Sutton - Electronic Structure of Materials PDF

Electronic Structure of
Materials
Electronic Structure
of Materials
ADRIAN P. SUTTON
Department of Materia/s, Oxford University
CLARENDON PRESS . OXFORD

Ox,ford Universily Press, Wallon Slreel, Oxford OX2 6DP
Oxford New York
Alhens Auck/and Bangkok Bombay
Calcurra Cape Town Dar es Salaam Delhi
Florence Hong Kong ISlanbul Karachi
Kuala Lumpur M adras Madrid M elbourne
Mexico Cily Nairobi Paris Singapore
Taipei Tokyo Toronto
and associated companies in
Berlin I badan
Oxford is a trade mark of Oxford Universily Press
Published in Ihe U niled Slales

by Oxford Universily Press Inc., New York
© Adrian P. Surron, 1993
Firsl published 1993

Reprinled with co"eclions, 1994
AII righrs reserved. No parl of Ihis publication may be

reproduced, stored in a retrieval system, or Iransmirred, in any
form or by any means, wilhoul Ihe prior permission in wri/ing of Oxford
Universily Press. Within Ihe UK, exceplions are allowed in respecl of any
fair dealing for Ihe purpose of research or privale stady, or cri/icism or
review, as permitted under Ihe Copyright, Designs and Palents Act, 1988, or
in the case of reprographic reproduction in accordanee with Ihe terms of
licenees issued by the Copyrighl Licensing Agency. E"4uiries coneerning
reproduction outside Ihose terms and in olher counlries should be sent to
the Rights Departmenl, Oxford Universily Press, at the address above.
This book is sold subjecl lo the condition that it shall nat,

by way of trade or otherwise, be lenl, re-sold, hired OUI, or otherwise
circulaled wilhoul the publisher's prior consenl in any form of binding
or cover olher than that in which il is published and wilhout a similar
condition inelading this condition being imposed
on the subsequent purchaser.
A calalogue record for this book is available from Ihe British Library
Library of Congress Cataloging in Publicalion Data

SuUon, Adrian P.
Electronic slroeture of materials / Adrian P. Sutton.
I neludes bibliographical references and indexo
1. Electronic slroeture. 2. Chemical bonds. 3. Solid state
physics. 4. Solid stale chemistry. l. Title.
QCI76.8.E4S875 1993 620.1'1299-<1c20 93-3940
ISBN O 19 851755 6 (Hbk) ISBN O 19 851754 8 (Pbk)
Printed in Greal Brilain by Bookcraft (Bath) LId

M idsomer N orton, Bath
Preface
Modem materials science is just as much concemed with deCects in crystals

and noncrystalline materials as it is with the properties oC perfect crystals.
For example, a materials science graduate should have sorne understanding
oCwhy ordinary window pane glass is transparent, or how the local electronic
structure at the surface oC a crystal differs Crom that oC the bulk. But in the
teaching oC introductory courses on electronic structure great emphasis is
still placed on k-space concepts such as Bloch's theorem and band theory.
These concepts are oC little help in understanding deCective crystals or
noncrystalline materials. In the past 20-30 years a real space picture oC the
electronic structure oC materials has emerged, which is more in line with a
chemist's way oC viewing materials than a physicist's, even though much oC
the work has been done by physicists. Rather than talking about the bands
oC the k-space approach the real space picture is in terms oC bond orders,
bond energies, and local densities oC states. A powerful theorem has been
derived which relates moments oC the local density oC states to the local
atomic environment. This has led to the realization that the local atomic
environment is the key to understanding deCective crystals and noncrystalline
materials. The aim oC this book is to provide an elementary introduction to
these modem ideas. However, the k-space approach has not been abandoned
and powerful relations between the two approaches in perfect crystals are
described.
I have assumed that the student has already been exposed to a course on
quantum mechanics, covering the hydrogen atom and the time-dependent
and the time-independent Schrodinger equations, although sorne oC this
material is reviewed in Chapter 1. Every effort has been made to keep the
mathematics as simple as possible. Dirac notation is used throughout the
book because I believe that it is much easier and transparent to work with
quantum states than it is to work with explicit Cunctions representing those
states. A reasonably self-contained description oC Dirac notation is given in
Chapter 2. The only other mathematics that appears in the book consists oC
elementary matrix algebra, partial differentiation, and surface and volume
integration. Although many oCthe ideas may be expressed in terms oC Green's
Cunctions their use has been avoided throughout.
Chapter 1 sets the scene by looking at the empirical database on cohesion
and structure. In recent years there have been two significant discoveries
which have brought a great deal oC order to the empirical database. The first
is the universal equation oC state Cor metals and the second is PettiCor's
structure maps Cor binary and pseudobinary compounds. The universal
vi Preface
equation of state is described in detaiJ beca use of its useful predictive powers,
but a detaiJed discussion of structure maps is deferred to Chapter 10. The
principal results of the solution of the Schr6dinger equation for the hydrogen
atom are reviewed, with particular emphasis on the consequences of
orthogonality of the eigenstates. Trends in the periodic table are surveyed,
including variations in the orbital energies along periods and down groups.
The classification of materials into conductors, semiconductors, and insu-
lators is introduced, and the notion of metal-insulator transitions is
mentioned, with a more complete discussion deferred to Chapter 12.
Chapter 2 begins with a review of Dirac notation. The LCAO picture of
bonding in the hydrogen molecule is described using both the time-
independent and time-dependent Schr6dinger equations. It is seen that the
hopping integral can be related to the probability per unit time of an electron
tunnelling from one atom to a neighbouring atom. The LCAO picture of
bonding in a heteronuclear diatomic molecule leads to the concepts of
covalency and ionicity, the latter leading to the concept of electronegativity.
The concepts of bond order and bond energy are introduced for the
heteronuclear diatomic molecule.
The concept of k-space in one dimension and energy bands is introduced
in Chapter 3 by considering chain molecules of increasing length from the
diatomic to the infinite. This leads naturally to Bloch's theorem, the concept
of k as a quantum number, Bloch functions and the one-dimensional
Brillouin zone, and the band structure as the Fourier transform of the
real-space hopping integrals. The relationship between the Fermi energy and
the number of electrons per atom is shown. The density matrix is introduced
as the link between band theory (reciprocal space picture) and chemical
bonds (real space picture). Bond order and bond energy in an infinite system
are illustrated for the infinite linear chain. The local density of states is
introduced and illustrated for infinite and semi-infinite linear chains. The
relationship between the local density of sta tes and the local atomic
environment is made explicit through the moments theorem. These ideas are
applied to a binary AB linear chain alloy. The relationship between the band
energy (i.e. the sum of energies of the occupied eigenstates of the whole
system) and the energies of individual bonds is also made explicit.
The generalization of the concepts introduced in Chapter 3 to two- and
three-dimensional crystals is carried out in Chapter 4. The formation of
energy bands in a square lattice and a simple cubic lattice in an atomic
s-state model is shown. The Brillouin zones in f.c.c. and b.c.c. lattices are
derived. The significance of k in the equation of motion of an electron under
an external force in a crystal is discussed, which leads to the concept of Bragg
reftection by the crystallattice. The concept of holes is introduced, and the
Fermi surface in two- and three-dimensional crystals is also discussed. The
density of states in two- and three-dimensional crystals is illustrated for a
square lattice and a simple cubic s-band model. The density matrix, bond
Preface vii
order, and bond energy are derived ror the simple cubic s-band model in the
lirnit of small band fillings where the free-electron approximation applies.
The application of the moments theorem in two- and three-dimensional
crystals is also discussed.
Chapter 5 is concemed with the atomic/chemical origins of band gaps. A
simple illustration of the atomic origins of sorne band gaps is given with the
example of a linear chain of atoms in which each atom is associated with
two s states. Chemical origins of band gaps are discussed with use of a binary
AB alloy. Band gaps arising from Peierls distortions are also discussed. The
classification of metals and insulators according to band theory is described
and the concept of the metallic bond as an unsaturated covalent bond is
introduced. The real space, chernical picture oC metals and insulators in terms
of unsaturated and saturated bonds is also discussed.
Chapter 6 is a case study in sp bonding using silicon as an example. The
prevalence of sp bonding in the periodic table is discussed and the varied
nature of crystal structures that it gives rise to is pointed out. The four
fundamental hopping integrals between s and p states on neighbouring atoms
are defined, and the angular dependences of the s-p and p-p hopping
integrals are described. The formation of sp hybrids and the structures that
tbey lead to are discussed. Simple analytic models of the electronic structure
oC crystalline silicon are derived and the condition for a band gap is expressed
in terms of a balance between the energy of bond formation and the energy
required to promote an electron from an s state to a p sta te. The band
structure of silicon in an Sp3 basis is discussed together with the bond orders,
bond energies, and promotion energies in the diamond cubic and hexagonal
diamond (wurtzite) structures.
The free electron and nearly free electron models are described in
Chapter 7. A simple argument is given for the need for a quantum mechanical
description of the electrons in a solido The simplest mode\ considers the
electrons confined to a box and the densities of states that this modelleads
to in one, two, and three dimensions are derived. The close similarity between
the free electron bands and the bands in sp bonded systems is pointed out.
The nearly free electron model is introduced and the equivalence of the nearly
free electron and LCAO pictures is emphasized. The pseudopotential is
introduced to explain why the electrons are nearly free. The failure of the
nearly free electron model for transition metals is also discussed. Screening
and the Thomas-Fermi screening length in a metal are introduced. Tbe
success of the independent electron approximation is explained in terms of
the elec~rons screening each other through exchange and correlation inter-
actions leading to the 'exchange-correlation' holeo
Properties of nearly free electron metals are discussed in Chapter 8. After
a brief review oC Fermi-Dirac statistics the concept of the contact potential
between two metals is described. The electronic contribution to the specific
heat of a metal is derived. The electronic and thermal conductivities of nearly
viii Preface
free electron metals are derived and the Wiedemann-Franz law is explained.
The Hall effect is described. The volume dependence of the cobesive energy
of a nearly free electron metal is discussed. Tbe small energies controlling
tbe relative stabilities of rival crystal structures at tbe same density are a1so
discussed. Tbis cbapter is tbe most conventional in tbe book, a1tbougb tbe
c10sing discussion of pair potentials does not normally appear in an
undergraduate texto
The transition metals are discussed in Cbapter 9. Friedel's rectangular
d-band model for tbe variation of tbe cobesive energy across tbe transition
metal series is described. Tbe variation of tbe equilibrium bond length,
cohesive energy, bandwidtb, and bulk modulus across tbe transition metal
series is modelled in the second moment approximation. Tbe derivation of
Finnis-Sinc1air potentials from tbe second moment approximation is a1so
given. Tbe nature of d-d bonding is discussed interms of tbe three
fundamental d-d bopping integrals and tbeir angular dependences. Tbe
cbanges in crystal structure across tbe transition metal series are related to
tbe fiftb and sixtb moments of tbe local densities of states. Tbe nature of tbe
metallic bond in a binary metallic alloy is contrasted witb tbe nature of tbe
ionic bond in an ionic solido
Cbapter 10 is conceroed witb tbe crystal structures ofbinary and pseudo-
binary compounds. Tbe cbapter begins witb a discussion of bybridization
involving s, p, and d functions. Tbe atomic factors influencing tbe structures
of compounds are enumerated. Tbe chapter concludes witb a discussion
of empirical structure maps and their use is illustrated for intermetallic
a1loys.
Modero, accurate, and predictive calculations of materials properties are
based on density functional tbeory and tbis is outlined in Chapter 11,
togetber witb a discussion of tbe Boro-Oppenbeimer and local density
approximations. Examples of density functional calculations for metals,
semiconductors, and insulators are given.
Cbapter 12 is conceroed witb tbe breakdown of band tbeory. Altbough
tbe topics covered are ratber advanced tbe treatment is qualitative, witb tbe
empbasis on simple pbysical understanding ratber tban matbematical rigour.
Electrons in noncrystalline materials are discussed and tbe concepts of a
continuous random network and electronic defects are introduced. Tbe
Weaire-Tborpe model for tbe existence of an energy gap in amorphous
silicon is presented. Tbe concept of electron 10ca1ization is introduced and
its importance in semiconductor devices is pointed out. Electron 10ca1ization
is described in general terms as a balance between a potential energy tending
to localize and a kinetic energy tending to delocalize. Polarons pro vide a
simple illustration of electron localization. Anderson localization is intro-
duced qualitatively. Metal-insulator transitions are discussed in terms of
competing terms. In tbis case tbe potential energy favours tbe insulating state
and tbe kinetic energy favours tbe metallic state. Tbe Mott and Hubbard
Preface ix
metal-insulator transitions are discussed and the polarization catastrophe
model is described.
There are 32 set problems which are designed for tutorial work to enable
Ihe student to test his or her understanding of the material, and also to
cxtend the material covered in the text into somewhat more advanced areas.
1 have also dreamt up 20 questions to help students prepare for the ordeal
of an examination.
The real space approach to electronic structure advocated in this book
stems from the pioneering work of the French school under J. Friedel and
Ihe Cavendish group under Volker Heine. My comparatively small involve-
ment in the field was germinated and sustained by an inspiring collaboration
wilh my friends David Pettifor and Mike Finnis. They are among the very
few solid-state physicists 1 have met with a deep interest in what 1 would
call materials science, and the contribution of their thinking is evident
throughout the book. Mike Finnis generously read the whole of the original
version of the manuscript and pointed out several mistakes and ineon-
sistencies. Carol Nichols made a number of perceptive suggestions on the
scope and content of sorne of the later chapters. Volker Heine gave me
constructive and encouraging comments on the content of the undergraduate
lecture course for which this book was written. 1 warmly thank Mike, David,
Caro~ and Volker. John Jakubovics, Ed Shelton and AHonso Ngan made a
number of small but important corrections, for which 1 am grateful.
1 would welcome comments, criticisms, suggestions, and corrections
from all readers, especially the students for whom this book was written.
Picase send them to me at the Department of Materials, Oxford University,
OXI3PH, UK.
Finally, my love and thanks go to Pat for her support and understanding.
Oxford A.P.S.
December 1992
Acknowledgements
1 am grateful to the following authors for permission to reproduce figures

or tables from their papers or books: Dr J. R. Smith (Figs 1.4 and 1.5, and
Table 1.1), Sir Alan Cottrell (Fig. 1.8), Professor D. G. Pettifor (Figs 1.6,
1.12, 1.13,8.5,9.9,9.11, 10.4, 10.5, and 10.6), Dr A. T. Paxton (Figs 6.11 and
11.2), Professor R. McWeeny (Figs 6.7 and 10.1), Professor W. A. Harrison
(Figs 1.7,6.10,7.6, and 12.5), Professor C. Kittel (Table 1.3), Dr Y. L. Moruzzi
(Fig. 7.7), Professor M. L. Cohen (Fig. 11.3), Professor S. C. Moss (Fig. 12.1),
Professor F. Wooten (Fig. 12.2), Professor E. A. Davis (Fig. 12.3), Professor
P. P. Edwards (Figs 12.8 and 12.11), Professor T. F. Rosenbaum (Figs 12.9
and 12.12), and Sir Brian Pippard (Fig. 4.7).
Table 1.1 is reprinted with permission from The American Physical Society
from J. H. Rose, J. R. Smith, F. Guinea, and J. Ferrante, Phys. Rev. B, 29,
2963 (1984). Table 1.3 is from Charles Kittel, Introduction to solid state
physics, 6th edition, Copyright © (1986) by John Wiley & Sons, Inc.,
reprinted by permission from John Wiley & Sons, Inc. Figure 1.4 is reprinted
with permission from The American Physical Society from J. H. Rose, J. R.
Smith, F. Guinea, and J. Ferrante, Phys. Rev. B, 29, 2963 (1984). Figure 1.5
is reprinted with permission from The American Physical Society from J. H.
Rose, J. R. Smith, F. Guinea, and J. Ferrante, Phys. Rev. B, 29, 2963 (1984).
Figure 1.7 is from ELECTRONIC STRUCTURE AND THE PROPER-
TIES OF SOLIDS by W. A. Harrison Copyright © 1980 W. H. Freeman
and Co. Reprinted by permission. Figure 1.8 is reprinted with permission
from The Institute of Materials from A. H. Cottrell, Introduction to the
Modern Theory of Meta/s, Institute of Metals, London (1988). Figure 1.12
is from D. G. Pettifor, Chapter 3 of Physica/ Metallurgy, eds R. W. Cahn
and P. Haasen, reprinted with permission from EIsevier Science Publishers
BY, Academic Publishing Division. Figure 1.13 is from D. G. Pettifor,
Chapter 3 of Physica/ Metallurgy, eds R. W. Cahn and P. Haasen, reprinted
with permission from EIsevier Science Publishers BY, Academic Publishing
Division. Figure 6.7 is reproduced by permission from Oxford University
Press from R. McWeeny, Cou/son 's Va/ence, Oxford University Press (1979).
Figure 6.10 is from ELECTRONIC STRUCTURE AND THE PROPER-
TIES OF SOLIDS by W. A. Harrison Copyright © 1980 W. H. Freeman
and Co. Reprinted by permission. Figure 7.6 is from ELECTRONIC
STRUCTURE AND THE PROPERTIES OF SOLIDS by W. A. Harrison
Copyright © 1980 W. H. Freeman and Co. Reprinted by permission. Figure
7.7 is reprinted from Y. L. Moruzzi, J. F. Janak, and A. R. Williams,
Ca/cu/ated e/ectronic properties of meta/s, Copyright (1978), pages 44, 45, 48,
xii Acknowledgements
49, 52, and 53, with permission from Pergamon Press Ud, Headington Hill
Hall, Oxford OX3 OBW, UK. Figure 8.5 is reprinted from Solid State
Commun., 49, D. G. Pettifor and M. A. Ward, 'An analytic pair potential for
simple metals', pp. 291-4, with permission from Pergamon Press Ltd,
Headington Hill Hall, Oxford OX3 OBW, UK. Figure 9.5 is reproduced with
permission from Academic Press Inc. (London) Ltd from D. G. Pettifor,
Solid State Physics, 40, 43 (1987). Figure 9.8 is reproduced with permission
from Institute of Physics Publishing Ltd from A. P. Sutton, M. W. Finnis,
D. G. Pettifor, and Y. Ohta, J. Phys. C: Sol. State Phys., 21, 35 (1988). Figure
9.11 is reproduced with permission from Academic Press Inc. (London) Ltd
from D. G. Pettifor, So lid State Physics, 40, 43, (1987). Figure 10.1 is
reproduced by permission from Oxford University Press from R. McWeeny,
Cou/son's Va/ence, Oxford University Press (1979). Figure 10.6 is reproduced
with permission from The Royal Society from D. G. Pettifor and M. Aok~
Phi/o Trans. R. SOCo Lond. A, 334, 439 (1991). Figure 11.2 is reprinted with
permission from The American Physical Society from A. T. Paxton, M.
Methfessel, and H. M. Polatoglou, Phys. Rev. B, 41, 8127 (1990). Figure 11.3
is reprinted with permission from The American Physical Society from M.
T. Yin and M. L. Cohen, Phys. Rev. B, 26, 5668 (1982). Figure 12.1 is
reprinted with permission from US Department of Energy from S. C. Moss
and J. F. Graczyk, Proceedings of the Tenth Internationa/ Coriference on the
Physics of Semiconductors, p. 661 (1970). Figure 12.2 is reproduced with
permission from Academic Press Inc. (London) Ltd from F. Wooten and D.
L. Weaire, So lid State Physics, 40, 1 (1987). Figure 12.3 is reproduced by
permission from Oxford University Press from N. F. Mott and E. A. Davis,
E/ectronic processes in non-crystalline materia/s, Oxford University Press
(1979). Figure 12.5 is from ELECTRONIC STRUCTURE AND THE
PROPERTIES OF SOLIOS by W. A. Harrison Copyright © 1980 W. H.
Freeman and Co. Reprinted by permission. Figure 12.8 is reprinted with
permission from P. P. Edwards and M. J. Sienko, J. Am. Chem. Soc., 103,
2967. Copyright (1981) American Chemical Society. Figure 12.9 is reprinted
with permission from The American Physical Society from T. F. Rosenbaum,
K. Andres, G. A. Thomas, and R. N. Bhatt, Phys. Rev. Letts., 45, 1723 (1980).
Figure 12.11 is reproduced with permission from The Royal Society of
Chemistry from P. P. Edwards and M. J. Sienko, Chemistry in Britain, 19,
39 (1983). Figure 12.12 is reprinted with permission from The American
Physical Society from H. F. Hess, K. DeConde, T. F. Rosenbaum, and G.
A. Thomas, Phys. Rev. B, 25, 5578 (1982).
Permission was sought (twice) from The Institute of Materials, London
for reproducing Figs 1.6, 10.4, and 10.5 but no reply was received. However,
author permission was obtained for all figures that have been reproduced in
this book.
Contents
1. Introduction 1
Aims of the book 1
The •universal' equation of state for metals 1
Structure maps 9
The hydrogen atom 9
Metals, semiconductors, and insulators 20
References 20
2. The diatomic molecule 21

Review of bras, kets, and all that 21
A homonuc1ear diatomic molecule: the hydrogen molecule 25
A heteronuc1ear diatomic molecule 32
Electronegativity 34
Bond energy and bond order 35
Reference 37
3. From the finite to the infinite 38

Chain molecules and k-space 38
Bond order in an infinite system 52
The density of states: total and local 58
Band energy and bond energy 63
The moments theorem 66
The binary AB alloy 72
References 73
4. Into two and three dimensions 74

Solids as giant molecules 74
The square lattice 74
The cubic lattice 78
Brillouin zones for the f.c.c and b.c.c. lattices 78
Equation of motion of an electron under an external force 80
Holes 84
The Fermi surface 87
The density of states in two- and three-dimensional crystals 88
The density matrix, bond order, and bond energy 92
The moments theorem applied to two- and three-dimensional
crystals 97
xiv Contents
5. Band gaps: origins and consequences 101
Introduction 101
Infinite linear chain with two s states per atom 103
Energy gap in a binary AB alloy linear chain crystal 106
PeierIs distortions 107
Metals, insulators, and the metallic bond 109
6. s-p bonding-a case study in silicon 112

s-p bonding 112
s-p bonding between two silicon atoms 113
Angular dependence of s-p and p-p hopping integrals 116
sp hybrids 117
Simple models of the electronic structure of tetrahedrally
bonded silicon 122
The band structure of silicon in a minimal atomic basis set 124
The bond order and bond energy in silicon in a minimal atomic
basis set 127
References 131
7. Free electron theory 132

Introduction to free electron theory 132
The free electron approximation 133
Electrons in a box 135
Density of states 138
Free electron bands and LCAO bands 139
The nearIy free electron model 144
The pseudopotential 151
Screening 153
Exchange and correlation 155
References 157
8. Properties of free electron metals 158

Fermi-Dirac statistics 158
Contact potential 159
Electronic specific heat 160
Electrical conductivity of metals 162
Thermal conductivity of metals 164
The Wiedemann-Franz law 164
The Hall effect 165
The cohesive energy oC simple metals and its volume dependence 166
Structural energy differences 169
Reference 171
Contenta xv
9. The transition metals 172
The transition metals 172
The Friedel model 174
The Friedel model in the second moment approximation 175
Finnis-SincJair potentials for computer simulations of
transition metals 179
d-d bonding 183
Changes in crystal structure across the transition metal series 187
Bonding in metallic alloys 189
References 193
10. Structural stability of oompounds 194

Hybridization and crystal structural stabiJity 194
Atomic factors influencing the structures of compounds 197
Structure maps 198
Applications of structure maps 201
References 203
11. Introduction to modern quantitative theory 204

Modero quantitative predictions of crystal structure and
stabiJity 204
The Boro-Oppenheimer approximation 205
Outline of density functional theory 205
Applications 209
References 214
12. Where band theory breaks down 215

Electrons in noncrystalline materials 215
The energy gap in amorphous silicon 219
Electron localization 221
Polarons 222
Anderson localization 224
Metal-insulator transitions, or, what is a metal? 225
References 235
Problems 236
Sample examination questions 250

Index 257
1
Introduction
Aims of the book

This book was written in order to replace two second-year undergraduate
courses that used to be taught in the Department of Materials at Oxford
University on free electron theory and band theory. Those courses considered
the electronic structure of perfect crystals and exploited the translational
symmetry of perfect crystals to set the problem up in reciprocal space. It was
recognized that they did not serve the needs of a modero degree in materials
science. Modero materials science is at least as conceroed with defects in
crystals and noncrystalline materials as it is with the properties of perfect
crystals. (In a phrase often attributed to F. C. Frank 'crystals are like
people-it is the defects in them that make them interesting'.) Band theory
and free electron theory have very little to say about such things, because
they rely on perfect translational symmetry and Bloch's theorem.
To understand the properties of defects in crystalline materials (interfaces,
dislocations,. point defects) and noncrystalline materials we have to take an
altogether different viewpoint and set the problem up in real space. These
ideas are not new. In the 1920s and 30s when the band theory of solids
was being worked out by solid state physicists, chemists were trying to
understand the properties of solids in terms of individual bonds and local
chemistry. Modero theory allows us to take either the real space viewpoint
or the reciprocal space viewpoint whenever it suits us best. That is the
approach taken here. At the end of this course you should understand the
'cans of spaghetti' that make up band structures as well as the relationship
between the local atomic environment and local electronic properties. Being
conversant with both the real space and reciprocal space viewpoints you will
be in a position to address, at least qualitatively, the electronic structure of
any material or defect you are likely to meet in your degree.
The 'universal' equation of state for metals

Before we get into electronic structure let us take a broad look at the
data base of cohesion and structure of metals. We do this for three reasons.
First, we want to set the scene by finding sorne rules that our electronic
structure theory will hopefully explain. Second, even if we cannot explain
the empirical results they are very interesting and important in their own
2 Introduction
right, because they enable predictions to be made. Third, empirical relations
are a valuable means oC ordering the data base and enabling us to see the
wood Cor the trees. We first take a look at the equation oC state oC metals.
In a fluid or a gas the equation oC state relates the pressure to the volume
and temperature oC the material. In the solid state the Cul! stress and strain
tensors replace the thermodynamic variables pressure and volume because
the material is no longer spherical!y symmetric on average. For example for
smal! elastic strains the elastic constant tensor relates the stress and strain
tensors, Corming Hooke's law. Hooke's law is thus an equation oC state Cor
smal! elastic strains and at a particular temperature, or range oC temperatures
where the elástic constants do not vary mucho In the solid state we can still
ask the question, what is the pressure required to change the volume oC the
material homogeneously? We will again find pressure, P, as a Cunction oC
volume, V, and temperature, T, that is P = P(V, T). In this section we shal!
reCer to P = P(V, T) as the equation oC state oC the material, but you should
be aware that in the solid state it is only part oC the Cull equation oC state
re1ating the stress and strain tensors and temperature. In particular the
equation oC state we shal! be discussing says nothing about the response oC
the material to a shear.
One oCthe most surprising discoveries oCthe 1980s was that the equations
oC state (i.e. P = P(V, T» oC many metals, when suitably scaled, Cal! on a
universal curve (Rose et al., 1984). Even more surprising was the discovery
that the same curve also describes metal-metal adhesion and chemisorption
oC selected materials, including sorne nonmetals. At the present time it is not
clear quite how universal the scaled relation is, i.e. whether it applies to all
metals. But it is sufficiently important that we should describe it, not least
beca use it enables us to see relations between the lattice parameter oC a metal,
its bulk modulus and cohesive energy, and the degree oC anharmonicity
shown by the metal, which is important Cor thermal expansiono
To begin let us introduce the Wigner-Seitz radius, rws, which is the radius
oC a sphere whose volume equals the volume per atom in the material. Thus
if there are N atoms per unit volume then
41t 3
N'-rws= 1. (1.1)
3
We shall use the Wigner-Seitz radius as a length scale which characterizes
interatomic separations in the metal.
Consider the binding energy, E b , oC the material as a Cunction oC rws , as
we change the volume oC the metal homogeneously. The binding energy is
the total energy oC the metal relative to the energy oC the same number oC
isolated atoms. The binding energy tends to zero when the atoms are
infinitely separated and it has a minimum, negative value at the equilibrium
value oC rws = rM' The binding energy at the minimum is E~. (See Fig. 1.1.)
Dift'erent materials have different binding energy curves because they have
The 'universal' equation of state for metals 3
Fig. 1.1 The binding energy, Eb , as a function of the Wigner-Seitz radius, 'ws,
for three metals. The minimum of each binding energy curve occurs at 'ws = ~s
and Eb = -IEgI·
different equilibrium values of the Wigner-Seitz radius and the binding

energy. The surprise is that they can be brought to lie on the same curve by
suitably scaling the length and energy axes.
Let us first scale the energy axis. The obvious thing to try is to 'normalize'
the energy axis by dividing by the equilibrium binding energy
(1.2)
The curves g(Tws) now have the same minimum at TwS of -1. (See Fig. 1.2.)
-i--t--1~~------~~=-----~rws
Fig. 1.2 The scaled binding energy, Eb/IEgI, as a function of the Wigner-Seitz
radius, 'ws, for the same three metals as in Fig. 1.1. The minima occur at
Eb/IEgI = -1, but different values of 'ws'
4 Introduction
Fig. 1.3 The definition of L in eqn (1.5).
We now have to scale the length axis so that all the curves in Fig. 1.2 Iie on
top .of each other. If we simply subtracted rWs from rws and plotted 9 as a
function of rws - rws the minima in Fig. 1.2 would all be brought to the
point (O, -1). But the curvatures of g(rws - rws) are different and therefore
the curves g(rws - rws) would still not overlap exeept at (O, -1). To take
account of the curvature we have to consider the second derivative of 9 at
rws = rws. In Fig. 1.3 we show a length L which is a measure of the distanee
over which the interatomíc forees act as the solid is expanded, and which is
determined by the second derivative of 9 at rws = rws. The length L is
obtained by approximating the mínimum by a parabola
_ 1 "(. ) (
9 (rws ) - - +. 9 rws
2
• )2
rws - rws (1.3)
which crosses the length axis when
2 )112 (1.4)
rws - rws ~ ± ( g"(rws) .
Therefore,
L = 2( 2 )1/2 = 2(2)112( IE:I )1/2. (1.5)
g"(rws) Eí:(rws)
The factor of 2(2)1 /2 is irrelevant to defining a length sea/e. We therefore
scale (rws - rws) by
/-
_ ( IE:I )112 . (1.6)
Eí:(rws)
The 'universal' equation of state for metals 5
o
-0.1 -
-0.2 -
".o
r.¡ -0.3
;;¡» OMo
» -0.4
oK
i.. -0.5 6Cu
oBa
~ -0.6 o Sm2•
:& vSm3+
~ -0.7
"
rn
-0.8
-0.9
-1.0
-1 o 2345678
Scaled separation a
Fig. 1.4 The scaled binding energy, Eb/IEgI, as a function ot the scaled length,
a, tor various metals. From Rose et al. (1984).
Thus, we define a scaled length a as foIlows

a = (rws - rws)/l (1.7)
and we plot the function g as a function of a. Thus we scale each Eb(TwS)
curve as foIlows
(1.8)
and it is then found that aIl the curves He on top of each other! This rnay
be seen in Fig. 1.4.
Well, so what! First of a11 this resuIt indicates that once the cohesive energy
IE~I, the bulk rnoduIus, B, and the Wigner-Seitz radius are known at the
equilibrium density then the energy-density relation over the whole range of
densities is known. The buIk rnodulus B enters the picture through eqn (1.6)
which defines the length scale
1= ( IE~I )1/2 = ( IE~I )1/2 (1.9)

