0 Oz, i.e., the response to an alternating magnetic field.
If the latter is parallel to the z axis, the response is zero, since this field does not
couple with the magnetisation. However, we can find the response to a transverse
field, e.g., a field applied in the x direction:
B(t) = B1 cos ωt x . (11.14)

B0 z
B0
M ω /γ
beff
0 B1 x
Fig. 11.3 Left: Larmor precession with relaxation in a ferromagnet. Right: Effective field in the
rotating frame
Experimentally, such a field is easily obtained by placing the sample in a coil per-
pendicular to the z axis. However, since the eigenmode of M is circularly polarised,
it is preferable to use a circularly polarised excitation field. Note that the linearly
polarised field (11.14) is precisely the sum of two circularly polarised fields of mag-
nitude B1 = B1 /2, rotating in opposite directions, viz.,
B+ (t) = B1 (cos ωt x + sin ωt y) and B− (t) = B1 (cos ωt x − sin ωt y) . (11.15)
To solve (11.13) with the excitation field B+ (t) it is convenient to use a frame Ox y z
rotating with angular speed ω about Oz. The time derivative of a vector A in the
stationary frame can be expressed relative to its value a in the rotating reference
frame by
dA da
= +∧a ,
dt dt
where = ωz. Under these conditions, the equation of motion (11.13) in the labo-
ratory frame is given in a completely analogous way in the rotating frame:
dm m − Ms z
= γ m ∧ beff − , (11.16)
dt T1
with a static effective field beff in the rotating frame given by (see Fig. 11.3 right)

ω ω − ωL
beff = B0 + z + B1 x = z + B1 x . (11.17)
γ γ
Setting
m = mx x + my y + (Ms − mz )z , (11.18)
we obtain
dmx mx
= (ω − ωL )my − ,
dt T1
dmy my
= γ B1 (Ms − mz ) − (ω − ωL )mx − , (11.19)
dt T1
dmz m
− = −γ B1 my + z .
dt T1
These equations are not linear, and will lead to special features to be investigated
further in the case of nuclear magnetic resonance. But for the moment, let us con-
sider the linear response, i.e., when mx and my are proportional to B1 , and neglect
all quadratic terms, in particular mz ∼ O(m2x + m2y ). We then have
dmx mx
= (ω − ωL )my − ,
dt T1
(11.20)
dmy my
= γ B1 Ms − (ω − ωL )mx − .
dt T1
Setting dmx /dt = dmy /dt = 0, we immediately obtain the stationary solutions for
which mx and my are independent of time in the rotating frame. These are reached
after a transient regime. Putting mx = χ+ (ω)B1 and my = χ+ (ω)B1 , it follows that
(ω − ωL )T12
χ+ (ω) = γ Ms ,
1 + (ω − ωL )2 T12
(11.21)
T1
χ+ (ω) = γ Ms .
1 + (ω − ωL )2 T12
In the laboratory frame, the magnetisation in the x direction is therefore2
Mx (t) = χ+ (ω)B1 cos ωt + χ+ (ω)B1 sin ωt .
2 Since B (t) is obtained from B (t) by changing the sign of ω in (11.15), we immediately obtain
− +
χ− (ω) = χ+ (−ω) and χ− (ω) = χ+ (−ω). The susceptibilities χ (ω) and χ (ω) corresponding to
the linearly polarised excitation field B1 = 2B1 are thus given by
1 1
χ (ω) = χ (ω) + χ+ (−ω) , χ (ω) = χ (ω) − χ+ (−ω) .
2 + 2 +
It is easy to check that the solutions given in (11.21) are such that, for narrow resonances, i.e.,
when ωL 1/T1 , the −ω components are negligible for ω ≈ ωL , and only the component of
the field rotating in the right direction gives a response close to ωL . We can therefore identify
χ (ω) and χ (ω) with χ+ (ω)/2 and χ+ (ω)/2, respectively, in practical cases. Note in particular
that the expressions for the real and imaginary parts of the susceptibilities given above satisfy the
Kramers–Kronig relations.
χ χ
(ω − ω L )T1
(ω − ω L )T1
Fig. 11.4 Frequency dependence of the real and imaginary parts, χ and χ , respectively, of the
magnetic susceptibility. They correspond to the dispersion and absorption signals of the ferromag-
netic resonance
The forms of χ+ (ω) and χ+ (ω), which are absorption and dispersion signals, are
shown in Fig. 11.4. They are even and odd in (ω − ωL ), respectively. The absorp-
tion signal has a Lorentzian form, precisely because we assumed an exponential
relaxation.
The resonant mode of the magnetisation gives rise to energy absorption at the
Larmor frequency. This resonance can be observed by placing the sample in an
inductance
L, in which there is a current of angular frequency ω. The impedance
iLω 1 + χ (ω) acquires a resistive term proportional to χ (ω) when ω reaches the
Larmor frequency. The resonance will therefore be detected by observing the energy
absorption at the frequency of the excitation field when ω crosses the value ωL . For
experimental reasons, it is more convenient to hold ω fixed and vary the applied
field. The dispersion signal can also be detected in a coil placed in the y direction,
i.e., orthogonal to the coil inducing the excitation field.
c. Ferromagnetic Modes in the Presence of Demagnetising Fields

or Magnetic Anisotropy
In practice, the existence of magnetic anisotropy and/or demagnetising fields will

alter the resonance conditions. We know that, even when it is uniform, the mag-
netisation of a ferromagnet is subject to a demagnetising field Hd determined by the
shape of the sample (see Chap. 9). For an ellipsoid of revolution Hd = −N·M, where
N is the demagnetisation field tensor. Dropping the relaxation term, the equation of
motion (11.11) is
dM
= γ M ∧ B0 − μ0 N·M . (11.22)
dt
If the field B0 is oriented along an eigenvector of N in the z direction, these equations

can be projected in the x and y directions. Terms of second order can be neglected,
assuming that M deviates little from B0 , and we may then write
dMx
= γ B0 + μ0 Ms (Ny − Nz ) My ,
dt (11.23)
dMy
= γ − B0 + μ0 Ms (Nz − Nx ) Mx .
dt
This implies that
d2 Mx
2
= γ 2 B0 + μ0 Ms (Ny − Nz ) − B0 + μ0 Ms (Nz − Nx ) Mx .
dt
The frequency of the eigenmode is thus given by

ω2 = ωL − γ μ0 Ms (Ny − Nz ) ωL − γ μ0 Ms (Nx − Nz ) , (11.24)
which only reduces to ωL in the case of a sphere, since then Nx = Ny = Nz = 1/3.

We can see from Fig. 11.5 that, for a given material, the ferromagnetic resonance
frequency depends on the sample shape.
Question 11.1. 1. For what values of the applied field should one expect to observe the
resonance for the three situations in Fig. 11.5. Compare with the experimental results.
(Assume that the long cylinder can be treated as infinitely long.)
2. Can you estimate the Landé factor and the magnetisation of the ferrite? Use the values
μB = 9.2740 × 10−24 J/tesla and h̄ = 1.05457 × 10−34 J s.
3. In your opinion, what leads to the shape and width of the observed lines?
Y 3 Fe 4.7 Al 0.3 O 12
μ app
15
10
0
2
μ app
0
2
4
0.25 0.30 0.35 0.40
Ba (tesla)
Fig. 11.5 Real and imaginary parts, μ and μ , respectively, of the magnetic permeability μ =
1 + χ as a function of the applied field at a frequency of 9.3 GHz. Measurements for spherical and
cylindrical ferrite samples with two directions of the applied field for the latter. From Berteaud,
A.J.: Le Journal de Physique et le Radium 23, 546 (1962), EDP Sciences copyright
Likewise, when there is magnetic anisotropy, the frequencies of the modes will be
perturbed. For example, for a uniaxial anisotropy with easy axis of magnetisation in
the z direction, an anisotropy energy term of the form −DSz2 leads to a macroscopic
anisotropy energy of the form Ea = K sin2 θ , where θ = (Oz, M). When M does
not lie exactly along the easy magnetisation axis, a restoring couple is exerted on it
which tends to bring it back along the easy axis. The most direct way to determine
the magnetocrystalline anisotropy is to measure this couple for different values of
the applied magnetic field. Note that, if M is close to an easy axis, i.e., if θ is small,
the couple is
dEa d2 Ea
C=− = −θ . (11.25)
dθ dθ 2
θ=0
This therefore adds to the couple B0 Ms θ due to the interaction with the applied
magnetic field. So it is just as though the anisotropy corresponded to a magnetic
field BK which adds to the applied field and modifies the resonance condition (see
Fig. 11.6). For the single-axis anisotropy we are considering here, C = −2Kθ , the
field associated with the anisotropy is BK = −2K/Ms , and the resonance condition

2K
ω = −γ B0 − . (11.26)
Ms
There is therefore a resonance even when no external magnetic field is applied, if

the coercive field is high enough for the sample to be prepared in a state where the
remanent magnetisation is saturated. In practice, in polycrystalline samples, the fer-
romagnetic resonance will be a superposition of resonances corresponding to the
different orientations of the crystallites. Its width will be determined by the distri-
bution of demagnetising fields or anisotropy energy. In general, the shape of the
spectrum is not determined by the relaxation time T1 .
120
[110] [100]
μ
100
[111]
80
60
40
20
0
0.3 0.32 0.34 0.36
Ba (tesla)
Fig. 11.6 Imaginary part of the magnetic permeability μ measured at 9.3 GHz in a garnet single
crystal for different orientations of the magnetic field with respect to the crystal axes. From Dillon,
J.F., Geshwind, S. Jaccarino, V.: Phys. Rev. 100, 750 (1955). With the kind permission of the
American Physical Society (APS). http://link.aps.org/doi/10.1103/PhysRev.100.750
11.3 Resonance in the Paramagnetic Regime 331
d. High Frequency Applications of Ferrites
Note that these resonant modes occur at high frequencies, around 30 GHz/tesla.
In these frequency ranges, which are very important for telecommunications and
correspond to centimeter wavelengths, magnetic materials are very useful in elec-
tronic devices. For such electromagnetic waves, the properties of magnetic materials
are dominated by the response to the magnetic component of the wave. At higher
frequencies, close to optical frequencies, the magnetic susceptibility vanishes, the
magnetic permeability tends to 1, and dielectric properties dominate. The insulat-
ing ferrites are thus used as filters, tuned to a ferromagnetic resonance. The fact
that the resonance only occurs for a wave that is circularly polarised in the right di-
rection allows one to manufacture polarising filters, and consequently isolators and
circulators.
11.3 Resonance in the Paramagnetic Regime
The phenomenon of ferromagnetic resonance introduced above leads to a wide

range of applications. Larmor precession, directly associated with the idea of an-
gular momentum, was introduced earlier for a single spin [2, Chaps. XI and XIII].
We shall show in Sect. 11.3.1 that this phenomenon also occurs for a thermody-
namic ensemble of weakly interacting spins. We shall then be able to introduce
electronic paramagnetic resonance (EPR) and nuclear magnetic resonance (NMR)
in Sect. 11.3.2. In the radiofrequency range of NMR, specific technical possibili-
ties lead to novel experimental methods (pulse NMR), and the observation of spin
echoes (Sect. 11.3.3). This will provide a glimpse of the extended field of applica-
tions in magnetometry (Sect. 11.3.4). NMR allows one to perform magnetic mea-
surements on the atomic scale due to the existence of couplings between electronic
and nuclear moments (Sect. 11.3.4).
11.3.1 Resonance for a Thermodynamic Ensemble of Spins
We consider localised moments, such as electronic paramagnetic entities diluted in

an insulator. The electron magnetic moment μ is proportional to the angular mo-
mentum of the electron, denoted here by h̄S, so that S is a dimensionless quantity
and μ = −gμB S = h̄γ S. The Zeeman interaction between this magnetic moment
and an applied field B0 in the z direction is
ĤZ0 = −h̄γ S · B0 = −h̄γ Sz B0 . (11.27)
For simplicity, we shall restrict here to the case of spin 1/2 (S = 1/2), for which the
Hamiltonian has two eigenstates (see Fig. 11.7) with energies E+ = −h̄γ B0 /2 and
Fig. 11.7 Quantum levels of +1 2 E + = − h̄ γ B 0 2

the Zeeman Hamiltonian
(11.27) for S = 1/2 h̄ ω L = −h̄ γ B0
−1 2 E − = h̄ γ B 0 2
E− = h̄γ B0 /2. The difference between the levels corresponds to the Larmor angular
frequency defined by
E+ − E− = h̄ωL = −h̄γ B0 . (11.28)
To detect this splitting of the quantum energy levels, we must use a wave of angular
frequency ωL that can induce transitions between states Sz = 1/2 and Sz = −1/2.
If an alternating magnetic field with frequency ωL is used, the Hamiltonian of the
perturbation is
ĤZrf = −h̄γ S · B1 cos ωL t . (11.29)
A transition can only occur between the states |1/2 and | − 1/2 if this Hamiltonian
has nonzero matrix elements between these two eigenstates of Sz . This in turn will
only be possible if B1 has a component perpendicular to z. Indeed, only the oper-
ators Sx and Sy have nonzero matrix elements between these two states. With B1
in the x direction, we recover the same conditions as those considered to detect the
ferromagnetic resonance.
The analogy goes much further than this, because it is easy to show that the time
evolution equation for the operator μ̂ in the presence of a time-dependent magnetic
field is precisely the same as the one obtained classically in (11.11), viz.,3
dμ̂
= γ μ̂ ∧ B .
dt
Likewise, if we now consider a macroscopic ensemble of independent spins placed

in a field B0 , the density operator for this system in thermodynamic equilibrium is
exp(−ĤZ0 /kB T)
ρ̂ = . (11.30)
Tr exp(−ĤZ0 /kB T)
At thermodynamic equilibrium, the magnetisation of the system is
h̄γ B0
μ̂z = Tr(ρ̂ μ̂z ) = NμB tanh , (11.31)
2kB T
and μ̂x = μ̂y = 0, where N is the number of paramagnetic centers per unit vol-
ume. If the spins only interact with a magnetic field B, possibly time-varying, it can

3 Simply write dS/dt = (i/h̄) ĤZ , S , with HZ = −h̄γ S · B, and use the commutation relations
between Sx , Sy and Sz to obtain this relation.
be shown that the time evolution equation of the density operator leads to a time
dependence of μ̂ satisfying4
d μ̂
= γ μ̂ ∧ B . (11.32)
dt
So this equation, used for a ferromagnet in Sect. 11.2.2, also applies to the mag-
netisation of a statistical ensemble of localised spins when interactions between the
spins are neglected. Such interactions can be neglected, for example, when the tem-
perature is well above the Curie or Néel temperatures.
The same experimental setups can be used to detect the ferromagnetic resonance
and the paramagnetic resonance of the electronic moments (EPR). Note also that,
in the paramagnetic state, the relaxation time T1 has a simple interpretation: it is
the time required to reach thermodynamic equilibrium. So if at time t = 0 a non-
polarised sample is placed in a field B0 , a time T1 will be required to reach the
magnetisation (11.31) corresponding to the difference in population between levels
−1/2 and 1/2. In many physical situations, experimental determination of T1 pro-
vides a way of investigating the interactions between the electronic moments and
the other degrees of freedom of the solid. It provides useful physical information
about the properties of the given solid.
11.3.2 Nuclear Magnetic Resonance (NMR)
The physical properties of solids, and in particular their magnetic properties, are
determined by the electronic states. On the other hand, the nuclear spin moments,
which do not affect these properties, provide an extremely useful probe for the elec-
tronic properties. Atomic nuclei are made up of neutrons and protons, which are spin
1/2 particles. They are assembled into quantum states in which the nuclear ground
state has a total spin I that may be integer or half-integer. The associated magnetic
moment μn is proportional to the magnetic moment of the proton μp , where the
multiplicative factor is analogous to the Landé factor for an atomic electronic mo-
ment. Each atomic nucleus thus has a specific magnetic moment μn = h̄γn I. The
gyromagnetic ratio γn is known to great accuracy for each of the stable isotopes in
the periodic table. Since μp 10−3 μB , the nuclear moments are extremely small,
as are their mutual interactions. As a consequence, they are almost always in a para-
magnetic state. An ordered nuclear magnetic state is only generally accessible for
temperatures of the order of 10−6 K. According to (11.31), we see immediately that,
in a given applied field, the nuclear magnetisation is about 106 times smaller than
the electronic magnetisation. It is therefore practically impossible to detect using
the methods described in Chap. 10.
4 This is established by
considering the time evolution equation of the density operator, namely
dρ̂/dt = −(i/h̄) ĤZ , ρ̂ .
However, everything we have just said for the electron paramagnetic moments
can be transposed to the nuclear moments. Although the nuclear magnetic suscep-
tibilities are weak, they can be detected by magnetic resonance. As we shall see
below, this technique can significantly increase sensitivity.
a. Technical Principles of NMR
We represent the absorption signal by a function f (ω) like the one given in (11.21),
whose area we assume to have been normalised to unity, viz., χ (ω) = af (ω). The
static nuclear susceptibility χ0 is found from the Kramers–Kronig relation (11.7)
evaluated at ω = 0 :

a ∞ f (ω ) a
χ0 = χ (0) =
dω ≈ ,
π −∞ ω π ωL
because f (ω) ≈ 0 when ω is any distance from ωL . If f (ω) has width ω, then the
intensity detected in an NMR experiment is
χ (ωL ) ≈ af (ωL ) ≈ π ωL χ0 f (ωL ) ≈ π χ0 ωL /ω .
The susceptibility measured in NMR is therefore increased by a factor of about

ωL /ω as compared with the static susceptibility. Since the line widths can be very
small in NMR, this leads to a significant gain in sensitivity, in fact up to 9 orders of
magnitude in high-resolution NMR in liquids.
Nuclear Larmor frequencies correspond to radiofrequencies of around 10 MHz
per tesla. In this frequency range, exceptionally stable coherent sources are avail-
able, with narrower spectral widths than the transitions to be observed. These are
obtained by electronic oscillators with frequency stabilised on the resonant mode
of a piezoelectric quartz crystal. Moreover, for such frequencies, very powerful am-
plifiers are also available. With radiofrequency pulses, these can be used to signifi-
cantly modify the populations of the spin quantum states. Such setups are useful for
studying nuclear relaxation.
b. Radiofrequency Pulse Spectroscopy
Consider the action of a very strong radiofrequency field B1 at a frequency close to

the Larmor frequency, applied for a time tw . In the rotating frame, the effective field
felt by the spin system is, according to (11.17),

ω
beff = B0 + z + B1 x . (11.33)
γ
If we are trying to detect the NMR of nuclei with a frequency spectrum extended
over a width ω/γ B1 , we will obtain (B0 + ω/γ ) B1 , and the field beff will be
very close to B1 x for all the nuclear moments to be observed (see Fig. 11.3 right).
In this case the equation governing the time evolution of m in the rotating frame is,
according to (11.16),
dm
= γ m ∧ beff ≈ γ m ∧ B1 ,
dt
if the duration tw of the radiofrequency pulse is much shorter than the relaxation
time T1 . As a consequence, during the pulse, m rotates about x with angular fre-
quency γ B1 .
Suppose we start out from thermodynamic equilibrium Mz = M0 , and that the
field B1 is applied from time 0 to time tw . In this case, at time tw , the magnetisation
will have rotated through an angle θ = γ B1 tw about the x axis. If we adjust tw so
that θ = π/2, the nuclear magnetisation lies along the y axis after the pulse, and then
remains fixed in the rotating frame. It therefore precesses at the Larmor frequency
ωL in the laboratory frame. This magnetisation rotating in the plane perpendicular to
B0 will induce a voltage oscillating at the frequency ωL in a coil placed in the (x, y)
plane. This signal will fade in time due to the distribution of Larmor frequencies in
the sample, and also the relaxation phenomena.
It is easy to check that the pulse experiment just described constitutes a determi-
nation in the rotating frame of m(t) as given in (11.5) and (11.6). The free preces-
sion signal is in fact the Fourier transform of the NMR spectrum. The two signals
90x
(a)
Start of acquisition
(b)
(c)
Fig. 11.8 Pulse and Fourier

transform NMR. (a) Ra-
diofrequency pulse in the x
direction. Free precession (d)
signal in phase (b) and in
quadrature (c). Spectrum 0 δω ω
obtained by complex Fourier
transform in absorption χ (ω)
(d) and in dispersion χ (ω)
(e). From [Callaghan, P.T.: (e)
Principles of NMR Mi-
croscopy. Oxford University
Press, Oxford (1991), p. 43]
χ (ω) and χ (ω) can be reconstructed by Fourier transform in the complex plane,
by detecting the two precession signals, in phase and in quadrature with the gener-
ator producing the radiofrequency pulses (see Fig. 11.8). This method is currently
the main technique for NMR observations. Note that the crucial condition for this
to work is γ B1 ω, so it will only succeed for sufficiently narrow spectra. This
is not always the case for magnetic materials.
In order to measure the relaxation time T1 , one could apply a pulse of length
2tw , rotating the magnetisation through π and thereby aligning it with the negative
z axis. After a delay of tD , we observe a free precession signal with a π/2 pulse. Its
magnitude is proportional to the magnetisation reestablished at time tD along the z
axis:

tD
Mz (tD ) = M0 − 2M0 exp − . (11.34)
T1
After a time much longer than T1 , thermodynamic equilibrium is regained. A mea-

surement can be made for a different value of tD , reproducing the time evolution
(11.34) in a step by step manner, whereupon T1 can be determined (see Fig. 11.9).
M0
π π/2
tD
π/2
tD
π/2
tD tD
π/2
(a) (b)
Fig. 11.9 (a) Measuring the relaxation time T1 . The equilibrium magnetisation is reversed at time
t = 0. It is then measured after a delay tD . Measurements made for various values of tD can be used
to reconstruct the nuclear relaxation. (b) The resulting evolution is shown for a spectrum compris-
ing two lines. The time tD is given in seconds. From [Sanders, J.K.M., Hunter, S.K.: Modern NMR
Spectroscopy. Oxford University Press, Oxford (1993), pp. 65–66]
11.3.3 Spin Echoes: Transverse Relaxation
In an experiment in which the magnetisation precesses freely, if there is a mag-

netic field distribution in the sample that can be related to microscopic physical
properties, the nuclear spins can lose their phase coherence by rotating in the (x, y)
plane, without exchange of energy with the thermodynamic bath. The free preces-
sion signal may disappear well before the magnetisation along the z axis has had
time to reappear. This suggests that the relaxation may be different in the (x, y)
plane and in the z direction. Indeed, for a saturated ferromagnet, it may be pos-
sible to treat the magnetisation as a classically evolving vector, with conservation
of |M|. But for a statistical ensemble of spins, |M| is not conserved during the evo-
lution to equilibrium. Here it is possible to produce a transverse magnetization as
the system is prepared in a state that establishes phase coherence within the nuclear
spin ensemble. But the microscopic couplings required to establish thermodynamic
equilibrium can destroy this transverse coherence without necessarily reestablish-
ing global thermodynamic equilibrium. This will be the case if dipole interactions
between nuclear spins are taken into account.
Indeed the dipole coupling Hamiltonian for two neighbouring nuclear spins I1
and I2 contains terms of type I1+ I2− /r3 which induce simultaneous reversals of I1
and I2 , e.g.,
z z
I = −1/2; I z = +1/2 ⇒ I = +1/2; I z = −1/2 .
1 2 1 2
Such transitions modify the dipole field felt by neighbouring spins, and hence lead
to a loss of transverse coherence without changing the macroscopic magnetisation
component Mz , because the change in the total spin is zero.
These considerations led F. Bloch to introduce two different relaxation times T1
and T2 for Mz and (Mx , My ), respectively, so that the equation of motion for M in
the laboratory frame (11.13) becomes
dM (Mz − M0 )z Mx x + My y
= γM∧B− − , (11.35)
dt T1 T2
where z is a unit vector along the z axis. This expression is not strictly accurate,
because there is no reason why the loss of coherence in the transverse plane mod-
elled by T2 should be exponential. Moreover, it may stem from different origins. A
width ωi can be associated with a static distribution of the magnetic field of ex-
perimental origin (e.g., non-uniformity of the magnet) or of the Larmor frequencies
due to microscopic differences within the sample. The local fields of dipole origins
between nuclear spins lead to a broadening ωdd of the spectrum. The total width
ω ≈ ωi +ωdd leads to a decrease in the free precession over a time T2∗ ≈ ω−1 .
The physical difference between these two sources of broadening can be demon-
strated by a sequence of pulses giving rise to a spin echo. After a π/2 pulse of
duration tw , a pulse of duration 2tw is applied at time τ > T2∗ , which rotates the
magnetisation through π about the y axis. This amounts to reversing the magnetisa-
tion distribution in the (x, y) plane. It can be seen from Fig. 11.10 that, at time 2τ ,
the magnetisation refocuses along the y axis, because the time evolution associated
with the local field distribution leading to ωi is reversed. Coherence thus reap-
pears at time 2τ in the transverse plane. The so-called spin echo signal increases in
the opposite direction to the free precession, to go to zero once again like the free
precession after a time T2∗ . However, its magnitude will be reduced because only
the static field distributions associated with ωi are reversible. Non-deterministic
(π 2)x (π)y
τ τ
FT FT (1/2 echo)
M0 M0 exp(−2 π T2 )
x
ν
y y y
ν x
x x
Fig. 11.10 Spin echoes. Sequence of radiofrequency pulses giving rise to a spin echo. The pulse
π applied at time τ in the y direction reverses the magnetisation distribution in the (x, y) plane.
The magnetisation then refocuses at time 2τ . From [Canet, D.: La RMN, Concepts et méthodes.
InterEditions, Paris (1991), p. 170]
exp(−t T2∗)
exp(−t T2)
Fig. 11.11 Relaxation time T2 . Spin echoes produced at different values of 2τ can be used to
determine the irreversible component of T2 . From [Freeman, R.: A Handbook of Nuclear Magnetic
Resonance. Longman Scientific and Technical, Oxford (1997), p. 270] Wiley copyright
processes like those leading to ωdd induce an irreversible loss that characterises a
true transverse relaxation time T2 . This can be determined by plotting the magnitude
of the echo as a function of 2τ (Fig. 11.11).
As a general rule, any evolution under the action of a given Hamiltonian pre-
serves coherence, and it may be possible to retrieve this coherence by methods that
reverse the equations of motion. However, non-deterministic terms increase the en-
tropy of the spin system in an irreversible way.
11.3.4 Applications
a. Magnetometry
NMR is clearly an important method for carrying out very accurate field measure-
ments. One only has to select a liquid for which the NMR line is very narrow,
because the fast Brownian motions of the atoms tend to average out the dipole inter-
actions between nuclear spins. Measuring the field reduces to measuring the Larmor
frequency. Since time and frequency standards are now very precise, the gyromag-
netic ratios are determined with great accuracy, allowing absolute measurements of
the magnetic field.
b. Distribution of B in the Mixed State of a Superconductor
In superconductors, we have seen that the magnetic induction varies significantly

in space in the mixed state. This leads to a distribution of Larmor frequencies for
the nuclear spins in the material. The shape of the NMR spectrum reconstitutes the
histogram of the magnetic fields. Close to Hc2 , singularities appear in the spectrum
for values of the magnetic field corresponding to the extrema of the field distribution.
In high-Tc superconductors, the line width of the observed resonance can be used to
deduce the London penetration depth.
However, for experimental reasons, NMR is not the best method for studying the
superconducting state. A related technique uses elementary particles called muons.
These behave like heavy electrons (or light protons), and have the property of de-
caying by emission of positrons in the direction of their spin. A muon whose spin is
initially polarised perpendicularly to the field B0 is implanted in the sample at time
zero. We then observe the direction of the emitted positrons when it decays. By re-
peating this experiment for a large number of events, the free precession signal of
the muon spin can be reconstructed statistically. This experiment is equivalent to an
NMR experiment, and can be used to determine λ(T). Since muons can be implanted
in almost any sample, it has been possible to make comparative measurements of λ
in a wide range of superconducting materials.
Atomic Scale Magnetic Measurements
We have seen above that changes in the magnetic induction in a material can be detected di-
rectly by a change in the nuclear Larmor frequency. In weakly magnetic materials, for which
the magnetisation is negligible, B = μ0 Ha and the nuclear Larmor frequency should be deter-
mined by the applied external field alone. It would be difficult to obtain information about the
physical properties of materials by this technique.
But we are forgetting here that the nucleus is a kind of atomic scale microscopic probe,
coupled to the electrons. Interactions like the dipole interactions between nuclear and electronic
spins are such that the nuclear spin feels a magnetic field associated with the polarisation of
the electronic magnetic moments. This means that the magnetic field felt by the nuclear spins
is modified with respect to the applied field. It is the spectroscopy of these fields that provides
atomic scale information about the immediate vicinity of the nuclei in the material. Let us
examine the different interactions between the nuclear spins and the magnetic moments of
electronic origins, known collectively as hyperfine interactions.
a. Hyperfine Interactions
The dipole interaction between the moments associated with a nuclear spin I and an electron
spin S separated by a displacement r is

μ0 γe γn h̄2 (I · r)(S · r)
Ĥdd = − I · S − 3 ,
4π r3 r2
where γn and γe are the nuclear and electronic gyromagnetic moments, respectively, and S and
I are dimensionless quantities. This dipole interaction diverges when r tends to zero, and is
only therefore valid for electrons with zero probability of being at the site of the nucleus. This
is the case for electrons in the p, d, or f shells.
On the other hand, the s electrons have nonzero probability of being at the site of the
nucleus. The Dirac Hamiltonian can be used to show that the corresponding interaction, called
the contact interaction Ĥc is scalar, and is given in this case by
μ0 8π
Ĥc = γe γn h̄2 I · Sδ(r) .
4π 3
Finally, the interaction with the magnetic field associated with the orbital angular momentum
of the electron is
μ0
Ĥorb = − γe γn h̄2 I · /r3 .
4π
These Hamiltonians can all be written in the form
Ĥeff = −γn h̄I · Beff ,
and we may consider that each electron induces a magnetic field Beff at the nuclear site. As
the temporal fluctuations of the electronic moments are very fast compared with the nuclear
Larmor frequency, the static component of Beff is its time average. The position of the NMR
for a given nucleus is thus determined by the time average Beff of the resultant of the fields
due to the different electrons in the material.
It is easy to see that the hyperfine interaction will vanish for filled electronic shells, because
they have zero total spin and total orbital angular momentum. When there is no applied field,
Beff can only be nonzero for materials in which there is a static spin or orbital magnetic
moment. This will be the case for ferromagnetic materials.
b. Chemical Shift
In substances where the electrons are paired in atomic or molecular levels, the static part of the
hyperfine coupling is only nonzero in the presence of an applied field B0 , and is proportional
to B0 , like the magnetisation. The resonance is shifted with respect to that of the free atom in a
gas. The relative shift Beff /B0 may be due to the orbital part of the hyperfine coupling. This is
the case, for example, for the displacement due to the orbital currents induced by the external
magnetic field in electronic or molecular shells close to the nucleus. Since this shift depends
on the electronic charge distributions, it is highly sensitive to the chemical environment of the
given atom, hence the name chemical shift.
These effects are generally small, but can be used to distinguish the nuclear spin resonances
of the different atoms depending on their environment. This has become a very powerful tool,
used universally in chemistry and biology. Routine chemical analyses are carried out by NMR.
It can also determine the 3D structures of biological molecules, with the help of multipulse
methods, which have reached an exceedingly high level of refinement.
11.4 Summary 341
c. NMR in Metals and Magnetic Materials
When the electron states are not paired, a component of Beff due to the electronic atomic
moment may arise via the contact hyperfine term. The Pauli spin susceptibility of a metal
induces a frequency shift given in %, whereas the chemical shifts are expressed in parts per
million (ppm). This comes about because the contact coupling is much stronger than the dipole
or orbital couplings, and the Pauli susceptibility is high compared with the susceptibilities of
orbital origins in insulating compounds.
In magnetic materials, the electronic moments are static at low temperatures, as compared
with their behaviour at the Curie or Néel temperatures. It follows that the static effective fields
are nonzero even when there is no applied field. For atomic nuclei carrying an electronic mo-
ment, this field will be very large (several tesla in general), and will give rise to a resonance at
the Larmor frequency ωL = γn Beff , even for zero applied field. The fields induced on the nuclei
distinct from those of magnetic atoms are generally weak.
Note that, in antiferromagnetic compounds, the magnetic susceptibilities are low, and it is
not always possible to determine whether a material is in fact antiferromagnetic by macroscopic
susceptibility measurements. In contrast, NMR measurements generally reveal the existence of
magnetic moments below the Néel transition. NMR often provides a way of characterising the
existence of antiferromagnetic phases.
Other experimental techniques are sensitive to the existence of a static hyperfine field. An
example is provided by the Mössbauer effect, which detects the existence of a hyperfine field
from the removal of degeneracy of the electronic energy levels of some radioactive ions, in-
cluding 57 Fe. Detection is achieved by monitoring the change in energy of a radioactive nuclear
decay transition.
11.4 Summary
The response of a magnetic system to a time-varying excitation is easy to study

in the linear response regime. For excitation by an alternating field B1 cos ωt, the
response M(t) is used to define a complex magnetic susceptibility

M(t) = Re χ (ω) − iχ (ω) B1 eiωt .
In this expression,
∞
χ (ω) = χ (ω) − iχ (ω) = m(t) exp(−iωt)dt
0
is the Fourier transform of the time response m(t) to an impulsive excitation pro-
portional to δ(t = 0), and its real and imaginary components χ and χ are re-
lated by the Kramers–Kronig relations. The energy absorbed by the material is
P = (ω/2)χ B21 , which is why the term χ is called the absorption.
Low frequency losses in ferromagnets can be explained by a response function
m(τ ) = (χ0 /τ ) exp(−t/τ ), defined by a relaxation time τ which depends on the mo-
bility of the Bloch walls.
In contrast, a saturated ferromagnet has a resonant mode at the eigenfrequency
ωL = −γ B0 which corresponds to the Larmor precession of the magnetisation
around the applied field. This precession is damped by the existence of a relaxation
time T1 which brings the magnetisation back into line with B0 . The time dependence
of M(t) is found by solving
dM M − M(B)
= γM∧B− .
dt T1
In ferromagnetic and ferrimagnetic materials, the frequency of the resonant mode
is shifted by the existence of demagnetising fields and magnetic anisotropy terms.
These modes lead to many applications in ultrahigh frequency electronic devices
(several GHz).
In paramagnetic spin systems, Larmor precession of the macroscopic magneti-
sation M0 aligned with the field in the z direction can also be observed. In the case
of the NMR, it is technically possible to study the pulse response of M0 . An in-
tense radiofrequency pulse B1 in the x direction, of duration tw and at the Larmor
frequency, rotates M0 through an angle θ = γ B1 tw about the x axis. By choosing
θ = π/2, the magnetisation is brought into the (x, y) plane. By detecting its pre-
cession after the pulse, one obtains the NMR spectrum by Fourier transform. Pulse
methods also provide a way of determining the longitudinal relaxation time T1 of
the nuclear magnetisation. A spin echo sequence is used to distinguish the decrease
in the transverse magnetisation due to the non-uniformity of the magnetic field from
that due to the transverse relaxation time T2 of the nuclear magnetisation.
The existence of hyperfine couplings between electronic and nuclear spins shifts
the NMR and provides a way of distinguishing the sites within a chemical structure,
or measuring magnetic properties on the atomic scale.
Question 11.1
1. According to (11.24), the ferromagnetic resonance frequency is given by

ω02 = ωL − γ μ0 Ms (Ny − Nz ) ωL − γ μ0 Ms (Nx − Nz ) , (11.36)
when the field is applied along the z axis.
Sphere. In this case, Nx = Ny = Nz = 1/3 and ω02 = ωL2 . The resonance condition
is thus obtained for Ba = Bs such that
ω0 = −γ Bs . (11.37)
Infinite cylinder with axis in the z direction. The cylinder axis is parallel to
Ba , which is in the z direction. In this case, we have Nz = 0, Nx = Ny = 1/2, and
hence
2
1
ω02 = ωL − γ μ0 Ms .
2
The resonance condition is thus obtained for the field Bc such that

1
ω0 = −γ Bc + μ0 Ms . (11.38)
2
Infinite cylinder with axis in the x direction. The cylinder axis is now perpen-
dicular to Ba , which is in the z direction. We thus have Nx = 0, Ny = Nz = 1/2,
and then
1
ω02 = ωL ωL + γ μ0 Ms .
2
The resonance condition is therefore obtained for a field Bc⊥ such that

1
ω0 = −γ Bc⊥ −γ Bc⊥ + γ μ0 Ms .
2
(11.39)
2
From (11.37) and (11.38), it follows that
1
Bc = Bs − μ0 Ms . (11.40)
2
Equations (11.39) and (11.37) yield

μ0 Ms
B2c⊥ − Bc⊥ − B2s = 0 ,
2
or .
μ0 Ms (μ0 Ms )2
Bc⊥ = + Bs 1 + ,
4 16B2s
whereupon
μ 0 Ms
Bc⊥ Bs + . (11.41)
4
It can be seen from Fig. 11.12 that the configurations (cylinder ⊥) and (cylin-
der ) give resonance fields on either side of the resonance for the sphere. The
difference is indeed twice as large for (c) as for (c⊥).
2. The resonance for the spherical ferrite is obtained for
Bs = 0.3338 ± 0.0009 tesla ,
or g = hν0 /μB Bs = 1.990 ± 0.005. Likewise, from (11.40), μ0 Ms /2 = Bs − Bc .

Measurement gives μ0 Ms = 0.12 ± 0.002 tesla.
Y 3 Fe 4.7 Al 0.3 O 12 Bs
μ app
Bc
Bc⊥
ΔB
ΔB
ΔB
μ app
Ba (tesla)
Fig. 11.12 Determinations of the resonance conditions and widths for the three geometrical shapes
of samples considered
3. Note that the the width of the resonance at half-maximum decreases from the
(c⊥) configuration to the (c) configuration to the (s) configuration. (This is also
in increasing order of resonance intensity.) This suggests that the width of the res-
onance can be associated with the fact that the demagnetising fields are not uni-
form. This is clearly the case for the cylinder which is not infinite. The resonance
is narrower in the (c) configuration than in the (c⊥) configuration, because the
first configuration is more symmetrical. The resonance also has noticeable width
for the spherical configuration. It might be thought that the line is Lorentzian
and that the width corresponds to a relaxation time T1 . In actual fact, T1 is very
long and even in this case it is the demagnetising field distribution associated
with imperfections in the sample that determines the line width (deviation from
sphericity, porosity of the ferrite).
Chapter 12
The Thermodynamics of Ferromagnets
Contents
12.1 Excited States and Low Temperature Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347

12.1.1 Magnons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
12.1.2 Low Temperature Thermodynamic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
12.1.3 Experimental Detection of Magnons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
12.2 The Magnetic Phase Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
12.2.1 Mean Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
12.2.2 Landau and Ginzburg–Landau Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
12.2.3 Critical Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
12.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
346 12 The Thermodynamics of Ferromagnets
The temperature dependence of the thermodynamic properties of solids is determined by the

excited energy levels above the ground state. In order to detect these excited states directly, a
choice method is to observe the scattering of neutrons produced by a nuclear reactor. A low temper-
ature finger is used to cool the neutrons emitted by the reactor, thereby reducing their energy. The
directions of their velocity vectors are selected by means of a monochromator, in which a specific
single crystal selects a Bragg diffraction of the neutrons. When these neutrons are scattered by the
sample, their wave vectors change and there is a loss or gain in energy that can be determined by
diffraction on a crystal analyzer. The photograph shows a three-axis spectrometer (monochromator
output axis, orientation of the single crystal sample, output analyser axis). In magnetic materials,
the magnon excitation spectra can be determined by spectroscopic analysis of this inelastic neutron
scattering mediated by the interaction between the spin of the neutron and the electronic atomic
moments. Likewise, neutron scattering by interaction with the atomic nuclear core is used to deter-
mine the phonon excitation spectra in solids (see Problem 9: Phonons in Solids). Photograph of the
TRISP spectrometer at the FRM-II reactor in Garching (Germany). Image courtesy of Keller, T.,
Keimer, B. who run this instrument on behalf of the Max Planck Institute for Solid State Research
12.1 Excited States and Low Temperature Properties 347
The discussion in Chap. 8 showed that the Heisenberg model provides a good
description of the ordering of magnetic moments in a magnetic material, especially
in the case of insulating compounds. However, this discussion considered only the
properties of the ground state, so it cannot inform as to the thermodynamic proper-
ties of magnetic materials. This will be the task of the present chapter. We begin by
discussing the low energy excited states in the Heisenberg model and deduce some
low temperature properties. We shall then examine the vicinity of the critical tem-
perature where an alternative approach will be needed, based on the molecular field
theory.
12.1 Excited States and Low Temperature Properties
12.1.1 Magnons
The excited states of the Heisenberg Hamiltonian (8.31) are obtained from the ferro-
magnetic ground state (8.32) by reversing one or more spins. To begin with, consider
the case in which a single spin is reversed. The state for which the spin at site R is
reversed is

+
|R = SR |0 = | ↑R | ↓R . (12.1)
R =R
But this state is not an eigenstate of the Hamiltonian. Indeed, for B = 0, the identity
(8.30) implies that

1 N 1 1
H|R = − − 1 J z|R − J |R − |R . (12.2)
4 2 2
2
R
The first term on the right-hand side comes from those terms in H not involving the
site R, while the second term comes from terms coupling this site with its z neigh-
bours. The sum over R is thus restricted to nearest neighbours of R, as indicated by
the prime over the summation symbol.
Introducing the energy E0 = −NzJ /8 of the ground state, (12.2) can be rewrit-
ten in the form

1
H|R = E0 |R − J |R − |R . (12.3)
2 R
Since Ĥ couples the states |R and |R , we can obtain the eigenstates of Ĥ as
linear combinations of the states |R. It is easy to see that, given the periodicity
of the lattice, these can be constructed as linear combinations of the states |R in the
following way:
1 ik·R
|k = √ e |R . (12.4)
N R
If as usual we now impose periodic boundary conditions on a cube of volume L3 ,

the components of k are quantised in the reciprocal lattice. It is straightforward to
obtain

J 1 ik·R
H|k = E0 |k + √ e (|R − |R ) . (12.5)
2 N R R
Setting R = R + R , the R are the lattice vectors from the origin to nearest neigh-
bours. The change of variables (R, R ) → (R , R ) makes the sums over R and R
independent and the order of summation can then be changed, which gives

J 1 ik·(R −R )
H|k = E0 |k + √ e |R − R − |R
2 N
R R

J
= E0 |k + |k − eik·R |k .
2
R
This in turn implies that

H|k = E0 + ε(k) |k . (12.6)
We thus find that |k is an energy eigenstate with excitation energy (taken above the
ground state)

1 − eik·R 1
ε(k) = J =J sin2 k · R . (12.7)
2 2
R R
The second equality here results from the symmetry of the lattice with respect to the
origin, which clearly implies a real value for ε(k).
These states |k, usually called spin waves or magnons, represent precession of
the spins around the z axis with an angular frequency ε(k)/h̄ and spatial period
2π/|k| (Fig. 12.1). In particular, for |k| → 0, the precession frequency tends to zero.
This is explained by the fact that, for k = 0, a global rotation of all the spins does not
alter their energies, owing to the invariance of the Hamiltonian under spin rotations.
a y
y
x
Fig. 12.1 Precession of spins about the z axis in a spin wave excitation. Note that each spin is out
of phase with respect to its neighbours
No restoring force is therefore generated. For nonzero values of k the restoring force
comes from the phase difference between neighbouring spins in their precessional
motion.
12.1.2 Low Temperature Thermodynamic Properties
It is important to note that the states |k are eigenstates of the operator Sz with a
lowered absolute eigenvalue with respect to that of the ground state:

z N
z
S |k = SR |k = − − 1 |k . (12.8)
2
R
As a consequence, the magnetic moment is also lowered. This has an important

consequence when we consider the finite temperature equilibrium properties, as the
thermally excited spin waves will lead to a reduction in the magnetisation.
To estimate this effect quantitatively, we must establish whether spin waves sat-
isfy boson or fermion statistics. Note that we obtain the same quantum state if we
first reverse the spin at site R1 and then reverse the spin at another site R2 , or vice
versa. More formally, we say that the spin operators
associated
with two different
sites commute, expressed mathematically by SRi , SRj = 0 if Ri = Rj . The wave
function does not therefore change sign when two spin reversals are exchanged, and
these reversals must therefore be treated as bosons.
Note that, for S = 1/2, the same spin cannot be reversed twice [(S+ )2 = 0], so
these are bosons with a short-range interaction that prevents one site from being
occupied by two bosons (sometimes called hard-core bosons). At low temperatures,
when not many spin waves are excited and the number of bosons is thus small,
this interaction will have no significant consequences. In a dilute gas, interactions
between particles play little role.
The number of spin waves with a given vector k is thus determined by Bose–
Einstein statistics:
1
n(k) = βε(k) , (12.9)
e −1
in which the chemical potential does not appear because the number of spin waves
excited is not fixed. We may now calculate the effect of the excited spin waves on
the magnetisation:
M(T) = −2μB Sz

1
= −2μB − N + n(k)
2
k

2Ω d3 k 1
= M(T = 0) 1 − . (12.10)
N (2π )3 eβε(k) − 1
In the last line, the sum over k has been replaced as usual by an integral. The general
case of this integral is not easy to evaluate, but at low enough temperatures, it can be
simplified. In that case, only states with low excitation energies, i.e., with low values
of |k|, contribute significantly, so (12.7) can be replaced by its quadratic expansion
in k about k = 0, i.e., ε(k) = αk2 . Setting q = β 1/2 k, we thus obtain

2Ω d3 q 1
M(T) = M(T = 0) 1 − (kB T)3/2 . (12.11)
N 3 αq
(2π ) e − 1
2
The integral, which cannot be evaluated analytically, is independent of temperature.

We obtain

M(T) = M(T = 0) 1 − const. × T 3/2 . (12.12)
This is Bloch’s law. It has been confirmed experimentally in many cases (see, for
example, Fig. 12.2). A detailed investigation of the integral in (12.11) determines
the value of the constant in Bloch’s law. It also shows that the terms omitted in the
truncated expansion of (12.7) give contributions going as higher powers of T than
T 3/2 , which can thus be neglected at low temperatures.
Note that in 1D and 2D the integral in (12.11) diverges. This suggests, at least
formally, that the first thermal correction to the magnetisation will be larger than the
order zero term (the correction at zero temperature). The natural interpretation of
this, as confirmed by more rigorous arguments, is that a ferromagnetic state at finite
temperature is not possible in 2D or 1D. This point is important when we come to
investigate magnetic effects in certain substances and structures that are currently
receiving much attention.
1
− NS ∑ k n(k) EuS
Fig. 12.2 Spontaneous mag-
netisation of EuO and EuS at EuO EuO
different temperatures. The
temperature is normalised by EuS
M M0
the Curie temperature and the

spontaneous magnetisation
by its value at zero tempera- 0 .36
TC − T
ture. Bloch’s law is satisfied 1. 17 TC
at low temperatures. From
Als-Nielsen, J., et al.: Phys.
Rev. B. 14, 4908 (1976). With
the kind permission of the
American Physical Society.
http://link.aps.org/doi/10.1103/
PhysRevB.14.4908 T TC
Note finally that spin waves also occur in antiferromagnets and ferrimagnets.
However, an important difference in antiferromagnets is that the energy is linear
rather than quadratic in k for small values of |k|.
12.1.3 Experimental Detection of Magnons
Spin wave modes can be studied directly by neutron scattering. Indeed, a neutron
with incident wave vector k can be scattered by a magnetic material by emitting
or absorbing a magnon. This scattering process is inelastic, because it changes the
energy and wave vector of the neutron. Indeed, the scattered neutron has wave vec-
tor k + q, where q is the wave vector of the absorbed magnon. Spectrometers with
three axes of rotation, like the one shown on p. 346, are used to carry out such
experiments. The first axis corresponds to that of the monochromator which specifies
the orientation of the incident neutron beam, while the second fixes the orientation
of the sample with respect to this beam, and the third selects the neutrons scattered
in a given direction. Writing down the conservation of energy and wave vector, the
magnon dispersion relation ε(k) can then be determined. Figure 12.3 shows the dis-
persion relation observed for a ferromagnetic cobalt compound (Co0.92 Fe0.08 ).
Question 12.1. Can you explain the shape of the experimental curve in Fig. 12.3? What
experimental quantities characterising ferromagnetism can be deduced from this ex-
periment? The magnetisation of this compound with face-centered cubic structure is
1.84μB /atom (neglecting the difference between Co and Fe). It has a Curie temperature
of 1,300 K. Can this information be used to check the applicability of mean field theory?
[111]
0.05 [110]
[100]
0.04
Magnon energy (eV)
Fig. 12.3 Magnon disper-

sion curves obtained by
0.03
inelastic neutron scattering
in Co0.92 Fe0.08 . The wave
vector is renormalised in the
0.02
three directions indicated.
From Sinclair, R.N., Brock-
house, B.N.: Phys. Rev. 120, 0.01
1639 (1960). With the kind
Physical Society. http://link.
aps.org/doi/10.1103/PhysRev. 0.1 0.2
120.1638 ka/2π
12.2 The Magnetic Phase Transition
The Curie temperature separates two different magnetic states or phases, one with
and one without spontaneous magnetisation (T ≤ TC and T ≥ TC , respectively).
This temperature thus characterises a phase transition. The theory of spin waves de-
scribed in the last section is adequate at low temperatures because the total number
of spin wave states remains small. However, in order to understand what is hap-
pening close to the phase transition, where the spontaneous magnetisation is small,
we would need a theory of dense spin wave gases. On the other hand, this is a dif-
ficult undertaking and not particularly instructive. In Chap. 8, we pursued another
approach called the mean field theory, based on the molecular field of Curie and
Weiss.
12.2.1 Mean Field Theory
In the framework of the mean field theory, we considered the part of the Heisenberg
Hamiltonian involving a spin at a given site R:

HR = SR · − J SR + 2μB B . (12.13)
R
The mean field approximation replaces the spin operators at neighbouring sites R
of R by their average value. The underlying hypothesis is that fluctuations about this
average value are small. Setting
M
SR −→ SR = − , (12.14)
2μB N
the Hamiltonian at site R then takes the simple form
HR = 2μB Beff · SR , (12.15)
where the spin at R is subjected to an effective field
zJ
Beff = B + λ∗ M , λ∗ = , (12.16)
4μ2B N
with z the number of nearest neighbours of the site R. For a spin S = 1/2, this led
simply to

μB Beff
M = NμB tanh . (12.17)
kB T
12.2 The Magnetic Phase Transition 353
A ferromagnetic solution was obtained in zero field below a critical temperature
zJ
TC = , (12.18)
4kB
which can be identified with the experimentally observed Curie temperature. For
an arbitrary value of the spin, the result (12.18) must be multiplied by a factor of
4S(S + 1)/3.
Note that this theory does indeed explain the self-stabilising nature of the ferro-
magnetic state. On the one hand, the nonzero magnetisation of site R is due, via the
effective field, to the nonzero magnetisation of the neighbouring sites R , while on
the other hand, site R contributes to the creation of a magnetisation at the sites R
via the very same effective field. This effect will propagate from site to site. Start-
ing from interactions only between nearest neighbours, we thus obtain a coherent
magnetic state over the whole crystal.
Note also that any nonzero value of the spontaneous magnetisation breaks the
spin rotational symmetry of the Heisenberg model. The Curie temperature thus cor-
responds to the symmetry breaking already discussed.
The mean field theory makes quite detailed predictions concerning certain physi-
cal quantities. Consider first the spontaneous magnetisation. From (12.17), the mag-
netic moment per atom for temperatures just below TC is given by

M √ T 1/2
m= = μB 3 1 − . (12.19)
N TC
Note in particular the square root growth, i.e., with infinite gradient just below TC .
For temperatures much lower than TC , the solution of (12.17) obviously tends
to unity, agreeing with the magnetic moment of the ground state in the Heisen-
berg model. However, it is easy to show that the solution of (12.17) tends to unity
exponentially when T → 0, whereas spin wave theory predicts a very different law,
going as T 3/2 [see (12.12)]. This disagreement is due to the fact that spin waves are
not correctly taken into account in the mean field theory.
Above TC , (12.17) can be used to find the behaviour in a weak applied field B. A
small magnetic moment is then expected, so the equation can be linearised to obtain
μ2B μ2 TC
m= Beff = B B + m . (12.20)
kB T kB T T
This yields a linear relation between m and B, whence the temperature dependence
of the magnetic susceptibility is
μ2B
χ (T) = N . (12.21)
kB (T − TC )
The magnetic susceptibility thus diverges as the Curie temperature is approached.