Eí:(rws) 121trwsB
where we have used
6 Introduction
Secondly, the length scale / is proportional to the width of the binding
energy curve, and therefore the width of the region over which Hookean
forces act. A crystal is strongly anharmonic if the Hookean region is confined
to small dilations from the equilibrium volume. The anharmonicity is
therefore measured by '1 = rws/l
'1 = ,. // = ( ws )3)1/2 .
127tB(r
(1.11)
ws IEbl
Table 1.1 gives the scaling length /, the dimensionless parameter '1 which
measures the anharmonicity, and 'ws, the equilibrium Wigner-Seitz radius,
for a range of metals.
Clearly, the observation that all the scaled curves have a universal form
is not much use unless we know its functional formo The following describes
it accurately
g(a) = -(1 + a + O.05a 3 ) e-a. (1.12)
Thus a1l the binding energy curves are described by
(1.13)
The apparent universality of the equation of state enables us to make sorne
generalizations. Suppose we have two materials with the same cohesive
energy IE:I. This means that the minima of the energy volume curves have
the same depth but they may have different positions ,ws.
But in order for
the two curves to be brought into coincidence by rescaling we see from eqn
(1.9) that it is necessary that B,ws is the same in both materials. Thus the
material with the larger equilibrium Wigner-Seitz radius must have a smaller
bulk modulus to compensate. Similarly, for two materials that share the same
'Ws it is necessary that the ratio of the cohesive energy to the bulk modulus
is the same: the greater the bulk modulus the greater the cohesive energy.
These observations are what one· would expect if the packing density were
the only criterion for enhanced bonding. For most close packed metals this
is indeed the dominant factor controlling cohesion. But we might expect
these generalizations to break down when directional bonding is important
because the bond strength is not determined by packing density but by
rehybridization. This is why we believe the 'universal' equation of state may
be more appropriate for metals than other materials.
At O K the equation of state is given by
(1.14)
and it can be shown that
Tabla 1.1 Scaling length 1 (eqn (1.9». equilibrium Wigner-Seitz radius ~s. and dimensionless parameter '1 (eqn (1.11»
characterizing the anharmonicity for a range of metals in the periodic table. For each element the first line gives 1 (in
Á). the second line gives r"ws (in Á). and the third lines give 'l. From Rose et al. (1984)
Li Be B C N O F Na
0.553 0.312
1.719 1.25
3.10 4.01
Na Mg Al Si P S CI Ar
0.562 0.316 0.336 0.344
2.080 1.77 1.58 1.68
3.70 5.60 4.71 4.88
K Ca Se Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As Se Br Kr
0.651 0.483 0.340 0.310 0.254 0.274 0.262 0.270 0.272 0.215 0.348
2.573 2.18 - 1.62 1.49 1.42 - 1.41 1.39 1.38 1.41 1.54 - 1.76
3.94 4.52 4.76 4.81 5.59 5.16 5.31 5.11 5.20 7.16 5.05
Rb Sr Y Zr Nb Mo Te Ru Rh Pd Ag Cd In Sn Sb Te I Xe
0.658 0.47 0.395 0.336 0.265 0.245 0.237 0.269 0.214 0.360
2.750 - 1.99 1.77 1.63 1.55 - 1.48 - 1.52 1.60 1.73 1.84
4.18 4.23 4.48 4.84 5.85 6.04 6.41 5.94 8.08 5.11
Cs Ba La Hf Ta W Re Os Ir Pt Au Hg TI Pb Bi Po Al Rn
0.714 0.558 0.373 0.330 0.274 0.247 0.230 0.237 0.236 0.331 0.303
2.977 2.46 - 1.74 1.62 1.56 1.52 - 1.50 1.53 1.59 - 1.90 1.93
4.17 4.41 4.66 4.92 5.69 6.15 6.52 6.47 6.75 5.74 6.37
Fr Ra Ae
Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu
0.648 0.478 0.467 0.404 0.393 0.506
2.02 2.27 1.99 1.96 1.94 1.99
3.11
Th Lw
0.483
1.99
4.12
8 Introduction
- Tbeorv Stainless 1.0 Rb

4.0 o Expenment sreel
i...
i
:::!l 2.0
...
~ -
8 0.5
...
~
o~~~ __ ____
~ ~
O~~~--::~--"JL,J
1.0 0.8 0.6 0.4 1.0 0.8 0.6 0.4
1.0 Cu 1.0 u
i~ i...
...
O.S
-
8°·S
~
0bC--..L...--:L--~ 0bL.--:'-::---:L--~
1.0 0.8 0.6 0.4 1.0 0.8 0.6 0.4
VIVo
10.0 _ Theory 1.0
PI Ag
8.0 o Experiment
i 6.0 o
i~0.5
~ 4.0
... 2.0
...
O~~~_~_~~ O~-..L...--:L--~~
1.0 0.8 0.6 0.4 1.0 0.8 0.6 0.4
1.0 Pd 1.0 Mo
i~ O.S
...
o~ __ __ ____
~ ~ ~~
1.0 0.8 0.6 0.4 0.8 0.6 0.4

VIVo
Fig. 1.6 Predicted (solid line) equations of state (based on eqn (1.15» for eight
metals. The circles are experimental data points. From Rose et al. (1984).
where BC is the bulk modulus at the equilibrium density. (You are asked to
show this in Problem 1.) This equation of state is shown in Fig. 1.5 for eight
different metals where it is compared with experimental data. The stainIess
steel 347 was taken as an alIoy of 70 per cent Fe, 19 per cent er, 11 per cent
Ni. As can be seen, the overaII agreement is very good indeed, especiaIly
when we recall that the ónIy input data were the equilibrium Wigner-Seitz
radius, the cohesive energy and the bulk modulus aII extrapolated to OK.
The hydrogen ato m 9
Using the definition of the bulk modulus
B = - V(oP/oVh (1.16)
we find that the bulk modulus varies with pressure as follows
2.3
(oB/oPh = 1 + - r¡. (1.17)
3
In a harmonic crystal the bulk modulus is independent of pressure. We see
from eqn (1.17) that the more anharmonic the crystal the higher r¡ and the
greater the variation of the bulk modulus with pressure, as we might expect!
Structure maps
Structure maps are a means of ordering the data base of crystal structures
of binary and ternary alloys. The goal of a structure map is to represent the
structures of all binary compounds of a given stoichiometry in a two-
dimensional plot in such a way that all those alloys with one type of crystal
structure appear in one region of the plot, which is separated from other
regions where other crystal structures appear. In Fig. 1.6 we show a structure
map for binary AB alloys. What determines whether a binary alloy adopts
one crystal structure or another? Obviously this is a very complicated
problem to which there is no simple answer that applies to aH aHoys
irrespective of the type of bonding. Indeed earlier attempts to obtain
structure maps with the remarkable degree of structure separation shown in
Fig. 1.6 failed because people tried to incorporate too many physical
parameters such as size, electron affinity, electron/atom ratio, and so on. The
reason for the success of Fig. 1.6 is that it was obtained entirely empirically.
We shall discuss how this was done in Chapter 10. But for the present we
simply note that the structure maps are an extremely convenient summary
of the structural data base that enables the crystal structures of pseudobinary
ternary alloys to be predicted. The structure maps present a clear challenge
to electronic structure theory, namely, what is the theoretical reason for their
success?
The hydrogen atom

The Schrodinger equation for the hydrogen atom can be sol ved exactly. Its
importance is that it provides us with a qualitative understanding of other
atoms because all atoms are spherically symmetric. What is unique about
the hydrogen atom is that the potential felt by the one electron it contains
varies inversely with distance from the nucleus. By contrast the additional
e1ectrons in other atoms introduce further Coulomb interactions. This has
the effect of lifting the degeneracy of states with the same principal quantum
10 Introduction
o 10 20 30 40 70 80 90 100
_ lA - IIA Bu Lu Ro Tb Pm Ce No Es Cm Np Th IVA VA VIA VIlA VIDa VIIIb VIIk lB DB IlIB IVB B VB VIB e VllB NOFH / /
/
F H ••• %
100 'iD Yo
:/
VD. / ..
~
,
90 VB
80 *.
[V8
dm
•
IIB
¡
70
m
VIne
VllIb
VIlla
60 VDA
,,/(. VlA /
50
VA
¡VA
/
/
/
/
i.
40
Th
Np
Cm
/
/
/
/
/
r 8
"'
No
eo /
/
/
30 /
Pm /
Tb /
/
Ho /
20 Lo /
/
"u
DA /
/
•
1 /
10 lA
/
¡ /
/
/
AB
Fig. 1.6 The sl'uclu,e map lo, binary AB alloys. See ehaple, 10 lo, lurthe,
details. F,om Pettilo, (1988).
number but different angular momenta in multielectron atoms. But the
angular character of states in multielectron atoms is the same as in hydrogen.
It is the angular character of the sta tes that is responsible for directional
bonding, which may then stabilize particular crystal structures.
We first recall that each quantum state of the hydrogen atom is labelled
by four quantum numbers:
1. n = 1, 2, 3, ... : the principal quantum number. The energy of the state is
proportional to l/n 2 and is independent of the other quantum numbers
in hydrogen.
2. 1 = O, 1, 2, ... , n - 1 is the angular momentum quantum number. States
with 1 = O, 1, 2, 3 are called s, p, d, f states respectively. The angular
momentum is given by JI(I + l)h/21t.
3. m = -1, -1 + 1, ... ,1- 1, 1 is the magnetic quantum number. This
quantum number gives the component of the angular momentum vector
in a particular direction which is arbitrarily designated as the z-axis. The
angular momentum in this direction is mh/21t.
4. m. is the electron spin and is 1/2 or -1/2 corresponding to up or down
spin.
We shall ignore the electron spin from now on except to say each In, 1, m)
state may be occupied by two electrons, one with spin up and one with spin
down. Thus, an s shell may contain 2 electrons, a p shell may contain 6
electrons, a d shell may contain 10 electrons, and an f shell may contain 14
electrons. A 'shell' means the set of states with the same principal quantum
number n.
Owing to the spherical symmetry ofthe hydrogen atom the wave functions
are separable
(1.18)
where R.,(r) is the radial component and lím(9, q¡) is a spherical harmonic
describing the angular component. Since these solutions are eigenfunctions
of a hermitian operator they are orthogonal to each other. This means that
f 'I'.'m(r, 9, q¡)'I'.·"m,(r, 9, q¡) dr =O (1.19)
unless n = n', 1 = 1', m = m' in which case the integral equals one. If the states
have different angular momenta then the orthogonality condition is satisfied
by the angular integration in eqn (1.19) because the spherical harmonics form
an orthonormal seto But if they have the same angular momentum quantum
number 1and magnetic quantum number m then the radial components have
r'
to satisfy the condition that
2
R.,(r)R.,,(r)r dr = O (1.20)
Fig. 1.7 The three s states oflowest energy for atomic hydrogen. The orbitals.
multiplied by r. are plotted as a function of distance. r. from the nucleus. From
Harrison (1980).
which can be satisfied only if there are nodes in the radial solutions. For
1 = O, i.e. s states, the radial Is, 2s, 3s, 4s solutions have O, 1,2,3 nodes
respectively, as shown in Fig. 1.7. Thus the orthogonality constraint intro-
duces oscillations into the radial wave functions, which increases their kinetic
energy because this is proportional to the Laplacian of the wave function.
The spatial extent of the wave functions, as measured by the position of the
final maximum or minimum, increases with the principal quantum number
n. In other words the orthogonality constraint forces higher sta tes further
from the nucleus: the exclusion principIe in action! On the other hand, the
2p, 3d, and 4f radial wavefunctions are nodeless because there are no lower
Iying p, d, or f sta tes respectively to which they have to be orthogonal. Thus
the np radial orbitals have n - 2 spherical nodes and the nd radial orbitals
have n - 3 sphericaI nodes. For this reason the 3d radial wave function is more
contracted than the 4s or 4p radial wave functions, which is highly sig-
nificant for understanding the properties of the transition metals. In Fig. 1.8
we show the probability 1'1'21 for the 3d, 4s, and 4p hydrogen atomic orbitals.
The state ofthe angular momentum ofthe eIectron determines the angular
dependence of the wave function and hence the angular dependence of the
probability distribution 1'1'21. The s orbitals have zero angular momentum
(l = O and m = O since Iml ~ 1). The wave function for an s orbital is
sphericalIy symmetric and is illustrated in Fig. 1.9. The p orbitals have one
unit of angular momentum, 1 = 1, and there are three p states corresponding
to m = -1, O, + 1: Yt -1' Yto, Yt l' However, we do not have to work with
these spherical harmonics if we do not want to, especialIy since two of them
are complex. We can always work with more convenient linear combinations
The hydrogen atom 13
o 5 10 15
Radial distance (A)
Fig. 1.8 The densities of 3d, 45, and 4p orbitals in hydrogen as a function of
distance from the nucleus. From Conrell (1988).
Fig. 1.9 The 1 s atomic orbital for the hydrogen atom.
of them if we prefer. It turns out that it is usually more convenient to work

with angular dependencies proportional to the component of radial distance
from the nucleus along one of the three orthogonal axes x, y, or z
Xlr
'I'nlm =
3 1/2
(411:) R. (r) 1
¡
ylr.
zlr
(1.21)
We shall use these forms extensively. There are three p orbitals oriented
along the x, y, z axes as shown in Fig. UO. Note that the three orbitals look
the same, except for the different orientations, and that they have odd parity,
Le. they change sign under an inversion. Unlike s orbitals, p orbitals are
directional. This is the origin of directional bonding in diamond and graphite
for example. Note that the sum of the squares of the three p orbitals produces
a function with spherical symmetry.
Linear combinations of the spherical harmonics corresponding to the five
d orbitals for 1 = 2 may be taken and a particularly convenient set is the
14 Introduction
z z z
y y + y
+ x --",,---lf-..e:;...-- x ---::01("---- x
Px Py p,
Fig. 1.10 Px' Py• and Pz atomic orbitals.
following
(3z 2 - r2)/(2(3)1/2r2)
(x 2 _ y2)!(2r2)
15)1/2
'1'.2.. = ( 4n R.ir) xy/r 2 (1.22)
yz/r2
zx/r2
These orbitals are shown in Fig. 1.11. Notice that they have even parity (ie.
they remain the same under an inversion). Despite their appearance in
Fig. 1.11 the orbitals are all equivalent in the sense that if we choose a
dift'erent coordinate system, a set of orbitals identica1 to eqn (1.22) is obtained
by linearly combining the original set. For example
z z
--"""'~--X--~IE:---X
dxy d",
Y
-_c:ij~ __ X
Fig. 1.11 dxyo d yz • d zx • d .. _,... and d.. atomic orbitals.

The angular character of the d orbitals is important for understanding the
transition metals and their compounds.
The angular momentum component m about the z-axis is easily found
from Figs. 1.9-1.11. Consider the Px orbital in Fig. 1.10. Take a circular path
around the z-axis at distance r. In one circuit of length 21tr we pass through
each of the positive and negative lobes of the Px orbital once. Thus this path
is of length one wavelength (A.). The linear momentum is given by the de
Broglie relationship p = h/A. = h/21tr. The angular momentum about the
z-axis is therefore rp = h/21t and hence m = ± 1. The same is true for the P.
orbital. By contrast s, Pz' and d 3z 2_,2 have circular symmetry about the z-axis
and therefore m = O. The orbitals d x • and dX 2_.2 change sign twice in one
circuit about the z-axis and therefore m = ± 2.
Moving onto the other e1ements of the periodic table we recall that the
degeneracy of states with the same principal quantum number but different
angular momenta is lifted by the presence of electron-electron interactions.
But despite the great complication of all the electron interactions it is still
possible to think of each electron moving independently of all the other
electrons. How? Well, we ignore the details ofthe electron-electron Coulomb
interactions and say that each electron moves in the electrostatic field of the
nuc1eus and all the other electrons. For example, for neon, which has 10
electrons, we say that each electron feels the nuc1eus and an average potential
due to the remaining nine electrons. The potential exerted by the remaining
nine electrons depends on their distribution, which is determined by their
wave functions. Thus the quantities we try to calculate, namely the wave
functions, also define in part the potentiaI acting on each wave function in
the SchrOdinger equation. This is an example of a self-consistent field
problem, which we discuss in Chapter 11. It is solved when the wave funetions
it genera tes give a potential that, when put back into the Schr6dinger
equation, yields the same wave functions! The angular dependence of the
wave function is just the same as for the hydrogen atom, so we will have the
s, p, d, . .. c1assification. Only the radial dependences will be different,
aIthough there will still be radial quantum numbers n, but with different
energies from those of the hydrogen atom. The usual order of the energies
in a given atom is as follows
Is < 2s < 2p < 3s < 3p < 3d ~ 4s, .... (1.23)
The reason why the energies in any given n shell inerease in the order ns,
np, nd, nf is easy to see. The increasing number of planar nodes in these
orbitals (O, 1,2,3 respectively) forces the e1ectron to spend an increasing
amount of time further from the nuc1eus where the attractive electrostatic discutir
field of the nuc1eus is more screened by the other electrons in the atom.
The electron configurations of atoms are then determined by the 'aufbau'
principie, according to which the lowest available states are filled subject to
the exc1usion principie, which asserts that no orbital may contain more than
16 Introduction
two electrons, with opposite spin. Thus the one electron in hydrogen goes
into a Is state which is thus only half occupied. Hydrogen can therefore form discutir el
bonds with other materials beca use there is space in the 1s shell for one significado
more electron. In helium the second electron fills the ls shell and thus helium de la frase
is chemically inert. In lithium, which has a nuclear charge of 3, the 3 electrons
again occupy the lowest available sta tes. Two of them go into the ls state
as in helium. The third goes into the n = 2 shell, but does it go into 2s or
2p? It goes into the 2s shell beca use the energy of the 2s shell is lower than
the 2p shell. Table 1.2 shows the electron configurations of the first 36
elements. You can see the pattem.
The chemical behaviour or valence of the elements is determined by the
outermost electron shell. The chemical valency reftects the extent to which
the outermost electrons are screened from the nuclear charge by the other
electrons in the atom. The + 1 valency of an alkali metal stems from the
lone outermost electron, which is more effectively screened than the lower
lying electrons. The outermost electron is thus more easily removed by
ionization. The - 1 valency of a halogen stems from the single vacancy in
the outer shell which is an attractive location for an added electron. The
outer shell of an inert gas atom is filled and its electrons are on average
equally screened from the nucleus both by each other and by the other
electrons in the atom. The inner filled shells are often referred to as core
electrons and the outer partially filled shell(s) are called valence electrons.
Core electrons do not participate in the formation oC bonds because their
energies are too low. The distinction between core and valence electrons
breaks down in the transition metals where the 4s shell is filled before the
3d shell.
In Fig. 1.12 we show the valence s and p energy levels computed for all
the elements self-consistently. We see that the 2s level is well below that of
the 2p level for the first row elements B to Ne. Notice that E, and Ep decrease aprox.
linearly across a given period, which is a result of the increasing nuclear
charge. If we had considered an ion oC nuclear charge + Ze with one electron
we would have found that E, and Ep would have decreased quadratically
with Z. The reason for the linear dependence of E, and Ep on Z is the
presence oC the other electrons. As we move down a given group the principal
quantum number of the valence shell increases and the valence s and p energy
levels become less strongly bound. As in the case oC hydrogen this is a
consequence oC the constraint that the valence states must be orthogonal to
all the lower lying core states, which raises the kinetic energy of the valence
state and pushes it Curther out Crom the nucleus. But there is an exception
to the left of group VB where the 4s level drops below the 3s leve!. This is
a consequence ofthe occupation ofthe 3d shell whose electrons do not screen
the core from the 4s electrons. The 4s electrons thereCore feel a more attractive
potential than their 3s counterparts in the preceding row. Finally, it is seen
that Ep - E, decreases as one goes Crom rigbt to leCt across the periodic tableo
Table 1.2 The electron configurations of the first 36 elements
Electron configuration
Z Element 15 25 2p 35 3p 3d 45 4p 4d 4f
1 H hydrogen 1
2 He he/ium 2
3 Li lithium 1
4 Be bery/lium 2
5 B boron 2 1
6 C carbon Filled 2 2 Number of electron5
7 N nitrogen (2) 2 3 in each 5tate
8 O oxygen 2 4
9 F f1uorine 2 5
10 Ne neon 2 6
11 Na sodium 1
12 Mg magnesium 2
13 Al a/uminium 2 1
14 Si silicon Filled 2 2
15 P phosphorus 2 3
16 S sulfur (2) (8) 2 4
17 CI ch/orine 2 5
18 A argon 2 6
19 K potassium
:
I 1
I
20 Ca ca/cium I 2
I
21 Sc scandium I 1 2
I
22 Ti titanium I 2 2
I
23 V vanadium Filled I 3 2
I
24 Cr chromium I 5 1
I
25 Mn manganese (2) (8) (8) I
I
5 2
26 Fe iron I
I
6 2
27 Co coba/t I
I
7 2
28 Ni nicke/ I
I
8 2
29 Cu copper I
I
10 1
30 Zn zinc I
I
10 2
31 Ga ga/lium 2 1
32 Ge germanium Filled 2 2
33 As arsenic 2 3
34 Se se/enium (2) (8) (18) 2 4
35 Br bromine 2 5
36 Kr krypton 2 6
18 Introduction
Or-~~--~~~~~~~-.~~--~~~~~~~-'
~ -10
1
rg
.!O!
-20
lAnA UD llIB IVB VBVIB VllB 2p
-<
Li Be_ B C N O F Ne
Na Mg- - Al Si P S a A
K ea·····ZnOaOe As Se Br Kr
Rb Sr _.- Cd In Sn Sb Te 1 Xe
Ca Ba-"-Hg Tí Pb Bi P Al Ro
Fig. 1.12 The valence S and p energy levels. From Pettifor (1983).
This is significant because a small vaIue of Ep - E, will enable s and p

electrons to hybridize and form compounds that cannot be stabilized with
only s or p electron bonding, as described in Chapter 6.
In Fig. 1.13 we show the self-consistent valence s and d energy levels across
~- -10
~
3d
·Ií 4d
~ -20 mA lVA VA VIA VDA .. vm... lB IIB
Se Tí V CrMnFeCoNiCoZn
y Zr NbMo Te RuRh PdAg Cd
Fig. 1.13 The valence s and d energy levels across the 3d and 4d transition
series. From Pettifor (1983).
Tabla 1.3 The electrical conductivity and resistivity of metals at 295 K. The conductivity (upper entry) is in units of 105
(ohm cm) -1 and the resistivity (Iower entry) is in units of 10- 6 ohm cm. From Kittel (1986)
Li Be B e N o F Na
1.07 3.08
9.32 3.25
Na Mg Al Si P S el Ar
2.11 2.33 3.65
4.75 4.30 2.74
K Ca Se Ti V er Mn Fe Co Ni Cu Zn Ga Ge As Se Br Kr
1.39 2.78 0.21 0.23 0.50 0.78 0.072 1.02 1.72 1.43 5.88 1.69 0.67
7.19 3.6 46.8 43.1 19.9 12.9 139 9.8 5.8 7.0 1.70 5.92 14.85
Rb Sr Y Zr Nb Mo Te Ru Rh Pd Ag ed In Sn(w) Sb Te I Xa
0.80 0.47 0.17 0.24 0.69 1.89 0.7 1.35 2.08 0.95 6.21 1.38 1.14 0.91 0.24
12.5 21.5 58.5 42.4 14.4 5.3 14 7.4 4.8 10.5 1.61 7.27 8.75 11.0 41.3
es Ba La Hf Ta W Re Os Ir PI Au Hg,;q. TI Pb Bi Po Al Rn
0.50 0.26 0.13 0.33 0.76 1.89 0.54 1.10 1.96 0.96 4.55 0.10 0.61 0.48 0.086 0.22
20.0 39 79 30.6 13.1 5.3 18.6 9.1 5.1 10.4 2.20 95.9 16.4 21.0 116 46
Fr Ra Ae
1"- Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu
0.12 0.15 0.17 0.10 0.11 0.070 0.090 0.11 0.13 0.12 0.16 0.38 0.19
""- 81
Th
67
Pa
59
U Np
99
Pu
89
Am
134
Cm
111
Bk
90.0
ef
77.7
Es
81
Fm
62
Md
26.4
No
53
Lr
0.66 0.39 0.085 0.070
15.2 25.7 118 143
20 Introduction
the 3d and 4d transition metal series. The two main features to note are:
(i) the energies of the s and d states decrease rapidly once the d shell is filled
at group lB where we find the noble metals. (ii) The valence 4d level becomes
progressively more strongly bound than the 3d level across the series. This
is because the 3d orbitals are so spatially compact, as we have already
remarked, and therefore their occupation leads to a significant Coulomb
repulsion termo
Metals, semiconductors, and insulators

Metals are distinguished by their high electrical conductivity. Table 1.3 shows
the electrical conductivity of metals at 295 K. This definition of a metal
is not very precise since sorne insulators have very small, but finite
conductivities at room temperature. Two further points help to define a
metal. Firstly, the conductivity is finite as the temperature approaches
absolute zero. By contrast the conductivity of a semiconductor or insulator
tends to zero as T -+ OK. Secondly, the electrical conductivity of a metal
decreases as T is increased, in contrast to a semiconductor or insulator. We
see in Table 1.3 that most elements of the periodic table are metals. The
nonmetals are on the right of the table.
We shall find that the 'metallic state' is referred to as a state that all matter
can in principIe be transformed into. Even an insulator can be transformed
into a metal if it is compressed sufficiently. Conversely a metal can be
transformed into an insulator if it is dilated sufficiently. We shall discuss the
basis of these transitions in Chapter 12. Thus our classification of matter
into metals, semiconductors, and insulators is dependent on the ambient
pressure. In the centre of a large planet like Jupiter we might find that even
the hydrogen there is metallic!
References
Cottrell, A. H. (1988). Introduction to the modern theory of meta/s. The Institute oC
MetaIs, London.
Harrison, W. A. (1980). E/ectronic structure and the properties of solids. W. H.
Freeman, San Francisco.
Kittel, C. (1986). Introduction to solid state physics (6th edn). John Wiley, New York.
Pettifor, D. G. (1983). Physica/ metallurgy (ed. R. W. Cahn and P. Haasen), Chapo 3.
Elsevier, Amsterdam.
Pettifor, D. G. (1988). Mat. Sci. Techno/., 4, 675.
Rose, J. H., Smith, J. R., Guinea, F., and Ferrante, J. (1984). Phys. Rev. B, 29, 2963.
2
The diatomic molecule
Review of bras, kets, and all that

In this section we shall review sorne of the mathematical ideas you should
have been exposed to in a course on quantum mechanics. If you find that
you still do not understand the material described in this section and would
like to read more, I suggest you try Chapter 8 of Feynman et al. (1966). It
looks weird at first but it is really very simple. You just have to become
familiar with it through using it and solving problems.
Before we get into quantum mechanics let us take a fresh look at something
that we are already familiar with and which is c10sely re1ated to the
mathematics we need to deve1op, namely vectors. A vector is something that
has both magnitude and direction. If I ask you to tell me the direction of a
vector you will be hard-pressed to do so unless we agree on sorne coordinate
system. It would be impossible to specify a wind direction if we did not have
sorne convention like the directions of the compass. Let us agree on a
coordinate system specified by the three basis vectors e" e2' e3 and for
simplicity we ensure that the basis vectors are orthonormal. Then a vector
v is represented in this basis as follows
(2.1)
In this basis v is represented by «v'e,), (v " e2), (v"e 3», where (v"e,), (v " e2),
and (V"e3) are simply numbers.
Now let us switch gears and consider how we represent a quantum state.
A wave function 'I'(r) is a representation of the quantum state 1'1') in real
space. The 1'1') is called a 'ket'. At each point r in space the quantum state
1'1') is represented by the function 'I'(r). Alternatively the quantum state
could be expanded in a set of orthonormal basis states {ltfJ)}
os también pueden tener una
ón en el espacio tridimensional. 1'1') = L ItfJ)<tfJl'I')" (2.2)
alll<P>
When we say that the set of basis states is orthonormal we mean that the
overlap <0"1") between any two members lO") and 1") ofthe set is zero unless
they are the same basis state in which case the overlap is 1. The overlap
between two states 10") and 1") is expressed as <0"1") where <0"1 is called a
'bra'. In eqn (2.1) the space in which the vector v and the basis vectors
lie is our familiar three-dimensional world and it is called a vector space. If
22 The diatomic molecule
there were N linearly independent basis vectors, rather than three, then the
vector space would be N dimensional. This would present us with no more
mathematical difficulties than working in three dimensions, but we would
not be able to draw pictures to represent the vectors in Cour and higher
dimensional spaces. In eqn (2.2) the •space' spanned by the basis states {It/J)}
is called a Hilbert space. If the set oC basis sta tes is finite the sum in eqn (2.2)
comprises a finite number oC terms and the Hilbert space has the same finite
number oC dimensions. Otherwise the Hilbert space has an infinite number
of dimensions. The expansion coefficients are the <t/J 1'1'), which are simply
numbers. They are the components oC the state 1'1') in the basis set {It/J)}.
For this reason we call 1'1') a state vector. The wave Cunction 'I'(r) is the
representation oC the state vector 1'1') in real space: 'I'(r) = <rl 'Y). In this
case the real space basis {Ir)} is continuous and infinite dimensional because
there is an infinite set oC vectors r. The parallel between eqn (2.1) and eqn
(2.2) is striking. It is even more obvious iC we rewrite eqn (2.1) as Collows
3
V =I e¡(e¡·v). (2.3)
i=l
It is important to realize the distinction we have made between a state

vector 1'1') and its representation in sorne basis. It is the same as the
distinction between a vector v and its representation in sorne basis set oC
vectors. The reason why vectors are so important mathematically is that
many oC their algebraic properties can be expressed without reCerence to any
particular basis set. This is the stuff oC 'vector algebra'. For example, the
scalar product a' b is independent oC the basis set chosen to represent
a and b. So, a vector can also be thought of as a purely abstract
entity obeying certain algebraic rules. We can always make it more concrete,
and visualize it, by representing it in sorne coordinate system iC we
want too But it would not only be cumbersome to keep writing a'b as
(a'e¡)(b'e¡) + (a'e2)(b'e2) + (a'e3)(b'e3) but strange since it is true Cor any
orthonormal set oC basis vectors el' e2, e3' Similarly it is neater and oCten
helpCul to use the abstract mathematical properties oC state vectors rather
than expressing them in sorne fixed basis seto An example oC this abstraction,
which is extremely useful, follows immediately from eqn (2.2). Since the
state vector 1'1') appears on both sides of the equation it follows that
L.1114» 14»<4>1 is the identity operator
I
81114»
1t/J)<4>1 = l.
.
(2.4)
We can use this result immediate1y to express the overlap between two
state vectors 1'1') and I~)
('1'10 = ('1'1110 = <'1'1 I It/J)<t/JIIO = I ('I'It/J) (t/JIO· (2.5)

81114» 811 14»
Review of bras. kets. and all that 23
The ordinary three-dimensional vector analogue of this expression is
(2.6)
A possibly more familiar representation of the overlap <'1' I~) is in terms of

wave functions 'I'(r) and ~(r)
<'1'10 = fan,pace 'I'*(r)~(r) dr = I.n,pace <'I'lr)<rl~) dr. (2.7)
The sum in eQn (2.5) over the basis vectors {14»} is replaced in eqn (2.7)
by an integral over all-space because the spatial representation is in a
continuous variable r. The analogue of eqn (2.4) for the identity operator in
the r representation is thus
1= f all space
Ir)<rl dr. (2.8)
We have already come a long way. Suppose we have two sets of basis
states {14»} and {Ii)} spanning the same Hilbert space. We can represent a
state vector 1'1') in both Hilbert spaces and obtain the expansion coefficients.
Using eqn (2.4) for the identity it is easy to find the relationship between
the expansion coefficients
<4>1'1') = <4>111'1') = L <4>li)<il'l') (2.9)

i
which relates the expansion coefficients <4>1'1') to the set of <il'l').

Let us consider the representation of an operator in sorne basis seto Let
L be an arbitrary operator which acts on a state vector 1'1') to produce a
new state vector 10
10 = LI'I')· (2.10)
Let us see how this works if we use the basis set {li)} to represent the state
vectors and the operator L. The coefficient GI~) in the representation of
10 in this basis set is given by
GIO = GILI'I') (2.11)
where Ii) is one of the basis sta tes. We can now use eqn (2.4) to get something
much more useful
<jlO = <jILI'I') = GILI L li)<ill'l') = L GILli)<il'l'). (2.12)

all i all i
We see that the jth component of the state vector 10 is related to all the
components of the state vector 1'1') through the quantities GILli). The
ordinary three-dimensional vector equivalent of this equation is that the
vector ~ is related to the vector '1' by a matrix L
(2.13)
or in component form
3
~j =
i= 1
L Lj¡'f'¡. (2.14)
Comparing this equation with eqn (2.12) we see that the quantities GILli)
and Lj¡ serve the same role. For this reason GILli) is called a matrix element
of the operator L.
Returning to eqn (2.10) we can obtain a useful formula for the operator
L in terms ofits matrix elements <iILij) by applying eqn (2.4) for the identity
twice
10 = LI'f') (2.15)
or
10 = L li)<iILI L IJ)GII'f') = L L li)<iILIJ)<JI'f')
all i allj all i allj
(2.16)
where we have used different dummy variables i and J in our double

application of eqn (2.4). In both sums we mean that the sum is to be taken
over aIl basis sta tes and li) and IJ) are just dummy representatives. lt would
have been misleading to use the same dummy representatives. Comparing
these last two equations we see that the operator L can be expressed as
L = L L li)<iILij)<JI
all i a1l j
(2.17)
since eqns (2.15), (2.16) are true for any state vectors 10 and 1'1'). If this
equation is true for any operator L then we should recover eqn (2.4) for the
identity operator if we re place L by l. It is easily verified that this is indeed
the case by using the orthonormality of the basis set.
Equation (2.17) ror the operator L enables us to write down immediately
the matrix elements of L in another basis set {I<p)}. Suppose we want the
matrix element <O"ILIK) where lO") and IK) are members ofthe basis set {I<p)}.
Using eqn (2.17) we obtain
<O"ILIK) = L L <O"li)<ilLij)GIK). (2.18)
alliallj
This equation allows us to transform the matrix elements of an operator

expressed in one basis set into matrix elements expressed in another basis
set. The transformation involves two sets of overlap matrix elements, <0"1 i)
and <J IK). between the basis sets. By contrast, in eqn (2.9), we saw that the
components of a state vector were related by only one set of overlap matrix
elements.
A homonuclear diatomic molecule 25
We have talked about a state vector 1'1') as being a quantum state vector.
In fact, quantum mechanics has not entered the picture so far in this section.
The state vector could equally well apply to classical quantities that we wish
to expand in sorne set of basis states. For example suppose we have a linear
chain along the x-axis of N billiard balls (obeying classical physics!)
connected by springs. Let Xi denote the displacement of the ith ball in the
x-direction from its equilibrium position. We can think of each ball as a
basis state and the ith such ball we write as li). The set of all N balls
comprises the 'basis set' {Ii)}. Then Xi is the representation of the ith basis
state in the position (x) representation: Xi = <xli). The 'state vector' 1'1')
denotes a state of all N balls
N
1'1') = L li)<il'l')· (2.19)
i=l
We project out the ith ball from the state vector by forming <il'l'). Then
<xl'l') is a scalar describing a state of the system in which the position of
the first ball is Xl (i.e. Xl = <xii», the position of the second ball is X2 (i.e.
X2 = <xI2» and so on with the position of the Nth ball given by X N
N
<xl'l') = L <xli)<il'l').
;=1
(2.20)
The chain of N billiard balls thus spans an N-dimensional Hilbert space. As

N tends to infinity so the Hilbert space becomes infinite dimensional. But
as you can see that is no big deal-all that happens is that more terms are
included in the sum in eqn (2.20).
We shall find the formulae that we have derived in this section extremely
useful. The link between the reciprocal and real space pictures of bonding
is achieved by simple transformations of basis states.
A homonuclear diatomic molecule: the hydrogen

molecule
Consider the H 2 molecule in its ground state. If we attempted to solve the
Schrodinger equation exactly we would soon come unstuck because there
are two electrons in this molecule, one from each hydrogen, and they interact
Coulombically. Instead, we shall follow a simple molecular orbital approach
and obtain a qualitative picture which nevertheless captures most of the
physics.
Let 1'1') denote a state vector of an electron in the molecule. What are
suitable basis states in which to expand 1'1')7 Imagine that we form the
molecule by bringing two isolated hydrogen atoms together. Let 11) denote
the electron state in the first atom and 12) denote the electron state in the
second atom. These states will be one of the states that we considered for the
isolated hydrogen atom in Chapter 1. Since we are interested in the ground
state of the molecule we shall assume that 11> and 12> are the ground state
1s, and let the energy of the electron in these states be Ef
and (2.21)
where H¡ and H2 are the Hamiltonians for the isolated atoms 1 and 2. We
now assume that the two basis states 11> and 12> form an adequate basis set
in which to expand the ground state 1'1'> of the hydrogen molecule
1'1'> = c¡l1> + c212>. (2.22)
Furthermore, we assume that the basis states 11> and 12> are orthonormal
so that <112> = <211> = O and <111> = <212> = 1. This assumption cer-
tainly breaks down when the two atoms get dose together and the 1s orbitals
start to overlap. But the main qualitative feaures of the molecular binding
survive this assumption. We explore the more plausible case of what happens
when the overlap <112> is not assumed to be zero in Problem 9. It follows
that C¡ = <11'1'> and c2 = <21'1'>. Our task is to find c¡ and C2 and the
energy of the state 1'1'> for the molecule.
The Schrodinger equation for the molecular state is
HI'I'> = EI'I'> (2.23)
or, substituting eqn (2.22) for 1'1'> we have
H(c¡l1> + c212» = E(c¡11> + c 212». (2.24)
To solve this equation we project it onto the basis states 11) and 12>
<1IH(c¡11> + c212» = <1IE(c¡11> + C212»,}
(2.25)
<2IH(c¡11> + c212» = <2IE(c¡11> + c212».
These two equations can be expressed as
Eoc¡ + H 12 c 2 = EC¡,}
(2.26)
H2¡c¡ + EOc2 = EC2'
We have used the orthonormality of the basis set to simplify the right hand
sides and we have expressed the <¡IHU> matrix elements as HIj. We have
also expressed Hll and HZ2 as Eo' Eo is not the same as E f because the
presence of the potential from the nearby second atom contributes an
additional term to Eo. Not surprisingly the 2 x 2 matrix with elements Hij
is called the Hamiltonian matrix. The integrals H¡¡ and H22 are called on-site
Hamiltonian matrix elements because they involve states only on the same
atom.t Equations (2.26) are simply the SchrOdinger equation, eqn (2.23),
t Although the potentials oC both atoms are contained in H.

expressed in matrix form rather than as a differential equation. They are
called the secular equations. For nontrivial solutions of C I and C2 we require
that the secular determinant be zero
Eo - E H12 l' = O. (2.27)

H 21 Eo - E
Expanding the determinant yieIds the following quadratic equation
E2 - 2EoE + E~ - H l2 H21 = O. (2.28)

Since the Hamiltonian is hermitian it must be represented by a hermitian
matrix. That means H 12 = H!l' But we have assumed real atomic orbitals
(ls orbitals) for the basis states (i.e. they are not complex) and therefore
the Hamiltonian matrix elements are real. Thus H 12 = H 21 = P sayo The
solutions to this equation are
Eb = Eo + P
and (2.29)
E. = Eo - P
The normalized state vector corresponding to the eigenvalue Eb is
(2.30)
while the normalized state vector corresponding to the eigenvalue E. is
(2.31)
Well, what does all this mean? We have found two states, one with energy
Eo + Pand the other with energy Eo - p. What is the sign of p? To answer
this we have to take a cIoser look at the matrix element H 12 = OIHI2). Let
us assume we can write the Hamiltonian operator for an electron in the
molecule as followst
(2.32)
where VI (r) is the electrostatic potential of the nucIeus in atom 1 and "í(r)
is the electrostatic potential of the nucIeus in atom 2. Therefore we can
t This Hamiltonian ignores the interaction between the two electrons in the molecule. It is
thereCore a description oC H; rather than H 2 • but that does not affect the argument too mucho
express H¡2 as follows
h2V 2
H 12 = 01 - 8n 2 m + V¡(r) + /I2(r)12)
h2 V 2
= 01 - 8n 2 m + V¡(r)12) + 01 /I2(r)12). (2.33)
Now <11 - WV 2 j8n2 m) + V¡(r)12) is zero because
and therefore 01- (h 2 V 2 j8n 2 m) + V¡(r)12) = Er(112) = O. Thus H 12 =

0I/l2(r)12). The potential /I2(r) is attractive to an electron (and hence
negative). Furtherrnore 11) and 12) are assumed to be Is states which are
never negative. Therefore H¡2, or p as we have called it, is negative. You
may feel somewhat uncomfortable with the idea we have assumed (112) = O
and yet we maintain that <II/12(r)12) is not zero. This inconsistency can be
eliminated, as in Problem 9, by removing the assumption that the overlap,
S = (112), is zero, but the results (despite the much greater effort!) are
qualitatively the same.
The molecular state l'I'b) has energy Eo + p, which is lower (i.e. more
negative) than Eo because p is negative. It corresponds to a symmetrical
combination of the basis sta tes, as shown in Fig. 2.1. The two electrons in
the H 2 molecule occupy this molecular state, with opposite spins. The
electronic contribution to the total energy is thus lower in the molecule than
for two isolated atoms by 21PI, and this is the source of the cohesion in the
molecule. Tbis is why we call1'l'b) the bonding state.
The molecular state 1'1'.) has energy Eo - P, which is greater than Eo. It
corresponds to an antisymmetrical combination of the basis states, as shown
in Fig. 2.2. Since an electron occupying this state would have a higher energy
than Eo its occupation would not lead to bonding. For this reason 1'1'.) is
called an antibonding state.
Fig.2.1 The wave function 'l'b(r) and the charge density Pb(r) of the bonding
state in a homonuclear diatomic molecule.
'f',(r) p,(r)
-----~--~~--~--~~r
Fig. 2.2 The wave function 'I'.(r) and the charge density p.(r) of the anti-
bonding state in a homonuclear diatomic molecule.
The reason why l'I'b) leads to bonding and 1'1'. )to antibonding is apparent
from Figs 2.1 and 2.2. Since 1'I'(rW is the probability of finding an electron
at r the charge density p(r) is proportional to 1'I'(r)1 2. The bonding state leads
to charge being heaped up in the region between the atoms, while in the
antibonding state the charge density has a cusp at the mid-point ofthe bond.
Heaping up charge between the atoms leads to bonding because it is
electrostatically attracted to both nuclei and, therefore, pulls the nuclei
together. We can see this directly from the charge density associated with
two electrons occupying the bonding wave function 'l'b(r) = <rl'l'b)
(2.34)
where
(2.35)
If the charge density in the molecule were simply a superposition of the

atomic charge densities Pl(r) and P2(r) then we would not have the bond
charge density Pbond (r), which we see in eqn (2.35) is a consequence of
interference between the atomic wave functions 'l'l(r) and 'l'2(r). This
interference is a purely quantum mechanical effect since there is no analogue
in classical physics. This result is sufficiently monumental that we state it
again: the chemical bond in the hydrogen molecule is a direct consequence
of quantum mechanical interference between the wave functions of the
constituent hydrogen atoms. The same is true in other covalent and
metallic bonds: bonding arises from interference of quantum mechanical
waves!
We shall now go through the analysis of the H2 molecule all over again,
but starting from the time-dependent Schrodinger equation. This is very
instructive because it gives us another insight into what those important
off-diagonal Hamiltonian matrix elements H 21 and H 12 mean physically. The
time-dependent Schrodinger equation for the state 1'1') is as follows
i ~ dl'l') = HI'I'). (2.36)

211: dt
Substituting eqn (2.22) for 1'1') we obtain the following coupled first order
differential equations
and
These equations have to be solved with sorne initial conditions for c l (t) and
cit). As before H 11 = H22 = Eo and H 12 = H2l = p. First order differential
equations with constant coefficients are solved by exponential functions, so
we try
and (2.38)
Inserting these in eqn (2.37) we again obtain the secular equations and find
that there are two solutions for (}) corresponding to the bonding and
antibonding states
(})b = 2n Eo + p and (2.39)
h
The solutions for Cl(t) and cit) are then given by
Cl(t) = a exp( -2ni Eo : P t) + b exp( -2m Eo ; P t)

and (2.40)
C2(t) = a exp ( - 2m. Eo h+ Pt ) - b exp ( . Eo h-

-2m Pt )
where a and b are arbitrary constants to be determined by the conditions

at t = O.
Suppose that at t = O we know the molecule is in the state 11) so that
Cl(O) = 1 and c2(O) = O. Then a = b = 1/2 and
Cl(t) = e-2dEot/h cos(2npt/h) and C2(t) = e-2KiEot/h sin(2npt/h).
(2.41)
i
The two amplitudes oscillate harmonically with time. The probability that
the molecule is in the state 12) is given by IC2(tW
(2.42)
The probability starts at zero, as it should, and rises to one in time h/4IPI,
and retums to zero in time h/2IPI, as shown in Fig. 2.3. At the same time
the probability that the molecule is in the state 11) is exactly out of phase,
as also secn in Fig. 2.3. The number of times the molecule passes from state
11) to state 12) per unit time is thus 2IPI/h. An electron in this molecular state
is thus vibrating back and forth between the two atoms with period h/2IPI.
~I " ,'
, 2
'" : 1c,(t)1
,
J ,,,
2
1c 2(t) I
O t
h h 3h h
41~1 21~1 41~1 I~I
Fig. 2.3 The probabilities IC1 (t)1 2 (broken line) and IC2(t)12 (sol id line) of an
electron occupying the atomic states 11) and 12) respectively as a function of
time in a diatomic molecule.
Since the ionization energy of a hydrogen atom in its ground state is

13.6 eV how is it possible for an electron to vibrate back and forth so freely
between the hydrogen atoms making up the molecule? The answer is that
the electron tunnels through the energy barrier, rather than going over it.
The quantum mechanics aboye has shown that provided Pis not zero there
is a1ways a finite probability of tunnelling taking place. Indeed the prob-
ability per unit time that the electron tunne1s or 'hops' from one atom to
the other is 2IPI/h. For this reason the off-diagonal elements of the
Hamiltonian matrix, such as p, are called 'hopping' parameters or hopping
integrals. This interpretation of P is quite consistent with our earlier
conc1usion that P represents the integral <11 J-í(r)12). The fact that this
integral involves both states 11) and 12) indicates that it involves the transfer
of an electron between the states. If the separation of the hydrogen atoms
increases, as it must during dissociation, then P decreases rapidly, and the
tunnelling probability with it. At the same time Eo approaches Er, the energy
of the free atomic state. Since the energy of the bonding state is Eo - IPI this
rapidly approaches Er. To obtain hopping integrals we can either calculate
them directly by evaluating integrals such as <11 J-í(r)12) or they can be
determined empirically by fitting experimentally observed molecular orbital
energies.
The picture of chemical bonding between homonuc1ear atoms that we
have gained consists of electrons becoming delocalized by tunnelling through
the energy barrier separating neighbouring atoms. When atoms are brought
together the electronic charge density is not simply a superposition of free
atomic charge densities beca1Jse electronic charge is also piled up in the
bonds. The build up of bond charge is a direct consequence of the quantum
mechanical tunnelling setting up interference between free atomic sta tes. The
greater the tunnelling rate the lower the energy of the bonding state relative
to the free atomic state.
A heteronuclear diatomic molecule

In the previous section we considered a homonuc1ear diatomic molecule and
showed how bonding and antibonding states arose. In this section we repeat
the analysis for a heteronuc1ear AB diatomic molecule. We again set the
hopping integral H 12 = H 21 equal to a parameter p, which we again take to
be negative. The on-site Hamiltonian matrix elements on the A and B atoms
are called EA and Ea and we assume EA > Ea. Writing the atomic states lA)
and lB) the molecular state is assumed to be sorne linear combination of
these sta tes
(2.43)
The time independent Schrodinger equation leads to the following secular
equations (compare eqn (2.26»
(EA - E)CA + PCa = O }
(2.44)
PCA + (Ea - E)ca = O.
The secular determinant againleads to a quadratic equation in E and the
bonding and antibonding eigenvalues are
Eb = + P2)1/ 2 }
8 _ (~2
(2.45)
E. = 8 + (~2 + P2)1/2
where 8 is the average on-site energy (EA + Ea)/2 and ~ equals (EA - E a )/2,
which is positive. Note that as ~ tends to zero we recover the results
(eqn (2.29» for the homonuc1ear diatomic molecule. We see that the effect
of the difference in the on-site energies is to increase the splitting between the
bonding and antibonding states. If the hopping integral p is zero then we
recover Ea and EA for the bonding and antibonding states respectively.
In the homonuc1ear case the charge densities associated with each atom
in both the bonding and antibonding states are equal. The charge density
associated with atom 1 is proportional to Ic¡j2 and it is easily seen from eqns
(2.30) and (2.31) that Ic¡j2 = IC212 = 1/2. The fact that EA> Ea in the
heteronuc1ear molecule results in it being energetically favourable in the
bonding state for sorne charge density to be transferred from the A atom to
the B atom where it can enjoy the lower on-site energy. In the antibonding
state the charge transfer is in the reverse direction. Inserting the eigen-
values, eqn (2.45), into the secular equations for the heteronuc1ear diatomic
molecule, eqn (2.44) we find that
d. 1 for the bonding state
(2.46)
for the antibonding state
A heteronuclear diatomic molecule 33
Antibonding state
Bonding state
~O~------------------~~x
Fig. 2.4 The ratio c¡/c¡, of the occupation of the atomic state on the A atom
to that on the B atom in an AB diatomic molecule, for the bonding and
antibonding molecular states as a function of x = ll/P = (EA - Ea )/2p.
where
&
x=-. (2.47)
p
These results are plotted in Fig. 2.4. It is seen that as x -+ O then cVc¡ -+ 1
for both the bonding and antibonding states as expected. For x» 1 we find
that ci/c¡ -+ Ofor the bonding state and cilc¡ » 1 for the antibonding state,
meaning that in the bonding state a11 the charge is transferred to the
B atom whereas in the antibonding state a11 the charge is transferred to the
A atom.
The upshot of a11 this is that when & is not zero the bond becomes
partia11y ionic because sorne charge transfer takes place. We can define the
polarity of a bond, meaning the extent to which it is polar or ionic, by the
quantity
(2.48)
and the covalency of a bond, meaning the extent to which it is covalent, by

1
IX
e
= (1
-----.:--:-=
+ X2)1/2 (2.49)
where x is given by eqn (2.47). It fo11ows that

(2.50)
The completely ionic limit is obtained when X tends to infinity and the
completely covalent limit when X = O. These simple ideas carry over to the
solid state as we11.
Electronegativity
We saw in the previous section that one consequence of the difference in
on-site energies of the heteronuclear diatomic molecule is that in the bonding
state charge is transferred from the atom with the higher on-site energy to
the atom with the lower on-site energy. Pauling defined electronegativity as
the power of an atom to attract electrons to itself from neighbouring atoms
in its environment. It is therefore a property of the atom and the environment
it finds itself in. An electronegativity scale is a set of numbers, one for each
element, such that if the electronegativity for atom A is greater than the
electronegativity for atom B then electronic charge flows from atom B to
atom A. There is nothing unique about such a scale beca use if we add a
constant to aH the electronegativities, or if we multiply aH the electro-
negativities by a constant amount, then we obtain the same ranking of the
elements in the scale. Consequently several scales are in use and may be
brought roughly into coincidence by shifting the zero and changing the units.
The electronegativity of an element is not a measurable quantity, but it is
useful in indicating trends in bonding beha viour.
According to the aboye definition minus the on-site Hamiltonian matrix
element should be a measure of electronegativity beca use the more negative
H¡¡ draws electrons onto itself from neighbouring atoms with less negative
H¡¡s. But there are problems with using on-site Hamiltonian matrix elements
as an electronegativity scale. Firstly, we have seen that the on-site Hamiltonian
matrix element for an atom depends on the environment into which the atom
is put-as we require for an electronegativity-but the dependence on the
local environment is not easily determined. Secondly, there may be several
on-site energies for a given atom corresponding to different valence orbitals,
for example 3s and 3p valence orbitals for Si. How do we average over them?
Pauling's scale was obtained empiricaHy and is shown in Table 2.1. It
shows the expected increase in electronegativity from left to right along the
periods (compare Fig. 1.12) and the decrease down columns of the periodic
table. It turns out that the difference in Pauling electronegativities is
roughly equal to the dipole moment of the bond expressed in debye units
(1 debye = 3.334 x 10- 30 Cm = 0.208 e- Á).
Mulliken introduced an alternative scale of 'absolute' electronegativities
defined in terms of measurable quantities. His electronegativity is defined by
(2.51)
where IPA is the ionization potential of atom A (the energy required to
remove an outer electron) and EA A is the electron affinity of atom A (the
negative of the energy of adding an electron). The reasoning behind this
definition is straightforward enough. An A-B bond would be covalent if the
energy of an A+B - bond is about the same as the energy of an A- B + bond.
Now the energy required to make an A+B- bond, for sufficiently weH
Bond energy and bond order 35
Table 2.1 The Pauling electronegativity values tor selected elements
H
2.2
Li Be B C N O F
0.98 1.57 2.04 2.55 3.04 3.41 3.98
Na Mg Al Si P S CI
0.93 1.31 1.61 1.90 2.19 2.58 3.16
K Ca Ge As Se Br
0.82 1.00 2.01 2.18 2.55 2.96
Rb I
0.82 2.66
Cs
0.79
separated A and B atoms, would be IPA - EAo whiJe the energy to make an
A-B+ bond would be IPo - EA A • The criterion for a nonpolar bond is thus
IPA - EAo = IPo - EAA , or
(2.52)
On the other hand if IPA + EA A > n~ + EAo then the energy of the A+ B -
bond is greater than the energy of the A- B + bond. In that case charge
transfer from the B atom to the A atom is favourable. This is entirely
consistent with saying that IPA + EAA (or !<IPA + EAA» is a suitable
electronegativity scale.
Empirica11y, it is found that the Pauling electronegativity scale can be
obtained quite accurately from the Mulliken scale by the relation
Pauling electronegativity scale = 0.336(XA - 0.615). (2.53)
Thus the scales are essentia11y equivalent. The Mulliken scale is defined in
terms of atomic properties and is independent of the environment into which
the atom is placed. Although this is not in accord with Pauling's original
conception of what electronegativity means it seems inevitable if we are to
come up with one number for each element in a scale that is supposed to
apply in a11 situations.
Bond energy and bond order

We now introduce two concepts that are at the heart ofthe real space picture
of bonding in the solid state, namely bond order and bond energy. Let us
return to the time-independent SchrOdinger equation for the heteronuc1ear
diatomic molecule. We can write the secular equations as a matrix equation
in the following way
(HAA HAo)(eA) = E(eA) (2.54)

HOA Hoo eo eo
where we have written the on-site energies as H and Hoo and the hopping
AAalready
integrals as HAO and HOA (HAO = H OA ). We have solved this equation
and obtained the two eigenvalues Eb and Ea for the bonding and antibonding
states. Let the eigenvectors be
and
for the bonding and antibonding states respectively. Thus, for the bonding
state we have
(2.55)
(::: :::)ei) = Ebei)·
Premultiplying both sides of this equation by the row vector (e!, cr) we
obtain
(e!, en(::: :::)(:i) = (e!,e~ + ere~)Eb· (2.56)

The right hand side is simply Eb because (e!,e~ + ere~) = 1 owing to the
normalization eondition for the bonding state. Expanding the left hand side
we obtain a physieally transparent breakdown of the eigenvalue
(2.57)
and similarly the eigenvalue of the antibonding state ean be expressed as
(2.58)
Let us now examine what eqn (2.57) means. Since e~e!, is just the probability
of finding the electron on the A atom in the bonding state, then the first
term issimply the energy contribution eoming from the time spent by the
electron on the A atom in the bonding state. Similarly, theseeond term is
the energy eoming from the time spent on the B atom in the bonding state.
The interesting term is the last term in curly braekets. This is the eontribution
to the eigenvalue of the bonding state from the time spent in the bond region:
it is the bond energy eontribution.
The bond energy contribution to the bonding eigenvalue can be expressed
as (e~er + e!,e~)HAO = 2e~e~HAO' where we have used HAO HOA= and the
faet that the eigenveetors are real. The term e~e~ is simply the inteference
term that we diseussed in eqn (2.35) and it is called the partial bond order
ofthe AB bond. The total bond order, or just 'bond order', is a sum ofpartial
bond orders taken over all occupied states. The bond energy is twice the
Reference 37
, J(, , ,
i
'g ,,
,,
, ,,
,,
.ii 0.5 , , .1 7C" ,
,,
,, ,,
,, ,
if-_........._--L---'--'---'1I-_ Number of electroos
234
Fig. 2.5 The bond order in the H2 molecule as a function of the number of
electrons in the molecu le.
bond order times the corresponding hopping integral. Thus, if there is just
one electron in the molecule it half fills the bonding state and the AB bond
order is c~c:. If there are two electrons in the molecule they can both
occupy the bonding state and the bond order is 2c~c:. If there are three
electrons in the molecule then the bonding state is fully occupied and the
antibonding state is half filled. The bond order is then 2c~c: + c~c~. If there
are four electrons (the maximum permitted by the exclusion principie)
then the antibonding state is al50 fully occupied and the bond order is
2(c~c: + c~c~) which is zero owing to the orthonormality of the eigenvectors
of a hermitian matrix.* The variation of the bond order with the number of
electrons in an H 2 molecule is shown in Fig. 2.5.
From the invariance of the trace of a matrix under a 5imilarity trans-
formation we know that the sum of the bonding and antibonding eigenvalues
equals the traceof the original Hamiltonian, i.e. HAA + Hoo. Adding eqns
(2.57) and (2.58) we see that the sum of the terms in the curly brackets must
be zero. That is, the bond energy contributions to the bonding and
antibonding states cancel. Thus, if both the bonding and antibonding states
are fully occupied (by two electrons each) then the total electronic energy is
just 2(HAA + Hoo) and there is no bond energy. This is exactly what we
would expect beca use the bond order is zero when both the bonding and
antibonding states are fully occupied.
Reference
Feynman, R. P., Leighton, R. B., and Sands, M. (1966). The Feynman lectures on
physics, Vol. III. Addison-Wesley, London.
• More precisely, iC we wrile Ihe eigeoveclors Ci) aod C~) as Ihe column veclors oC a 2 x 2
malrix e Iheo bolh Ihe columos and rows oC e are orthooormal.
3
From the finite to the infinite
Chain molecules and k-space

In this section we shall consider a series of hypothetical molecules consisting
of chains of between two and an infinite number of hydrogen atoms. One
aim is to show that the quantum mechanical description of bonding in an
infinite linear chain is essentially no different from what we have already
discussed in Chapter 2 for a diatomic molecule. In this way we hope to
convince you that solids may be thought of as infinite molecules, but, of
course, in three dimensions. Another aim is to set up the machinery that we
need to discuss an infinite periodic system, such as k-space, Bloch functions,
Brillouin zones, the Fermi energy, and so on. Once we have done that then
moving onto a three-dimensional perfect crystal is more complicated, but
no new ideas are needed.
Our H. chain molecules do not exist in reality because, for example, it is
energetically favourable for the H3 molecule to dissociate into H 2 and H. In
other words, the lengths of the two bonds in the H3 molecule spontaneously
distort, so that one increases and the other decreases in length. Similarly, if
we managed to produce an infinite chain of hydrogen atoms with the bond
lengths all the same the molecule would spontaneously distort so that pairs
of hydrogen atoms would come c10ser together and form H 2 molecules. The
periodicity of the chain would thus be doubled. This is called a Peierls
distortion. Although H. chains do not exist there are sorne very important
chain molecules that are built up by adding units, namely the alkanes and
alkenes. Polyethylene and polyacetylene are long chain molecules that have
a carbon chain as the backbone. Starting from ethane, C 2 H 6 , we can add
CH 2 groups and make longer and longer chains: C 3H 8 , C 4 H lO , C S H 12 •
C 6 H 14, ••. ,C.H 2.+ 2 . These are the alkanes and as n tends to infinity we
get polyethylene. Similarly starting from acetylene, C 2 H 4 , we can add (CHh
groups and make longer and longer chains: C4 H 6 , C 6 H 8 , •.• , C.H.+ 2 • These
are the alkenes and as n tends to infinity we get polyacetylene. Although the
electronic structure of these molecules is obviously more complicated than
our H. chains there is neverthe1ess a good deal of similarity between how
one sets up the calculations.
So, let us now consider a linear chain of N hydrogen atoms as shown in
Chain molecules and k-s pace 39
-0----0--0
2 3 4 N-2 N-I N
Fig.3.1 A linear chain of N hydrogen atoms.
Fig. 3.1. Each hydrogen atom is associated with an s state and we assume
that the set of these states on the N hydrogen atoms forms a complete basis
set in which to expand a molecular state 1'1') for the chain
N
1'1') = LeN)· (3.1)
j-l
Here, the s state on atomi has been denoted by the ket li). Our task is to
find the coefficients eJ and the energy of the molecular state 1'1'). To do this
we have to solve the Schrodinger equation for the molecular state
HI'I') = EI'I'). (3.2)
Inserting eqn (3.1) for 1'1') into eqn (3.2) we obtain