This is the Curie–Weiss law. This increase in the susceptibility compared with
the situation for spins without exchange interactions can be explained by the fact
that a given spin ‘sees’ not only the applied field, but also, through the exchange
interaction, the magnetisation of the neighbouring spins. It is therefore exposed to
a higher field in a given applied field, and so has a higher magnetisation than a
free spin.
We may wonder how to interpret the infinite susceptibility at the Curie tempera-
ture. It could not of course be an infinite magnetic moment for a weak applied field.
Indeed, at the temperature T = TC = zJ /4kB , expanding the hyperbolic tangent
function to third order in (12.17), we find that
zJ 3
B− m =0, (12.22)
12μ4B
that is 1/3
12μB B
m = μB . (12.23)
zJ
Consequently, the magnetisation begins by growing faster than linearly in B (which
does indeed correspond to an infinite susceptibility), but clearly remains less than or
equal to NμB .
Finally, by (12.15), we see that the internal energy U(T), i.e., the expectation
value of the Hamiltonian, is proportional to the square of the spontaneous magneti-
sation. As a function of temperature, it thus has a different gradient above and below
the Curie temperature. As a consequence, the specific heat C(T) = Ω −1 ∂U/∂T has
a discontinuity at TC .
12.2.2 Landau and Ginzburg–Landau Theories
We thus find that the mean field approximation gives a reasonable explanation for
the behaviour of the thermodynamic properties at the paramagnetic–ferromagnetic
transition. We shall show that this approach can be generalised in a simple way to
many other phase transitions, following Landau.
a. Landau Theory of Phase transitions
As in any problem of themodynamic equilibrium, we consider the partition function

−βE
Z = Tr e−βH = e n
, (12.24)
n
where β = 1/kB T and the sum over n includes all the eigenstates of the Hamiltonian
(the Heisenberg Hamiltonian in this case). We now observe that the Heisenberg
Hamiltonian commutes with the z component of the spin, and hence also with the
z component of the magnetisation. Consequently, the sum in (12.24) can be rear-

ranged to give

−βEn
Z= e = e−βL(T,Mz ) . (12.25)
Mz n Mz
The prime on the sum over n implies that the sum only includes states for the given
value of Mz . The Landau function L is then defined as

L(T, Mz ) = −kB T ln e−βEn . (12.26)
n
Naturally, L will be as hard to calculate as the full partition function. However, there
are two points to note here:
1. The Landau function can be considered as the free energy associated with a given
magnetisation. Like any free energy, L must therefore be extensive, i.e., propor-
tional to the volume Ω of the system.
2. L must have the same symmetry properties as the Hamiltonian. In particular,
when there is no magnetic field, we must have L(T, −Mz ) = L(T, Mz ).
In addition to these two simple observations, Landau assumed that L could be
expanded in a Taylor series for small values of Mz

L(T, Mz ) = Ω (T) + a(T)Mz2 + b(T)Mz4 − HMz + · · · . (12.27)
We assume here that Mz remains small, as happens near the Curie temperature, and
hence that terms of higher order than 4 can be neglected. The form of the coefficients
, a, and b can only be ascertained by detailed microscopic calculation. However,
the coupling with the external magnetic field H, assumed to be oriented in the z
direction, is determined by the fact that Mz commutes with the Hamiltonian. It fol-
lows that the energy of a magnetisation Mz depends on the field according to
En (Mz , H) − En (Mz , H = 0) = −ΩHMz . (12.28)
We now observe that, in the thermodynamic limit Ω → ∞, which applies for any
macroscopic system, the sum over Mz in (12.25) is dominated by the largest term,
i.e., the term minimising L. Up to corrections that tend to zero for Ω → ∞, the true
free energy of the system is then given by
F(T, H) = L(T, H, Mmin ) , (12.29)
where Mmin is the value of Mz minimising L. Furthermore, the equilibrium value of

the magnetisation is simply Mmin .
In order to obtain a stable minimum of L, the constant b in (12.27) must be
positive. For zero applied field, there are then two situations (see Fig. 12.4): either a
is also positive, giving rise to a unique minimum at Mz = 0 (paramagnetic state), or a
Fig. 12.4 Generic shape of L(Mz)

L(T, Mz ) for a > 0 and a < 0
a >0
a< 0
Mz
is negative, in which case there are two degenerate minima at Mz = 0, corresponding

to a ferromagnetic state.
Without knowing more about the temperature dependence of the coefficients,
the most natural assumption for describing a paramagnetic to ferromagnetic phase
transition is a linear variation:
a(T) = a (T − Tc ) , (12.30)
with a > 0. It is then easy to check that the qualitative description at the end of the
last section is confirmed by Landau’s theory. We may add that this theory contains
two phenomenological parameters, a and b, but makes predictions about (at least)
four physical quantities, leading to non-trivial experimental tests.
The Landau theory exploits very little microscopic information about the given
system. It is based primarily on considerations of symmetry and internal consis-
tency. For this reason, it applies to a very broad range of phase transitions: mag-
netic phase transitions, structural transitions in solids, order–disorder transitions in
alloys, phase transitions in liquid crystals, and so on. Naturally, in each case one
must identify a physical variable characterising the ordered state, usually called the
order parameter. In ferromagnetic systems, we took the magnetisation, while in
antiferromagnetic systems it would be the alternating magnetisation, in structural
transitions the atomic displacements, and so on. In every case, an expansion of type
(12.27) can be made, and similar conclusions are drawn: the order parameter varies
as (Tc − T)1/2 , the specific heat has a discontinuity, and there are divergent suscep-
tibilities. In short, Landau theory predicts a universal behaviour for thermodynamic
quantities near a phase transition.
b. Ginzburg–Landau Functional
In Landau’s theory, the magnetisation (or more generally, the order parameter) is
treated as a global variable characterising the system. In particular, in a uniform
body and far from any surfaces, this implies a constant magnetisation throughout
space. However, one often has to study non-uniform systems, as exemplified by
the magnetic domains and walls discussed in Chaps. 6 and 9. Ginzburg and Lan-
dau generalised the theory to cover this situation. They observed that in general the
homogeneous configurations are the most stable, while inhomogeneous configura-
tions, when possible, have a certain energy cost. The simplest case, and realistic in
most situations, is simply an energy associated with the magnetisation gradient, of
the kind already discussed in Chap. 9. Since the magnetisation is now space depen-
dent, the Landau function must be written as an integral over the whole volume of
the sample:

2
L T, M(r) = Ω(T) + d3 r a(T)M2 + b(T) M2 + c(T)(∇M)2 − H · M .
(12.31)
For reasons of stability, the constant c must be positive. Note also that we have rein-
stated the vectorial nature of the magnetisation, neglected in the above description
of the Landau theory to simplify the discussion.
With the simplest boundary conditions, the Ginzburg–Landau functional L intro-
duced above is once again minimised by a constant magnetisation state, as in the
Landau theory. However, in non-uniform situations, the full description by (12.31)
must be used. This is of particular importance when studying superconductivity, for
which Ginzburg and Landau originally developed their theory. Indeed, in this case
the general principle of gauge invariance imposes a well-determined form for the
coupling between the order parameter and the electromagnetic field. A consequence
of this coupling is the quantisation of the magnetic flux through a superconduct-
ing loop. This effect in turn allows unambiguous determination of the charge of
the superconducting carriers. However, it should be noted that these experiments
were only carried out after the BCS theory had been developed, at which point the
existence of pairs was already well accepted.
Another consequence of the Ginzburg–Landau model (12.31) is that there are
now deviations from the minimal L state. These deviations cost a finite energy, no
longer proportional to the volume of the system. As a result, in the expansion in a
sum of terms with fixed Mz , similar to the one in (12.25), we can no longer restrict
to a single dominant term. We must actually sum over all possible configurations,
which is clearly a much more difficult task. This is of crucial importance in a deeper
understanding of phase transitions, as we shall discuss briefly in the next section.
12.2.3 Critical Behaviour
We have just seen that physical quantities vary in a singular way near phase tran-
sitions. All such singularities are commonly referred to as critical phenomena, and
such behaviour is characterised by a set of critical exponents.
It is interesting to compare the theory with experimental results. The temper-
ature dependence of the spontaneous magnetisation of EuO and EuS is shown
in Fig. 12.2. The two compounds have the same crystal structure, with magnetic
moments carried by the Eu2+ ions, with spin S = 7/2. Note the almost vertical
slope of the magnetisation close to TC , as predicted by the mean field theory. How-
ever, more careful analysis of the experimental results shows that the exponent of
the singularity is not 1/2, but 0.36. Note on the other hand that the behaviour for
T TC agrees extremely well with the spin wave theory.
The magnetic susceptibility of nickel is shown in Fig. 12.5. There is over-
all agreement with the Curie–Weiss law, but we find similar deviations to those
observed for the magnetisation: the critical behaviour χ ∝ (T − TC )−γ is not given
by γ = 1, but rather by γ = 4/3. Similar discrepancies with respect to the mean field
theory are also observed in other physical quantities, and this in a quite general way,
independently of the substance under investigation. The experimental observations
are summed up in Table 12.1, where the critical exponents are defined.
The differences between the critical behaviour observed experimentally and the
predictions of the mean field and Landau theories have been the subject of a great
deal of theoretical physics over several decades. The main problem was the lack
Fig. 12.5 Inverse magnetic 3

susceptibility of nickel as
a function of temperature.
χ −1 is not found to vary lin-
early with temperature, thus 2
(ρ /χ) × 10 −4
defining a critical exponent

γ greater than unity. From
Weiss, P., Forrer, R.: Ann.
Phys. 5, 153 (1926), EDP 1
Sciences copyright
350 400 450 500

T (◦ C)
Table 12.1 Critical phenomena. The first column defines the critical exponents γ , β, α, and δ
specifying the behaviour of physical quantities close to TC . The values predicted by the mean field
theory and the experimental data are given in the second and third columns, respectively. The other
columns give the exact results for specific models, viz., the 2D and 3D Ising models, and the 3D
Heisenberg model
Mean field Experiment 2D Ising 3D Ising 3D Heisenberg
χ(T) ∝ (T − TC )−γ 1 1.3–1.4 7/4 1.24 1.39
M(T) ∝ (TC − T)β 1/2 ≈1/3 1/8 0.324 0.362
C(T) ∝ |T − TC |−α 0 −0.1–0.1 log 0.110 −0.115
M(B, T = TC ) ∝ |B|1/δ 3 ≈5 15 4.82 4.82
12.3 Summary 359
of exact or approximate theoretical methods for understanding the physics and

making reliable predictions. A variety of methods were thus implemented. One was
to simplify as far as possible the original model (the Heisenberg model in our case)
to obtain an exactly soluble model.
The most spectacular success here was the solution of the Ising model by Onsager
in 1944. This is obtained by replacing the spin operators in the Heisenberg model by
their z components, simply dropping the x and y components. The eigenstates then
become very simple, although the partition function remains extremely difficult to
calculate. Even today, it can only be done in the one- and two-dimensional cases. In
one dimension, there is no phase transition, while in two dimensions, Onsager’s
solution leads to the exponents in Table 12.1. Note the considerable difference
between the exact results and those of the mean field theory. Apart from a few spe-
cific applications, e.g., to surface phase transitions, the importance of the Onsager
solution is mainly conceptual: it clearly demonstrated that the critical exponents
could be very different from those predicted by the mean field theory.
Exactly soluble models are scarce and mainly concern 1D and 2D systems. In many cases, a
realistic alternative is a (low or high temperature) series expansion of the partition function
and other physical quantities. Very precise results can now be obtained in this way. We can
accurately obtain the critical temperatures of various models (the mean field theory typically
overestimates them by 20–40%), together with their critical exponents. From a quantitative
point of view, this method is often satisfactory, but the underlying physical mechanisms of the
critical behaviour remain obscure.
The general theory of critical phenomena was finally developed by Wilson at the beginning
of the 1970s (Nobel prize in 1982). He introduced the idea of the renormalisation group and
applied it to the Ginzburg–Landau theory of (12.31). This approach provides a general physical
picture of critical phenomena and at the same time a quantitative calculational tool. In particu-
lar, it explained the approximate nature of the Landau theory. In fact, the Landau function in the
form (12.27) is not correct: assuming uniform magnetisation, L contains non-integer powers
of Mz . The idea of the renormalisation group subsequently proved fruitful in many other areas
and has become one of the most general tools of theoretical physics today.
12.3 Summary
In the Heisenberg model, the first excited states of a ferromagnetic material are not
obtained by reversing a spin, but rather by producing a rotation of each spin with
respect to its neighbours, leading to a total variation Sz = 1. The end of each vector
S thus describes a wave called a spin wave for these excitations. They have an energy
spectrum given by

2 1
ε(k) = J sin k· R ,
2
R
where the sum is taken over nearest neighbours of the origin. Their occupation num-
ber is given by Bose–Einstein statistics. At nonzero temperature, spin waves are
excited and correspond to a reduction in the equilibrium magnetisation given by

M(T) = M(T = 0) 1 − const. × T 3/2 .
This is confirmed by experiment.

The paramagnetic–ferromagnetic phase transition can be explained by a mean
field theory. Here the effect of the spins neighbouring a given site is replaced by a
uniform mean field Beff = B + λM. The self-consistent solution for the magnetisa-
tion provides a determination of the Curie temperature and the temperature depen-
dence of M in the ferromagnetic and paramagnetic states. It also shows that there is
a discontinuity in the specific heat.
The mean field approximation is an example of a more general methodology
initiated by Landau which starts with a simplified expansion of the free energy as a
function of the magnetisation:

L(T, Mz ) = Ω (T) + a(T)Mz2 + b(T)Mz4 − HMz + · · · .
The magnetisation M is then determined as the variable minimising the Landau

functional L. This method is quite general and applies to many phase transitions, for
which it predicts universal behaviour of thermodynamic quantities at the transition.
Deviations from the predictions of the Landau theory are observed experimentally.
The general theory of critical phenomena explains these differences and gives a
better account of fluctuations in the vicinity of the phase transition.
Question 12.1
At the center of the Brillouin zone, for small values of k, the dispersion relation
of the spin waves takes the form ε(k) = AJ k2 when k is taken in a given direc-
tion of the reciprocal space. The experimentally observed relation agrees with this
expectation, but with a constant term:
ε(k) = C + AJ k2 .
There is no difficulty in determining J from exact knowledge of the crystal struc-

ture and the normalisation factor in the experiment.
Note that this calculation corresponds to a Heisenberg model, for which the ori-
entation of the magnetisation is indeterminate in space. The spin wave mode for
k = 0 is the one for which the overall magnetisation rotates in space without phase
shifts between neighbouring moments. This justifies the fact that ε(k) = 0 for k = 0
in the Heisenberg model (M can be rotated in space at no energy cost). However, in
the presence of a magnetocrystalline anisotropy, for example, a rotation of M with

respect to the easy magnetisation axis has an energy cost related to the anisotropy
energy, and this explains the constant term C. There is a gap in the excitation
energies of the spin system. It can then be shown that the energy of the mode at
k = 0 is the energy associated with the ferromagnetic resonance in zero applied
field C = h̄γ BK (see Chap. 11).
Note that the magnetisation does not correspond to 1μB /atom, indicating that the
spin S is not in fact 1/2. The spin wave calculation must be generalised to S = 1/2
and g = 2. This is perfectly feasible, and J could then be determined experimen-
tally. In a mean field model, the Curie temperature is itself related to J and S. It
thus provides another experimental determination of J . We may then be able to
check whether the mean field theory is quantitatively correct by comparing these
two experimental values for J .
Chapter 13
Problem Set
Contents
Problem 1 Debye–Waller Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367

Problem 2 Reflectance of Aluminium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Problem 3 Band Structure of YBa2 Cu3 O7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
3.1 Isolated Copper–Oxygen Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
3.2 Isolated Copper–Oxygen Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
3.3 Chain and Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
3.4 Realistic Models of YBa2 Cu3 O7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Problem 4 Electronic Energy and Stability of Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Problem 5 Optical Response of Monovalent Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Problem 6 One-Dimensional TTF-TCNQ Compounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
6.1 Isolated Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
6.2 Experimental Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
6.3 Dimerised Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
6.4 Peierls Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Problem 7 Insulator–Metal Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7.1 Tight-Binding Method for Hydrogen-Like Orbitals . . . . . . . . . . . . . . . . . 419
7.2 Interactions Between Electrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
7.3 Alkali Elements and Hydrogen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
7.4 Insulator–Metal Transition in Si–P . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Problem 8 Cyclotron Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
8.1 Real and Reciprocal Space Paths of an Electron State . . . . . . . . . . . . . . . 441
8.2 A Semiconductor: Silicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
8.3 Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
Problem 9 Phonons in Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.1 Einstein Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.2 Debye Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
9.3 Experimental Detection of Phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
9.4 Thermodynamic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
9.5 Resistivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
Problem 10 Thermodynamics of a Thin Superconducting Cylinder:
Little–Parks Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
Problem 11 Direct and Alternating Josephson Effects in Zero Magnetic Field . . . . . . . . . . . 481
11.1 Model Josephson Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
11.2 Realistic Josephson Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
11.3 Josephson Junction in a Microwave Field . . . . . . . . . . . . . . . . . . . . . . . . . 483
364 13 Problem Set
Problem 12 Josephson Junction in a Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

12.1 Current Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
12.2 Screening of the Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
12.3 Josephson Plasma Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Problem 13 Magnetisation of a Type II Superconductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Problem 14 Electronic Structure and Superconductivity of V3 Si . . . . . . . . . . . . . . . . . . . . . . . 523
Problem 15 Superconductivity of NbSe2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
Problem 16 Magnesium Diboride: A New Superconductor? . . . . . . . . . . . . . . . . . . . . . . . . . . 541
16.1 Atomic and Electronic Structure of MgB2 . . . . . . . . . . . . . . . . . . . . . . . . 541
16.2 Superconductivity of MgB2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
Problem 17 Electronic Properties of La2 CuO4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
Problem 18 Properties of an Antiferromagnetic Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
18.1 Preliminaries: The Ferromagnetic Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
18.2 Antiferromagnetic Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564
18.3 Susceptibility in the Antiferromagnetic State . . . . . . . . . . . . . . . . . . . . . . 565
Problem 19 Magnetism of Thin Films and Magneto-Optic Applications . . . . . . . . . . . . . . . . 573
Problem 20 Magnetism of a Thin Film . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
20.1 Uniform Magnetisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 580
20.2 Non-uniform Situations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
20.3 Detailed Investigation of Non-uniform Situations . . . . . . . . . . . . . . . . . . 583
366 13 Problem Set
While physics is often considered from a formal point of view, and this is necessary at a certain
level of understanding, the basis of the scientific method is observation and experiment. For this
reason, experimental observations underpin the many problems presented hereafter. In the picture,
Claire and David are now quite clear that their qualitative understanding of the experimental phe-
nomena must be set in the form of models that can account for the observations. Peter, a keen
experimenter, prefers to make a careful record of the facts before moving on to this stage
Problem 1: Debye–Waller Factor 367
Problem 1: Debye–Waller Factor1
Experiments using X-ray diffraction by crystals show that the intensity of Bragg
spots depends on temperature. Figure P1.1 shows the decrease in intensity of some
of these peaks with temperature for aluminium, which has a face-centered cubic
structure. This decrease is due to thermal vibrations of the atoms around their equi-
librium positions.
The aim here is to calculate the effect of these thermal vibrations on the structure
factor S(K), bearing in mind that this is given by

S(K) = fj (K)e−iK·rj ,
j
where rj and fj (K) are the position in the primitive cell and the form factor of each
atom belonging to the basis, respectively.
We thus seek to evaluate the thermal average S(K)(T). Let uj (t) be the displace-
ment of each atom j from its equilibrium position rj :
rj (t) = rj + uj (t) .
It is assumed that each atom undergoes independent fluctuations about its equi-
librium position (Einstein model) and that it can be described by an isotropic 3D
harmonic oscillator of mass Mj and angular frequency ωj .
(200)
(400)
Intensities (arbitrary units)
(600)
Fig. P1.1 Effect of tempera-

ture on the intensity of (h00) (800)
reflections for aluminium.
(h00) reflections with odd h
are forbidden in an fcc struc-
ture. From Nicklow, R.M.
(1000)
Young, R.A.: Phys. Rev. 152,
591 (1966). With the permis-
sion of the American Physical
Society (c 1966 APS). http://
link.aps.org/doi/10.1103/Phys
Rev152.591 T (K)
1 This problem has been designed with G. Montambaux.

368 13 Problem Set
1. Express the structure factor S(K)(T) in terms of u2j = |uj |2 , the mean
squared displacement of each atom in the material basis. Hint: Use the property
of the harmonic oscillator that eiAx = e−A x /2 , where x is the displacement
2 2
and A is a scalar.
2. Calculate the mean squared displacement u2j of a harmonic oscillator as a
function of its mass Mj and angular frequency ωj and the temperature. Recall
that the average potential energy of the harmonic oscillator is equal to half the
average total energy.
In the following, it is assumed that the crystal is monatomic. The mass of the atoms
is M and the angular frequency of their oscillations about their equilibrium position
is ω.
3. When kB T h̄ω, calculate the intensity I(K, T) diffracted by the crystal in the
direction K. Give an expression for I(K, T)/I0 (K), where I0 (K) is the intensity
diffracted by the atomic lattice at rest.
4. At zero temperature, show that the mean squared displacement of an atom u2

does not vanish, and that the factor by which the intensity is reduced is given by

h̄K 2
I(K, T = 0) = I0 (K) exp − .
2Mω
5. Comment on the experimental results. In particular, discuss (a) the temperature

dependence, (b) the dependence on the diffraction vector, and (c) explain the
nonlinearity of the curves at low temperature.
Solution
1. For atoms at positions rj (t) = rj + uj (t), the structure factor is

S(K) = fj exp − iK · rj − iK · uj (t) .
j
At finite temperature T, the displacements uj (t) are finite and the thermal average
of the structure factor is given by

S(K)(T) = fj exp(−iK · rj )Wj (T) ,
j
where
!
Wj (T) = exp − iKuj (t) .
The reduction factor Wj (T) is called the Debye–Waller factor. uj is the displace-
ment of a 3D harmonic oscillator described by the energy
|pj |2 1
E= + Mj ωj2 |uj |2 .
2Mj 2
We have
! !
e−iK·uj = exp − iKx ujx − iKy ujy − iKz ujz

1 2 2
= exp − Kx ujx + Ky ujy + Kz ujz .
2 2 2 2
2
Furthermore, u2jx = u2jy = u2jz = |uj |2 /3, and hence,

K2 2
Wj (T) = exp − |uj | .
6
2. For the harmonic oscillator, the average value of the kinetic energy is equal to
the average value of the potential energy and half the average total energy Ej .
The average energy of an oscillator is given by
h̄ωj h̄ωj
Ej = + β h̄ω .
2 e j −1
An atom oscillates in three directions, and this corresponds to three independent
oscillators. Hence,

1 3 h̄ωj h̄ωj
Mj ωj2 |uj |2 = + β h̄ω . (13.1)
2 2 2 e j −1
370 13 Problem Set
3. At high temperatures such that kB T h̄ωj , this average value tends to
1 3
Mj ωj2 |uj |2 = kB T .
2 2
This classic result can be obtained directly using the energy equipartition theo-
rem. There is a corresponding Debye–Waller coefficient

K 2 kB T
Wj (T) = exp − .
2 Mj ωj2
For a monatomic crystal, all the atoms have the same form factors f (K) and
Debye–Waller coefficient W(T), and

S(K)(T) = f (K)W(T) exp − iK · rj .
j
The intensity of the diffraction spots is the square of the structure factor:

2 kB T
I(K, T) = I0 (K) exp −K , (13.2)
Mω2
where
2

I0 (K) = f (K) e−iK·rj
j
is the intensity diffracted by the atomic lattice at rest.

4. At zero temperature, the zero point motion of the atoms is such that u2 does not
vanish, but is given by (1/2)Mω2 |u|2 = 3h̄ω/4. Therefore,

h̄K 2
I(K, T = 0) = I0 (K) exp − .
2Mω
5. Figure P1.1 shows the logarithm of the intensity diffracted in a direction K as a

function of temperature. (a) At high temperature, we do indeed observe a limiting
linear behaviour described by (13.2) (see Fig. P1.2 left). (b) The gradient of these
straight lines is proportional to K 2 ∝ h2 (see Fig. P1.2 right). (c) At low temper-
atures, we observe a deviation due to the zero point motion. The mean squared
displacement u2 is no longer linear in T. At any temperature, it is given by
(13.1) and the above ratio is therefore given by

I(K, T) h̄K 2 1 1
ln =− + ,
I0 (K) Mω eβ h̄ω − 1 2
(200)
(400)
Intensities (arbitrary units)
Gradient (arbitrary units)

(600)
(800)
(1000)
20 40 60 80 100
T (K) h2
Fig. P1.2 Left: Exponential decay of the intensity at high temperatures. Right: Slopes of the
straight lines on the left are proportional to h2
which explains the shape of the curves obtained. Experiments at low tempera-
tures would be required to make a better assessment of the applicability of the
model.
Problem 2: Reflectance of Aluminium 373
Problem 2: Reflectance of Aluminium
Aluminium (1s2 2s2 2p6 3s2 3p1 ) crystallises into a face-centered cubic Bravais lat-
tice. The side of the conventional unit cell is a = 4.05 Å. The overlaps of the orbitals
of the inner shells 1s2 , 2s2 , and 2p6 between nearest neighbours are very small,
leading to narrow bands of tightly bound electrons in the solid. In contrast, the three
valence electrons are weakly bound and can be treated as nearly free electrons.
1. Find the density n of nearly free electrons in aluminium.
2. Assuming these electrons to be completely free, find the radius of the Fermi
sphere and its position with respect to the first Brillouin zone of aluminium. Find
the value of the Fermi energy EF in eV.
Now consider the Bragg plane bounding the first Brillouin zone in the kx direction.
3. Give an expression for the energy of the free electrons for the electron states
belonging to this Bragg plane as a function of ky and kz . Graph the surface
E(ky , kz ) corresponding to the vectors k of this Bragg plane and locate the Fermi
level.
4. When there is a periodic potential, how will the eigenenergies of these states
be modified in the nearly-free electron approximation? Hint: Assume that the
matrix element of the periodic potential VK associated with the vector K of the
reciprocal lattice corresponding to the Bragg plane is such that |VK | EF .
A photon incident upon an aluminium surface can only be absorbed by a Bloch
electron of the metal with quasi-momentum k if the electron can be excited into
an unoccupied state with the same quasi-momentum k. The absorption of photons
by electrons can be determined by measuring the reflectance of aluminium, i.e., the
reflected intensity from a polished aluminium surface, as a function of the incident
photon energy.
5. Can you explain the experimentally measured reflectance of aluminium shown in
Fig. P2.1? Deduce the value of |VK |.
1.0
0.95
Reflectance
0.90
Fig. P2.1 Reflectance of alu-

minium for different values
of the incident photon energy. 0.85
From Bennet, H.E., Silver,
M., Ashley, A.J.: J. Opt. Soc. 0 2.5 5
Am. 53, 1089 (1963) h̄ω (eV)
374 13 Problem Set
Solution
1. There are three nearly-free electrons per aluminium atom. In the fcc lattice, the
volume of the primitive cell, which contains one aluminium atom, is a3 /4, so the
number of nearly-free electrons per unit volume is
−3
n = 3/(a3 /4) = 12/a3 = 0.18 Å = 1.8 × 10−31 m−3 .
2. The radius of the Fermi sphere is
kF = (3π 2 n)1/3 = (36/π )1/3 (π/a) = 1.1272(2π/a) .

The Brillouin zone is the polyhedron in Fig. P2.2. The point furthest from the
origin Γ is one of the vertices W of the square faces which belong to two Bragg
planes: the face of the cube through the point (0, 2π/a, 0) and the plane bisecting
and perpendicular to the vector Γ A (2π/a, 2π/a, 2π/a). W is therefore equidis-
tant from Γ and the vertex A of the cubic cell, with coordinates (0, 2π/a, π/a).
Hence |Γ W|2 = 1.25(2π/a)2 . It follows that
kF ≈ 1.008|Γ W| .
This shows that the first Brillouin zone is completely contained in the free-
electron Fermi sphere. The Fermi energy is EF = h̄2 kF2 /2m0 = 11.65 eV.
3. The energy of states corresponding to a vector k of the Bragg plane (2π/a, ky , kz )
is given by

E(ky , kz ) = (h̄2 /2m0 ) (2π/a)2 + ky2 + kz2 .
This is the paraboloid of revolution in Fig. P2.3 (left), with minimum
E0 = (h̄2 /2m0 )(2π/a)2 = 9.16 eV .
The intersection with the Fermi sphere is the circle of radius

ki = (11.65 − 9.16)/9.16(2π/a) = 0.52(2π/a) .
kz
A
L
W
Γ
kx
X ky
K
Fig. P2.2 First Brillouin zone 4π /a

of aluminium
Problem 2: Reflectance of Aluminium 375
E E
ki EF h̄ω = 2VK
E0+VK h̄ω = 2VK
E0
E0−VK
W kz kz
ky ky
Fig. P2.3 Energy of states corresponding to a vector k of the Bragg plane (2π/a, ky , kz ) for a free
electron approximation (left). In presence of a periodic potential the degeneracy of these states is
lifted (right)
√
The square of side π 2/a bounding the first Brillouin zone in this plane lies
entirely within the projection of this circle. This is to be expected since all states
in the first Brillouin zone are occupied.
4. In a periodic potential, states corresponding to the Bragg plane must be handled
using degenerate perturbation theory. For all k corresponding to the Bragg plane,
the dominant matrix element VK of the potential is the same, and the degeneracy
is removed in the same way: E(ky , kz )±VK . We obtain two paraboloids by vertical
translations through ±VK , as shown in Fig. P2.3 (right). The Fermi energy is
barely changed in a nearly-free electron approximation, insofar as |VK | EF .
Occupied states are shown in the same figure.
5. Degeneracy is removed in such a way that incident photons of energy h̄ω = 2|VK |
can be absorbed by a transition of an electron from an occupied state k of the
lower paraboloid in Fig. P2.3 (right) to a free state of the upper paraboloid. As
there are many possibilities for the same energy, absorption will be significant.
This is precisely what is observed for the experimental results in Fig. P2.4. High
1.0
0.95
Reflectance
0.90
0.85
Fig. P2.4 Observed reflect- 0 2.5 5 h̄ ω (eV)

ance of aluminium h̄ω =1.5 eV
376 13 Problem Set
absorption leads to a marked drop in reflectance, occurring for h̄ω ≈ 1.5 eV, and
then
|VK | ≈ 0.75 eV .
Since this is much less than EF , the nearly-free electron approximation is vindi-
cated.
Problem 3: Band Structure of YBa2 Cu3 O7 377
In 1986, a family of oxides was found that proved to be superconducting with much
higher critical temperatures than metals and alloys. The compounds in this oxide
family all contain planes of copper and oxygen with Cu atoms at the nodes of a
square lattice (see Fig. P3.1). Some members of this family also contain copper–
oxygen chains.
An example is YBa2 Cu3 O7 . The real structure is shown in Fig. P3.2 (left). The
simplification used in this problem is shown in Fig. P3.2 (right). The primitive cell of
this compound contains two copper–oxygen planes (levels 2 and 3) and one family
of chains, e.g., level 1.
The aim here is to investigate the band structure of YBa2 Cu3 O7 in a simpli-
fied way using the tight-binding approximation (LCAO). 3.1 examines an isolated
copper–oxygen chain. 3.2 then looks at an isolated copper–oxygen plane and two
Fig. P3.1 CuO2 square lattice

a
Cu
O
y
Chains Level 1
CuO
Ba
Chains
YBa2Cu3O7 Level 2
Planes Y
CuO2 Planes
Cu(2)
Level 3
O(3)
c O(2)
O(1) z
b y
Ba
a x Level 4
Cu(1)
O(4)
Fig. P3.2 Real structure of YBCO7 (left) and simplified structure used in this problem (right)
2 This problem has been designed with C. Hermann and T. Jolicœur.

378 13 Problem Set
coupled planes, while 3.3 deals with the coupling between a plane and the nearest
chains. Finally, 3.4 compares the results obtained in this way with a more detailed
calculation and experimental data concerning YBa2 Cu3 O7 .
Throughout this exercise, we only take into account those orbitals with energy
levels close to the Fermi level of the solid, i.e., the 3d orbitals of copper and the
2p orbitals of oxygen. To simplify, we assume that, within the basis, these atomic
orbitals combine to form one atomic orbital per Bravais lattice point. This orbital
will be denoted by φ1 (r) for the chains (3.1) and φ2 (r) for the planes (3.2 and
3.3). These orbitals are assumed to be isotropic, real, and normalised. We apply
the LCAO method to these orbitals. We also assume that the orbitals of two neigh-
bouring sites barely overlap, and take them to be orthogonal for simplicity. There is
no need to write down the relevant Hamiltonians explicitly. All the necessary matrix
elements will be given.
Note: Questions 4 and 7 in 3.2 are not essential for tackling the neighbouring ques-
tions.
3.1: Isolated Copper–Oxygen Chain
Consider an isolated chain of copper and oxygen atoms. Let a be the Cu–Cu distance
and y the unit vector along the chain.
1. Using Fig. P3.3, specify the Bravais lattice and basis of a copper–oxygen chain.
2. What is the associated reciprocal lattice? Specify the corresponding first Bril-
louin zone.
3. Give the form of the Bloch functions in the tight-binding approximation. Let k be
the corresponding wave vector.
4. Considering only nearest neighbours, calculate the dispersion relation EC (k).
Express the result in terms of the matrix elements

EC = d rφ1 (r)ĤC φ1 (r) ,
0 3
V2 chain = V = − d3 rφ1 (r)ĤC φ1 (r + ay) ,
y
x
Fig. P3.3 Copper–oxygen

chains
where y is the unit vector along the chain and ĤC is the Hamiltonian of an elec-
tron in the chain.
5. Plot the dispersion relation, assuming V > 0. If the electron occupation of this
band is one electron per unit cell, what is the shape of the Fermi ‘surface’? What
happens if the electron occupation is very low? Give the effective mass m∗ of the
electrons in that case.
3.2: Isolated Copper–Oxygen Plane
Consider an isolated copper–oxygen plane:

1. From Fig. P3.1, specify the Bravais lattice and basis of a copper–oxygen plane.
What is the associated reciprocal lattice. Specify also the corresponding first
Brillouin zone.
As in 3.1: consider one orbital per basis, denoted φ2 (r). Let ĤP be the Hamiltonian
of an electron in the plane.
2. Give the form of the Bloch functions in the tight-binding approximation. Let k be
the corresponding wave vector. Calculate the dispersion relation EP (k), consid-
ering only nearest neighbours. Use the matrix elements

EP0 = d3 rφ2 (r)ĤP φ2 (r)
and

VP = − d3 rφ2 (r)ĤP φ2 (r + ax) = − d3 rφ2 (r)ĤP φ2 (r + ay) ,
where x, y are unit vectors along the x and y axes.

3. If the electron occupation of this band is one electron per unit cell, what is the
shape of the Fermi ‘surface’? If the number of electrons is 1 ± δ with δ 1,
sketch the Fermi surface in the Brillouin zone. What happens if the band is almost
empty?
Now, and only for this question, consider second nearest neighbours in the plane
using the matrix element

VP = d3 rφ2 (r)ĤP φ2 (r ± ax ± ay) ,
where the notation ± indicates that the four matrix elements are equal by symmetry.
Assume that VP > 0.
380 13 Problem Set
4. For one electron per unit cell, what is the new value of the energy for those
values of k that corresponded to the Fermi surface in the last question? Deduce
the approximate position of the new Fermi surface in the region 0 ≤ kx , ky ≤ π/a.
In the compound YBa2 Cu3 O7 , the planes are in fact coupled into pairs [levels 2 and
3 are coupled in Fig. P3.2 (right)]. We treat the case of an isolated double plane. This
complicates the square lattice basis considered above. We now consider that there
are two orbitals per site of this basis, one for each plane. These are φ2 (r) and φ2 (r +
cz), where z is the vector joining the two planes. Use the following LCAO function:

ψk (r) = exp(ik · Rj ) Ak φ2 (r − Rj ) + Bk φ2 (r + cz − Rj ) ,
j
where the sites Rj run over the points of the Bravais lattice of the plane z = 0 with j
as index. The coefficients Ak and Bk are adjustable parameters.
5. Show that the LCAO function satisfies Bloch’s theorem. The Hamiltonian of an
electron in the double plane is denoted by ĤDP .
Project Schrödinger’s equation ĤDP |Ψk = E(k)|Ψk onto the functions φ2 (r) and
φ2 (r + cz). Simplify the problem by neglecting overlaps between distinct sites, i.e.,
assume that the orbitals φ2 (r) and φ2 (r + cz) are orthogonal.
6. Derive a homogeneous linear system of equations in which the unknowns are the
coefficients Ak and Bk . Specify the coefficients of the system in the form of matrix
elements of ĤDP . Assume that the matrix elements involving only the orbitals of
a given plane are the same as those of ĤP introduced in question 2. For matrix
elements involving orbitals from both planes, keep only the one involving the
same site, viz.,

d3 rφ2 (r − Rn )ĤDP φ2 (r + cz − Rn ) = T .
7. How many bands are there? Give their dispersion E(k) as a function of EP (k)
and T. Plot the result for 0 ≤ kx = ky ≤ π/a. Assuming T small, and when there
are two electrons per unit cell, one from each plane, locate the Fermi energy on
the band diagram E(k) for the given direction. Deduce the shape of the Fermi
surface in the region 0 ≤ kx , ky ≤ π/a, with the help of question 3.
3.3: Chain and Plane
In YBa2 Cu3 O7 , there are also copper–oxygen chains, as discussed in 3.1 and shown
in Fig. P3.2 (right). Consider now a plane coupled with a lattice of chains. To de-
scribe the combined CuO2 plane in level 2 and Cu–O chains in level 1, use the same
Bravais lattice as in 3.2, but consider two orbitals per primitive cell, viz., φ1 (r) of
the chain and φ2 (r) of the associated plane.
1. Start by examining the lattice of Cu–O chains in level 1 using the tight-binding
approximation. These chains are barely coupled together in YBa2 Cu3 O7 and we
may completely neglect matrix elements involving different chains. Using 3.1,
give the dispersion ECL (k). Plot this function for k = (0, ky ) and −π/a ≤ ky ≤
+π/a. Plot the constant energy curves of ECL (k) in the region −π/a ≤ kx , ky ≤
+π/a. With one electron per unit cell in this band, sketch the Fermi ‘surface’.
2. Now investigate the coupled problem of the plane and the chains using the wave
function

Ψk (r) = exp(ik · Rj ) Ck φ1 (r − Rj ) + Dk φ2 (r + cz − Rj ) .
j
The one-electron Hamiltonian is now ĤPC . Assume that the matrix elements
within a given plane are the same as those of ĤP (see question 2 of 3.2: Iso-
lated Copper–Oxygen Plane) and that, in the chain lattice, they are the same as
those of ĤC (see question 4 of 3.1 and question 1 of 3.3). Using the arguments
in 3.2, show that there are two bands E± (k) and express them in terms of EP (k),
ECL (k), and the matrix element

T = d3 rφ1 (r − Rn )ĤPC φ2 (r + cz − Rn ) .
3. Assume that T = 0 and that there are two electrons per unit cell, one per unit cell
of the plane and one per unit cell of the chain lattice. Assume also that EP0 = EC0 .
Plot the Fermi surface of the ensemble in the square 0 ≤ kx , ky ≤ +π/a.
4. Now consider the case T = 0, still with EP0 = EC0 . This coupling is only im-
portant where the Fermi surfaces of the plane and the chains used to intersect.
By examining the neighbourhood of the point k = (π/2a, π/2a), make a qual-
itative sketch of the Fermi surface for the electron occupation of question 3 in
3.3: Chain and Plane, then for an occupation number close to this. What sim-
ple remark can be made about the wave functions at the edge of the region
0 ≤ kx , ky ≤ +π/a? This may be important for explaining the Josephson effect
in YBa2 Cu3 O7 [see Combescot, R., Leyronas, X.: Phys. Rev. Lett. 75, 3732
(1995)].
3.4: Realistic Models of YBa2 Cu3 O7
1. In fact the structure of YBa2 Cu3 O7 comprises two weakly coupled planes (levels
2 and 3) and a chain lattice (level 1 coupled to 2) in the primitive cell. Each
isolated CuO2 plane is described by the dispersion obtained in question 7 of 3.2.
This system is more weakly coupled to the chains by T T in question 4 of
3.3. Figure P3.4 (left) shows the results of a more detailed calculation. Give a
qualitative interpretation of the different parts of the Fermi surface.
382 13 Problem Set
ky ky
π π π π π π
0, a , 0, ,
a a a a a
photoemission
positrons
(0,0) π (0, 0) π kx
,0 kx ,0
a a
Fig. P3.4 Left: Fermi surface obtained from a detailed calculation, adapted from results of Yu, J.,
et al.: Phys. Lett. A 122, 203 (1987) Right: Data taken by photoemission and positron annihilation
adapted from Pickett, W.E., et al.: Science 255, 46 (1992)
2. In Fig. P3.4 (right), black points represent photoemission measurements of the

band structure and white points correspond to results obtained by another tech-
nique using positron annihilation. What parts of the band structure are these
techniques able to reveal?
Solution
Isolated Copper–Oxygen Chain
1. The Bravais lattice of a chain is the set of points Rn = nay, where n ∈ Z and y is
a unit vector along the chain axis. The basis then comprises one copper atom at
R0 = 0 and one oxygen atom at ay/2.
2. The reciprocal lattice is then Kp = (2π/a)p y, where p ∈ Z. The first Brillouin
zone is the interval [−π/a, +π/a].
3. Using the notation adopted in this book, we have
1 ika
ψk (r) = √ e φ1 (r − ay) ,
Nn
where Nn is the number of unit cells.

4. We obtain
EC (k) = EC0 − 2V cos ka .
5. Restricting to the first Brillouin zone, this relation is shown in Fig. P3.5. The
function EC (k) is symmetric under
π
k −→ −k , EC (k) −→ 2EC0 − EC (k) .
a
So half the states are in the sub-interval [−π/2a, +π/2a]. One electron per unit
cell corresponds to a half-filled band, and the Fermi energy is then EC0 . The Fermi
‘surface’ reduces to the two points k = +π/2a and k = −π/2a. If the occupation
is very low, only the bottom of the band close to k ≈ 0 is occupied. Then

1 h̄2
EC (k) EC0 − 2V 1 − k2 a2 , whence m∗ = .
2 2Va2
E C (k)
V
E C0
π π k
− +
a a
Fig. P3.5 Dispersion relation
of the chain obtained in ques-
tion 4, identical to the one −V
given in Chap. 1
384 13 Problem Set
Isolated Copper–Oxygen Plane
1. The Bravais lattice is square: Rn,m = nax + may, n, m ∈ Z. The basis can be taken
as CuO2 with Cu at (0, 0) and two oxygens at ax/2 and ay/2. The reciprocal
lattice is square: Kp,q = (2π/a)px + (2π/a)qy, p, q ∈ Z. The first Brillouin zone
is then the square
π π
π π

− ,+ along kx × − , + along ky .
a a a a
2. We have
plane 1 ik·Rj
ψk (r) = √ e φ2 (r − Rj ) ,
Nn j
where Nn is the number of unit cells and j indexes the sites of the Bravais lattice.
Further,
!
plane plane 1 ik·(Rj −Ri )
EP (k) ψk |ĤP |ψk = e d3 rφ2 (r − Ri )ĤP φ2 (r − Rj ) .
Nn
i,j
The diagonal terms are all equal to EP0 . A given site Ri has four nearest neigh-
bours: Ri + ax, Ri − ax, Ri + ay, and Ri − ay. For just these cases, we have a
nonzero matrix element equal to −VP . Hence,

EP (k) = EP0 − VP eik·ax + e−ik·ax + eik·ay + e−ik·ay ,
and
EP (k) = EP0 − 2VP (cos kx a + cos ky a) .
3. The constant energy terms have the form
cos kx a + cos ky a = C .
We have the symmetry

π π
kx −→ − kx , ky −→ − ky , C −→ −C .
a a
The half-filled state thus corresponds to C = 0, which reduces to
π
|kx | + |ky | = .
a
The Fermi ‘surface’ comprises four straight-line segments [see Fig. P3.6 (center)].
The general shape of the constant energy curves has been given in Chap. 3,
Fig. 3.7. If the number of electrons is 1 + δ, the immediately adjacent states
π ky
+
a
π π
−a + kx
a
π
−
a
ky
π π ky
+ +
a a
π π kx
+ a kx +
a
Fig. P3.6 Fermi surface calculations for a 2D square lattice with nearest neighbour hopping, for
one electron per unit cell (center), (1 + δ) electron per unit cell (right) and for (1 − δ) electron per
unit cell (left)
are filled [see Fig. P3.6 (right)] and if the number is 1 − δ the surface becomes
connected [see Fig. P3.6 (left)]. If the band is almost empty, k ≈ 0 and EP (k) has
almost circular level curves as seen in Fig. 3.7.
4. In the calculation for question 2, for each site Ri , we take into account second
plane plane
neighbours, viz., Ri ± ax ± ay. This produces an extra term in ψk |ĤP |ψk
equal to

+VP eik·(ax+ay) + eik·(ax−ay) + e−ik·(ax−ay) + e−ik·(ax+ay) ,
and then
EP (k) = EP0 + 2VP (cos kx a + cos ky a) + 4VP cos kx a cos ky a .
In the region 0 ≤ kx , ky ≤ π/a, the Fermi surface of question 3 is kx + ky = π/a.

With second nearest neighbours, the energy on this straight line is equal to
EP = EP0 − 4VP cos2 kx a .

386 13 Problem Set
Fig. P3.7 Fermi surface tak- ky

ing into account second near-
est neighbour contributions π
a E = E P0
EF
(0,0) π kx
a
When VP > 0, this energy remains unchanged at the center of the square, but is al-
ways reduced for the other values of kx . The constant energy curve EP0 thus moves
toward the corner of the Brillouin zone. For the half-filled band, the new Fermi
surface takes the form shown in Fig. P3.7. (Note that it divides the Brillouin zone
into two equal areas.)
5. If r is translated to r + R0 for R0 in the Bravais lattice, the dummy sum changes
by
Rj → Rj + R0 , ψk (r + R0 ) = eik·R0 ψk (r) .
6. First project onto φ2 (r) to obtain
φ2 |E(k)|ψk = E(k)Ak ,
since we neglect non-local overlaps. Then we have

φ2 |ĤDP |ψk = Ak e ik·Rj
d3 rφ2 (r)ĤDP φ2 (r − Rj )
j

+Bk eik·Rj d3 rφ2 (r)ĤDP φ2 (r + cz − Rj ) .
j
The first sum only involves matrix elements within a plane. These are equal to
the matrix elements of HP . This first sum is thus equal to EP (k) as obtained in
question 2. In the second sum, only the term Rj = 0 is nonzero and equal to T.
Therefore,
φ2 |ĤDP |ψk = EP (k)Ak + TBk = E(k)Ak .
As the two planes enter the expression for ψk (r) in a symmetric way, the projec-
tion onto φ2 (r + cz) leads in a similar manner to

φ2 (r + cz)|E(k)|ψk = TAk + EP (k)Bk = E(k)Bk .
E(k)
ky
E+ E+
E P0 E−
π kx = ky
a
(0,0) kx
E−
Fig. P3.8 Two bands obtained in the tight-binding approximation for a bilayer of CuO2 planes.
Dispersion relation of the two bands obtained in the direction kx = ky (left). Fermi surface obtained
for one electron per CuO2 unit cell per plane (right)
The coefficients Ak and Bk thus satisfy the system

/
Ak EP (k) + Bk T = Ak E(k)
.
Ak T + Bk EP (k) = Bk E(k)
7. There are eigenstates if the determinant of this system is zero:

EP (k) − E(k) T
=0.
T EP (k) − E(k)
There are two solutions, hence two bands of dispersion
E± (k) = EP (k) ± T .
Along kx = ky , EP (k) = EP0 − 4VP cos kx . The two bands E± are related to one
another by translation [see Fig. P3.8 (left)]. By symmetry, the Fermi energy for
one electron per plane remains equal to EP0 . This Fermi energy cuts the surface
E+ at a level curve of question 3 with occupation 1 − δ, but cuts the surface E− at
a level curve with occupation 1 + δ. The Fermi surface thus comprises two arcs
[Fig. P3.8 (right)].
Chain and Plane
1. Use the LCAO function

1 ik·Rj
ψkCL (r) = √ e φ1 (r − Rj ) .
NCL j
388 13 Problem Set
ECL (0, k y) ky
π /a
π /a
kx
π π ky
− +
a a
Fig. P3.9 Band structure for a lattice of chains: Dispersion relation ECL (k) (left) and constant
energy level curves in the region −π/a ≤ kx , ky ≤ +π/a (right)
Only the matrix elements of ĤC come in when calculating ECL (k) (see question
3 of 3.1), because there is no coupling between chains. As a consequence,
ECL (k) = EC (ky ) ,
as shown in Fig. P3.9 (left). The constant energy curves of ECL (kx , ky ) are thus
straight lines at fixed ky [see Fig. P3.9 (right)]. According to question 5 of 3.1,
with one electron per unit cell, the Fermi surface thus comprises two straight
lines ky = +π/2a and ky = −π/2a, as shown in Fig. P3.10.
2. Project ĤPC |Ψk = E(k)|Ψk onto φ1 (r) and φ2 (r + cz). The calculation is similar
to the one in question 6 of 3.2. For φ1 (r), we obtain
Ck ECL (k) + Dk T = Ck E(k) ,
and with φ2 (r + cz), we obtain
Ck T + Dk EP (k) = Dk E(k) .
ky
+π /2a
π/a
kx
−π /2a
Fig. P3.10 Fermi surface of
the chains for one electron per
unit cell
Setting the determinant equal to zero, this yields two bands:

1 2
E± (k) = EP (k) + ECL (k) ± EP (k) − ECL (k) + 4T 2 .
2
3. Since EP0 = EC0 , the plane and chain bands are filled equally. The Fermi surface
then comprises the surfaces of these two ensembles [see Fig. P3.11 (left)].
4. The surfaces intersect in (π/2a, π/2a) when T = 0. If T = 0, the result of ques-
tion 2 above shows that the equality EP = ECL no longer holds for E+ and E− : the
levels repel one another. Level crossing disappears and we obtain the qualitative
result shown in Fig. P3.11 (right). At the edge of the Brillouin zone, the Bloch
states correspond to states completely within the plane or completely within the
chains.
ky ky
π /a
+π /2a
(0, 0) kx (0,0) π /a kx
Fig. P3.11 Chain and plane Fermi surface for different hopping T between chain and plane (left)
T = 0, (right) T = 0
Realistic Models of YBa2 Cu3 O7
1. A system of two planes with VP = 0 will have a Fermi surface made up of two
segments in [0, π/a]2 , as shown in Fig. P3.12 (left). If there is a chain lattice as
ky ky
+π /a
Double plane
Double plane
Chain
kx
(0,0) +π/a kx
Fig. P3.12 Fermi surfaces for the bilayer of CuO2 planes (left) and for the bilayer coupled to a
CuO chain (right)
390 13 Problem Set
well, it will also give a branch that will avoid crossing the branches of the double
plane by the hybridisation phenomenon of questions 3 and 4 of 3.3. This gives
Fig. P3.12 (right), which agrees with Fig. P3.4 (left).
2. The dispersion points due to photoemission [see Fig. P3.4 (right)] coincide with
contributions coming from double planes. In contrast, positron annihilation sees
the chain contribution. A single technique was not enough initially to investigate
the whole Fermi surface. It has since been viewed by higher resolution ARPES
experiments, and matches the results of a full calculation illustrated in Fig. P3.13.
Fig. P3.13 Full calculation of the 3D band structure. Surfaces due to double planes are very close
together, while those due to chains are widely spaced. The Fermi surface is almost cylindrical,
because the hopping integrals between cells are very small in the c direction. Image courtesy
of O. Andersen and I. Mazin from results published in Andersen, O.K., Liechtenstein, A.I., Ro-
driguez, O., Mazin, I.I., Jepsen, O., Antropov, V.P., Gunnarsson, O., Gopalan, S.: Physica C 185–
189, 147–155 (1991)
Problem 4: Electronic Energy and Stability of Alloys 391
Problem 4: Electronic Energy and Stability of Alloys3
It is important for industrial applications of metals to investigate alloys in order

to obtain as broad a range of properties as possible. For some elements with very
similar ionic radii, alloys can be obtained in which the atoms of the two metals
are randomly distributed at the lattice points of a Bravais lattice. This happens in
particular for many copper-based alloys, denoted here by Cu–B, where the element
B could be zinc, for example (giving brass), or aluminium. Let Cu1−x Bx denote the
composition of such an alloy, where 0 < x < 1 is the concentration of B atoms. These
crystals can be considered as perfect monatomic crystals, where the Bravais lattice
points are occupied by identical ‘average’ atoms, each with probability (1 − x) of
being a Cu atom and probability x of being a B atom.
It is observed experimentally that these alloys can adopt different stable ther-
modynamic structural forms depending on the concentration x. For example, phase
diagrams like those in Fig. P4.1 for Cu–Zn and Cu–Al are observed. Note in partic-
ular that, at low concentrations x, the stable phase at low temperature is the α phase.
This is face-centered cubic (fcc) like pure copper. Above a certain concentration xc ,
the stable phase is the β phase, which is body-centered cubic (bcc).
The British physicists Hume and Rothery tried to understand the origin of this
transition α → β, i.e., from fcc to bcc, along with other structural changes at higher
concentrations x. X-ray crystallography observations showed that, in these alloys,
the average volume occupied by each atom is the same in the two structures. Let af
and ab be the sides of the cubic conventional cells in each structure.
1. Determine the value of af /ab corresponding to the same volume per atom in these
two structures.
In the remainder of this problem, we assume that this condition is always satisfied,
and in particular during the transition fcc → bcc, with ab = 3 Å.
T (◦ C) T (◦C)
1000 1000
900 β γ
900
α
800 α β
γ
800
700
600 700
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5
x (Zn) x (Al)
Fig. P4.1 Phase diagrams for Cu–Zn and Cu–Al alloys
3 This problem has been designed with Y. Quéré.

392 13 Problem Set
To represent the electronic band structures of these alloys, we assume that the
electrons in the filled atomic shells of the Cu+ , Zn2+ , and Al3+ ions form bound
states that fill the deep electronic bands arising from the corresponding atoms. Elec-
trons in the outer valence shells of the atoms will be treated as nearly free and sub-
ject to a periodic potential Vat (r) which may be considered as an average potential.
Hume and Rothery observed that the concentration values xc at which the fcc →
bcc transitions occur in the phase diagrams generally correspond to a well defined
value νc of the number of conduction electrons per atom.
2. Find the average number ν(x) of conduction electrons per atom as a function of
x for the alloy Cu1−x Bx , in the cases B = Zn and B = Al. What is the range of this
function? Determine the value of νc for Cu–Zn and Cu–Al using Fig. P4.1.
Consider now that the conduction electrons behave as independent electrons in a

constant potential (free electrons), and the temperature of the alloy is T = 0 K.
3. Calculate the radius kF of the Fermi sphere for electrons in the alloy with fcc
structure, as a function of ν and 2π/af .
4. What can you say about kF for the bcc alloy?
5. Sketch the energy dependence of the density of states D(E) and the position of
the Fermi level for a given value of ν, in the case of the fcc and bcc structures.
6. State the type and primitive cell of the reciprocal lattices associated with the fcc
and bcc structures. For each of these structures, what is the shortest distance
between the origin and a point of the reciprocal lattice?
7. Show that, for x = 0, the Fermi sphere is entirely contained within the first Bril-
louin zone (BZ) of the fcc and bcc structures.
When the number ν of electrons per atom increases, kF also increases.

8. Find the ‘critical’ values of ν which correspond to a Fermi sphere that is tangen-
tial to the surface of the first Brillouin zone closest to the origin. Denote these by
ν(fcc) and ν(bcc).
Now assume that the conduction electrons behave as nearly free electrons in a peri-
odic potential V(r).
9. Sketch the shape of the curve E(k) in the direction of the vector k0 from the
surface of the BZ closest to the origin. In what way does this differ from the one
for free electrons?
10. Using the E(k) of question 9, explain the form of the density of states for nearly
free conduction electrons shown in Fig. P4.2.
Assume now that the Fermi surface can always be treated as spherical.
11. What radius kF does the point G of the curve in Fig. P4.2 correspond to?
Fig. P4.2 Density of states D(E) G

of conduction electrons in a
nearly free electron approxi-
mation M H
L
EM EG E
The Fourier component of the periodic potential for the vector K of the reciprocal
lattice corresponding to the Bragg plane is VK = 0.5 eV for both the fcc and the bcc
structure.
12. Give the approximate value of EG − EM ?
13. For different values of ν, compare qualitatively the positions of the Fermi level
as given by the free electron and nearly-free electron hypotheses.
14. On the same graph, sketch the energy dependences of the densities of states
Dfcc (E) and Dbcc (E), indicating the points M and G corresponding to the two
structures. For different values of ν, compare the position of the Fermi level in
the fcc and bcc structures.
Assume that, at T = 0 K, the stable structure of the alloy corresponds to the mini-
mum of its total electronic energy.
15. What is this structure when ν = 1.2 and ν = 1.45?
16. What values of ν would the stability limits move to if the Fermi surface were
no longer taken to be spherical and if the shape determined by the periodic
potential V(r) were taken into acount?
394 13 Problem Set
Solution
1. For a monatomic crystal, the average volume Ω per atom is the volume of the
primitive cell, i.e., Ωf = a3f /4 for an fcc structure and Ωb = a3b /2 for a bcc struc-
ture. Equating these, we have
af /ab = 21/3 1.26 .
2. An atom in the alloy contributes on average (1 − x) electrons in the case of Cu

and 2x electrons in the case of Zn (or 3x electrons in the case of Al). We thus have
ν(x) = 1 + x for B = Zn and ν(x) = 1 + 2x for B = Al. ν thus takes values between
1 and 2 in the first case and 1 and 3 in the second. The transition α → β occurs
for x 0.4 in the case of Zn and x 0.20 in the case of Al, so that νc 1.4.
3. We have ν 1/3
kF = (3π 2 n)1/3 = 3π 2 ,
Ω
implying
1/3
2π 3ν
kF = .
af 2π
4. kF is the same for the bcc structure, since we assumed that the primitive cell has
the same volume Ω.
5. The Fermi level is thus the same for each structure. For free electrons, the density
of states per unit volume is (see Sect. 3.3)
3/2
m0
D(E) = √ E1/2 .
π 2h̄3
2
This parabolic form is shown in Fig. P4.3.