N N
L ejHU) = E LeN)· (3.3)
J- 1 j- 1
This is equivalent to eqn (2.24) for the H 2 molecule if we put N = 2. Our

next step is exactly the same as in eqn (2.25), namely that we multiply from
the left by abra, say <pi, where p is one of the sites of the N atoms. We
then obtain
N N
L ej<pIHli) = E L ej<pli). (3.4)
j= 1 j= 1
This is the secular equation for the chain of N atoms (compare eqn
(2.26) for the H 2 molecule). To solve it we have to assign the Hamiltonian
elements <pIHli) and the overlap matrix elements <pli). The simplest
assumption for the overlap matrix e1ements is that the basis set is ortho-
normal, that is <plp) = 1 and <pli) = O for Ip) and U) at different
sites. The simplest assumption for the Hamiltonian matrix elements is that
they are all zero except for 'on-site' matrix e1ements <pIHlp), which
we set equal to IX, and matrix elements between states on neighbouring
atoms (the 'hopping' integrals), which we set equal to p. IX and pare
simply parameters in this treatment-to determine them we could either
try to evaluate the integrals that they correspond to or we could try to fit
them to experimental data on the spectrum of energy levels of the molecular
states.
Let us now see how eqn (3.4) works out with these assumptions for the
40 From tha finita to tha infinita
matrix elements. We have
IXC 1+ PC2 = EC 1 (for p = 1)
pC 1 + IXC2 + PC3 = EC2 (for p = 2)
PC2 + IXC3 + PC4 = EC3 (for p = 3)
(for p = j) (3.5)
PCN-2 + IXCN - 1 + pC N = ECN-l (for p = N - 1)

PCN-l + IXCN = ECN (for p = N).
There are N coupled linear equations here, and we have to solve them! But,
it is not as bad as it looks. First of aU we note that the equations have aU
the same form except for p = land p = N, the end atoms. AU the equations
can be simplified by first dividing by p. For example, for p = j we get
C)-l - xc) + C)+l = O (for p =j) (3.6)
where
x = (E - IX)/P. (3.7)
Let us try the solution c) = ei)8 where i = (_1)11 2 and Ois to be determined
(do not confuse i in the exponential with the label of atom i). At this stage O
may be real, imaginary, or complexo Inserting this guess into eqn (3.6) we get
e 1U - 1)8 _ x ei)8 + elU + 1)8 = O
or
(3.8)
If we replace O by - O we get the same solution for X. Therefore we could
have c) = A e lJ8 + B e- 1J8, where A and B are arbitrary constants and still
eqn (3.8) would be satisfied. How do we determine A and B? WeU, do not
forget the end atoms. For p = 1 we have C2 = XCI' Therefore,
(3.9)
and this equation is satisfied by setting A = - B. Thus the equation C 2 = Xc!
pins down A and B to the extent that cJ = A(e lJ8 - e- 1J8 ) where A is still an
arbitrary constant and O satisfies eqn (3.8). We may rewrite c) as D sinjO
where D = 2iA, i.e. yet another arbitrary constant. The equation for the other
end atom, CN -1 = XCN, is used now to pin down the allowed values of O
D sin(N - 1)0 = xD sin(NO)
= 2D cos Osin(N O)
or
sin(N - 1)0 = 2 cos Osin(NO)
= sin(NO) cos O - cos(NO) sin O (3.10)
where the constant D is cancelled and remains arbitrary (D is determined
below by the condition that 1'fI) is normalized). Equation (3.10) is easily
simplified as follows
or
tan(N8) + tan 8 = °
sin N8 cos 8 + sin 8 cos N8 = °
cos 8cos N8
which implies that
sin(N + 1)8 = O
and therefore
8= mn where m is an integer. (3.11)
N+ 1
Thus 8 is real There are N distinct solutions corresponding to m = 1,2, ... , N.
Other values of m simply reproduce one of these N solutions. We have now
sol ved the problem we set ourselves. The allowed energies are given by eqns
(3.7), (3.8), and (3.11)
E = IX + 2PCOS( mn ) (3.12)
N +1
where
m = 1,2, ... , N.
Let the mth molecular state be denoted by 1'fI(m) and let the expansion
·
coeffi Clents be ej(m) where 1. = 12
, , ... , NTh
. en ej(m)·IS glven
• by
ej
(m) _
-
D(") .
Sin
( mjn ) (3.13)
N+ 1
where D(m) is still an arbitrary constant that is determined by the normal-
ization condition <'fI(m) 1'fI(m) = 1
D(m) = ( 2 )1/2. (3.14)

N +1
It may seem that this derivation has been somewhat tortuous. In fact,
there is an easier way to the solution, as follows. In addition to the N real
atoms of the chain let us add one imaginary atom before atom 1, called
atom O, and one after atom N, called atom N + 1. We now expand the
molecular state 1'fI) in the basis of N real atomic states and the two imaginary
atomic states 10) and IN + 1). We keep our feet on te"a firma by demanding
that the expansion coefficients eo and eN+ 1 are zero! This simply ensures
that the molecular state is confined to the real atoms 1 to N. The reason for
doing this is that the equations for p = 1 and p = N in eqn (3.5) now have
the same form, exemplified in eqn (3.6), as all the other values of p in the
chain. As before, the general solution to eqn (3.6) is eJ = A el}6 + B e- iJ6, but
42 From the finite to the infinite
E=a - - 41~1
.....
N=l 2 3 4 5 N=oo
Fig. 3.2 The eigenvalue spectra for linear chains of length N = 1,2,3,4.5, ... ,00
hydrogen atoms. In the limit of N -+ 00 a band of eigenvalues of width 41PI centred
at E = ~ is formed.
now j is allowed to vary between O and N + 1. Demanding that Co = O then

gives A = -B and demanding that CN+l = O gives sin(N + 1) 9 = O and
hence eqn (3.11). Well, it is often possible to find an easier way after a difficult
derivation has been completed. We shall obtain a very similar result in
Chapter 7 for free electrons in a box, and by the same procedure of
demanding that the eigenstate is zero outside the box.
Let us check that this general solution for a chain of N hydrogen atoms
reproduces our earlier solution for the H 2 molecule, eqns (2.29)-(2.31).
Setting N = 2 in eqn (3.12) we get two solutions corresponding to m = 1
and m = 2. For m = 1 we have E = IX + Pwhile for m = 2 we have E = IX - p.
These are the same as eqn (2.29) because there we set IX equal to Eo. The
parameter IX simply sets the zero of energy. It is also readily confirmed that
eqns (3.13), (3.14) reproduce the two solutions for the H 2 molecule in eqns
(2.30), (2.31).
In Fig. 3.2 we show the eigenvalue spectrum, eqn (3.12), for chains of
1, 2, 3, 4, 5 and an infinite number of hydrogen atoms. As the number of
atoms in the chain increases the width of the spectrum tends to 4jPj. In the
infinite limit the energy gap between successive eigenvalues shrinks to zero
and the variable 9 of eqn (3.11) is a continuous variable between O and 1(.
The eigenvalue spectrum is ca11ed an 'energy band' and the bandwidth is
directly proportional to the hopping integral p. But although 9 is a
continuous variable in the infinite limit it is still the case that the eigenstates
of eqn (3.13) are labelled by 9 in that they are distinguished by having
different values of 9 lying between O and 1( in eqn (3.11). For this reason 9
is called a quantum number of the chain, in the same way as the angular
momentum is a quantum number of an atomic state.
Let us take a look now at the eigenstates of the chain, and in particular
let us start with the chain of length 5. It is helpful to think of those imaginary
atoms at positions O and 6, just before and after the chain. The fact that Co
and C6 must be zero means that there are always nodes in the molecular
sta tes at those points. The eigenstates are shown in Fig. 3.3. It is seen
(Fig. 3.3(a», that the lowest energy eigenstate, IX + (3)tp (remember that P
is negative), is the most bonding state with no nodes between atoms 1 and 5.
(a) a+~..J3=E
(b) a+~=E
(e) o a=E
,,
,,
"
(d) O a-Il=E
--
,,
. ,,
,
(e) O a-~..J3=E
Fig. 3.3 Molecular orbitals for the five eigenstates of the linear chain of five
hydrogen atoms (shown by small circles) in order of ascending energy. The solid
line shows the molecular orbital and the broken line shows the envelope
sin(mjn/6), where m = 1,2, ... , 5 in (a), (b), ... , (e), and j (which refers to the
atom number) varies from 1 to 5 in each figure. Notice that as the energy increases
the number of nodes in the molecular orbital increases, until, in the highest energy
state (e), there is a node between everysuccessive pair of atoms. AII the molecular
orbitals have nodes at atoms zero and six (not shown).
The next eigenstate (Fig. 3.3(b), has one node in the middle of the chain
and its energy is IX + {J. The next eigenstate (Fig. 3.3(c)), has two nodes, at
atoms 2 and 4, and its energy is IX. The fourth eigenstate (Fig. 3.3(d), has a
node at atom O and between atoms 1 and 2 and between atoms 4 and 5,
with eigenvalue IX - {J. Finally, the fifth eigenstate (Fig. 3.3(e)), has a node
between each pair of atoms in the chain and is the most antibonding state
with eigenvalue IX - (3)t {J. Thus the lower two states, with energy lower
N 1
N-2 ..
3
6 5
Fig. 3.4 A ring of N hydrogen atoms, formed by bonding atom 1 to atom N of

the linear chain of N atoms shown in Fig. 3.1.
than ex, may be described as bonding states, while the upper two, with energy
greater than ex, as antibonding states and the middle state, with energy ex, as
neither bonding nor antibonding. It is clear that the more nodes in the
molecular state the higher its energy.
The same qualitative picture oC bonding and antibonding states carnes
over to the eigenstates oC the infinite linear chain. But it is much easier to
discuss the states oC the infinite chain by using a trick that we shall oCten
exploit, namely periodic boundary conditions. This amounts to nothing more
than closing the chain on itselC to form a ring, as shown in Fig. 3.4.
The Hamiltonian is exactly the same as beCore except that atoms 1 and
N are now nearest neighbours and thus the Hamiltonian matrix element
between them is p, rather than zero. What physical argument can we give
to justify treating an infinite linear chain as an infinite ring oC atoms? The
reasoning is that when N is infinite the atoms in the bulk oC the linear chain
are unaware oC what happens to the end atoms. Thus whether the chain is
linear or a closed ring should not affect the eigenvalues in the limit oC an
infinite chain. The problem again reduces to eqn (3.6) but now the boundary
conditions are
and (3.15)
where atoms Oand N + 1 are the imaginary atoms that we introduced earlier.
The general solutions to eqn (3.6) become
2mn
e lN8 = 1 => 8 = - - where m = O, 1,2, ... ,(N - 1). (3.16)
N
The corresponding normalized eigenstates are given by
(1ft) _ 1 (. 2nj m)
e¡ - (N)1/2 exp 1 N (3.17)
- - a-2p
= = a-p
E=a--- ==a
- - - a+2p - - - a+2p - - - a+2p

N=! N=3 N=4 N=5
o
N=oo
Ú
Fíg. 3.5 The eigenvalue spectra of rings of 3, 4, 5, ... , 00 hydrogen atoms. As
in Fig. 3.2 the spectrum becomes a band of width 41PI, centred at E = IX, in the
limit N -+ oo.
and the eigenvalues by
E = IX + 2{J cos ( 27tm)

N . (3.18)
In Fig. 3.5 we show the eigenvalue spectrum for N = 3, 4, 5, ... an infinite

number of atoms in the ringo In the limit of an infinite ring, Oagain becomes
a continuous variable between O and 27t, and the eigenvalue spectrum
becomes
E = IX + 2{J cos O. (3.19)
Since this is a periodic function of O we may redefine the range of O to be

between -7t and +7t. We do this purely for convenience and because it
emphasizes the degeneracy between +0 and -O, as shown in Fig. 3.6. Now
let us discuss the eigenstates of the infinite ringo The expansion coefficients
E(9)
a-2p
, ,
~'~--~----~--~----~'~-'9
-Jt: :Jt
a+ 213
Fíg. 3.6 The eigenvalue spectrum (or band structure) for the infinite ring, plotted
as a function of 0, when there are only nearest neighbour hopping integrals.
become
1 "8
ej(O) = (N)1/2 e'l . (3.20)
At the bottom oC the band where O = O we see that all the expanslOn
coefficients are equal. ThereCore, there are no nodes in this molecular state
and that is why it is the most bonding state. On the other hand, at the top
oC the band where O = ± 11: we see that ej(O) alternates in sign Crom one atom
to the next around the ringo This is the most antibonding sta te. As beCore
we group together aIl the states with energies below (X and call them bonding
states and aIl those with energies aboye (X and call them antibonding states.
In the middle oC the band at O = ± 11:/2 the states are neither bonding nor
antibonding.
What is the physical significance oC the Cact that the eigenstates are in
general complex Cor the ring but real Cor the linear chain? The answer is
related to the Cact that the ring can carry an electric current but the linear
chain cannot (if it is not connected to anything else!). This is discussed Curther
in Problem 14.
We have already pointed out that in the infinite Iimit the eigenstates are
labeIled by O and Cor the ring we can choose the range oC O to be -11: to
+ lt. If we go outside this range then we simply reproduce one oC the states
inside the range. Ois continuous because there is an infinite number oC states
between - l t and + lt. Since O specifies the energy and the expansion oC the
eigenstate in the basis set it is the appropriate quantum number Cor the
infinite ringo In eqn (3.20) we see that the eigenstate labeIled by O is a wave
travelling around the ring with wavelength 2lt/O. The degeneracy oC the
eigenstates labelled by Oand - Osimply means that the energies oC the waves
travelling in opposite directions but with the same wavelength are the same.
This is an example oC time reversal symmetry because a wave travelling in
one direction reverses its direction if time were to reverse, and the energy oC
the state cannot depend on the direction oC time. Because the eigenstates Cor
the ring are waves the quantum number O is caIled the wave vector. (As we
are considering a one-dimensional problem the 'vector' is only a scalar. But
in two- and three-dimensionallattices it becomes a vector with two and three
components as we shaIl see in Chapter 4.)
Let us recapitulate what we have achieved so Car Cor the ring oC N atoms.
By considering a ring rather than a linear chain oC N atoms we eliminated
the 'surface atoms' at the ends oC the chain and aIl atoms in the ring are
equivalent. The molecular state 1'1') was then expanded (eqn (3.1)) as a linear
combination oC N atomic states U) around the chain. The Schrodinger
equation becomes a set oC N coupled linear equations. I think you will agree
that at this point the task oC determining the N expansion coefficients looks
pretty hopeless as N tends to infinity. But by guessing the solution ej = e 1j8
we reduce the problem in the Collowing remarkable way:
Chain molecules and k-s pace 47
1. Inc1uding the normalization constant, 1f(N)'/2, the molecular state
becomes
1 ~"o
1'1') = 1/2 L. el} Ij). (3.21)
(N) j~ 1
2. Putting this into the SchrOdinger equation, HI'I') = EI'I'), we get

N N
L: evoHIj) =E L: e1jOlj) (3.22)
j~l j~l
and multiplying from the left by the bra <pi we get

N N
L: evo<pIHij) =E L: eijo<plj). (3.23)
}~ 1 }~ 1
3. Since <plj) = O except whenj = p this becomes

N
L: e iJ8 <pIHij) = E e ip8 (3.24)
j~l
or, taking the e ip8 factor over to the left hand side
N
L: eiU-P)o<pIHlj) = E(8). (3.25)
}~1
4. The matrix elements <pIHIj) are aH zero except whenj = p or j = p + 1

or j = p - 1. When j = p we have <pIHlp) = cx. When j = p + 1 or
j = p - 1 we have <pIHij) = p. Thus there are only three terms in the
sum on the left hand side that are not zero
or cx + 2p cos 8 = E. (3.19)
5. The key point to notice here is that we get this answer for aH values of
p between 1 and N. This is because all atoms in the ring are equivalent.
In other words, it does not matter which atom p you choose in the
ring, the hopping integrals, p, to its two neighbours, and the on-site
Hamiltonian matrix element, ex, are always the same. If the hopping
integrals between successive atoms around the ring varied we would
deduce different values of the energy E for different values of p. That
would simply mean that eqn (3.21) could not be an eigenstate of the
system.
6. The restrictions on the aHowed values of 8 are determined by the condition
that the molecular state is periódic so that eN = Co and CN + 1 = C l' This
gives eqn (3.16).
AH this is fine except that we have not given any idea how we •guessed'
the solution, eqn (3.21). Since this is the key to our success we had better
try to understand where it carne from. Equation (3.21) is caHed a Bloch
fimction and it is based on the central theorem of the whole of band theory,
namely Bloch's theorem. Bloch's theorem is derived from the translational
symmetry of the ring-the fact that aH the atoms in the ring are equivalent
by a simple translation around the ringo We shaH not give a rigorous proof
here, but instead we give a plausibility argument that is based on our
observation that aH atoms in the ring are equivalent. Physically, this means
that any observable property of the ring must be invariant as we go around
the ring from one atom to the next. This does not mean that the molecular
state itself has to be invariant from one atom to the next because states are
not observable quantities in quantum mechanics. But certainly the prob-
ability of observing an electron at any point around the ring has to have
the translational symmetry of the ringo The probability of finding an electron
at x around the ring is 1'I'(xW. Therefore we are saying that translational
symmetry requires
1'I'(xW = 1'I'(x + mW (3.26)
where the distance between successive atoms around the chain has been set
to one, and the points x and x + m are separated by m atomic spacings. In
order to satisfy eqn (3.26) the wave functions 'I'(x) and 'I'(x + m) must differ
only by a phase factor and it would be reasonable to expect that the phase
factor depends only on the separation between the points, i.e. m, because of
the translational symmetry around the ring
'I'(x + m) = el~m'l'(x) (3.27)
where 4>.. is the phase factor. Now the wave function 'I'(x) = <xl'l') is
expressed in terms of the atomic orbitals <xlj) as foHows
N
'I'(x) = L cj<xlj).
j-1
(3.28)
Similarly, the wave function .'I'(x + m) can be expressed in terms of the

atomic orbitals <x + mlj)
N
'I'(x + m) = L cj<x + mU)·
j-1
(3.29)
Now in order to satisfy eqn (3.27) we first rewrite <x + mlj) as <xU - m)
and eqn (3.29) as
N
'I'(x + m) = L cj+ .. <xlj).
j-1
(3.30)
Therefore to satisfy eqn (3.27) we must have

(3.31)
This relation must be true for any j and m, and since the right hand side
depends only on m we must have
c.J = A e lj6 (3.32)
where A is an arbitrary constant and e is any real number. The arbitrary
constant A is determined by the normalization condition and is equal to
1/(N)I!2.
Bloch's theorem may now be stated in the one dimensional case
'I'(x + m) = e lm6 '1'(x). (3.33)
It follows immediately from this that 'I'(x) can always be represented as a
periodic function, say p(x), with the periodicity ofthe one-dimensionallattice,
times aplane wave e l6x
(3.34)
Equations (3.33) and (3.34) are equivalent statements of Bloch's theorem in
one dimensiono The same theorem applies in two and three dimensions, as
discussed in Chapter 4.
Bloch's theorem tells us immediately what the coefficients Cj for a ring of
N atoms must be. The severe restriction of translational symmetry allows
no freedom in the choice of these expansion coefficients. Looking at eqn
(3.21) mathematically, we see that the molecular state is simply a Fourier
series in which the coefficients are the atomic basis states. In eqn (3.25) we
see that the E( e) relation is ahro a F ourier series in which the coefficients
are Hamiltonian matrix elements. This is an extremely important point to
grasp. It helps you to appreciate that the various wiggles that you see in a
band structure diagram, i.e. a plot of E vs e, are related directly to hopping
integrals. To make this c1earer imagine that not only are there hopping
integrals to first neighbours but also to second and third neighbours. Let us
call these hopping integrals PI' P2' and P3 as shown in Fig. 3.7. Then the
E(e) relation is simply given by:
N
E(e) = L el (j-P)6<pIHIj)
j= I
= ri + PI(e I6 + e- 16 ) + P2(e 216 + e- 216 ) + P3(e 3i6 + e- 3i6 )

= ri + 2PI cos e + 2P2 cos 2e + 2P3 cos 3e. (3.35)
Fig. 3.7 Hopping integrals p" P2' and P3 between first, second, and third
neighbours respectively in an infinite ringo
E(9)
--------~-+-------.9
-lt lt
Fig. 3.8 The eigenvalue spectrum for the infinite ring, plotted as a function of
8, when there are hopping integrals between first, second, and third neighbours,
as shown in Fig. 3.7. Notice that there is now more structure in the spectrum
compared with Fig. 3.6 where there is only nearest neighbour hopping.
In Fig. 3.8 this one-dimensional E(8) re1ation is shown. Second-neighbour

interactions, P2' give rise to oscillations in E(8) with half the wavelength oC
the first neighbour interactions. Thus, finer and finer wiggles in an E(8)
relation are determined by longer and longer range hopping integrals.
We have already remarked that 8 is called the wave vector. In most books
it is called k and the space in which k is allowed to vary is called k-space.
From now on we shall call 8 the wave vector k. What is this k-space? Well,
consider the infinite ringo In that case all molecular states may be labelled
uniquely by a wave vector that lies between -n and n. Suppose we had said
that the spacing between the atoms oC the ring were 'a' and not 1. What
would the range oC wave vectors be then? In Bloch's theorem, eqn (3.33), the
phase Cactor would become eh""k and thereCore the restriction on k would be
Crom -n/a to + n/a. Thus 'k-space' Cor the infinite ring is simply a straight
lineo We shall see in Chapter 4 that Cor a two-dimensionallattice k-space is
aplane and Cor a three-dimensional lattice k-space is a continuous three-
dimensional space. In all three cases k-space is the reciprocal space oC the
1-, 2-, or 3-dimensional real space that the lattice is embedded in. For an
infinite number oC atoms in the ring k-space is continuous and any value oC
k along a line is permissible, although all states oC the ring lie within any
2n/a range in k-space and are simply repeated outside that range. For a finite
number N oC atoms in the ring the restriction that e;Nka = 1 restricts k to be
2nm/Na where m is any integer. Thus Cor a finite value oC N the reciprocal
space (k-space) becomes discrete, meaning that k may only be an integer
multiple oC2n/Na. Put another way, the number oCstates that we can squeeze
into unit distance along the k-axis is N a/2n or L/2n where L is the length
oC the ringo This is called the 'density of states' in k-space, d(k)
d(k) = Na/2n. (3.36)
E(k)
...- - - - + - - - - -.... First Brillouin zone

-_-n~ffl-----+------n~ffl-~k
Fig. 3.9 The first Brillouin zone in the one-dimensional infinite ring of atoms
with atomic spacing B.
The Brillouin zone is the region of k-space in which all eigenstates of the
ring may be labelled uniquely. For both finite and infinite rings all states of
the ring may be labelled uniquely within any 2n/a range along the k-axis.
Thus the Brillouin zone for a ring of atoms (finite or infinite) could be -n/a
to +n/a, or O to 2n/a, or 103n/a to 105n/a. It is customary to choose the
region of k-space that is c10sest to the origin, k = O, as the Brillouin zone.
For a ring this is -n/a to + n/a, as shown in Fig. 3.9.
The method we have described in this section is called the k-space, or
reciprocal space, method. The hallmark of the approach is the pivotal role
of Bloch's theorem which leads to the Fourier transformation from a basis
set of atomic states in 'real space' into eigenstates that are labelled in k-space.
The whole approach is predicated on the existence of translational symmetry.
If one atom of the chain were to move slightly c10ser to one of its
neighbours, and thereby change the hopping integrals, the entire edifice of
k-space would collapse. But such movements are happening all the time in
real solids: they are called thermal vibrations! We would not expect the
chemistry of the ring (for example, bond energies and charges on the atoms)
to change that much if the atomic movement were small. Our chemical
instincts tell us that there must be an alternative way of looking at the
problem that is not so precariously dependent on the existence of trans-
lational symmetry. And indeed there is. To see it we have to be more precise
about what we would not expect to change that much if one atom of an N
atom ring moved by, say, 1 per cent of a bond length. The movement changes
four entries in the N by N Hamiltonian matrix, and therefore all the
eigenstates of the ring would be affected, and not necessarily by only an
amount of order 1 per cent. So the individual eigenstates are sensitively
dependent on the atomic positions, and therefore they are not helpful in
understanding the chemistry of the ringo But we would be surprised if the
atomic movement of 1 per cent changed the bond order (introduced on p. 36)
far away from the atom. We would also be surprised if the atomic movement
changed the amount of charge on a remote atom. Somehow all the changes
in the expansion coefficients of the eigenstates at those remote atoms must
cancel out. The bond order and the charge on an atom are quantities that
we can associate with atoms in real space. Going into k-space does not help
us to understand individual bonds. We say that they are real space concepts,
as opposed to k-space concepts such as the band structure E(k) and Bloch
functions. In the next section we shall see how these real space concepts may
be obtained from the k-space picture that we have derived for the ring of N
atoms. The key finding is that although individual eigenstates may behave
in an often bizarre way, sums over eigenstates are much more stable and
predictable! And chemistry is about interactions between atomic orbitals,
which are linear combinations (sums) of eigenstates.
Bond order in an infinite systam

In Chapter 2 we introduced the concept of the bond order for the H 2
molecule. In this section we shall develop this concept for an infinite system
and take the infinite ring as a prototypical example.
Let us begin by thinking about the electronic charge density in the infinite
ring with atomic spacing a. The eigenstates of the ring are labelled by the k
vector, which we choose to define uniquely in the range -n/a to + n/a. The
normalized molecular eigenstate l'Pt ) is expressed in terms of atomic states
as follows
1 ~ .t.
l'Pt ) = N 1 / 2 /~'1 el Jali) (3.37)
where N is the (infinite) number of atoms in the ringo (Apologies for the
notation here! The i, j, k, and a in the exponent refer to four completely
different things: i is (_1)1/2, j is the atom Iying between 1 and N, k is the
wave vector Iying between -n/a and n/a, and a is the atomic spacing.) The
probability of finding an electron at x occupying this state around the ring
is l'Pt(x)12 = <xl'Pt)<'Ptlx). The electronic charge density at x for an
electron in this state is therefore - el'Pt (x Wwhere e is the electronic charge.
This is the charge density arising from just one electron in one molecular
orbital <xl 'lit)· The total charge density in the ring is the charge density
contributed by all occupied molecular orbitals. What determines whether a
molecular orbital is occupied? The exc\usion principIe says that each
molecular state may be occupied by a maximum of two electrons. Therefore,
we simply occupy the lowest energy molecular states (at zero Kelvin), with
two electrons each, until we have the required number of electrons in the
ringo If there are v electrons per atom then the M lowest energy molecular
states (O S; M S; N) are occupied where
2M
V=-. (3.38)
N
Here M and N are infinite but their ratio is finite! If the atoms of the ring
were neutral hydrogen atoms then v would be 1. We shall find it useful to
allow v to be any number between Oand the maximum possible, 2. AII states
E(k)
Unoccupied
_~ ___ +-_--'-_.¡-~< states
Occupied
states
Fig. 3.10 The Fermi energy, EF . Al! states in the infinite ring with energy up to
EF are occupied (at O K) and all states aboye EF are unoccupied. There are two
states with energy EF at k = ±kF .
up to sorne energy EF are occupied, while all states aboye EF are unoccupied,
as shown in Fig. 3.10.
EF is called the Fermi energy. The Fermi energy is determined entireIy by
v. Corresponding to EF there is a Fermi wave vector kF given by
(3.39)
The Fermi wave vector is found by using the density of states in k-space,
eqn (3.36). Since there are Na/2n states per unit length of k-space then
Na
2-kF = M, (3.40)
2n
where the prefactor of 2 takes account of the fact that for each state at k
there is another state at - k with the same energy. Hence,
nv
kF=-· (3.41)
2a
Thus for v = 2 we have kF = n/a, which is at the top of the band of energies.
The band is said to be fulL The Fermi energy is ca1culated from the Fermi
wave vector by eqn (3.39).
Now we can calculate the total charge density arising from the M occupied
molecular orbitals. Each molecular orbital can be occupied by two electrons,
each of charge -e. Since we sum over the occupied molecular orbitals that
are labelled by k up to kF , and k is a continuous variable, we have to perform
an integral
Na J,kF
p(x) = - k - o <x I'l'k ><'I'klx> dk. (3.42)
2n k= -kF
The factor of N a/2n before the integral arises from replacing a sum over
discrete k values k = 2nm/N a, where m is an integer, by an integral. (The
worrying infinite factor of N will be cancelled shortly.) Now let us express
this charge density in terms of the atomic orbitals <xli). Before we use eqn
(3.37) for l'I'k) it is instructive to write l'I'k) in the following way
l N
l'I'k) = NI/2 JI cj(k)\j) (3.43)
where cj(k) is the expansion coefficient of the kth molecular state in the jth
atomic state, and it is given by eqn (3.37) as
Cj(k) = e ikja • (3.44)
The charge density p(x) becomes
N N
= -e L L ppj<xlj)<plx) (3.45)
)= I p= I
where
(3.46)
This looks messy but if you take a closer look you will see it is very
transparento In the last line of eqn (3.45) we see that the total charge density
consists of a sum of terms of two kinds:
1. 'Diagonal' terms where P = j, giving contributions like pii<xlj)Glx).
This is the charge density that we may associate with atom j in the ringo
2. 'Off-diagonal' terms where p oF j, giving contributions like
pp)<xlj)<plx) + pjp<xlp)Glx) = (Ppj + pjp)<xlj)<plx)
since the s atomic orbitals are not complex functions. This is the charge
density we may associate with the bond between atoms p and j.
The matrix elements Ppj are expansion coefficients of the total charge
density in the atomic orbital basis set. They are called density matrix elements
and the N by N matrix with elements Ppj is called the density matrix. We
see from eqn (3.46) that the off-diagonal elements of the density matrix are
the bond orders that we introduced in Chapter 2 (see eqn (2.58) and the
discussion thereafter).
Using eqn (3.44) for the expansion coefficients cik) let us now evaluate
the density matrix elements p p ) in eqn (3.46). We shall see that the bond
order depends on k F , i.e. the number of electrons per atom:
kF
= ~ eikU- p)a dk. (3.47)
1t J,k= -ky
In evaluating this integral we have to be a little bit careful. Let us consider
the case where j = p first. It is then easy
F
PjJ=~ fk dk=2ak F • (3.48)
nJk=-kF n
Is it right? Yes, because (a) it is independent of which atom we look at (and
all atoms in the ring are equivalent), (b) when kF = Othen no molecular states
are occupied and. hence the charge associated with each atom is zero,
(c) when k F = n/a, the maximum allowed in the Brillouin zone, then the
charge on each atom is two e1ectrons-which agrees with the maximum
allowed by the exclusion principie for an s state, and finally (d) the charge
on each atom increases linearly with k F in agreement with the density of
states in k-space, eqn (3.36).
Now let us do the more interesting integral where j #- p. In that case eqn
(3.47) becomes
a J,kF .. a [eikFU-p)a - e-ikFU-p)aJ
PJP. =- e,kU-p)adk=_
'( • )
n k=-kF n IJ-pa
2 sin[kF(j - p)a]
- (3.49)
n j-p
(You will become very familiar with integrals of this kind (if you are not
already) when you study diffraction.)
First of all notice that pp } depends only on the difference between p and
j. In other words P12 is the same as P7S' This simply reflects the translational
symmetry of the ringo In Fig. 3.11(a) we plot P12, which is the density matrix
element, or bond order, for any nearest neighbour bond as a function of IkFI.
It is a maximum when the band is half-filled at k F = n/2a, and it is zero
when the band is empty (k F = O) or full (k F = n/a). Why?
When k F = O the band is empty. Therefore there can be no bonding and
the bond order is zero. Between kF = O and IkFI = n/2a we occupy the lower
half of the band where all the states are bonding sta tes. Remember we
assigned them as bonding states because their energies are below those of the
free atom oc. But between IkFI = n/2a and n/a we occupy antibonding states
in addition to the bonding states. This leads to a weakening of the bonds and
the bond order is thus reduced, until eventually the number of bonding states
(a) (b)
Fig.3.11 The bond order P12 between nearest neighbours in the infinite ring as
a function of (a) the Fermi wave vector and (b) the Fermi energy.
occupied is equal to the number of antibonding states occupied, at IkFI = n/a,

and the bond order is zero. This seems to imply that the bond order is
proportional to the difference between the number of bonding sta tes and
antibonding sta tes occupied in the bond.
We can make this more precise. in the following way. Consider the bond
order between atomic states li) and Ii). Form a normalized bonding state
1
lb) = 21 / 2 (Ii) + Ii»), (3.50)
and a normalized antibonding state