6. The reciprocal lattices of the fcc and bcc structures are of types bcc and fcc,
respectively, with cubic conventional cells of side 4π/af and 4π/ab . Their first
Brillouin zones (BZ) are shown in Fig. P4.4. In the first case, the shortest distance
between reciprocal lattice points is the one between a vertex and the center of the
conventional cell. In the second case, it is the distance between a vertex and the
D(E)
Fig. P4.3 Density of states E

per unit volume for free
electrons EF
kz kz
K
K
k0 k0
O kx O
kx
ky ky
4π/af 4π/ab
Fig. P4.4 First BZ of the fcc structure (left) and of the bcc structure (right)
center of an adjacent face. Hence,

√ √
4π 3 4π 2
dfcc = , dbcc = .
af 2 ab 2
7. The distances from the origin of the reciprocal lattice to the plane of the first
Brillouin zone closest to the origin are half the distances calculated in question
6 for each lattice. For the Fermi sphere to lie within the first BZ, we must have
kF < d/2. For ν = 1, we find
2π dfcc 2π dbcc 2π
kF = 0.781 × , = 0.866 × , = 0.89 × .
af 2 af 2 af
We thus find that the condition for inclusion is satisfied in each case.
8. The critical values of ν are determined by the condition kF (ν) = d/2. We find
ν(fcc) = 1.363 , ν(bcc) = 1.48 .
9. The curve has the same parabolic shape as for free electrons, except for k
close to k0 , where it changes curvature and joins with a horizontal tangent onto
E(k0 ) = E0 − |VK |. Here E0 = (h̄2 /2m0 )k02 and VK is the Fourier component of
the periodic potential for the vector K of the reciprocal lattice corresponding
to the Bragg plane. At k0 , there is a gap 2|VK |. Figure P4.5 shows E(k) in an
unfolded band structure diagram.
10. The curve in Fig. P4.5 comprises 4 parts that can be related to the points in
Fig. P4.6:
(i) Between the origin and M, the energy is less than EM and the correspond-
ing states are close to plane waves, for which the density of states has the
parabolic shape for free electrons.
396 13 Problem Set
Fig. P4.5 Dispersion E(k) in E

the direction k0
G
EM
M
k
k0
Fig. P4.6 Density of states D(E) G

of nearly free conduction
electrons
M H
EM EG EG E
(ii) Between M and G, the curve E(k) is distorted in the direction k0 , and has
more states per energy interval due to the curvature. In the other directions,
E(k) remains close in behaviour to the function for free electrons. As a
consequence, the density of states D(E) increases.
(iii) Between G and H, the energy is less than the gap EG −EG in the direction
k0 and neighbouring directions, and there are no states corresponding to
these vectors k. Only those states associated with the other wave vectors
contribute to the density of states, which therefore decreases.
(iv) Above H, the energy lies above EG . There are once again states in the di-
rection k0 and neighbouring directions, and D(E) begins to increase again.
(We are assuming that |VK | is small enough to ensure that the Fermi sur-
face always lies within the first BZ for directions well separated from k0 .
This is the case for metals like Cu–Zn or Cu–Al.)
11. Assuming that the Fermi surface remains spherical, the point G corresponds to
a situation in which this Fermi surface touches the BZ at k0 , so kF is given by
the condition in question 8, i.e., kF = d/2. It follows that
2π 2π
kF (fcc) = 0.866 , kF (bcc) = 0.89 .
af af
12. Still assuming that the Fermi surface is spherical, EG = h̄2 k02 /2m0 . The point M
corresponds to the point M where the curve E(k) in the direction k0 begins to
move significantly away from the free electron curve. The position of M cannot
be estimated from the second order perturbation calculation carried out earlier in
this book, because it was assumed there that Ek − Ek−K is small compared with
|VK |. Now, close to the Bragg plane, and in particular at M , the last two terms
are of the same order of magnitude. We must therefore revert to the original
formula found by diagonalising Ĥ = (p2 /2m0 ) + V(r) in the two-state basis |k,
|k − K. It is easy to show that
.
2
εk−K + εk (εk−K − εk )2
E(k) = − + VK 2 ,
2 2
with εk = h̄2 k2 /2m0 and k0 = K/2. For k = K/2 − q, the expansion to order q2
of the above expression is

h̄2 q2 h̄2 εK/2 2
E(k) = εK/2 + − VK 1 + 2
q
2m0 4m0 VK

h̄2 2 εK/2
= εK/2 − VK + q 1−2 .
2m0 VK
This shows that E(k) has parabolic shape near the BZ. We may estimate that
the point M corresponds to the vector q such that the second term above is of
order −VK . Since εK/2 > VK , we obtain
√
q/k0 VK /εk0 2 , EM εk0 − 2VK .
With the above numerical values, EG (fcc) ∼ 8 eV, EG (bcc) ∼ 8.45 eV, and
EG − EM ∼ 1 eV .
13. If we take ν such that kF lies below the point M (depending on the structure),
the density of states curve is the same as for free electrons and the Fermi level is
also similar. When ν is such that kF (calculated for the spherical Fermi surface)
lies between M and G, the density of states is higher than for free electrons,
and for a given total number of electrons, the last occupied state will have lower
energy than in the case of free electrons. The Fermi level will thus be lower.
Beyond G, the Fermi level gradually climbs to reach the level corresponding to
free electrons.
14. Given the numerical values found for the positions of the points M and G for
the fcc and bcc structures, the density of states curves have the shapes shown in
Fig. P4.7. In the same way, the Fermi level EF (fcc) will be lower than the level
EF (bcc) for ν corresponding to a kF between M1 and D, where D is the point
such that the integral of the density of states between the origin and D is the
same for the two structures. Beyond D and up to the point D , the Fermi level of
the bcc structure is lower than that of the fcc structure.
15. The total electronic energy varies in the same way as the Fermi energy. The
most stable structure will thus be the one giving the lowest Fermi energy. For
ν = 1.2, we are slightly above the point M1 and the stable structure is fcc. But
398 13 Problem Set
Fig. P4.7 Densities of states. D(E ) G2

Continuous curve: fcc struc- G1
ture. Dashed curve: bcc
M1
structure. Dotted curve: free M2
electrons fcc (1)
D
D
bcc (2)
E
E F ( ν = 1.20) E F ( ν = 1.45)
when ν = 1.45, it is the bcc structure that is the most stable, corresponding to a
point close to G2 on the curve of D(E). There is thus a limit to the stability of
the fcc structure for νc = 1.40.
16. Considering a Fermi surface for nearly free electrons, there is a distortion of
the sphere in the direction k0 and equivalent directions. The surface tends to
protrude in these directions. The Fermi surface will thus touch the surface of the
Brillouin zone for lower values of kF than those calculated above. The energy EG
is shifted to the left, but the energy EM is not significantly altered. The density
of states begins to drop for a lower energy. More detailed numerical calculations
must be carried out because the point G no longer corresponds exactly to contact
between the two surfaces. The density of states can continue to increase through
an increase in the number of states for directions close to k0 .
Problem 5: Optical Response of Monovalent Metals 399
Problem 5: Optical Response of Monovalent Metals
The alkali elements have an almost spherical Fermi surface and a free electron
model can explain most of their electronic properties. We thus expect the Drude
model to account correctly for electronic transport. The conductivity predicted by
this model was examined in Question 4.1. The real part σ1 (ω) of the conductivity is
obtained directly by experiment from the reflectivity of a light wave on a metal film.
For potassium, measurements of σ1 (ω) give the result shown in Fig. P5.1.
1. Does the Drude model apply to this experimental result? How could one deter-
mine experimentally the limiting value of σ1 as ω → 0?
To understand the result in Fig. P5.1, we examine the influence of the band structure
on the optical reflection coefficient of the metal in the case of the alkali elements.
Recall that the alkali elements are monovalent and crystallise into body-centered
cubic crystals. Let a be the side of the cube.
2. Characterise the reciprocal lattice and first Brillouin zone. Specify the reciprocal
space directions corresponding to points N of the first Brillouin zone closest to
its center O. Find k0 = |ON|.
For the valence electrons, the band structure of the alkali elements is very well
accounted for in the nearly-free electron approximation.
3. Sketch the first two energy bands in a direction Ok0 : (a) in an extended zone
scheme (b) in a restricted zone scheme.
We may assume that the matrix element of the periodic potential V(r) of the lattice,
which removes the degeneracy at the edge of the Bragg zone, is small enough to
ensure that the Fermi surface is well separated from the edges of the Brillouin zone.
In this case it is easy to find the equations En (k) for the energy bands far from the
Bragg planes by treating V(k) as a perturbation (to second order).
1.5
Re{σ (ω )} (1014 s−1 )
0.5
K
Fig. P5.1 Real part of the 0.5 1 1.5 2

conductivity of potassium h̄ω /EF
400 13 Problem Set
Recall that, if |k are plane wave functions with eigenenergies h̄2 k2 /2m0 , the
matrix elements of a periodic potential satisfy
k |V(r)|k = VK δ(k − k − K) ,
where K is a vector of the reciprocal lattice. For simplicity, we may assume V0 = 0.

4. Give the equations En (k) for the first two bands n = 0 and n = 1, far from the
Bragg planes, in terms of k0 and the VK . Consider only those terms in K giving
dominant contributions to second order.
Apart from the processes whereby a free electron gas can absorb electromag-
netic waves that were discussed in Question 4.1, a photon can also be absorbed
if it has enough energy to excite an electron of given wave vector k into an
unoccupied electron state of the same wave vector in a folded band structure
diagram.
5. On the band structure diagram sketched for question 3, indicate the interband
transition with lowest possible energy h̄ωs .
6. Assuming that the perturbation terms found in question 4 are totally negligible
for k ≤ kF , find h̄ωs as a function of the Fermi energy EF .
7. Can you explain how σ1 (ω) differs from the prediction of the Drude model, as
obtained for potassium (see Fig. P5.1).
Solution
1. The Drude model gives

σ0
σ1 (ω) = .
1 + ω2 τe2
The conductivity σ1 (ω) does indeed fall off steeply with frequency. The hump
on the right of Fig. P5.1 is clearly not described by the Drude model. The ex-
perimental results correspond to optical frequencies and cannot be used to reach
low frequencies. We could obtain σ (0) by measuring the resistivity of the metal
film. Note that the steep decline as a function of ω suggests that ωτe 1 for the
frequencies considered experimentally. This therefore corresponds in the Drude
2
model to σ1 (ω) σ (0)/τe ω−2 . We may thus plot the experimental points ob-
tained for σ as a function of ω−2 .
Figure P5.2 shows that the points are aligned and extrapolate to σ ∼ 1013 s−1
for ω → ∞. This implies that the Drude model gives a fair account up to
h̄ω/EF ∼ 0.6. The Drude contribution to σ1 (ω) is plotted in Fig. P5.3. A further
1.2
σ (10 14 s−1 )
0.8
0.4
Fig. P5.2 Conductivity as a

0 4 8 12 16
function of (EF /h̄ω)2 with ex-
trapolation for ω → ∞ (E F /h̄ ω )2
1.5
Re σ1 ( ω ) (10 14 s −1 )
0.5
Fig. P5.3 Conductivity as a

0.5 1 1.5 2
function of h̄ω/EF . Dashed
curve: Drude contribution h̄ ω /EF
402 13 Problem Set
Fig. P5.4 Reciprocal lattice

and first Brillouin zone
N
O
4π
a
contribution is observed for h̄ω/EF > 0.6. Hence τe and σ (0) = ne2 τe /m can be
obtained directly. (But note that the normalisation of σ1 and the units are not
shown correctly in Fig. P5.1.)
2. The reciprocal lattice is fcc with lattice constant 4π/a. The Bragg planes are
the orthogonally bisecting planes of the vectors joining O to its nearest neigh-
bours. They bound the first Brillouin zone, which is a regular dodecahedron (see
Fig. P5.4). The closest points N are the centers of the faces
√
2π √ 1 π 2 4.44
|k0 | = ON = 2 = = = |k0 | .
a 2 a a
3. Figure P5.5a, b show the changes to the free electron parabola in unfolded and
folded band structure diagrams, respectively.
4. We have
p̂2
Ĥ = + V(r) .
2m
In the unfolded band structure,
| k |V(r)|k|2
E(k) = εk + k|V(r)|k + , (13.3)

εk − εk
k =k
where
εk = h̄2 k2 /2m0 , k|V(r)|k = 0 , k|V(r)|k = VK δ(k − k − K) .
Consequently, only the terms k = k − K contribute to changing E(k) from the

free electron parabola, where K is a vector of the reciprocal lattice.
In the folded band structure, k = k − K corresponds to k, so the third term of
(13.3) only couples states of the same vector k in the folded band structure.
Therefore, for the first band, far from k = k0 , we have
2
h̄k2 2m0 V2k
E1 (k) = − 2 0
+··· ,
2m0 h̄ (2k 0 − k) 2 − k2
+2k0
Free
electron
parabola
−2k0
h̄ωs
EF
−k0 0 k0 2k0 k −k0 0 k k0 2k0 k

(a) (b)
Fig. P5.5 Changes to the free electron parabola in (a) unfolded and (b) folded band structure
diagrams
because the unperturbed state with vector k in the second band corresponds to
k − 2k0 . Likewise,
2
h̄2 (k − 2k0 )2 2m0 V2k
E2 (k) = + 2 0
+··· ,
2m0 h̄ (2k0 − k)2 − k2
far from k = k0 and k = 0. Here we neglect terms due to higher bands for which
the energy denominators are large.
5. The lowest energy interband transition is the one involving electrons at EF . It is
indicated as h̄ωs in Fig. P5.5.
6. Neglecting the perturbation terms, the bands are purely parabolic. Then,

h̄2 2k0 2
h̄ωs = (kF − 2k0 ) − kF = EF 1 −
2 2
−1 .
2m0 kF
kF is determined for free electrons, i.e.,
Ω 4 3 N
πk = ,
8π 3 3 F 2
with N = 1 electron per atom, Ω = a3 /2 (atomic volume), and then
kF = (6π 2 )1/3 /a = 3.90/a ,
and
h̄ωs = 0.64EF .
404 13 Problem Set
7. The absorption induced between bands n = 0 and n = 1 explains the increase in

σ1 (ω) for h̄ω > h̄ωs = 0.64EF .
Colour of Transition Metals
The interband transition h̄ωs is observed for different alkali elements, as shown in Fig. P5.6. It
contributes to an absorption term that covers almost the whole visible spectrum. Indeed, EF is
small (1–3 eV in the series Li, K, Na) and the threshold below 2 eV. In addition, the absorp-
tion is not very selective, since the empty band is broad. The corresponding metals have no
‘colour’.
On the other hand, the noble metals Cu, Ag, and Au are monovalent transition metals
(3d10 4s, 4d10 5s, 5d10 6s). The d and s electron bands overlap, which leads to the situation
shown schematically in Fig. P5.7. In this case, optical transitions between the d band and the
Fermi level occur in the visible part of the spectrum, at energies below h̄ωs (here EF ∼ 7–8 eV).
As the d electrons have a very high density of states (a narrow band containing 10 electrons),
the absorption h̄ωd is intense. It appears just above 2 eV for Cu and Au, which explains their
‘red’ and ‘yellow’ colours, respectively. For silver, the d band is further from the Fermi level
and h̄ωd ∼ 4 eV, so the whole spectrum is reflected.
Fig. P5.6 Interband transi-

tions for three alkali elements.
Adapted from [4, p. 296]
1.5
Re σ1 (ω ) (10 14 s −1)
h̄ω = 0.64EF
Rb
0.5 Na
0.5 1 1.5 2
¯ω
E E
h̄ωd h̄ωs
h̄ωd
EF EF
D(E) 110
Fig. P5.7 Interband transitions for noble metals

Problem 6: One-Dimensional TTF-TCNQ Compounds 405
Problem 6: One-Dimensional TTF-TCNQ Compounds4
Some organic compounds like TTF-TCNQ are made up of weakly coupled chains.
In TTF-TCNQ, tetrathiafulvalene (TTF) and tetracyanoquinodimethane (TCNQ)
chains are weakly bound. This gives these compounds a 1D aspect. The aim in
this problem is to study some characteristic features of these 1D compounds.
6.1: Isolated Chains
In a very simple approximation, we shall describe these compounds as single chains

of N identical molecules, each the same distance a from the next, as shown in
Fig. P6.1. Let R be the coordinate of the th molecule, so R = a. We assume
the usual periodic boundary conditions. We calculate the band structure of each
chain in the tight-binding approximation. The quantum states of the electrons in the
last occupied orbital of molecule are modelled by a single non-degenerate orbital
state |R of energy E0 with R |Rm = δm . We assume that the Hamiltonian Ĥ of
one electron in the chain can be written in the form
p2
N
Ĥ = + V(r − R ) ,
2m
=1
where V(r −R ) is the potential at site . We only take into account hopping integrals
involving the two nearest neighbours. These are denoted as they were in Chap. 1 by
−t0 and −t1 (with t0 and t1 positive).
1. Recall the general form |Ψk for an eigenstate of this Hamiltonian together with
the associated energy eigenvalue Ek . Sketch the shape of the dispersion curve in
the first Brillouin zone.
2. Sketch the wave functions |R with s symmetry, together with the potential con-
tributing to the various hopping integrals entering the expression for Ek .
3. Assume that each molecule gives a conduction electron to the band. What can be
deduced about the electrical transport properties of this chain? Find the Fermi
wave vector kF and the Fermi energy EF .
4. Calculate the total electronic energy of the chain. Hint: It simplifies to integrate
over k.
Fig. P6.1 Simple model for R R R +1

−1
a molecular chain
4 This problem has been designed with F. Rullier-Albenque.

406 13 Problem Set
6.2: Experimental Observations
The structure of TTF-TCNQ is shown in Fig. P6.2. The TTF and TCNQ molecules
are stacked in the direction b, aligned with the chains. (The parameter b is identified
with a in 6.1 and 6.3). The temperature dependence of the conductivity parallel to
the TTF-TCNQ chains is shown in Fig. P6.3.
1. What can be said about the electrical resistivity of this compound?
Figure P6.4 shows topographical images obtained by scanning tunnelling microscopy
of the surface of a TTF-TCNQ crystal in the plane ab at two different temperatures
(70 and 36 K).
2. On Fig. P6.4a, indicate the primitive cell at the surface. Do you recognise the
material basis of Fig. P6.2?
NC CN
C C
NC CN
b
TCNQ
S S
C C
S S
a
TTF c
Fig. P6.2 Chemical structure of TTF-TCNQ (left) and three dimensional stacking of the molecules
(right)
σ̃ = σ (T )/σ (300 K)
Fig. P6.3 Temperature de-

pendence of the conductivity
parallel to TTF-TCNQ chains.
From Cohen, M.J., Coleman,
L.B., Garito, A.F., Heeger,
A.J.: Phys. Rev. B 10, 1298
(1974). With the permission
of the American Physical
Society ( c 1974 APS). http://
RevB.10.1298 T (K)
(a) (b)
Fig. P6.4 Topographical images obtained by scanning tunnelling microscopy of the surface of a
TTF-TCNQ crystal in the plane ab at two different temperatures: (a) 70 K and (b) 36 K. Images
courtesy of Wang, Z.Z.: LPN, Marcoussis
3. The structure is radically different at 36 K. How would you characterise it along

the chains? Across the chains?
6.3: Dimerised Chain
Although the experimental situation in Fig. P6.4b is more complicated, we shall use
a simple model of structural deformation to describe changes in electronic proper-
ties. We assume that the chains dimerise at low temperatures, with each molecule
moving a distance ±u as shown in Fig. P6.5.
1. What is the primitive cell for this structure and the size of the first Brillouin zone?
The aim now is to calculate the new band structure of this compound, still using the
tight-binding approximation. The Hamiltonian for an electron in the chain can now
be written
p2
N/2
−
Ĥ = + V(r − R+
) + V(r − R ) .
2m
=1
2. Explain why solutions of the Hamiltonian will have the form

2 2ika
N/2
−
|Ψk = e ak |R+
+ bk |R ,
N
=1
with |ak |2 + |bk |2 = 1.
a
u −u u −u u −u u
Fig. P6.5 Dimerisation of the R+−1 R−−1 R+ R− R ++1 R−+1 R++2

chain
408 13 Problem Set
In order to find the functions |Ψk and their associated energies Ek , we write down
the eigenvalue equation Ĥ|Ψk = Ek |Ψk for the energies and project onto each of
the states |R+ −
and |R in turn. We still assume that the only nonzero hopping
integrals are those involving nearest neighbours, and use the notation
−t0 = R+ + + − − −
|V (r)|R = R |V (r)|R ,
−t1 = R+ + − + − −
|V (r)|R = R |V (r)|R ,
−t1 = R− + + − − +
−1 |V (r)|R = R−1 |V−1 (r)|R ,

where V+ = V(r − R−
)+ m=l V(r − Rm ) and similarly for V− .
3. Show that we obtain a system of two equations in ak and bk , one of which is

ak E0 − Ek − t0 + bk − t1 − t1 e−2ika = 0 ,
and derive an expression for the other.

4. Calculate the two energy eigenvalues Ek1 and Ek2 in terms of E0 , t0 , t1 , and t1 .
Sketch the band structure, specifying the values at the center and the edge of the
Brillouin zone. What happens to the band structure when u = 0?
5. Would you say this result is compatible with the observations of Fig. P6.3?
For a more accurate analysis of the temperature dependence of the conductivity σ

of TTF-TCNQ, Fig. P6.6 shows the logarithm of σ as a function of 1/T down to the
lowest temperatures.
6. What can you say about the temperature dependence of the conductivity at the
lowest temperatures?
ln σ T (K)
40 20 13.3 10 8 6.7 5.7 5 4.4
8

pendence of the conductivity 0
σ of TTF-TCNQ down to the
lowest temperatures. From
Ferraris, J., Cowan, D.O., −8
Valatka, N., Jr., Perlstein,
J.H.: J. Am. Chem. Soc. 95,
948 (1973). With permission −16
0 50 100 150 200 250
of the American Chemical
Society,
c 1973 10 3/T (K−1)
6.4: Peierls Transition
In this section the aim is to investigate the stability of the dimerised state at T = 0.
From now on, we consider only the nearest-neighbour contribution to the potentials
V .
1. For small displacements u compared with a, justify the following approximations:
t0 ≈ t0 , −t1 ≈ −t1 − αu, and −t1 ≈ −t1 + αu with α > 0.
2. Calculate the change in the total electronic energy resulting from dimerisation
and show that the gain in electron energy compared with the situation u = 0 can
be written in the form

B 1
E = −ANu2 ln − ,
u 2
specifying the values of A and B in terms of t1 and α.
It is given that
π/2
1/2
1 2 4 1
1 − (1 − z2 ) sin2 x dx = 1 + z ln − ,
0 2 |z| 2
for small z.
3. Assuming that the atoms are coupled harmonically by an elastic force with stiff-
ness constant K, calculate the static deformation energy of the chain as a function
of N, K, and u.
4. Discuss the instability of the ion–electron system with regard to dimerisation of
the chain. Find the static displacement ueq at equilibrium as a function of K, t1 ,
and α.
In TTF-TCNQ, charge transfer between chains is such that a chain carries 0.50
electrons per molecule rather than 1 as assumed above.
5. Can you explain the structure actually observed in Fig. P6.4b?
410 13 Problem Set
Solution
Isolated Chain
1. According to Chap. 1, we have
1 ika
N
|ψk = √ e |R ,
N =1
and for all ,

p2
+ V(r − R ) |R = Ĥ0 |R = E0 |R ,
2m
−t0 = R |V (r)|R ,
−t1 = R−1 |V (r)|R = R |V (r)|R+1 ,
where
V (r) = V(r − Rm ) .
m=1,m=
As shown in Chap. 1, the dispersion curve in the first Brillouin zone has the form
Ek = E0 − t0 − 2t1 cos ka ,
shown in Fig. P6.7.

2. From the definition of −t0 , the wave function φ(r − R ) and the potential V (r)
come into the calculation of the transfer integral, where r|R = φ(r − R ). The
representation in Fig. P6.8 is similar to the one in Fig. 1.3c. For the integral −t1 ,
the wave functions at two nearest neighbour sites enter as shown in Fig. P6.9.
3. In k space, each state occupies a ‘volume’ 2π/Na, and taking the spin degeneracy
into account, the band contains 2N states. If each molecule gives 1 electron to
the band, it is thus half-filled. We then find kF = π/2a and EF = E0 − t0 (see
Fig. P6.7). We therefore expect metallic behaviour.
E0 − t0 + 2t1
E0 − t0
k
Fig. P6.7 Dispersion curve E0 − t0 − 2t1

and occupied states in first −π /a 0 π /a
Brillouin zone
φ (r − R )
V (r)
−1 +1 +2
Fig. P6.8 Elements used to calculate −t0
φ (r − R ) φ (r − R +1)
V (r)
−1 +1 +2
Fig. P6.9 Elements used to calculate −t1
4. The total electronic energy of the system is

π/2a
dk t1
E=2 E(k) × = N E0 − t0 − 4 .
−π/2a 2π/Na π
Experimental Observations
1. In Fig. P6.11, for the temperature range 60–300 K, the conductivity goes down
as the temperature goes up (or the resistivity increases with temperature). This is
characteristic of metallic behaviour. Below 60 K, the opposite kind of behaviour
is observed. The system becomes an insulator.
2. Figure P6.12 is a vertical cross-section in the plane ab of the structure shown
in Fig. P6.10. The TTF or TCNQ chains are stacked in the b direction. One
protrudes from the surface, while the other is indented. Several possible primitive
cells are drawn on Fig. P6.12a. Unit cell 1 is also reproduced in Fig. P6.10.
3. At low temperatures, the surface topography observed by STM changes radically.
The surface A corresponding to the high-temperature primitive cell is easy to
make out. However, the periodicity of the structure is totally different. Along
the chains, the period is much longer, apparently multiplied by a factor of 4
(example B). In the direction transverse to the chains, the period appears to be
doubled, but in fact more careful observation shows that it is in fact multiplied
by 4 (example C) (see the comments at the end of the problem).
412 13 Problem Set
Fig. P6.10 Structure of TTF-

TCNQ, showing a primitive
cell on the surface area corre-
sponding to Fig. P6.4a
b
1
a
c
Fig. P6.11 Temperature dep-

endence of conductivity, dis-
playing the metallic and insu-
lating temperature regimes
Metal
σ̃
Insulator
T (K)
(a) (b)
Fig. P6.12 Surface topography of the TTF-TCNQ crystal at two different temperatures: (a) 70 K
and (b) 36 K, showing several possible primitive cells (see text)
Dimerised Chain
1. The new unit cell in Fig. P6.13 has length 2a. The Brillouin zone is the line
segment [−π/2a, +π/2a].
2. The most general form of the wave function is

2
N/2
−
|ψk = ak |R+
> +bk |R ,
N
=1
with |ak |2 + |bk |2 = 1. Applying Bloch’s theorem with period 2a, the two coef-
ficients ak and bk can be written in the form
/
ak = ak e2ika ,
with |ak |2 + |bk |2 = 1 ,
bk = bk e2ika ,
whence the proposed form of |ψk .

3. The eigenvalue equation is
Ĥk |ψk = Êk |ψk .
Picking out the site R+

, we may write Ĥk in the form
p̂2
Ĥk = + V(r − R+ + +
) + V (r) = Ĥ0 + V (r) .
2m
The projection onto the ket |R+
then yields

N/2

R+ +
|Ĥ0 + V (r)| e2ikma ak |R+ −
m + bk |Rm
m=1

N/2

= Ek R+
| e2ikma ak |R+ −
m + bk |Rm ,
m=1
or

ak e2ikla E0 + e2ikla (−t0 ) + bk e2ikla (−t1 ) + e2ik(−1)a (−t1 ) = ak Ek e2ikla .
a 2a
u −u u −u u −u u
R +−1 R−−1 R+ R− R ++1 R −+1 R++2

Fig. P6.13 Dimerised chain
414 13 Problem Set
Therefore,
ak (E0 − t0 − Ek ) + bk (−t1 − t1 e−2ika ) = 0 .
Likewise, the projection onto |R−

yields
ak (−t1 − t1 e+2ika ) + bk (E0 − t0 − Ek ) = 0 .
4. The system of two equations in two unknowns only has nonzero solution if its
determinant vanishes:
(E0 − t0 − Ek )2 − (+t1 + t1 e−2ika )(t1 + t1 e−2ika ) = 0 ,
that is, if

Ek1,2 = E0 − t0 ± t12 + t12 + 2t1 t1 cos 2ka ,
or
Ek1,2 = E0 − t0 ± (t1 + t1 )2 − 4t1 t1 sin2 ka .
We conclude that
/
Ek1,2 = E0 − t0 ± (t1 + t1 ) , for k = 0 ,
Ek1,2 = E0 − t0 ± |t1 − t1 | , for k = ±π/2a .
A band gap thus appears at the edge of the Brillouin zone (see Fig. P6.14). When
u = 0, we have t0 = t0 and t1 = t1 = t1 . Then Ek1,2 = E0 − t0 ± 2t1 | cos ka|, and the
band structure is as shown in Fig. P6.15 in the Brillouin zone [−π/2a, +π/2a].
This is indeed the band structure found in question 5 of 6.3 in the case of a non-
dimerised chain, where the portions corresponding to line segments [π/2a, π/a]
and [−π/2a, −π/a] have been translated by −π/a and +π/a, respectively.
E 0 − t0 2|t1 − t1 |
t1 + t1
k
−π π
0
2a 2a
Fig. P6.14 Band gap at the edge of the Brillouin zone

E0 − t0 + 2t1
E0 − t0 k
E0 − t0 − 2t1
−π −π π π
0
a 2a 2a a
Fig. P6.15 Band structure when u = 0
5. When each molecule contributes one electron to the band, the lower band is
fully occupied and the upper band is empty. We thus expect insulating behaviour
in the case of the dimerised chain. This agrees qualitatively with Fig. P6.11,
where a metal–insulator transition occurs at 60 K, associated with the structural
deformation.
6. For an insulator or semiconductor, the conductivity is always given by
e2 τe e2 τh
σ = ne + nh ,
me mh
where ne and nh are the electron density in the conduction band and the hole
density in the valence band, respectively. The temperature dependence of σ is
governed at low T by those of ne and nh , which are determined by statistical ther-
modynamics. This generally leads to an activated temperature dependence for σ ,
i.e., going as exp(−δE/kB T), where δE is related to the width of the band gap.
Note: The electron–hole symmetry of the band structure requires a Fermi level
in the middle of the band gap Eg . According to (3.56), for kB T EF ,
ne = nh = Nc (T) exp(−Eg /2kB T) ,
where Nc (T) is determined by the effective electron mass, the same in this case
as the effective hole mass. [Replace m0 by me in (3.57).] This therefore leads to
σ = σ0 exp(−Eg /2kB T) .
At low temperatures, the slope of ln σ as a function of 1/T can thus be used

to estimate Eg ∼ 60 K (see Fig. P6.16). In fact, since there are many defects in
compounds like TTF-TCNQ, and these cause extrinsic doping, the value of Eg
obtained in this way is lower than the actual band gap.
416 13 Problem Set
Fig. P6.16 Estimating the ln σ T (K)

band gap from the graph of
ln σ vs. 1/T 40 20 13.3 10 8 6.7 5.7 5 4.4
8
−8
−16
0 50 100 150 200 250
10 3/ T (K−1 )
Peierls Transition
1. When only nearest neighbours are considered in the definition of V , the hopping
integrals −t0 , −t1 , and −t1 can be written in the form
−t0 = R+ − − +
|V(r − R−1 ) + V(r − R )|R ,
−t1 = R− − − +
|V(r − R−1 ) + V(r − R )|R ,
−t1 = R− − − +
−1 |V(r − R−1 ) + V(r − R )|R .
Compared with the non-dimerised situation (u = 0, distance between atoms a),

the overlap integrals now involve either smaller distances (a − u) or larger dis-
tances (a + u). If u is small, the overlap integrals can be estimated by a truncated
expansion to first order in u about the non-dimerised situation.
For t0 , to first order in u, the contributions of the two potentials centered at − and
( − 1)− will cancel, and we obtain t0 = t0 . For t1 and t1 , it is the contributions
from the wave functions centered either at + and − (for − t1 ), or at + and
( − 1)− (for − t1 ), which really matter, and we may write
/
−t1 = −t1 − αu ,
−t1 = −t1 + αu .
For the energies Ek1,2 , we then obtain
0
1
1 α 2 u2
Ek = E0 − t0 ± 2t1 21 − 1 − 2
1,2
sin2 ka .
t1
2. The total electronic energy of the system is written as in question 4 of 6.1:

π/2a dk
E=4 E1 (k) .
0 2π/Na
Therefore,
0
1
π/2a 1
21 − 1 − α u sin2 kadk .
4Na 2 2
E = N(E0 − t0 ) − t1
π 0 t12
Setting z = αu/t1 and making the change of variable x = ka, we obtain

π/2
4N
E = N(E0 − t0 ) − t1 1 − (1 − z2 ) sin2 xdx .
π 0
For small z (small u), this is equal to

4N 1 α 2 u2 4t1 1
E = N(E0 − t0 ) − t1 1 + ln − .
π 2 t12 αu 2
The gain E in electronic energy compared with the non-dimerised situation is

thus equal to

2N α 2 u2 4t1 1
E = − ln − .
π t1 αu 2
Setting A = 2α 2 /π t1 and B = 4t1 /α, the gain in energy E is given by

B 1
E = −ANu2 ln − .
u 2
3. If the atoms are harmonically coupled by an elastic force, the deformation energy
due to dimerisation is
⎡ ⎤
1 ⎣ +
N/2 N/2
2⎦
Ed = K (R − R− 2
− a) + (R+ −
+1 − R − a) ,
2
=1 =1
that is
Ed = 2NKu2 .
4. The change in the total energy of the system compared with the non-dimerised
situation is therefore equal to

B 1
ET = −ANu ln −
2
+ 2NKu2 .
u 2
418 13 Problem Set
Stability is achieved for a displacement ueq such that ∂ET /∂u = 0, i.e., for

B 2K
ueq = exp − .
e A
or
4t1 π Kt1
ueq = exp − 2 .
αe α
5. We showed above that, for a chain with one electron per molecule, the electron–
ion system could be unstable in its ground state with respect to a distortion that
produces a band gap at the Fermi level. In this particular case, the distortion
results from dimerisation.
Polyacetylene provides a simple example of an insulating polymer due to dimeri-
sation of the polymer chain. It is only by doping that it becomes a conducting
compound.
If the occupation of the chain is x electrons per molecule, the corresponding
Fermi wave vector is kF = π/2a. A gap can appear at kF , allowing a gain in
electronic energy, and it is then associated with a distortion of period π/kF .
When x < 1, π/kF > 2a, and this explains the period of about 4a observed in
Fig. P6.12b.
Note: This is the general case of the so-called Peierls transition, named after the physicist who
first described it. Note that, for an occupation x = 1, π/kF a is not necessarily an integer. A
structure can be obtained whose period is not commensurate with the period of the original
lattice, which corresponds to the structure observed in Fig. P6.12b. In the 3D compound, the
distortions of the various chains can accommodate a quasi-ordering in 3D. This explains the
periodicity in the direction transverse to the chains in Fig. P6.12b.
Problem 7: Insulator–Metal Transition 419
Problem 7: Insulator–Metal Transition
The aim here is to identify the parameters determining whether a material is con-
ducting or insulating. In 7.1, we apply the tight-binding (LCAO) method to a crystal
of hydrogen-like atoms to determine the band width of the electronic states arising
from the localised atomic state. In 7.2, the important role played by interactions be-
tween electrons will be examined for the simple example of the hydrogen molecule.
A condition for the solid described in 7.1 to be metallic is then obtained. It will be
compared with experimental observations of hydrogen and the alkali elements in
7.3. A full analogy with doped semiconductors will be made in 7.4. We can then un-
derstand how the physical properties of these materials can change when the donor
concentration is increased.
The calculations in questions 4 and 5 of 7.1 are useful for understanding the rest of the
problem, but not absolutely essential. In 7.2, the problem can be resumed at question 5 if
the reader has not fully processed the preceding questions. 7.4 is independent of 7.2. The
reader need only understand the physical meaning of 7.1 and 7.2 to be able to answer.
7.1: Tight-Binding Method for Hydrogen-Like Orbitals
We consider a crystalline solid comprising a stack of monovalent atoms arranged in

a body-centered cubic structure with conventional unit cell of side b. It is assumed
that the electronic energy bands obtained from inner electron shells are extremely
narrow and well separated from the band arising from electrons of the outermost s
shell of the atom. The latter can be described by the tight-binding approximation. If
ε is the eigenenergy and |φ(r) the wave function of the s state corresponding to the
isolated atom, we look for a crystal wave function of the form

|ψk (r) = Ajk |φ(r − Rj ) ,
j
where the sum over j runs over all sites of the crystal.
1. Specify the primitive cell of the lattice. What is its volume?
It is assumed that:
(a) The potential seen by an electron in the crystal is the sum of the atomic poten-
tials V(r − Rj ).
(b) ϕ(r − Ri )|ϕ(r − Rj ) = δij .
(c) The only terms ϕ(r − Rj )|V(r − R )|ϕ(r − Rm ) that are nonzero are those
involving two nearest-neighbour sites j and m :
−α = ϕ(r − Rj )|V(r − Rm )|ϕ(r − Rj )

−γ = ϕ(r − Rj )|V(r − Rj )|ϕ(r − Rm ) .
420 13 Problem Set
2. Specify the eigenenergy Ek associated with the wave function |ψk (r).
In the case of hydrogen atoms, the only ones considered up to the end of 7.2: Inter-
actions Between Electrons, the corresponding values of V(r), ϕ(r), ε are
q2 e2
V(r) = − =− ,
4π ε0 r r

1 r
ϕ(r) = exp − ,
(π a30 )1/2 a0
ε = −e2 /2a0 ,
where a0 is the Bohr atomic radius and q the elementary charge.
3. Express α and γ as integrals involving a0 and distances rA and rB between elec-
trons at two nearest neighbour sites A and B.
The quantity γ is found analytically to be

e2 R
γ= 1+ exp(−R/a0 ) , R = |AB| .
a0 a0
The quantity α is obtained analytically by noting that it corresponds to the poten-

tial energy of a point charge q placed at B, subjected to the effects of a spherical
charge distribution with electrostatic density qϕ(rA )2 centered at A, for which the
electrostatic potential is easily found. This can be done by decomposing the charge
distribution into two judiciously chosen parts.
4. Find α in terms of R and a0 .
5. Write −α as the sum of two terms, one of which exactly cancels the Coulomb
interaction between the nuclear charges. Check that the second term of α can
be neglected in comparison to γ in the limit R/a0 1, and the assumption (b)
above only actually applies in this limit.
6. Find the width W of the resulting band of electronic states as a function of R and
a0 .
7. What is the Fermi level at T = 0 K? Is the resulting state metallic or insulating?
Does it depend on b?
7.2: Interactions Between Electrons
The result obtained above does not accord with physical reality. This is because
we completely neglected the Coulomb interactions between the electrons. We shall
illustrate this effect in the simple case of a hydrogen-like molecule, comprising two
atoms in which the protons are located at rA and rB and the electrons at r1 and r2 .
The total electronic Hamiltonian for this system is
p21 p2
Ĥ = + V1A + V1B + 2 + V2B + V2A + V12 ,
2m 2m
where
e2 e2 e2 e2
V1A = − , V1B = − , V2A = − , V2B = − ,
|r1 − rA | |r1 − rB | |r2 − rA | |r2 − rB |
and
e2
V12 = + .
|r1 − r2 |
Since Ĥ is independent of the electron spins, the eigenfunctions can be written as a

product of space and spin wave functions. For the two-electron spatial wave func-
tion, we could adopt the basis
|ϕ1A ⊗ |ϕ2B , |ϕ1B ⊗ |ϕ2A , |ϕ1A ⊗ |ϕ2A , |ϕ1B ⊗ |ϕ2B ,
with the obvious notation |ϕ1B = |ϕ(r1 − rB ). However, it makes more sense to
note that Ĥ is invariant under exchange of the protons A and B, effected by the
operator T̂AB , which implies that Ĥ and T̂AB have a common basis of eigenfunc-
tions. Therefore, the eigenfunctions of Ĥ are either symmetric or antisymmetric
under exchange of A and B, and we may choose the following basis of spatial wave
functions:
1
|Φ1 = √ |ϕ1A ⊗ |ϕ2B + |ϕ1B ⊗ |ϕ2A ,
2
1
|Φ2 = √ |ϕ1A ⊗ |ϕ2B − |ϕ1B ⊗ |ϕ2A ,
2
1
|Φ3 = √ |ϕ1A ⊗ |ϕ1A + |ϕ1B ⊗ |ϕ2B ,
2
1
|Φ4 = √ |ϕ1A ⊗ |ϕ2A − |ϕ1B ⊗ |ϕ2B .
2
Furthermore, since electrons are fermions, the Pauli exclusion principle requires
the total wave function, including space and spin components, to be antisymmetric
under exchange of electrons 1 and 2.
1. Which of the functions |Φ1 , |Φ2 , |Φ3 , and |Φ4 are associated with a spin
singlet state, and which with a spin triplet state? Let |ΦnM be the total wave
functions thus obtained, with M = S for a singlet and M = T for a triplet.
2. From the properties given in 7.2: Interactions Between Electrons, explain without
calculation why the states obtained in question 1 are eigenstates of Ĥ.
422 13 Problem Set
The aim now is to find the eigenvalues and eigenstates of Ĥ when R a0 (with
R = rB − rA ). As we have seen in 7.1, we will then be able to neglect integrals of
type α and retain only those of type γ . Moreover, among all the matrix elements of
V12 , we shall retain only the biggest, involving two electrons on the same atom. The
corresponding energy, called the intra-atomic Coulomb energy, is
U = ϕ1A | ⊗ ϕ2A |V12 |ϕ1A ⊗ |ϕ2A .
3. Show in this limit that the eigenstates found in question 2 have eigenenergies 2ε
and 2ε + U, and that the 2 × 2 matrix representing Ĥ in the subspace spanned by
the other two states is

2ε + U −2γ
.
−2γ 2ε
4. Specify the energies of the two-electron states. Note that, in the limit R → ∞,
only two levels remain. What eigenstates do they correspond to?
Including the Coulomb interaction V12 between the electrons therefore disallows
double occupation of an atomic orbital in the ground state. We may thus apply an
independent electron description of the kind discussed in 7.1, provided that the spin
degeneracy is removed in the one-electron state of energy ε, giving rise to a state
of energy ε and an excited state of energy ε + U. In the hydrogen molecule, the
effect of the γ terms is, as in 7.1, to remove the degeneracy in each of these states.
Extending this remark to a crystal, it can be shown that the electron levels at ε and
ε + U give rise to energy bands with the width W determined in question 6 of 7.1.
5. Assuming this result, under what conditions is the solid insulating? When does it
become metallic?
In order to make a quantitative estimate of when the insulator–metal transition will

occur, we must estimate U.
6. Express U in its integral form for hydrogen. Use the result for α in question 4 of
7.1 to show that U = 5e2 /8a0 .
7. Check that the metal–insulator transition occurs when the body-centered cubic
solid has a number nIM of atoms per unit volume given by
1/3
nIM a0 = C ,
where C is a constant of order unity to be determined.

8. Show that C depends little on the chosen crystal structure by considering the
case of a simple cubic crystal.
7.3: Alkali Elements and Hydrogen
The method of 7.1 and 7.2 applies to monovalent atoms, such as the alkali elements,
but the numerical result found in question 7 of 7.2 does not apply directly. Indeed,
the valence electrons of the alkali elements are in ns states where, e.g., n = 2 for Li,
n = 3 for Na. The corresponding radial wave functions ϕn do decrease exponentially
at long range, but become small for r close to zero, because the s electron barely
penetrates the inner electron shells. One can use a pseudopotential approximation,
with an infinite potential for r < Rc , where the ionic radius Rc increases with the
atomic number. It is not difficult to see
that, if the wave function is approximated by
ϕn (r) = const. × exp − (r − Rc )/a0 for r > Rc , the condition of question 7 of 7.2
is altered to a first approximation by replacing R by R − 2Rc .
1. Use the values of Rc and the lattice constants of the body-centered crystal lattices
of the various alkali elements given in Table 13.1 to decide whether this model
suggests the metallic behaviour observed for these crystals.
Hydrogen solidifies at low temperatures (T ∼ 15 K) into an insulating molecular

crystal. The hydrogen molecules are located at the lattice points of a face-centered
cubic lattice with lattice constant 5.4 Å.
2. Calculate the number of hydrogen atoms n0 per unit volume in the solid state.
Can one explain why this phase should be insulating?
3. Can you imagine how one could obtain metallic solid hydrogen? In your opinion,
under what conditions could certain alkali elements become insulating?
Table 13.1 Ionic radius Rc and lattice constant for various alkali elements
Li Na K Rb Cs
Rc (Å) 1.72 2.08 2.57 2.75 2.98
a (Å) 3.49 4.23 5.23 5.59 6.05
7.4: Insulator–Metal Transition in Si–P
The model exposed in 7.1 and 7.2 is difficult to test in detail for hydrogen or the
alkali elements. We shall see, however, that it applies to doped semiconductors with
a few minor modifications. Indeed, if we consider silicon doped with phosphorus,
at zero temperature the phosphorus atom (valence 5) substituting a silicon atom
(valence 4) traps its excess electron in a hydrogen-like orbit with an energy level
below the conduction band. The ionisation energy of the donor is the energy of the
electron in its enormous Bohr orbit. If a large number nd of P atoms are introduced
per unit volume, we obtain a disordered system of hydrogen-like orbits. Assuming
that this disordered state can be modelled by a cubic crystal with lattice constant
424 13 Problem Set
equal to the average distance between donors, we expect the model of 7.1 and 7.2 to
be applicable.
1. Adapt the condition obtained in question 8 of 7.2 to the case of silicon doped
with phosphorus.
2. What band structure would you expect for different concentrations nd of phos-
phorus? Sketch the energy dependence of the electron densities of states for Si–P
in the three cases nd nIM , nd ∼ nIM , and nd nIM . Indicate the position of the
Fermi level at zero temperature.
The resistivity of Si–P has been carefully measured for various phosphorus concen-
trations. It can be seen from Fig. P7.1 that the behaviour changes significantly for
different values of nd .
3. Find the concentration nIM from the experimental results. Given that the dielec-
tric constant of Si is ε = 11.75 and the effective mass is me /m0 = 0.33, compare
the experimental value of C with the one obtained in question 8 of 7.2.
In the following, the aim will be to understand the physical behaviour observed in
the two extreme cases nd nIM and nd nIM .
4. From Fig. P7.1, deduce the order of magnitude of the electron collision time τ at
low temperatures in the most metallic case. What could be the physical origin of
the observed temperature dependence? Why is it less than in pure metals?
5. What characteristic energy of the band structure in question 2 can be deduced by
analysing ρ in the insulating case? Estimate the order of magnitude and compare
with the ionisation energy of phosphorus.
6. How would you expect this energy to depend on the concentration nd ?
10
Si–P
3.75 3.70
(10 24/m 3 )
1

ρ ( Ω cm)
pendence of the resistivity

(logarithmic scales) for four 10−1
samples of Si–P. Here nd
is given in units of 1024
atoms/m3 . From Rosen-
3.84
baum et al.: Phys. Rev. B
27, 7609 (1983). With the 10−2 7.0
permission of the Amer-
ican Physical Society c
1983. http:/link.aps.org/doi/ 10−2 10−1 1 10 10 2
10.1103/PhysRevB.27.7509 T (K)
Fig. P7.2 Magnetic suscepti- 10 −2

bility (per kg of Si) measured 1.1 K
at three different temperatures 4.2 K
as a function of the concen- 77 K
tration nd (in atoms per m3 ).
10 −3
χ (J/tesla 2 kg)
Logarithmic scales. Adapted
from Alloul, H., Dellouve, P.:
J. de Physique C8-49, 1185
(1988)
10 −4
10 −5
10 23 10 24 10 25 10 26
nd (m−3 )
The insulator–metal transition is also accompanied by a major change in the elec-

tronic magnetic properties. This is shown in Fig. P7.2, which gives the temperature
dependence of the electronic magnetic susceptibility χ . Very different behaviour is
observed in the cases nd nIM and nd nIM .
7. In the case nd nIM , what magnetic behaviour is expected? Give the expected
expression for χ . In Fig. P7.2, plot the calculated expression for the three tem-
peratures T = 1.1 K, T = 4.2 K, and T = 77 K. For what limiting case do the
results accord with observation? The density of Si is 2.33 × 103 kg/m3 .
In the limiting case nd nIM , we may assume that the impurity and conduction
bands merge into a single parabolic band.
8. Give the expected expression for the magnetic susceptibility in this case. How
does it depend on nd and T? From what concentration nd is this assumption
valid?
426 13 Problem Set
Solution
Tight-Binding Method for Hydrogen-Like Orbitals
1. The cubic conventional cell of the body-centered cubic lattice contains two lattice
nodes. The Bravais lattice can be constructed in infinitely many different ways,
e.g., specifying the three basis vectors (see Fig. P7.3)
b
b1 = bx , b2 = by , b3 = (x + y + z) .
2
The volume of the primitive cell is b3 /2. A more symmetrical cell is given by
b b b
b1 = (y + z − x) , b2 = (z + x − y) , b3 = (x + y − z) .
2 2 2
2. Ek is found exactly as done in Chap. 3. For a crystal with N atoms, this gives

|ψk (r) = N −1/2 eik·Rj |ϕ(r − Rj ) ,
j

Ek = ψk (r)|Ĥ|ψk (r) = ε − α−γ eik·(Rj −Rj ) .
j neighbour j neighbour
of j of j
In the body-centered cubic lattice, an atom has 8 neighbours (the centres of the
cubes for which this atom occupies a vertex). The coefficient of γ is then

b b b b
exp i (εx kx + εy ky + εz kz ) = 8 cos kx cos ky cos kz ,
2 2 2 2
εx =±εy =±εz =±1
b3
b2
Fig. P7.3 Constructing the
Bravais lattice of the bcc b1
x
crystal
and then

b b b
Ek = ε − 8α − 8γ cos kx cos ky cos kz , (13.4)
2 2 2
where k belongs to the first Brillouin zone. Here γ and 8α correspond to t1 and
t0 , respectively, in Sect. 3.2.3. The reciprocal lattice of the body-centered cubic
lattice is face-centered cubic with conventional cell of side 4π/b, and the first
Brillouin zone is shown in Fig. P4.4 (right) (see Problem 4: Electronic Energy
and Stability of Alloys).
√
3. R = b 3/2 is the distance between nearest neighbours. Let A and B be two atoms
at this distance R. With the notation in Fig. P7.4,

1 r A e2 rA 3
α= exp − exp − d rA
π a30 a0 r B a0

e2 2rA 1 3
= exp − d rA . (13.5)
π a30 a0 rB

e2 r A e2 rB 3
γ = exp − exp − d rA . (13.6)
π a30 a0 r A a0
4. α/q is the electrostatic potential created at B by a charge distribution
q ϕ(rA )|ϕ(rA ) .
The integral giving α can be decomposed into an integral over the sphere of
radius R and an integral over the region outside this sphere:

e2 2rA 1 3 2rA 1 3
α= exp − d rA + exp − d rA .
π a30 rA <R a0 rB rA >R a0 r B
To determine the first integral, the charge can be assumed to be concentrated at

A, by Gauss’ theorem. The second integral corresponds to the potential produced
at B by the charge distribution outside the sphere. Since the resulting electric
field within the sphere is zero, by Gauss’ theorem again, the electric potential is
rA
rB
A B
R
Fig. P7.4 Notation for ques-

tion 3
428 13 Problem Set
constant. It thus suffices to work it out at A, so that

∞
e2 1 R 2rA 2rA 1
α= exp − 4π rA drA +
2
exp − 2
4π rA drA
π a30 R 0 a0 R a0 r A
∞
e2 1 ∞ 2rA 2rA 1 1
= exp − 4π r 2
dr
A A + exp − − 4π r 2
dr
A A
π a30 R 0 a0 R a0 rA R

e2 4e2 ∞ 2rA 1 1 2
= + 3 exp − − r drA .
R a0 R a0 rA R A
Hence
∞
e2 4e2 R
α= + x − x exp(−2x)dx ,
R R R/a0 a0
where we have set x = rA /a0 . Integrating by parts with

R
u= −x x , dv = exp(−2x)dx ,
a0
we obtain
∞
e2 2e2 R
α= + − 2x exp(−2x)dx .
R R R/a0 a0
A second integration by parts gives

e2 e2 R 2R e2 e2 a0 2R
α= − 1+ exp − = − 1+ exp − .
R R a0 a0 R a0 R a0
(13.7)
5. The first term of −α is precisely the Coulomb interaction energy between two
charges +q and −q at A and B. It thus balances the Coulomb interaction energy
between the protons in the total energy of the system of electrons plus nuclei. For
the change in the electronic levels, we thus retain only the second term of α in
the expression (13.4) for Ek . According to the statement of the problem,

e2 R R
γ= 1+ exp − . (13.8)
a0 a0 a0
For R/a0 1,

e2 R R e2 e2 2R
γ exp − , α − − exp − ,
a0 a0 a0 R a0 a0
which yields

e2

α − γ .
R
Note: From the expression (13.6) for γ , it is clear that, when R/a0 1, the
charge distributions are concentrated around A and B (see Fig. P7.5). The domi-
nant contribution to γ comes from rA ∼ rB ∼ R/2, so that the factor exp (−R/a0 )
dominates the expression (13.8) for γ . The same remark leads to

ϕ(r − rA )|ϕ(r − rB ) ∝ exp(−R/a0 ) 1 ,
and this justifies neglecting the overlap integrals ϕ(r − Rj )|ϕ(r − Ri ) in ques-
tion 2.
6. Since the term 8α is negligible in (13.4), the atomic level ε gives rise in the solid
to an unshifted energy band on average (see Fig. P7.6). The lowest level is ε −8γ ,
obtained for k = 0, and the highest level is ε + 8γ , obtained for k = 2π x/b, i.e.,
for six vertices of the polyhedron representing the first Brillouin zone, viz., those
with 4 adjacent faces [see Fig. P4.4 (right) of Problem 4: Electronic Energy and
Stability of Alloys]. The band width is W = 16γ .
7. The resulting band has N spin degenerate states and can therefore accommodate
2N electrons. There are N electrons to attribute. The Fermi level at T = 0 is
in the middle of the band for all R. Although the band narrows very quickly
when R increases, it is always half filled and thus corresponds to a metallic state.
This clearly contradicts what is observed physically because, when the atoms are
widely separated, the electrons are localised on atomic sites and no conduction
can occur.
rA rB
A B
Fig. P7.5 Charge distributions for R a0
ε + 8γ
ε
ε − 8γ
Atom Solid
Fig. P7.6 Atomic level ε and resulting energy band
Interactions Between Electrons
1. We have
1
|Φ1 = √ |ϕ1A ⊗ |ϕ2B + |ϕ1B ⊗ |ϕ2A ,
2
430 13 Problem Set
1
|Φ2 = √ |ϕ1A ⊗ |ϕ2B − |ϕ1B ⊗ |ϕ2A ,
2
1
|Φ3 = √ |ϕ1A ⊗ |ϕ2A + |ϕ1B ⊗ |ϕ2B ,
2
1
|Φ4 = √ |ϕ1A ⊗ |ϕ2A − |ϕ1B ⊗ |ϕ2B .
2
The Hamiltonian is spin-independent, so the eigenfunctions separate into a prod-
uct of a spatial wave function and a spin wave function. For fermions, the Pauli
exclusion principle requires the total wave function to be antisymmetric with
respect to interchange of the electrons (1, 2). Therefore, for a symmetric spa-
tial wave function under the exchange (1, 2) there corresponds an antisymmetric
wave function under the exchange (1, 2), i.e., a spin singlet state. Likewise, for an
antisymmetric spatial wave function, there corresponds a triplet spin state. This
means that only |Φ2 is associated with a spin triplet, while |Φ1 , |Φ3 , and |Φ4
are all associated with spin singlets. The basis is thus denoted by
|Φ1S , |Φ2T , |Φ3S , |Φ4S .
Note that the Pauli exclusion principle has had the effect of reducing the number
of possible states from 16 to 6.
2. The Hamiltonian Ĥ commutes with T̂AB and also with the total spin operator
S2 = (s1 + s2 )2 . A common eigenbasis can thus be found for these three opera-
tors. Since |Φ2T is an eigenstate of S2 orthogonal to the three other states, which
are spin singlets, |Φ2T is also an eigenstate of Ĥ. Furthermore, |Φ1S and |Φ3S are
symmetric under exchange of A and B, while |Φ2T and |Φ4S are antisymmetric
under exchange of A of B. The eigenstates of Ĥ are thus either a linear combina-
tion of |Φ1S and |Φ3S or a linear combination of |Φ2T and |Φ4S . Since |Φ2T is
an eigenstate of Ĥ, |Φ4S is thus also an eigenstate of Ĥ.
3. We seek the energy eigenvalues of the states |Φ2T and |Φ4S . To do this, we set
Ĥ = Ĥ0 + V̂12 with
p21 p2
Ĥ0 = + V1A + V1B + 2 + V2A + V2B .
2m 2m
Consider, for example, |Φ2T , which has eigenenergy given by

E2T = Φ2T |Ĥ0 + V̂12 |Φ2T = Φ2T |Ĥ0 |Φ2T + Φ2T |V̂12 |Φ2T . (13.9)
With the assumption concerning the matrix elements of V̂12 , the second term is
zero because
1
|Φ2T = √ |ϕ1A ⊗ |ϕ2B − |ϕ1B ⊗ |ϕ2A
2
does not contain any terms where the two electrons are at the same site. To cal-
culate the first term of (13.9), note that

p21
Ĥ0 |ϕ1A ⊗ |ϕ2B = + V1A + V1B |ϕ1A ⊗ |ϕ2B
2m

p22
+|ϕ1A ⊗ + V2A + V2B |ϕ2B
2m

= (ε + V1B )|ϕ1A ⊗ |ϕ2B + |ϕ1A ⊗ (ε + V2A )|ϕ2B .
(13.10)
We obtain a nonzero matrix element in (13.9) by taking the bra containing ϕ2B |
for the first term of (13.10), and ϕ1A | for the second. This shows that, in the
calculation of E2T , the cross terms disappear and
1
E2T = 2 × ϕ1A | ⊗ ϕ2B |H0 |ϕ1A ⊗ |ϕ2B
2
= 2ε + ϕ1A |V1B |ϕ1A + ϕ2B |V2A |ϕ2B
= 2ε − 2α 2ε .
Likewise for
1
|Φ4S = √ |ϕ1A ⊗ |ϕ2A − |ϕ1B ⊗ |ϕ2B ,
2
we have
E4S = Φ4S |Ĥ0 |Φ4S + Φ4S |V12 |Φ4S = 2ε − 2α + U 2ε + U .
Applying the same argument to |Φ1S and |Φ3S , we immediately find
Φ1S |Ĥ|Φ1S = 2ε − 2α 2ε ,
Φ3S |Ĥ|Φ3S = 2ε − 2α + U 2ε + U .
Regarding the off-diagonal matrix element, Φ3S |Ĥ|Φ1S = Φ3S |Ĥ0 |Φ1S . Taking
the term |ϕ1A ⊗ |ϕ2B in |Φ1S , and considering (13.10) and
1
Φ3S | = √ ϕ1A | ⊗ ϕ2A | + ϕ1B | ⊗ ϕ2B | ,
2
it is clear that we must associate the second term of Φ3S | with the first term of
(13.10) and vice versa. This yields

1 1
Φ3 |Ĥ|Φ1 =
S S
ϕ1B |V1B |ϕ1A + ϕ2A |V2A |ϕ2B × 2 = −2γ .
2 2
432 13 Problem Set
The matrix representing Ĥ in the subspace spanned by |Φ1S and |Φ3S is thus

2ε −2γ
Ĥ = .
−2γ 2ε + U
4. The associated eigenenergies are
(2ε − E)(2ε − E + U) − 4γ 2 = 0 ,
or

1
E± = 2ε + U ± U + 16γ
2 2 .
2
Note that E+ > 2ε + U and E− < 2ε, leading to the level structure in Fig. P7.7.
U is an intra-atomic quantity that is independent of the distance between the two
atoms. However, γ decreases quickly when the atoms are far apart. In the limit
rB − rA → ∞, we have γ → 0, E+ → 2ε + U, and E− → 2ε. The corresponding
eigenstates are then |Φ1S for E− and |Φ3S for E+ .
The energy 2ε corresponds to the states |Φ2T and |Φ1S for which only terms
of the type |ϕ1A ⊗ |ϕ2B arise, i.e., in which one electron is localised on each
orbital. The energy 2ε + U corresponds to the states |Φ4S and |Φ3S for which
only terms of the type |ϕ1A ⊗ |ϕ2A arise, i.e., with two electrons on the same
orbital.
Consequently, in the ground state of the system, the two electrons are not simul-
taneously in the same orbital. In the excited state, two electrons coexist in the
same atom, which is thus an H− ion, and U is the extra energy to be supplied to
get two electrons into the same atom.
5. In the solid, as in 7.1, hybridisation due to the γ term broadens the electron states
at ε and ε + U by W. The band due to ε can only accommodate N electrons.
It is thus fully occupied for W < U, and the material is an insulator. On the
other hand, for W > U, the bands overlap and the system becomes metallic. This
happens if the lattice parameter b is varied. The insulator–metal transition occurs
for U = W (see Fig. P7.8).
E+
2 ε + U = E 4S E 4S = E 3S = 2 ε + U
U U
2ε = E 2T E2T = E 1S = 2 ε
E−
(a) (b)
Fig. P7.7 Level structure obtained for the two electron states when (a) γ = 0, (b) γ = 0
ε +U W ε +U W
EF
ε W
ε W
0 1/b
Fig. P7.8 Left: Broadening of the states at ε and ε + U. Right: Overlap of the bands for W > U
Fig. P7.9 Notation for question 6 r2

r 12
A r1
6. By definition, with the notation in Fig. P7.9,

2
1 2r1 e2 2r2 3 3
U= exp − exp − d r1 d r2
π a30 a0 |r2 − r1 | a0
2
1 2r 1 2r 2 e
= d3 r1 exp − exp − d3 r2 .
π a30 a0 a0 r12
The second integral over r2 corresponds exactly to the one in (13.5) for the cal-
culation of α. So according to (13.7), with r1 = R,

e2 1 1 a0 2r1 2r1 3
U= − 1+ exp − exp − d r1 .
π a30 r1 a0 r1 a0 a0
Hence,
4e2
∞ a
a0

2r1

2r1
2
r1 dr1
0
U= − 1+ exp − exp −
a0 0 r1 r1 a0 a0 a0 a0
2 ∞
4e
= x exp(−2x) − (x2 + x) exp(−4x) dx .
a0 0
After several integrations by parts,
5e2
U= . (13.11)
8a0
7. The metal–insulator transition occurs for
U = W = 16γ ,
434 13 Problem Set
Fig. P7.10 Finding the con-

dition for the metal–insulator
transition
R
exp −
a0
−1
5 R
1+
5 128 a0
128
−1 1 R/a 0
so using (13.8) for γ and (13.11) for U (see Fig. P7.10),

R 5
exp − = .
a0 128(1 + R/a0 )
Graphically, the solution is seen to correspond to R/a0 1. Numerically,

√ we
obtain R/a0 =√ 5.5. For the body-centered cubic lattice, R = b 3/2, and n =
2/b3 , so R = ( 3/22/3 )n−1/3 and R/a0 = 5.5 corresponds to the condition for
the insulator–metal transition
1/3
nIM a0 = 0.218 = Cbcc .
8. In the simple cubic crystal, there are only six nearest neighbours, and the condi-
tion is W = 12γ or
R 5
exp − = ,
a0 96(1 + R/a0 )
which yields R/a0 = 4.7. Since n = R−3 , we then obtain
1/3
nIM a0 = 0.213 = Csc .
For a face-centered cubic lattice (12 neighbours), we find
1/3
nIM a0 = 0.204 = Cfcc .
The constant C depends little on the crystal structure.
Alkali Elements and Hydrogen
1/3
1. The condition nIM a0 = 0.218 or R/a0 = 5.5 for a body-centered cubic lattice
√
becomes (RIM − 2Rc )/a0 = 5.5, or RIM /a0 = 5.5 + 2Rc /a0 with R = b 3/2. We
obtain the values shown in Table 13.2. We always have R < RIM and the density
is well above the density required to obtain a metallic state.
Table 13.2 Key parameters for alkali elements

Li Na K Rb Cs
b (Å) 3.49 4.23 5.23 5.59 6.05
Rc /a0 1.13 1.79 2.51 2.79 3.19
RIM /a0 7.76 9.08 10.52 11.08 11.88
R/a0 5.70 6.91 8.55 9.13 9.89
The approximation here is somewhat rough. Indeed, although the pseudopotential approxima-
tion takes into account the existence of the ionic core (see Fig. P7.11), the radial wave func-
tions do not have the simple form used here. However, for a given distance R, the integrals γ
are much larger than for hydrogen atoms, and the approximation is not disastrous, since we did
not expect to determine the exact density at which the alkali metals would become insulating.
The above results nevertheless suggest something which is confirmed by experimental ob-
servations: it is easier to make Cs or Rb insulating than Li or Na. When R < RIM , the integrals
are such that W U, and in this case U no longer plays the dominating role in the electronic
properties. In this case, the free electron model is applicable and the interactions between elec-
trons introduce minor modifications.
Fig. P7.11 Pseudopotential 2Rc 2Rc
approximation |ϕ |
A B
V (r)
2. The atomic volume is v = n−1 = a3 /8 = 19.7 Å3 . This gives n1/3 a0 = 0.196, just
less than the density required for atomic hydrogen to give a metallic solid.
In fact, in the case of molecular hydrogen, the electronic bands arise from energy levels cor-
responding to molecular orbitals, and an insulator–metal transition can occur when the fully
occupied band resulting from the binding orbital overlaps with a higher energy band. We must
therefore compare the hopping integral γ (the analog of γ ) between molecular orbitals, which
decreases exponentially with the distance between hydrogen molecules, with the energy sep-
aration of the molecular levels, which has the same order of magnitude as U. It is not hard to
show that γ for the molecular solid is likely to be less than γ for the monatomic solid of the
same density. One thus expects the density required to obtain a molecular conducting state to
be well above the value given by the above condition.
3. It is clear that hydrogen must be compressed to obtain a metallic state.
Pressures of the order of 200 GPa (2 Mbar) would nevertheless be required to obtain a metallic
state. In the laboratory, such pressures can be only yet be obtained by transient methods, viz., a
shock wave produced by a violent explosion. Very high density hydrogen must certainly exist
in some planets of the Solar System. The magnetic field of Jupiter detected by Voyager I in
1975 can only be explained by the existence of metallic hydrogen in this planet, at pressures of
several hundred GPa.
436 13 Problem Set
As far as the alkali elements are concerned, to make them insulating, their den-
sity must be reduced. (In the liquid phase, close to the critical point, Cs and Rb
become insulating.) Another way to obtain a metal–insulator transition is to pro-
duce an alloy of the alkali element with a noble gas. This would tend to distance
the alkali atoms from one another.
This effectively produces a metal–insulator transition, but one has a disordered system of
hydrogen-like atoms rather than a crystal, and the transition is in this case more closely re-
lated to the existence of disorder (Anderson transition) than to a consequence of interactions
between electrons (Mott transition).
Insulator–Metal Transition in Si–P
1. The phosphorus nucleus and the extra electron it contributes to the silicon lat-
tice constitute a large hydrogen-like atom. The interaction potential between the
electron and nucleus is reduced by the high dielectric constant ε, and the effec-
tive electron mass is reduced (me < m0 ). The main effect is to transform a1 to
ap = εa1 m0 /me . If the phosphorus atoms are distributed over a regular lattice,
the arguments of 7.2 can be applied, and we conclude that
1/3
nIM ap 0.20 .
2. The energies of states related to the impurity are Ed = Ec − E0 , where E0 =

e2 /2ap is the ionisation energy of the donor, and EP− = Ed + U corresponding to
a phosphorus that traps two electrons (P− ). The latter is slightly higher than Ec
because U = 5e2 /8ap = 5E0 /4. It thus lies in the conduction band of silicon.
The case nd nIM corresponds to the one described in Fig. 4.12a in Chap. 4 of
the lecture notes, except that the P− band is superposed on the conduction band.
EF lies between Ed and Ec at T = 0, and the impurity band is fully occupied
(see Fig. P7.12). As nd increases, the widths of the two impurity electronic bands
also increase. Close to the conduction band, the electronic band structure is as
shown in Fig. P7.13. For nd > nIM , the bands intersect and a metal is obtained
(see Fig. P7.14).
D(E )
nd n IM EF
Total density of states

Valence band
P 0 states Conduction band
P−state
Ev E
Fig. P7.12 Electronic band structure for the case nd nIM

D(E ) Total density of states

nd n IM
EF
Ed EG Ec E P− E
Fig. P7.13 Electronic band structure for a P concentration just below that of the metal–insulator
transition
D(E )
nd > n IM
Ed EF Ec E P− E
Fig. P7.14 Electronic band structure for the case nd > nIM
3. The experiment identifies metallic behaviour for nd ≥ 3.75 × 1024 /m3 (ρ is con-
stant when T → 0) and insulating behaviour for nd ≤ 3.70 × 1024 /m3 (ρ diverges
when T → 0). Note that measurements have to be made at very low temperatures!
Taking nIM ∼ 3.74 × 1024 /m3 , we obtain
ε 11.75
0.53 × 10−10 = 0.29 .
1/3
C = nIM a1 = (3.74 × 1024 )1/3
me /m0 0.33
This is close to the calculated value C ∼ 0.20–0.22.

4. The electron density in the metallic band is nd . The conductivity is
nd q2 τ
σ= = ρ −1 .
me
For nd = 7 × 1024 /m3 , we have ρ 5 × 10−5 Ω m, so
me 0.33 × 9.1 × 10−31

τ= = ∼ 3.3 × 10−14 s .
nd q ρ
2 7 × 10 × (1.6 × 10−19 )2 × 5 × 10−5
24
438 13 Problem Set
ρ increases slightly with increasing temperature, as would be expected in a metal.

This happens because 1/τ is a sum of two contributions:

1 1 1
= + .
τ τ defects τ vibrations
In a pure metal, the second term dominates the first at high temperatures, but
here the donor atoms in disordered positions constitute scattering centers for the
electrons, and the first term dominates the second across almost the whole tem-
perature range, which explains the almost negligible dependence of τ −1 on T. At
T = 0 the collision time is comparable with τ ∼ 10−14 s obtained in a pure metal
like aluminium at T 300 K.
5. In the insulating case, the resistivity is due to thermally excited carriers in the
conduction band (or the P− band). For nd nIM , the number ne of such carriers
is given by
ne ∝ exp(−EG /2kB T) , with EG = Ec − Ed = E0 .
When nd approaches nIM , the distance EG between the filled band and the empty
band decreases (see Fig. P7.15). The resistivity is given by ρ = (ne rμ)−1 , and
since μ = eτ/me is practically independent of temperature, the variation of ρ is
dominated by that of ne :
ρ ∝ exp(EG /2kB T) .
In Fig. P7.1, for nd = 3.7 × 1024 /m3 , ρ does indeed diverge exponentially as T
decreases. We have log ρ = EG /2kB T + constant.
If log ρ is plotted as a function of T −1 for 2K < T < 10 K, we therefore obtain a
straight line, as in Fig. P7.16. For ρ1 /ρ2 = 10, we obtain T1−1 − T2−1 = 0.33 K−1 .
This implies that EG /kB = 2 log 10/0.33 = 14 K.
Ed EF Ec
EG
E
Fig. P7.15 Insulating case. Ed Ec
As nd approaches nIM , the
gap EG decreases EG
10
Si–P
3.75
1 1
ρ (Ω cm)
10 −1 10 −1
3.84
10 −2 10 −2
7.0
10−2 10−1 1 2 510 102 0 0.2 0.4 0.6

T (K) 1/T (K−1)
Fig. P7.16 Determination of EG from log ρ = EG /2kB T + const.
This should be compared with E0 = (E1 /ε 2 )(me /m0 ), where E1 = 13.6 eV is the
energy of the Bohr atom. This gives E0 /kB = 400 K. We observe that EG E0 ,
since the concentration considered is very close to the insulator–metal transition.
6. EG must decrease from the value E0 when nd is very low, and vanish for nd ∼ nIM
(see Fig. P7.17).
7. In the limit nd nIM , the electrons are localised on phosphorus sites. This is a
classic case of a paramagnetic solid. For electron spins S = 1/2, Curie paramag-
netism is obtained with
μ2B
χC = nd .
kB T
The susceptibility grows linearly with nd and is inversely proportional to T. This

does seem to be the case for nd < 1024 /m3 . The susceptibility is given per kg of
silicon. In this case, for example, for nd = 1023 atoms/m3 , we find
1023 (0.927 × 10−23 )2 2.67 −4

χC = −23
= 10 SI .
2.33 × 10 1.38 × 10 T
3 T
We can now plot the three straight lines in Fig. P7.18 for χC (1.1 K), χC (4.2 K),
and χC (77 K).
EG
E0
Fig. P7.17 EG as a function nd

of nd n IM
440 13 Problem Set
10 −2
1.1 K
χC (1.1 K) 4.2 K
77 K
10 −3
χ (J/tesla 2 kg)
1/3
χP ∝ nd
χ C (4.2 K)
10 −4
χ C (77 K)
10 −5
10 23 10 24 n IM 10 25 10 26
nd (m−3 )
Fig. P7.18 Magnetic susceptibility. Fit of low concentration data with a Curie susceptibility χC
and high concentration data with a Pauli susceptibility χP (see text)
8. For nd nIM , we have a metallic state. In this case, the susceptibility is a

temperature-independent Pauli susceptibility. This is indeed observed for nd >
1025 /m3 . The susceptibility is determined by the electronic density of states at
the Fermi level:
χP = 2μ0 μ2B DΩ (EF ) .
If for high enough nd the metallic band can be treated as a simple parabolic band,
then taking the zero energy at the bottom of this band, we can use (3.47) and
(3.49) obtained in Chap. 3 for a free electron gas, replacing m0 by the effective
mass me :
3/2
1 2me 1/2 h̄2 2 2/3
DΩ (E) = EF , EF = 3π nd .
4π 2 h̄2 2me
This yields
1/3
3 me 1/3
χP = μ0 μ2B nd .
π4 h̄2
We thus find a susceptibility that approaches this behaviour asymptotically for

nd > 1025 /m3 (see Fig. P7.18).
Note: The insulator–metal transition observed in Si–P cannot be explained as simply as sug-
gested here. The effects of disorder must be included along with the interactions between elec-
trons. A great deal of experimental and theoretical work is currently under way to elucidate the
respective roles of these two effects in various insulator–metal transitions.
Problem 8: Cyclotron Resonance 441
Problem 8: Cyclotron Resonance5
The aim here is to investigate the frequency dependence of the conductivity of a

conducting material in the presence of an applied magnetic field B = μ0 H. For a
given value of B, the electrons can engage in an oscillatory motion with a character-
istic frequency. This is the cyclotron resonance mode. The energy associated with
a small alternating applied electric field can be absorbed by the electrons if its fre-
quency is tuned to the frequency of the cyclotron mode. In experimental setups, one
generally observes the increased absorption at fixed frequency when the magnitude
of the magnetic field is modified.
8.1: Real and Reciprocal Space Paths of an Electron State
Consider to begin with a free electron subjected to a magnetic field B. The plane
wave packet representing the electron has momentum p and group velocity vg =
h̄k/m0 , and the Lorentz force acting on the electron is
F = −evg ∧ B .
1. What are the paths of the electron in real space and its wave vector k in recipro-
cal space? What is the period of the motion? Does it depend on the initial value
of k?
Now consider an electron in a solid. It is represented by a Bloch wave packet of

quasi-momenta centered on k and of energy En (k). In the presence of a magnetic
field, the electron is subjected to a force
F = −e vn,k ∧ B ,
where vn,k is the group velocity of the electron in the corresponding Bloch state.
2. Prove that the path C of the point k representing the electronic state in reciprocal
space lies on a surface of constant energy E.
3. Show that its motion is planar and periodic.
4. Show that the total force on the electrons of a fully occupied band is zero.
Consider a Bloch state belonging to constant energy surface in the form of an ellip-
soid of revolution about the z axis, centered on a point k0 of the Brillouin zone:

(kx − k 0x ) 2 (k y − k0y )2 (kz − k0z )2
E = h̄2 + + ,
2m∗⊥ 2m∗⊥ 2m∗
where m∗⊥ and m∗ are the transverse and longitudinal effective masses.
5 This problem has been designed with C. Hermann and B. Sapoval.

442 13 Problem Set
5. Show that the equation of motion of the electrons in real space is

7 7
dv 7 1 7
= −7 7
7 m∗ 7 ev ∧ B ,
dt
where we have set v = vn,k to simplify.
Let θ = (Oz, B) be the angle between the axis of revolution of the ellipsoid and the
magnetic field, which lies in the (x, z) plane.
6. Show that the equation of motion has an oscillating solution with v = v0 eiωt for
ω = ωc = eB/mc ,
where mc does not depend on the initial Bloch state and is a function of θ , m∗⊥ ,
and m∗ , to be determined.
8.2: A Semiconductor: Silicon
Silicon with atomic configuration (Ne) 3s2 3p2 crystallises into a diamond cubic
structure with lattice parameter a, illustrated in Fig. P8.1. It has a face-centered
cubic Bravais lattice.
1. What is the basis of the primitive cell? What is the reciprocal lattice?
The dispersion curves corresponding to electrons in the 3s2 3p6 shells are shown in
Fig. P8.2 in the two directions (100) and (111) of the reciprocal lattice.
2. Count the states in the bands of Fig. P8.2 and indicate the occupied electron
states. Check that silicon is an insulator at zero temperature.
3. On Fig. P8.2, indicate the band gap and specify its value. What are the vectors
k0 and energies corresponding to the conduction electrons and holes excited at
nonzero temperatures? (1 rydberg = 13.6 eV.)
4. Do you think cyclotron resonances can be observed in silicon at all tempera-
tures?
5. How many cyclotron resonance frequencies would one generally expect to ob-
serve for the band structure of silicon given in Fig. P8.2?
y
Fig. P8.1 Crystal structure of a
Si, similar to diamond x
Fig. P8.2 Band structure of −1.0

Si, represented in two high
−1.2
symmetry directions of the
−1.4
Energy (rydberg)
Brillouin zone
−1.6
−1.8
−2.0
−2.2
−2.4
−2.6
−2.8
2π 2π
k= a
1 1 1
2 2 2 k=0 k= a
(100)
[111] [100]
Fig. P8.3 Defining the plane θ B

x Oz in which the field lies z
y y
x
x
We now assume that the field B is applied in a plane x Oz which bisects the unit
cell cube (see Fig. P8.3). For θ = π/6 and a frequency ν = ω/2π = 2.4 × 1010 Hz,
we observe the signal shown in Fig. P8.4 when the magnitude of the magnetic field
is varied. If B is rotated in the plane x Oz, peaks 2 and 3 are found to move, while
peaks 1 and 4 do not.
6. Which electrons are responsible for peaks 2 and 3? Which one of them splits into
two if the field is shifted outside the plane x Oz?
7. Which electrons are responsible for peaks 1 and 4?
Fig. P8.4 Absorption as a 10

function of the field applied 3 ν = 2.4 × 10 Hz
in the direction θ = π/6 T = 4.2 K
Absorption
in the plane x Oz. From 2

Dresselhaus, G., Kip, A.F., 4
Kittel, C.: Phys. Rev. 98, 1
368 (1955). With the per-
mission of the American
Physical Society ( c 1955
0 0.1 0.2 0.3 0.4 0.5 0.6
APS). http://link.aps.org/doi/
10.1103/PhysRev.98.368 μ0 H (tesla)
444 13 Problem Set
8. From the experimental results, deduce the effective masses of the electrons
and holes in silicon. (For free electrons, the cyclotron resonance occurs at
B = 0.86 tesla for the given frequency.)
In the above calculation, we neglected collisions. They can be taken into account
by introducing a friction term into the equation of motion, in the presence of the
oscillating electric field.
9. Without calculation, indicate the effect it should have. Is this visible in the ex-
perimental curve? What should one expect to see if the temperature is raised or
lowered in the above experiment?
10. By reconsidering your answer to question 4, can you guess how the experiment
in Fig. P8.4 was carried out?
8.3: Metals
The dominant contribution to the conductivity in a metal comes from electrons at

the Fermi level which can absorb energy from the electric field excitation Eeiωt
associated with the incident electromagnetic wave.
1. Check that, if the Fermi surface is spherical, the cyclotron angular frequency is
the same for all electrons at the Fermi level.
2. For an arbitrary Fermi surface, show that the angular frequency ωc of the motion
is related to the area S of the path (C) in k space by
2π eB dE
ωc = .
h̄2 dS
&
Use the fact that the period T = 2π/ωc of the motion is T = (C) dt.
3. The Fermi surface of copper is shown in Fig. P8.5. What electronic orbits do you
expect to dominate the frequency response of the conductivity when the field is
applied in the direction (111)?
Fig. P8.5 Fermi surface (111)

of copper
Solution
Real and Reciprocal Space Paths of an Electron State
1. We have m0 vg = h̄k, so
dvg
m0= −evg ∧ B .
dt
For B z, vgz = constant and vg⊥ = (vgx , vgy ) satisfies
d2 vg⊥ e2 2
m0 + B vg⊥ = 0 .
dt2 m0
Hence, k undergoes a circular precession with an angular frequency
eB
ω = ωc = ,
m0
which is independent of the initial momentum of the electron. The electron has a
spiral motion in real space with the same angular frequency.
2. We have vn,k = (1/h̄)∇k E and
dk e
= − 2 ∇k E ∧ B .
dt h̄
Therefore, k · B = constant, and since ∇k E is normal to the surface of constant
energy E, the point k moves tangentially to the constant energy surface, following
an orbit (C) in the plane perpendicular to B (see Fig. P8.6).
3. If this section of the constant energy surface is a closed curve, as is generally the
case, the motion is periodic.
4. Electrons in a fully occupied band are such that the average value over the
band of the velocity of the Bloch states is identically zero. The Bloch states
kz
B
dS dk /dt
ky
(C) dk⊥
Fig. P8.6 Orbit C of point k kx

in reciprocal space
446 13 Problem Set
corresponding to states k on the same orbit (C) have representative states that
rotate on this orbit. This means that all the states of the Brillouin zone are con-
served in the motion induced by the field. The average velocity of the electrons
thus remains identically zero.
5. With

(kx − k0x )2 (ky − k0y )2 (kz − k0z )2
E = h̄2
+ + ,
2m∗⊥ 2m∗⊥ 2m∗
we obtain ⎛ ⎞ ⎛ ⎞
vx h̄(kx − k0x )/m∗⊥
⎜ ⎟ ⎜ h̄(k − k )/m∗ ⎟
⎝ vy ⎠ = ⎝ y 0y ⊥⎠ ,
vz h̄(kz − k0z )/m∗
and hence,
⎛ ⎞ ⎛ 1/m∗ ⎞⎛
h̄dkx /dt
⎞
dvx /dt ⊥ 0 0
⎝ dvy /dt ⎠ = ⎜ ∗ ⎟⎜ ⎟
⎝ 0 1/m⊥ 0 ⎠ ⎝ h̄dky /dt ⎠ ,
dvz /dt 0 0 1/m∗ h̄dkz /dt
so finally,
7 7
dv 7 1 7
= −7 7
7 m∗ 7 ev ∧ B .
dt
6. With v = v0 eiωt , this equation becomes

eB
0 = iωv0x + cos θ v0y ,
m∗⊥
eB eB
0=− ∗ cos θ v0x + iωv0y + ∗ sin θ v0z ,
m⊥ m⊥
eB
0=− sin θ v0y + iωv0z .
m∗
There are solutions other than v0 = 0 for

2 2 cos θ sin2 θ
2
ω =e B
2
+ ∗ ∗ .
m∗2
⊥
m⊥ m
The cyclotron frequency is thus

eB 1 cos2 θ sin2 θ
ω= , with = + .
mc m2c m∗2
⊥
m∗⊥ m∗
It is independent of initial conditions and is the same for all orbits perpendicular
to B. In fact it depends only on the orientation of the magnetic field with respect
to the axis of revolution of the ellipsoid.
A Semiconductor: Silicon
1. The basis of the rhombohedral primitive cell comprises two silicon atoms [see
Fig. P8.7 (left)]. The reciprocal lattice is body-centered cubic. The first Brillouin
zone is shown in Fig. P8.7 (right).
2. There are two silicon atoms (3s2 3p2 ) per primitive cell. For bands arising from
the 3s2 3p6 levels, this therefore represents 8 electrons per primitive cell. These
will thus fill the states of the first four bands in the first Brillouin zone. The
occupied levels are shown in Fig. P8.8. Higher bands are empty and silicon is an
insulator at zero temperature.
3. The band gap is shown in Fig. P8.8. It is a little less than 0.1 rydberg across,
i.e., about 1 eV. At T > 0, electrons will be excited to the bottom of the con-
duction band, which corresponds to vectors k0 ≈ 0.8(2π/a, 0, 0) and −k0 . With
those corresponding to the equivalent directions ky and kz , there are therefore six
equivalent electronic pockets [see Fig. P8.7 (right)]. The states freed at the top of
the valence band are close to k = 0 and correspond to hole pockets in the three
last valence bands, of which two are degenerate.
4. To observe the cyclotron resonances, a significant number of electrons and holes
must be excited. In principle, this means that the experiment must be carried out
at a high enough temperature to populate electron and hole pockets.
5. For each of these electron and hole pockets, the first term in the Taylor expansion
of the energy has the general form
B
kz
z θ1
θ2
θ2
kx
y ky
a 4 π /a
x
Fig. P8.7 Si real space and reciprocal lattice primitive cell within a basis comprising two Si atoms
(full spheres) (left), reciprocal lattice and first Brillouin zone, showing the six thermally populated
ellipsoidal electronic pockets (right)
448 13 Problem Set
Electrons
−1.0
−1.2
Gap
−1.4
Energy (rydberg) −1.6
−1.8 Holes
−2.0
−2.2
−2.4
−2.6
−2.8
k = 2aπ 1 1 1
2 2 2 k=0 k= 2π
a
(100)
[111] [100]
Fig. P8.8 Band structure showing the band gap

2 (kx − k0x )2 (ky − k0y )2 (kz − k0z )2
E = h̄ + + .
2m∗x 2m∗y 2m∗z
For the holes, k0x = k0y = k0z = 0. Given the symmetries of the cubic crystal,
the x, y, and z axes are equivalent, and the constant energy surfaces are therefore
spheres with different effective masses m∗h1 and m∗h2 , as can be seen on Fig. P8.8.
For the electrons, we have for example k0x = k0y = 0, and k0z ≈ 0.8 × 2π/a. The
symmetry of the cubic cell under rotation through π/2 about the z axis implies
equivalence of the x and y axes, and hence m∗x = m∗y . The constant energy surfaces
are thus six paraboloids of revolution about the kx , ky , and kz axes, characterised
by two longitudinal and transverse effective masses, viz., m∗ and m∗⊥ = m∗x = m∗y ,
respectively [see Fig. P8.7 (right)].
A cyclotron frequency is expected for each of the two pockets of holes and for
each pocket of conduction electrons. Since ω depends only on the orientation
of the magnetic field with respect to the axis of revolution of the ellipsoid, the
paraboloids k0 and −k0 correspond to the same cyclotron frequencies. There
will thus be three cyclotron frequencies for the conduction electrons for an arbi-
trary direction of the magnetic field. We therefore expect a total of five cyclotron
frequencies for an arbitrary orientation of the field.
6. Four cyclotron frequencies can be seen in Fig. P8.9. Peaks 2 and 3 are shifted if
the field orientation is altered. They cannot therefore correspond to the holes, for
which the rotational symmetry implies a cyclotron frequency that is independent
of θ . They thus correspond to the electron ellipsoids. With the chosen orienta-
tion of B in the plane bisecting xOy, the angles θ between the field B and the
x and y axes are equal and the corresponding cyclotron frequencies are equal.
Resonance 3, the most intense, therefore corresponds to the four paraboloids
with axes the x and y axes. Cyclotron resonance 2 corresponds to the two z axis
Fig. P8.9 Assignment of the

four resonance frequencies
(see text)
3 ν = 2.4 × 10 10 Hz
T = 4.2 K
Absorption
2
4
1
0 0.1 0.2 0.3 0.4 0.5 0.6

μ0 H (tesla)
paraboloids. If the field is rotated out of the plane bisecting xOy, resonance 3
splits into two distinct resonances, because the x and y axis paraboloids are now
distinguished.
7. Resonances 1 and 4 correspond to the holes in the two valence bands. Resonance
4 corresponds to the highest resonance field and thus also the highest effective
mass. From the bending of the energy bands shown in Fig. P8.8, it is therefore
the flattest doubly degenerate hole band with the highest energy, and therefore
the most intense signal.
8. Resonance 1 occurs for B 0.13 tesla, with ω/B = e/m∗h1 . Since ω/0.86 T =
e/m0 , it follows that
m∗h1 0.15m0 .
Resonance 4 occurs for B 0.44 tesla, and corresponds to
m∗h2 0.5m0 .
Resonance 2 occurs for B 0.18 tesla, so mc1 0.20m0 . It corresponds to θ1 =

π/6, and
1 3/4 1/4
= + ∗ ∗ .
m2c1 m∗2
⊥
m⊥ m
Resonance 3 occurs for B 0.29 tesla, so mc2 0.34m0 . It corresponds to θ2 ,
shown on Fig. P8.7 (right), and

1 1/8 7/8
= + ∗ ∗ .
m2c2 m∗2
⊥
m⊥ m
The above equalities imply that
m∗⊥ 0.18m0 , m∗ 1.1 m0 .

450 13 Problem Set
9. A cyclotron periodic motion can only be well defined if electrons undergo sev-
eral cyclotron precessions without suffering collisions. The resonance will not be
observed if the collision time τ is shorter than the cyclotron period 2π/ωc . Col-
lisions will broaden the resonance by an amount that depends on 1/τ , and which
will be greater for higher values of 1/ωc τ . Under the given experimental condi-
tions, ωc is fixed and it is the field B that varies. The widths at half maximum B
of the resonances increase with Bc , and it can be seen that B/Bc 0.2. This
therefore suggests that the collision time determines these widths with ωc τ ≈ 10.
If the temperature is lowered, the collision time will increase and the cyclotron
resonances should narrow. On the other hand, raising the temperature should
broaden the cyclotron resonances.
Note: In principle, collision times differ for electrons and the various hole pockets.
Indeed, the values of B/Bc are 0.12 and 0.30 for the holes (1 and 4, respec-
tively). For resonances 2 and 3, which correspond to electrons, the collision times
should be the same. It can be checked that in this case B/Bc has the same value
0.15 for the two resonances.
10. We observe two contrasting effects of temperature. We would like to increase the
number of excited carriers by raising the temperature, but then the resonances
broaden and can no longer be detected. But if we lower the temperature to nar-
row the resonances, the number of thermally excited carriers is reduced. Since
the number of carriers goes as exp(−Eg /kB T), where Eg is the band gap energy,
we lose far more carriers when T is decreased than we gain in spectral reso-
lution, so there is no particular advantage in lowering the temperature. In fact,
with Eg 1 eV, the number of thermally excited carriers in silicon at T = 4.2 K
is too low to be able to detect cyclotron resonances. The experiment shown in
Fig. P8.9 was thus carried out by illuminating the sample with a laser in which
the photons have an energy slightly greater than Eg . In this way, electron–hole
pairs can be excited optically and enough free carriers are created to be able to
observe the cyclotron resonances with all the benefits of a long collision time at
low temperatures.
Metals
1. For a metal with spherical Fermi surface, question 6 of 8.1 implies that the cy-
clotron frequency does not depend on the angle θ . Therefore, as for holes in
silicon, all electrons at the Fermi level correspond to the same cyclotron fre-
quency, with effective mass determined by the curvature of E(k) at the Fermi
level.
2. By question 2 of 8.1,
dk e
= − 2 ∇k E ∧ B .
dt h̄
The vector dk = dk is tangential to (C) in the plane perpendicular to B (see

Fig. P8.6). The period of the motion is given by
% %
dk h̄2
T= dt = .
(C) (C) |∇k E| eB
Now, dE = |∇k E| dk⊥ , so

%
h̄2 1 h̄2 dS
T= dk⊥ dk = .
eB dE (C) eB dE
and finally,
2π eB dE
ωc = .
h̄2 dS
3. The Fermi surface of copper is almost spherical, but the contacts of the Fermi sur-
face with the first Brillouin zone correspond to parts of the Fermi surface which
deviate markedly from sphericity. When the field is applied parallel to one of the
cube diagonals, some cyclotron orbits at the Fermi level are very different from
orbits corresponding to a sphere. Note that, if we consider a cross-section of the
Fermi surface parallel to the (111) axis, we find two regions in which the density
of cyclotron orbits with similar characteristics is very high. They correspond to
orbits for which the Fermi surface can be approximated by a tangential cylinder
with axis parallel to B. Figure P8.10 shows orbits at the bellies (B) and at the
necks (N) of the Fermi surface, which lead to the two families of orbits, called
extremal orbits (belly orbits and neck orbits), which dominate the cyclotron res-
onance spectrum. The belly orbits (B) are not significantly modified if B moves
away from the (111) axis, while the neck orbits (N) only exist for B (111).
(N)
(B)
(N)
Fig. P8.10 Belly and neck or-

bits on the Fermi surface of (111)
copper
Problem 9: Phonons in Solids 453
Problem 9: Phonons in Solids6
The aim in this problem is to describe the atomic vibrational modes in a crystal,
along with their influence on the thermodynamic properties and resistivity of metals.
9.1: Einstein Model
We begin by considering the simplest quantum model that can account for atomic
vibrations in a crystal containing N atoms (or ions) with equilibrium positions at
the lattice points. This model due to Einstein treats the small-amplitude oscillations
of the crystal as harmonic vibrations of its atoms about their equilibrium positions.
These atomic vibrations, with angular frequency ω, are assumed to be mutually
independent. The Hamiltonian for the ensemble of atoms is then a sum of Hamilto-
nians of independent 3D harmonic oscillators.
Recall the quantisation of the harmonic oscillator energies:

1
En = n + h̄ω ,
2
where n = 0, 1, 2, 3, . . . . An energy quantum h̄ω in the system is called a phonon.

Thermal excitation of vibrations in the solid thus corresponds to an increase in the
number of phonons with temperature.
1. Write down the partition function for this ensemble of harmonic oscillators and
deduce the temperature dependence of the internal energy.
2. Determine the contribution of these vibrational modes to the specific heat of the
crystal, and check that, at high temperatures, it does indeed correspond to the
classically expected Dulong–Petit law.
In non-magnetic insulating solids, the specific heat measured at low temperatures

varies as a power law in T, usually T 3 .
3. Explain why the specific heat of electronic origins is negligible in such solids.
Show that the Einstein model does not apply for low temperatures.
9.2: Debye Model
A weak point of the Einstein model is that it treats the atomic motions as completely
uncorrelated. In reality, whenever an atom moves away from its equilibrium posi-
tion, this leads to forces on its neighbours. Atomic motions are therefore coupled.
6 This problem has been designed with S. Biermann and G. Montambaux.

454 13 Problem Set
The Debye model is the simplest one taking this fact into account. We begin by
considering a linear chain of N atoms, and vibrations that can occur along the chain.
We assume that, between two neighbouring atoms of mass M a distance a apart,
there is an elastic force fn,n+1 = −K(un − un+1 ), where un a is the displacement
of atom n from its equilibrium position. The Hamiltonian for this system is thus

p2 1
n
H = + Mω0 (un+1 − un ) ,
2 2
n
2M 2
√
where the eigenfrequency ω0 is related to the stiffness constant K by ω0 = K/M.
a. Eigenmodes
The motion of an atom is thus coupled with that of its neighbours. We start by deter-
mining the eigenfrequencies of this chain of atoms. We impose periodic boundary
conditions, identifying atom N + 1 with the first atom of the chain.
4. Show that, if the motions are decomposed into modes,
1
un (t) = √ ũq (t)eiqna ,
N q
1
pn (t) = √ p̃q (t)eiqna ,
N q
the total Hamiltonian can be written as a sum of independent harmonic oscillator

terms, viz.,

|p̃q |2 1
H = + Mωq |ũq | ,
2 2
q
2M 2
where the frequencies ωq of these new oscillators are given by

qa

ωq = 2ω0 sin .
2
This relation can be used to determine the speed of sound c in the solid in the long
wavelength limit ωq = cq.
5. Why is the speed of sound greater in a solid than in a gas?
b. Quantisation: Phonons
This Hamiltonian thus describes an ensemble of independent harmonic oscillators

corresponding to the various eigenmodes. The energy of each mode involves a
different quantum number nq through the relation
Eq = (nq + 1/2)h̄ωq .
6. What is the average energy of a harmonic oscillator at finite temperature? Show

that the average energy of the atomic chain is

1 1
E(T) = nq + h̄ωq , nq = β h̄ω .
q
2 e q −1
Deduce by analogy with photons that each classical mode can be represented
quantum mechanically by bosonic excitations, viz., phonons.
c. Linear Diatomic Chain
We now consider more realistic crystalline solids in which the primitive cell contains
several atoms. A simple illustration of the resulting changes in the eigenmodes can
be obtained by considering a linear chain comprising two types of atom with masses
M1 and M2 , a distance a apart. To simplify, we assume that nearest neighbours are
always coupled via the same stiffness constant K.
7. Write down the equations of motion in a semiclassical approximation and find
the eigenfrequencies.
8. Show that, in the limit q → 0, these eigenfrequencies become

1 1 2K
ωO
2
= 2K + , ωA
2
= q2 a2 .
M1 M2 M1 + M2
9. Represent the eigenfrequency spectrum in the first Brillouin zone. Why do you
think the two branches are called the acoustic branch and the optical branch?
What happens if M1 → M2 ?
9.3: Experimental Detection of Phonons
Phonon modes can be detected by inelastic neutron or X-ray scattering, in which

the incident particle absorbs or emits a phonon (see Chap. 12). Energy and momen-
tum conservation are used to determine the dispersion curves ω(q) of the phonons
involved in the scattering process. Figure P9.1 shows just such an experimental de-
termination for graphite. This is a hexagonal crystal obtained by superposing 2D
graphene sheets in which the atoms are arranged in a honeycomb pattern, with two
carbon atoms per unit cell. The dispersion curves ω(q) are represented in the direc-
tions of highest symmetry of the first Brillouin zone for this structure.
456 13 Problem Set
h̄ ωq (meV)
c∗
150
100 Γ
M
K
50
0
Γ K M Γ
Fig. P9.1 Phonon modes detected in graphite by inelastic X-ray scattering, plotted in directions
Γ M, MK, and Γ K of the hexagonal first Brillouin zone of graphite shown on the right. Figure
courtesy of M. Mohr, from Mohr, M., et al.: Phys. Rev. B 76, 035439 (2007)
Note that the atomic vibrations are not generally restricted to longitudinal vibra-
tions, and the atoms also vibrate in directions transverse to the wave propagation
vector.
10. On Fig. P9.1, indicate the ‘acoustic’ and ‘optical’ modes. Can you explain the
characteristics of the modes observed experimentally?
9.4: Thermodynamic Properties
Given dispersion relations ω(q) like the ones in Fig. P9.1, describing the phonon
eigenmodes, we can determine the density of eigenmodes g(ω), i.e., the number
g(ω)dω of oscillators with frequencies between ω and ω + dω. Their effects on ther-
modynamic properties can thus be found in a completely general way.
11. Show that the specific heat is given by
2
∞ β h̄ω
C = kB g(ω)dω .
0 2 sinh β h̄ω/2
12. Deduce that, at high temperatures, the specific heat is given by the Dulong–Petit
law, whatever the frequency distribution g(ω).
In contrast, the low temperature behaviour of the specific heat depends on the form
of the function g(ω) at low frequencies. The Debye model takes the phonon spec-
trum to be a straight line ω(q) = c |q|, and bounds it by the Debye frequency ωD
such that the total number of modes is 3N.
13. Explain why this model is realistic for low temperature properties. Determine
g(ω) and the Debye temperature θD = h̄ωD /kB .
14. Show that this Debye model explains the observed behaviour C ∝ T 3 . What
would be the T dependence in a d-dimensional space?
9.5: Resistivity
The aim here is to study the temperature dependence of the resistivity due to
phonons in pure metals.
15. Show that, at high temperatures, the number of phonons in a mode q is propor-
tional to T. Why should the resistivity be proportional to this number?
At low temperatures, the almost universal temperature dependence ρ(T) of the re-
sistivity shown in Fig. 4.8 goes as T 5 (Bloch–Grüneisen law).
16. Use conservation of quasi-momentum to show that only phonons with energies
close to kB T can be absorbed or emitted during an electron collision. What
can you deduce about the number of phonons that can take part in a collision
process?
In order to justify the observed low temperature behaviour, we must also understand
why the different collision processes suffered by the electrons do not equivalently
limit the current in the direction of the applied electric field.
17. Show that this leads to an extra factor 1 − cos θ in the scattering probability,
where θ is the scattering angle. Does this produce a T 5 dependence in the resis-
tivity? Would another temperature-dependent factor be conceivable?
458 13 Problem Set
Solution
Einstein Model
1. With the quantisation of the levels

1
En = n + h̄ω , n = 0, 1, 2, . . . , (13.12)
2
the canonical partition function ZN is given as a function of temperature by
∞
3N 3N 3N
e−β h̄ω/2 eβ h̄ω/2
Z= e−βEn = = . (13.13)
1 − e−β h̄ω eβ h̄ω − 1
n=0
It follows that the energy is

∂ log Z h̄ω h̄ω
U=− = 3N + β h̄ω . (13.14)
∂β 2 e −1
2. The specific heat is therefore
2
∂U β h̄ω
C= = 3NkB . (13.15)
∂T 2 sinh(β h̄ω/2)
3. In non-magnetic insulating solids, the valence and conduction bands are sepa-
rated by a band gap Eg generally greater than 1 eV. In the ground state at T = 0,
the conduction band is empty and the valence band is fully occupied. When T in-
creases, a small number of electrons are excited into the conduction band. There
is indeed an increase in internal energy of electronic origins, but it is very small
because it is determined by the occupation factor, which goes as exp(−Eg /kB T).
When T → 0, this same exponential factor will thus also dominate the behaviour
of the specific heat. The latter is therefore low, even for semiconductors.
Likewise for the Einstein model, the temperature dependence of the specific heat
per atom obtained in (13.15) is dominated at low T by the factor exp(−h̄ω/kB T).
Once again this arises because of the energy gap h̄ω between the ground state
and excited states. The Einstein model cannot therefore explain the dependence
C ∝ T 3 found experimentally at low T in non-magnetic insulating solids.
Debye Model
a. Eigenmodes
4. The Hamiltonian is thus

p2n 1
H = + Mω02 (un+1 − un )2 . (13.16)
n
2M 2
The motions are first decomposed into ‘modes’:
1 1
un (t) = √ ũq (t)eiqna , pn (t) = √ p̃q (t)eiqna . (13.17)
N q N q
As for the electronic band structure, periodic boundary conditions lead to quan-
tisation q = 2π m/Na of the values of q, and the translational symmetry of the
crystal lattice allows us to restrict the representation of the modes to values of
q belonging to a primitive cell of the reciprocal lattice, here the first Brillouin
zone qa ∈ (−π , π ]. The number of values of q and hence the number of modes
is precisely the number N of atoms.
To rewrite the Hamiltonian, note that
1
p2n = p̃q p̃q eina(q+q ) = |p̃q |2 , (13.18)
n
N q
n,q,q
because, for the quantised values of q given above, we have

N
N

eiqna = 0 , for q = 0, and then ei(q+q )na = Nδq,−q .
n=1 n=1
Likewise,
1 iqna iqa
un+1 − un = √ uq e (e − 1) , (13.19)
N q
1
(un+1 − un )2 = uq uq (eiqa − 1)(eiq a − 1)eina(q+q ) (13.20)
n
N n,q,q
qa
= 4 sin2 |uq |2 . (13.21)
q
2
The Hamiltonian can then be rewritten as a sum of independent harmonic oscil-

lator terms:
460 13 Problem Set

|p̃q |2 1 qa

H = + Mωq |ũq | ,
2 2
with ωq = 2ω0 sin . (13.22)
q
2M 2 2
This dispersion relation is symmetrical with respect to q = 0. The modes for

q < 0 and q > 0 represent identical modes propagating in opposite directions. The
dispersion relation can thus be represented in the half-Brillouin zone (0, π/a), as
shown on the right of Fig. P9.2.
5. In the long wavelength limit, (13.22) is linear:

ωq = cq , c = ω0 a = a K/M . (13.23)
Long wavelength vibrations correspond to sound waves in the solid, with c the
speed of sound. According to (13.23), the speed of sound is determined by the
interatomic distance a and the coupling constant K between atoms. Even though
interatomic distances are greater in gases than in solids, the coupling constants
K are much weaker in gases, so sound propagates much faster in solids than in
gases, e.g., 1,300 m/s in lead, 3,000 m/s in copper, and 5,100 m/s in aluminium.
In a gas like nitrogen, at room temperature and pressure, it is a mere 350 m/s.
b. Quantisation: Phonons
6. The Hamiltonian (13.22) thus describes an ensemble of quantised harmonic os-

cillators. It has eigenvalues
1

E= nq + h̄ωq , (13.24)
q
2
ωq
Optical
mode
ωO Acoustic 2ω 0
modes
√
ω0 2
π π π
− − 0
a 2a a
Fig. P9.2 Phonon dispersion relation for a monatomic linear chain (right) and for a diatomic linear
chain (left)
where nq = 0, 1, 2, . . .. As we saw in (13.14), the energy of a harmonic oscillator

at finite temperature is
h̄ω h̄ω
U= + β h̄ω . (13.25)
2 e −1
The energy of the chain at finite temperature is then

1 1
E(T) = nq + h̄ωq , nq = . (13.26)
q
2 eβ h̄ωq −1
We see therefore that the vibrational excitations of the chain involve N inde-
pendent modes. The vibrational energy in each mode is quantised. One mode of
quasi-momentum q involves nq phonons of energy h̄ωq . The number of phonons
excited in a mode represents the energy accumulated in this mode, which is clas-
sically directly related to the amplitude of the vibration by

Eq = Mωq2 u2q . (13.27)
The analogy with photons is clear. We recover the duality between wave and par-
ticle: photons are particles corresponding to excited electromagnetic modes in
the vacuum. In the same way, the energy quanta called phonons can be thought
of as quasi-particles associated with the wavelike displacement modes excited in
a solid. This analogy also extends to magnons, which are magnetic excitations
associated with the wavelike modes of spin waves in materials exhibiting mag-
netic order at T = 0. The numbers of photons, phonons, or magnons thermally
excited are unbounded and obey Bose–Einstein statistics. These excitations can
thus legitimately be considered as bosons, the total number of which is not con-
served, i.e., the chemical potential is zero.
The creation and annihilation operator formalism developed for photons can be
generalised to phonons. Indeed, it is commonly used in textbooks on quantum
mechanics or statistical physics to describe the harmonic oscillator, and was ap-
plied in Chap. 12 to the case of magnon modes. We shall see later that, during an
interaction between an incident particle and a solid, the particle can be scattered
by absorbing or emitting a phonon. In this process, energy and momentum, or
rather quasi-momentum, are conserved, up to addition of the reciprocal lattice
constant (a reciprocal lattice vector in the 3D case).
c. Diatomic Linear Chain
7. In this case, atoms of masses M1 and M2 are located at 2na and (2n + 1)a, respec-
tively. If u2n and u2n+1 denote their respective displacements, the Hamiltonian is

p2 p22n+1 1
H = 2n
+ + K(u2n+1 − u2n ) .2
(13.28)
n
2M1 2M2 2
462 13 Problem Set
This leads to equations of motion that couple the motions of the atoms of different
masses:
d2 u2n
M1 = −K(2u2n − u2n−1 − u2n+1 ) (13.29)
dt2
d2 u2n+1
M2 = −K(2u2n+1 − u2n − u2n+2 ) . (13.30)
dt2
Here we look for solutions for harmonic modes of wave vector q, defining dis-
tinct vibrational amplitudes uq and vq for the atoms of masses M1 and M2 , re-
spectively:

u2n (t) = uq ei(2qna+ωq t) , u2n+1 (t) = vq ei q(2n+1)a+ωq t . (13.31)
We obtain the two equations

− M1 ωq2 uq = −K 2uq − vq (eiqa + e−iqa ) , (13.32)

−M2 ωq2 vq = −K 2vq − uq (eiqa + e−iqa ) , (13.33)
or

uq M1 ωq2 − 2K + vq (2K cos qa) = 0 , (13.34)

uq (2K cos qa) + vq M2 ωq2 − 2K = 0 . (13.35)
There are only non-trivial solutions for uq and vq if the determinant of the system
is zero, i.e., if

1 1 4K 2
ωq4 − 2K + ωq2 + sin2 qa = 0 . (13.36)
M1 M2 M1 M2
This second order equation in ωq2 has solutions
1/2
1 1 1 1 2 4
ωq2 =K + ±K + − 2
sin qa . (13.37)
M1 M2 M1 M2 M1 M2
8. For q → 0, the modes are thus given by

1 1 2K
ωO
2
= 2K + , ωA
2
= q2 a2 . (13.38)
M1 M2 M1 + M2
9. There are two values of ωq for each q, i.e., two branches on the dispersion curves.
As for the monatomic chain, the values of ωq are the same for q > 0 and q < 0, so
the phonon dispersion relations can be represented in half of the Brillouin zone.
Here the primitive cell of the chain has dimension 2a, and the dispersion curves
are represented in (−π/2a, 0), on the left of Fig. P9.2.
One branch of the dispersion curves vanishes for q → 0 and has the same
q dependence as for the monatomic chain. Since this solution corresponds to
uq /vq → 1 for q → 0, the two types of atom vibrate in phase. For q → 0, there
is therefore a long wavelength rigid displacement of the crystal structure. This
corresponds to what is expected in mechanics for an acoustic deformation mode
of the medium. It is therefore called the acoustic branch, as for the monatomic
chain.
Regarding the other branch, we find that uq /vq → −M2 /M1 for q → 0 and the
two types of atom then vibrate completely out of phase. In ionic crystals like
NaCl, where the two types of atom carry opposite charges, this vibrational mode
can be excited by coupling with an electromagnetic wave, since its electric field
will exert opposite forces on the different atoms. As this phonon mode can be
excited optically, phonon branches for which the energy does not vanish when
q = 0 are traditionally known as optical branches, even when the atoms carry
charges of the same sign.
Note that, when M1 = M2 , the chain that previously had a primitive cell of length
2a becomes once again a monatomic chain of lattice constant a. Its Brillouin
zone is then (−π/a, π/a), and the two branches
2K
ωq2 = (1 ± cos qa)
M1
are degenerate for q = π/2a, where they correspond to solutions uq → ±vq . They
do indeed correspond to the solution ωq = (4K/M1 )1/2 sin |qa/2| obtained for the
monatomic chain, which is folded in the half-Brillouin zone.
Experimental Detection of Phonons
10. In graphite, composed purely of carbon, all the atoms have the same atomic
mass. However, there are two carbon atoms per unit cell, with non-identical en-
vironments. Whether the atomic sites are distinguished by a difference in their
mass, their position in the unit cell, or the force constant K, the result is in
all cases analogous to the one obtained above for the diatomic chain. Experi-
mentally, it is found that the phonon dispersion curves determined for graphite
have acoustic and optical branches (see Fig. P9.3). Since we are considering
3D vibration modes, there are three non-degenerate accoustic modes and three
non-degenerate optical modes in this case. The structure of graphite is highly
anisotropic. The interatomic distance from plane to plane is 3.35 Å. This is de-
termined by Van der Waals interactions. On the other hand, within a plane, C–C
bonds correspond to a distance of only 1.42 Å. As a consequence, vibration
464 13 Problem Set
200 LO
TO
LO TO
150
LA
LA
ZO
h ωq (meV) 100
ZO
TA
50 TA
ZA ZA
0
Γ K M Γ
Fig. P9.3 Acoustic and optical phonon modes: longitudinal (LA, LO), transverse in the graphite
plane (TA,TO), and transverse to the graphite plane (ZA, ZO). Note the lower energy of the latter
modes transverse to the graphite planes (ZA, ZO) are naturally of much lower
energy than those within a plane.
Phonon mode calculations with the harmonic coupling model can be gener-
alised to the case where the couplings are not limited to nearest neighbours.
In Fig. P9.3, continuous curves are modes calculated taking into account up to
fifth neighbours. The agreement with experiment is remarkable.
These observations are in fact very general. The total expected number of
phonon modes is equal to the number of degrees of freedom of the crystal.
For a crystal with p atoms per unit cell, we expect 3p phonon branches, each
involving N modes for a crystal with N primitive cells. This is illustrated in
Fig. P9.4, which shows an experimental determination of the phonon modes in
the metallic compound Sr2 RuO4 . The Fermi surface of this compound is shown
on the cover of this book. It has the same atomic structure as La2 CuO4 (see
Problem 17: Electronic Properties of La2 CuO4 ). In this case, there are p = 7
atoms per unit cell, hence 21 phonon branches, as observed by inelastic neutron
scattering. Note that the three acoustic modes split away from most of the optical
modes. Once again, harmonic oscillator calculations up to fifteenth neighbours
account exceptionally well for the experimental observations.
Thermodynamic Properties
11. The specific heat is

2 2
β h̄ωq ∞ β h̄ω
C = kB = kB g(ω)dω ,
q
2 sinh β h̄ωq /2 0 2 sinh β h̄ω/2
(13.39)
where g(ω) is the mode density.
24
16
ν (THz)
0
0 0.5 1 0.5 1 0.5 1 0.5 1
Fig. P9.4 Experimental phonon dispersion curves obtained by inelastic neutron scattering on
Sr2 RuO4 . Only shown are the x dependences of the 21 branches in the direction (x00) of the recip-
rocal lattice associated with the body-centered tetragonal structure of this compound. Continuous
curves: Dispersion relations calculated using the harmonic oscillator approximation up to fifteenth
neighbours. Figure courtesy of M. Braden, from Braden, M., Reichardt, W., Sidis, Y., Mao, Z.,
Maeno, Y.: Phys. Rev. B 76, 014505 (2007)
12. At high temperatures, i.e., kB T greater than the energy h̄ω0 of the eigenmode of
highest frequency, the specific heat is given by the Dulong–Petit law, viz.,
C = 3NkB ,
whatever the frequency distribution g(ω). However, the low temperature be-
haviour depends on the low frequency behaviour of the function g(ω).
13. For kB T h̄ω0 , the exponential factor cuts off the integral at kB T, so we
only need to know g(ω) for low frequencies, where the spectrum is linear, i.e.,
ω = c|q|. The number of oscillators g(ω)dω with frequencies in the range from
ω to ω + dω is
3
L 3 L3 ω2
g(ω)dω = 3 4π q2 dq = dω . (13.40)
2π 2 π 2 c3
The Debye model shown schematically in Fig. P9.5 amounts to introducing a

cutoff frequency such that the total number of modes remains equal to 3N :
ωD
g(ω)dω = 3N . (13.41)
0
466 13 Problem Set
Fig. P9.5 Phonon density of g(ω)

states in the Debye model
0 ωD ω
It follows that
1/3 1/3
N
ωD = c 6π 2 ,
V
and the mode density becomes

/
9Nω2 /ωD
3 , for ω < ω ,
D
g(ω) = (13.42)
0 for ω > ωD .
The Debye temperature θD is defined by kB θD = h̄ωD . It takes values between

72 K for mercury and 640 K for silicon.
14. To determine the low temperature behaviour of the specific heat in the Debye
model, we write
2
ωD β h̄ω ω2 kT 3 β h̄ωD /2 x4
C = kB 9N 3 dω = 72NkB dx .
0 2 sinh β h̄ω/2 ωD h̄ωD 0 sinh2 x
(13.43)
As β tends to infinity, the upper bound of the right-hand integral tends to infinity
and the integral is found to have the value π 4 /30. The specific heat then becomes
3
π4 T
C = 12 NkB . (13.44)
5 θD
Hence C goes as T 3 at low temperatures. In a d-dimensional space, it would go

as T d .
Resistivity
15. We saw earlier that the resistivity of a metal is due to electron scattering by
defects perturbing the periodicity of the crystal structure. When there are no
impurities, phonons are the main cause of this deviation from strict periodicity.
Classically, the resistivity is therefore dominated by the scattering cross-section
due to these vibrational modes, itself proportional to the square of the mean
atomic displacement, viz., u2q for the given mode. As we saw in question 6, the
latter is proportional to the number of phonons in this mode.
When kB T > θD , the number of phonons in the mode ωq is given by
1 kB T
nq = ∼ . (13.45)
exp(hωq /kB T) − 1 hωq
It is proportional to T. Since this is true for all phonon modes, the resistivity due
to phonons must indeed be linear in temperature, as observed experimentally,
for example, for copper and its alloys (see Fig. 4.9).
In more detail, the electron scattering processes that dominate the resistivity
involve the scattering of an electron with emission or absorption of a phonon
of energy h̄ω(q). If the initial electronic state has wave vector k and energy εk ,
it can only be scattered into a final unoccupied electronic state of wave vector
k = k ± q. Furthermore, energy conservation implies
h̄ω(q) = ±(εk±q − εk ) .
This restricts the wave vectors of phonons that can scatter the electron k to a
2D surface in the space of wave vectors q. Since the maximal phonon energy,
typically h̄ωD 0.02 eV, is much less than the Fermi energy EF 1 eV, these
conditions restrict available scatterings to phonons of wave vector q connecting
a state k to a state k on the constant energy surface εk , i.e., almost EF , to sat-
isfy the rule that state k is occupied and state k is not. At high temperatures
T > θD , scattering probabilities are proportional to the occupancy nq of these
phonons which are all proportional to T. So we find that the resistivity is indeed
proportional to T.
16. At low temperatures, the temperature dependence is radically different, in fact,
going as T 5 . We shall see that this results from a number of different factors
restricting the processes in which phonons can scatter electrons.
Note that, extending the above argument, it might be thought that the total num-
ber of phonons excited at temperature T would govern the scattering process. In
the Debye approximation, it is given by

1
nq = dωD(ω) ∼ T3 . (13.46)
q
exp(hω/k B T) − 1
However, the above constraints restrict the number of phonons that can take part
in scattering processes. Indeed, for T θD , only phonons with h̄ω(q) < kB T
can be absorbed or emitted by scattering by electrons at the Fermi level. For
absorption, this is obvious, because these are the only modes to be significantly
populated. For emission, it is due to the need to find an unoccupied final state
468 13 Problem Set
into which the electron can be scattered. This state is lower than the initial state
by an amount hωq . Since the unoccupied states of lowest energy are located at
kB T below the Fermi energy, while the occupied states of highest energy are at
kB T above it, only phonons of energy kB T can actually be emitted.
Since ω(q) = cq in this Debye limit, the surface described above containing the
vectors q of phonons involved in scattering has linear dimension ∝ T, and hence
size T 2 . The effective number of phonons available for scattering processes thus
goes as T 2 .
17. It is easy to see that not all collision processes affect the electron drift current
in the same way. Indeed, a scattering process in which an electron in state k is
scattered into a state k that is almost collinear with k is much less efficient than
one in which k goes to k = −k. This introduces a geometrical factor
1 − cos θ ∼ (k − k )2 = q2 ∝ T 2 (13.47)
into the scattering cross-section for electron transport. This therefore leads to a
temperature dependence going as T 4 for the resistivity.
The extra factor of T comes from the fact that, in the above argument, we assumed that the
scattering cross-section for electrons by phonons is independent of q = k − k. In fact, for
small q, the electron–phonon coupling constant has the form [4, Chap. 26]
2
gq h̄ω(q)EF cq .
At low temperatures, this introduces the missing factor of T in the scattering cross-section that
needs to be taken into account when describing electron transport.
A more mathematical discussion of the scattering cross-section is possible, starting with
the Fermi golden rule [2, Chap. 17]. An account can be found in [10, Chap. 9]. It leads to the
result

T 5 TD /T x5
ρ(T) ∼ dx x . (13.48)
ωD θ (e − 1)(1 − e−x )
At low temperatures, the integral is constant and a T 5 dependence is obtained. At high temper-
atures, the integral goes as T −4 and we recover the linear temperature dependence discussed
above. Equation (13.48) is the Bloch–Grüneisen law.
Problem 10: Thermodynamics of a Thin Superconducting Cylinder 469
Problem 10: Thermodynamics of a Thin Superconducting

Cylinder: Little–Parks Experiment
The aim in this problem is to investigate the thermodynamics of a superconducting

cylinder of outer radius R as a function of its thickness d (see Fig. P10.1). We shall
be particularly interested in the temperature dependence of its critical field. The tube
is cooled in zero field at a temperature T < Tc . A small external field B = μ0 H is
then applied parallel to the z axis of the tube.
1. We assume that d is much greater than the London penetration depth λ. Specify
the spatial configuration of the fields and the currents in the superconductor.
The field is now applied in the normal state, and the superconductor cooled in an
applied field H.
2. Does the field or current configuration change, and if so, in what way?
The conductivity of the metal is measured by placing two contacts for the current I
and two for the voltage V along the tube, as shown in Fig. P10.1. The tube is cooled
in zero field until T < Tc and the variation of V is monitored as the applied field H
is increased.
2. Explain how V varies with H.
3. For what type of superconductor and under what experimental conditions could
one thereby determine the condensation energy in the bulk superconductor?
Experiments show that the thermodynamic critical field measured in a bulk sample
obeys the relation 2
T
Hc (T) = Hc (0) 1 −
b
,
Tc
d B = μ0 H
R
V I
Fig. P10.1 Thin superconducting

cylinder
470 13 Problem Set
where μ0 Hc (0) = 0.03 tesla and Tc = 3.73 K for tin. We now consider the case where
d is not infinitely large compared with the penetration depth λ. The tube is cooled in
a much higher applied field than the critical field Hcb (T) of the bulk material, down
to T < Tc , and V is measured for a decreasing field H at fixed T. The transition fields
Hcd (T) measured for a tin tube of thickness d = 2500 Å are plotted in Fig. P10.2.
5. Show that Hcd (T) > Hcb (T), and give a qualitative explanation.
6. Assuming that the radial profile of the field is similar to the one observed for a
thin film of thickness d = 2a, as studied earlier in Chap. 6, find the expected
value of the critical field Hcd (T) in the limit d λ according to the London
phenomenological model.
For the bulk material, the T dependence of λ is well modelled by

4 −1/2
T
λ(T) = λ0 1− ,
Tc
with λ0 = 600 Å for tin.