1
la) = 21/ 2 (Ii) -li»), (3.51)
between them. The number of electrons in the occupied eigenstate l'I't) in

the bonding state lb) is simply 2('I'tlb) (bl'l't), and similarly the number of
electrons in the antibonding state la) is 2('I'tla)<al'l't), where the factor
of 2 accounts for spin degeneracy (i.e. two electrons can occupy any quantum
eigenstate, with opposite spins). What we have done here is to project the
eigenstate l'I't) onto the bonding and antibonding states between the atomic
orbitals li) and Ii). Summing over all the occupied eigenstates between
k = -kF and k = +kF , the difference between the number of electrons in
occupied bonding states and occupied antibonding states is
r~ -tF 2('I'kl b) <bl'l't) -

a
2n 2('I'tl a ) (al'l't) dk. (3.52)
Substituting eqns (3.50) and (3.51) for lb) and la) into this expression we
find that it is simply 2p¡j' i.e. 2 times the bond order between atomic states
li) and li). We therefore arrive at the following useful interpretation of the
bond order:
The bond order is one half of the difference between the number of electrons
in bonding and antibonding states in the corresponding bond.
Returning to eqn (3.49) for the bond order Ppj we observe that as Ip - ji
increases, Le. as the separation between atoms p and j increases, the bond
order decreases in magnitude and displays an increasing number of oscil-
lations with band filling IkFI. Why? As the separation between the atoms
increases the coupling between the atoms must decrease beca use it involves
hops through intermediate atoms and thus the probability of a hop from p
to j decreases. Since the quantum mechanical bond depends on the prob-
ability of such hops we can expect the bond order to decrease as the
separation between the atoms increases. The reason for the increasing
number of oscillations is apparent if we look at the phase factors e ikja along
the chain for particular eigenstates. To focus the discussion consider P13,
which is the bond order for next nearest neighbour bonds
PI3 = ~ sin(2k Fa). (3.53)

71:
In Fig. 3.12 we show schematically the phase factors for the states k = O,
n/2a, and n/a, corresponding to the bottom, middle, and top of the bando
At the bottom of the band the state is bonding between all pairs of atoms
because all the atoms are in phase. In the middle oC the band (k = n/2a) the
state is neither bonding nor antibonding for first neighbour pairs, but it is
antibonding for next nearest neighbour pairs because they are now 180 out 0
of phase. At the top of the band the state is antibonding for the nearest
neighbour pairs but it is bonding for next nearest neighbours. This explains
the oscillation in PI3 which is absent in P12'
In phase
Bonomof o o o o
r-
o
-....o o o k=O
band
t t t t t t t t
180' out of phase
Middleof o o o o
r-
o o
-....o o k =!t/2a
band
~ t ~ t
Inpbase
,,-
Topof o o o o ---o o o o k =!tIa
band
~ t ~ t ~ t ~ t
Fig. 3.12 Schematic illustration of the phase factors (represented byarrows) for
the states k = O. 7I:/2a. and 7I:/a, for the infinite ring corresponding to the bottom,
middle. and top of the bando
The density of states: total and local

We now come to an extremely useful concept, the density of states. In many
texts the density of sta tes is introduced as a concept that applies to a whole
crystal, chiefly because those texts consider only perfect crystals where aH
atomic sites are equivalent. In this section we shaH generaIize the notion of
a density of sta tes to the concept of a local density of states, that applies to
one atom and is just as valid in an amorphous material as it is in a perfect
crystal. But we begin where most texts start: the density of states in an infinite
crystal.
Consider again the infinite ring of atoms. In eqn (3.36) we introduced the
density of states in k-space. This is simply the number of states in k-space
per unit length of k, and it is given by d(k) = Na/21t for the ringo We could
pose a slightly different question: what is the density of states as a function
of energy rather than wave vector? The range of energies in which states lie
is C( + 2fJ ~ E ~ C( - 2fJ. The density of states D(E) is going to be highest at
those energies where E does not vary much with k in Fig. 3.6, because then
for a given interval !:J.E there are more states in k-space. More precisely D(E)
for the infinite ring is given by
dS
D(E) = dS = dS dk = dk = 2 Na 1 N 1
dE dk dE dE 21t 2fJa sin ka - -; (4fJ2 - (E - C()2)1/2'
dk
(3.54)
The total number of states up to energy E has been represented by S, and
dS/dk is the density of states in k-space, N a/21t. The modulus of dE/dk is
taken because D(E) cannot be negative (because it refers to a number of
states), and it is evaluated using eqn (3.19) (where O has been replaced by
k). The factor of 2 arises beca use for each energy there are sta tes at k and
- k. The density of states D(E) given in eqn (3.54) is for the infinite ring,
and that is why there is a prefactor of N. The density of states for each atom
of the chain is thus
(3.55)
and this is plotted in Fig. 3.13. Note that it is a minimum in the middle of
the band where the slope of E(k) is greatest and it diverges at the band edges
where dE/dk tends to zero. If we integrate the density of states d(E) over
the whole band we get 1, which simply means that there is one state
associated with each atom.
Since aH atoms in the ring are equivalent the total density of states D(E) is
simply the number of atoms N, times the density of states on one atom d(E).
---L------~a~----~---.E
a-213
Fig. 3.13 The density of states per atom in an infinite ring of hydrogen atoms
(eqn (3.55».
0----0----<>---0---<;1--- 00
1 2 3 4 5
Fig. 3.14 A semi-infinite linear chain of atoms.
The same is true in 2- and 3-dimensional perfect crystals. But suppose we

consider a semi-infinite linear chain, as shown in Fig. 3.14. Clearly we can
no longer expect the density of states on atoms 1, 2, 3, ... etc. to be the same,
although we would not expect the density of states on atom 1000 to be very
different from that of an atom in an infinite ringo In a three-dimensional
crystal this statement is equivalent to saying that we do no! expect the density
of states on a surface atom to be the same as that of an atom in the bulk
of the crystal. Notice that we have switched from talking about a density of
states per atom, for the case where the atomic sites are al1 equivalent, to a
density of states on an atom where atomic sites are no longer equivalent. The
implication is that we can associate a density of states with each atom, and
that inequivalent atomic sites will have different densities of states. We are
using here the concept of a local density of states, d(E), as opposed to the
total density of states D(E) for the whole system. We are saying that the
local density of states varies with the local atomic environment. This
extremely important concept was introduced by Friedel (1954).
Can we be more precise about these ideas and write down a formula for
the local density of states? Consider an eigenstate l'Pk) which is expressed
in an orthonormal basis set of atomic states la), with expansion coefficients
<al'Pk)
(3.56)
•
AH atoms for which the expansion coefficient <al'Pk) is not zero contribute
to this eigenstate. F or this eigenstate the probability of finding an electron
at the particular basis state In) is l<nl'Pk)12. We can speak of l<nl'Pk)1 2 as
being a factor with which to weight the contribution to the total density of
states, D(E), from the eigenstate l'Pt ) at site n. Let the energy of this
eigenstate be Ek. If there is more than one eigenstate with energy E = Ek we
must sum the weights l<nl'l't>12 from aH of them to get the total weighting
factor at energy E for the local density of states for the basis state In>. But
before we can proceed any further we need a way of writing down the
contribution of the eigenstate l'I'k> to the total density of states D(E).
The total density of states D(E) is reaHy just a counter. The total number,
S(E), of states up to energy E is obtained by counting up the number of
eigenstates with energies less than E. The contribution of the eigenstate l'I't>
to the total number of states up to energy E is zero, unless E ~ Et whereupon
the contribution is one. For example, for our linear chain of five atoms (see
Fig. 3.3) S(E) is zero for E < C( + 3 1/2 fJ, 1 for C( + 31/2fJ:s; E < C( + fJ, 2 for
C( + fJ :s; E < c(, 3 for C( :s; E < C( - fJ, 4 for C( - fJ :s; E < C( - 31/2 fJ, and 5 for
E ~ C( - 31/2 fJ. Thus, there are steps in S(E) of unit height at the five
eigenvalues. Now, from the definition in eqn (3.54) the density of states D(E)
is the derivative of S(E): D(E) = dS(E)/dE. This derivative is zero at aH
energies except the five eigenvalues where it is undefined because S(E) is
discontinuous.
What we have shown is that S(E) and D(E) are reaHy discontinuous
functions, but we can understand them better by approximating them by
more familiar continuous functions and then taking the limit that the
approximations go to zero. As an example consider the foHowing approxi-
mation to the contribution to D(E) from the eigenstate with energy Ek
1 e
f(E - Et) = - 2 (E E )2 (3.57)
1I:e + - t
This continuous function is a Lorentzian centred at E = Et with a width e,

as shown in Fig. 3.15. If this approximation is any good we must obtain a
value of 1 when we integrate f(E - Et) from E = - ce to E = + ce in order
for the state to contribute one to the total number, S, of states in the system.
Indeed we do, for any finite value of e
(E - E)JE = + <Xl = 1.
f + <Xl
-1 2 e
-<Xl 11: e + (E - Et)2
dE = -1 [ tan- 1
11: e
t
E=-<Xl
(3.58)
¡(E-El)
I
En
Fig. 3.15 The Lorentzian described byeqn (3.57). As e tends to zero the width
of the Lorentzian tends to zero and its height tends to infinity, but the area under
it remains 1 at all times.
As the width e of the Lorentzian approaches zero its height tends to infinity
and the integral of f(E - El) over any small range of energies centred on
El tends to 1. In this way f(E ~ El) becomes a better approximation to the
contribution of the state at E = El to the density of states. In the limit that
e tends to zero we obtain an exact description of this contribution. The
function that we obtain by taking the limit that e -+ O is called a delta
function, b(E - El)
1 e
b(E - El) = Lt - 2 (3.59)
.~O 1t e + (E - Ek)2
lt is zero except at E = El where it is infinite, and it has the important

property that its integral over any range of energies inc1uding E = El is 1.
The conclusion we have reached is that the contribution of the eigenstate at
E = Ek to the total density of states is b(E - El) and therefore D(E) may
be expressed as follows
D(E) = L b(E - El)· (3.60)
all Ere
In the limit of an infinite system, where the number of eigenstates becomes

infinite and the energy interval between successive delta function contri-
butions to D(E) becomes infinitesimally small, then D(E) becomes a smooth,
continuous function. But it is still useful to keep in mind that it is really a
bunch of delta functions that are so c10sely spaced together we cannot resolve
them.
Now we are in a position to write down an expression for the local density
of states, d.(E), associated with the basis state In> at energy E. Each
contribution b(E - El) to the total density of states must be weighted by
the factor l(nl'Pk>12 and thus we arrive at the following expression for d.(E)
d.(E) = L
all Erc
l(nl'Pl>1 2b(E - El) (3.61)
Since this is so important let us go over it again. We pick a basis state In>
and an energy E. Only those eigenstates with eigenvalues El equal to E
contribute to the local density of states d.(E). To make sure we get all
contributions we sum over aH eigenstates of the system and the delta function
picks out a11 states with energies equal to E. Almost all the eigenstates will
not contribute to d.(E) because the delta function will be zero at the
particular energy E. But for those eigenvalues equal to E we work out
l(nl'Pl>12, multiply by b(E - El)' and add aH these terms together to
give d.(E).
The total density of states is recovered by simply summing all local
densities of sta tes
D(E) = L d.(E) = L b(E - El)' (3.62)
all n all EJe
where we have used the normalization condition of the eigenstate
¿ l<nl'l'k)1 2 = 1. (3.63)
all PI
Let US apply these ideas to the semi-infinite linear chain, shown in

Fig. 3.14, and compute the local density of states on the •surface' atom,
number 1. To do this we shall use the eigenstates for the finite linear chain
that we derived in eqn (3.13) and take the limit that the number of atoms
N tends to infinity. With an interatomic spacing of a the eigenstates of the
finite chain may be written as
2 )1/2 N
I'ID = ( ¿ sin(kja)Ij), (3.64)
N +1 j=l
where
mn
k=---,- and m= 1,2,3, ... ,N. (3.65)
(N + l)a
Therefore, the weighting factor for atom 1 is given by
1<lI'l'k)1 2 = 2 sin2 ka. (3.66)

N +1
From eqn (3.65) we see that there is one eigenstate along the k-axis in each
n/«N + l)a) interval. Therefore, in the limit of N tending to infinity
the density of eigenstates in k-space, dS/dk, is continuous and equal to
«N + l)a)/n. The relationship between the eigenvalues El and k is
El = ex + 2p cos ka (see eqn (3.12» where k is given by eqn (3.65). The total
density of states for the semi-infinite chain is therefore
dS
D(E) = dS = dS dk = _d_k = (N + l)a 1 (3.67)
dE dk dE· dE n 2Pa sin ka
dk
where E = ex + 2p cos ka. From eqn (3.60) this is equal to ¿,IIE. t5(E - El)'
To get the local density of states, d 1(E), associated with basis state 11> we
have to multiply each one of these delta functions by the weighting factor
1<II'1't)12, given by eqn (3.66), and thus we obtain
d 1(E)
_ (N + l)a 1 2. 2 k
- Sin a
n 2Pa sin ka N + 1
1 .
=-sInka
np
= _1_ (4p2 _ (E _ ex)2)1/2. (3.68)

2np2
d(E)
~~----~a~--~~~E
,,
a-2p
Fig. 3.16 The local density of states, given by eqn (3.68), for the end ato m of
a semi-infinite linear chain of hydrogen atoms.
This local density of states is plotted in Fig. 3.16, and it should be compared
with the local density of sta tes for an atom in an infinite ring, shown in
Fig. 3.13. Note that the local density of states for the 'surface' atom is peaked
at the middle of the band and is zero at the band edges. This is the opposite
to Fig. 3.13. The integral of the local density of states over the whole band
is 1 as it should be (try it!).
Our simple calculations in this section have established something very
significant: the local density of states is dependent on the local atomic
environment. In fact there is an astonishingly direct and powerful link
between the local atomic environment and the local density of states that is
known as the moments theorem. This theorem enables us to deduce features
about the shape of the local density of states purely from the topology of
the local environment without all the hassle of determining the eigenstates
of the system. We shall explore this link later in the chapter.
Band energy and bond energy

The procedure for finding the electronic energy levels for the H 2 molecule
is essentially the same as it is for an HN molecule where N is finite or infinite.
We find the eigenstates of the system and populate the lowest energy states
with electrons according to the exclusion principie until we have the required
number of electrons in the system. The sum of the occupied eigenvalues is
called the band energy, E band
Eband =2 LE•. (3.69)
n occupied
The factor of 2 is for spin degeneracy. In an infinite system the eigenvalues

are very closely spaced and this sum is replaced by an integral over energy
involving the total density of states D(E)
EF
E band =2
f ED(E) dE.
The integral is taken from the bottom of the band to the Fermi energy EF
(3.70)
which is the energy of the highest occupied state at zero Kelvin. (At a finite
temperature we have to inelude the Fermi-Dirac distribution function, but
that is an added complication that we defer until Chapter 8.)
In simple theories of cohesion the band energy is considered to be the
source of the binding of the molecule or solido The thinking is that valence
e1ectrons enter molecular sta tes that have energies lower than the atomic
sta tes and thereby lower the energy of the system. As the atoms get eloser
together the hopping integrals increase in magnitude and the band width
increases. As the band width increases the energy lowering that is achieved
by occupying molecular states is greater still. So why do molecules and solids
not collapse to form very dense matter? Of course the answer is that once
atoms become so elose together that the core electrons start to overlap there
is a very strong repulsion, the source of which is partly electrostatic but it
is mainly derived from the exelusion principie. At equilibrium there is a
balance between this repulsion tending to drive atoms apart and the band
energy favouring eloser atoms.
One obvious problem with the band energy is that it is an absolute energy
as written in eqn (3.70). Going back to the HN molecule we recall that the
parameter IX, which is the on-site Hamiltonian matrix element, simply
determines the centre of gravity of the bando If we add 100 eV to IX the whole
band is rigidly shifted up by 100 eV, as may be seen from eqn (3.19). But
c1early this shift does not alfect the cohesive energy of the solid beca use we
subtract from the band energy the energies of the electrons in their original
atomic states, which is just IX. It makes more sense, therefore, to discuss the
quantity defined by
EF
E bond = 2
f (E - IX)D(E) dE (3.71)
which is called the bond energy. At each energy E the factor (E - IX) is the
energy of the molecular state relative to the atomic state IX. Alternatively,
since IX is just a constant it may be taken outside the integral
EF fEF
E bond = 2 f ED(E) dE - 2IX D(E) dE
(3.72)
where N. is the total number of electrons in the system.
We have called the quantity in eqn (3.72) the bond energy because when
we bring atoms together and ask for the change in electronic energy, as we
are in eqn (3.72), that change must come from the formation of bonds. There
is an exact transformation ofthe bond energy that makes this explicit, which
we shall now develop. It is essentially the same analysis as our discussion of
the bond energy in the H 2 molecule at the end of Chapter 2. It is helpful to
think of a finite molecule again and then take the limit that the number of
atoms becomes infinite. For a finite molecule the integral in eqn (3.72)
becomes a discrete sum over the eigenvalues Ek
Ebond = 2 1: (Ek - oc) (3.73)
k occupied
where we have assumed that each occupied molecular state has two
electrons in it. We recall that Et is an eigenvalue ofthe Schrodinger equation,
so that
Et = <'PtIHI'ID
= 1: 1: <'Ptli><iIHU><il'Pk > (3.74)
i j
where we have inserted a complete set of orthonormal atomic basis states

{Ii)} twice (see eqns (2.4), (2.18». The sums run over aII the basis states.
Equation (3.74) is simply a rearrangement ofthe Schrodinger equation. The
matrix elements <¡IHU> are the Hamiltonian matrix elements Hij and <il 'lit >
is the expansion coefficient of the eigenstate l'Pt> corresponding to the basis
state U>. Substituting this expression for Et into the bond energy, eqn (3.73),
we get
Ebond = 21:1: 1: <'Pkli)WIHU> - oc<ili»<iI'Pk >
l j 1: occupied
= 1: 1: (2 1: <i1'Pt><'Ptli»WIHU> -oc<ili». (3.75)

i J k occupied
We recognize the term in large brackets as the density matrix element Pji
(see eqn (3.46». Furthermore <¡U> = O unless ¡ = i in which case it is 1.
Since <iIHli> = oc we deduce that there are no terms in the sum arising from
i = i. Our final result is therefore the following
(3.76)
It is easy to see that a similar expression exists for E band
(3.77)
In going to the limit of an infinite molecule we simply replace the discrete

sums over occupied states in eqn (3.73) by integrals. The results, eqns (3.76),
(3.77), remain valid.
The bond energy is seen in eqn (3.76) to involve literally a sum of individual
bond energies. The energy of the bond between sta tes Ip> and Iq> for instance
is pIH/Hqp + POpHIH/' The density matrix elements Ppo and POP are complex
conjugates of each other, as are HIH/ and H op . Thus the bond energy is real,
as it should be, and it is proportional to the bond order and the hopping
integral.
The band energy in eqn (3.77) consists ofthe bond energy plus the 'on-site
energies' p¡;H¡;, which is simply the number of electrons at site i times ex.
Although this exact transformation was derived by Coulson (1939) its
usefulness in understanding solids has been appreciated only in the last 10-20
years. The usefulness of the transformation derives from the fact that it relates
the k-space picture of solids, in which we work only in terms of eigenstates
of the whole system defined in k-space, eqns (3.70), (3.71), to the real space
picture of solids, in which we work with atomic orbitals, bonds, and charges
on atoms, eqns (3.76), (3.77). The transformation does not rely on trans-
lational symmetry and therefore eqns (3.76), (3.77) are just as valid in an
amorphous solid as they are in a crystalline solido But the real power of the
transformation derives from the fact that we do not need to know the eigenstates
of the system at al/ provided we can evaluate the density matrix elements by
some other way. In this way we can escape the severe restrictions of
translational symmetry (without which we cannot hope to find the eigen-
states of 1023 atoms) and study defective crystals, amorphous solids,
and liquids. The importance of this in materials science cannot be over-
emphasized!
There are methods of determining the density matrix elements which do
not need the eigenstates of the system. One of them, the 'recursion method',
is too advanced for the present book but we shall develop sorne of the ideas
on which it is based. The whole philosophy is completely different from
traditional solid state physics and one begins by 'throwing out k-space' to
quote Volker Reine. Instead the emphasis is on the local atomic environment
and how that determines the density matrix. This is a much more chemical
approach to solid state physics. One of the central pieces of this view of
solids is the moments theorem.
The moments theorem

This theorem, which was derived by Cyrot-Lackmann (1968), relates
moments of the local density of sta tes to the topology of the local atomic
environment. Consider the local density of states for atom i. From eqn (3.61)
we have
d¡(E) = L
all Ek
c5(E - Ek)<il'l'k> <'I'kl i>. (3.78)
Let us consider the 'nth moment' of this local density of states
¡tI") = f
wholc band
(E - H¡¡)"d¡(E) dE. (3.79)
For example, ¡tIO) = 1 beca use this is simply the integral of the local density
di(E) ,
,
Fig. 3.17 A local density of states with a large negative third momento The centre
of gravity is at H¡¡. Notice the long tail at energies below H¡¡ and the compressed
peak above H¡¡.
oC states itself over the whole band (which is 1 owing to the normalization
condition oC the basis (ili> = 1). The first moment, JI~l), is the centre oC
gravity oC the local density oC states relative to H¡¡. We shall see shortly that
JI?) is always zero because the centre oC gravity oC the local density oC states
is Hil' The second moment, JlI 2 ), is the moment oC inertia oC the local density
oC states relative to the centre oC gravity. The square root oC Jl12) is a measure
oC the width oC the local density oC states in the root mean square sense. The
third moment measures the skewness oC the local density oC states about the
centre oC gravity. A large negative value oC JlP) corresponds to a long tail in
the local density oC states below the centre oC gravity oC the band and a more
compressed peak aboye the centre oC gravity, as shown in Fig. 3.17. The
Courth moment measures the tendency Cor a gap to Corm in the middle oC
the band, as shown in Fig. 3.18. A low value oC the normalized Courth
moment, JlI4 )/(JlI 2 )2, corresponds to two well-separated peaks or bimodal
behaviour; in the local density oC states, whereas a large value corresponds
to a central peak or unimodal behaviour. As we shall see in Chapter 9 higher
moments are important in questions oC the stability oC competing crystal
structures, but the shape oC the local density oC sta tes is most easily interpreted
in terms oC the second, third, and Courth moments.
Inserting eqn (3.78) Cor d¡(E) into eqn (3.79) the nth moment oC the local
dilE)
__L-______~----~~--.E
H¡¡
Fig. 3.18 A bimodal local density of states, which is to be expected when s.

given by eqn (3.89), is less than 1.
density of states becomes
/ll") = f L
whole band aU E"
(E - HiJ" ó(E - Et)<il 'l't) <'I't Ii) dE. (3.80)
Expanding (E - H¡i)" in a Taylor series about E = Et we find that the

integral of (E - Hu)" times ó(E - Et) is just the first term of the series, i.e.
(Et - Hu)". Therefore, summing over all eigenvalues Et we get
/ll") = L <il'l't)(Et - Hu)"<'I'tl i). (3.81)

a1l Ek
We interpret this as the i-i matrix element of an operator W, where
W = L l'I't)(Et - H/i)"<'I'tl. (3.82)

.u EJe
This operator is simply
W= (H - Hu)" (3.83)
as may be easily verified by using
H= (3.84)
aH ellcoltates
It follows that
/ll") = <il(H - H¡i)"li). (3.85)
Thus, /lIO) = <ili) = 1, and ¡.¡JI) = <il(H - Hu)li) = H/l- H¡¡<ili) = 0, which
confirms that the centre of gravity of the local density of states is always
Hu. But what about /l12)? To evaluate /l12) we use the completeness of the
basis set
/l12) = <il(H - HU)2Ii) = <il(H - Hll)(H - Hll)li)
= L¡o <ilH - Hllli') <i'IH - H¡¡Ii).

.
(3.86)
The sum over i' is over the whole basis seto But in a nearest-neighbour
model, only those atoms i' that are neighbours of atom i have nonzero
Hamiltonian matrix elements <iIHli'). Moreover, <ilH - H¡¡Ii) = 0, and
<ilH - Huli') = Hil" for i # i'. Thus we may simplify eqn (3.86) as follows
(3.87)
Each term Hil' H¡.¡ describes an electron starting at site i. hopping out to a
neighbouring site i'. and hopping back to i. This is described as a path of
length two hops from atom i. The second moment of the local density of
sta tes is thus the sum of all such paths of length two hops. and it is illus-
trated in Fig. 3.19(a) for a two-dimensional hexagonal crystal. In a perfect
crystal where the coordination number is z and the nearest neighbour
0Ji~
(a) Hexagonal network
O . JiO 11¡2) = 6Ji2
I
O O
°iO
(b) Hexagonal network
O 11¡3) = 12Ji3
O O
Fig. 3.19 Paths contributing to (a) the second and (b) the third moments of
the local density of states in a two-dimensional hexagonal crystal.
hopping integral is p, the second moment of the local density of sta tes is
thus Zp2. The root mean square width ofthe band is thus proportional to zl/2.
It is straightforward to generalize this result to higher moments. For
example, the third moment is given by
11)3) = ¿ ¿ HwHi'i',Hj"j (3,88)

1':;I:II"#lori'
which is the sum of a11 paths of length three hops starting and ending on
site i, as illustrated in Fig. 3.19(b), Thus we arrive at the moments theorem:
The nth moment of the local density of states on atom i is the sum of all paths
of length n hops starting and ending at site i.
Let us pause for a moment to appreciate this remarkable result. The

determination of the local density of states is almost invariably hard work.
But without knowing the precise form of the local density of sta tes we can
obtain, through the moments theorem, exact values of the moments of
the local density of states simply by counting c10sed paths around the
site in real space, Knowing only the first four moments we can make
intelligent comments about the shape of the local density of states, as we
described earlier in this section, Thus by simply looking at the atomic
environment surrounding two nonequivalent sites we can make qualitative
comments, guided by exact results, about differences in their local densities
of states.
Let us try the theorem out on two local densities of sta tes that we evaluated
earlier for the infinite ring and the surface atom on the semi-infinite linear
chain. The local density of states for an atom in the infinite ring is given in
eqn (3,55) and the local density of states for the end atom of the semi-infinite
linear chain is given in eqn (3,68), We could evaluate the moments of these
local densities of states by evaluating the appropriate integrals. But suppose
we did not know the functional forms of local densities of states. We can
(a)
ce::::n O 0----00 11;2) =~2
(b)
c@) O
+
0----00
11;4) =2~4
~ 0----00
(e) -00--0 O
~ 0--00
+
_00 --O>----'~___--<O)_--O-- 00
(d) _00 +-o---.o)_---<~)---o-- 00
_00 +-O---.O)_---<%@)___--<O___--o--oo
_00 +-O>----.~___--<O)---O-- 00
Fig. 3.20 Graphieal evaluation of the seeond and fourth moments of the local
densities of states for (a), (b) the end atom of a semi-infinite linear ehain and
(e), (d) an ato m in an infinite ringo In (a) there is only one path of length two
hops and henee ¡l~2) = p2. There are two paths of length four hops and henee
¡l~4) = 2p4. In (e) there are two paths of length two hops and henee ¡l}2) = 2p2.
In (d) there are three paths of length four hops to the right and there are another
three to the left, henee ¡l}4) = 6p4.
use the moments theorem to write down the moments straight away.
Consider the end atom of the semi-infinite linear ehain first. The third and
all other odd moments are zero beeause there are no closed paths eomprising
an odd number of hops. Hence the local density of states must be an even
funetion. The seeond moment of the local density of sta tes is given by the
sum of all paths of length two hops starting and finishing on the end atom.
As shown in Fig. 3.20(a) there is only one sueh path and henee /1;2) = f12.
The fourth moment is given by the sum of all closed paths of length four
hops. As shown in Fig. 3.20(b) there are two sueh paths; henee /1;4) = 2f14.
These values may be eonfirmed by direet integration using eqn (3.68). Now
eonsider the infinite ringo Again we find the odd moments are all zero, and
therefore the local density of states must be even. There are now two paths
of length two hops, as shown in Fig. 3.20(e), and hence /I~2) = 2f12. Similarly
there are now six paths of length four hops, as shown in Fig. 3.20(d), and
hence Jl~4) = 6/34. These values may also be confirmed by direct integration
using eqn (3.55).
What could we deduce from these moments about the local densities of
state if we do not know their functional forms? Firstly, the 'width' of the
local density of states for an atom in the infinite ring is greater, in the root
mean square sense, than that of the end atom of the semi-infinite linear chain
because ~2) is double Jl!2). Secondly, the 'normalized' fourth moment for an
atom in the infinite ring, Jl~4)/(Jl~2»2 = ~, is less than the normalized fourth
moment for the end atom of the semi-infinite linear chain, /1!4)/(/1!2»2 = 2.
We have already alluded to the fact that the fourth moment enables us to
discriminate between unimodal and bimodal forms of the local density of
states. It turns out that the precise criterion is as follows. We first calculate
the dimensionless parameter s
/1(4)/1(2) _ (/1(2»3 _ (/1(3»2
s = '-----'-----:(::..../1(;:;'2)c'-):;-3-"----'- (3.89)
If s ;;:.: 1 we have unimodal behaviour. If s < 1 we have bimodal behaviour.

Notice that if /1(3) = O then
/1(4)
S = (2) 2 - 1. (3.90)
(/1 )
Since s = ! for the atom in the infinite ring the local density of states has
a bimodal form, whereas s = 1 for the end atom of the semi-infinite chain
indicates that its local density of states has a unimodal formo Both densities
of states are symmetric about E = IX. Thus we would sketch something like
Fig. 3.21 for the two local densities of states. As can be seen by comparison
with the exact local densities of sta tes in Figs 3.13 and 3.16, these diagrams
are pretty close to the exact densities of states. The local atomic environment,
thriJugh the numbers of closed paths and the moments theorem, has given us
insight into the forms of the local densities of states.
d,(E) d,(E)
(a) (b)
Fig. 3.21 Approximate local densities of states for (a) the end ato m of a
semi-infinite linear chain and (b) an atom in an infinite ring, deduced from their
first four moments. Compare with the exact local densities of states in Figs 3.16
and 3.13 respectively.
Tha binary AS alloy

Finally in this chapter let us illustrate the concepts that we have gained by
considering an infinite ring of alternating A and B atoms. Let the on-site
Hamiltonian matrix elements be eA and eR, with eA > eR, and let the hopping
integrals be zero except between nearest neighbours where they are {J. In
Problem 15 you are asked to show that there are now two bands of states: a
band of bonding states Iying between e - (&2 + 4{J2)1/2 and eR and a
band of antibonding states Iying between eA and e + (&2 + 4{J2)1/2, where
B = (eA + eR)/2 and & = (eA - eR)/2. The total density of states for the alloy
has contributions from both A and B atoms. The local densities of states,
nA(E) and nR(E) on the A and B atoms are different reflecting the differences
in the properties of free A and B atoms. The centres of gravity of the local
densities of states nA(E) and nR(E) are eA and eR respectively. You are asked
to derive these local densities of states in Problem 16. Here, we shall find
out as much as we can about the local densities of states using the moments
theorem.
The moments theorem is slightly more complicated now that there are
two on-site energies eA and eR. We shall find that this leads to nonzero odd
moments. Consider first the second moments ~r¡) and ~~2). They are both
equal to 2{J2, as before for the infinite ring of atoms of just one kind. Thus
the widths of the local densities of states, in the root mean square sense, are
the same. The third moment ~r) is given by
<AI(H - eA)3IA> = L L <AIH - eAlm> <mlH - eAU><iIH - eAIA> (3.91)
m j
where lA> is a state on an A atom and the sums over j and m are over all
atomic states. There are no contributions to this sum from paths of three
hops length because there are no such paths which start and finish on the
same atom. However, we can hop out to a neighbour m "# A, wmch is a B
atom, pick up the factor <mlH -' BAlm> = eR - eA' and hop back to A. There
are two such paths, and in this way find ~r) is given by 2{J2(eR - eA). Similarly
J1.~3) is given by - 2{J2(eR - eA). Thus ~r) is positive and equal and opposite
to ~~3). Hence the local density of states nA(E) is skewed positively and nR(E)
is skewed negatively.
The fourth moment ~i4) is given by
<AI(H - eA)4IA>
= L L L <AIH -
.. j •
eAlm> <mlH - BAU><jIH - eAln><nIH - eAIA>. (3.92)
It is easily shown that in addition to the six c10sed paths of length four hops,
giving 6{J4, there are two c10sed paths of length two hops and two •on-site'
terms, giving 2{J2(eR - eA)2. Thus,
~i4) = 6{J4 + 2{J2(eR - eA)2. (3.93)
Referenc.. 73
Tbis is the same as JJL4). The Cact that JJ(4) is greater Cor the AB alloy than
Cor the infinite ring oC just one kind oC atom indicates that the local densities
oC states in the alloy case have a smaller tendency Cor bimodal behaviour
and more oC a tendency Cor unimodal behaviour. However, by calculating
the parameter s, given by eqn (3.89), it is seen tbat tbe local densities oC
states never attain a unimoda1 Corm (Cor which s ~ 1).
References
Coulson, C. A. (1939). Proc. R. Soco Lond. A, 139, 413.
Cyrot-Lackmann, F. (1968). J. Phys. Chem. Solids, 19, 1235.
Friedel, J. (1954). Adv. Phys., 3, 446.
4
Into two and three dimensions
Solids as giant molecules

In the last chapter we saw that an infinite chain of interacting atoms may
simply be regarded as a one-dimensional molecule. If the chain does not
consist of a periodic sequence of atoms the task of determining the eigenstates
of the infinite system is impossible because it involves finding the eigenvalues
and eigenvectors of a matrix of infinite dimensions. But if periodicity is
present then Bloch's theorem enables us to write down the eigenstates ofthe
infinite system almost at a stroke as Bloch functions. In this chapter we shall
show how these ideas are readily applied to two- and three-dimensional
infinite perfect crystals. In this way we shall develop the standard machinery
of the band theory of solids.
However, we shall certainly not abandon our real space picture of solids
in terms of local densities of states, the moments theorem, and bond orders.
The relationships between the real and reciprocal space pictures that we
deve10ped in the last chapter for one-dimensional systems will be shown to
be just as valid in two and three dimensions. Thus, we can use Bloch's
theorem to analyse the eigenstates of an infinite periodic system and we can
use the local densities of states and bond orders to analyse the chemical
bonding.
The square lattice

We now consider an infinite two-dimensional square lattice which we
decorate with one atom at each lattice site. For simplicity we shall again
allow only one s state at each atomic site. This simplification enables us to
focus on the geometrical aspects of the problem, which remain the same for
any square lattice regardless of the number of basis states we associate with
each atomic site. In Chapter 6 we shall consider a real solid, namely silicon,
where the smallest number of basis states that we need to describe the band
structure reasonably well is one s state and three p states per atom. In order
to avoid surface atoms at the edge of the two-dimensional lattice we again
apply periodic boundary conditions along both dimensions of the square
lattice. In one dimension we saw that this was achieved by joining the ends
of a linear chain to form a ringo In two dimensions you can imagine that it
is equivalent to putting a very slightly distorted square lattice onto the
surface of a sphere. The distortion of each unit cell tends to zero as the radius
of the sphere tends to infinity.
Given that we have translational symmetry in two directions we can apply
Bloch's theorem as we did in the one-dimensional case in eqn (3.33). In two
dimensions the theorem becomes
(4.1)
where T is any one of the translation vectors of the two-dimensionallattice.
The vector k is called the wave vector and is the means of labeJling the
quantum eigenstates of the lattice. Now let us apply this theorem to our
'molecular sta te , for the lattice described as a linear combination of atomic
states
l'Pk> = I ck(R)IR>, (4.2)
R
where IR> denotes the atomic state at the lattice vector R. The ck(R)s are
the expansion coefficients for the molecular state, which we do not know
and want to determine. The sum in eqn (4.2) is taken over alllattice sites R
of the two-dimensionallattice. We have labelled the molecular states in eqn
(4.2) with a k in anticipatíon of the result that they wiJI be labelled by the
wave vector.
Using eqn (4.2) the molecular state at r + T is
<r + TI'Pk> = Ick(R)<r + TIR>

R
=I
R
ck(R)<rIR - n· (4.3)
Bloch's theorem demands that this equals

<r + TI'Pk> = eik.T<rl'Pk >
= eik.TI ck(R)<rIR>
R
= eik.TI ck(R - T)<rIR - n. (4.4)

R
Comparing the coefficients of <rlR - T» in eqn (4.3) and eqn (4.4) we

realize that
ck(R) = eik.Tck(R - T). (4.5)
This can be satisfied only if
(4.6)
where N is a constant that can be absorbed into the normalization constant
for the molecular eigenstate. As in the one-dimensional case Bloch's theorem
fixes the expansion coefficients and thus we do not have to worry about
76 Into two and three dimensions
diagonalizing an infinite dimensional matrix-it is all done! Precisely the
same argument applies in three dimensions as wel!.
The molecular eigenstates of the two-dimensional lattice are thus
l'I'k> = normalization constant x L eik'RIR>, (4.7)
R
where the normalization constant is determined by the condition

<'I'k I'Pk > = 1. Thus, if we say there are N 2 lattice sites in the infinite
two-dimensional lattice (and hence N2 is infinite!) then the normalization
constant is lj(N2 )1 / 2. Here we are making our usual assumption that the
basis states form an orthonormal set. Therefore the eigenstates of the
two-dimensional lattice are
(4.8)
This is as far as Bloch's theorem will take uso We now insert this eigenstate
into the Schrodinger equation, HI'I'k> = E(k)l'I'k>
1
(N2 )1/2
LR eik'RHIR> = (NE(k) L eik'RIR>.
)1/2 R
(4.9)
2
Projecting this equation onto an arbitrary atomic state IR'> by multiplying

from the left with the bra <R'I we get
E(k) = L eik'(R-R')<R'IHIR>. (4.10)
R
In a nearest neighbour model an electron can hop from an atom to any of

its four nearest neighbours in the square lattice. Let the nearest neighbour
hopping integral, <R'IHIR>, be p, which is negative for s states. Let the
on-site Hamiltonian matrix element <R'IHIR'> be IX. Although the sum in
eqn (4.1 O) is over all N 2 lattice sites only five of them are not zero in this
nearest neighbour model. If we displace atom R' to the origin its nearest
neighbours are at (1, O), ( -1, O), (O, 1), and (O, -1). Thus we get
(4.11)
where k = (k x , ky) and 'a' is the nearest neighbour distance, which is the
lattice constant of the two-dimensionallattice.
This equation still imposes no restriction on k. However, if we have two
wave vectors k and k' that differ by any vector G = (21tja)(m, n), where m and
n are integers, they will denote states that have exactly the same energy E(k).
Allowing m and n to take on al! integer values we generate another square
lattice in which the lattice constant is 21tja. This is the reciprocal lattice of
the square lattice. Not only are the energies of the states l'I'k> and l'I'k+G>
equal but the phase changes that the states undergo, following a lattice
translation T, are also equal
<r + TI'I'k> = eik'T<rl'l'k>
and
(4.12)
where we have used eiG' T = 1, which holds Cor any reciprocallattice vector
G and any direct lattice translation vector T. Thus there is no means oC
distinguishing between the eigenstates l'I'k> and 1'I'k+G>' All eigenstates oC
the system may be represented uniquely within a region oC k-space defined
by -1t/a S;; k"" ky S;; 1t/a, which is the first Brillouin zone. Any vector k lying
outside the Brillouin zone may be brought to He within it by adding sorne
reciprocal lattice vector. This is called the reduced zone scheme. The first
Brillouin zone Cor the square lattice is shown in Fig. 4.1. In the extended
zone scheme we do not restrict k to He within the first Brillouin zone and
allow it to range over the whole oC k-space. In this way we see that E(k) is
a periodic Cunction oC k with the periodicity oC the reciprocal lattice. The
two schemes are equivalent, and it is simply a matter oC convenience as to
which one chooses to represent the band structure E(k).
It is not difficult to show that there are N2 states contained within the
first Brillouin zone. Let there be N", lattice sites in the (1, O) direction oC the
square lattice, where N", is infinite really. Similarly there are Ny sites in the
(0,1) direction, so that N",Ny = N2 • The imposition oC periodic boundary
conditions along the (1, O) and (O, 1) directions leads to the conditions that
2m1t
e'
"k N
x "" =1 =>k", =-N
",a
' where m is any integer
and (4.13)
2n1t
eik.N"a =1 => k =-
y Nya '
where n is any integer.
. I"X"'/'
, --'0'/"'~-,
FirstBrillouin
zone
"X' . X" •
. .
.... . .
....
-- --
.¡
... .. I
I ... .. I
X
,, ,, 21tla
~
21tla
Fig. 4.1 The construction of the first Brillouin zone for a square reciprocallattice.
of lattice constant 21[/a. The broken lines show the perpendicular bisectors drawn
between the central site and the neighbouring reciprocal lattice vectors. The first
Brillouin zone is shown with solid lines and is the inner envelope of the broken
lines.
The spacings between successive allowed values of kx and k, are 2n/Nxa and
2n/N,a respectively. There are Nx states in the range -n/a < kx :5: + n/a and
N, states in the range -n/a:5: k, :5: + n/a. Thus in the first Brillouin zone
there are NxN, = N2 states altogether.
The eubie lattiee

Having seen how to apply Bloch's theorem in two dimensions in the last
section we can immediately generalize the result and apply it in three
dimensions to a simple cubic lattice with lattice constant a. Again we apply
periodic boundary conditions in all three directions. Following the same line
of reasoning as before we deduce that the molecular states of the infinite
three-dimensional crystal are
(4.14)
where the only change is that the normalization constant involves the
number, N3 , of lattice sites in the three-dimensional crystal and R is now a
lattice vector of the three-dimensional crystal. The wave vector k has three
components (k x, k" k.). Putting this eigenstate into the SchrOdinger equation
we find that the corresponding eigenvalue is
E(k) = IX + 2/J(cos kxa + cos k,a + cos k.a). (4.15)
We have again adopted a nearest-neighbour model in which the hopping
integral between nearest neighbours is /J and IX is the on-site Hamiltonian
matrix element. The eigenvalues fall within a range of values between IX + 6/J
and IX - 6/J, which is called the energy bando
The first Brillouin zone is a cube in k-s pace centred on the origin with
side 2n/a. The reciprocallattice is a cubic lattice with lattice constant 2n/a.
The first Brillouin zone contains N3 states.
Brillouin zones for the f.e.e. and b.e.e. lattiees

Apart from the metal polonium there are no elements with the simple cubic
structure. Most metals have either the f.c.c., b.c.c., or h.c.p. structures. In this
section we shall construct the Brillouin zones for the f.c.c. and b.c.c. lattices.
The region of k-space that is associated with the first Brillouin zone is defined
by the condition that any point in the first Brillouin zone is nearer to the
origin of the reciprocal lattice than to any other reciprocallattice site. It is
constructed by (i) picking a reciprocal lattice vector G and drawing a line
from the origin to the point G, (ii) drawing the plane which perpendicularly
bisects this line, (iii) repeating this procedure for all the vectors of the
reciprocal lattice, and (iv) finding the inner envelope of these planes. This
Brillouin zonas of tha f.c.c. and b.c.c. lattices 79
procedure is shown for a square lattice in Fig. 4.1. lt is straightforward to
apply the procedure to any other reciprocal lattice.
Consider an f.c.c. lattice in which the primitive translation vectors are
t 1 = a/2[1 1 O], t 2 = a/2[1 O 1], and t 3 = a/2[0 1 1]. The corresponding
reciprocal lattice vectors are given by
t "1 -_ 27t t2 X t 3 - _ 27t [1

- 1 -1],
t 1 ·(t2 xt 3 ) a
t *-2
2 - 7t t 3 xt 1 -_27t[1
- -1 1], (4.16)
t 1 ·(t 2 x t 3 ) a
t "3 -_ 27t t 1 X t 2 -_ -27t [ - 1 1 1].

t 1 ·(t 2 x t 3 ) a
These reciprocal lattice vectors are the basis vectors of a b.c.c. lattice in which
the lattice constant is 47t/a. Thus, the reciprocallattice of an f.c.c. lattice is
a b.c.c. lattice. The first BrilJouin zone is shown in Fig. 4.2. The reciprocal
lattice vectors 27t/a( 1 1 1) give rise to eight hexagonal faces and the six
square faces arise from the reciprocal lattice vectors 47t/a(l O O) .
The primitive lattice translation vectors of a b.c.c. lattice are
t 1 = a/2[1 1 -1], t 2 = a/2[1 -1 1] and t 3 = a/2[ -1 1 1]. The corre·
sponding reciprocal lattice vectors are given by
" t2 X t3 27t
t 1 =27t =-[1 1 O],
t 1 ·(t2 x t 3 ) a
t2
* -_ 27t t 3 X t 1 _ 27t [1
-- O 1], (4.17)
t 1 ·(t 2 x t 3) a
"_ 2 t 1 X t 2 _ 27t [O 1 1].
t 3 - 7t --
t 1 ·(t 2 x ' 3 ) a
Fig. 4.2 The first Brillouin zone for the face-centred cubic lattice.
Fig . 4.3 The first Brillouin zone for the body-centred cubic lattice.
These are the primitive lattice vectors of an f.c.c. lattice with lattice constant
41t/a. The first Brillouin zone is shown in Fig. 4.3. The 12 sides of the figure
arise from the 12 reciprocallattice vectors 21t/a(1 1 O).
The band structure E(k) is a function ofthe three-dimensional wave vector
within the BrilIouin zone. It is usualIy plotted a10ng particular directions
within the Brillouin zone, for example from the origin to the centre oC one
of the faces of the zone, or one of the corners of the zone.
Equation of motion of an electron under an

externalforce
We have seen that the quantum eigenstates in a perfect crystal are
labelled by a wave vector that may be uniquely defined within the first
Brillouin zone. Let us consider what happens to an electron occupying a
state l'I'k) when we apply an electric field ~ to the crystal. Prior to the
application of the electric field the group velocity of the electron is given by
(see Problem 13)
(4.18)
The electron experiences a force - e~ due to the electric field and it

accelerates. Since its energy depends only on k the value of k changes.
Let us calculate the rate at which k changes. The amount of work done by the
electric field when the electron is displaced by "k t5t in the time interval
c5t is
(4.19)
This must equal the change in the energy of the electron due to the
concomitant change in k
(4.20)
Equating eqn (4.19) and eqn (4.20) and using eqn (4.18) we deduce that
h dk
-e<; = --o (4.21)
2n dt
This equation is exactly the same as for a free electron, Le. an electron moving
through a vacuum. The total momentum of a free eIectron is hkj2n. For an
electron in the periodic potential of a crystal we call hkj2n the crystal
momentum of the electron. The significance of the crystal momentum is that
its rate of change is determined by the extemaIlyappliedforce-ee.Itis
>,
not the momentum of the electron in the Bloch state l'Pk which is given
by the rest mass of the electron times the group velocity, eqn (4.18) (see
Problem 13). However, it is the crystal momentum that enters in equations
such as the conservation of momentum and so it behaves as though it were
the true momentum of the electron in the crystal.
The distinction we have made between the true momentum and the crystaI
momentum of the electron can be made more forcefulIy if we say that the
rate of change of the crystal momentum is equaI to an effective mass, m·, of
the electron times the rate of change of the group velocity. The reason for
introducing the effective mass is that this is the mass that the electron in the
crystal will appear to have when it is subjected to the extemaIly applied
electric field. The interactions between the electron and the periodic potential
of the crystaI are thus alI subsumed into this effective mass. For a one-
dimensional crystal
k• dV h dk
m -= --. (4.22)
dt 2n dt
Since Vk = (2njh) dE(k)jdk the left hand side of this equation becomes
m* -=m
dVk * -2n d2 E(k) dk .
2
(4.23)
dt h dk dt
Equating (4.22) and (4.23) we get
(4.24)
Let us put in our one-dimensional band structure for an infinite linear chain
with periodic boundary conditions, eqn (3.19)
E(k) = CI. + 2p cos ka. (3.19)
We get
h2
m* = - 2 2 sec(ka), (4.25)
8n pa
which is plotted in Fig. 4.4. The prefactor - h2 j(8n 2 Pa 2 ) is positive because
p is negative. In the bottom half of the band, where Ikl ;S; nj2a, the effective
m*
-+--~~~~~~--~~--~~~k
: -1tla :-1t/2a o :, x/2a :7tla
:( ,
,
,
,
,
,
,
,
,
,,
,
,
,
,
,
,
, ,
Fig, 4,4 The effective m" plotted as a function of wave vector k for the
one-dimensional band E(k) = C( + 2P cos ka.
mass is positive but it becomes infinite at the infiection points + n/2a. In the
top half of the band the elfective mass is negative. What is going on
physically?
Consider the motion in real space of the electron under the action of the
electric field. Integrating eqn (4.21) for dk/dt we find that
2ne~
k=ko---t (4.26)
h '
where ko is the value of k at time t = O. The group velocity of the electron
as a function of time is therefore given by this equation, eqn (4.18) and
eqn (3.19)
Vk = - 4?a sin(ka _ 2n:~ t)a. (4.27)
This shows that the electron is moving backwards and forwards within the
crystal under the infiuence of the applied constant electric field ~. The origin
of this oscillatory behaviour may be traced to the group velocity formula in
k-space, eqn (4.18). For our one-dimensional band we have
4npa .
vk = - - - sm(ka). (4.28)
h
Consider what happens as the wave vector approaches the Brillouin zone
boundary at k = n/a. As k increases towards n/a it is decelerating until at
k = n/a it is stationary. As k exceeds n/a the group velocity becomes negative.
Thus the oscillatory behaviour is caused by the electron reversing its motion
at the Brillouin zone boundaries in k-space.
This is nothing more than Bragg reflection by the crystal lattice. The
condition Cor a particle to be Bragg reflected Crom a state with crystal
momentum hk/21t to a state with crystal momentum hk'/21t is that
k - k' = G, (4.29)
where G is a reciprocal lattice vector oC the crystal. Recall that the Bloch
Cunction 'l'k(r) is aplane wave e ik ·, times a Cunction, uk(r), which is periodic
in the crystal lattice. A Cunction which is periodic in the crystal lattice can
always be expanded in a Fourier series
(4.30)
The sum in this series is taken over all reciprocallattice vectors oC the crystal.
It Collows that a Bloch function is a mixture oC Fourier components and that
the coefficients, ck(G), oC the mixture change as k varies throughout the
Brillouin zone. When we treat the nearIy Cree electron approximation in
Chapter 7 we will see that as k approaches a Brillouin zone boundary
corresponding to the reciprocallattice vector G the Fourier component with
wave vector k - G increases, until at the zone boundary there is an equal
mixture oC components with wave vectors k and k - G, Corming standing
waves. That is why the group velocity at the zone boundaries is zero.
The picture we have is oC the electron being accelerated by the external
electric field so that its wave vector increases linearIy with time. As the wave
vector approaches the Brillouin zone boundary there is an increasing amount
oC Bragg reflection until at the zone boundary standing waves are set up and
the electron comes to a stop. At that point we can say that the electron has
been Bragg reflected beca use with increasing time the electron moves against
the electric field, signiCying that its wave vector has become k - G. There is
no discontinuity oC the particIe velocity in real space, as seen in eqn (4.27).
When the electron is Bragg reflected it slows down, reaches zero velocity
and is accelerated in the reverse direction.
When the electron is Bragg reflected it transCers its momentum into the
crystal lattice. Thus even though the total momentum oC the electron +
crystal system is increasing under the action oC the externally applied electric
field the momentum oC the electron oscillates. At the bottom oC the band the
electron accelerates 'normally' under the action oC the electric field and
almost all oC the impulse imparted by the electric field is taken up by an
increase in the crystal momentum oC the eIectron. However, as the wave
vector approaches the top oC the band more and more oC the electron
momentum is transCerred to the crystallattice through Bragg reflection. This
results in the eIectron slowing down and eventually moving against the
electric field, as though it had a negative mass. This explains the behaviour
oC the effective mass.
If the electron is oscillating backwards and forwards under the action oC
a DC electric fieId how can a current flow? The answer is that the electron's
wave vector does not change very much at all before it is scattered by a
lattice vibration, so that the wave vector never increases linearly with time
through the whole Brillouin zone. This strikes one as somewhat odd because
scattering is normally regarded as a source of electrical resistance, but the
reasoning outlined here indicates it is essential for a current to flow!
Equation (4.28) shows that, in the absence of an applied electric field, the
states at ±k have equal and opposite group velocities. Therefore, there is
no net current in the material in the absence of an electric field, as we might
expect. In order for a current to flow in the presence of an electric field the
balance between electrons occupying states with +k and -k wave vectors
must break down. Again, this is achieved by scattering of the electrons by
thermal vibrations and defects in the crystal, provided there are unoccupied
sta tes for the electrons to be scattered into. (Remember that each state may
hold only two electrons, with opposite spins.) If all states in the band are
filled it is not possible to alter the balance between occupied sta tes with ± k
wave vectors because there are no available unoccupied states to accept a
scattered electron. Electrons can pass through such a material only if
unoccupied sta tes in higher energy bands can be accessed by thermal
excitation. Therefore, such a material is an insulator at O K. Conversely, if
the band is partially occupied, scattering of the electrons into unoccupied
sta tes can take place, and the material will allow electrons to flow through
it so that it is a metal. This reinforces our earlier point that scattering is
essential to electronic conduction in normal (i.e. not superconducting)
metals.
Notice that the hopping integral P comes into the formulae for both the
group velocity and the efl'ective mass. The greater the absolute value of the
hopping integral the easier it is for the electron to hop from atom to atom
and hence the greater its group velocity and the smaller its efl'ective mass.
Finally,let us apply the equation of motion F = (h(2rc) dk(dt to an electron
in a magnetic field. The Lorentz force acting on the e1ectron is - ev x B,
where v is the group velocity. Thus,
h dk 2rce
- - = -evk xB = - - VkE(k) x B. (4.31)
2rc dt h
Therefore the electron moves on a constant energy surface in k-s pace under
the action of an applied magnetic field.
Holes
Unoccupied orbitals in a band are called holes. A hole acts in an applied
electric or magnetic field as though it were a particle with a positive charge.
It is important to realize that the hole is simply a shorthand way of describing
the behaviour of all the electrons in the partially filled bando When we ascribe
Holes 85
properties to the hole such as charge, mass, velocity, and energy what we
mean is that the totality oC all the electrons in the band responds to an
applied electric or magnetic field as though it were replaced by the holeo
In the previous section we argued that a Cully occupied band cannot
conduct electricity because there is no way oC destroying the balance between
the occupation oC states with equal and opposite wave vectors. This balance
is maintained after an electric field is applied because, although the wave
vectors evolve in time according to eqn (4.26), they all change by the same
amount. If J.(Ebottom to E F) is the current carried by electrons in states with
energies between the bottom oC the band and EF and J.(EF to Etop) is the
current carried by electrons in states between EF and the top oC the band then
J.(Ebottom to E F) = -J.(EF to E top ). (4.32)
Thus, the current that is carried by electrons occupying orbitals up to only
EF in a partially occupied band is precisely the same as would be obtained
by regarding the current carriers not as being the electrons in the occupied
orbitals but positively charged carriers (Le. holes) in the unoccupied orbitals.
The holes are positively charged beca use the current they carry has the
opposite sign (see eqn (4.32» to the current carried by the electrons. Note
that this is an eitherjor situation! We either regard the electrons as the
carriers or the holes: we cannot mix them and have sorne oC the current
carried by electrons and sorne by holes.
Under what circumstances would we be perverse enough to regard the
current carriers in a partially occupied band as being positively charged holes
in unoccupied orbitals, when there are plenty oC bona fide partic1es (i.e.
electrons) around to carry the current? The answer is when the band is
almost completely filled with electrons, because then the properties of
~ 1023 electrons cm - 3 in the solid are described by a much smaller number
of holes, typically ~ 1014 _10 18 holes cm -3 (i.e. 10- 9 to 10- 4 oC the electron
concentration). If this seems bizarre consider the following analogy with
which you may be familiar. When we discuss dilfusion in solids by a vacancy
mechanism we talk about the jump rates of the vacancy and the activation
energy for vacancy migration. This is simply a shorthand way oC describing
the atomic motions that take place when an atom jumps into a vacant site.
Similarly, the dilfusional flux is determined by the concentration of vacancies
(boles!) in the crystal structure. It would be more cumbersome and much
less transparent to set the dilfusion problem up in terms of the motion oC
all the atoms in the crystal, without regard to the location and concentration
of vacant sites, even though dilfusion occurs by atoms jumping from site to
site. In the same way when the band is nearly full it makes much more sense
to discuss the motion of the vacant electron orbitals, the holes, than the
motion of the much greater number of electrons.
So how do holes behave? We have already seen that they are assigned the
opposite charge of an electron. Their other properties are:
1. If an electron is missing in the state k. then the wave vector assigned to
the hole is k h = - k •. This is because if the band were filled the total wave
vector of all the electrons would be zero, because for every state at k there
is another state at -k. Thus if the band is filled except for a state at k.
then the total wave vector of all the occupied electron states is - k., and
this is the wave vector that is assigned to the hole because the hole is an
equivalent description of the behaviour of all the occupied electron states.
2. The energy of a hole Eh(k h) is the negative of the energy of the electron
missing at k. = - kh • The energy of the hole at k h is the energy required to
remove an electron at - k h • The nearer the e1ectron is to the bottom of the
band the more energy is required to remove the electron from the bando
3. The group velocity of the hole is equal to the group velocity of the missing
electron. This is because the slope of Eh(k h) is equal to the slope of
-E.( -k.), which is equal to the slope of E.(k.). See Fig. 4.5.
4. The effective mass of the hole is the negative of the effective mass of the
electron. As seen in Fig. 4.5 the curvature of the hole band d 2 Eh/dP has
the opposite sign of the curvature of the electron bando Near the top of
the valence band the effective mass of the e1ectrons is negative so the
effective mass of the holes is positive.
5. The equation of motion of a hole is
h dk
---
2n dt
h
= Fh = +/e/(E + "h x B). (4.33)
where Fh is the force acting on the hole due to the electric field E and the
magnetic field B. Tbus, the equation of motion of tbe hole is that of a
particle of positive charge e.
E
Slopes are \he same
--------T7.~_r-E~_r------·k
Fig. 4.5 Group velocity of the hole with wave vector k h is equal to the group
velocity of an electron missing from state k. = -kh . The slope of Eh (kh ) is equal
to the slope of E.{k.).
The Fermi surface 87
The Fermi surface

We have already diseussed the Caet that a filled band eannot conduct
electricity. In a partially filled band current ean ftow because there are
unoccupied sta tes available to receive scattered eleetrons and carry the
current. These sta tes have energies that are infinitesimally cIose in energy to
the energy oC the highest occupied states, which (at absolute zero) is the
Fermi energy. Thus, provided the density oC states at the Fermi energy is
not zero we ha ve a metal. The energy surface in k-s pace corresponding to
the Fermi energy is called the Fermi surfaee. If the band is filled there is no
Fermi surCace separating occupied and unoceupied sta tes in k-space. Thus
a metal may be defined as a material with a Fermi surface, with the
supplementary eondition that the density oC states is not zero at the Fermi
energy. Many oC the properties oC metals are a direct consequence oC the
presence of a Fermi surface.
Consider the Fermi surface for the square lattice that we considered at
the beginning of this ehapter. The band structure is given by eqn (4.11). If
the energy of the highest occupied energy levels is EF then the Fermi surface
is defined by
EF = ex + 2fJ(cos kxa + eos kya). (4.34)
This 'surfaee' is just a line in our two-dimensional example. In Fig. 4.6 we
(a) (b) (e)
(d)
Fig. 4.6 The Fermi surfaee (Iine) for a two·dimensional band E = IX +

2fJ(eos kxa + eos kya). (a) Small band filling where the oeeupied states (shaded)
oceupya eirele at the centre of the first Brillouin zone. (b) Half·filled band where
EF = IX. where the Fermi surfaee just touehes the Brillouin zone. (e) Almost filled
band where all but circular quadrants at the Brillouin zone eorners are oceupied.
(d) Same as (e) but in an extended zone seheme.
Fig. 4.7 The Fermi surface 01 copper as determined experimentally by Pippard.