7. How is Hcd (T) expected to behave near T = Tc ? Keep only the first term of the
expansion in t = 1 − T/Tc .
8. How do the experimental results compare quantitatively with this result in the
London model?
In fact, in the above argument, we have neglected flux quantisation, and also the
microscopic physical origins of superconducting pair condensation. We begin by
considering the first of these features.
9. Show that the superfluid current density Js is quantised and given by

ns h̄q Φ
Js = N − ,
2m∗ R Φ0
where m∗ = 2m0 is the mass of the superconducting carriers, q = 2e is their

charge, Φ is the magnetic flux, Φ0 = h/2e is the flux quantum, and ns is the
density of superconducting electrons.
1
2R = 0.92 μm
μ 0 H (mtesla)
0.5
Fig. P10.2 Fields Hcd (T)

3.70 3.75
measured for a tin tube of
thickness d = 2500 Å T (K)
Fig. P10.3 Variation of V

versus the applied field for 3.72 K
different temperatures. From
Groff, R.P., Parks, R.D.: Phys.
Rev. 176, 567 (1968), with the 3.715 K
10.1103/PhysRev.176.567 3.70 K
3.67 K
3.65 K
In fact, experimentally, one measures the behaviour of V, as shown in Fig. P10.3,

for 2R = 1.35 μm, showing that there are transitions at several field values for a
given value of T. (Figure P10.2 concerns only the transition corresponding to the
weakest field.) This suggests that, in the above arguments, we have neglected some
contribution to the free energy that is sensitive to flux quantisation. We shall express
it in the form E∗ = γ Js2 , where γ is a constant of proportionality.
10. What could such a contribution correspond to physically? What coefficient γ
would you use?
Figure P10.4 shows all transitions obtained as a function of T, for experiments car-
ried out on tin tubes of different radius R but the same thickness d = 2, 500 Å. Curves
joining the experimental points are guides to the eyes.
11. How does this extra free energy term depend on the applied field? For what
values of H does it vanish? What would you expect if this term was the only one
responsible for the dependence of Tc on H?
12. Check the main features that can be explained this way against the experimental
results.
The London hypothesis fixes the condensation energy in the superconducting state as
1
2
Gs − Gn = − μ0 Hcb (T) ,
2
but does not explain its physical origin. Under the London hypotheses, the density
of superconducting carriers and the condensation energy are assumed to be inde-
pendent of the dimension of the superconducting material. But we have seen that
condensation into Cooper pairs implies that ns can only vary in space over a dis-
tance greater than a characteristic length.
472 13 Problem Set
6 2R = 1.55 μm 2 R = 1.35 μm 6
μ0 H (mtesla)
4 4
2 2
6 2R = 1.20 μm 2R = 0.92 μm 6
μ0 H (mtesla)
4 4
2 2
0
3.60 3.65 3.70 3.75 3.65 3.70 3.75
T (K) T (K)
Fig. P10.4 Transitions obtained for different values of T for tin tubes of different radius R but
the same thickness d = 2, 500 Å. From Groff, R.P., Parks, R.D.: Phys. Rev. 176, 567 (1968), with
the permission of the American Physical Society ( c 1968 APS). http://link.aps.org/doi/10.1103/
PhysRev.176.567
13. If the material considered here is type I and if d < λ, can ns vary over the
thickness of the tube?
To account for the fact that ns does not necessarily have the value expected for a
bulk superconductor, Ginzburg and Landau suggested that, close to Tc , the zero-
field free energy of condensation in the superconducting state could be written as an
expansion in terms of ns of the form
Gs − Gn = Ans + Bn2s /2 ,
where A and B may depend on t = 1 − T/Tc . Its value is determined by minimising

with respect to ns , which leads to conditions on A and B that can be obtained for the
bulk material.
14. To obtain ns = 0 for T > Tc , show that A must change sign at Tc , and that B
must be positive.
The simplest form is thus A = α(Tc − T) = αTc t, taking α and B constant near Tc .
15. Show that in this case ns = α(T − Tc )/B when T < Tc for a bulk superconductor.
16. Using the experimental forms of Hcb (T) and λ(T) close to T = Tc , determine α
and B as a function of Hc (0), λ0 , and Tc .
For the thin tube in an applied field, surface currents flow around the supercon-
ductor. We therefore add the free energy term E∗ considered above and the energy
difference between superconducting and normal states is given by
1
Gs − Gn = Ans + Bn2s + γ Js2 .
2
17. Show that the critical temperature depends on the applied magnetic flux and can
be written in the form

ξ02 Φ 2
Tc (Φ) = Tc 1 − 2 N − ,
R Φ0
where ξ0 is the coherence length of the superconductor. Does this result explain
the experimental observations?
18. Combining this with the solution to question 7, do we obtain a quantitative
explanation for all the experimental results?
474 13 Problem Set
Solution
1. The tube is cooled in zero field at a temperature T < Tc (0). The flux through the
cylinder, which is initially zero, remains zero when the field B = μ0 H outside is
increased, provided that H < Hcb (T) for a type I superconductor and H < Hcb1 (T)
for a type II superconductor. The magnetic induction remains zero inside the
superconductor, because persistent currents are induced throughout a thin layer
of thickness λ at the cylinder surface (see Fig. P10.5).
2. When the system is cooled in an applied field H < Hcb (T) [or Hcb1 (T)], the flux
within the cylinder remains unchanged (up to Φ0 ). Currents flow on the outer
face and on the inside of the cylinder in such a way as to cancel B in the bulk of
the superconductor (see Fig. P10.5).
3. V is zero as long as H < Hcb (T) [or Hcb2 (T)]. It acquires a value independent
of the field, associated with the resistance of the tube in the normal state when
H > Hcb (T) [or Hcb2 (T)].
4. If the superconductor is type I (type II, respectively), this measurement gives
Hcb (T) [respectively, Hcb2 (T)]. For a type I superconductor, this determines the
condensation energy in the superconducting state, which is given by
1
2
Gs − Gn = − μ0 Hcb (T) .
2
5. Near Tc , the experimental results are approximated by the straight line

T
Hcb (T) = 2Hc (0) 1 − ,
Tc
the dashed line in Fig. P10.6, with μ0 Hc (0) = 0.03 tesla and Tc = 3.73 K for
tin. We observe that this straight line lies below the experimental curve Hcd (T)
B=0
d B
R
B B
V I
Fig. P10.5 Persistent currents induced by a field B in a superconducting cylinder after cooling in
zero field (left) and upon cooling in the applied field (right)
Fig. P10.6 Linear approxi-

mation to Hcb (T) near Tc for 1
a bulk sample (dashed line),
2R = 0.92 μm
μ 0 H (mtesla)
and temperature dependence H cd (T )
of the critical field expected
0.5
for the small thickness (d < λ) Hcb (T )
of the cylindrical supercon-
ductor (thick continuous
curve)
3.70 3.75
T (K)
obtained for a cylinder of thickness d = 2, 500 Å, for T > 3.68 K. This corre-
sponds to the fact that, for a thin superconductor compared with the penetration
depth, the magnetic induction will not be zero in the superconductor, so the field
exclusion energy cost is reduced compared to the bulk superconductor. It is thus
natural to find a higher critical field than the one observed for the bulk sample.
6. If the field profile is similar to the one for a thin film of thickness d = 2a with
d λ (which strictly applies for R d), the increase in the critical field is then
given as in Chap. 6 by
√ λ(T) b
Hcd (T) = 2 3 Hc (T) .
d
7. Near T = Tc , setting t = 1 − T/Tc , we have

−1/2
λ(T) = λ0 1 − (1 − t)4 λ0 (4t)−1/2 ,
and then
√ λ0
Hcd (T) 2 3 Hc (0)t1/2 .
d
8. For λ0 = 600 Å, d = 2, 500 Å, and μ0 Hc (0) = 0.03 tesla, we have
√ 6
μ0 Hcd (T) = 2 3 0.03t1/2 = 0.025t1/2 (tesla) .
25
The corresponding prediction, plotted as a continuous curve in Fig. P10.6, lies
well above the curve observed experimentally, showing that there is another
energy term affecting the critical field.
9. Consider a circuit (C) in the superconductor. Since this necessarily lies less than
a distance λ from the edge of the tube, the quantised quantity is
%
m
Φ∗ = Φ + 2 2 j · dl = NΦ0 ,
q φ (C)
476 13 Problem Set
with N an integer and the notation used in Chap. 5 (φ 2 = ns /2 and m = m∗ ).

Since j = Js varies little across the tube thickness, (C) can be taken on the outer
diameter of the tube, whereupon
q 2 ns
2π RJs = (NΦ0 − Φ) ,
2m∗
and

ns h̄q Φ
Js = N − .
2m∗ R Φ0
10. The energy to be taken into account here is the kinetic energy of the supercur-
rents. These are given by Js = −ns evs , and their kinetic energy is
1 m∗ m∗ 2
E∗ = ns v2s = J ,
2 2 ns q2 s
so that
γ = m∗ /ns q2 .
11. Substituting in the quantised version of Js , we obtain

ns h̄2 Φ 2
E∗ = N − .
4m∗ R2 Φ0
The constant in front of the bracket is independent of H for fixed T, so this

kinetic energy depends only on H through the flux Φ. Consequently, for fixed
N, E∗ is quadratic in Φ, i.e., quadratic in the applied field H. This kinetic energy
term vanishes for Φ = NΦ0 . The energy cost becomes enormous if Φ deviates
much from NΦ0 . It is easy to show that this kinetic energy term is minimised if N
changes by one unit each time |Φ − NΦ0 | > Φ0 /2. Thermodynamic equilibrium
is achieved by this adjustment of N which minimises the current. The latter
grows linearly and changes sign when Φ/Φ0 is a half-integer (see Fig. P10.7).
The corresponding dependence of E ∗ on Φ is also shown in the figure. When
the applied field has no other effect on the value of Tc , we therefore expect a
periodic dependence of Tc on the field, where the latter resumes its maximal
value each time Φ = NΦ0 .
12. This periodicity is indeed observed in the experimental curves (see Fig. P10.8).
The maxima occur for values of the field H that are multiples of a fixed value
Hp for each of the four tubes. The minima of Tc also occur for values equal
to half this period Hp . In fact, Hp varies approximately linearly with R−2 (see
Fig. P10.9). From the slope of the straight line, we obtain
Fig. P10.7 Dependence of Js Js

and E∗ on Φ
Φ /Φ 0
E∗
Φ /Φ 0
−2 −1 0 1 2
6 2R = 1.55 μm 2 R = 1.35 μm 6
μ0 H (mtesla)
4 4
2 2
6 2R = 1.20 μm 2R = 0.92 μm 6
μ0 H (mtesla)
4 4
2 2
0
3.60 3.65 3.70 3.75 3.65 3.70 3.75
T (K) T (K)
Fig. P10.8 Transitions obtained for different values of T for tin tubes of different radius R but the
same thickness d = 2, 500 Å
Φ0 = π R2 μ0 Hp (π/4)(0.92)2 (3)10−15
1.99 × 10−15 weber,
which agrees reasonably well with the value Φ0 = 2.07 × 10−15 weber given
that the experimental error in the tube diameter is at least 5%. Note also that
the amplitude of the oscillations in Tc (H) increases as R−2 , as predicted by the
above equation (see Fig. P10.9). We conclude that this kinetic energy term does
account for most of the experimental results.
478 13 Problem Set
Fig. P10.9 Linear depen- μ 0 Hp ΔTc,max

dence of Hp and the amplitude (mtesla) 3 (mK)
of the oscillations in Tc (H)
on R−2
2 100
1 50
0.5 1 1.5
(2R)−2 (μm−2)
13. Since d < λ, and since λ < ξ in a type I material, the fact that d < λ implies
d < ξ , which implies that ns cannot vary significantly over the thickness of the
tube.
14. Setting
Gs − Gn = Ans + Bn2s /2 ,
and minimising this free energy with respect to ns yields
ns = −A/B ,
with B > 0 for this condition to correspond to a minimum. Since ns 0, A and

B have opposite signs for T < Tc (hence A < 0), and the same sign for T > Tc
(to avoid there being a physical solution with ns positive). Moreover, for T =
Tc , we must have ns = 0, and hence A = 0 for T = Tc . The parameter A must
therefore grow steadily and change sign at Tc .
15. With A = α(Tc − T) = αTc t, where α < 0 and B is constant close to Tc , we do
indeed obtain ns = α(T − Tc )/B when T < Tc for the bulk superconductor, and
A2 1 α 2 Tc2 2
Gs − G n = − =− t .
2B 2 B
16. This value ns = α(T − Tc )/B = −(αTc /B)t should be compared with the exper-
imental expression for the temperature dependence of λ given in Chap. 5. Close
to Tc , this takes the form
1/2
m0
λ(T) = λ0 (4t)−1/2 ,
ns e2 μ0
or
ns e2 μ0 λ20
4t .
m0
We may thus conclude that
αTc 4m0
− = .
B e μ0 λ20
2
Moreover, the experimental parabolic shape of Hcb (T) implies that, near Tc , we
have Hcb (T) = 2Hc (0)t, that is
1
2 1 α 2 Tc2 2
Gs − Gn = − μ0 Hcb (T) = −2μ0 Hc2 (0)t2 = − t .
2 2 B
We thus have
2
−αTc m0 = eμ0 λ0 Hc (0) .
17. We have
1 m∗ Js2
Gs − Gn = α(Tc − T)ns + Bn2s + .
2 ns q2
With the form of Js obtained in question 9, we observe that the free energy E∗
is linear in ns , so
1 m∗ 2
Gs − Gn = α(Tc − T)ns + Bn2s + ns vs ,
2 4
where vs = h̄(N − Φ/Φ0 )/m∗ R is independent of ns . Minimising this free energy
with respect to ns , we obtain
m∗ 2
Bns = −α(Tc − T) − v .
4 s
The superconducting state disappears when ns = 0, i.e., when
m∗ 2
Tc (Φ) = Tc + v .
4α s
We thus obtain

ξ02 Φ 2
Tc (Φ) = Tc 1− 2 N − ,
R Φ0
where
h̄2
ξ02 = −
4αm∗ Tc
defines the coherence length of the superconductor. This is similar to the expres-
sion in question 11, but does not involve a value of ns independent of the applied
flux. It is in fact the vanishing of ns that defines the critical temperature under
any experimental conditions. The other experimental aspects are described in
the same way as in question 11.
480 13 Problem Set
18. For Φ = NΦ0 , flux quantisation does not affect Tc . Its dependence on the applied
field is then entirely due to the free energy associated with flux penetration. The
value
μ0 Hcd (T) = at1/2 , a = 0.025 tesla ,
estimated in question 7 can therefore describe the temperature dependence of the
fields corresponding to the maxima of Tc of the curves Tc (H). Indeed, the dotted
lines in Fig. P10.8 show quite clearly that these fields do vary as (Tc − T)1/2 .
The numerical constant a = 0.0416 obtained experimentally is the same for all
the tubes and is slightly greater than indicated above.
Note: Continuous curves in Fig. P10.8 are the results of a detailed calculation
in which the magnetic energy is included in the Ginzburg–Landau formalism.
Problem 11: Direct and Alternating Josephson Effects in Zero Magnetic Field 481
Problem 11: Direct and Alternating Josephson Effects

in Zero Magnetic Field
The aim here is to study the characteristics of a Josephson junction subjected to a

static or alternating excitation. The Josephson effect is a tunnelling of the macro-
scopic wave function through an insulating barrier between two superconductors
(1 and 2). Recall first the characteristic equations relating the current I and the volt-
age V across the terminals of the junction to the phase difference γ = θ2 − θ1 of the
wave function on either side of the junction:
dγ
I = Ic sin γ , h̄ = 2eV . (13.49)
dt
11.1: Model Josephson Junction
The junction is connected to a current source and we assume that initially I = 0 and
γ = 0. The current I(t) supplied by the current source is slowly increased up to a
constant value I0 , where |I0 | ≤ Ic .
1. What stationary state is obtained for the junction from (13.49)?
2. What energy is supplied by the current source in order to establish the current I0
starting from I = 0?
We now connect the junction to a constant voltage V = V0 at time t = 0, the earlier

values of I and γ being zero.
3. What current passes through the junction?
4. Consider the case where V0 = 10 μV. Do you think this current could be detected
directly by means of an ammeter?
11.2: Realistic Josephson Junction
In the model Josephson junction described above by (13.49), the current cannot
exceed Ic and the direct voltage is zero. In a real Josephson junction, if we start with
I = 0 and apply a voltage source V across the junction, the experimentally observed
characteristic representing the steady-state current I measured for different values
of V is as shown in Fig. P11.1 (left).
5. Explain the physical origin of the current arising for large V and the observed
asymptotic behaviour.
482 13 Problem Set
V I
I
Ic
J Rn
Vc V 0 Ic I
−Ic
Fig. P11.1 Realistic Josephson junction: Experimentally observed characteristics I(V) (left) and
V(I) (center) and equivalent circuit when supplied with current I (right)
Likewise, when a current is passed through the Josephson junction, the observed
characteristic V(I) is as shown in Fig. P11.1 (center).
6. Explain the physical origin of the current for I > Ic and the asymptotic behaviour
observed for large I.
We thus represent the realistic Josephson junction with current supply by the equiv-
alent electric circuit shown in Fig. P11.1 (right), where the ideal Josephson junction
J described by (13.49) is connected in parallel with a resistance Rn .
7. Write down the equation for the behaviour of the phase γ for this equivalent
circuit.
8. Show that, for I(t) = I constant with |I| < Ic , we recover the same response as
for the ideal junction.
For I > Ic , the phase γ is no longer bounded. The equation found in question 7
remains unchanged if γ is increased by 2π . Therefore, the time dependence of γ
can be decomposed into a linear variation on which is superposed a periodic varia-
tion with period τ . There corresponds a periodic variation of V(t) about an average
value V.
9. Determine τ and hence V as a function of I, using the fact that
2π dx 2π
=√ . (13.50)
0 1 − b sin x 1 − b2
10. Is the result obtained here compatible with the characteristic of Fig. P11.1
(center)? Examine the asymptotic behaviour. What is the physical meaning
of Rn ?
V (μvolt)
Attenuation of microwave power (dB)
−max −50 −45 −40 −38 −36 −34 −32 −30 −29 −28 −26 −24
600
450
300
150
0
−150
−300
−450
−600
I (arbitrary units)
Fig. P11.2 Characteristics V(I) observed for an Nb/NbO/Nb junction for different values of the
applied microwave irradiation power (72 GHz). From Grimes, C.C., Shapiro, S.: Phys. Rev. 169,
397 (1968). With the permission of the American Physical Society (
c 1968 APS). http://link.aps.
org/doi/10.1103/PhysRev.169.397
11.3: Josephson Junction in a Microwave Field
We consider a Josephson junction Nb/NbO/Nb (superconductor–insulator–super-

conductor), with a current supply, and study the characteristic V(I) of this junction
when it is irradiated by a microwave field of angular frequency ω. The measured
characteristics V(I) are shown in Fig. P11.2 for different powers of the microwave
source at frequency ω/2π = 72 GHz. Note that, since the junction is symmetric
by construction, the characteristics are symmetric with respect to the origin, i.e.,
V(−I) = −V(I). The curves in the figure have thus been shifted arbitrarily along the
horizontal axis, with increasing microwave power from left to right.
11. Give a qualitative description of the observed phenomena, without attempting

to interpret them for the moment.
12. Looking back at the results of questions 3 and 4, can you identify the main
physical effect observed here?
In order to give a quantitative explanation for the experimental results in Fig. P11.2,
we shall now examine the case of a junction with a voltage supply. Analytic calcula-
tions are not feasible for a junction with a current supply. However, we assume that
the observed effects will be similar.
It is assumed that the effect of the microwave field is to superpose upon the
voltage V a microwave voltage vs cos ωt of strength vs proportional to the amplitude
of the incident microwave field.
13. Write down the voltage–phase Josephson equation (P11.1) under these condi-
tions and deduce an expression for the current. The relation
484 13 Problem Set
+∞

eib sin x = Jn (b)einx
n=−∞
may be useful, where Jn is the order n Bessel function of the first kind, a real-
valued function satisfying J−n (b) = (−1)n Jn (b).
14. Is the current expected to change at zero frequency? Can you explain certain
experimental features noted in question 11 as observed in Fig. P11.2? Hint: For
small x,
x2 1 x n
J0 (x) 1 − , Jn (x) .
4 n! 2
Solution
Model Josephson Junction
The starting point here is

dγ
I = Ic sin γ , h̄ = 2eV .
dt
1. When I is increased to I0 , γ increases as a function of time to reach the value
γ = arcsin(I0 /Ic ). During this variation dγ /dt is nonzero and a voltage V(t) ap-
pears, but dγ /dt vanishes when I = I0 , and then V = 0. The stationary state thus
corresponds to
γ = arcsin(I0 /Ic ) , I = I0 , V =0 .
2. The energy supplied by the current source is

∞
h̄Ic ∞ dγ (t)
W = I(t)V(t)dt = sin γ (t) dt
0 2e 0 dt
h̄Ic ∞ h̄Ic
=− cos γ (t) 0 = (1 − cos γ ) .
2e 2e
The energy of the junction therefore increases and is minimal for γ a multiple
of 2π .
3. When a voltage V0 is applied at t = 0, with I = 0 and γ = 0 initially, we obtain

2eV0 2eV0
γ (t) = t, hence I(t) = Ic sin t .
h̄ h̄
The phase grows continuously and I is an alternating current at the Josephson

frequency
νJ = 2eV0 /h ,
so νJ /V0 = 483.6 × 1012 (Hz/volt). This assumes that eV0 < 2, where is
the superconductor gap. Otherwise one also expects superconducting pairs to be
broken, an effect neglected in this model.
4. For V0 = 10 μvolt, νJ 5 GHz. An alternating current at such a high frequency
could not of course be detected by a simple device like an ammeter. One might
attempt to detect the electromagnetic power emitted from the junction at this
frequency. This is possible but difficult, since the signal is very weak!
486 13 Problem Set
Realistic Josephson Junction
5. For the characteristic observed in Fig. P11.1 (left), a current flows when the ap-
plied voltage is such that eV exceeds eVc = 2, where is the superconductor
gap. Indeed, this voltage would be required to break the Cooper pairs and give
rise to ‘normal’ electrons. The latter can tunnel from 1 to 2. Such an electronic
tunneling current is dissipative, unlike superconducting pair tunneling. When V
is increased, the number of broken pairs is also increased, and for large enough
V, there will be no more superconducting pairs. The tunneling current therefore
tends asymptotically toward the linear characteristic for electron tunneling be-
tween normal metals, such as would be observed above Tc for this same junction.
Its slope is 1/Rn , where Rn is the normal state junction resistance.
6. In a setup with a current supply, when I exceeds Ic , superconducting pairs are
broken and ‘normal’ electrons appear. It is the tunneling current of these elec-
trons that adds to the current of superconducting pairs. When I is increased,
the number of broken pairs also increases and for large enough I, tunneling by
superconducting pairs disappears. The tunneling current therefore tends asymp-
totically to the linear characteristic V = Rn I for tunneling by normal electrons.
7. In the circuit of Fig. P11.1 (right), the current I is the sum of the currents in the
two branches, viz.,
V dγ h̄
I = Ic sin γ + = Ic sin γ + ,
Rn dt 2eRn
and then
dγ 2eRn Ic dt
= .
I/Ic − sin γ h̄
8. For I < Ic , we find that γ increases up to the value I = Ic sin γ . As in question 1,

this leads to V = 0.
9. For I > Ic , we find that γ is no longer bounded, increasing indefinitely, according
to the above relation. γ increases by 2π over a period τ , and hence 2eV =
h̄dγ /dt = h/τ . We thus have

h̄ 2π dγ h 2 2 −1/2
τ= = I − Ic ,
2eRn Ic 0 I/Ic − sin γ 2eRn
using the given integral, and this yields

V = Rn I 2 − Ic2 .
√
10. V is indeed zero up to I = Ic and grows as I − Ic thereafter. The limiting
behaviour does indeed correspond asymptotically to V = Rn I. The above equa-
tion gives a good description of the behaviour in Fig. P11.1 (center).
Josephson Junction in a Microwave Field
11. We note the following points with regard to Fig. P11.3:

(a) At zero microwave power, the observed characteristic does correspond to a
realistic Josephson junction with current supply, as in Fig. P11.1 (center).
(b) Voltage plateaus appear in the characteristic curves for discrete values of
the voltage, which seem to be quantised and independent of the microwave
power.
(c) The width of the voltage plateau at V = 0, initially 2Ic , seems to steadily
diminish as the microwave power increases.
(d) New plateaus have current widths which grow to a maximum, then decrease
with increasing microwave power.
(e) For strong irradiation, the characteristic is almost linear, and its slope is the
large-current limit corresponding to the characteristic without microwave
irradiation, i.e., the tunneling resistance Rn of the junction in the normal
state.
12. As soon as V = 0 across the terminals of the junction, there is an alternating
Josephson current, according to questions 3 and 4. The incident microwave
field produces an alternating electric field in the junction. It seems likely that
this field couples with the alternating Josephson field in the junction and that
a resonance effect occurs when the frequencies coincide. Indeed, a Josephson
frequency of 72 MHz corresponds to a voltage of about 150 μvolt. The observed
plateaus, known as Shapiro steps after the physicist who first demonstrated this
V (μvolt) Attenuation of microwave power (dB)

−max −50 −45 −40 −38 −36 −34 −32 −30 −29 −28 −26 −24
600
450
300
150
0
−150
−300
−450
−600
I (arbitrary units)
Fig. P11.3 Characteristics V(I) observed for an Nb/NbO/Nb junction for different values of the
applied microwave irradiation power (72 GHz)
488 13 Problem Set
phenomenon, occur at integer multiples of 150 μvolt. This implies that all mul-
tiples of the irradiation frequency are detected. These observations thus reveal
the existence of the alternating Josephson effect.
13. With V(t) = V0 + vs cos ωt, solving the voltage–phase relation, we obtain
2eV0 2evs
γ (t) = γ (0) + t+ sin ωt .
h̄ h̄ω
This leads to a current

2eV0 2evs
I = Ic sin γ (0) + t+ sin ωt .
h̄ h̄ω
As the current–phase relation is non-linear, we see that the current response is

not at the frequency of the microwave excitation. Indeed, it also contains the
alternating Josephson frequency of questions 3 and 4 and the harmonics of this
frequency.
This response can be Fourier analysed using the given expression:
+∞

eib sin x = Jn (b)einx ,
n=−∞
where Jn is the order n Bessel function of the first kind. We then have
+∞
+∞

ei(a+b sin x) = Jn (b)ei(a+nx) = (−1)n Jn (b)ei(a−nx) .
n=−∞ n=−∞
In the second equality, we used J−n (b) = (−1)n Jn (b). Taking the imaginary part
of this expression with
2eV0 2evs
a = γ (0) + t = γ (0) + 2π νJ t , b= ,
h̄ h̄ω
we obtain
+∞

2evs
I = Ic (−1) Jnn
sin γ (0) + (2π νJ − nω)t .
n=−∞
h̄ω
14. We check the points (b)–(e) noted for question 11:

(b) Note that the current I due to superconducting pairs has a direct component
each time 2π νJ = nω, i.e., for all Josephson frequencies that are integer
multiples of the microwave irradiation frequency ω. We thus have a direct
current of superconducting pairs for all discrete values Vn of V0 such that
2eVn = nh̄ω, i.e., Vn = n(h̄ω/2e).
(c) The amplitude of this current depends on the value of γ (0), and can take
any value |I| < Ic Jn (2evs /h̄ω) for the voltage Vn .
This dependence on γ (0) is exactly analogous to what was obtained for the
characteristic without irradiation at V0 = 0. There the phase adjusted itself
to fix the component of the supercurrent in the junction.
In the experiment where the junction is supplied with a current, the pair
current adds to the one present when there is no irradiation, due to the
resistance in parallel with the junction (that is, to the tunneling of ‘nor-
mal’ electrons). So in the presence of the microwave field, the super-
current at zero voltage can take any amplitude up to the maximal value
|I| < Ic J0 (2evs /h̄ω), which itself depends on the irradiation power. Expand-
ing J0 to lowest order in its argument, we obtain
2
evs
|I| < Ic 1 − 4 .
h̄ω
This explains the reduction of the plateau at zero voltage as a function of

irradiation power.
(d) Likewise, since the Jn (x) go as xn for small x, the current plateaus at higher
voltages grow broader with the irradiation power according to

2Ic evs n
2Ic Jn (2evs /h̄ω) = .
n! h̄ω
This shows that the first order plateau appears first, then the second order
plateau, and so on. At increasing powers, Jn (x) goes through a maximum
and subsequently oscillates. The first plateau at 150 μvolt can be seen to
increase and then fall off with the power.
(e) The condition 2eV1 = h̄ω can be considered to correspond to pair tunneling
helped by absorption of a microwave photon of energy h̄ω. Likewise, at the
Vn plateau, n photons are absorbed simultaneously. For strong irradiation,
the absorption of many photons breaks the superconducting pairs, and the
normal electron current dominates the response. We then recover the linear
characteristic which is the strong current limit when there is no microwave
irradiation.
Problem 12: Josephson Junction in a Magnetic Field 491
Problem 12: Josephson Junction in a Magnetic Field7
The aim here is to understand the effect of a magnetic field on certain properties of a
Josephson junction: current distribution and analogies with type II superconductors
(12.1), screening of the magnetic field inside a junction (12.2), and the observation
of plasma resonance phenomena (12.3).
Note: 12.3 is independent of 12.2.
Consider a Josephson junction comprising an insulator placed between two identi-

cal superconductors, as in Fig. P12.1. The two superconductors occupy the regions
x < −w/2 and x > w/2, labelled S1 and S2 , respectively. The insulator is in the
region J specified by
−w/2 < x < w/2 , −L/2 < y < L/2 .
The system is assumed infinite in the z direction. The local magnetic induction lies
along the z axis, viz., B = Bz, with B ≥ 0. The field applied outside corresponds to
the magnetic induction Ba z = μ0 Ha z.
Recall that an electric current flows between the two superconductors by tunnel-
ing of Cooper pairs. The local current density is
jx (−w/2, y) = jx (w/2, y) = j0 sin δ(y) , (13.51)
where j0 is characteristic of the junction (it can be shown that j0 = −|j0 | is negative)
and δ(y) is the local phase difference between the two superconductors. (In general
y
Ba
L
2
C
w w
−
2 2
x
Fig. P12.1 Configuration of

the Josephson junction and S1 J S2
applied field. The super-
conducting (S1 , S2 ) and
insulating (J) regions are L
−
indicated 2
7 This problem has been designed with P. Ledoussal.

492 13 Problem Set
w/2
q
δ(y) = θ (w/2, y) − θ (−w/2, y) − dxAx (x, y) ,
h̄ −w/2
with q = −2|e|, but here we can choose a gauge with Ax = 0.)

Let I be the total current through the junction, from S1 to S2 , per unit length in
the z direction, and λ the London penetration depth. The superconductors S1 and S2
are type II with an applied field Ha Hc1 .
In 12.1 and 12.2, no voltage is applied and the problem is thus time-independent.
In the geometry of Fig. P12.1, we write B = B(x, y)z and
j = jx (x, y)x + jy (x, y)y .
We have
j = jM + j J , (13.52)
where jJ is the Josephson current and jM the Meissner screening current when we
set j0 = 0 (i.e., when each superconductor is considered alone). We thus have jM = 0
in the region J.
12.1: Current Distribution
Throughout this section we assume that |j0 | is small enough to justify completely
neglecting its influence on the field, which is therefore uniform and equal to the
applied field B = Ba outside the superconductors, in particular in region J. We begin
by setting j0 = 0 (and hence I = 0).
1a. Write down the two equations relating j and B in a superconductor. Deduce
the equation determining B inside the superconductors S1 and S2 .
1b. Without solving these equations, sketch the current lines jM when B > 0. Over
what thickness does jM flow?
1c. Express B = B(x) as a function of Ba in the three regions when L/2 − |y| λ.
We now assume that L λ and hence neglect the edge regions (L/2 − |y| ∼ λ) in
all calculations. We consider j0 = 0.
2a. Write down the equation relating j, θ , and A in a superconductor.
2b. Calculate δ(y) in terms of δ(0), Ba , the flux quantum Φ0 = h/2|e|, λ, w, and
j0 . Hint: Consider the closed path C of Fig. P12.1. Use the equation obtained
in question 2a on the portion C of C lying in the superconductors, and relate
δ(y) − δ(0) with the magnetic flux Φ(y) through the surface generated by C.
Neglect the contribution from jJ in the superconductors.
In the following, Φtot denotes the total flux in the junction, defined here as the total
magnetic flux in S1 , S2 , and J.
3. Calculate the total current I in the junction and express it in terms of Φtot .
Now set δ(0) = 0 (so that I = 0).

4. Plot the current jJx (y) as a function of y for Ba corresponding to the two cases
(a) Φtot = 2Φ0 , (b) Φtot = 4Φ0 .
5. Sketch the current lines jJ in the three regions S1 , S2 , and J for the two cases.
Explain the qualitative analogy with a vortex arrangement in the mixed phase
of a superconductor.
12.2: Screening of the Magnetic Field
We no longer assume B to be uniform in the region J, as the aim now is to estimate

the spatial variation of B in the junction (screening). In the region J, we no longer
neglect the effect of the current jJ on the field, which becomes a function of y. We
still assume that |j0 | is small, so that we may continue to neglect |jJ | |jM | in
the superconductors (but not in the junction) when calculating the field. We also
assume that jJy = 0 throughout the region J and hence that B = B(y) and jJx (y) only
depend on y.
1. Show that, for L/2 − |y| λ, the field profile has the form
/
B(y)e−(x−w/2)/λ (region S1 ) ,
B(x, y) = (13.53)
B(y)e(x+w/2)/λ (region S2 ) .
We now determine B(y) and δ(y).

2. Show that δ(y) satisfies the differential equation
d2 δ(y) 1
= 2 sin δ(y) , (13.54)
dy2 λJ
and express the characteristic length λJ as a function of the flux quantum Φ0 ,

|j0 |, w, and λ. Hint: Use Maxwell’s equation in the junction:
dB
= μ0 jJx , (13.55)
dy
and the closed path C and C defined in question 2b (Fig. P12.1).
We now determine the boundary conditions for solving (13.54) for fixed Ba and I.
= (dδ/dy)
Define B± = B(0, ±L/2) and δ± y=±L/2 .
3. Write δ± as a function of B . Show that B = B ± μ I/2. Relate the total
± ± a 0
flux Φtot in the junction to δ+ − δ− .
494 13 Problem Set
In order to analyse the solutions of (13.54), it is useful to note the analogy with
the equation for the oscillations of a pendulum, if δ is the angle measured from the
highest point (position of unstable equilibrium), and if we take y proportional to
time.
4a. Write down the conserved first integral (analogous to the total mechanical
energy of the pendulum), viz.,
1
E = δ 2 + U(δ) .
2
Fit the constant of the potential energy U(δ) so that it vanishes at δ = 0.
From now on we consider the case δ(0) = 0, I = 0, and Ba > 0. By symmetry, we

may choose a solution with δ(−y) = −δ(y), and therefore

δ+ = δ− , δ+ = −δ− . (13.56)
Note that δ(y) is a strictly decreasing function and E ≥ 0.

4b. Express E as a function of Ba and Φtot .
We now investigate in turn the three following regimes: Ba very large (question 5),
Ba very small (question 6), and Ba of intermediate value (questions 7, 8, and 9). We
begin by examining the situation where E is large (kinetic energy much greater than
the potential energy).
5. Show that we then recover the situation in which the field is uniform in the
region J, B = Ba , with
2π
δ(y) = − (2λ + w)Ba y . (13.57)
Φ0
Show that this limit corresponds to
B B∗ , (13.58)
where B∗ is a characteristic field, and give an expression for this characteris-

tic field up to a numerical prefactor. Show that B∗ vanishes when |j0 | → 0 or
λJ → +∞.
6a. Show that, if B is very small, then (13.54) can be linearised. Deduce the equa-
tion satisfied by B(y) in this limit. Discuss the analogy with Meissner screening
in a superconductor. What is the penetration depth?
6b. Calculate B(y) and δ(y). Show that, if the junction is long enough, there will
be total screening of the field within it. What is the exact condition on Ba , Φ0 ,
λ, w, λJ , and L for the linearised equation to provide a good approximation?
Show that, for large L, we do indeed recover the condition B B∗ .
In the following we shall consider an intermediate field Ba of order B∗ . Consider

the solution with E ≈ 0 such that the pendulum has just enough kinetic energy to go
above the point δ = 0.
7a. Plot U(δ) and indicate the trajectory. Show that this situation corresponds to
very long junctions L λJ . Show that δ is given at the edges y = ±L/2 by
2
Ba
cos δ± = 1 − 2 . (13.59)
B∗
Give an exact expression for B∗ .

7b. What value of the total flux Φtot does Ba = B∗ correspond to?
7c. In the analogy with a type II superconductor, what characteristic field H does
Ba = B∗ correspond to? Explain. Is screening total for B > B∗ ?
8. What condition must λJ satisfy for the superconductors S1 and S2 to be effec-
tively in the Meissner phase?
9. Find the analytic solution for δ(y) and B(y) when E = 0. Show that we recover
the solution obtained for question 6b when Ba B∗ and L λJ and that it
describes the profiles of B and j more precisely near the edges over the whole
interval 0 < Ba < B∗ . Note the similarity with the equation describing the
shape of a Bloch wall and the differences in the boundary conditions.
12.3: Josephson Plasma Resonance
When a voltage V is applied, the Josephson current is still determined by (13.51),

but the phase difference acquires a time dependence:
∂δ 2|e|V
=− , (13.60)
∂t h̄
where w/2
V = V1 − V2 = Ex dx
−w/2
is the potential difference across the terminals of region J (using the gauge Ax = 0).
Assume that Ba > 0 and Ex are uniform and time-independent in region J and
outside the superconductors. (As in 12.1, we neglect the effect of jJx on B in the
junction.)
1. Show that the Josephson current is a plane wave

jJx (y, t) = j0 sin δ(0) + k0 y + ω0 t . (13.61)
496 13 Problem Set
Determine k0 , ω0 , and the velocity v0 of the wave as a function of the constant

voltage V0 applied across the terminals of the junction. In the analogy with
the type II superconductor, what corresponds to the velocity v0 ?
We now consider the problem for zero applied field Ba = 0, but when there is a
nonzero total Josephson current I, still neglecting the effect of the latter on the field.
We may thus assume that B = 0 everywhere and δ(y) = δ is uniform. We study small
oscillations of the electric field Ex (t), the current jx (t), and δ(t) in the junction as a
function of time, assuming all these quantities to be spatially uniform.
2. Use Maxwell’s equation in the presence of a time-dependent electric field to
show that
d2 δ
+ ωJ2 sin δ = 0 . (13.62)
dt2
Find ωJ and the corresponding frequency νJ (Josephson plasma mode).
In the following experiment, small voltage oscillations were measured while excit-
ing the system at a fixed microwave frequency νp , as a function of the direct current
I injected into the junction. A resonance is observed for a certain current I = Ip .
Figure P12.2 shows the values of νp obtained as a function of the current injected
into an Sn/SnO/Sn junction.
3. Show that (13.62) explains the experimentally observed dependence of νp
on Ip . What qualitative dependence of j0 on the superconductor gap can be
deduced from the temperature dependence? Estimate Tc .
1.0
Sn–SnO–Sn
0.9 H=0
1.15 K
2.05 K
0.8 2.53 K
3.05 K
3.26 K
0.7
Fig. P12.2 Dependence of νp4
on Ip2 at different tempera-
νp4 (10 10 Hz)4
0.6
tures. Current values Ip have
been normalised by the criti- 0.5
cal current I0 of the junction
measured at the lowest tem- 0.4
perature. From Dahm, A.J.,
0.3
Denenstein, A., Finnegan,
T.F., Langenberg, D.N., 0.2
Scalapino, D.J.: Phys. Rev.
Lett. 20, 859 (1968). With the 0.1
0 0.2 0.4 0.6 0.8 1.0
10.1103/PhysRevLett.20.859 Ip2
Solution
Current Distribution
1a. The equations determining the current and magnetic field in a superconductor
(in the Meissner phase) are
μ0 j = curl B (Maxwell) , (13.63)
λ2 μ0 curl j = −B (London) , (13.64)
where λ2 = m0 /ns e2 μ0 = m/4φ 2 e2 μ0 is the London penetration depth, with

m0 the electron mass, ns the density of superconducting electrons, φ 2 = ns /2
the density of Cooper pairs, m the mass of a pair, and −|e| the electron charge.
We also have div j = div B = 0. These imply
1
∇ 2B = B. (13.65)
λ2
With the geometry of Fig. P12.3, the field B = B(x, y)z is determined by
solving

∂2 ∂2 1
+ 2 B(x, y) = 2 B(x, y) , (13.66)
∂x 2 ∂y λ
in S1 and S2 , with B(x, y) = Ba on the boundaries of S1 and S2 .

1b. The current j = jM is obtained from the solution B(x, y) of (13.66). Its compo-
nents are then
y Ba
λ
S1 J S2
Fig. P12.3 Geometry of the

junction and currents induced
by the applied field λ
498 13 Problem Set
∂B(x, y) ∂B(x, y)
μ0 jx (x, y) = , μ0 jy (x, y) = − , (13.67)
∂y ∂x
with the boundary conditions jy (x, ±L/2) = 0 for |x| > |w|/2. The current
flows through a layer of thickness roughly equal to λ near the surfaces. jM
is parallel to the edges very close to them (at distances ξ and λ). It
decreases roughly exponentially over distances of the order of λ as we move
away from the surfaces of S1 and S2 . The current lines are shown in Fig. P12.3.
1c. For L/2 − |y| λ (far from the horizontal edges), the current flows in the
x = 0 and B(x, y) = B(x). The solution described in
y direction. Therefore, jM
question 1a is thus
⎧
⎪
⎨ Ba , |x| < w/2 ,
B(x) = Ba e (x+w/2)/λ , x < −w/2 , (13.68)
⎪
⎩
Ba e−(x−w/2)/λ , x > w/2 .
2a. In a superconductor, (5.25) gives
h̄
μ0 λ2 j = ∇θ − A , (13.69)
q
where q = −2|e| is the charge of a Cooper pair.

2b. The closed path C is the rectangle shown in Fig. P12.4, with sides y = 0, y,
and x = ±X. We integrate (13.69) over the portion C of C within the super-
conductor (where it is valid):

h̄
μ0 λ2 j · dl = ∇θ · dl − A · dl + A · dl . (13.70)
C q C C C\C
$Choosing the gauge Ax = 0, the integral along the missing path is zero, i.e.,
C\C A · dl = 0. Moreover,
y
Ba
L
2
C
w w
− x
2 2
Fig. P12.4 Josephson junc-

tion configuration and il- S1 J S2
lustration of the rectangular
circuit C and its part circuit
C within the superconductor −
L
used to integrate (13.69) 2

−w/2 ∞

j · dl = dx + dx jx (x, 0) − jx (x, y) + O e−|X|/λ ≈ 0 ,
C −∞ w/2
(13.71)
since is zero for L/2 − |y| λ, far from the edges which we neglect here,
jM
x
and since we neglect jJ . Taking |X| → +∞, and with φ0 = h/2|e|, Eq. (13.70)
reduces to

Φ0
− ∇θ · dl = A · dl = Φ(y) ,
2π C C
or

Φ0 Φ0
− ∇θ · dl = − θ (w/2, y) − θ (w/2, 0)
2π C 2π

+ θ (−w/2, 0) − θ (−w/2, y)
Φ0
=− δ(y) − δ(0) . (13.72)
2π
In addition,
+∞ y
Φ(y) = Bz dxdy = B(x, y)dxdy = (2λ + w)Ba y , (13.73)
" −∞ 0
where " is the surface generated by C. Hence,
2π
δ(y) = δ(0) − (2λ + w)Ba y . (13.74)
Φ0
3. The total current in the junction is

L/2 L/2
2π
I= dyjx (y) = j0 dy sin δ(0) − (2λ + w)Ba y , (13.75)
−L/2 −L/2 Φ0
or
sin(π Φtot /Φ0 )
I = j0 L sin δ(0) , (13.76)
π Φtot /Φ0
where Φtot = (2λ + w)Ba L is the total flux in the junction (S1 , S2 , and J).
4. When δ(0) = 0 and Φtot = nΦ0 , we write

y y
δ(y) = −nπ , jJx (y) = |j0 | sin nπ . (13.77)
L/2 L/2
Figure P12.5 represents (13.77) for n = 2 and n = 4.

500 13 Problem Set
δ jx
nπ
−L/2 | j0 | y
y L/2
−L/2 L/2
−nπ
Fig. P12.5 Variation across the junction (left) of the phase of the wave function and (right) of the
current for n = 2 (dashed line) and n = 4 (full line)
y Ba y Ba
S1 J S2
x x
J S2
S1
λ λ
Fig. P12.6 Illustration of the periodic current configurations in the junction for n = 2 and 4
5. We now plot the maximum current lines corresponding to question 4. For

n = 2 and 4, they close naturally, and the configuration is periodic (see
Fig. P12.6). This configuration of pair currents tunneling through the junction
is reminiscent of the currents associated with a periodic row of n vortices
along the y axis, as would appear in a thin type II superconductor above the
critical field Hc1 . The average spacing between these vortices, called Joseph-
son vortices, is a = Φ0 /[(2λ + w)Ba ] here.
Screening of the Magnetic Field
1. We continue to neglect |jJ | |jM | in the superconductors, and jM still lies

along the y direction. Since the London equation is
B ∂ ∂
= (μ0 jx ) − (μ0 jy ) , (13.78)
λ 2 ∂y ∂x
we still have
1 ∂2 ∂2 ∂2
B= + 2 B≈ 2B (13.79)
λ2 ∂x 2 ∂y ∂y
inside the superconductors, which amounts to assuming that λJ λ (see
below). The solution with B = B(y) in region J is
⎧
⎪
⎨ B(y) , |x| < w/2 ,
B(x, y) = B(y)e (x+w/2)/λ , x < −w/2 , (13.80)
⎪
⎩
B(y)e −(x−w/2)/λ , x > w/2 .
It remains to determine B(y).

2. We have
∂B
= μ0 jJx .
∂y
We still assume that jJy = 0, so ∂B/∂x = 0 and ∂jJx /∂x = 0, so jJx only depends
on y. Then
∂
B(y) = μ0 j0 sin δ(y) . (13.81)
∂y
Furthermore, we still assume that jM J
x and jx can be neglected on the closed
contour. Therefore, the relation (13.72), viz.,
Φ0
Φ(y) = − δ(y) − δ(0) ,
2π
is still valid. Differentiating, we have
d 2π dΦ 2π
δ(y) = − = − (2λ + w)B(y) , (13.82)
dy Φ0 dy Φ0
d2 δ(y) 2π 1
= −μ0 j0 (2λ + w) sin δ(y) = 2 sin δ(y) , (13.83)
dy 2 Φ0 λJ
with
.
Φ0
λJ = . (13.84)
2π μ0 |j0 |(2λ + w)
The length λJ diverges when |j0 | → 0.

3. According to (13.82), the boundary conditions are
2π
δ+ = δ (L/2) = − (2λ + w)B+ ,
Φ0
(13.85)
2π
δ− = δ (L/2) = − (2λ + w)B− .
Φ0
502 13 Problem Set
Integrating μ0 jx = ∂B/∂y in the junction, B is found by superposition as the

sum of the field at I = 0 and the field produced by the current I :
B+ = B(L/2) = Ba + μ0 I/2 ,
(13.86)
B− = B(−L/2) = Ba − μ0 I/2 .
Integrating (13.82), we find the total flux Φtot in the junction:
2π
δ+ − δ− = − Φtot . (13.87)
Φ0
Note that we also have

L/2 L/2d2 δ(y)
I = j0 sin δ(y) = j0 λ2J 2
= j0 λ2J (δ+ − δ− ). (13.88)
−L/2 −L/2 dy
4a. There is an analogy with a pendulum, where δ is the angle measured from the
highest point:
d2 δ
m2 − mg sin δ = 0 .
dt2
Comparing with (13.83), putting t → y, we find that λ−2J → g/ = ω0 corre-

2
sponds to the natural frequency of the pendulum. The conserved total energy is

1 dδ(y) 2 1
E= − 2 (1 − cos(δ) = K + U(δ) . (13.89)
2 dy λJ
4b. We now consider δ(0) = 0 and I = 0, but Ba > 0. We can thus choose a solution
with δ(−y) = −δ(y) and
2π
δ+ = δ− =− (2λ + w)Ba , δ+ = −δ− < 0 . (13.90)
Φ0
Since δ(0) and U vanish for y = 0, whereas the kinetic energy does not, we
must have E > 0. This means that there are no trajectories that return along
the same path, and then δ(y) is a monotonic decreasing function. δ− can then
be related to the total flux by
Φtot
δ− = π , (13.91)
Φ0
and E can be identified with its value at y = −L/2, δ = δ− :
1 2 1
E = (δ− ) − 2 1 − cos(δ− ) , (13.92)
2 λJ
or
2
1 2π 1 Φtot
E = E(Ba , Φtot ) = (2λ + w) Ba − 2 1 − cos π
2
.
2 Φ0 λJ Φ0
(13.93)
The problem is fully determined. Integrating the equation expressing energy

conservation, using the fact that δ(y) is a monotonic decreasing function,
we obtain a further (complicated) relation which can in principle be used to
calculate Φtot in terms of Ba and L :
δ− dδ
L=
δ+ 2E + 2(1 − cos δ)/λ2J
+π Φtot /Φ0 dδ
= . (13.94)
−π Φtot /Φ0 2E(Ba , Φtot ) + 2(1 − cos δ)/λ2J
5. We assume that the total energy E is large (strong field). In this case, the
pendulum will oscillate very fast and we may neglect the potential energy U.
We thus obtain
2π
δ(y) = − (2λ + w)Ba y . (13.95)
Φ0
This solution is valid for

2
1 2 1 2π 1
E ≈ (δ+ ) = (2λ + w) Ba2 2 , (13.96)
2 2 Φ0 λJ
which is equivalent to
.
Φ0 μ0 |j0 | Φ0
B λ J μ 0 j0 = = . (13.97)
2π (2λ + w) 2π λJ (2λ + w)
We therefore define the characteristic field

504 13 Problem Set
Φ0
B∗ = 2 , (13.98)
2π λJ (2λ + w)
where the numerical constant will be justified in question 7a. It does indeed
vanish for λJ → +∞ and hence when j0 → 0.
6a. For very small B, we have δ− very small and so can linearise to obtain
1
∂y2 δ(y) = δ(y) . (13.99)
λ2J
Differentiating, this also leads to
1
∂y2 B(y) = B(y) , (13.100)
λ2J
which is just the Meissner equation (13.53), where λJ plays the role of the
London penetration depth. We conclude that the field B(y) is screened and
cannot penetrate the junction beyond a depth λJ .
6b. We find the solutions
cosh(y/λJ )
B(y) = Ba , (13.101)
cosh(L/2λJ )
2π sinh(y/λJ )
δ(y) = − (2λ + w)λJ Ba . (13.102)
Φ0 cosh(L/2λJ )
The linearisation condition δ+ π thus becomes
Φ0
Ba coth(L/2λJ ) . (13.103)
2(2λ + w)λJ
For long junctions, this condition is equivalent to B B∗ , but for short junc-
tions L λJ , we may in fact exploit the results of question 5 and the condition
becomes
Φtot Φ0 . (13.104)
7a. We now assume that E ≈ 0. This corresponds to a pendulum started at δ− > 0

with a speed δ− < 0 that is just enough to get it past the highest point δ = 0
+
with speed 0 . The function U(δ) is shown in Fig. P12.7. The trajectory
spends a long time near the highest point (an infinite time in the limit E → 0
because its speed then vanishes there), so E ≈ 0 corresponds to very large L.
Equation (13.93) can be rewritten using (13.98) to obtain
U(δ )
δ+ δ−
Fig. P12.7 The function U(δ)

1 Ba 2
E(Ba , Φtot ) = 2 2 − (1 − cos δ− ) , (13.105)
λJ B∗
so E = 0 corresponds to
2
Φtot Ba
cos δ− = cos π = 1−2 . (13.106)
Φ0 B∗
7b. The above equation, relating δ+ to Ba , only has solutions for Ba < B∗ . The
limiting case Ba = B∗ corresponds to δ− = π , hence Φtot = Φ0 , or exactly one
flux quantum through the junction.
7c. We thus have the following analogy:
• Ba < B∗ corresponds to the Meissner phase of a type II superconductor,
with total screening inside. For Ba < B∗ , even for very large L, the total
flux remains bounded by Φtot < Φ0 .
• Ba = B∗ corresponds to the field Hc1 for which vortices first appear in
the superconductor. For Ba > B∗ (and L large), the pendulum goes round
several times, and this corresponds to the presence of several vortices,
hence the mixed phase. For L large, the number of vortices per unit length,
or the reciprocal of their spacing, n/L = 1/a = Φtot /LΦ0 is a function
of Ba which vanishes for Ba ≤ B∗ , becomes nonzero for Ba > B∗ , and
converges to n/L = 1/a = (2λ + w)Ba /Φ0 for Ba B∗ .
• For Ba > B∗ , the pendulum goes round more than once. For all y, we may
write
/ >
1 B(y) 2
E= 2 2 − 1 − cos δ(y) , (13.107)
λJ B∗
so the field attains its maximal value Bmax for δ = (2m + 1)π and its min-
imal value Bmin for δ = 2mπ , with
2 2
1 2 Bmin Bmax
Eλ = = −1 . (13.108)
2 J B∗ B∗
For (Bmax ≥)Ba > B∗ , we have Bmin > 0 and the field is no longer fully
screened inside.
506 13 Problem Set
8. To observe these phenomena, the applied external field H ∗ = B∗ /μ0 must be

very small compared with Hc1 for the superconductors S1 and S2 . Since in
general μ0 Hc1 ∼ Φ0 /λ2 and here
Φ0
B∗ ∼ ,
λJ (2λ + w)
we must have, to within an order of magnitude,
w
λJ λ + ≈λ , (13.109)
2
for w is usually very small (nanometers).