Note that the Fermi surface touches the Brillouin zone at the hexagonal {1 1 1}
faces.
sketch the shape of the Fermi surface at three band fillings. At small band
filJings, where EF is dose to the bottom of the band at ex + 4fJ (remember
p < O), we may expand the cosines in eqn (4.34)
(4.35)
Thus, the Fermi surface is a cirde for small band fillings (see Fig. 4.6(a».
We shall see in Chapter 7 that this corresponds to the free electron
approximation.
At intermediate band fillings the cirde develops bulges towards the
Brillouin zone boundaries until when EF = ex, i.e. the band is half-filled, the
Fermi surface just touches the Brillouin zone boundaries at ± (n/a, O) and
±(O, n/a). This is shown in Fig. 4.6(b). Finally, in Fig. 4.6(c) we show the
Fermi surface for an almos! filled bando In the extended zone scheme,
Fig. 4.6(d), we see that the unoccupied states are confined to pockets in
k-space.
In Fig. 4.7 we show the Fermi surface of copper. It is seen that the
Fermi surface touches the Brillouin zone boundaries on the {1 1 1} faces.
The density of sta tes in two- and three-dimensional

crystals
The density of sta tes for a one-dimensional crystal was introduced in eqn
(3.54). The density of states in higher dimensions has the same meaning: the
Density of states in two- and three-dimensional crystals 89
density of sta tes D(E) is the rate at which the total nurnber of states is
increasing with energy at the energy E. Thus D(E) dE is the nurnber of states
in the energy interval E to E + dE. We can also define a local density of
states as in eqn (3.61), which is the density of states projected onto a
particular basis sta te.
Consider the square lattice again. Using eqn (4.13) we find that the density
of states in the two-dirnensional k-space is
2
D(k) = Nx Nya (4.36)
(2n)2
where we recognize Nx Ny a 2 as the (infinite) area of the two-dirnensional
square lattice. To find the density of states as a function of energy let us pick
an energy Eo. This corresponds to sorne curve in k-space where eqn (4.11)
is satisfied, as shown in Fig. 4.8. If we increase the energy to Eo + dE the
curve is displaced slightly outwards, by an arnount which is greater in those
parts of the curve where the gradient of the energy is srnallest. Let Idkl be
the rnagnitude of the normal vector displacernent at k of the line correspond-
ing to the energy Eo. Then Idkl = dE/IVkEI. The area in k-space enclosed
between the two energy surfaces is thus
area =
iE=Eo
dI
IVkE(k)1
,
where the integral is a line integral taken around the curve E(k) = Eo. To
(4.37)
get the density of states D(Eo) we rnultiply this area in k-space by the nurnber
of states per unit area of k-space given by eqn (4.36)
D(E o) = Nx Nya2
(2n)
2
i E=Eo
dI
IVkE(k)1
• (4.38)
E(k) = Eo
/\,
E(k) = Eo + dE '. '
,
Fig.4.8 The energy surface (solid line) E(k) = Eo tor the two-dimensional band
E= IX + 2P(coskxa + coskya). Thebroken lineshowstheenergysurfaceE(k) =
Eo + dE. Notice that the separation between the two energy surfaces increases
as IVkEI decreases.
This is the total density of states for the whole square lattice, and that is
why it is infinite. To get the density of sta tes per lattice site we just divide
by NxNy • Using eqn (4.11) we find that
IVkE(k)1 = 2Pa(sin 2(k xa) + sin2(k ya»1!2. (4.39)
We are not able to evaluate the integral in eqn (4.38) analytically, but let us
consider two important cases. The first is a small band filling where we saw
in eqn (4.35) that the band structure reduces to a free electron formo Taking
the limit that kx and ky tend to zero in eqn (4.39) we find
(4.40)
where k = Ikl. The integral in eqn (4.38) is now trivial because it reduces to
an integral around a circ1e of radius k and we find the density of states, d(E),
per atom beco mes
al 1 1
(4.41)
d(E) = (21t)2 21Pla2k 21tk = 41tIPI·
Therefore the density tends to a finite value of 1/41tIPI at small band fillings.
The same result is obtained for a nearly full bando (Why?)
Consider the density of sta tes in the middle of the band where E = cx. This
is also the energy at which the Fermi surface just touches the Brillouin zone
boundaries, as shown in Fig. 4.6(b). The line in k-space corresponding to
this energy is
cos kxa + cos kya = O. (4.42)
Using this relation we deduce that IVkE(k)1 is given by
IVkE(k)1 = 2(2 1!2)IPlalsin kxal. (4.43)
Putting this into the integral in eqn (4.38) we find that the density of states
per atom is given by
a 1t/a
- 2(21!2)1t2IPlln(tan(kx/2» o
= infinity! (4.44)
Thus the density of states is infinite in the middle of the bando The density
of states for the square lattice is shown in Fig. 4.9, and it is seen that our
two Iimiting cases agree with this curve.
Density of states in two- and three-dimensional crystals 91
d(E)
00 00
--------- ,--¡-----
41t1~1
--~~----~~-------L~-.E
a +413 a a -413
Fig. 4.9 The density of states for the two-dimensional square lattice s-band
model. Notice that the density of states tends to a constant finite value at the
band edges where the free electro n approximation applies.
When we consider the cubic lattice the expression Cor the density oC states
is much the same. The density oC states in k-s pace is
3
D(k) = N"NyN.a (4.45)
(2lt)3
where we recognise N"NyN.a 3 as the volume of the infinite crystal. The

density oC states per atom in k-space is just (aj2lt)3. Following the same
argument as in two dimensions the density of sta tes per atom as a function
oC energy is given by the Collowing surface integral
d( Eo ) -- ( -a
2lt
)3 f E=Eo
dS(k)
IVkE(k)l·
(4.46)
The integral is taken over the surCace in k-space where the energy equals Eo.
It is even more difficult! We shall consider analytically only the limit of small
band fillings. U sing eqn (4.15) Cor the band structure for the simple cubic
lattice s-band model we have, in the limit oC small wave vectors,
(4.47)
Therefore,
(4.48)
From eqn (4.47) we see that the surface ofintegration in eqn (4.46) reduces
to a sphere in k-space, Cor small band fillings, oC radius
k
= (E - IX - 6P)1/2 (4.49)
-pa 2 .
Therefore, the density oC states, d(E), per atom, for small band fillings
d(E)
----~~--L---~~--.E
a+613 a a-613
Fig. 4.10 The density ot states tor the simple cubic s-band modelo
becomes
= (2 nY 27a~2k = 4n21~13/2 (E -
a
d(E) (tI + 6P»1/2. (4.50)
The important point here is that at the bottom oC the band the density oC
states varies as the square root oC E. This is true Cor all crystal structures in
three dimensions. The same applies to the density oC sta tes near the top oC
the bando In Fig. 4.10 we show the density of states over the whole bando It
is seen that it does not diverge to infinity anywhere, but there are energies
where the slope changes discontinuously. These are called van Hove
singularities.
The density matrix, bond order. and bond energy

Let us consider the charge density p(r) in the simple cubic crystal in the case
where the band is only slightly filled. We make the approximation oC eqn
(4.47) Cor the band structure E(k), in which the Ferroi surface is spherical.
We shall derive the density matrix for the simple cubic crystal, which will
enable us to see the Corro of the bond orders. Thus, we shaIl gain a real space
picture oC bonding in the crystal which is analogous to the real space analysis
oC Chapter 3 Cor the one-dimensional crystal. However, we emphasize that
the Cunctional Corros we derive are valid only in the limit oC smaIl band fillings.
The charge density is given by
p(r) = - 2e L l'I'k(r)1 2 (4.51)
k occupied
where the Cactor oC 2 accounts Cor the Cact that each state may be occupied
by two electrons oC opposite spin, and the sum is taken over aIl occupied
sta tes. This expression is exact regardless oC the band filling, but in general
the evaluation ofthe sum can be done only numericaIly (i.e. with a computer).
However, in the limit oC smaIl band fiIlings the sum can be done analyticaIly,
and that is what we now considero
The density matrix. bond arder. and bond energy 93
The first step is to replace the sum by an integral over k-space
L:
k occupied
-+ N3 (21ta)3
- i Ikl :<;kF
die (4.52)
where k F is the Fermi wave vector and N 3 = NxNyN. is the number of atoms
in the crystal.
The next step is to write down an expression for l'I'k(r)1 2 using eqn (4.14).
We introduce a small change of notation and denote a lattice site by R,.
where the 'm' labels the lattice site. Thus
where <rIR,.) is the atomic orbital at site R,.. Therefore l'I'k(r)12 is given by
Putting this together the charge density becomes

N3 N3
p(r) = -e L: L: p ... <R.lr)<rIR,.)
m= 1,.=1
(4.55)
where P... are elements of the density matrix. As in the one-dimensional case
in eqn (3.45) the density matrix elements are the expansion coefficients of
the charge density in the atomic orbital basis functions. The density matrix
elements are given by
p,.. = 2(~)3
21t
rJ
Ikl :<; lF
eik·(Rm-Rn ) die. (4.56)
This is very similar in form to the one-dimensional integral that we had to

evaluate for the density matrix element in the one-dimensional crystal but
the present integral is a volume integral over a sphere in k-space. It is easily
evaluated using spherical polar coordinates in which the volume element die
becomes k 2 sin (J dk d(J dtjJ. The most convenient choice for the z-axis is along
the bond R,. - R., in which case k·(R.. - R.) becomes klR,. - R.I cos (J.
Thus,
( )3ilF
P... = 2 ~ dk J.0 d(J f2. dtjJk 2 sin (J eillRm - Rnl COI 9. (4.57)
21t l=O 9=0 4>=0
The integrand does not depend on tjJ and so the tjJ integration is just 21t. The
integration over 9 is easy once it is recognized that sin 9 d9 is -d(cos 9)
r dk J
3 tF
P... = 2 a -1 -d(cos 9)k 2 elkIRm-R.J co.6
41t Jt=o
2
cos6=1
= 2 a r dk JI
3 kF
d(cos 9)k 2eitIRm-R.1 cos6
2
41t Jt=o c.. 6=-1
a3 J,kF [eltIRm-R.1 cos 6Jeo.6=1

= 2 -2 dkk 2
41t t=o ikIR .. -R.1 ... 6=-1
= a: r tF
dk k sin(kIR.. - R.!)
1t Jt=o IR.. - R.I
= (akF)3 (sin(kFIR.. - R.I) - (kFIR.. - R.!) cos(kFIR.. - R.!)) (4.58)
2
1t (kFIR.. - R.!)3 .
The integration over k was evaluated by parts. This function is shown in

Fig. 4.11. Note that when IR.. - R.I = O we have
(ak F )3
P.... = 31t 2 (4.59)
which follows directly from eqn (4.56) by setting the integrand to 1. This is
the number of electrons per atom. But the interesting result is the expression
for the bond order P... (i.e. when m # n). For a given Fermi wave vector it is
seen that the bond order oscillates and decays with increasing separation,
IR.. - R.I, of the two atoms. Note. that the bond order is not zero
even when the Hamiltonian matrix element connecting the two atoms is
zero. The Hamiltonian matrix couples only nearest neighbours but the bond
order extends beyond first neighbours. The first zero in P... occurs when
IR.. - R.lkF ~ 4.493 and the first minimum is approximately 8 per cent of
p(x)
x =4.493
sinx-xcosx
p(x) =
x3
Fig.4.11 The function p(x) = (sin x - x cos x)/xl.
The density matrix, bond order, and bond energy 95
the value of Pmm. Although our model of the solid is very simple many of
the features we have found here survive in more sophisticated modeis.
The bond energy was defined in two equivalent ways in eqn (3.71) and
eqn (3.76). Let us demonstrate their equivalence for the present model.
Consider first the bond energy per atom as the sum of energies of the bonds
between an atom and its six neighbours, eqn (3.76). For this we need the
bond order between neighbouring atoms, which we call POI. Using eqn (4.58)
we can write it down straight away
(4.60)
The bond energy per atom is thus six times this bond order times the
Hamiltonian matrix element coupling nearest neighbours
(4.61)
We should keep in mind that this result is valid only for small band fillings.
This means that it is valid only when kFa « 1.
Now let us evaluate the bond energy per atom using the density of states
per atom, eqn (4.50), which was calculated in the same approximation of a
small band filling. The bond energy is given in eqn (3.71) in terms of the
density of sta tes. In this representation the bond energy is given by
f
EF (E - IX) 1/2
E bond =2 , 2 3/2 (E - (IX + 6P)) dE. (4.62)
a+ip 471: IPI
The Fermi energy, EF , is related to the Fermi wave vector, kF , through eqn
(4.47). This integral is straightforward enough provided it is remembered
that P is negative! Making the change of variable u = E - (IX + 6P) we get
1
E bond = 2 3/2 (HEF - (IX + 6P)]5/2 - 4IPI[EF - (IX + 6P)]3/2). (4.63)
271: IPI
Using eqn (4.47) we have
(4.64)
Therefore,
(4.65)
Is this the same result as eqn (4.61)? On the face of it, it does not loo k like
it is! But remember that these results are valid only in the limit that kFa « 1.
With this in mind let us expand the bond energy in eqn (4.61) to fifth order
in kFa and see if we get the same result as eqn (4.65). If we do then that is
all we are entitled to expect with the approximations we have made
E bond = 6~ (sin(kFa) - (kFa) cos(kFa» (4.61)

¡¡
~ -61PI (k a) _ (k F a)3 + (k F a)3 _ (k a)(l _ (k F a)2 + (k F a)4))

¡¡2 F 6 120 F 2 24
= -Ifl (2(ak F )3 - !(akF )3)

¡¡
which is the same as eqn (4.65).

It we want to go beyond the approximation of small band fillings we have
to evaluate the integration over the occupied states in the Brillouin zone in
eqn (4.56). This is a nontrivial task and is most easily done on a computer.
In Fig. 4.12 we sketch the bond order between neighbouring atoms, POI' as
a function of the number of electrons per atom, which can be a maximum
oC two in an s bando We observe that the bond order is a maximum when
the band is half CulJ. That is because only bonding states are occupied when
the band is less than half filled. At higher band fillings the bond order
decreases because antibonding states are occupied. When the band is full
there are as many bonding states occupied as antibonding states, and the
bond order is zero. This is consistent with our statement on p. 57 that the
bond order is proportionaJ to the difference between the number oC electrons
in bonding and antibonding states in the bond.
FinalJy we emphasize an important distinction between the total charge
density p(r) and the bond order POI' We have seen, in eqn (4.55), that the
bond order is one term in the expansion of the charge density in the atomic
orbital basis functions. It is very tempting to think that a larger charge
POI
- L - - - - - - I ' -=-- - - - ' " - - _ Number of

o 2 electrons
peralOm
Fig. 4.12 The nearest neighbour bond order in the simple cubic lattice s-band
model plotted as a function of the number of electrons per atom. The bond order
increases to a maximum when the number of electrons per atom is one and all
bonding states are occupied. but all antibonding states are empty.
Moments theorem applied to 2- and 3-dimensional crystals 97
density in the solid implies that all the tenns in the expansion of the
charge density are simply increased in proportion to the increase in the
charge density. For a given lattice constant an increase in the charge density
implies that the number of electrons per atom is increased. Figure 4.12 shows
that the bond order can actually decrease, with increasing electron to atom
ratio, for more than half-filled bands. In that case the increased electron to
atom ratio results in a smaller bond charge density contribution to the total
charge density. In effect the charge density distribution becomes more
atomic-like in the second half of the band, until, when the band is full, the
charge density becomes a superposition of atomic charge densities with no
bonding charge density at all. The point to be grasped here is that the bond
order, or bond energy, does not necessarily increase when the charge density
is increased. The widely prevalent view that increasing the charge density in
a metal necessarily results in stronger bonds (i.e. an increase in absolute
tenns of the bond energy) is plainly wrong. On the contrary, it will weaken
the bond if it results in more antibonding states being occupied, and it will
result in a smaller bond charge density contribution to the total charge
density. On the other hand, since the charge density in a bond is proportional
to the bond order, a large bond charge density does indicate a large bond
order and therefore a strong bond.
The moments theorem applied to two- and

three-dimensional crystals
The moments theorem was introduced on p. 66. The theorem applies in two
and three dimensionsjust as well as in one dimensiono The only complicatioD
is that with the higher dimensionality the number of closed paths iDcreases
rapidly. On the other hand the theorem is arguably even more useful in two-
and three-dimensional solids because of the greater difficulty in computing
the density of sta tes by integrations over the Brillouin zone.
Consider the simple cubic lattice s-band model again. The second moment
of the local density of sta tes for atom O is the sum of all paths of length two
hops, starting and ending on atom O. Since only the six neighbours are
coupled to atom Othrough the Hamiltonian the second moment is 6p2. The
root mean square width of the band is thus 6 1/2 1PI. There are no closed paths
comprising an odd number of hops and therefore the density of states is even
about the centre of the band, as seen in Fig. 4.10. In Fig. 4.13 we show the
closed paths of length four hops, of which there are 54 in the simple cubic
lattice. Thus the fourth moment is 54p4 , and the nonnalized fourth moment,
J.I.&4)/(J.I.&2)2, is l The parameter s, defined by eqn (3.89), is t and, therefore,
we expect the density of states to show bimodal behaviour. Looking at Fig.
4.10 we see that it does show bimodal behaviour, but the minimum between
the two peaks is very shallow.
=24~4
Fig. 4.13 To illustrate the paths contributing to JlÓ4) in the simple cubic crystal.
Suppose we introduce a substitutional impurity at site i. We can model

tbe impurity by giving it a different on-site Hamiltonian matrix element,
(iIHli), from the other atoms in the simple cubic crystal, which was IX. Let
(iIHli) = e, and we assume the hopping integrals between the impurity and
the host atoms remain equal to p. Crudely speaking this model assumes that
the impurity has the same size as the host atoms, but a different electro-
negativity. What can we say about the local density of states, d¡(E), at the
impurity site? Bloch's theorem is no longer applicable to the solid because
the presence of the single impurity has destroyed the translational symmetry
of the crystal! It is therefore much harder to determine the eigenstates
of the crystal + impurity system. This is a cIassic problem (known as the
Slater-Koster problem) tbat can be solved exactly using more sophisticated
tecbniques (Green's functions) than we shall develop in this book. But the
moments theorem will give us a good deal of qualitative insight into the form
of tbe local density of states d¡(E), witbout the bassle of figuring out the
eigenstates.
The nth moment of the local density of states d¡(E) is given by
¡.t!") = (il(H - e)"li). (4.66)
As usual ¡.tIO) = 1 owing to tbe normalization condition <i Ii) = 1 and
¡.tll) = Oindicating tbat the centre of gravity of d¡(E) is e. Tbe second moment
is still given by the sum oC closed patbs oC length 2 hops and is again 6p2.
Thus, so far the moments are tbe same as for the host atoms in the absence
of the impurity. But higber moments differ. The third moment has six
contributions arising from a hop out to a neighbour, an on-site term
(i IH - elj) = IX - e, and a hop back: ¡.tf3) = 6P2( IX - e). Thus the local
density oC states, d¡(E), is not symmetric about tbe centre of gravity e, but
skewed positively or negatively depending on tbe sign oC (IX - e). The fourth
moment comprises the same cIosed paths of Cour bops as in the pure crystal,
giving 54p4, plus six new paths arising from a hop out to a neighbour, an
Moments theorem applied to 2- and 3-dimensional crystals 99
on-site term (jl(H - e)lj) <i1(H - e)lj) = (O! - 8)2, and a hop back to atom
i giving 6P2(0! - e)2. Thus ~4) = 54p4 + 6P2(0! - e)2. Thus the normalized
fourth moment is larger for the impurity site than for the host atoms,
indicatingthat d,(E) is more unimodal. More precisely, the local density of
states has a unimodal form when s ~ 1, that is when lO! - el ~ 31/2 1PI. Using
(8) a-E>O -a-+'":6":;"p--....E-.J..a----a~-6:-:p:-+ E
d(E)
(b) a-E<O --~~-----L~----~~·E

a+6p a a-6p
d(E)
(c) a-E« O
-a~+~6~P-----aL-----~E-a~-~6P~--E~L~E
Fig. 4.14 Schematic illustrations of the local density of states at a substitutional

impurity site, d¡(E), compared with the local density of states at a host simple
asdeduced fromthefirstfourmoments. (a) IX - e > 0, (b) IX - e < 0, (e) IX - e«

leads to a localized state at EL aboye the top of the bando
°
cubic lattice site, do(E), as a function of the difference of on-site energies IX - e,
this infonnation we show d¡(E) for the two cases (O( - 6) > O and (O( - 6) < O
in Fig. 4.14 (p. 99). We have drawn the irnpurity density of states with the
sarne band edges as the host density of states. The full so!ution to the prob!ern
indicates that if 1(0( - 6)1 exceeds sorne critica! value then states can be split
offfrorn the band as shown in Fig. 4.14(c) at E = EL. However, even in this
case the rnornents we have ca!culated rernain exacto
5
Band gaps: origins and
consequences
Introduction
Our discussion of the formation of energy bands in Chapters 3 and 4 was
based on a simple s-band model. We associated one s state with each atom
and we found just one band of states. Of course, the s-band model is a gross
simplification because an atom is associated with more than just one s sta te.
Depending on the value of the principal quantum number there may be
s, p, d, ... states within each shell of the atom. At higher energies there is a
continuum of free electron energy states, which far from the atom are
represented by plane waves. We are concemed only with the valence electron
states of the atom because the core states are too strongly bound to the
atom for significant tunnelling to occur to neighbouring atoms. In the solid
sta te, under normal pressures, core states remain bound to the core and
therefore they do not give rise to energy bands. In this chapter we shall find
that there is one band for each valence state in a unit cell of the crystal.
Thus if there are Na atoms in the unit cell and N. valence states associated
with each atom then there are N.N. bands. The ranges of energy over which
the bands lie may or may not overlap. Band gaps are ranges of energy in
which no band lies. Band gaps are not only important for the electrical and
optical properties they impart to the material but, as we shall see in
Chapter 10, they can also stabilize a crystal in a particular structure.
Figure 5.1 is a schematic illustration of the changes in the e1ectron energy
levels of an infinite set of atoms in the crystal as the lattice parameter of the
crystal is reduced from 1 m to 1 Á. We have depictedjust two valence states
for each atom. When the atoms are separated by 1 m the two valence sta tes
of each atom do not interact with sta tes on other atoms (i.e. the probability
of electron tunnelling between the atoms is negligible) and thus there are
two energy levels that are infinitely degenerate. When atoms come within
the tunnelling range of each other (of order 10 Á) there is a finite probability
of an electron hopping from one atom to the next and we see the onset of
band formation. The bandwidth increases in proportion to the increase of
the tunnelling probability, which is determined by the hopping integral.
Eventually, at sorne critical density, the energy bands start to overlap and
increasing the density further leads to wider bands and greater overlap of
102 Band gaps: origins and consequences
Energy
gap
Decreasing
atomic seperation
Fig.6.1 The two energy levels on each atom of an infinite crystal broaden into
bands as the lattice parameter is reduced from a very large value, where electron
hopping does not occur, to the value, 80' pertaining at equilibrium in the crystal.
At 80 the energy bands are separated by a gap.
the bands. However, iC the equilibrium density oC the material is below the
critical density then there is an energy gap between the bands.
The two atomic states on the left oC Fig. 5.1 could have two quite distinct
origins. The first possibility is that they are two atomic states on the same
atom. For example they could reCer to a 18 state and a 2s state on the same
atom. The second possibility is that they reCer to states on two distinct atoms
in the unit cell oC the crystal. For example they could reCer to a state on an
A atom and a state on a B atom in an AB alIoy crystal. Both possibilities
are consistent with the view that band gaps have an atomic or chemical
origino As we shall see in Chapter 12, this picture enables us to understand
the existence oC band gaps in amorphous materials as well as crystalline
materials. On the other hand, the picture we shall describe in Chapter 7 oC
band gaps arising Crom Bragg reflection at Brillouin zone boundaries is not
applicable to amorphous silicon Cor example. In this chapter, thereCore, we
shall continue to develop the picture oC the electronic states oC solids as sta tes
oC giant molecules, but we shall allow more than one atomic state in the unit
cell. This will inevitably complicate the mathematical analysis, but only in
the sense that there will be more algebra. The key ideas have already been
introduced.
A possibly misleading Ceature oC Fig. 5.1 is the Cact that the two bands oC
states are drawn emanating Crom two atomic states on the leCt. We might
be tempted to think that the sta tes in the upper band are Cormed Crom linear
combinations oC the upper atomic state on atoms throughout the crystal.
This is wrong. We shall see in the next section that states in the bands are
mixtures oC both atomic states. The giant molecular states are Bloch
Cunctions not oC pure atomic states but oC linear combinations oC atomic
Infinite linear chain with two s states per ato m 103
states called hybrids. This mixing together of atomic states is called
hybridization, which is discussed in Chapter 10 in connection with crystal
structures. In the next section we shall iJIustrate the concept of hybridization
by retuming to the infinite linear chain, but with two s states on each
atom.
Infinite linear chain with two s states per atom

Consider an infinite linear chain of atoms with atomic spacing 'a' and
periodic boundary conditions applied to the ends of the chain. Each atom
is associated with two s states, 11) and 12). The Hamiltonian matrix elements
are all zero except for hopping integrals between neighbouring atoms and
on-site matrix elements. Let E, and E2 be the on-site matrix elements <lIH 11)
and (2IHI2) and we assume E2 > E" Let 1m, 1) and 1m, 2) denote the states
11) and 12) on atom m. The nonzero hopping integrals are
«m, 11Hlm ± 1, 1) = P,)

«m, 11Hlm ± 1,2) = P12)
1 . (5.1)
«m, 21Hlm ± 1,2) = P2)
Thus we have not assumed that the hopping integrals between different
atomic s sta tes on neighbouring atoms are the same. Our model has five
parameters: E" E2, P" P,2' and P2'
Since there is translational symmetry we may apply Bloch's theorem.
However, since there are two sta tes at each atomic site the Bloch state will
be a linear combination of the atomic states
This is the same as eqn (3.37) except that the single atomic state 1m) at site
m has been replaced by a hybrid state cin)(k)lm, 1) + c~n)(k)lm, 2), i.e. a linear
combination ofthe atomic states on atom m. To ensure that this hybrid state
is normalized we require
(5.3)
The superscript (n) anticipates the fact that at each k there wiIJ be two
eigenstates which we shalllabel with n = 1 and n = 2. The label 'n' is called
the band indexo Thus the quantum numbers for the eigenstates of the system
are the wave vector k as before and the band index n. The two eigenstates
at each k wiIJ have different energies and thus there wiJI be two bands. As
in eqn (3.37), the sum in eqn (5.2) is taken over all atomic sites m and the
factor of 1/N l /2 is to normalize the Bloch state, where N is the (infinite)
number of atomic sites of the chain.
The Schrodinger equation is
HI'I'~') = E~')I'I'~') (5.4)
and inserting eqn (5.2) for l'I't) we get
¿ eik"'(c~')(k)Hlm, 1) + c~')(k)Hlm, 2»
m
= E~') ¿ eik"'(c~n)(k)lm, 1) + c~n)(k)lm, 2». (5.5)

m
Projecting this equation onto the state 10, 1) by multiplying from the left by
<0,11 we get
(E¡ + 2p¡ cos(ka) - E~n)c~n) + 2P12 cos(ka)cr) = 0, (5.6)
and similarly, projecting eqn (5.5) onto 10,2) by multiplying from the left
by <0, 21 we get
2P¡2 cos(ka)c\n) + (E 2 + 2P2 cos(ka) - E~n)cr) = O. (5.7)
Equations (5.6) and (5.7) are the secular equations. There are two equations
beca use there are now two atomic basis states in the unit cell. For nontrivial
solutions we require the secular determinant to be zero
(E¡ + 2p¡ cos(ka) - E~n) 2P12 cos(ka)
(5.8)
2p 12 cos(ka)
Expanding the determinant we get a quadratic equation with two roots
2
E1/J = 9¡(k) ; 9ik) + (( 9¡(k) ; 92(k)Y + 4M2 COS 2(ka)r (5.9)
and
where
9¡(k) = E¡ + 2p¡ cos(ka) and 92(k) = E2 + 2P2 cos(ka). (5.11)
These two solutions are plotted as a function of k in Fig. 5.2. We recognize
9¡(k) as the band structure for the infinite linear chain if there were only
states 11) associated with the atoms. Similarly, 92(k) is the band structure
for the infinite linear chain if there were only states 12) associated with the
atoms. The band structures 9¡(k) and 92(k) are also shown in Fig. 5.2. The
lower band, E1¡), lies below both 9¡(k) and 92(k) while the upper band, E~2)
lies aboye 9¡(k) and 92(k), except at k = ±n/2a where EP) = 9¡(k) and
Ef) = 92(k).
As an aside we observe that the effect of coupling states 11> and 12) on
adjacent atoms together, through P12' is to increase the energy difference
between the two bands 9¡(k) and 92(k). This effect diminishes as the energy
Infinite linear chain with two s states per atom 105
E(k)
,
~ .... ,
-+------r-----~----_.------+__k
-1C/a -1C/2a 1C/2a 1C/a
Fig. 5.2 The energy bands, Eí ) and EI ), given byeQn (5.9) and eQn (5.10),
1 2
for an infinite ring in which each atom is associated with two s states with on-site
energies e1 and e2' The curves labelled g1 (k) and g2(k) are the energy bands in
the absence of hopping, P12' between states 11) and 12) on adjacent atoms (see
eqn (5.11». Notice that Eí1) is less than g1 (k) and Eí2 ) is greater than
g2(k) at all wave vectors except k = ±n/2a.
difference IgI(k) - g2(k)1 increases and as IP121 decreases. Extending the

atomic basis set to inelude much higher energy atomic states that are only
weakly coupled to the existing basis set does not significantly alter the
calculated band structure. This is the justification for using a limited atomic
basis set to calculate band structures. The point is that one needs to inelude
only those atomic states with energies that lie in the range where we want
the energy bands. Inelusion of much higher energy atomic states, for example
with higher angular momenta or higher principal quantum numbers, leads
to only small refinements of the calculated bands.
At any particular value of k there is obviously an energy gap between EL I )
and EL2 J, but this is not what we mean by an 'energy gap'. When we speak
of an energy gap we mean that there is a range of energies where there are
no eigenvalues for all wave vectors. Thus there is a gap in the density of
states as a function of energy. This is achieved in Fig. 5.2 if the maximum
of the lower band is below the minimum of the upper band, which is attained
when the parameters 8h 82' PI' P2' and Pl2 satisfy a simple inequality. This
is illustrated in Fig. 5.3 where the corresponding density of sta tes is also
shown on the right of the band structure.
The energy gap shown in Figs 52 and 5.3 is described as indirect because
the maximum in the lower band occurs at a different wave vector from the
minimum in the upper bando An electron making a transition between these
E E
_----.00
-------------;- -------------f
---__-00
&;rg;g~p
---- -------- ---------~
_ _ _ _ _ _ _ _ _ _ _ _ _ 1_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ .00
~------'------~.. k '-----------<.d(E)
-1tla o '/tIa
Fig. 5.3 On the left we see the energy bands El') and El2 ) shown in Fig. 5.2.
On the right the corresponding density of states, d(E), is plotted as a function
of energy. There is an indirect energy gap between the maximum of El') and the
minimum of Ef).
two states has to undergo a change ofwave vector, and that means a phonon
(Le. a crystal lattice vibration) has to be involved in order to conserve
momentum. The significance of involving a phonon is that the probability
of the transition occurring is very significantly reduced, which has consider-
able implications for the optical as well as the electrical properties of the
material. On the other hand, if the wave vectors at the maximum in the
lower band and the minimum in the upper band coincide the gap is called
a direct gap and a transition between these two states does not require the
participation of a phonon.
So, band gaps may arise from extending the basis set on each atom to
include higher energy atomic states.
Energy gap in a binary AB alloy linear chain crystal