Note: Likewise, to justify neglecting the edge regions L/2 − |y| ∼ λ in the
expressions obtained here, we must have min(L, λJ ) λ.
9. For E = 0, the equation integrates formally to

δ δ
y dδ 1 dδ
= √ = = ln | tan(δ/4)| , (13.110)
λJ 2(1 − cos δ) 2 | sin(δ/2)|
as for the Bloch wall. Since the integral diverges at δ = 0, this solution cor-
responds to L = +∞ (bearing in mind that, for E = 0, the time required for
the pendulum to go past its highest point is strictly infinite). The equation can
nevertheless be integrated, taking into account the boundary conditions (not
the same as for the wall) in the following way:
δ−
L/2 − |y| 1 dφ
= = ln tan(δ− /4) − ln tan(|δ|/4)| . (13.111)
λJ 2 |δ| sin(φ/2)
Therefore,

|y|−L/2 δ−
δ(y) = −sgn (y)4 arctan e tan , (13.112)
4
with
2
B
δ− = arccos 1 − 2 = 2 arcsin(B/B∗ ) , 0 ≤ δ− ≤ π .
B∗
Since L is in principle strictly infinite, this solution actually corresponds to

the limit L → +∞ with |y| − L/2 fixed, so it describes the profile near the
edges (over a layer of thickness roughly λJ λ). The field B(y) is obtained
by differentiating:

1
1 + tan2 arcsin(B/B∗ )
2
B(y) = Ba e(|y|−L/2)/λJ . (13.113)
2 1 ∗
1+e2(|y|−L/2)/λJ
tan arcsin(B/B )
2
In the limit B B∗ (δ− small), we recover a simple exponential, which is

indeed the large L limit, for fixed L/2−|y|, of the relation (13.101). The above
formula gives a valid description of the screening profile for arbitrary 0 < B ≤
B∗ . In particular, for B = B∗ , we have the simple solution
Ba
B(y) = . (13.114)
cosh (|y| − L/2)/λJ
Josephson Plasma Resonance
1. Here V = V0 and Ba are uniform throughout the region J and constant in time.
In the superconductors, B(x) is as in 12.1 and the argument with the closed
path in question 2b applies identically to determine ∂δ/∂y. Since V determines
∂δ/∂t, we find
δ(y) = δ(0) + ω0 t + k0 y , (13.115)
2|e|V0
ω0 = − , (13.116)
h̄
2π
k0 = − (2λ + w)Ba , (13.117)
Φ0
and the current jJx (y) is therefore a plane wave with phase velocity
ω0 V0
v0 = − =− , (13.118)
k0 (2λ + w)Ba
in the y direction. Returning to the analogy between jJx (y) = j(y − v0 t) and the
current distribution that would be produced by a periodic arrangement of n =
Φtot /Φ0 vortices along the y axis in a type II superconductor, we observe that,
under an applied voltage, these vortices flow with a constant velocity in the y
direction.
2. Maxwell’s equation in vacuum is
curl B = μ0 j + ε0 μ0 ∂t E . (13.119)
508 13 Problem Set
In an insulator, we simply replace ε0 → εε0 , where ε is the dielectric constant.

Since here B = 0 everywhere and V = wEx (E uniform), we obtain in the x
direction,
1 ∂V
0 = μ0 jx + εε0 μ0 . (13.120)
w ∂t
Combining with jx = j0 sin δ and ∂δ/∂t = −2|e|V/h̄, this yields
d2 δ
+ ωJ2 sin δ = 0 , (13.121)
dt2
with
2|e|wj0 2|e|w|j0 |
ωJ2 = − = , (13.122)
εε0 h̄ εε0 h̄
which is indeed positive. As can be seen by considering small oscillations δ 1

and linearising (13.115) around zero, ωJ is an angular frequency. The corre-
sponding frequency, called the Josephson plasma frequency, is
.
ωJ |e|w|j0 |
νJ = = . (13.123)
2π π εε0 h
3. The experiment measures the natural oscillation frequency νp of the Joseph-

son junction. A resonance is observed when the imposed external frequency is
equal to the eigenfrequency of the system. The experiment is carried out in the
presence of a direct current I, so the average phase is δ = δ0 determined by
I = I0 sin δ0 , (13.124)
with I0 = j0 L. In the presence of this current, the small natural oscillations of

the system are obtained by setting δ(t) = δ0 + φ(t) in (13.115) with φ(t) small.
We may therefore linearise in φ to obtain
d2 φ
+ (ωJ2 cos δ0 )φ = 0 . (13.125)
dt2
The new angular frequency is thus ωp2 = ωJ2 cos δ0 , related to the current by
2
I
ωp4 = ωJ4 1− , (13.126)
I0
exactly as observed in Fig. P12.2. The abscissa at the origin gives I02 as a func-
tion of temperature. It can be seen from Fig. P12.8 that I02 decreases when the
Fig. P12.8 Temperature 1

dependence of I02 I02
0.8
0.6
0.4
0.2
0
0 1 2 3 4
T (K)
temperature is raised. It thus varies in the same way as the superconducting gap
(T). The latter vanishes at the normal–superconducting transition at T = Tc
so we may estimate that Tc ≈ 3.54 K. The temperature dependence of I02 is
found to be almost linear near Tc with
I02 = α(T − Tc ) . (13.127)

The experiment gives α = 0.50 and Tc = 3.54 K according to Fig. P12.8. We
also know that the superconductor gap (T) ∼ (Tc − T)1/2 near Tc . Hence I0
varies approximately as (T) and should saturate for T Tc , as can be seen
from Fig. P12.8.
Note on Gauge Invariance
We have
∂A
E = −∇V − , (13.128)
∂t
where V is the local electric potential and A is the vector potential. Combining this equation
with the two following equations which hold in a superconductor, viz.,
h̄
λ2 μ0 j = ∇θ − A , (13.129)
q
∂j
E = λ2 μ0 , (13.130)
∂t
we deduce that
∂θ q 2|e|
=− V = V. (13.131)
∂t h̄ h̄
The time derivative of the phase, which is not gauge invariant, is proportional to the electric
potential, itself not gauge invariant. On the other hand, the gauge invariant phase
2
q
δ = θ2 − θ1 − A
h̄ 1
satisfies
510 13 Problem Set

∂δ q 2 2|e| 2
=+ E=− E. (13.132)
∂t h̄ 1 h̄ 1
Being a physical observable, the electric field has to be gauge invariant. In the geometry we
have here, we may use the gauge Ax = 0, and the two observables coincide, but this is not
generally the case. In this problem, we had V = V1 − V2 .
Problem 13: Magnetisation of a Type II Superconductor 511
Problem 13: Magnetisation of a Type II Superconductor
The magnetisation curves of type II superconductors often exhibit marked hystere-

sis. In particular, in superconductors with high critical temperature, for a sample in
the form of a platelet, and with a field applied parallel to the plane of the platelet,
the measured cycle is totally asymmetric with respect to M = 0 (see Fig. P13.1).
1. On the figure, indicate the region where a perfect Meissner effect prevails. How
would you interpret the field for which the absolute value of the magnetisation
suddenly decreases?
To understand one of the mechanisms that opposes the transition to a state of

thermodynamic equilibrium, we consider vortices in a superconductor in a non-
equilibrium situation. We restrict here to type II superconductors for which λ ξ .
Then the second London equation, viz.,
λ2 μ0 curl j(r) + B = 0 , (13.133)
holds for a field B z and the current j(r) in the superconducting region r > ξ . The
core of the vortex parallel to the z direction, of radius ξ , is assumed normal.
Applying flux quantisation with a single quantum Φ0 = h/2e per vortex, in the
limit ξ λ, we find
−1
μ 0 M (10 −4 tesla)
T = 91 K
−2
T = 90 K
−3 YBa 2 Cu 3 O 7 T = 89 K
T = 88 K
−4
0 1 2 3 4 5
μ0 Ha (10 −4 tesla)
Fig. P13.1 Hysteresis cycles for YBa2 Cu3 O7 . Figure courtesy of M.M. Konczykowski (LSI, Ecole
Polytechnique) from Konczykowski, M., Burlachkov, L.I., Yeshurun, Y., Holtzberg, F.: Phys. Rev.
B 43, 13707 (1991). http://link.aps.org/doi/10.1103/PhysRevB.43.13707
512 13 Problem Set
⎧ r
⎪ Φ0
⎪
⎨ 2π λ2 K0 λ , for r ≥ ξ ,
Bz = (13.134)
⎪
⎪ Φ0 ξ
⎩ K0 , for r ≤ ξ .
2π λ2 λ
In the above, K0 (x) is the Bessel function such that K0 (x) (π/2x)1/2 e−x for x 1
and K0 (x) = ln x for x 1.
2. Express the current j(r) in terms of K1 (x) = −dK0 (x)/dx and represent it graph-
ically.
Such a vortex can exist in the superconductor out of thermodynamic equilibrium,

and in particular when the external applied field is zero. Its energy per unit length is

1 2
Wv = B + λ2 μ20 j2 dS , (13.135)
2μ0 S
where integration is over a plane circular surface S perpendicular to the z axis, whose
radius is allowed to tend to infinity.
3. Specify the physical meaning of the two terms in the above integral.
To handle energy questions concerning vortices, it can be shown that the flux quanti-
sation condition can be directly taken into account by modifying the second London
equation (13.133) to
λ2 μ0 curl j + B = Φ0 δ(r)z = V(r) , (13.136)
and reducing the normal

$ core of the vortex to a point. δ(r) is a 2D Dirac delta
function, for which S δ(r)dxdy = 1 for any integration surface S including r = 0.
In this case, the solution of (13.136) for Bz extends the first relation of (13.134) to
r = 0, but in order to correctly take into account the field in the vortex core, we shall
take the value of Bz at r = 0 to be the one given by the second relation of (13.134).
By carrying out vector transformations of (13.135), it can be shown that the latter
transforms to
%
1
Wv = B · V(r)dS + λ2 (B ∧ curl B) · d , (13.137)
2μ0 S C
where d is a length element on the path (C) enclosing S.

4. Show that the integral around the path (C) is zero in (13.137) when the radius Rc
of this path tends to infinity.
5. Deduce the energy Wv of the vortex per unit length.
We now consider two vortices oriented along the z axis and located at r1 = −ax and
r2 = ax.
6. Specify the spatial dependence of the magnetic field induced by these two
vortices.
The energy W2v of the ensemble of two vortices is given by the same equation,
namely (13.137), in which the vorticity is now

V(r) = Φ0 δ(r − ax) + δ(r + ax) z . (13.138)
7. Determine W2v and deduce the interaction energy between the two vortices.
8. From this interaction energy, deduce the force induced per unit length of vortex
1 on vortex 2. Show that this force can be written in the form
f12 = j12 ∧ Φ0 z , (13.139)
where j12 is the current created by vortex 1 at the core of vortex 2. (This force
can be considered as a Lorentz force exerted by the magnetic field of one vortex
on the current created by the other.)
We now consider two vortices with opposite vorticity, i.e., such that the fields at
the core of the vortices at ax and −ax lie along the positive and negative z axes,
respectively.
9. Give the corresponding expressions for B and j. What values do they take on the
plane x = 0? In what direction is the force induced by one vortex on the other?
Consider a semi-infinite superconductor occupying the half-space x > 0, and a

vortex parallel to z located at ax.
10. What boundary conditions do we expect for B and j at x = 0 when no field is
applied outside?
11. Show that these boundary conditions can be obtained if the fields and currents
for x > 0 are those produced by superposing the field created by the vortex at ax
and the one created by its image in the surface located at −ax.
A field Ha z is now applied in the half-space x < 0.

12. Determine the distributions of the field and the current jS in the superconductor
for x > 0 when there is no vortex.
If a vortex is at ax, with an external field Ha , it is subject to two forces: fI due to

the interaction with its image, and fS due to the interaction with the surface currents
associated with penetration of the field Ha . Assume that the latter also results from
the Lorentz force, i.e.,
fS = jS ∧ Φ0 z . (13.140)
514 13 Problem Set
13. Find the forces fI and fS .

14. Assuming that the vortex forms at the surface and is located at a distance a = ξ
from the surface, what is the value Hp of Ha for which the vortex will be pushed
inside the superconductor?
15. Recalling that
Φ0 λ
Hc1 = ln ,
4π μ0 λ2 ξ
compare Hp and Hc1 .
The aim now is to analyse the various branches of the hysteresis cycles observed in
Fig. P13.1.
16. What remarkable behaviour is revealed by Fig. P13.1 for the branches corre-
sponding to decreasing Ha ?
When H > Hp , the vortices penetrate into the superconductor. The measured mag-
netisation contains two contributions. One is MS , associated with surface currents,
and the other is associated with vortices in the material. We assume that the dis-
tribution of vortices that have penetrated is spatially uniform, i.e., uniform average
magnetic induction within the superconductor. In this case, only surface currents
contribute to the measured magnetisation.
17. What can we say about the variation of these surface currents and the variation
of the vortex density along the branches of the cycle examined in question 16?
18. How may we explain this behaviour?
Solution
1. The perfect Meissner effect is observed up to a point Hp that depends on the

temperature T. (These points are clearly visible in Fig. P13.1. They are indi-
cated on Fig. P13.7.) If the magnetisation were totally reversible, we would have
Hp (T) = Hc1 (T). The fact that it is irreversible shows that
Hp (T) ≥ Hc1 (T) . (13.141)
Hp (T) is the field for which vortices can penetrate the bulk of the sample.
2. We have μ0 j = curl B and
⎧
⎨ − 1 ∂Bz i = Φ0 K r i , for r > ξ ,
⊥ 1 ⊥
j⊥ = μ0 ∂r 2π μ0 λ3 λ (13.142)
⎩
0, for r < ξ .
j⊥ and its dependence on r are shown in Fig. P13.2a, b.

3. We have

1
Wv = (B2 + λ2 μ20 j2 )dS .
2μ0 S
The first term here is the energy associated with the magnetic field produced by
the vortex. Indeed, since we are considering a non-equilibrium thermodynamic
state, the vortex is itself the source of a field. The second term is the kinetic
energy of the supercurrents. Indeed, if the number of superconducting electrons
per unit volume is ns and their velocity at r is v(r), we have j(r) = ns ev(r) and

1 2 2 1 m 2 2 2 1 2
λ μ0 j (r) = n e v (r) = mv (r) ns .
2 2 ns e2 s 2
To get from (13.135) to (13.137), we observe that
z j⊥ (r)
Bz
y
j⊥ (r)
r
0
S x ξ λ r
(C)
(a) (b)
Fig. P13.2 (a) Representation of the current j⊥ around the vortex and (b) radial variation of j⊥
516 13 Problem Set
μ0 j = curl B ,
and hence that

λ2 μ20 j2 = λ2 μ0 j · curl B .
Now,
j · curl B = B · curl j + div (B ∧ j) .
We thereby obtain
%
1
Wv = B · (B + λ2 μ0 curl j)dS + λ2 μ0 (B ∧ j) · d
2μ0 S (C)
%
1
= B · V(r)dS + λ 2
(B ∧ curl B) · d , (13.143)
2μ0 S (C)
which is Eq. (13.137) in the question (but this proof was not required).
4. Bz (r) and curl B go as e−r/λ /r for large r, so
% 2π 1 −2r/λ 2π −2r/λ
(B ∧ curl B) · d ∝ e dθ ∼ e −→ 0 .
(C) 0 r r
5. The energy per unit length of the vortex is found by evaluating the first integral
in (13.137), with the result

1 1 Φ0
Wv = B · V(r)dS = B(r) · Φ0 δ(r)zdS = Bz (0) ,
2μ0 S 2μ0 S 2μ0
and hence
Φ02
Wv = K0 (ξ/λ) . (13.144)
4π μ0 λ2
6. The magnetic fields and currents satisfy (13.136), which is linear, so the fields
and currents produced by two vortices are simply additive. With two vortices
located at r1 = −ax and r2 = ax, we thus have

Φ0 |r − ax| |r + ax|
Bz (r) = K 0 + K0 . (13.145)
2π λ2 λ λ

7. With V(r) = Φ0 δ(r − ax) + δ(r + ax) z, choosing an integration surface normal
to the two vortices and containing them, then letting it tend to infinity, the second
term in (13.137) vanishes, while the first gives

Φ02 ξ 2a
W2v = 2K 0 + 2K 0
4π μ0 λ2 λ λ

Φ02 2a
= 2Wv + K0 , (13.146)
2π μ0 λ2 λ
The interaction energy between the two vortices is thus

Φ02 2a
W12 = K0 . (13.147)
2π μ0 λ2 λ
8. The force exerted by the vortex at −a on the one at +a is given by

1 ∂W12 Φ02 2a
f12 = − x=+ K1 x. (13.148)
2 ∂a 2π μ0 λ3 λ
This force is positive in the x direction. The two vortices repel one another. As
can be seen from Fig. P13.3, the force can indeed be expressed in the form
f12 = j12 ∧ Φ0 z , (13.149)
where j12 is the current produced by vortex 1 at the position of vortex 2, accord-
ing to (13.142) of question 2.
9. For two vortices of opposite vorticity,

Φ0 |r − ax| |r + ax|
Bz (r) = K 0 − K0 . (13.150)
2π λ2 λ λ
Note that, in the plane x = 0, we have |r − ax| = |r + ax| and hence Bz (r) = 0.
The symmetry of Fig. P13.4 is such that the total current in the plane x = 0 lies
along the y axis and the
contributions of the two vortices are additive in this
plane. With V(r) = Φ0 δ(r − ax) − δ(r + ax) , we obtain
B z1 Bz 2
y
j 12
Fig. P13.3 Repulsive force
between two vortices with r1
the same vorticity and placed a 0 a f 12 x
symmetrically with respect to
the plane x = 0
518 13 Problem Set
z
Bz2
r1 ′
f 12 r2
j1⊥ a 0 a x
j2⊥ ′
j 12
j
B z1
Fig. P13.4 Situation in the plane x = 0 for two vortices with opposite vorticity, placed symmetri-
cally with respect to the plane x = 0

Φ02 2a
W20 = 2Wv − 2K 0 . (13.151)
4π μ0 λ 2 λ
The interaction energy has changed sign compared with the previous case, and
the two vortices attract one another with a force f12 opposite to the one obtained
in (13.149) and given by
f12 = j12 ∧ Φ0 z . (13.152)
10. For a vortex parallel to the surface of a semi-infinite superconductor, when no

field is applied outside the superconductor, we expect B = 0 at the surface. Sur-
face currents flow in the plane of the surface and can only be oriented along the
y axis if the vortex is parallel to the z axis.
11. We can therefore obtain the solution for the field and current by taking the vortex
and its image at −ax with opposite vorticity. Indeed, we saw in questions 9 and
10 that we obtain Bz = 0 and currents in the plane x = 0 along the y axis. The
solution for Bz (r) is therefore precisely (13.150) for x > 0.
12. The London equation gives (see Fig. P13.5)
x Ha x
B = μ0 Ha exp − , jS = exp − y . (13.153)
λ λ λ
13. The force acting on the vortex due to the interaction with its image was found in
question 9 and given by (13.152). It pushes the vortex toward the surface. The
force due to the applied field Ha is given by
Ha x
fS = jS ∧ Φ0 z = Φ0 exp − x . (13.154)
λ λ
Fig. P13.5 Field and current z

at the surface of the super-
conductor in the absence of μ 0 Ha
vortices
jS
0 λ x
Fig. P13.6 Forces on a vortex z

lying parallel to the surface
of the superconductor, in the Bz
presence of a field B
y
jS
a a
0 x
fI fS
−B z
It pushes the vortex toward the interior of the superconductor. These two forces
thus act in opposite directions (see Fig. P13.6).
14. When a vortex is located at the surface, it can only penetrate the bulk of the
superconductor if the field Ha is large enough for the force fS to exceed the
attractive force of its own image. The penetration threshold Hp for vortices is
thus specified by

Φ0 Hp ξ Φ02 2ξ
exp − = K1 . (13.155)
λ λ 2π μ0 λ3 λ
15. Since ξ/λ 1, we have

2ξ λ ξ
K1 , exp − 1,
λ 2ξ λ
which implies that
Φ0 λ Hc1
Hp = = Hc1 . (13.156)
4π μ0 λξ ξ ln λ/ξ
Hp is thus of the same order of magnitude as the critical thermodynamic field

Hc = (λ/ξ )Hc1 . Owing to their interaction with the surface, vortices forming at
520 13 Problem Set
(4)
0
(3)
−1 (1) (2)
μ0 M (10 −4 tesla)
T = 91 K
−2
T = 90 K
Hp T = 89 K
−3 YBa 2 Cu 3 O 7
T = 88 K
−4
0 1 2 3 4 5
μ 0 Ha (10 −4 tesla)
Fig. P13.7 Hysteresis cycles for YBa2 Cu3 O7
Hc1 cannot therefore penetrate into the superconductor as long as Ha < Hp ∼ Hc .

We say that there is a surface barrier.
16. Beyond Hp , vortices can penetrate and surface currents gradually decrease on
the second branch (2) of the hysteresis cycle in Fig. P13.7. When the field is
reduced on branch (3) of the hysteresis cycle, the magnetisation varies with a
slope almost equal to −1 [branch (3) is almost parallel to the straight line cor-
responding to the perfect Meissner effect]. This goes on until the magnetisation
has almost vanished. Note that branch (4) of the cycle corresponds to virtually
zero magnetisation.
17. If the vortices are uniformly distributed throughout the material, the average
magnetic induction Bint over a volume with elementary dimension greater than
the distance between vortices is almost uniform. The decrease in the surface
currents can be understood from the fact that the magnetic induction varies from
μ0 Ha to Bint over the distance λ. It is this reduction in the surface currents that is
detected by a drop in the measured magnetisation. When the field decreases on
branch (3), the gradient −1 corresponds to a reduction of the screening currents,
but the vortices do not leave the sample. At the end of branch (3), when M = 0,
we have μ0 Ha = Bint and the average magnetic induction is uniform, so there is
then no more surface current. When Ha is reduced, M remains zero and μ0 Ha =
Bint right along branch (4). There the vortices gradually leave the sample.
18. The existence of fI and fS corresponds to a dependence of the vortex energy on
the distance from the surface which has a maximum. It thus constitutes a sur-
face barrier for penetration or expulsion of vortices. When H > Hp , the surface
barrier vanishes and the flux penetrates, but the surface currents fall off.
The function M(Ha ) on branch (2) of Fig. P13.7 is such that the surface barrier
is permanently zero. When Ha is reduced on branch (3), the surface barrier reap-
pears and the flux can no longer be excluded along branch (3). It does indeed
correspond to a decrease in the surface currents without modification of the vor-
tex density. In contrast, on branch (4), when M = 0, the surface currents vanish,
μ0 Ha = Bint permanently, and the vortices are gradually excluded from the su-
perconductor. We return at Ha = 0 to a situation in which almost no vortices are
pinned.
Problem 14: Electronic Structure and Superconductivity of V3 Si 523
Problem 14: Electronic Structure and Superconductivity

of V3 Si8
In the 1960s, relatively high critical temperatures Tc were discovered in supercon-

ducting metal alloys based on vanadium and niobium. Whereas Tc = 5 K for vana-
dium and Tc = 8 K for niobium, Tc is 17 K in V3 Si and 23 K in Nb3 Ge. The best
superconducting coils used today are made from Nb3 Sn (Tc = 18 K).
The aim here is to obtain a qualitative understanding of these high Tc values by
investigating the electronic structure of these compounds. Their crystal structure is
A-15, defined as follows for V3 Si. The Si atoms occupy sites in a body-centered
cubic lattice with parameter a, while the V atoms are distributed along the faces of
the cube at positions

1 1 3 1 1 1 1 3 1 1 3 1
0, , , 0, , , , 0, , , 0, , , ,0 , , ,0 .
4 2 4 2 2 4 2 4 4 2 4 2
1. Sketch a primitive cell of V3 Si, indicating all the atoms of the basis?
2. Is it possible to predict a priori whether the compound V3 Si is metallic or insu-
lating according to a band theory for independent electrons?
It can be shown that the bands at the Fermi level are mainly built up from 3d orbitals
of vanadium (bands resulting from the s levels are broad and contribute little at the
Fermi level). To calculate the d band structure, we apply the LCAO method to these
orbitals and assume that the orbitals of two neighbouring sites are orthogonal.
3. Compare the distances between near-neighbour vanadium atoms located on the
same face and on different faces. Deduce that the V atoms can be described as
forming three series of mutually independent chains in the three space directions,
and that the problem reduces to studying the band structure in a single direction.
We consider a row of V atoms along the z axis.

4. Indicate the corresponding first Brillouin zone.
Assume that the 3d orbitals of each vanadium atom are degenerate with energy E0 .
5. Considering only nearest neighbours, express the dispersion relation E(k) in
terms of the energy E0 and hopping integrals t0 and t1 , assuming that they are
the same for all the 3d orbitals.
6. State and plot the corresponding density of states D(E).
In the following, we take t0 0.
7. Determine the Fermi energy in terms of E0 and t1 .
8 This problem has been designed with F. Rullier-Albenque.

524 13 Problem Set
z dxz z dyz z dxy

y y y
x x x
z z
y y
x x
dz2 dx2 −y2
Fig. P14.1 Spatial representation of the five different 3d orbitals
In fact the hopping integrals t1 are not the same for the different 3d orbitals. Their
spatial representations are shown in Fig. P14.1.
8. Show that, by symmetry, the 3d orbitals fall into three different groups with
increasing hopping integrals aong the chains.
Take the values of the hopping integrals to be t1 /4, t1 /2, and t1 . We still assume that
the integrals t0 are negligible.
9. Describe the new band structure, indicating the atomic orbitals associated with
each subband, and plot the resulting density of states D(E).
10. Indicate the approximate position of the Fermi level expected in each case.
Figure P14.2 shows the temperature dependence of the specific heat of a V3 Si single
crystal.
T (K)
5 10 15 20 25
0.3
C/T (J/mole K2 )
0.2
0.1
0
0 200 400 600 800
T 2 (K2 )
Fig. P14.2 Temperature dependence of the specific heat of a V3 Si single crystal. Adapted from
Muto, Y., Toyota, N., Noto, K., Akustu, K., Isimo, M., Fukase, T.: J. Low Temp. Phys. 34, 617
(1979)
11. What characterises the superconducting transition? Determine the value of Tc

in this sample.
12. How could you deduce the superconducting gap from these data? (Do not
actually do this.) The value found is = 2.4 meV. Can V3 Si be described by
the BCS theory?
13. From these data, find the electronic specific heat Cv = γ T in the normal state.
For vanadium, γ = 8.8 mJ/mole/K2 .

14. Can you suggest an explanation for the high value of Tc in V3 Si based on the
band structure model and these experimental data?
In contrast, Mo3 Ge has a very low Tc = 1.5 K.

15. Is this compatible with the above model?
526 13 Problem Set
Solution
1. The primitive cell is cubic. The basis contains two V3 Si molecules, as shown in
Fig. P14.3.
2. There is an even number of V atoms and of Si atoms, and hence an even number
of electrons in the primitive cell. It is not therefore possible to predict a priori
whether this compound is metallic or insulating using a band theory for indepen-
dent electrons.
3. The distance
√ between near-neighbour vanadium atoms on a given face is a/2,
but (a/4) 6 ≈ 1.22a/2 for atoms belonging to orthogonal faces. The hopping
integrals between vanadium atoms will be greater along the three directions x, y,
and z. We may thus consider that the vanadium atoms form three series of or-
thogonal chains numbered 1, 2, and 3 in Fig. P14.3. These chains can be ob-
tained from one another by symmetry. It suffices therefore to study the electronic
structure along some given direction.
4. As the distance between V atoms is a/2, the Brillouin zone is a straight line
segment [−2π/a, 2π/a] along the z axis.
5. The dispersion relation (given in Chap. 1) is
ka
E(k) = E0 − t0 − 2t1 cos .
2
This band is ten-fold degenerate: (2 spins) × (5 orbitals).
6. The density of states per V atom is, assuming t1 > 0,
1 1
D(E) = ×2×5 ,
2π t1 1 − (E − E + t )2 /4t2
0 0 1
2 1 y
z
1
V
Si
Fig. P14.3 Primitive cell and basis of V3 Si, showing the three linear V chains directed along the
three axes
2.5
D(E) × 2πt1/10
2
1.5
0.5
0
−3 −2 −1 0 1 2 3
EF (E − E0)/t1
Fig. P14.4 Density of states D(E) per V atom, showing the Fermi level
for |E − E0 + t0 | < 2t1 . The two-fold spin degeneracy and five-fold orbital degen-
eracy of the band have been included. This density of states is shown for t0 = 0
in Fig. P14.4.
7. The outermost electronic shells of vanadium are 3d 3 4s2 . There are thus 3 elec-
trons per V atom in the d band. The Fermi energy EF is thus given by
EF
D(E)dE = 3 .
E0 −2t1
This integral is more easily calculated in k space, since the density of states is
$k
constant there. It becomes 10(a/2π ) 0 F dk = 3, or kF = 3π/5a, and then
3π
EF = E0 − 2t1 cos ≈ E0 − 1.18t1 .
10
The position of the Fermi level is shown in Fig. P14.4, where the density of states
is plotted as a function of the energy.
8. Consider for example a chain of V atoms along the z axis. The hopping integrals
t1 get bigger as the overlap of the atomic orbitals along z increases. Examining the
d orbitals we find that the dz2 orbitals, which yield a high probability along the z
axis, correspond to the largest t1 and will therefore form the broadest band. Then
come the dxz and dyz orbitals, which yield significant probabilities in the (x, z)
and (y, z) planes. Finally, the dxy and dx2 −y2 orbitals, which are perpendicular to
the z axis, correspond to the smallest t1 and will thus form the narrowest bands.
9. The band structure now comprises three distinct bands arising respectively from:
• the dz2 orbitals: E(k) = E0 − 2t1 cos(ka/2), with two-fold degeneracy,
• the dxz and dyz orbitals: E(k) = E0 − t1 cos(ka/2), with four-fold degeneracy,
528 13 Problem Set
• dxy and dx2 −y2 orbitals: E(k) = E0 − (t1 /2) cos(ka/2), with four-fold
degeneracy.
The corresponding densities of states are:
2 1
D1 (E) = ,
2π t1 1 − (E − E )2 /4t2
0 1
8 1
D2 (E) = ,
2π t1 1 − (E − E )2 /t2
0 1
and
16 1
D3 (E) = .
2π t1 1 − 4(E − E )2 /t2
0 1
These densities of states are plotted in Fig. P14.5. There are now six singularities
located at E = (E0 ± 2t1 ), E = (E0 ± t1 ), and E = (E0 ± t1 /2).
10. If there were only the bands 1 and 2, we would obtain EF = E0 (half-occupancy).
In the present case, the Fermi level will lie between E0 − t1 /2 and E0 and closer
to E0 − t1 /2 than to E0 . A more careful calculation shows that, when the three
d electrons of vanadium are placed, we begin by filling bands 1 and 2 up to
the energy E0 − t1 /2, which amounts to placing ∼ 0.84 electrons in band 1 and
1.36 electrons in band 2. The Fermi level will thus be determined by placing the
remaining ∼ 0.8 electrons in the three bands starting from the energy E0 − t1 /2.
Owing to the divergence of the density of states of the third band for E0 − t1 /2,
we expect the Fermi level to be very close to this energy. Solving numerically,
we do indeed find EF ≈ E0 − 0.41t1 . The corresponding Fermi level is shown in
Fig. P14.5.
2.5 1 2 3
D(E) × 2πt1/10
1.5
0.5
0
−3 −2 −1 0 1 2 3
EF (E − E0)/t1
Fig. P14.5 Densities of states for the three distinct bands

11. The superconducting transition is revealed by the discontinuity in the specific

heat, implying that Tc is in this case equal to 16.7 K.
12. We expect the specific heat to go as C ∝ exp(−/kB T) in the superconduct-
ing state. Plotting the logarithm of C as a function of 1/T, we should therefore
obtain a straight line with slope /kB . In fact, in the present case this only ap-
plies for the very low temperature data, and we obtain = 2.4 meV. According
to BCS theory, is related to Tc by 2 = 3.52kB Tc . In the present case we
find kB Tc = 1.39 meV, so that 2/kB Tc = 3.45, which is close to 3.52. We may
conclude that V3 Si is well described by the BCS theory.
13. In a metal, we know that the specific heat is the sum of electron and phonon
contributions and takes the form C = γ T + AT 3 . This is indeed what we observe
in Fig. P14.2 since C/T varies linearly with T 2 in the normal state. Extend-
ing this straight line down to T = 0, we obtain the coefficient γ directly as
55 mJ/mole/K2 .
14. The coefficient γ of the electronic specific heat is related to the density of states
at the Fermi level. The fact that γ is much smaller in vanadium than in V3 Si
thus shows that D(EF ) is higher in the latter. This agrees with the simplified
calculation of the band structure, which shows that the Fermi level is located
close to a singularity in the density of states. According to the BCS theory, Tc is
given by

1
kB Tc = 1.14h̄ωD exp − .
V0 D(EF )
We thus expect Tc to increase as D(EF ) increases. This explains the high values
of Tc observed for V3 Si.
Note: It can be shown quite generally that Tc is high when the Fermi level lies
near a van Hove singularity in the density of states. This is the case for the A-15
compounds, e.g., V3 Si, V3 Ga, Nb3 Sn. This is confirmed by the fact that Tc is highly
sensitive to the order of the atoms in chains for this type of compound. If the chains
are disordered, as occurs for instance if the compound is non-stoichiometric, the
singularities in the density of states are broadened, whereupon D(EF ) and hence Tc
decrease very fast.
15. The molybdenum atom has one more electron than vanadium in its outer d shell
and the simplified calculation of the band structure shows that the Fermi level in
Mo3 Ge will be further from the van Hove singularity. We thus understand that
D(EF ) and hence Tc will be smaller.
Problem 15: Superconductivity of NbSe2 531
Problem 15: Superconductivity of NbSe2
NbSe2 is a quasi-2D lamellar material that cleaves easily. The cleavage planes are
almost perfect on the atomic scale and give rise to clean observations by scanning
tunneling microscopy. The structure of the lamellas is shown in Fig. P15.1. The
equilateral triangles with Se at the vertices have side a = 3.45 Å. The thickness of a
lamella is e = 2.75 Å.
1. What is the primitive cell of a lamella?
In one crystalline form called 2H-NbSe2 , the lamellas are organised in such a way
that the Se triangles have alternating orientations as shown in Fig. P15.2. The dis-
tance between lamellas is d = 3.52 Å.
2. Specify a primitive cell of the 3D crystal and calculate its volume.
From an electronic standpoint, the lamellas can be considered weakly coupled, and
the band structure is then very similar to that of an isolated lamella. The atomic
electronic structures of niobium and selenium are 4d 4 5s and 4s2 p4 , respectively.
3. Can we say whether a lamella is metallic or insulating?
The scanning tunneling microscope can measure the electron current j between the
metallic tip and the cleaved surface of the sample as a function of the tip–sample
distance and the voltage V applied to the tip. When the tip is held at a fixed distance
from the surface, we obtain the tunneling characteristic G(V) = dj/dV.
Se
Nb
Fig. P15.1 Structure of one
lamella of NbSe2
Fig. P15.2 Three-dimensional

structure of 2H-NbSe2 show-
ing the alternating stacking
orientations of the Se triangles
532 13 Problem Set
Ba = 0
1
G(V )
–2 0 2 4
Voltage (mV)
Fig. P15.3 Tunneling characteristic of NbSe2 at T = 0.3 K in zero applied field. http://link.
aps.org/doi/10.1103/PhysRevLett.62.214
4. Express the characteristic as a function of the densities of states of the tip and
surface.
NbSe2 is superconducting with critical temperature Tc = 7.2 K. The characteristic

curve observed by tunneling at T = 0.3 K is shown in Fig. P15.3. In the figure, the
conductance is normalised to unity for values of |V| greater than 4 mV. The tungsten
tip has energy-independent density of states for energies over a range of 0.1 eV on
either side of its Fermi energy.
5. How does the density of states compare with the one expected according to BCS
theory?
6. Estimate the width of the superconducting gap.
7. How does it compare with the BCS prediction for the given value of Tc ?
8. Given that there is one electron per Nb in the conduction band, calculate the
London penetration depth λL of NbSe2 (consider the 3D structure).
A magnetic field Ha is applied perpendicular to the cleaved surface of the sample

and the microscope tip moved parallel to the surface at height z. At each point of the
surface, G(V) = dj/dV is measured for two values of the voltage, viz., V = 0 mV
and V = 3 mV. The values of G(0)/G(3) for different x and y are used to produce a
grey-scale image of the surface, where 0 = white and 1 = black. For an applied field
Ba = μ0 Ha less than 0.023 tesla, the image of the sample is completely white.
9. What can you deduce?
For a field Ba greater than 0.02 tesla, spots appear on the image. In particular, for a
field of 1.041 tesla, the image is the one shown in Fig. P15.4.
10. Explain the physical origin of this observation. Would you expect to see the
observed structure?
11. Do an accurate check to see whether the lattice parameter corresponds to the
one expected for the given applied field.
Fig. P15.4 Grey-scale image

of the cleaved surface for
a field of 1.041 tesla. From
Hess, H.F., Robinson, R.B.,
Dynes, R.C., Valles, J.M.,
Jr., Waszczak, V.: Phys. Rev.
Lett. 62, 214 (1989). With
the permission of the Ameri-
can Physical Society (
c 1989
10.1103/PhysRevLett.62.214
6000 Å
In a simplified model, a vortex can be represented by a normal metal tube with

radius equal to the coherence length ξ .
12. What is the tunnelling characteristic G(V) expected at the center of a vortex?
13. Can you use the image to estimate a value for ξ ?
When Ba exceeds 1.83 tesla, the observed image becomes completely black.
14. Can this observation be used to estimate the penetration depth λ? How does it
compare with the value found for λL in question 8?
In a field of less than 0.05 tesla, the vortices are distant from one another and their
electronic structure can be examined more carefully by plotting the characteristics
G(V) as a function of the distance d from the center of the vortex. The results of
such a study are shown in Fig. P15.5, where the characteristics G(V) for distances
d = 0–585 Å are compared with the one obtained for zero field.
It can be shown that, when there is a supercurrent js , the gap in a BCS supercon-
ductor is lowered by a kinematic effect according to
−→ − h̄kF |vs | ,
where vs is the average velocity of Cooper pairs. Recall that the field B around a
vortex has the form Bz (d) = B0 K0 (d/λ), where z is the vortex axis, d the distance to
the center of the vortex, and K0 (x) a Bessel function.
15. Give the expression for the resulting supercurrent js (d). Note that K1 (x) =
−K0 (x) and B0 is determined by the flux quantisation condition.
534 13 Problem Set
5
d (Å)
Ba = 0.05 tesla
G(V ) 0
4 65
130
3 195
260
325
2
390
455
1 520
Ba = 0 585
0
−4 −2 0 2 4
V (mV)
Fig. P15.5 Normalised conductances G(V) are shifted by 0.75 for the curve acquired at 585 Å and
a further 0.25 for each of the other curves. From Hess, H.F., Robinson, R.B., Waszczak, J.V.: Phys.
Rev. Lett. 64, 2711 (1990). With the permission of the American Physical Society ( c 1990 APS).
http://link.aps.org/doi/10.1103/PhysRevLett.64.2711
16. Show that the variation of the gap can be written locally in the form

πξ d
(d) = 1 − K1 .
2λ λ
Over what distance range does this formula apply?

17. Show that, far from the vortex core, the curves of Fig. P15.5 vary in accordance
with this kinematic effect.
Using the curves in Fig. P15.5, we find the values of as a function of d given in
Table 13.3.
Table 13.3 Superconducting gap as a function of the distance d from a vortex core
d 130 195 260 325 390 455 520
– 0.33 0.41 0.47 0.52 0.57 0.61 0.7
18. Given that K1 (x) ∼ 1/x for x 1, find a value for ξ . Deduce a value for λ.
19. Is the curve observed at the vortex core the one expected from question 12?
Solution
1. A primitive cell of the lamella contains one NbSe2 unit. A possible primitive cell
is shown in Fig. P15.6.
2. The alternating orientation of the lamellas is such that a primitive cell of the
3D crystal contains two√NbSe2 units, as shown in Fig. P15.7. The volume of the
primitive cell is v = a2 c 3/2, with a = 3.45 Å and c = 2(2.75 +3.52) = 12.54 Å,
therefore v = 129.25 Å3 .
3. The number of electrons per primitive cell of the lamella in the valence orbitals
is 5 + (2 × 6) = 17, an odd number. According to band theory, this necessarily
implies a metallic state, since at least one band will be partially occupied.
4. Recall that
+∞
j = AT D1 (E)D2 (E + eV) f (E) − f (E + eV) dE .
−∞
For constant D1 (E) = D1 (EF ),

+∞
dj ∂f (E − eV)
= AT D1 (EF ) D2 (E ) dE ,
dV −∞ ∂V
Se
Nb
Fig. P15.6 Possible primitive cell for the NbSe2 lamella
Fig. P15.7 Primitive cell of 3D crystal containing two NbSe2 units

536 13 Problem Set
and for T = 0, this reduces to

dj
= −AT eD1 (EF )D2 (EF + eV) .
dV
5. The density of states observed by electron tunneling is similar to the one shown
in Chap. 7 for Sn (see Fig. 7.10), and this satisfies the BCS theory. There is
indeed a range of values |V| 0.7 meV for which D2 (EF + eV) = 0, and this
corresponds to the gap. However, the singularity at eV = is not as sharp as
would be expected from BCS theory at T = 0:
⎧
⎨ D(E ) |E − EF |
F , for E − EF > ,
DBCS = [(E − EF )2 − 2 ]1/2
⎩
0, for E − EF < .
This may be because the measurement was made at nonzero temperature

(T = 0.3 K). However, in the case of Sn, for similar values of T and , exper-
imental and theoretical values correspond to a much smaller broadening of the
transition, as shown by the comparison done in Fig. P15.8. This broadening of
the density of states singularities thus has a distinct origin, such as a distribution
of gap values in NbSe2 . According to Fig. 7.10, if we take into account a thermal
broadening ∼ 0.1 meV, we deduce that at zero temperature,
min (0) = 0.7 ± 0.05 meV , max (0) = 1.1 ± 0.05 meV .
Note: The structure of the NbSe2 lamellas in the ab plane is not symmetrical.
This anisotropy causes an anisotropy of the superconducting properties which may
explain the anisotropy of the gap, why the vortex lattice is not hexagonal, and so on.
7. In BCS theory, we expect
2Δ max
B=0
1
G(V )
2 Δ min
−2 0 2 Voltage (mV)
Fig. P15.8 Comparison of the data for NbSe2 with expectation for a thermally broadened BCS
density of states
1.14 = 2kB Tc , hence = 1.75kB Tc .
For Tc = 7.2 K, this would correspond to = 1.75 × 25/300 = 1.05 meV. There
is therefore only a small discrepancy between the average gap 0.9 ± 0.05 meV
and the value expected from the BCS theory.
8. We have λ2L = m0 /ne2 μ0 , with n = (2/129.25) × 10+30 /m3 , which gives
λL = 428 Å .
9. If the image is completely white, this means that the superconducting gap is
present at all points of the sample. There is no normal region and we thus have a
perfect Meissner effect for μ0 Ha < 0.023 tesla. Therefore,
μ0 Hc1 = 0.023 tesla .
10. The appearance of black spots tells us that the superconducting state is mod-
ified in the black regions. The resulting lattice means there is a mixed state.
The observed structure corresponds to a vortex lattice. This lattice appears to be
hexagonal in Fig. P15.9. However, if we measure the lattice spacings in direc-
tions 1, 2, and 3, we find a triangular lattice, but not perfectly hexagonal. Indeed
2
1
3 h
6000 Å
Fig. P15.9 Vortex lattice for the mixed state

538 13 Problem Set
the segments shown by dark lines, which correspond to 10 lattice constants are
such that 10a1 → 81 mm, 10a2 → 87 mm, and 10a3 → 87 mm.
11. The magnetic induction is such that BS = Φ0 if S is the area of the primitive cell
of the vortex lattice. To determine S, we can consider the blue parallelogram
in Fig. P15.9, for example, which corresponds to 56 cells of the lattice and has
area
69 × 50.5
bh = 36 × 106 Å2 , and then S = 0.1994 × 106 Å2 ± 0.5% .
(106)2
This corresponds to B = 2.07/1.994 = 1.038 ± 0.005 tesla, which is perfectly

compatible with the applied field of 1.041 tesla.
12. The characteristic for a normal region should be G(V) = 1.
13. Black regions correspond approximately to a region where G(V) 1. The aver-
age diameter of a black spot is 3 ± 0.3 mm, which corresponds to
ξ = 85 ± 9 Å .
14. When the sample gives a black image, it is completely in the normal state and
μ0 Hc2 1.83 tesla .
We thus have Hc2 /Hc1 = (λ/ξ )2 , or (λ/ξ )2 = 1.83/0.023 = 79.5, which gives
λ ∼ 760 ± 150 Å ,
slightly greater than λL . This is generally the case, as explained in Chap. 7.

15. The flux quantisation condition yields
Φ0
B0 = ,
2π λ2
and μ0 j = curl B yields j⊥ = −μ−1

0 ∂Bz /∂r, hence

Φ0 d
js (d) = K1 .
2π μ0 λ3 λ
16. We have js = ns evs = npairs qvs . Hence

h̄kF Φ0 d
(d) = 1 − K1 ,
ns e 2π μ0 λ3 λ
5 d (Å)
G(V ) B = 0.05 tesla 0
4 65
130
195
3
260
325
2 390
455
1 520
B=0 585
0
−4 −2 0 2 4
V (mV)
Fig. P15.10 Characteristic G(V) near a vortex
with
m0 h hvF
λ2 = , Φ0 = , ξ= ,
ns e2 μ0 2e π
and therefore

πξ d
(d) = 1 − K1 .
2λ λ
This relation is only valid outside the vortex core, i.e., for d > ξ .
17. It is clear from Fig. P15.10 that the gap decreases as we approach the vor-
tex core, for d decreasing from 585 to 195 Å. It even seems to disappear for
d ∼ 130 Å.
18. For d/λ 1, we have
πξ 1
= 0 1 − .
2 d
Figure P15.11 shows as a function of 1/d. Taking into account the condition
d/λ 1, i.e., small values of d, the best agreement with the above equality
is given by the straight line in Fig. P15.11. It has slope p = 78 mV Å, which
permits the estimate
2 2 78
ξ= p= Å = 71 Å .
π 0 π 0.7
From (λ/ξ )2 = 79.5, we therefore deduce
λ = 630 Å .
540 13 Problem Set
Fig. P15.11 Variation of the Δ (mV)

measured gap with the dis- 0.7
tance d to the vortex core
0.6
0.5
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6
10 3 /d ( Å) −1
19. At the center of the vortex, we would expect G(V) = 1, so the fact that we
observe a peak in the density of states at the Fermi level suggests that, in the
vortex core, the electrons do not behave altogether as they would in a normal
metal.
Note: The ‘normal’ electrons are confined within the vortex core and interact with
the superconducting interface. This leads to a closer spacing of the levels near the
Fermi level than in the normal metal. However, it can be checked that the integral
of the density of states as a function of energy is equal to unity, as the light and dark
regions have the same area in Fig. P15.10. This indicates that the number of electron
states accessible in the vortex core remains the same as in the normal state.
Problem 16: Magnesium Diboride: A New Superconductor? 541
Problem 16: Magnesium Diboride: A New Superconductor?
MgB2 is an extremely simply binary compound, to be found on the shelf of al-

most any chemistry laboratory. It is quite astonishing that its superconductivity,
with abnormally high Tc = 39 K for an apparently conventional material, was
only discovered in 2001, while research groups have been carrying out systematic
searches since the end of the 1970s to find the maximum possible value of Tc for
electrotechnical applications. The aim here will be to present the characteristics of
this material and its superconductivity, determined in the two years following its
discovery.
16.1: Atomic and Electronic Structure of MgB2
The lamellar atomic structure of this compound is depicted in Fig. P16.1. It com-
prises alternating honeycomb planes of B and planes of Mg. The latter are arranged
vertically above the centers of the B hexagons.
1. Specify the 3D Bravais lattice of MgB2 , together with a primitive cell and its
basis.
2. What is the reciprocal lattice of this structure and its first Brillouin zone?
3. Can we deduce immediately whether this compound is insulating or conducting
according to a band theory for independent electrons?
4. Experiment tells us that MgB2 is a conductor. What can we conclude regarding
its band structure, if such a band theory is applicable?
In this structure, Mg is almost in the form Mg2+ . Its core levels and the 1s2 levels
of B form narrow fully occupied bands.
5. If the role of the Mg were simply to donate its two valence electrons to B, what
well known band structure would the valence electrons of the boron plane adopt?
Mg
Fig. P16.1 Lamellar atomic

structure of MgB2
542 13 Problem Set
The calculated band structure for these valence electrons is shown for several
directions of reciprocal space in Fig. P16.2. The directions can be identified in
Fig. P16.3, which shows the Fermi surface in a unit cell of the reciprocal lattice.
6. Plot the first Brillouin zone of question 2 on Fig. P16.3, taking Γ1 as zone center.
7. The Fermi surface comprises four sheets corresponding to the last occupied
states of different bands. Identify the curves E(k) in Fig. P16.2 giving rise to
the different sheets of the Fermi surface and associate them pairwise.
8. How could one find the velocity of an electron corresponding to a Bloch state
represented by a point on the Fermi surface?
9. Identify the two sheets of the Fermi surface corresponding to electrons propagat-
ing in the B plane and the two sheets corresponding to bands with a more 3D
character.
The various band are built up from σ orbitals hybridising the boron 2s and 2px , 2py
orbitals on the one hand and the π orbitals built up from the boron 2pz orbitals on
the other.
4.0
2.0
Fig. P16.2 Calculated band EF
0.0
Energy (eV)
structure for the Mg valence

electrons for several direc-
tions of reciprocal space. −2.0
From Ann, J.M., Pickett,
W.E.: Phys. Rev. Lett. 86, −4.0
4366 (2001). With the per-
mission of the American −6.0
APS). http://link.aps.org/doi/ −8.0
10.1103/PhysRevLett.86.4366 Γ M K Γ A L
Fig. P16.3 Fermi surface in

a unit cell of the reciprocal
lattice. From Kortus, J.,
Mazin, I.I., Belachenko, K.D.,
Antropov, V.P., Boyer, L.L.:
Phys. Rev. Lett. 86, 4656
RevLett.86.4656
10. What are the dominant features of the different sheets of the Fermi surface?
11. What are the differences with the band structure of graphene?
12. Can you explain the effect of the Mg2+ ions on the band structure?
16.2: Superconductivity of MgB2
a. Zero-Field Properties
Electron tunneling experiments have been carried out to investigate the supercon-
ducting properties of MgB2 . When a single crystal of MgB2 is cleaved along a
plane parallel to the B planes, the tunneling characteristics dI/dV parallel to c
were observed at different temperatures and the results plotted in Fig. P16.4. For
improved visibility, the curves are shifted vertically by 0.5 units for each tempera-
ture, while the conductivity is always equal to unity for V = −15 mV.
1. What electronic states are probed by this type of experiment? What quantity can
be deduced?
2. Do we observe the value expected from BCS theory given that Tc = 39 K?
A tunneling experiment on a sample cleaved along the plane (a, c) gives a com-
pletely different result at low temperatures, as can be seen from Fig. P16.5.
3. Explain why certain electronic bands do note contribute to tunnelling when the
tip is parallel to the c axis. What causes the modification observed in Fig. P16.5?
4. What is the physical origin of the two new peaks?
6
38.8 K
34.8 K
5
29.9 K
Normalised conductivity
25.0 K
4
20.0 K
15.0 K
3
10.0 K
Fig. P16.4 Tunneling charac- 7.5 K

2
teristics dI/dV parallel to c in 5.0 K
an MgB2 single crystal at
1.9 K
different temperatures. From 1
Eskildsen, M.R., Kugler, M., 320 mK
Lévy, G., Tanaka, S., Jun, J.,
Kazakov, S.M., Karpinski, J., 0
−15 −10 −5 0 5 10 15
Fisher, O.: Physica C 385, 169
(2003). c Elsevier V (mV)
544 13 Problem Set
σ = dI/dV (G Ω−1)
6 1.2
4 0.8
2 0.4
0 0
−20 −10 0 10 20
V (mV)
Fig. P16.5 Low temperature tunnelling characteristic dI/dV parallel to the plane (a, c) in an MgB2
single crystal. From Eskildsen, M.R., Kugler, M., Lévy, G., Tanaka, S., Jun, J., Kazakov, S.M.,
Karpinski, J., Fisher, O.: Physica C 385, 169 (2003).
c Elsevier
b. Superconductivity in an Applied Field
When a magnetic field is applied along the c axis of the sample, the tunnelling char-
acteristic dI/dV along c depends on the spatial position of the tip over the sample.
For example, for an applied field of 0.05 tesla, we find that dI/dV does not depend
on the voltage V at certain points of space. Moving the tip a distance r from such
a point, we find that the characteristic dI/dV varies as shown in Fig. P16.6. In this
figure, the curves dI/dV are shifted vertically by an amount proportional to the tip
displacement and the total distance investigated is 250 nm.
5. What produces this result? What can you conclude regarding the superconduct-
ing characteristics of MgB2 ?
250 nm
−10 −5 0 5 10
V (mV)
Fig. P16.6 Voltage dependence of dI/dV as the tip moves away from a point where the character-
istic is independent of V. From Eskildsen, M.R., Kugler, M., Tanaka, S., Jun, J., Kazakov, S.M.,
Karpinski, J., Fisher, O.: Phys. Rev. Lett. 89, 187003 (2002). With the permission of the American
Physical Society (c 2002 APS). http://link.aps.org/doi/10.1103/PhysRevLett.89.187003
6. What characteristic length of the superconducting state can be deduced from this
experiment? Estimate this length. Is it compatible with the BCS theory, given that
the average velocity of the electrons at the Fermi level is |vF | = 5 × 105 m/s?
In a higher applied field, we find that many similar regions appear. Figure P16.7
maps the value of dI/dV taken at V = 0 across the sample surface. The value of
dI/dV is represented by a scale from blue to red.
7. Given that the applied field is B = 0.5 tesla, can you determine the size of the
region in the figure?
Another experimental method for studying the influence of a magnetic field on
the properties of a superconductor is neutron diffraction. The field is applied
perpendicularly to the surface in the direction of the c axis of a single-crystal sample.
A monokinetic neutron beam produces diffraction peaks which are used to recon-
struct the reciprocal lattice plane perpendicular to the c axis, as shown in Fig. P16.8
(left) for B = 0.5 tesla. The direct beam that would end up at the center of the figure
is blocked beyond the sample and hence not detected.
8. Analyse these experimental results. Are the observations compatible with the
STM results? The neutrons used here have energy corresponding to a tempera-
ture of 9.5 K. What is the angle of diffraction for the observed spots?
9. What other characteristic length of the superconducting state can in principle
be ascertained through this experiment?
10. How could it be determined?
In an applied field of 0.9 tesla, the diffraction pattern changes to the one shown in
Fig. P16.8 (right).
11. What real space feature does this experimental result reflect?
12. Can you suggest the origin of the observed effect by considering the electronic
band structure investigated in 16.1?
Fig. P16.7 Map of dI/dV

taken at V = 0 across the sam-
ple surface. From Eskildsen,
M.R., Kugler, M., Tanaka,
S., Jun, J., Kazakov, S.M.,
Karpinski, J., Fisher, O.:
RevLett.89.187003
546 13 Problem Set
Fig. P16.8 Neutron diffraction patterns. Reciprocal lattice plane perpendicular to the c axis. Left:
B = 0.5 tesla. Right: B = 0.9 tesla. From Cubitt, R., Eskildsen, M.R., Dewhurst, C.D., Jun, J., Kaza-
kov, S.M., Karpinski, J.: Phys. Rev. Lett. 91, 047002-1 (2003). With the permission of the Ameri-
can Physical Society ( c 2003 APS). http://link.aps.org/doi/10.1103/PhysRevLett.91.047002
Solution
Atomic and Electronic Structure of MgB2
1. The boron planes have exactly the same honeycomb structure as the graphene
sheets making up the 3D structure of graphite. The 2D Bravais lattice of
graphene can be specified by the two vectors a et b in Fig. P16.9a. In the or-
thonormal frame (ex , ey , ez ), they are given by
√ √
3a a 3 3a a 3
a = ex − ey , b = ex + ey .
2 2 2 2
The corresponding primitive cell of the honeycomb structure contains 2 boron
atoms at the points (0, 0) and (2/3, 2/3) relative to the frame (a, b). Graphite, in
which there is no atomic plane between the graphene sheets, has a hexagonal
3D Bravais lattice obtained by adjoining the vector c = cez . This is also the
Bravais lattice (a, b, c) of MgB2 . In addition to the two boron atoms at (0, 0, 0)
and (2/3, 2/3, 0), the primitive cell contains an Mg at (1/3, 1/3, 1/2), as shown
in Fig. P16.9b.
2. The reciprocal lattice of the 2D hexagonal lattice is also hexagonal√with axes
a∗ , b∗ rotated by 90◦ with respect to a, b, where |a∗ | = |b∗ | = 4π/a 3. Its first
Brillouin zone is a hexagon centered on the Γ point (see Fig. P16.10a). The
reciprocal lattice of the 3D hexagonal lattice is obtained by adjoining the vector
c∗ = (2π/c)ez . Its first Brillouin zone is a cylinder with hexagonal base centered
on the Γ point (see Fig. P16.10b).
3. Magnesium is bivalent, so has an even number of electrons, like B2 . The primi-
tive cell thus contains an even number of electrons. According to a band theory
for independent electrons, it may be insulating or conducting, depending on the
band width and occupancy.
y
a
ey b
c
Mg
ex x
a B
b
a
(a) (b)
Fig. P16.9 (a) Two-dimensional Bravais lattice of graphene. (b) Three-dimensional hexagonal
Bravais lattice of MgB2
548 13 Problem Set
a∗ b
c∗
b∗
a
Γ
Γ
(a) (b)
Fig. P16.10 (a) Reciprocal lattice of the 2D hexagonal lattice showing axes a∗ , b∗ rotated by 90◦
with respect to a, b and the hexagonal first Brillouin zone centered on the Γ point. (b) Reciprocal
lattice of the 3D hexagonal lattice showing the vector c∗ = (2π/c)ez and cylindrical first Brillouin
zone with base centered on the Γ point
4. The fact that MgB2 is metallic necessarily means that at least two bands overlap,
and hence that there are at least two partially occupied bands at the Fermi level.
This gives rise to two independent sheets for the Fermi surface.
5. If the role of the Mg was simply to donate its 2 valence electrons to the B,
there would be 4 valence electrons per boron atom, and we would recover pre-
cisely the same situation as in graphene. The two band structures would be
identical.
6. Part of the first Brillouin zone is shown by the white lines in Fig. P16.11, where
a second primitive cell, identical to the one in Fig. P16.3 has been represented
below the point Γ1 . The part shown corresponds to only one third of the first
Brillouin zone. The rest is obtained by two rotations, through 120◦ and −120◦
Fig. P16.11 Fermi surface in

two unit cells of the recipro-
cal lattice. White lines repre-
sent part of the first Brillouin
zone
4.0
2.0
0.0 EF
Energy (eV)
−2.0
−4.0
−6.0
−8.0
Γ M K Γ A L
Fig. P16.12 Calculated band structure for the Mg valence electrons for several directions of recip-
rocal space. The bands of Fig. P16.2 are identified here by choosing colours which correspond to
the sheets of the Fermi surface shown in Fig. P16.11
about the c∗ axis. In fact, the representation here could have been restricted to
one quarter of the volume shown, given the symmetries of the primitive cell.
7. Following the curves E(k) in Fig. P16.12, it is easy to identify those that reach
the value EF at definite points of the line segments Γ K, Γ M, and AL. One can
then identify the bands indicated by continuous green and dark blue curves in
Fig. P16.12 which give rise to the green and blue Fermi surfaces lying parallel
to the c∗ axis in Fig. P16.11. A light blue band cuts the Fermi level near the
point M on Γ M and near the point K on Γ K. It corresponds to the blue Fermi
surface forming a ring around the point Γ1 in Fig. P16.11. Finally, the fourth
band (continuous red curve) cuts the Fermi surface near L on the axis AL. This
corresponds to the red Fermi surface shown by a ring around the c∗ axis. In the
first Brillouin zone, it decomposes into two half-rings, symmetrical with respect
to the horizontal plane passing through Γ1 .
8. The velocity of an electron corresponding to a Bloch state represented by a point
k of the Fermi surface is given by
v = h̄−1 ∇k E .
It is thus normal to the constant energy surface at the point corresponding to the
given Bloch state.
9. The blue and green sheets with almost cylindrical shape parallel to the c∗ axis
are such that the velocity of electrons at the Fermi level in real space is per-
pendicular to the c axis and hence lies in the plane spanned by a and b. Con-
sequently, these bands have quasi-2D behaviour at the Fermi level. Note that
the conductivity of these bands is hole conductivity. In contrast, the other bands
550 13 Problem Set
have electron velocities at the Fermi level which point in all space directions,
and thus exhibit more 3D behaviour. One of these bands, the blue one, corre-
sponds to holes, while the red one corresponds to electrons.
10. The bands giving rise to the quasi-cylindrical Fermi surfaces correspond to
carriers propagating in the plane (a, b). They are therefore mainly formed by
hybridisation of the spx et spy orbitals. The 3D bands contain a dominant
contribution from the π orbitals built up from the pz orbitals, which can be
hybridised in the plane and also give rise to an interplane hopping integral, ulti-
mately responsible for the 3D nature of the bands.
11. For graphene, as discussed in Sect. 3.3.4, the spσ bands have low energy and
do not reach as far as the Fermi level. Likewise, it was noted that bonding and
antibonding bands, π and π ∗ , respectively, meet at the Dirac points. Regarding
MgB2 , it is clear from Fig. P16.2 that these two bands also meet at K. However,
in MgB2 , the Dirac point is totally involved in excited states and plays no role
at the Fermi level. Likewise, for graphite, the 3D interactions make the bands
overlap, but the Fermi level remains close to the Dirac point. The electronic
properties of graphite are thus close to those of hole-doped graphene (see the
explanation below).
12. The Mg2+ ions produce an attractive Coulomb potential for the electrons.
Since the π electrons are closer to the Mg than the spσ electrons, they are
more strongly attracted by the Mg2+ ions, so their energy decreases. This ex-
plains the lower energy levels of the π bands as compared to the spσ bands
when MgB2 is compared with graphene. Likewise, the antibonding π ∗ bands
have higher electron density close to the Mg2+ than the bonding π bands.
Comparing with graphene, this explains why the antibonding π ∗ bands have
lower energy than the bonding bands. Furthermore, the Coulomb potential of
the Mg2+ ions attracts the π electrons toward the Mg, and this increases the
hopping integrals between π orbitals along the c axis. This explains the more
3D nature of the π bands as compared with graphite, where they retain a 2D
character.
Calculated Band Structures of MgB2 and Graphite
The band structures of MgB2 and graphite can be directly compared in Fig. P16.13. The bond-
ing spσ bands, shown by white circles, are in fact very similar in the two electronic structures,
but lower in energy in graphite. The bonding π and antibonding π ∗ bands of MgB2 , shown by
dotted curves, are not so different from those of graphene either, and they do indeed meet at the
K point. But in contrast to graphite, the bonding and antibonding bands of B overlap, a further
reason for MgB2 to be conducting. We thus have two π contributions at the Fermi level: the
red and blue 3D bands. If the Fermi level of MgB2 is raised by electron doping, for example
by partially substituting Al3+ for Mg2+ , we may eliminate the holes in the spσ bands, but the
π bands will remain metallic.
4.0
2.0
MgB2
0.0 EF
Energy (eV) −2.0
−4.0
−6.0
−8.0
Γ M K Γ A L
4.0
2.0 Graphite
0.0 EF
Energy (eV)
−2.0
−4.0
−6.0
−8.0
Γ M K Γ A L
Fig. P16.13 Calculated band structures of MgB2 (upper) and graphite (lower). From Ann,
J.M., Pickett, W.E.: Phys. Rev. Lett. 86, 4366 (2001). With the permission of the American
Physical Society (
c 2001 APS). http://link.aps.org/doi/10.1103/PhysRevLett.86.4366
Superconductivity of MgB2
a. Zero-Field Properties
1. In an electron tunneling experiment, an electron is injected into a state acces-

sible to the material, or an electron is extracted from an occupied state. Since
electrons are paired in a superconductor, such an experiment thus probes excited
electron states (or excited holes). In a superconductor, electrons can only be in-
jected from a normal metal in states excited above the superconducting gap. It
is therefore this gap that gets measured in such experiments.
2. In the experiment considered in Fig. P16.14, the band gap is clearly visible
at low temperatures. It gradually closes as the temperature is raised. The value
measured at zero temperature in Fig. P16.14 is = 2.8 meV. The energy depen-
552 13 Problem Set
Fig. P16.14 Tunneling char-

acteristics dI/dV parallel to 6
c in an MgB2 single crystal at 38.8 K
different temperatures 34.8 K
5
29.9 K
Normalised Conductivity
25.0 K
4
20.0 K
15.0 K
3
10.0 K
7.5 K
2
5.0 K
1.9 K
1
320 mK
2Δ π
0
−15 −10 −5 0 5 10 15
dence of the excited states does indeed match the prediction of BCS theory (as
does the observed temperature dependence). However, in BCS theory, the value
of is directly related to Tc by = 1.75kB Tc . For the value Tc = 39 K obtained
by resistivity and Meissner effect, one should obtain a gap 6 meV, twice the
observed value. BCS theory would not therefore appear to be completely valid.
3. When electrons are injected with the tip parallel to the c axis, their velocities are
mainly in the direction of the c axis. This will therefore be much less sensitive
to electron states with velocities in the plane spanned by a and b, hence much
less sensitive to electrons in the pσ orbitals. It would therefore seem that the
modification observed in the excited states when the tip is parallel to the plane
(a, b) in Fig. P16.15 comes from the contribution of these pσ electrons.
2 Δ pσ
2Δπ
a o
)
−1
6 1.2
σ = dI /dV (G Ω
4 0.8
2 0.4
0 0
−20 −10 0 10 20
Fig. P16.15 Low temperature tunneling characteristic dI/dV parallel to the plane (a, c) in an MgB2
single crystal
Note: This argument implicitly assumes that tunneling conserves k , the momen-
tum component parallel to the sample surface. This is true for an elastic tunneling
process, i.e., without energy loss, which have been the only ones considered in the
lecture notes. However, inelastic processes can transfer electrons into Bloch states
with v perpendicular to c.
4. The two peaks appearing in Fig. P16.15 are therefore excited states that we
may try to associate with electronic states in the pσ bands. The gap thus differs
between the two types of electronic state! It therefore seems that the conden-
sation energy of electrons in the superconducting state differs for pσ and π
electrons. The gap obtained for the former in Fig. P16.15 is pσ = 7.7 meV,
whereas an isotropic value of π = 2.8 meV is obtained for the other gap. It is
the average condensation energy of the two types of electron that gives rise to
the overall superconductivity of the material. The average value of is in better
agreement with the one expected on the basis of BCS theory.
b. Superconductivity in an Applied Field
5. The constant characteristic dI/dV corresponds to the one observed in the normal
material. There are therefore normal regions in the superconducting material.
This is what happens in a vortex. But if there are vortices, this means that MgB2
is a type II superconductor.
6. In Fig. P16.6, when the tip moves away from a vortex, the tunneling characteris-
tic reaches a minimum at V = 0 which decreases gradually until dI/dV becomes
identical to the characteristic observed in zero field. The amplitude of the vari-
ation at V = 0 corresponds to a gradual increase in the density ns of supercon-
ducting pairs in the material as r increases. The dependence of the minimum on
r, as plotted in Fig. P16.16, can be used to estimate the coherence length
dI/dV
(V = 0)
1.0
Fig. P16.16 Voltage depen- 1.8

dence of dI/dV as the tip
moves away from a point
1.6
where the characteristic is
independent of V. From
Eskildsen, M.R., Kugler, M., 0.4
Tanaka, S., Jun, J., Kazakov,
S.M., Karpinski, J., Fisher, O.: 0.2
2ξ
(2002). With the permission 0.0
of the American Physical −150 −100 −50 0 50 100 150
Society (
c 2002 APS) Position (nm)
554 13 Problem Set
ξ ≈ 500 Å .
BCS theory would imply
h̄vF h̄vF
ξ= = .
π 1.75π kB Tc
With vF = 5 × 105 m/s and Tc = 39 K, we obtain ξ = 180 Å. Since the measured
value of ξ is greater by a factor of about 2, this implies that it is associated with
a small value of , close to the value of the gap π .
7. In a high field, we see that the vortices arrange themselves into an ordered
hexagonal structure with parameter a . Since the flux is quantised and has the
value Φ0 per primitive cell of the vortex lattice, the macroscopic magnetic in-
duction, which is a spatial average, is B = Φ0 /S, where S is the area of the
primitive cell of the hexagonal lattice, i.e.,
√
2 3 Φ0
a = .
2 B
It follows that
a = 69 nm .
By measuring 3a to within 3% in Fig. P16.7, we can determine the dimensions

of the figure, viz., 215 nm × 215 nm to within 3%.
8. The diffraction peaks observed in Fig. P16.8 (left) correspond to neutron diffrac-
tion by the magnetic field distribution. We thereby directly observe the 2D
reciprocal lattice of the field distribution. Since the latter is a hexagonal lattice
of side a , it has a reciprocal lattice of side
√
a∗ = 4π/a 3 .
From the figure we find 2a∗ = 0.0218 Å−1 to within 3%, which yields
a = 665 ± 20 Å ,
agreeing with the direct STM observation of the vortex lattice for the same field
B = 0.5 tesla. The angle of diffraction is
2d sin θ = λ ,
√
where d = a 3/2 = 576 ± 20 Å is the distance between lattice rows parallel to
one of the sides of the lattice triangle, and λ is the de Broglie wavelength of the
neutrons, i.e., λ = h/p = h/(2mn ε)1/2 . This gives

λÅ = 0.286(εeV )−1/2 .
For energy ε corresponding to 9.5 K, ε = 0.819 meV and λ = 100 Å. We then
have
100
sin θ ≈ θ = 5◦ .
1152
Note: To see this diffraction, the sample thus had to be tilted at an angle θ in order
to satisfy the Bragg condition, and this for various directions in reciprocal space
corresponding to nearest neighbours.
9. Neutrons have spin 1/2 and are deflected by the magnetic field they encounter
along their trajectory. Diffraction is thus due to the magnetic field distribution
between vortices. This clearly has the same periodicity as the superconducting
pair distribution and corresponds to the same real space lattice. This is indeed
borne out. On the other hand, the intensities of the diffraction peaks are de-
termined by the Fourier transform of the magnetic field distribution, which is
directly related to the penetration depth λ.
10. We may thus estimate λ by measuring the intensity of the Bragg peaks and
hence the structure factor of the field distribution. When B μ0 Hc1 , which is
the case here, the lattice spacing is not very large compared with ξ , and B varies
only slightly between vortices. In this case, the structure factor falls off rapidly
with distance in reciprocal space. This explains why only those Bragg peaks
close to the origin are actually observed. In the present case, λ was determined
by studying the variation of the intensity of the Bragg peak for different values
of the applied field and a value of λ of the order of 1,000 Å was obtained.
11. For B = 0.9 tesla, we still have a hexagonal reciprocal lattice in Fig. P16.8
(right), and its unit cell grows, which does indeed imply a smaller unit cell in the
real space vortex lattice. This is to be expected, since the quantisation
√ condition
implies that a ∝ B−1/2 . We do indeed find that a is multiplied by 9/5 = 1.34.
However, the orientation of the reciprocal lattice has changed between 0.5 and
0.9 tesla. The same therefore goes for the vortex lattice in real space.
12. In the neutron experiment considered here, the vortex lattice forms a large sin-
gle crystal with fixed orientation relative to the crystal axes. There is therefore
an anisotropic energy term favouring the observed choice of orientation. The
rotation of the vortex lattice in high applied fields indicates that this anisotropy
term changes with the applied field. This is related to the fact that the supercon-
ductivity gets contributions from both the pσ and π bands. As the superconduc-
tivity of the π band corresponds to the small superconducting gap, it has lower
condensation energy and is more easily destroyed for increasing magnetic field
than the superconductivity due to pσ . The superconductivity is associated with
both types of electron in zero field, but gradually becomes dominated by the pσ
contribution. The anisotropy term associated with the latter favours a different
orientation of the vortices relative to the crystal axes.
556 13 Problem Set
Further Notes: In fact the experiment shows a gradual rotation of the axes of the
vortex lattice and a gradual change in the values of λ and ξ between 0.5 and 0.9 tesla.
The value of ξ corresponding to the pσ bands is less than the one measured in a
weak field for the π band. The value of Hc2 is also higher. The existence of differing
superconducting gaps for different bands is a novel feature which has been revealed
by the study of MgB2 and is nowadays quite commonly found in multiband metallic
systems.
The fact that the vortex lattice is linked to the crystal lattice is in itself also a novel feature.
In an isotropic superconducting material, the vortex lattice would in principle have no reason
to depend on the crystal lattice. Vortices generally appear in a ‘polycrystalline’ form, because
order nucleates at several independent points of the sample, and with random orientations.
Since the many vortex crystals that form are distributed randomly, their Bragg spots have a
single distance from the origin and a random orientation. They thus form a diffraction ring for
each lattice distance.
The existence of an anisotropic term can be understood qualitatively when we realise that
the energy term associated with the field penetration can be minimal for certain configurations
of the current lines relative to the crystal axes. In materials for which the coherence length
ξ is small, these terms are of increasing importance close to Hc2 when the distance between
vortices becomes of the same order of magnitude as ξ . In MgB2 , since the crystal lattice itself
has hexagonal symmetry, only the orientation of the vortex lattice relative to the crystal lattice
can be affected. In cubic materials, this can lead in some cases to the stabilisation of a square
vortex lattice in strong fields, rather than a hexagonal one.
Problem 17: Electronic Properties of La2 CuO4 557
Problem 17: Electronic Properties of La2 CuO4
The compound with chemical formula La2 CuO4 has the 3D structure with body-
centered tetragonal conventional cell shown in Fig. P17.1. Note that it comprises a
stack of LaO and CuO2 planes.
1. On Fig. P17.1, indicate the primitive vectors and basis of a primitive cell in the
3D crystal.
The electronic configurations of the neutral atoms making up this structure are
La : (Xe) 5d1 6s2 , Cu : (Ar) 3d10 4s , O : 1s2 2s2 2p4 .
2. According to an LCAO calculation, is this compound insulating or metallic?
The LCAO calculation shows that the La and oxygen in the LaO planes are in the
configurations La3+ and O2− corresponding to fully occupied shells. Bands result-
ing from these orbitals are at energy levels well below those of the Cu(3d) and
O(2p) bands of the oxygens in the CuO2 planes. It is thus the latter that constitute
the valence or conduction levels of the crystal. It therefore suffices to consider the
electronic structure of the CuO2 plane.
3. Specify the primitive cell of the CuO2 plane, along with its reciprocal lattice and
first Brillouin zone.
We shall assume that the last occupied band corresponds to one orbital per Cu site
of atomic energy E0 . The band is calculated using the LCAO approximation, only
taking into account nearest-neighbour hopping integrals t0 and t1 .
4. Find the eigenergies E(kx , ky ) and plot the constant energy curves in the first
Brillouin zone. What is the curve corresponding to an occupancy of one electron
per unit cell?
Cu
Oxygen
La
Fig. P17.1 Crystal structure

of La2 CuO4 , showing body-
centered tetragonal conven- a
a
tional cell
558 13 Problem Set

pendence of the intensity of La 2CuO4
the neutron diffraction peak
at (π/a, π/a). Adapted from
Diffracted intensity
(arbitrary units)
Yamada, K., Kudo, E., Endoh,
Y., Hikada, Y., Oda, M.,
Suzuki, M., Murakami, T.:
Solid State Commun. 64, 753
(1987)
100 200 300

T (K)
Conductivity measurements show that La2 CuO4 is insulating. Moreover, neutron

diffraction produces Bragg peaks at the four points (±π/a, ±π/a) of the plane
below 300 K. The intensity of these peaks increases at low temperatures and sat-
urates at T = 0 K, as shown in Fig. P17.2. X-ray diffraction only picks up the Bragg
spots corresponding to the reciprocal lattice found in question 3.
5. What can you deduce about the ground state of the electronic system of the CuO2
plane? Specify the primitive cell of the real lattice of the CuO2 plane for T <
300 K.
A fraction of the La3+ is replaced by Sr, which has the ionic form Sr2+ in the
crystal. In the compound La1.85 Sr0.15 CuO4 it is observed that the spots at the four
points (±π/a, ±π/a) of the plane disappear in a neutron diffraction experiment.
Moreover, this compound is superconducting with a critical temperature of 40 K.
6. What conclusions can be drawn from these observations? According to the LCAO
approximation, what should be the Fermi surface of this metal?
Solution
1. Like the body-centered cubic cell, the body-centered tetragonal cell has two Bra-
vais lattice points per conventional cell. The latter contains twice the chemical
formula La2 CuO4 (see Fig. P17.3a). A possible primitive cell containing the for-
mula La2 CuO4 only once is shown in Fig. P17.3b. It is constructed from the
vectors a(−1, 0, 0), a(0, 1, 0), and (−a/2, a/2, c/2).
2. 2La and 4O give an even number of electrons per unit cell. However, copper
has an odd number of electrons. The primitive cell thus contains an odd number
of electrons. There will therefore be at least one partially occupied band in an
LCAO calculation and this leads to a metallic state.
3. The square primitive cell in the CuO2 plane containing one Cu atom and two O
atoms is shown in Fig. P17.4a. The square reciprocal lattice of side 2π/a and the
first Brillouin zone are depicted in Fig. P17.4b.
4. The LCAO calculation gives E(kx , ky ) = E0 − t0 − 2t1 (cos kx a + cos ky a). The
constant energy curves are
E0 − t0 − E
cos kx a + cos ky a = .
2t1
These are almost circular curves around k = 0 for E E0 − t0 − 2t1 and around
k = (π/a, π/a) for E E0 − t0 + 2t1 . For one electron per cell, the last filled level
corresponds to a half-filled band, or
E = E0 − t0 ,
Cu
Oxygen
La
a x
(a) (b)
Fig. P17.3 Crystal structure of La2 CuO4 . (a) Conventional cell containing twice La2 CuO4 .
(b) Possible primitive cell
560 13 Problem Set
y ky
kx
x
2π/a
(a) (b)
Fig. P17.4 (a) Square primitive cell of the CuO2 plane. (b) Square reciprocal lattice and first
Brillouin zone
Fig. P17.5 Constant energy π

curves and Fermi surface
2
ky a 0
−1
−2
−π
−π −2 −1 0 1 2 π
kx a
cos kx a = − cos ky a .
This is represented by the square in Fig. P17.5, which would be the Fermi surface
of the corresponding metal.
5. Since neutrons and X rays do not give the same Bragg spots below 300 K,
this suggests that magnetic order occurs at this temperature. The further spots
obtained with neutron diffraction at the four points (±π/a, ±π/a) can be used to
determine the reciprocal
√ lattice of the magnetic order. It corresponds to a square
lattice of side π 2/a, with area equal to half that of the reciprocal lattice of
the crystal structure (see Fig. P17.6a). The real magnetic lattice thus has a unit
cell of area 2a2 , twice that of the crystal lattice. The ground state is therefore
antiferromagnetic with two sublattices.
ky y
ky
kx
2π /a
kx
(a) (b)
Fig. P17.6 (a) Reciprocal lattice of the magnetic structure. (b) Real space two-sublattice antifer-
romagnetic structure showing the primitive cell (shaded) and the two antiparallel Cu spin states of
the magnetic basis
Since the La3+ and O2− in the LaO planes donate one electron per primitive cell of the atomic
structure to the CuO2 plane, it seems likely that the electronegativity of the oxygen is satisfied
and that the Cu is in a state Cu2+ , or 3d9 . Therefore, the Cu2+ do carry the magnetic moment.
The antiferromagnetic primitive cell is shown in Fig. P17.6b. This natural hypothesis regarding
magnetism is borne out experimentally by measuring the intensities of the magnetic spots,
which show that the corresponding structure factor is indeed that of Cu. The orientation of
the moments in the unit cell cannot be obtained directly from this data, and are indeed more
difficult to determine experimentally.
6. Replacing xLa3+ by xSr2+ amounts to transferring x fewer electrons per primitive

cell to the CuO2 planes. This seems to be enough to destroy the antiferromagnetic
state and make the system metallic. Whereas one electron was lacking to saturate
the 3d shell of Cu, 1 + x are now lacking. Formally, there are therefore 1 − x
electrons or 1 + x holes in the last occupied CuO2 band. If we attempt to interpret
this metallic state in a band model with 1 − x electrons, this corresponds in the
case x = 0.15 to a curve of constant E with area 0.85/2 times the area of the first
Brillouin zone. This would be roughly the area enclosed by the dotted curve in
Fig. P17.5.
In reality, the compound La2 CuO4 is a Mott–Hubbard insulator, and the holes created by
substituting Sr are in fact holes on the oxygens in the planes. We are still a long way from
understanding this type of metallic state, in which a magnetic moment associated with the hole
on the Cu coexists with holes on the oxygen that play a dominant role in electron transport
properties. The many conceptual difficulties sketched here explain in part the considerable
effort required to understand the origins of high-Tc superconductivity.
Problem 18: Properties of an Antiferromagnetic Solid 563
Problem 18: Properties of an Antiferromagnetic Solid9
The aim here is to determine the magnetic response of an antiferromagnetic solid in

the mean field approximation and to evaluate the exchange interactions from these
measurements.
18.1: Preliminaries: The Ferromagnetic Case
Consider an ensemble of N atoms each carrying a magnetic moment μ̂i associated

with a spin 1/2, i.e., μ̂ = −2μB Ŝ. The moments are non-interacting. They are sub-
jected to a magnetic field H0 parallel to the z axis.
1. Recall the expression

Mz = N μz = f (H0 /T)
for the temperature and field dependence of the magnetisation. Obtain the
susceptibility

−1 ∂M

χ = μ0 .
∂H0 H0 →0
What type of magnetic behaviour is characterised by the above laws?
The moments subjected to the field H0 also interact with their nearest neighbours
by the exchange interaction, with a positive exchange constant J .
2a. Explain why in the molecular field approximation the magnetic state of the N
magnetic moments is determined by
Mz = f (Heff /T) , (13.157)
where f is the function defined in question 1. Give the expression for Heff in
terms of J , M, H0 , N, and the number z of nearest neighbours of an atom.
2b. In zero field, (13.157) reduces to M = g(M/T). In terms of J , what is then
the temperature Tc below which the system becomes ferromagnetic?
2c. For a weak external field, deduce the temperature dependence of the suscep-
tibility χ for T > Tc .
9 This problem has been designed with J.C. Tolédano.

564 13 Problem Set
18.2: Antiferromagnetic Transition
Among the N magnetic moments considered in 18.1, N/2 occupy the corners of ad-
jacent cubes forming the crystal lattice in a solid body, while N/2 occupy the centers
of these cubes. The dipoles are thus located at the lattice points of two identical cu-
bic sublattices A and B, shifted by one cube half-diagonal from one another.
Each moment interacts by the exchange interaction with its six nearest neigh-
bours of the same sublattice, with the positive exchange constant J1 > 0, and with
its eight nearest neighbours of the other sublattice, with the negative exchange con-
stant −J2 (with J2 > 0). The Hamiltonian for this system is therefore

H = −J1 SaR ·SaR − J1 SbR ·SbR + J2 SaR ·SbR

+ 2μ0 μB H0 · SaR + SbR ,
where the first three sums are over pairs of spins and the last two are over the N/2
spins of a sublattice. Indices a and b denote variables associated with sublattices A
and B, respectively.
3. Adapting (8.37) of Chap. 8, write down HRa , the Hamiltonian containing all
terms of H involving a spin SRa at a given site of sublattice A. Similarly for
HR for a given spin SR .

b b
4. Letting Ma and Mb be the magnetisations of the two sublattices A and B,

deduce by analogy with 18.1 that the magnetic state of the system is deter-
mined in the molecular field approximation by
1 1
Ma = f (Haeff /T) , Mb = f (Hbeff /T) , (13.158)
2 2
where
Haeff = H0 + WMa − W Mb , Hbeff = H0 + WMb − W Ma ,
and f is the function defined in question 1. Express W and W in terms of J1 ,

J2 , and N.
Equations (13.158) determine the appearance in zero external field of a nonzero

magnetisation on each of sublattices A and B below a temperature TN .
5. Explain qualitatively why the magnetisations of the sublattices should satisfy
Ma = −Mb . Using this, calculate TN as a function of W and W .
When there is an external field H0 , the total magnetisation M = Ma + Mb is no

longer zero.
0 100 200 300 400 500
3
2.5 α -MnS Ma
MnF2
2
Susceptibility χ
( π /2 − φ )
1.5
H0
0.5
Mb
0
0 100 200 300 400 500
T (K)
Fig. P18.1 Susceptibility measurements on MnF2 and α-MnS (left). Tilting of the magnetisation
in a small transverse external field (right). Adapted from Bizette, H., Tsai, B.: Comptes-Rendus de
l’Académie des Sciences 1575 (1954)
6. From (13.158), calculate the temperature dependence of the susceptibility

χ = μ−1
0 (∂M/∂H0 )H0 →0 for T > TN (M and H0 small).
In the following, we set
Nμ0 μ2B (W − W)
Θ= ,
2kB
Neutron diffraction experiments show that, in MnF2 and α-MnS, the manganese
moments are in an antiferromagnetic ground state with two sublattices, below the
Néel temperatures TN = 68 K and TN = 150 K, for MnF2 and α-MnS, respectively.
Magnetic susceptibility measurements give the results shown in Fig. P18.1 (left).
7. How could you deduce J1 and J2 from these experimental results? Can the
constant J1 be negative?
18.3: Susceptibility in the Antiferromagnetic State
We now consider T < TN , and assume that the equilibrium directions of the mag-
netisations Ma and Mb in zero external field are parallel to the z axis. A small field
H0 is then set up perpendicular to the initial magnetisation direction.
8a. Show using the Hamiltonian of question 3 and the expressions for W and W
obtained in question 4 that that the energy U of the magnetised lattices A
and B is
U W
= − (M2a + M2b ) + W Ma Mb − H0 (Ma + Mb ) . (13.159)
μ0 2
566 13 Problem Set
Assume that the ‘transverse’ H0 does not affect the magnitudes |Ma | = |Mb | = M0
of the magnetisations, but tends to tilt them toward the field direction, as shown in
Fig. P18.1 (right).
8b. Find the tilt angle φ that minimises the energy. Deduce the transverse sus-
ceptibility χ⊥ . How does it depend on T < TN ? How does it connect with the
susceptibility found in question 4?
8c. What further contribution to the energy could have been considered here?
What would be its effect?
A small field H0 is now applied parallel to the z axis and in the same direction as
Ma . We assume that the changes in the magnitudes of Ma and Mb have opposite
sign:
Ma = M0 + (H0 ) , Mb = −M0 + (H0 ) .
9. Use (13.158) to find the temperature dependence of the longitudinal suscepti-

bility χ near TN . Show that χ (T = 0) = 0 and that χ (TN ) = χ⊥ (TN ).
10. On Fig. P18.1 (left), sketch the behaviour expected for χ and χ⊥ in the case
of MnF2 . Are neutron diffraction measurements absolutely necessary to deter-
mine the characteristics of an antiferromagnetic solid?
11. Can there be domains and domain walls in an antiferromagnetic solid?
Solution
Preliminaries: The Ferromagnetic Case
1. For an ensemble of non-interacting spin 1/2 moments, the magnetisation is
μ0 μB H0
Mz = N μz = NμB tanh = f (H0 /T) . (13.160)
kB T
Differentiating this,
N μ2B
χ= .
kB T
These equations characterise the paramagnetic behaviour.

2a. The Hamiltonian for the system of N moments is

H = −J SR ·SR + 2μ0 μB H0 · SR ,
RR R
where the first sum is over neighbouring pairs of spins and the second over all
spins. The terms in the Hamiltonian involving a given spin SR are

HR = SR · −J SR + 2μ0 μB H0 ,
R
where the sum is over the z nearest neighbours of the spin SR . In the molecu-
lar field approximation, SR is replaced by SR = −Mz /2NμB , its statistical
average. Then,

zJ Mz
HR = S R + 2μ0 μB H0 = 2μ0 μB Heff SR ,
2NμB
which defines the effective field

zJ
Heff = H0 + Mz .
4Nμ0 μ2B
For such a system, (13.160) is applicable with H0 replaced by Heff .

2b. In zero external field H0 , the equation Mz = f (Heff /T) reduces to

zJ Mz
Mz = NμB tanh β ,
4N μB
568 13 Problem Set
which only has nonzero solution if βzJ /4 > 1, that is, for temperatures
T < Tc = zJ /4kB . Below Tc , the system acquires a permanent magnetisa-
tion and becomes ferromagnetic.
2c. In a weak external field H0 and above Tc , the magnetisation is also weak. It is
given by the expansion

μ0 μB H0 Tc Mz Nμ0 μ2B H0 Tc
Mz = NμB tanh + + Mz .
kB T T NμB kB T T
It thus varies as Mz = Nμ0 μ2B H0 /kB (T − Tc ), and the susceptibility χ

diverges as
Nμ2B
χ= .
kB (T − Tc )
Antiferromagnetic Transition
3. The Hamiltonians HˆRa and HˆRb which describe the coupling of the spins SaR
and SbR with their surroundings are, respectively,

ˆ
H = S · −J1
a a a
S + J2 b
S + 2μ0 μB H0 ,
R R R R
R R

HˆRb = SbR · −J1 SbR + J2 SaR + 2μ0 μB H0 ,
R R
where sums are over nearest neighbours.
4. In the molecular field approximation, SaR is replaced by its average value
SaR = −Ma /NμB . Likewise, SbR = −Mb /NμB . (There are N/2 moments
in each sublattice.) Consequently,

6J1 Ma 8J2 Mb
HˆRa = SaR · − + 2μ0 μB H0 = 2μ0 μB Haeff ·SaR .
NμB NμB
The effective field Haeff is
3J1 4J2
Haeff = H0 + WMa − W Mb , W= , W = .
Nμ0 μ2B Nμ0 μ2B
Likewise, Hbeff = H0 + WMb − W Ma . In the mean field approximation, sub-

lattices A and B in effective fields Haeff and Hbeff , respectively, are statisti-
cally independent and we may apply the relation found in question 1 to each
sublattice. Then
NμB NμB
Ma = tanh(βμ0 μB Haeff ) , Mb = tanh(βμ0 μB Hbeff ) .
2 2
5. The negative exchange constant −J2 tends to orient the magnetisations Ma
and Mb in opposite directions. Then since the two sublattices have identical
structure, we can deduce that Ma = −Mb . Substituting this into the previous
equations, we then have
NμB
Ma = tanh βμ0 μB H0 + (W + W )Ma . (13.161)
2
In zero field, the equation
NμB μ0 μB (W + W )Ma
Ma = tanh
2 kB T
has a nonzero solution if T is less than the temperature TN determined by
Nμ0 μ2B (W + W ) 3J1 + 4J2

kB TN = = >0 . (13.162)
2 2
6. In a weak external field, and above TN , the magnetisation of a sublattice is

small and (13.161) can be linearised:
Nμ0 μ2B
Ma = (H0 + WMa − W Mb ) ,
2kB T
Nμ0 μ2B
Mb = (H0 + WMb − W Ma ) .
2kB T
Adding these, we obtain

Nμ0 μ2B W − W
M = M a + Mb = H0 + M ,
kB T 2
which yields
Nμ2B Nμ0 μ2B (W − W) 4J2 − 3J1

χ= , kB Θ = = .
kB (T + Θ) 2 2
(13.163)
7. Neutron diffraction experiments are used to determine the temperature TN
below which antiferromagnetic order arises. (At the end of this problem, we
shall see how TN can be deduced from measurements of χ .) Furthermore,
above TN , the susceptibility varies as 1/(T + Θ). We can thus find Θ and
deduce J1 and J2 from these two characteristic temperatures:
570 13 Problem Set
Fig. P18.2 Obtaining Θ graph- 2.5

ically from the plot of
χ −1 vs. T
2
1.5
χ −1
α -MnS
1
0.5 MnF2
Θ Θ
0
−600 −400 −200 0 200 400
T (K)
TN + Θ TN − Θ
J2 = kB , J1 = kB .
4 3
Figure P18.1 (right) shows the temperature dependence of χ above TN for the
two compounds. We must first check that the variation does indeed have the
form 1/(T + Θ). This is generally done by plotting 1/χ as a function of T.
This should give a straight line that cuts the horizontal axis at −Θ. This has
been done in Fig. P18.2 for the experimental results of Fig. P18.1 (right).
We obtain Θ 108 K for MnF2 and Θ 482 K for α-MnS. Using the TN
values obtained by neutron scattering, we find J2 44 K and J1 −13 K
for MnF2 and J2 138 K and J1 −110 K for MnS.
In both cases the coupling between sublattices is antiferromagnetic (J2 > 0),
as required to obtain an antiferromagnetic state. We observe that, for both
MnF2 and α-MnS, the magnetic state occurs even with antiferromagnetic cou-
pling within each sublattice (J1 < 0, see the definitions of J1 and J2 and
their signs in the statement of the problem). In fact, from (13.162), the anti-
ferromagnetic state described here is possible provided that J2 > −3J1 /4,
which is satisfied for the values found for MnF2 and α-MnS. For J1 > 0, we
would obtain Θ < TN .
Susceptibility in the Antiferromagnetic State

8a. Replacing the spins SR by their thermal average, taking into account the fact
that the first two sums in the Hamiltonian involve 3N/2 terms, i.e.,
1
3N/2 = (6 neighbours × N/2 spins ) ,
2
and that the third sum involves 4N terms (N/2 spins of one sublattice
connected to 8 neighbours on the other sublattice), the average energy is

3N Ma 2 3N Mb 2 Ma Mb
U=− J1 − J1 + 4NJ2
2 NμB 2 NμB (NμB )2
−μ0 H0 (Ma + Mb ) ,
that is
U 3J1 4J2
=− (M2a + M2b ) + Ma Mb − H0 (Ma + Mb ) .
μ0 2Nμ0 μB
2 Nμ0 μ2B
Given the expressions for W and W , we obtain (13.159) as required.

8b. We have
U
= −WM20 − W M20 cos 2φ − 2H0 M0 sin φ .
μ0
Minimising this with respect to φ, we obtain sin φ = H0 /2W M0 . The mag-
netisation along the field H0 is 2M0 sin φ, so the transverse susceptibility is
χ⊥ = (μ0 W )−1 . It is independent of temperature. It is easy to check that χ⊥
is equal to the value of χ (T = TN ) above the transition, which is given by
Nμ2B
χ⊥ = χ (T = TN ) = . (13.164)
kB (TN + Θ)
8c. We must take into account the magnetic anisotropy. This determines the mag-
netisation directions in space. It can be described by an energy of the form
K sin2 φ. This changes the susceptibility to
1
χ= .
μ0 W + K/2M02
9. Setting Ma = M0 + and Mb = −M0 + , the first equation of (13.158)

becomes
M0 + = μB tanh βμ0 μB M0 (W + W ) + H0 + (W − W ) .

2
Since H0 and are assumed small here, the hyperbolic tangent can be
expanded about
x = βμ0 μB M0 (W + W ) .
Eliminating M0 = (NμB /2) tanh x, it follows that
N μ0 μ2B H0 d
= tanh x tanh x = tanh x = 1 − tanh2 x .
2 kB T + Θ tanh x dx
572 13 Problem Set
Fig. P18.3 Temperature de- 2.5

pendence of longitudinal and
χ⊥ TN
transverse susceptibilities. 2 χ
Adapted from Bizette, H.,
Tsai, B.: Comptes-Rendus χ⊥
de l’Académie des Sciences 1.5
1575 (1954)
10 −4 χ
1
χ
0.5
0
100 200 300
T (K)
The susceptibility χ = 2/μ0 H0 is
Nμ2B tanh x
χ = . (13.165)
kB T + Θ tanh x
If T = 0, we have x → ∞ and tanh x = 0, whereupon χ (T = 0) = 0.

If T = TN , we have M0 → 0 and hence x → 0 with tanh x = 1, so that
Nμ2B
χ (T = TN ) = .
kB (TN + Θ)
Note that, with an applied field H0 parallel to Ma , this solution corresponds

to a high Zeeman energy for the two sublattices together. In reality, the
Zeeman energy cost can reduce to zero if the magnetisations of the sublat-
tices tilt to orient themselves in a direction perpendicular to H0 . Only some
source of anisotropy could prevent the magnetisation tilting in this way. But
the magnetisation will do this anyway if the applied field exceeds a critical
value called the spin flop field. (This field is determined by the values of the
anisotropy and exchange energies.)
10. The calculated values are plotted in Fig. P18.3 as continuous lines for χ
and χ⊥ . The measured values for MnF2 are also plotted there and found to
be in perfect agreement with expectations. It should be pointed out that such
susceptibility data do allow one to establish the existence of antiferromagnetic
order at TN .
11. There may be domains, associated with an interchange of the directions of
Ma and Mb (antiphase domains).
Problem 19: Magnetism of Thin Films and Magneto-Optic Applications 573
Problem 19: Magnetism of Thin Films and Magneto-Optic

Applications
Consider a thin ferrimagnetic single-crystal film of thickness d = 4 μm in the z di-

rection and infinite in the x and y directions. We assume that the exchange energy
is large enough for the magnetisation of the film to be uniform on a length scale
greater than the interatomic distance. To begin with, consider a film A of a mate-
rial with negligible magnetocrystalline anisotropy. The magnetisation is subject to
demagnetising fields.
1. Write down the demagnetisation energy of the film. In which direction does the
magnetisation M point when there is no applied field?
The film is subjected to a magnetic field Ha in a direction making an angle ψ with
the z axis. The system is taken round a hysteresis cycle.
2. Determine the cycle of the magnetisation in the direction of the applied field for
ψ = π/2.
3. Repeat for ψ = 0.
The composition of the ferrimagnetic material is modified in such a way as to sig-
nificantly increase the uniaxial anisotropy. The new film B has the same thickness
and its easy axis of magnetisation is perpendicular to the film. Setting θ = (Oz, M),
the anisotropy energy per unit volume of the film is
Ea = K sin2 θ .
4. If we assume that the magnetisation of the film is still uniform, what condition
must K satisfy for the zero-field magnetisation to differ from the one obtained for
film A?
We assume in the following that this condition is satisfied for film B. The aim now is
to find out how these thin, insulating, transparent ferrimagnetic films A and B, can
be used as indicators in magneto-optical imaging to carry out magnetic observations
and measurements. An aluminium film is thus deposited on them to act as a mirror
and they are observed by microscope under polarised light.
The polarisation of a polarised light wave will rotate through an angle dα
given by
dα = VMz dz ,
when it passes through a slab of magnetic material of thickness dz in the direction
of propagation z, where V is the Verdet constant and M the magnetisation of the
medium. This is known as the Faraday effect in magneto-optics (see Fig. P19.1).
Figure P19.2 shows a suitable experimental arrangement. We analyse the polari-
sation of light crossing film A or B and reflected by the metal film. Since the angle
of rotation α of the polarisation is small, the intensity of the light detected by the
CCD camera is proportional to α. This intensity is grey-scaled.
574 13 Problem Set
Fig. P19.1 Faraday rotation Ferrimagnetic film

of the polarisation of a light M
beam propagating through
a magnetised ferrimagnetic α
sample
Polarised z
light
5. What would you expect to observe for the above films A and B when there is no
applied field?
Figure P19.3 shows the image actually observed for film B.

6. Can you explain this observation qualitatively? Which of our assumptions con-
cerning film B is not borne out?
7. Without calculation, can you say which energy terms determine the width of the
ribbons observed in Fig. P19.3?
We wish to use films A and B as indicators for the magnetic structure inscribed on
a plane magnetised object C. To do this, C is placed under film A or B, in contact
Polariser
Sight
Lamp
Analyser
Beam
splitter
Objective
Ha Indicator A or B
Mirror (Al)
Solenoid Sample (C)
Fig. P19.2 Experimental magneto-optics setup. The enlarged view shows the disposition of
indicator A or B on the sample C
Fig. P19.3 Magneto-optic 300 μm

image of film B in the absence
of any applied field
with it, and we observe any changes in the reflected light. Let H(x, y) be the magnetic
field produced by C. We assume it uniform to begin with.
8. Show that only the component of H(x, y) normal to the film will produce a visible
effect for film A. But what about film B?
Figure P19.4 (left) is the image observed with film B when C is a credit card with
number in the form of a bar code on the magnetic strip. Note the difference of scale
with Fig. P19.3.
9. Can you explain the phenomena apparent in this image?
Figure P19.4 (right) shows a different region of the same credit card observed with
film A.
10. Which of the two films A and B could be used to make an exact measurement
of the magnetic field H(x, y) induced by C? Explain why. What is the maximal
value of H that could be measured in this way?
2mm
Fig. P19.4 Credit card. Left: Image of bar code using film B. Right: Image of another region of
the card using film A. Images 19.1–4 courtesy of K. Van der Beek (LSI, Ecole Polytechnique)
576 13 Problem Set
Solution
1. The demagnetisation tensor is diagonal with Nz = 1, Nx = Ny = 0. The demag-

netisation energy is
1 1
Ed = μ0 M · N · M = μ0 Mz2 .
2 2
For zero applied field, it is minimal when the magnetisation lies in the plane of
the film. Under these conditions, the magnetic poles are repelled to infinity.
2. When a magnetic field Ha is applied in a direction making an angle ψ with the z
axis, setting θ = (Oz, M), the total energy is
1
ET = μ0 M02 cos2 θ − μ0 M0 Ha cos(ψ − θ ) .
2
The same calculation was carried out in Chap. 9 for the hysteresis cycle when
there is a uniaxial anisotropy. We simply make the change of variables
π π 1
α= −θ , φ= −ψ , K = μ0 M02 ,
2 2 2
to recover the same equations. Indeed, the anisotropy due to the demagnetising
fields has the same form as the uniaxial anisotropy. The difference here is just
that the easy axis lies in the x direction of the plane. We obtain the same result,
namely for ψ = π/2, i.e., φ = 0, the field applied parallel to an easy axis leads
to a square hysteresis cycle. The reversal field is simply Ha = −M0 .
3. On the other hand, for ψ = 0, i.e., φ = π/2, the applied field is transverse with
respect to the easy axis and the magnetisation perpendicular to the film, Mz = M0
for Ha > M0 , rotates continuously when Ha < M0 , so that Mz = Ha .
4. The ferrimagnetic film B has two anisotropy terms with antagonistic easy mag-
netisation axes, the uniaxial anisotropy corresponding to an easy axis perpendic-
ular to the film. The total anisotropy energy is

1 1
Ea = μ0 M02 cos2 θ + K sin2 θ = K + μ0 M02 − K cos2 θ .
2 2
For K > μ0 M02 /2, the sign of the anisotropy term changes and the zero field
magnetisation points in a direction perpendicular to the plane.
5. For film A, the magnetisation M lies in the plane of the film. It is therefore
perpendicular to the direction of propagation of the light. There is no Fara-
day effect and hence no rotation of the polarisation. For film B, the mag-
netisation is perpendicular to the film, and the polarisation is rotated through
α = VMd. In both cases we thus expect uniform polarisation of the reflected
light.
6. The image of Fig. P19.3 shows that the polarisation of the light is rotated by dif-
ferent amounts at different points on film B. Therefore, the magnetisation of this
film is not uniform. While a magnetisation direction perpendicular to the plane
of the film minimises the uniaxial magnetocrystalline anisotropy, it does not min-
imise the demagnetisation energy. Quite the contrary, if the magnetisation were
uniform, the latter would actually be maximal. In order to minimise the demag-
netisation energy, magnetic domains appear that are alternately magnetised up
and down. The Faraday rotation of the light polarisation is in opposite directions
for domains magnetised up and down, and this is what leads to the lighter and
darker strips in the image.
7. The configuration of these domains is the one minimising the sum of three energy
terms, viz., the exchange energy which imposes a locally uniform magnetisation,
the magnetocrystalline anisotropy energy which tends to align the magnetisa-
tion with the easy axis, and the demagnetisation energy which tends to minimise
the field produced outside. In order to minimise the demagnetisation energy, the
width of the domains must be reduced as far as possible to minimise the demag-
netising fields of the magnetic domains and reduce the field produced outside.
However, this cannot be done without increasing the number of Bloch walls. The
energy involved in creating these interfaces is what prevents an infinite subdivi-
sion. So in practice it is the minimisation of the sum of the demagnetisation en-
ergy and the energy of the Bloch walls that determines the width of the domains.
8. For film A, the magnetisation lies in the plane of the film and can only acquire
a transverse component under the effect of a perpendicular field. There is thus
no Faraday effect if H(x, y) lies parallel to the film. For film B, a field parallel
to the film will rotate the magnetisation of the domains and reduce the normal
component of their magnetisation. With a sufficiently strong applied field, the
domain structure can completely disappear so that the magnetisation lies in the
plane. This disappearance of the domains is easy to observe magneto-optically.
9. The credit card has a magnetic bar code which encodes 0 and 1 by the direction
of the magnetisation. The field produced by the magnetic zones has a component
in the perpendicular direction which magnetises indicator B in such a way as to
broaden domains corresponding to the field orientation. In Fig. P19.4 (left), the
domains have completely disappeared in regions corresponding to the bar code.
This shows that the indicator magnetisation was saturated there. In these zones,
the magnetisation is M = ±M0 z, corresponding to the same magnetisation as in
the domains of the map observed in zero field. In the regions of the map where
there is no bar code, the field is zero and the indicator domains are visible. It can
be checked that the Faraday rotation is the same in the domains of the film in
zero field and in the magnetised zones (the image is made up solely of whites
and blacks).
10. With indicator A, one can also observe the field distribution induced by the
credit card in Fig. P19.4 (right). Recall that, in this case, only the field com-
ponent perpendicular to the plane of the film is active. If Hz (x, y) = Hz is
578 13 Problem Set
uniform, the magnetisation of the film is Mz = Hz , according to question 3.

If Hz (x, y) M0 , a spatial non-uniformity in Hz (x, y) will be faithfully reflected
by the same non-uniformity in Mz (x, y) = Hz (x, y), provided that the spatial vari-
ation of Hz (x, y) is only significant over large distances compared with the char-
acteristic dimensions of the Bloch walls of the indicator. In this case, the image
provides a direct measurement of the spatial variation of Hz (x, y). The image in
Fig. P19.4 (right) contains all shades of grey, and constitutes a faithful repro-
duction of Hz (x, y). If K μ0 M02 /2, the field that can be measured is limited
only by the value of the anisotropy field 2K/μ0 M0 . To measure strong fields,
film A should have the greatest anisotropy possible.
With film B, the pattern of magnetic domains is influenced by the fields both
parallel and perpendicular to the film, and these processes are not generally
linear in the field. It cannot therefore be used to carry out accurate measure-
ments of Hz (x, y). This can be seen by comparing the left and right images of
Fig. P19.4. In the first, no variation is visible in the magnetisation at the edges
of the regions corresponding to the bar code. A weak field Hz suffices to saturate
the magnetisation of the film.
The only possible linear response for film B would arise for a field Hz (x, y) much
weaker than the coercive field Hc of the film. In this case, the transverse field
broadens (narrows) the domains depending on their direction of magnetisation.
Integrating the light intensity over much greater spatial distances than the width
of the ribbons, we obtain a quantity proportional to Hz (x, y). However, in so
doing, we lose a great deal of spatial resolution, and this approach can only be
used to measure large-scale spatial variations.
Problem 20: Magnetism of a Thin Film 579
Problem 20: Magnetism of a Thin Film10
The aim here is to investigate the magnetisation configuration in a thin magnetic

film as a function of its thickness. The three parts of the problem can be tackled
independently. However, it is important to obtain a general grasp of the issues here.
General Introduction
Near a surface, the magnetic anisotropy increases, mainly due to the lower symmetry
of the environment in which the atoms find themselves. This has consequences for
the magnetic properties of thin films, which may be relevant when they are used for
the purposes of magnetic recording.
If we assume uniform magnetisation (constant magnitude M and parallel orien-
tation of spins), the anisotropy energy per unit volume of a thin film contains three
terms:

1 2Ks
EA = KV − μ0 M 2 + sin2 θ , μ0 = 4π × 10−7 . (13.166)
2 t
Here θ is the angle between the magnetisation and the normal to the film, and KV
is the anisotropy energy per unit volume of the bulk material. The second term, the
shape anisotropy, arises from the demagnetisation energy of the film, assumed of
infinite extent (see Problem 19). The third term is due to the different environments
of atoms located close to the surface, which have fewer nearest-neighbour atoms
and therefore see an anisotropic exchange field. This anisotropy tends to impose a
magnetisation perpendicular to the surface on these surface layers. It has less and
less effect as the thickness t of the film increases, whence its form Ks /t, where Ks is
the anisotropy energy per unit area.
In most cases, KV is either zero (cubic symmetry) or very small. We shall take
KV = 0 throughout this problem. The image underlying (13.166) is that, despite
the contradictory instructions given to the spins at the surface and in the bulk, the
exchange interaction is strong enough to align all the spins in the same direction.
We shall ascertain the conditions for this to provide a good description.
To do this, we consider a Heisenberg model on a simple cubic lattice. We assume
that the film is infinite in the x and y directions and of thickness t = La in the z
direction, where L is the number of atomic layers and a is the lattice parameter [see
Fig. P20.1 (left)]. We treat the spins as classical vectors of unit length. The mag-
nitude of the spin will be absorbed by redefining the various constants. The spins
make an angle θ with the z axis [see Fig. P20.1 (right)]. Under these conditions, the
orientations of the spins only depend on z, the spins in plane i making angle θi with
the z axis [see Fig. P20.1 (left)]. The energy per unit area is then
10 This problem has been designed with H. Schulz.

580 13 Problem Set
z=t
θL M (z)
a z
θ (z)
θi
ψ
B
θ1 x
z=0
Fig. P20.1 Left: Simple cubic lattice. Right: Notation for spins and magnetic field
A
L−1 L
1
E= (θl − θl+1 )2 − μ0 M 2 a sin2 θl
a 2
l=1 l=1

L
−BMa cos(θl − ψ) + Ks (sin2 θ1 + sin2 θL ) . (13.167)
l=1
The first term, with A > 0, describes the exchange interaction, while the coupling
with a magnetic field making angle ψ with the z axis [see Fig. P20.1 (right)] is
described by the term in BM. The lattice parameter a is introduced to give the vari-
ous constants the right physical dimensions.
20.1: Uniform Magnetisation
1. Assuming the θl independent of l, and when B = 0, show that (13.167) reduces to

(13.166), up to a multiplicative constant t. For the relevant case Ks > 0, deduce
that there is a critical thickness tc for which the orientation changes suddenly
from θ = π/2 to θ = 0 (or to θ = π ). Find tc . Recall that KV = 0 throughout.
2. Consider the case B = 0. Generalise (13.166) to this case. For t < tc , show that
there are metastable situations for a perpendicular field (ψ = 0), whereas there
are not for a parallel field (ψ = π/2). Plot the hysteresis curves (magnetisation
vs. applied field) for the two cases ψ = 0 and ψ = π/2.
When there is a uniaxial anisotropy, the equation of motion for the magnetisation is
modified. We have
dM
= γ M × (B + λMz ) , (13.168)
dt
4Ks
λ= − μ0 , (13.169)
tM 2
with γ = gμB /h̄ and Mz = (0, 0, Mz ). Consider an applied field in the z direction for
a thin film t < tc .
3. Linearising (13.168), find the ferromagnetic resonance frequency when the field
B is varied in the hysteresis diagram determined earlier.
Fig. P20.2 Ferromagnetic t = 20 Å

resonance field Br vs. angle T = 291 K
ψ between the field and the 0.8
normal to the film. Figure
courtesy of Hurdequint, H.:
Laboratoire de Physique des 0.6
B r (tesla)
Solides, Orsay
0.4
0.2
0
−90 −60 −30 0 30 60 90
ψ (degrees)
We now consider a thicker film, with t > tc .

4. What is the resonance frequency for a field applied in the x direction, i.e., with
ψ = π/2?
Even for t > tc , a strong enough field parallel to the z axis can align the magnetisation
with that axis.
5. Find the minimal value B0 of the field required to do this. Using the result of
question 3, give the resonance frequency for B > B0 . (For weaker fields, the res-
onance frequency vanishes.)
The ferromagnetic resonance of a cobalt film of thickness 2 nm is observed with a

spectrometer at fixed frequency ω/2π = 9.23 GHz, while varying the strength and
orientation of the applied field B. In Fig. P20.2, the field Br corresponding to the
ferromagnetic resonance is plotted against the angle ψ between the field and the
normal to the film. The values of the resonance field for ψ = 0 and ψ = π/2 should
correspond to the situations investigated above.
6. Given that μ0 M = 1.7 tesla for cobalt at room temperature, find the critical thick-
ness tc for cobalt. Could thin cobalt films be used to make perpendicular mag-
netic recordings? (Note that, for electron spins, ω/2π B is 27.98 GHz/tesla.)
20.2: Non-uniform Situations
7. Write down the equations minimising the energy in (13.167). Note that the
surface variables θ1 and θL satisfy different equations to the bulk variables
θ2 , . . . , θL−1 . Here and in the following, take B = 0.
Assume in the following that the variables θl vary slowly with l. We then have θl =
θ (z), where z = la is treated as a continuous variable. Hence,
582 13 Problem Set
dθ d2 θ
θl+1 − θl = a , 2θl − θl+1 − θl−1 = −a2 . (13.170)
dz dz2
8. Show that the equations for the bulk variables become the differential equation
d2 θ
4A + μ0 M 2 sin 2θ = 0 , (13.171)
dz2
while the equations for θ1 and θL become the boundary conditions

dθ
−2A + K̃s sin 2θ (0) = 0 , (13.172)
dz z=0

dθ
2A + K̃s sin 2θ (t) = 0 , (13.173)
dz z=t
where K̃s = Ks − μ0 M 2 a/2.
In the following, K̃s > 0.

9. For B = 0, show that there are three trivial solutions to (13.171). What physical
situations do they correspond to?
Comments
In the following, still assuming B = 0, we shall consider the possibility of solutions

in which θ is space dependent. We note first that the symmetry of the problem
requires θ (t/2 + z) = θ (t/2 − z). The function θ (z) thus takes its maximal value θm
in the middle of the film: θm = θ (t/2). We define the exchange length by
.
2A
Λ= . (13.174)
μ0 M 2
After the transformation

sin θ = sin θm sin φ , (13.175)
and setting φ0 = φ(z = 0), the solution to (13.171) has the form
φ(z)
z dφ
= = F φ(z), sin2 θm − F φ0 , sin2 θm . (13.176)
Λ φ0 1 − sin2 θm sin2 φ
In particular, in the middle of the film, θ (t/2) = θm and hence φ(t/2) = π/2, so
t
= F(π/2, sin2 θm ) − F(φ0 , sin2 θm ) , (13.177)
2Λ
where F(φ, m) is a special function of the two variables φ and m called the incom-
plete elliptic integral of the first kind, defined by
4 4
F (π /2, m)
F (φ , m) 3
2
2
1
0 1
0.0 0.5 1.0 1.5 0.0 0.2 0.4 0.6 0.8 1.0
φ m
Fig. P20.3 (a) Dependence of F(φ, m) on φ for (bottom to top) m = 0, 0.2, 0.4, 0.6, 0.8, and 1.
(b) Dependence of F(π/2, m) on m
φ dx
F(φ, m) = . (13.178)
0 1 − m sin2 x
Figure P20.3 plots some examples. When m → 1, we have
1
F(π/2, m) ≈ln 16/(1 − m) .
2
The solution (13.176) still contains two undetermined constants, θm and φ0 . Com-
bining (13.176) with the boundary condition at z = 0, these are found to be related by

ρ + 1/ρ − (ρ + 1/ρ)2 − 4 sin2 θm
sin φ0 =
2
, (13.179)
2ρ sin2 θm
with
K̃s
ρ= . (13.180)
μ0 M 2 A/2
Note that sin2 φ0 is an increasing function of θm . Equations (13.177) and (13.179)
completely determine θ (z) for given thickness and constants specifying the
material.
20.3: Detailed Investigation of Non-uniform Situations
For thin enough film (small t), the solution θ = 0 is stable, minimising the surface
anisotropy energy. We now consider a situation where the surface anisotropy is very
large, i.e., ρ → ∞. Equation (13.179) then gives ρ0 = 0.
10. Sketch the function θ (z) approximately, choosing two or three values of the
parameter θm .
11. Are these non-uniform situations observable? What experimental methods could
you use?
584 13 Problem Set
12. Still for ρ = ∞, use (13.177) to show that there is a critical thickness tc below
which the solution (13.176) no longer exists, and we therefore have θ = 0. Sketch
the dependence of θm on t approximately. Sketch the function θ (z) for t just above
tc and for t tc . Comment on the result.
As in the previous case, for ρ < ∞, the critical thickness is given by the solution to
(13.177) with θm = 0.
13. Simplify this expression and obtain an explicit form for tc as a function of ρ and
Λ alone.
14. Show that, for ρ > 1, the solution given by (13.176) exists for any value of the
thickness t > tc . Use the fact that, in this case, φ0 < π/2 even when θm = π/2.
Plot θ (z) for t ≈ tc and for t tc .
15. For small surface anisotropy ρ < 1, show that t remains finite when θm = π/2
and that there is therefore an upper critical thickness tc2 , and specify its
dependence on ρ. Use the fact that, in this case for θm → π/2, we also have
φ0 → π/2.
If the two angles are close to π/2, (13.179) implies
ρ2
(π/2 − φ0 )2 = (π/2 − θm )2 . (13.181)
1 − ρ2
The integral in (13.177) thus simplifies. For t > tc2 , we have a homogeneous solution
θ = π/2. The situation is completely dominated by the shape anisotropy.
16. With the numerical values found in 20.1, what are the corresponding orders of
magnitude?
17. Plot a phase diagram in the (ρ, t) plane. Comment on the respective roles of
the exchange interaction, surface anisotropy, and shape anisotropy in the three
phases.
Solution
Uniform Magnetisation
1. We have
1
E = − μ0 M 2 aL sin2 θ + 2Ks sin2 θ
2

1 2Ks
= − μ0 M + 2
t sin2 θ . (13.182)
2 t
For small enough t, surface anisotropy dominates (Ks > 0). We thus minimise
E by setting θ = 0 or θ = π . For greater thickness, shape anisotropy dominates,
giving θ = π/2. The critical value of t is found by setting the prefactor of sin2 θ
to zero, that is
tc = 4Ks /μ0 M 2 . (13.183)
2. For B = 0, the calculation leading to (13.182) generalises to give

1 2Ks
E = − μ0 M + 2
t sin2 θ − BMt cos(θ − ψ) . (13.184)
2 t
For t < tc , set

4Ks
B0 = − μ0 M . (13.185)
tM
When ψ = 0 and |B| < B0 , the function E(θ ) has two local minima at θ = 0 and
π . Metastable situations are thus possible for B perpendicular to the film. For
B > 0, the solution θ = 0 is the equilibrium case and θ = π is metastable, while
the situation is reversed for B < 0. When |B| > B0 , there is only one minimum at
θ = 0 or θ = π .
When ψ = π/2, there is only one local minimum for E(θ ). When |B| < B0 , it
corresponds to sin θ = B/B0 , and when |B| > B0 , it is at θ = ±π/2, with the
same sign as B. The two cases are depicted in Fig. P20.4.
3. For Bẑ, the equation of motion is
dMx dMy
= γ (B + λMz )My , = −γ (B + λMz )Mx . (13.186)
dt dt
In the linear response limit Mz = ±M, we have precession with angular frequency
ω2 = γ 2 (B + λM)2 ,
where λM = ±B0 , depending on the sign of M, and then
ω2 = γ 2 (B ± B0 )2 . (13.187)
586 13 Problem Set
Mz Mx
−B0
B B
−B0 B0 B0
Fig. P20.4 Hysteresis curves for ψ = 0 (left) and ψ = π/2 (right). Thick lines are stable and thin
lines are metastable
This frequency vanishes at the ends of the two metastable branches of Fig. P20.4.
4. In this case M x̂, and we must write down the equations for My,z , viz.,
dMy dMz
= γ (B − λMx )Mz , = −γ BMy .
dt dt
Linearising, i.e., putting Mx M, we have
ω2 = γ 2 B(B − λM) . (13.188)
5. We may return to (13.184) for the case ψ = 0 and λ < 0, with

4Ks
λ= − μ0 . (13.189)
tM 2
It transpires that θ = 0 is the absolute minimum of this function if
B ≥ B0 = |λ|M . (13.190)
For B ≥ B0 , the geometry is as in question 3 above. In this regime, we thus obtain

the result (13.187). Since λ < 0 here, ω = 0 for B = B0 . Subsequently, ω increases
linearly with B for B > B0 .
6. It is the situations in question 4 for ψ = π/2 and question 5 for ψ = 0 that
apply here. We write down the corresponding results, but slightly modifying the
notation. For ψ = π/2, the resonance condition is
2
ω
= Br (Br + μ0 M − BA ) , (13.191)
γ
whereas for ψ = 0, we have

ω
= Br − μ0 M + BA , (13.192)
γ
where BA = 4Ks /tM. In the present case, the fixed spectrometer frequency gives
ω 9.23
= T = 0.33 T . (13.193)
γ 27.98
For ψ = 0, we have Br = 0.85 T, and with μ0 M = 1.7 T, we obtain BA = 1.18 T.
When ψ = π/2, we have Br = 0.15 T, and from (13.191), we obtain BA = 1.12 T.
To determine the conditions under which there can be magnetisation perpendic-
ular to the film, we write (13.182) in the form

1 4Ks 1 2 nm
E= − μ0 M Mt sin θ =
2
1.18 T − 1.7 T Mt sin2 θ .
2 tM 2 t
(13.194)
The coefficient of sin2 θ becomes positive if t < 1.39 nm, so the magnetisation
will be perpendicular to the film if
t < tc = 1.39 nm .
This effect has in fact been observed for films of thickness 1.13 nm [Chappert, C.,
et al.: Phys. Rev. B 34, 3192 (1986)].
Non-uniform Situations
7. Differentiating E with respect to θl in (13.167), we find that

2A 1
(2θl − θl+1 − θl−1 ) − μ0 M 2 a sin(2θl ) = 0 , 2 ≤ l ≤ L−1 , (13.195)
a 2
2A 1
(θ1 − θ2 ) − μ0 M 2 a sin(2θ1 ) + Ks sin(2θ1 ) = 0 , l=1, (13.196)
a 2
2A 1
(θL − θL−1 ) − μ0 M 2 a sin(2θL ) + Ks sin(2θL ) = 0 , l = L . (13.197)
a 2
8. Using equations (13.170), (13.195)–(13.197) transform immediately into the re-
quired differential equation and boundary conditions.
9. θ = 0, π are the uniform solutions of (13.171). They correspond to magnetisation
perpendicular to the film in the positive and negative z directions. These solutions
minimise the surface anisotropy energy, but maximise the shape anisotropy en-
ergy. In contrast, for θ = π/2, the magnetisation is parallel to the film, minimising
the shape anisotropy energy, but maximising the surface anisotropy energy.
588 13 Problem Set
Fig. P20.5 Sketched curves θ (z)

for θ(z) for three different π /2
values of θm . Lower curve:
θm 0. Upper curve:
θm π/2. Center curve:
Intermediate value of θm
Detailed Investigation of Non-uniform Situations
10. For small θm , the function F(φ, sin2 θm ) can be approximated by a linear function
of φ, as can be seen in Fig. P20.3, so that
φ(z) ≈ z/Λ . (13.198)
Furthermore, under these conditions, sin θ (z) ≈ θ (z) for all z. From (13.175), we
then obtain
θ (z) = sin θm sin(z/Λ) . (13.199)
For θm ≈ π/2, the function F(φ, sin2 θm ) diverges almost everywhere when φ →
π/2. φ(z) can be obtained graphically from Fig. P20.3. Under these conditions,
we have θ (z) ≈ φ(z), and the same plot also gives θ (z), except in the immediate
vicinity of z = t/2, where φ(z) approaches its maximum π/2 with a gentle but
nonzero slope, while θ (z) reaches its maximum with zero slope. The curves are
shown schematically in Fig. P20.5 for θm ≈ 0, θm ≈ π/2, and for an intermediate
value. In particular, we observe that the parameter θm increases with the film
thickness.
11. To identify such situations, we can try to measure the magnetisation transverse
to the film. This might be feasible with SQUID sensitivity. However, there are
some obvious problems, e.g., sample alignment, the need to orient the per-
pendicular component by applying a magnetic field, demagnetising field, etc.
Clearly, θm must be large enough. The magnetisation profile can be ascertained
by a technique in which polarised neutrons are reflected from the surface. This
was developed by C. Fermon at the Léon Brillouin research institute, Saclay
(France) in 1996.
12. When θm → 0, the above solution (13.198) becomes exact. This solution satis-
fies the boundary conditions for a thickness t = π Λ. Since θm increases with the
film thickness, there is no non-constant solution satisfying the boundary condi-
tions below the critical thickness given by
tc = π Λ . (13.200)
Fig. P20.6 Dependence of θm θm

on the film thickness t π/2
tc t
Below this thickness, the constant solution θ (z) = 0 minimises the energy.
From Fig. P20.3, we obtain the dependence of θm on the thickness, as shown in
Fig. P20.6. For t close to tc , we find that θm varies as (t − tc )1/2 . Plots of θ (z)
are the same as in Fig. P20.5. Small θm corresponds to t tc , and θm ≈ π/2 to
t tc .
Comments
• In the case investigated here, ρ = ∞, the surface anisotropy is large, and a mag-
netisation is imposed perpendicularly to the film at surfaces z = 0, t.
• For t < tc , the shape anisotropy is too small to ‘pull’ the magnetisation from the
perpendicular orientation.
• For t tc , the shape anisotropy is just sufficient to shift θ slightly from θ = 0,
giving rise to a slightly arched curve θ (z).
• For t tc , the shape anisotropy aligns the magnetisation parallel to the film
almost everywhere, except in a region of thickness approximately Λ close to the
surface, where the surface anisotropy still dominates.
13. In contrast to the case ρ = ∞, for ρ < ∞, the surface magnetisation of the film
is not strictly perpendicular. In particular, from (13.179) for θm → 0, we obtain
1
sin2 φ0 = . (13.201)
1 + ρ2
Equation (13.177) then yields

π 1
tc = 2Λ − arcsin = 2Λ arctan ρ . (13.202)
2 1 + ρ2
14. For ρ > 1, when θm → π/2, (13.179) yields

1
sin2 φ0 → <1. (13.203)
ρ2
The second term on the right-hand side of (13.177) thus remains finite when
θm → π/2, while the first diverges. Consequently, there is always a solution to
590 13 Problem Set
(13.177), for all values of t > tc . Since φ0 = 0, we also have θ (0) = θ (t) = 0,
and the magnetisation is not perpendicular, even at the surface. The curves θ (z)
therefore have the shapes shown in Fig. P20.7.
15. Equation (13.177) can be rewritten in the form
π/2
t dφ
= . (13.204)
2Λ φ0 1 − sin θm sin φ
2 2
Since φ0 = π/2 − δ here, with δ small, the sin2 under the square root sign can
be replaced by a truncated expansion about φ = π/2, namely
sin2 φ ≈ 1 − (φ − π/2)2 = 1 − δ 2 .
Changing variables to φ = π/2 − φ, we obtain

δ
t dφ
=
2Λ 0 cos2 θm + sin2 θm φ 2
1 δ
= arcsinh . (13.205)
sin θm cot θm
Here θm → π/2, so sin θm → 1 and cot θm = π/2 − θm . Using (13.181), we
obtain
t ρ
= arcsinh = arctanh ρ . (13.206)
2Λ 1 − ρ2
The maximal allowed value of θm is π/2, corresponding to magnetisation par-
allel to the film. The thickness is then given by (13.206). For thicker films, the
only solution is θ (z) = π/2. As a consequence, there is an upper critical thick-
ness tc2 , above which the magnetisation is parallel to the film everywhere:
tc2 = 2Λarctanh ρ . (13.207)
This thickness increases for higher surface anisotropy and lower exchange
energy.
θ (z)
π /2
Fig. P20.7 Lower curve: θ(z)

for t tc . Upper curve: θ(z)
for t tc . Center curve: θ(z)
for an intermediate value of t z
Fig. P20.8 Phase diagram t/2Λ tc2

showing the three possible
M x
configurations of the spatial
variation of the magnetisation
of a ferromagnetic film,
depending on its thickness
t and surface anisotropy II III
parameter ρ
π tc
I M z
ρ
16. We have .
tc − 2a 2A
ρ= , Λ= .
Λ μ0 m2
The experiment in 20.1 yields the value μ0 M = 1.7 tesla. A is determined by
the exchange energy, which is not specified in the statement of the problem.
The Curie temperature and crystallographic constants of Co give the estimate
ρ ∼ 0.1. The window for observing a configuration in which the magnetisation
is not uniform is thus very narrow. In order to observe such a situation, it would
be preferable to use a material with a low value of A, i.e., with a low Curie
temperature.
17. There are 3 regions (see Fig. P20.8):
(I) t < tc . Surface anisotropy dominates over shape anisotropy, and the mag-
netisation is perpendicular. This region grows in size when the exchange
interaction increases, making the magnetisation more rigid.
(II) t > tc2 . Shape anisotropy dominates. A strong exchange interaction also
stabilises this phase, provided that the surface anisotropy is not too large.
(III) In the rest of the diagram, surface anisotropy dominates the film surfaces,
but inside, shape anisotropy plays the dominant role. This region becomes
significant when the exchange interaction is weak.
Appendix A
Physical Constants
Units
Angstrom 1 Å = 10−10 m (∼ size of an atom)

Fermi 1 fm = 10−15 m (∼ size of a nucleus)
Electron-volt 1 eV = 1.60218 × 10−19 J ←→ 11 600 K
Magnetic induction tesla (1 gauss ←→ 10−4 tesla)
Fundamental Constants
Planck constant h = 6.6261 × 10−34 J s

h̄ = h/2π = 1.05457 × 10−34 J s
Speed of light c = 299 792 458 m s−1
Vacuum permeability μ0 = 4π × 10−7 H m−1 (ε0 μ0 c2 = 1)
Boltzmann constant kB = 1.38066×10−23 J K−1 = 8.6174×10−5 eV K−1
Avogadro number NA = 6.0221 × 1023
Electron charge e = 1.60218 × 10−19 C
Electron mass m0 = 9.1094 × 10−31 kg
Proton mass mp = 1.67262 × 10−27 kg
Neutron mass mn = 1.67493 × 10−27 kg
Fine structure constant α = e2 /h̄c = 1/137.036
(dimensionless)
Classical electron radius re = e2 /m0 c2 = 2.818 × 10−15 m
Compton wavelength λc = h/m0 c = 2.426 × 10−12 m
of the electron
Bohr radius a1 = h̄2 /m0 e2 = 0.52918 × 10−10 m
Hydrogen ionisation energy EI = m0 e4 /2h̄2 = α 2 m0 c2 /2 = 13.6057 eV
Bohr magneton μB = qe h̄/2m0 = −9.2740 × 10−24 J T−1
= −5.7884 × 10−5 eV T−1
Nuclear magneton μN = qh̄/2mp = 5.0508 × 10−27 J T−1
= 3.1525 × 10−8 eV T−1
DOI 10.1007/978-3-642-13565-1, c Springer-Verlag Berlin Heidelberg 2011
Appendix B
Some Useful Functions and Relations
Euler Function
∞
(t) = xt−1 e−x dx = (t − 1)(t − 1) = (t − 1)!
0
√
(1/2) = π.
Riemann ζ Function
∞
1
ζ (t) =
nt
n=1
with some specific values
t 2 3 4
ζ π 2 /6 1.202 π 4 /90
Integrals of the Fermi–Dirac Function
1 1
f (x) = and − df /dx =
ex + 1 4 cosh2 (x/2)
∞ ∞
f (x)dx = − (df /dx)x dx = ln 2.
0 0
For t > 1
∞ ∞ ∞
xt−1 1 xt
f (x)xt−1 dx = dx = dx = (1 − 21−t )(t)ζ (t),
0 0 ex + 1 4t 0 cosh2 (x/2)
which yields some specific values
595
596 Appendix B Some Useful Functions and Relations
∞ ∞
1 x21 π2
f (x)xdx = dx = ζ (2) =
2
,
0 8 0 cosh (x/2) 2 12
∞ ∞
1 x3 3 3
f (x)x dx = 2
2
dx = ζ (3) ≈ 1.202 ≈ 1.8,
0 12 0 cosh (x/2) 2 2
∞ ∞
1 x4 21 7 4
f (x)x3 dx = 2
dx = ζ (4) = π .
0 16 0 cosh (x/2) 4 120
Bessel Functions
The Bessel functions of order α are solutions of the differential equation
d2 y dy
x2 2
+ x + (x2 − α 2 )y = 0.
dx dx
They are important to solve Laplace’s equation in cylindrical (for integer α) or
spherical (half-integer α) coordinates.
The Bessel functions of the first kind Jn (x) with integer n are solutions which are
constant for x → 0 and are oscillating decreasing functions for x → ∞. They have
an integral form given by

1 π
Jn (x) = cos(nu − x sin u)du
π 0
with
J−n (x) = (−1)n Jn (x).
For n ≥ 0, they can be expanded for small x into
x2 1 x n
J0 (x) ≈ 1 − and Jn (x) ≈ .
4 n! 2
The asymptotic form for x → ∞ is
1
2 2 nπ π
Jn (x) ≈ cos x − − .
πx 2 4
Sinusoidal functions of sinusoidal functions can be developed into series of

Bessel functions using
∞

ix sin φ
e = Jn (x)einφ .
n=−∞
The modified Bessel functions of order α are solutions of the differential equation
d2 y dy
x2 + x − (x2 + α 2 )y = 0.
dx2 dx
Appendix B Some Useful Functions and Relations 597
The modified Bessel functions Kn (x),with integer n, are solutions which diverge
for x → 0, and vanish exponentially for x → ∞. Their integral representation is
given by
1 π ∞
Kn (x) = exp in exp(−ix sht − nt) dt ,
2 2 −∞
with
K−n (x) = Kn (x).
Their expansion for small x is
n
1 2
K0 (x) ≈ − ln x and Kn (x) ≈ .
2n! x
Their asymptotic forms for x → ∞ are

1
2 2
Kn (x) ≈ exp(−x).
πx
These functions are related to one another through

d n
x Kn (x) = −xn Kn−1 (x),
dx
which gives as an example
dK0 /dx = −K−1 = −K1 .

Appendix C
Standard Notation
A(T, μ) Grand canonical potential

M Atomic mass
I Electric current
T Temperature
U Internal energy
V Potential difference
Z Atomic number
ZG (T, μ) Grand partition function
e Electron charge
h Planck constant (= 2π h̄)
kB Boltzmann constant
s Electron spin (s = 1/2)
Φ Flux of magnetic induction around a contour (C)
β Reciprocal of kB T
χ, χ̃ Magnetic susceptibility (χ̃ = μ0 χ)
δ(x) Dirac delta
δij Kronecker delta
γ Gyromagnetic ratio
μ Chemical potential
μB Bohr magneton
A Vector potential
Ba Applied magnetic induction (= μ0 Ha )
F Force
Ha Applied magnetic field
M Magnetisation
j Electric current density
k Wave vector
pi Momentum of i th electron
vi Velocity of i th electron
μ Atomic magnetic moment
599
Appendix D
Specific Notation
ak0 ,k1 Atomic form factor [= f (k1 − k0 )]

BJ (x) Brillouin function with total angular momentum J
as a function of x = gJ μ0 μB Ha /kB T
Ce (T) Specific heat • electronic
Cn • in normal metal state
Cs • in superconducting state
Cv • at constant volume
D Coefficient characterising magnetic anisotropy energy per ion (see K)
D(E) Density of states • of energy E per atom and per spin direction
Dd (E) • of a free electron gas in d dimensions
Ds (E) • excited in the superconducting state
DΩ (E) • of energy E per unit volume
Dχ • at the Fermi level obtained by the Pauli susecptibility
Dc • at the Fermi level obtained by the specific heat
Dc (E − Ec ) • of the conduction band of a semiconductor
Dv (E − Ec ) • of the valence band of a semiconductor
E1 , E2 , . . . En Atomic energy levels
Ev Atomic energy level of a valence electron
Ek Electronic energy of state with quasi-momentum k (single band)
EF Fermi energy
EZ Energy of Zeeman coupling between magnetisation and magnetic field
Ed Demagnetisation energy
Ean Anisotropy energy as a function of orientation of M (see HA )
Ewall Energy of a Bloch wall per area of primitive cell
Ec Energy at minimum of conduction band (semiconductor)
Ev Energy at maximum of valence band (semiconductor)
Ec Condensation energy in superconducting state
Eg Semiconductor band gap (= Ec − Ev )
Eh Energy of hole
Gs (T, Ha ) Gibbs free energy • of superconducting state
Gn (T, Ha ) • of normal metal
Gsf (T, Ha ) • of superconducting film
Gms (T, Ha ) • of mixed state
601
602 Appendix D Specific Notation
Hc , Hc (T) Critical thermodynamic field of a superconductor

Hc Critical field of thin superconducting film
Hc1 Lower critical field for type II superconductor
Hc2 Upper critical field for type II superconductor
ĤZrf Hamiltonian for Zeeman coupling with a radiofrequency field
z 2
ĤA Magnetic anisotropy Hamiltonain, e.g., HA = D R (SR )
HK Anisotropy field
Hd Demagnetising field
Ic Critical current of a Josephson junction
J Total angular momentum of an atomic state (J = L + S)
JNN Current in tunnel junction • between two normal metals
JNS • between normal metal and superconductor
J Exchange interaction • between two spins S1 and S2
J0 • interatomic
K Magnetic anisotropy coefficient = −NDS2 = −DM02 /Ng2 μ20
L Spatial dimensions of a macroscopic crystal
L Latent heat at normal–superconducting transition
L Lorenz number

L T, M(r) Ginzburg–Landau functional
L(T, Mz ) Landau function
M0 (T) Spontaneous magnetisation of ferromagnet
Ms Magnetisation at saturation
Malt Alternating magnetisation in antiferromagnet with two sublattices
Ne Number of electrons in solid
Nn Number of atomic nuclei in solid
N Demagnetisation tensor
Pi Momentum of i th atomic nucleus
RW Wilson ratio relating χP and Ce in metal
Ss Entropy • of superconducting state
Sn • of normal state
S(k) Structure factor (primitive cell)
Tc Critical temperature of superconductor
TC Curie temperature
TN Néel temperature
T1 Relaxation time • spin lattice
T2 • spin–spin
T̂Ri Operator effecting translation of crystal lattice by Ri
U(r) Magnetic potential
V(r1 − r2 ) Interaction between two electrons at r1 and r2
Vat Atomic potential
VK Fourier component of periodic potential for vector K of reciprocal lattice
Vc (r) Coulomb potential
Vk Fourier component of attractive potential V(r1 − r2 )
between two electrons due to electron–phonon interaction
Appendix D Specific Notation 603
V0 Interaction between electrons due to electron–phonon coupling

in point interaction limit
a Distance between nearest neighbour atoms
c Specific heat capacity per particle
d Width of tunnel barrier
f (E) Fermi population factor
gJ Landé factor of electron in ground state with total angular momentum J
jen Energy flux
kF Fermi wave vector for free electron gas
Mean free path • of particle
e • of electron
(me )αβ Effective electron mass tensor [(mh )αβ for hole]
me Effective mass • of electron (isotropic case)
m∗ • of charge
m(t) Magnetic response to pulse excitation at t = 0
n Band index (band structure of solid)
n or ne Number density • of conduction electrons
np • of Bravais lattice points
ns • of electrons condensed in superconductor
nh • of holes in semiconductor
q Electric charge of pair in superconductor (q = −2e)
t1 Hopping integral • between nearest neighbours
tn, • between sites n and
un,k (r) Coefficient of eik·r in Bloch function ψn,k (r)
v Volume of primitive cell in crystal
vF Electron velocity at Fermi level
z Number of nearest neighbours of atom
k Width • of wave packet in quasi-momentum space
r • of wave packet in real space
Δ Supeconducting band gap
Ω Volume of solid
Φ0 Flux quantum (= h/2e)
Φ∗ Fluxoid (quantised quantity in superconductor)
α Optical absorption coefficient
α Exponent for isotopic effect
α = a/λ where a is thickness of superconducting film
α−n Overlap integral between atomic sites and n
χ (ω) Real part of alternating magnetic susceptibility (dispersion)
χ (ω) Imaginary part of alternating magnetic susceptibility (absorption)
χd Diamagnetic susceptibility
χn (r) Atomic wave function • level n
χv (r) • valence state
χP Pauli susceptibility of metal
δ Binding energy of Cooper pair
ε Dielectric constant of semiconductor
604 Appendix D Specific Notation
εk Energy of an electronic plane wave eigenstate (h̄2 k2 /2m0 )

ε(k) Spin wave dispersion relation
|φ(r)|2 = φ 2 Density of superconducting pairs (= ns /2)
γ Linear term in electronic specific heat of metal (Ce = γ T)
κ Thermal conductivity
κ = λ/ξ Ratio of penetration depth to coherence length
λ , λ(T) Penetration depth
λ0 • at zero temperature
λL • London
λ Spin–orbit coupling coefficient (λl · s)
λ∗ Molecular field coefficient
μe Electron mobility
ωc Cyclotron angular frequency of charge in magnetic field
ωD Debye angular frequency
ωL Larmor angular frequency
ψnk (r) Bloch function with band index n and quasi-momentum k
ρ(r) Electronic density at r
σ , σe , σh Electron or hole conductivity
σ (ω) Alternating conductivity of a metal
σs (ω) Alternating conductivity of a superconductor
T Transmission coefficient of a tunneling barrier
τ Relaxation time due to wall mobility in a ferromagnet
τe Relaxation time of electron velocity distribution and electron collision rate
θ(r) Phase of wave function ψ(r) (superconductor)
θD Debye temperature characterising phonon spectrum in solid
ξ Coherence length in the superconducting state
ζ Thickness of Bloch wall
Beff Effective magnetic induction at a site in a magnetic compound
J Total angular momentum for one ion
Hd Demagnetising field
K Vector in reciprocal lattice
Rl Vector in real lattice
|R Excited state of Heisenberg model: single site at R
has spin reversed with respect to ferromagnetic ground state
|Ri , χ0 (r − Ri ) Atomic state and wave function for a nucleus at Ri
a1 , a2 , a3 Vectors specifying unit cell • of real lattice
a∗1 , a∗2 , a∗3 • of reciprocal lattice
beff Effective field in rotating frame at magnetic resonance
E Electric field
jc Critical current density of a superconductor
jh Hole current density
h̄k Quasi-momentum or crystal momentum
k0 Incident wave vector
k1 Scattered wave vector
kh Quasi-momentum of hole
Appendix D Specific Notation 605
δk Displacement of Fermi surface in quasi-momentum space

under effect of a force
|k Eigenstate of Heisenberg model corresponding to a spin wave of wave vector k
m Magnetisation in the rotating frame
vn,k Average velocity of electron in Bloch state ψn,k
ve Drift velocity of conduction electrons
vh Hole velocity
vg Group velocity of wave packet
References
Quantum Mechanics and Statistical Physics
1. Balian, R.: From Microphysics to Macrophysics, Vols. I and II. Springer-Verlag, Berlin,
Heidelberg, New York, NY (1991)
2. Basdevant, J.L., Dalibard, J.: Quantum Mechanics. Springer-Verlag, Berlin, Heidelberg (2002)
3. Georges, A., Mézard, M.: Introduction á la théorie statistique des champs. Cours de l’Ecole
Polytechnique, Palaiseau, France
Solid State Physics
4. Ashcroft, N.W., Mermin, N.D.: Solid State Physics. Saunders College Publishing, Philadel-
phia, PA (1976)
5. Dugdale, J.S.: The Electronic Properties of Metals and Alloys. Edward Arnold, London (1977)
6. Ibach, H., Luth, H.: Solid State Physics. Springer-Verlag, Berlin, Heidelberg, New York, NY
(1995)
7. Kittel, C.: Introduction to Solid State Physics, 7th edn., Wiley, New York, NY (1996)
8. Olsen, J.L.: Electron Transport in Metals. Interscience, New York, NY (1962)
9. Voos, M., Drouhin, H.J., Drévillon, B.: Semi-conducteurs et composants. Cours de l’Ecole
Polytechnique, Palaiseau, France
10. Ziman, J.M.: Electrons and Phonons. Oxford University Press, Oxford (1960)
Superconductivity and Magnetism
11. Abragam, A.: Principles of Nuclear Magnetism. Oxford University Press, Oxford (1994)
12. Becker, R., Döring, W.: Ferromagnetismus. Springer, Berlin (1939)
13. Chikamuzi, S.: Physics of Ferromagnetism. Clarendon Press, Oxford (1997)
14. Cullity, B.D.: Introduction to Magnetic Materials. Wesley, Reading, MA (1972)
15. Evetts, J.: Concise Encyclopedia of Magnetic and Superconducting Materials. Pergamon Press,
Oxford (1992)
16. Lévy, L.P.: Magnetism and Superconductivity. Springer-Verlag, Berlin, Heidelberg, New York,
NY (2000)
17. Orlando, T.P., Devlin, K.A.: Foundations of Applied Superconductivity. Addison Wesley,
Reading, MA (1991)
607
608 References
18. Rose-Innes, A.C., Rhoderick, E.H.: Introduction to Superconductivity. International Series in

Solid State Physics, Pergamon Press, Oxford (1978)
19. Tilley, D.R., Tilley, J.: Superfluidity and Superconductivity. Graduate Students Series in
Physics, Institute of Physics Publishing, London (1990)
20. Tinkham, M.: Introduction to Superconductivity. McGraw-Hill Inc, New York, NY (1996)
Subatomic Physics
21. Rougé, A.: Introduction à ła Physique Subatomique. Editions de l’Ecole Polytechnique,

Palaiseau, France
Periodic Table of the Elements
Translated with the agreement of the author from the french edition of ref (16):
“Magnétisme et Supraconductivité” by L. Lévy, Savoirs Actuels, InterEditions,
CNRS Editions (1997).
1 HEX
– H
1s
110 –
Hydrogen
3 CC 4 HEX
1.76 Li 0.21 Be
[He]2s [He]2s 2
400 55.1 1000 166
Lithium 0.03 Beryllium
11 CC 12 HEX
1.46 Na 1.34 Mg
[Ar]3s [Ar]3s 2
150 37.7 318 82.3
Sodium Magnesium
19 CC 20 FCC 21 HEX 22 HEX 23 CC 24 CC 25 CUB 26 CC 27 HEX
1.97 K2.9 Ca10.8 Sc

3.41 Ti
9.04 1.46 V 16.6 Cr 5.0 Mn Fe – Co
[Ar]4s [Ar]4s 2 [Ar]3d4s 2 [Ar]3d 24s 2 [Ar]3d 34s 2 [Ar]3d 54s [Ar]3d 54s 2 [Ar]3d 64s 2 [Ar]3d 74s 2
100 24.6 230 – 359 – 380 – 390 – 480 – 400 127 420 130 385 –
Potassium Calcium Scandium 0.4 Titanium 5.35 Vanadium Chromium Manganese Iron Cobalt
37 CC 38 FCC 39 HEX 40 HEX 41 CC 42 CC 43 HEX 44 HEX 45 FCC
2.43 Rb 3.64 Sr 10.1 Y2.91 Zr
8.4 Nb 2.1 Mo 4.06 Tc3.3 4.6 Ru Rh
[Kr]5s [Kr]5s 2 [Kr]4d 5s 2 [Kr]4d 25s 2 [Kr]4d 45s [Kr]4d 55s [Kr]4d 65s [Kr]4d75s [Kr]4d 85s
56 21.5 147 45.7 256 – 250 – 275 61.6 380 Molyb- – 351 – 382 – 350 –
Rubidium Strontium Yttrium 0.5 Zirconium 9.25 Niobium 0.92 denum 7.8 Technetium 0.5 Ruthenium Rhodium
55 CC 56 CC 57 HEX 72 HEX 73 CC 74 CC 75 HEX 76 HEX 77 FCC
3.53 Cs 2.72 Ba
10.1 *La 2.4 Hf 5.84 Ta 1.22 W 2.4 Re 2.35 Os 3.15 Ir
[Xe]6s [Xe]6s 2 [Xe]5d6s 2 [Xe]4f 145d 26s 2 [Xe]4f 145d 36s 2 [Xe]4f 145d 46s 2 [Xe]4f 145d 56s 2 [Xe]4f 145d 66s 2
[Xe]4f 145d 9
50 18.4 110 42.3 132 Lanth- – – – 225 – 310 – 416 – 400 – 430 –
Cesium Barium 4.9 anum 0.13 Hafnium 4.4 Tantalum 0.015Tungsten 1.7 Rhenium 0.65 Osmium 0.14 Iridium
87 CC 88 – 89 FCC
– Fr – Ra –
† Ac
[Xe]7s [Xe]7s 2 [Xe]7s 26d
– – – – – –
Francium Radium Actinium
58 FCC 59 HEX 60 HEX 61 – 62 ROM 63 CC
– Ce – Pr – Nd – Pm – Sm – Eu
* Lanthanides 2
[Xe]4f 6s 2
[Xe]4f 6s 3 2
[Xe]4f 6s 4
[Xe]4f 6s 2
[Xe]4f 6s5 2
[Xe]4f 6s 2 6 2 7
139 – 152 – 157 – – – 166 – 107 –
Cerium Praseodymium Neodymium Promethium Samarium Europium
90 FCC 91 TET 92 ORT 93 ORT 94 MCL 95 –
† Actinides 4.69 Th – Pa
10.9
– U – Np – Pu Am
[Rn]6d 27s 2 [Rn]5f 26d 7s 2 [Rn]5f 36d7 s 2 [Rn]5f 46d7 s 2 [Rn]5f 67s 2 [Rn]5f 77s 2
100 – – – 210 – 188 – 150 –
1.37 Thorium 1.3 Protactinium 1.1 Uranium 0.08 Neptunium Plutonium Americium
2 HEX
– He
1s 2
26 –
Helium
5 TET 6 DIA 7 HEX 8 CUB 9 MCL 10 FCC
– B– C – N – – O – F Ne
[He]2s 2p [He]2s 2p 2 [He]2s 2p 3 [He]2s 2p 4 [He]2s 2p 5 [He]2s 2p 6
1250 – 1800 – 79 – 46 – – – 63 –
Boron Carbon Nitrogen Oxygen Fluorine Neon
13 FCC 14 DIA 15 CUB 16 ORT 17 ORT 18 FCC
1.26 Al – Si – P – S – Cl – Ar
[Ar]3s 2p [Ar]3s 2p2 [Ar]3s 2p3 [Ar]3s 2p4 [Ar]3s 2p5
2 6
[Ar]3s p
394 136 625 – – – – – – – 63 –
1.18 Aluminum Silicon Phosphorus Sulfur Chlorine Argon
28 FCC 29 FCC 30 HEX 31 ORT 32 DIA 33 ROM 34 HEX 35 ORT 36 FCC
_ Ni 0.67 Cu 0.6 Zn 0.62 Ga – Ge
– –As – Se – Br Kr
[Ar]3d 84s 2 [Ar]3d104s [Ar]3d 10 [Ar]3d104s 2p [Ar]3d 104s 2p 2 [Ar]3d104s 2p3 [Ar]3d104s 2p4 [Ar]3d104s 2p5 [Ar]3d104s 2p6
375 _ 315 81.6 234 110 240 120 360 – 285 – 150 – – – 85 –
Nickel Copper 0.9 Zinc 1.08 Gallium Germanium Arsenic Selenium Bromine Krypton
46 FCC 47 FCC 48 HEX 49 TET 50 TET 51 ROM 52 HEX 53 ORT 54 FCC
10 Pd 0.66 Ag 0.63 Cd 1.8 In
1.84 Sn 0.63 Sb – –Te – I Xe
[Kr]4d 10 [Kr]4d 105s2 [Kr]4d 105s2 [Kr]4d 105s2p [Kr]4d 105s2p2 [Kr]4d 105s2p3 [Kr]4d 105s2p4 [Kr]4d 105s2p5 [Kr]4d 105s2p6
275 _ 215 63.8 120 86.8 129 100 170 200 127 139 – – – 55 –
118
Palladium Silver 0.56 Cadmium 3.4 Indium 3.75 Tin Antimony Tellurium Iodine Xenon
78 FCC 79 FCC 80 ROM 81 HEX 82 FCC 83 ROM 84 CUB 85 – 86 FCC
6.68 Pt 0.7 Au 2.2 Hg 2.83 Tl 3.14 Pb
0.084 – Bi – Po – At Rn
14 10 2
[Xe]4f 145d 96s [Xe]4f 145d 106s [Xe]4f 145d 106s 2 4f 5d 6s p 4f145d106s2p2 4f145d106s2p3 4f145d106s2p4 4f145d106s2p5 4f145d106s2p6
230 170 64.2 100 82.6 96 94.6 88 110 120 115 – – – – – – –
Platinum Gold 4.16 Mercury 1.37 Thallium 7.23 Lead Bismuth Polonium Astatine Radon
64 HEX 65 HEX 66 HEX 67 HEX 68 HEX 69 HEX 70 FCC 71 HEX

_ Gd _ Tb _ Dy _ Ho _ Er
_ _Tm Yb 10.22 Lu
[Xe]4f 75d 6s 2 [Xe]4f 96s 2 [Xe]4f 106s 2 [Xe]4f 116s 2 [Xe]4f 126s 2 [Xe]4f 136s 2 [Xe]4f 146s 2 [Xe]4 f 145d 6s 2
176 _ 188 _ 186 _ 191 _ 195 _ 200 _ _ _
118 207
Gadolinium Terbium Dysprosium Holmium Erbium Thulium Ytterbium Lutetium
96 _ 97 _ 98 _ 99 _ 100 _ 101 _ 102 _ 103 _
_ Cm _ Bk _ Cf _ Es _ Fm _ Md _ No _ Lr
[Rn]5f 76d 7s2 [Rn]5f 76d 27s2 [Rn]5f 96d 7s2
Curium Berkelium Californium Einsteinium Fermium Mendelevium Nobelium Lawrencium
Z Crystal Structure
2 13 CFC
Sommerfeld constant (mJ/mole ) Symbol
1.29 Al
[Ar]3s2p Configuration
Debye temperature (K) 394 136
1.18 Aluminium
Fermi temperature (× 1000 K)
Tc (K)
Index
A Bohr radius, 593

Absorption, 209, 321, 328 Boltzmann constant, 593
Acceptor, 126 Bose condensation, 222
Alkaline earth metals, 204 Boson, 222, 349
Allowed energy band, 15 Bragg diffraction, 24, 37, 39
Alloy, 33 Bragg peak, 41, 43
Alloys, electronic energy and stability of, 391 Bragg plane, 39, 59, 66
Alternating magnetisation, 252 Bravais lattice, 26, 28, 38
Aluminium, Reflectance of, 373 Brillouin function, 244
Angle-resolved photoemission Brillouin zone, 66, 221
spectroscopy, 89
Antiferromagnetic Solid, 563 C
Antiferromagnetic Transition, 564 Causality, 321
Preliminaries Case, 563 Chemical shift, 340
Susceptibility, 565 Coercive field, 289
Antiferromagnetism, 252, 341 Coherence length, 185, 207, 222
Atomic force microscopy, 310 Collision, 103
Atomic vibration, 118 Collision time, 118
Atom with partially filled shell, 237 Compton wavelength, 593
Avogadro number, 593 Condensation energy, 210, 215
Conduction band, 77, 115
B Constant energy surface, 69
Band gap, 62, 67, 76, 84, 215 Contact interaction, 340
Band structure, 15 Conventional cell, 28, 30, 31
extended zone, 57 Cooper pair, 161, 213
restricted zone, 57 Critical current, 324
BCS theory, 215 Critical exponents, 357
Binding energy, 213 Critical field, 150, 178, 183, 186
Bitter method, 288, 307, 308 lower Hc1 , 191
Bloch function, 13, 14, 54, 111 upper Hc2 , 192
Bloch’s law, 350 Critical temperature, 147, 217, 259
Bloch theorem, 13, 53 Cryogenics, 204
Bloch wall, 283, 285, 322, 324 Crystal lattice, 26, 28, 38
energy of formation, 287 Crystallography, 205
thickness, 287 Crystal momentum, 16, 56
Body-centered cubic lattice, 29 Cubic lattice, 29, 32
Bohr magneton, 593 Cuprate superconductor, 33, 65, 259
613
614 Index
Curie paramagnetism, 235, 244 Fermi sphere, 73, 108

Curie temperature, 251, 353 Ferrimagnetism, 252, 255, 322
Curie–Weiss law, 353 Ferrites, 304
Cyclotron Resonance, 441 Ferromagnetic resonance, 325, 329
Electron State, 441 Ferromagnetism, 236, 249, 340
Metals, 444 hard, 290
Silicon, 442 soft, 290
Field effect (doping), 126, 130
D Fine structure constant, 593
Debye frequency, 220 First Brillouin zone, 16, 57, 59, 65
Debye temperature, 119 Flux quantum, 160, 161, 192
Debye–Waller Factor, 367 Fluxoid, 159
Defect, 116, 119 Form factor, atomic, 43
Demagnetisation energy, 282, 284 Fourier transform, 336
Demagnetising field, 279, 298, 303, 328 Free energy, 178, 184
Density functional theory, 7 Free precession, 335
Density of states, 17, 71, 86, 235 Fullerene, 33, 260
Diamagnetism, 177, 234
Diffraction, 35 G
Bragg, 37 Gap, 220
Dipole interaction, 244, 277, 337 Gauge, 159
Disorder, 44 Giant Magneto-Resistance, 264
Dispersion, 321, 328 Glass, 33
Dissipation, 320 Graphene, 29, 78, 126, 130
Donor level, 125 Group velocity, 108, 111
Drude model, 102 Gyromagnetic ratio, 324
E H
Easy magnetisation axis, 272 Hall effect, 126, 306
Effective mass, 114 Hall voltage, 297
Elastic collision, 119 Hartree approximation, 7
Electrical conductivity, 103, 108 Hartree–Fock approximation, 7
Electron, 593 Heisenberg model, 247
nearly free, 56, 61, 66 Helium, 4
Electron microscopy, 308 High Temperature Superconductors, 223, 259,
Electron mobility, 103 261, 306, 309
Electron–phonon interaction, 206, 220, 221 High Temperature superconductors, 32, 65, 89
Energy bands, 20 Hole, 121
Entropy, 179 Hopping integral, 10
EPR, 331 Hubbard model, 256
Ewald construction, 40 Hund rules, 240
Exchange anisotropy, 273 Hydrogen atom, 236
Exchange constant, 246 Hyperfine interaction, 340
Exchange energy, 285 Hysteresimeter, 304
Exchange Hamiltonian, 246 Hysteresis cycle, 276, 289, 304
Exchange interaction, 245, 354
Excited state, 208, 216 I
Impurity, 116, 119
F Insulator, 75
Face-centered cubic lattice, 30, 34, 43, 59, 69 Insulator–Metal Transition, 419
Faraday balance, 300 Alkali Elements and Hydrogen, 423
Faraday effect, 307 Hydrogen-Like Orbitals, 419
Fermi–Dirac statistics, 108 Insulator–Metal Transition in Si–P, 423
Fermi energy, 73, 108 Interactions Between Electrons, 420
Index 615
Interface, 188 Magnetisation, 299

Interface energy, 188 remanent, 289
Intermediate state, 303 rotation of, 274
Ising model, 359 of superconductors, 301
Isotope effect, 206, 220 Magnetocrystalline anisotropy, 272
Magneto-optical effects, 308
J Magnetoresistance, 297, 306
Josephson effect, 165, 217 Magnon, 348, 351
Josephson Effects in Zero Magnetic Field, 481 Matthiessen’s law, 118
Josephson Junction in a Microwave Field, Maxwell equations, 156
483 Mean field approximation, 251, 352, 359
Model Josephson Junction, 481 Mean free path, 106
Realistic Josephson Junction, 481 Meissner effect, 154, 177
Josephson Junction in a Magnetic Field, 491 Metal, 75, 341
Current Distribution, 492 Microstructure, 323
Josephson Plasma Resonance, 495 Miller indices, 38
Screening of, 493 Mixed state, 185, 188, 306
Joule effect, 118 Molecular field, 249
Monovalent metal, 221
K Monovalent Metals, Optical Response of, 399
Kerr effect, 307 Mössbauer effect, 341
Kramers–Kronig relations, 321, 334 Mott–Hubbard insulator, 256
L Multielectron atoms, 237
La2 CuO4 , 557 Muon, 339
Landau function, 355
Landau theory, 354 N
Landé factor, 240, 324, 333 NbSe2 , 531
Lanthanide, 204 Near-field microscopy, 310
Larmor frequency, 297, 324, 339 Néel state, 252
Larmor resonance, 150 Néel temperature, 252
Latent heat, 179 Neutron, 205, 593
Lattice planes, 27 NMR, 331, 339
Levitation, 155 high resolution, 334
Linear atomic chain, 11 Noble metals, 204
Linear response, 320 Nuclear magneton, 593
Liquid vortex phase, 306 Nuclear spin, 333
Little–Parks Experiment, 469
London equations, 152, 156 O
Lorentz microscopy, 309 Ohm’s law, 103, 156
Lorenz number, 107, 111 One-dimensional system, 5
Optical absorption, 84
M Order parameter, 356
Magnesium Diboride, 541 Orientational disorder, 34
Atomic and Electronic Structure, 541 Overlap integral, 19
Superconductivity, 543
Magnetic anisotropy, 272, 285, 330 P
Magnetic domain, 283 Paramagnetic insulator, 257
rotation of, 289 Pauli exclusion principle, 107, 213, 238, 245
Magnetic force microscopy, 311 Pauli paramagnetism, 178
Magnetic hysteresis, 273, 275, 289 Pauli spin paramagnetism, 235
Magnetic pole, 278 Pauli susceptibility, 341
Magnetic potential, 278 Penetration depth, 156, 182, 185, 192, 339
Magnetic susceptibility, 300 Periodic boundary conditions, 12
complex, 320 Perovskite structure, 32
616 Index
Persistent current, 149, 156 Superfluidity, 222

Phase transition, 352, 354 Susceptibility, Pauli, 82
Phonon, 118 Symmetry axes, n-fold, 28
Phonons in Solids, 453 Symmetry breaking, 248
Debye Model, 453
Detection of, 455 T
Einstein Model, 453 Thermal conductivity, 105
Resistivity, 457 Thermodynamic potential, 178
Thermodynamic Properties, 456 Thin Film, 579
Photoemission, 88 Non-uniform Situations, 581, 583
Planck constant, 593 Uniform Magnetisation, 580
Plastic crystal, 34 Thin film, 181
Potential, atomic, 6 Thin Films and Magneto-Optic Applications,
Pressure, 204, 205 573
Primitive cell, 26 Tight-binding approximation, 63, 64, 68
Proton, 593 Transfer integral, 10
Pseudogap phase, 259 Transition metals, 204, 221
Pulse response, 321 Transverse relaxation, 338
Triplet state, 239
Q TTF-TCNQ Compounds, 405
Quasi-momentum, 16 Dimerised Chain, 407, 413
Isolated Chains, 405
R Observations, 406
Rayleigh regime, 289 Peierls Transition, 409, 416
Reciprocal lattice, 35, 38, 43 Tunnel effect, 85, 217
Reflectivity, 209 Two-fluid model, 157
Relaxation time, 103, 336 Type I superconductor, 188
Renormalisation group, 359 Type II Superconductor, 511
Rotating frame, 326 Type II superconductor, 188
Rydberg constant, 593
V
S V3 Si, 523
Scanning tunneling microscopy, 85, 87 Vacancy, 120
Semi-classical approximation, 108 Valence band, 77
Semiconductor, 31, 76, 77, 204 Van der Waals interaction, 222
doped, 123 Vortex, 192
Semi-metals, 204 Vortex pinning, 198
Singlet state, 238
Slater determinant, 8 W
Solid solution, 33 Wave packet, 108
Specific heat, 81, 179 Wigner–Seitz cell, 27
Spin echo, 337 Wilson ratio, 83
Spin lattice relaxation time, 325
Spin–orbit interaction, 240 X
Spin singlet, 213 X ray, 35, 205
Spin wave, 348, 351
Spontaneous magnetisation, 235 Y
SQUID, 166, 217, 299 YBa2 Cu3 O7 , Band Structure, 377
Structural defect, 302 Chain and Plane, 380
Structure factor, 42 Isolated Copper–Oxygen Chain, 378
Superconductivity, 147, 177 Isolated Copper–Oxygen Plane, 379
Superexchange interaction, 252 Realistic Models, 381

(Graduate Texts in Physics) Henri Alloul (Auth.) - Introduction To The Physics of Electrons in Solids-Springer-Verlag Berlin Heidelberg (2011) PDF

Uploaded by

Copyright:

Available Formats

(Graduate Texts in Physics) Henri Alloul (Auth.) - Introduction To The Physics of Electrons in Solids-Springer-Verlag Berlin Heidelberg (2011) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Graduate Texts in Physics) Henri Alloul (Auth.) - Introduction To The Physics of Electrons in Solids-Springer-Verlag Berlin Heidelberg (2011) PDF

Uploaded by

Copyright:

Available Formats

What is the purpose of the Graduate Texts in Physics series?

What is the purpose of the Graduate Texts in Physics series?

What are some characteristics of the textbooks produced by Ecole Polytechnique?

What are some characteristics of the textbooks produced by Ecole Polytechnique?

GRADUATE TEXTS IN PHYSICS

Professor Richard Needs

Professor William T. Rhodes

Professor H. Eugene Stanley

Introduction to the Physics

Translated by Stephen Lyle

Original French edition published by Édition de l’École Polytechnique, Palaiseau,

ISSN 1868-4513 e-ISSN 1868-4521

Library of Congress Control Number: 2010932620

c Springer-Verlag Berlin Heidelberg 2011

Cover design: WMXDesign GmbH, Heidelberg

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Recent Trends in Condensed Matter Physics

Magnetism and Superconductivity

How do magnetism and superconductivity fit in to the general context of condensed

A Brief Guide to the Course

A Guided Choice of Problems

1 The Quantum Mechanics of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Crystalline Solids: Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3 Electronic Structure of Solids: Metals and Insulators . . . . . . . . . . . . . . . . 51

3.1.3 Electrons in a Weak Periodic Potential . . . . . . . . . . . . . . . . . . 56

4 Electron Transport in Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5 Introduction to Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

5.2 Difference Between a Perfect Conductor and a Superconductor:

6 Thermodynamics of Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

7 Microscopic Origins of Superconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . 201

7.4 High-Tc Superconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

8 Magnetism of Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

9 Magnetic Anisotropy, Domains, and Walls . . . . . . . . . . . . . . . . . . . . . . . . . 269

10 Measurements in Magnetism: From the Macroscopic

10.1.3 Hysteresimeters: Toroidal Geometry . . . . . . . . . . . . . . . . . . . . 303

11 Spin Dynamics and Magnetic Resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

12 The Thermodynamics of Ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

13 Problem Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

6.2 Experimental Observations . . . . . . . . . . . . . . . . . . . . . . . 406

Appendix A Physical Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593

Appendix B Some Useful Functions and Relations . . . . . . . . . . . . . . . . . . . . 595

Appendix C Standard Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599

Appendix D Specific Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601

Orsay, October 2010 Henri Alloul

1.1 General Hamiltonian for a Solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

H. Alloul, Introduction to the Physics of Electrons in Solids, Graduate Texts in Physics, 1

Image obtained by scanning tunneling microscopy of the surface of a single-crystal TTF-TCNQ

In this chapter, we discuss the quantum description of a solid. As we shall see in

1.1 General Hamiltonian for a Solid

1.1.1 The Need for Approximations

where we have replaced the Coulomb electron–nucleus and electron–electron inter-

1.1.2 Wave Functions and Pauli Exclusion Principle

In the second line of (1.4), we obtained a separable form of He , written as a sum

Ψ (r1 , r2 , . . . , rNe ) = ψk1 (r1 )ψk2 (r2 ) . . . ψkNe (rNe ) , (1.8)

with energy eigenvalue

1.2 Energy Bands

1.2.1 Tight-Binding Approximation

corresponding to the quantum states

H1 |Rl = Ev |Rl + Vl (r) |Rl , (1.14)

−tn,l = Rn | Vl (r) |Rl , (1.18)

ψk |ψk = δk ,k ,

Rn |Rl = αl−n = αn−l , (1.40)