On p. 72 and in Problem 15 we discussed an infinite linear chain of
alternating A and B atoms. In that case each atom was associated with only
one state, and two bands of states were found, one bonding and one
antibonding. The energy gap is the difference in energy between the
maximum of the bonding sta tes and the minimum of the antibonding states,
which is just BA - Bu' In this case the band gap is a consequence of the
presence of two distinct atoms in the unit cell. In Problem 20 you will see
that a very similar result holds for an s-band model of a binary AB alloy
crystal with the NaCl structure.
Peierls distortions 107
Peierls distortions
In one-dimensional crystals (linear chains) band gaps are always introduced
by periodic distortions of the chain. Such distortions are known as Peierls
distortions.
Consider an infinite ring of atoms with spacing a, in which the hopping
integral is P and the on-site energy is zero. The band structure is
E(k) = 2p cos ka. Suppose we now move every second atom so that there
are now two hopping integrals Pi and pz in the unit cell, as shown in Fig. 5.4.
The period of the chain is doubled and therefore the Brillouin zone now
extends from -n/2a to + n/2a. Let 1m, 1) and 1m, 2) denote the atomic states
on atoms 1 and 2 in the mth unít cell. The eigenstates of the chain are now
expressed, using Bloch's theorem, as follows
\TI(n) _
Tk - L:'" e;kz"'(c~')(k)lm, 1) + c~)(k)lm, 2» (5.12)
m=-oo
where c~n)(k) e ikZ ... and c~)(k) e;k2... are the coefficients of atomic states
1m, 1) and 1m, 2). Inserting this eigenstate into the Schrodinger equation,
H'I'~n) = Eln)'I'ln), we obtain the following secular equations
(Pi e-;kZa + Pz)c~)(k) = Eln)c~n)(k)}

(5.13)
(PZ + Pi eU.Za)c~n)(k) = Eln)c~n)(k)
which lead to the following energy bands
Eln) = ±«Pi - Pz)Z + 4PiPZ cOS
Z
ka)l I Z. (5.14)
These bands are sketched in Fig. 5.5, and it is seen that a gap is opened up
at the Brillouin zone boundaries k = ±n/2a of magnitude 21Pi - Pzl.
If the number of electrons per atom is one the energy band of the original
undistorted chain is half-filled, and the Fermi wave vector is kF = ± n/2a.
After the distortion the lower band shown in Fig. 5.5 is filled and the upper
band is empty. This has two consequences. The first is that the sum of the
energies of the occupied electronic states is lower in the distorted chain
because the gap depresses the energy of states near kF • The second is that
the linear chain is transformed from a metal to an insulator.
p p p p p p p
(a)
Fig. 5.4 (a) A perfect infinite ring in which all nearest neighbour bond lengths
and hopping integrals are the same. (b) A distorted infinite ring in which every
second atom is moved to the right so that there are now two hopping integrals
Pi and PZ. and the period of the ring is doubled.
E(k)
,,,
,
: Energy gap = 21~1 - ~21
_x~n~a-------------+------------~~------~k
Fig. 5.5 The energy bands for the distorted infinite ring shown in Fig. 5.4(b).
Notice the energy gap of magnitude 21P, - P21 at the Brillouin zona boundaries
k = ±x/2a.
Although the sum of the occupied electronic eigenvalues is lowered there

is an energy cost associated with the distortion of the chain because there
is now an elastic strain energy. The strain energy varies quadratica!ly with
the magnitude of the distortion of the bond lengths. On the other hand,
IP 1 - P21 varies linearly with the distortion of the bond length, at least for
small bond distortions. The energy lowering of the occupied states is
proportional to IPi - P21, at least near the Brillouin zone boundaries. Thus
the electronic energy lowering is linear in the distortion of the bond lengths,
and therefore there is always a net lowering of the total energy by distorting
the bonds in this way for a half-filled bando
If Ihe band of the original undistorted chain were only one third full then
the same argument shows that the total energy of the system would be
lowered by a periodic distortion with a wavelength of 3a. This leads to a
Brillouin zone extending from k = -n/3a to k = n/3a, and there are now
three bands, with gaps between them at the Brillouin zone boundaries, as
shown in Fig. 5.6.
Generalizing this result we see thal for a Fermi wave vector, kF , in the
undistorted chain the total energy of the system is lowered by a periodic
distortion of wavelength n/kF' If the wavelength, n/k F, is an integer multiple
(say n) ofthe atomic spacing a then the original single band ofthe undistorted
chain is split into n sub-bands with gaps between them at the Brillouin zone
boundaries at k = ± k F • The distortion is then said to be cornmensurate with
the atomic spacing, a. On the other hand, if the wavelength, n/k F , is an
irrationa! multiple of the atomic spacing a then the original single band of
the undistorted chain is split into an infinite number of levels. The distortion
is then said to be incommensurate with the atomic spacing, a, and the
Brillouin zone collapses to the point k = O.
In conclusion, there is a!ways a gain in energy if an idea! infinite ring of
Metals. insulators. and the metallic bond 109
----~----+-----~---.k
-1t/3a 1t/3a
Fig. 5.6 The energy bands for a distorted infinite ring in which the wavelength
of the distortion is three atomic spacings. Notice that there are now three bands
in the first Brillouin zone extending from k = -1t/3a to 1t/3a. and that there are
energy gaps between them.
atoms undergoes a distortion with a wavelength n/kF' The ring is then con-
verted from a one-dimensional metal to an insulator. Thus, one-dimensional
metals are not stable energetically. In three dimensions the wave vector of
the periodic distortion lies at a point on the Fenni surCace, and only those
states near this point will be split by the distortion, with the rest of the Fenni
surCace remaining unaffected. Whether a sufficient energy lowering can
be achieved to make this distortion favourable depends on how Ilat the Fermi
surCace is near this wave vector.
Metals. insulators. and the metallic bond

On p. 84 we notedthat a fuIl band cannot conduct electricity. There are no
states available in the band to enable the electron gas to acquire a net
momentum because a11 states are fully occupied and the exclusion principie
prevents us from putting any more electrons into states a1ready occupied
with two electrons. Thus, under the action of the applied field nothing
changes. There are always unoccupied higher energy bands in the crystal,
but iftheir energies are separated by a finite amount (i.e. the band gap) from
the highest occupied states then the material is an insulator at absolute zero.
If the energy gap is not too large compared with kT it is possible to excite
thennally sorne electrons from occupied states into unoccupied states in the
next energy bando This is a semiconductor.
Each energy band can hold 2N electrons where N is the number of
primitive cells in the crystal. Therefore, to fil1 a band we require that the
number of electrons per primitive cell is even. Thus a material can be an
insuIator only if the number of electrons in a primitive cell is an even integer.
For example diamond, silicon, and gennanium have four valence electrons
per atom and two atoms in the primitive cell. There is an even number of
electrons and the four lowest energy bands are fully occupied. There is an
energy gap separating the occupied and unoccupied bands and therefore the
pure crystals are insulators at absolute zero. On the other hand, Group 11
metals such as Ca and Mg have two valence e1ectrons per atom. The reason
why they are not insulators is that the energy bands overlap, so that before
one band is filled the other starts to be filled. However, the density of states
at the Fermi energy is not very high and they are not very good metallic
conductors. An example of overlapping bands was given in Fig. 5.2. The
alkali metals, Li, Na, K, etc. have one valence electron per atom and therefore
they must be metallic.
The great success of band theory is to give an explanation for why so
many materials are metals. Its shortcomings are really two-fold: (i) it ignores
electron-electron repulsions and, as we shall see in Chapter 12, this is the
reason why many (thousands) compounds that are expected to be metallic
are in fact insulators; (ii) its complete reliance on Bloch's theorem does not
enable us to address the electronic structure of amorphous materials, and in
particular, to explain why sorne of them, such as amorphous silicon, have
band gaps.
The chemist's view of band gaps and the distinction between metals and
insulators is more intuitively appealing as it is couched in real space. The
band gap is the finite energy required to take an electron out of a bonding
state and put it into an antibonding state. For example, consider diamond.
If we regard the diamond crystal as a (very) large carbon molecule we can
ask how dilferent are the C-C bonds from the C-C bond we find in ethane
(C 2 H 6 ) for example. In both cases each carbon atom is at the centre of a
tetrahedron of neighbours and chemically speaking each covalent bond is
saturated, that is each bond contains two electrons. It tums out that the
bond orders and bond energies of the C-C bonds are very similar. In this
picture the band gap represents the finite energy required to remove an
electron from a bonding state between two carbon atoms in the diamond
crystal and place it in an antibonding sta te, in analogy with the finite energy
required to excite an electron in a bonding state in the ethane molecule into
an antibonding state. Clearly, this chemical picture of the origin of band
gaps is not restricted to cases where there is translational symmetry.
According to the aboye picture the metallic bond is an unsaturated covalent
bond. Because the band is not full there are less than two electrons per bond
and each bond is unsaturated. Sorne authors speak of the metallic bond as
involving itinerant electrons that are free to roam over the entire crystal,
and c1aim that it is not possible to speak of 'bonds' in the same way as one
would in an insulator. Our viewpoint is quite dilferent. In both insulators
and metals the electronic eigenstates extend over the entire crystal, and, in
both metals and insulators it is possible to construct the density matrix and
talk about charge associated with atoms and bond orders. There is no conflict
with the exc1usion principie here and the equivalence of the descriptions is
Metals. insulators. and the metallic bond 111
mathematicallyexact. In the independent electron approximation (i.e. where
we ignore electron-electron repulsions) the difference between metals and
insulators does not involve the localization of the electrons from extended
eigenstates into bonds. It is a question only of occupation of the available
eigenstates and energy gaps in the spectrum of eigenstates. Density matrix
elements are nothing more than the cumulative result of the interference
between particular (localized) atomic basis states of aH occupied (extended)
eigens tates.
One of the essential properties of a metal is that an electric field is screened
by the e1ectrons. Suppose we introduce a point positive charge into a metal
and ask for the electrostatic potential in the metal caused by this charge.
(We shall treat this problem quantitatively in Chapter 7.) The electric field
of the charge attracts electrons until the electron-electron repulsion caused
by their enhanced local concentration cancels the attraction of the positive
charge. More remote electronic charges see the original positive charge and
the neutralizing c10ud of electrons surrounding it. Thus, they see a neutral
object which has no long range electrostatic field and we say that the
positive charge is screened. In metals the distance over which screening is
achieved is remarkably short-it is about a nearest neighbour distance. The
consequence of such highly efficient screening is that all atoms in the
metal, including impurities and atoms in structural defects, remain 10caHy
charge neutral at aH times longer than it takes electrons to respond to a
perturbation, i.e. longer than ~ 10- 16 s.
By contrast, in an insulator electric fields can be sustained over larger
distances. Thus it is possible to have charged defects in insulators. But there
is not a discontinuous reduction in the screening if a small energy gap is
introduced at the Fermi energy of a metal. In other words the screening
efficiency does not plummet from being perfect to that of a wide band gap
insulator. The relevant energy scale with which to compare the band gap,
El' at. the Fermi energy is the total width, W, of the two bands separated
by the gap. If Ea/W is smaH then screening is still very efficient and atoms
remain almost charge neutral.
6
s-p bonding-a case study
in silicon
s-p bonding
In this chapter we shall enlarge the basis set that we assume Cor each atom
to inc1ude p electrons. It is necessary to do this in order to understand the
cohesion oC many elements oC the periodic table and their compounds. In
Chapter 9 we shall consider bonding caused by interactions between d sta tes.
By surveying the elements oC the periodic table we see that s-p bonding
occurs among the valence electrons oC the elements oC groups I-VII. The
elements oC the first three groups have densely packed metallic structures, i.e.
f.c. c., h.c.p., and b.c.c. As we go down the elements oC group IV Crom C to
Si, Ge, Sn, and Pb, the stable crystal structures change Crom three-Cold
coordination in graphite, to Cour-Cold coordination in the diamond cubic
structure in Si and Ge, to 12-Cold coordination in f.c.c. Pb. In group V (apart
Crom N which Corms diatomic molecules) the pnictides take structures based
on the stacking oC three-Cold coordinated buckled layers oC atoms. In
group VI the chalcogenides take structures based on two-Cold coordinated
helical chains. The group VII halogens crystallize as diatomic molecules (and
thus with a coordination number oC 1) which are held together by weak van
der Waals interactions. Thus, s-p bonding in the elements gives rise to a
wide variety oC crystal structures and coordination numbers. Using an atomic
basis set oC s and p states Cressoni and PettiCor (1991) have shown that it
is possible to understand these trends by looking at the third and Courth
moments oC the local densities of states and the number of electrons per
atom. For example, the stability of the metals of the first three groups in
c1ose-packed crystal structures follows from the fact that the third moment
of the local density of sta tes in these structures is large and negative, which
in tum follows from the large number of three-membered rings. Therefore,
the local density of states is skewed with a long tail below the centre of
gravity (see Problem 19). The electronic energy is thus relatively low when
the band is less than half full, i.e. less than four electrons per atom.
In this chapter we shall consider crystalline Si in detail. The reason for
choosing Si is that it is very important in semiconductor technology and an
understanding of its electronic structure will help to understand its many
varied applications in devices. We shall start by looking at interactions
between s and p states on adjacent atoms. These are slight1y more compli-
cated than interactions between just s states, with the important new
ingredient that the interactions are directional. This directionality is the
origin of directional bonding, and it may be present in metals as well as
insulators. We shall then consider a range of approximations for the band
structure of Si in which we consider hybrids of s and p states as the basis
states rather than the s and p states themselves. Since hybrids are just linear
combinations of the atomic states and the eigenstates of the crystal are just
linear combinations of the atomic sta tes why do we bother to go through
the intermediate step of constructing hybrids? The answer is that the hybrids
give us physical and chemical insight which suggests a range of simplifying
approximations. These insights are particularly useful when we come to
consider amorphous silicon in Chapter 12. But if we want the band structure
of crystalline Si without making these simplifying approximations we do not
bother with the hybrids and give the job to a computer. We shall show band
structures for silicon computed in this way. We conclude with a discussion
of bonding in silicon based on calculations of bond orders and occupancies
of atomic states.
s-p bonding between two silicon atoms

The electron configuration in a free silicon atom is 1s 22s22p 6 3s 23p2 and thus
there are four valence electrons in the outer shell, two of them occupying an
s state and two of them occupying p states. The 3d states are empty, and
although they are necessary for a good description of the higher energy
antibonding states (in the 'conduction band') we shall not inelude them in
the basis set. Qur basis set consists of the 3s state and the three 3p states,
3px, 3py, and 3P., and we shall omit the principal quantum number '3' in
our labelling of them and call them Is), Ix), Iy), and Iz), respectively. These
states form an orthonormal set. This basis set is the smallest set of atomic
states we can use to describe the bonding sta tes of silicon. It is called a
'minimal basis set'o
Consider two Si atoms separated by a distance d. Let the basis sta tes
on the first atom be IS1), IX1)' IY1)' and IZ1> and on the second atom
be Is 2), IX2)' IY2), and IZ2)' The overiap between these states in normal
crystalline Si is actually quite large, about 0.5. However, using a theory
called 'chemical pseudopotential theory' (see Heine (1980) Cor details)
it is possible to transform the basis set into an orthonormal basis
set with the overiap now appearing in the on-site Hamiltonian matrix
elements e. = <slIHlsl) = <s2IHls2) and 6p = <xlIHlx l ) = <Y1IHIY1> =
<zlIHlzl) = <x2IHlx2> = <Y2IHIY2) = <z2IHlz2)' Although this has the
drawback that 6, and 6 p vary with the atomic environment it enables us to
treat the basis states as an orthonormal set. Furthermore the transformation
does not affect the angular character of the basis states so that, for example,
114 s-p bonding-a case study in silieon
Fig.8.1 The hopping integral between the s orbital on atom 1 and the Px orbital
on atom 2 is zero by symmetry.
we may eontinue to treat the IXI> states as a Px state. We ehoose the z-axis
to be along the bond.
Many of the hopping integrals between the two atoms are zero. For
example, eonsider the hopping integral (sIIHlx2>' as shown in Fig. 6.1. This
is zero because the eontribution from the positive lobe of the p orbital is
eancelled by an equal and opposite contribution from the negative lobeo The
rule for a nonvanishing hopping integral is that the two atomie sta tes at
either end of the bond share the same angular momentum eomponent
about the bond axis. (See p. 15 if you need reminding of how to find the
angular momentum of an orbital about a particular axis.) There are just four
nonzero hopping integrals, as shown in Fig. 6.2. Hopping integrals between
states with no angular momentum about the bond axis are labelled with a
CT (greek 's') and hopping integrals with angular momentum m = ± 1 about
the bond axis are labelled with a n (greek 'p'). We now eonsider these four
hopping integrals in tum
(88a)
~ (8pa)
(a) (b)
~
(ppa) (pplt)
(e) (d)
Fig. 8.2 The four fundamental hopping integrals between s and p orbitals: (a)
(ssu). (b) (spa), (e) (ppa), and (d) (ppx).
a. <sdHls2) is the hopping integral we have worked with in earlier chapters.
It is called (ssu), and is shown in Fig. 6.2(a). As we have noted before
(ssu) is a negative quantity because the s orbitals have the same signo
b. <sdHlz2> is called (spu), and is shown in Fig. 6.2(b). The P. orbital has
zero angular momentum about the z-axis. Note that <zIIHls2> = -(spu).
The hopping integral (spu) is positive because the negative lobe of the
IZ2> orbital is nearer to the (positive) ISl) orbital than the positive lobeo
C. <zdHlz2> is called (ppu), and is shown in Fig. 6.2(c). This hopping
integral is a positive quantity because it is dominated by the contribution
from the positive lobe of the IZl) orbital overlapping with the negative
lobe of the Iz 2> orbital.
d. <xlIHlx2> is called (ppn), and is shown in Fig. 6.2(d). It is the same as
<YIIHIY2>' The Px and Py orbitals have angular momentum ± 1 about
the z-axis. The hopping integral (ppn) is a negative quantity because it
is dominated by the overlap between the two positive lobes and the
overlap between the two negative lobes.
To summarize, the signs of the hopping integrals are as follows
(ssu) < O; (spu) > O; (Ppu) > O; (ppn) < O. (6.1)
We emphasize that the (Ppu) interaction is positive. This means that if we
had a linear chain along z with one P. state on each atom the band structure
would be E(k) = Bp + 2(ppu) cos ka, as shown in Fig. 6.3. Because (ppu)
is positive the energy E(k) decreases from k = O to the Brillouin zone
boundaries at k = ±n/a. The eigenstate at k = ±n/a is sketched in Fig. 6.4
and it is seen that adjacent lobes on neighbouring atoms now have the same
sign so that the energy is lowest.
For a given separation, d, the magnitudes of the four hopping integrals
depend on the extent to which the relevant orbitals overlap. For example,
E(k)
---_-n~a+---~--~----~-----n/~a--·k
,,
Fig. 6.3 A one-dimensional energy band arising from (ppu) interactions. Note
that the energy is a maximum at k = O and a minimum at the Brillouin zone
boundaries. in contrast to energy bands arising from (550') interactions.
116 s-p bonding-a case study in silicon
k = ±1t/a
Fig. 6.4 The arrangement of Pz orbitals at k = ±n/a. corresponding to the energy

minimum in Fig. 6.3.
the two P. orbitals in the (ppO') interaction in Fig. 6.2(c) point towards each
other and the overIap is very large. Thus we expect l(ppO')1 to be larger than
the other three hopping integrals. On the other hand the extent of the overlap
in the (pp1t) integral, Fig. 6.2(d), is considerably smaller and therefore we
expect l(pp1t)1 to be correspondingly smaller. For a given separation, d,
Harrison (1980) expresses the ratios of the hopping integral s as follows
(ssO'):(spO'):(ppO'):(pP1t) = -1.40: 1.84:3.24: -0.81. (6.2)
Angular dependen ce of s-p and p-p hopping integrals

When we assign p states to an atom we must choose a coordinate system
to define the directions along which the Px, P" and pz orbitals lie. As we
rotate one atom around another atom, whose centre is the origin of the
coordinate system, the (ssO') hopping integral is obviously invariant, see
Fig. 6.5(a). However, the s-p and p-p hopping integrals change.
In Fig. 6.5(b) we show an interaction between an s state at the origin and
a P. orbital on an atom where the bond axis is at an angle 9 to the z-axis.
We may express the pz orbital as a linear combination of p orbitals in the
rotated coordinate system. Let the rotated coordinate system be denoted by
primes (') and let z' be along the rotated bond axis and x' be normal to both
the rotated bond axis and the axis of rotation, as shown in Fig. 6.5(b).
The P. can be expressed as a linear combination oC orbitals p~ and
p~: p~ cos 9 + p~ sin 9, as shownin Fig. 6.5(c). The s orbital at the origin is
the same in the rotated and unrotated coordinate systems. Now the
Hamiltonian matrix element between the s orbital at the origin and the p~
orbital is (spO'), while that between the s orbital and the p~ orbital is zero.
In effect what we have done here is resolve the P. orbital into components
parallel and perpendicular to the bond axis.
SimilarIy, when we consider the interaction between two p orbitals on
neighbouring atoms we resolve both p orbitals parallel and perpendicular
to the bond axis. In Fig. 6.5(d) we show the interaction between a p orbital
along the unit vector al on one atom and a p orbital along the unit vector
a 2 on an adjacent atom. The bond axis is along the unit vector d. The
components (al'd) and (a2·d) parallel to the bond axis give a contribution
(a, 'd)(a2 'd)(PpO') while the components a, - (al ·d)d and a2 - (a 2·d)d nOf-
mal to the bond axis give a contribution (a, - (al 'd)d)'(a 2 - (a2'd)d)(Pp1t)
<a,IHla 2) = (a, 'd)(a2'd)(ppO') + (al - (a,'d)d)'(a2 - (a 2 'd)d)(pp1t) (6.3)
z·
,,
,
"
a
"x,
(spcs)eosa
iT-1-'--- - - - ... z
(a) (b)
x
z· ;( z·
"
J?"
,
~,""a
- ,, +
" ,
= ,,
- ' ,,
+
~"'"
+
'" x' ',,- x' '",- x'
p, = p~cosa + p~sin9
(e) (d)
Fig. 6.5 (a) The interaction (ssa) is independent of (J. (b) The interaetion
between an s orbital at the origin and a Pz orbital is (spa) eos (J where (J is the
angle between the bond axis and the z-axis. (e) The orbital Pz may be expressed
as a linear eombination of p~ and p~ orbitals in the rotated eoordinate system.
(d) The interaction between p orbitals pointing along the unit veetors B, and B2
separated by the unit bond vector d.
It is the angular dependence of the s-p and p-p hopping integrals that is
the origin of directional bonding in sp-bonded materials.
Sp hybrids
The concept of hybridization was introduced on p. 103 as the construction
of new states from linear combinations of atomic states on the same atom.
In this section we shall discuss the construction of hybrid states as linear
combinations of s and p states on the same atom. The reason for considering
such sp hybrids is that they often allow much greater insight into the band
structure and stability of a particular crystal structure than working with
the atomic states directly. The construction of sp hybrids is based on two
principIes:
1. Bonding is maximized when the extent to which orbitaJs on adjacent
sites overlap spatially is maximized. In eqn (6.2) we saw that the hopping
integral (ppu) is greater in magnitude than the other hopping integrals
beca use the Jobes of the p orbitals point towards each other and thus the
overlap is maximized. The larger the hopping integral the greater the bond
energy because the band width is correspondingly larger, and therefore the
difference in energy between the most bonding and most antibonding sta tes
is greater. It follows that if we construct hybrids which lead to a maximum
overlap between the hybrids on neighbouring sites we shall account for most
of the bonding energy of the system by considering the interactions between
these hybrids and discarding other interactions where the overlap is much
smaller. This is called the principie ofmaximum overlap.
2. The hybrid orbitals at a given atomic centre should be orthogonal. This
is the principie of orthogonality. This principie is not a physical requirement
but it exists to ensure a physically transparent breakdown of the total valence
eleetronie eharge of the molecule or solid into separable contributions, one
from each hybrid. This is already the case for the atomic basis sta tes on each
atom. The total charge is the integral oC the charge density taken over the
molecule or solid. If h¡(r) and h/r) are two hybrid orbitals on the same
atomic state then the total charge associated with 'I'(r) = cihi(r) + eJhj(r) is
given by
f 'I'*(r)'I'(r) dr = /C;l2 f h¡(r)*h¡(r) dr + /c J/2 f hJ(r)*hJ(r) dr
+ ercJ f h¡(r)*hJ(r) dr + cTc¡ f

hj(r)*hi(r) dr. (6.4)
If h¡(r) and hJ(r) are orthonormal then the last two integrals are zero and
the first two integrals are equal to 1. The total charge is then separated into
contributions, /C;l2 and /eJ/2, from the individual hybrids with no overlap
terms to complicate the picture.
We can construct a p orbital on a given atom with any required orientation
simply as a normalized linear combination of the available Px, Py, and P.
orbitals. For example, suppose we :.vant a p orbital in the plane z = O at an
angle 8 to the x-axis, as shown in Fig. 6.6. This is just Px cos (J + Py sin 8.
Now suppose we have two equivalent hybrid orbitals on the same atom (i.e.
they differ only in their orientation)
h1 = N(s + ;'Pl)' h2 = N(s + ;'P2) (6.5)
y
,
,,
8
---:;,..f=-_...l...-_ .t
(pxcos8 + pysin8)
Fig. 6.6 A normalized P orbital in the plane z = O at an angle (J to the x-axis.

This is P. cos (J + Py sin (J.
where N is the normalization constant (1 + A. 2 )-1/2 and Pl and P2 are normal-
ized p orbitals with arbitrary orientations. The mixing ratio of p orbitals to
s orbital in the hybrids h 1 and h2 is A. 2. The overlap (htlh2> is given by
<hllh2> = N 2(l + A.<plls> + A.<SIP2> + A. 2<Pllp2»' (6.6)
which we want to be zero to satisfy the principie of orthogonality. The
integrals <plls> and (SIP2> are zero by symmetry. To evaluate <Pllp2> we
may arbitrarily let the orbital Pl point along the x-axis, and resolve the
orbital P2 into parallel, Px cos 012 , and perpendicular, Py sin 012 , components.
Then
(6.7)
Inserting this value in eqn (6.6) and requiring that <hllh2> = O leads to the
result that
(6.8)
°
Since A. 2 is positive the angle between the hybrids, 12 , must be greater than
90°. The angle between the hybrids varies with the amount of s-p mixing.
With this preparation we can now discuss the three principal types oC s-p
hybridization involving equivalent hybrids on the same atom:
1. sp hybrids. We mix an s orbital with one p orbital, leaving the other
two p orbitals unchanged, and Corm two equivalent hybrids
1
h1 = 2 1/2 (s + Px)
(6.9)
1
h2 = 2 1/2 (s - Px).
In the case oC the hybrid h 1 the positive lobe oC the Px orbital adds to the
positive s orbital to produce an enlarged positive lobe along + x and, simi-
larly, a smaller negative lobe along - x results from the partial cancellation
between the negative lobe of the Px orbital and the s orbital. The opposite
holds Cor h2 • See Fig. 6.7(a). The s orbital content of these hybrids is 1/2.
2. Sp2 hybrids. We mix an s orbital with two p orbitals, leaving the third
p orbital unchanged, and form three equivalent hybrids. The s content of
each hybrid is 1/3. Therefore 1/(1 + A. 2 ) = 1/3 and hence A. = 2 1/2 . From eqn
(6.8) we have cos 0= -1/A. 2 = -1/2 and therefore the three hybrids are at
120° to each other in aplane. If the hybrids lie in the x-y plane then
h1 = (s + 21/2Px)/3 1/2
h
2
= (s _ Px + 31/2py)j31/2
21/2 2 1/2 (6.10)
h = (s _ Px _ 31/2py)j31/2.
3 21/2 2 1/2
'? h¡ z
C9E:J (a)
x
:~--
I
I
h
¡....-___... 1
--- --;-"1
/
,,.,."
/ I
I
I
I
I
I
I
I
I
(e)
x
(b)
Fig. 8.7 Principal types of s-p hybridization: (a) sp hybrids pointing in opposite
directions along the same axis, (b) Sp2 hybrids at 1200 to eaeh other in aplane,
(e) Sp3 hybrids at eos( -1/3) .. 1090 to eaeh other, pointing towards the eorners
of a tetrahedron. From MeWeeny (1979).
These three hybrids are shown in Fig. 6.7(b). The unaffected P. orbital
remains normal to the x-y plane and is orthogonaI to all three hybrids.
3. Sp3 hybrids. We mix an s orbital with all three p orbitals and formfour
equivalent hybrids. The s content of each hybrid is now 1/4. Therefore
1/(1 + ,t2) = 1/4 and hence,t2 = 3. It follows from eqn (6.8) that thehybrids
are at cos -1( -1/3) .. 1090 to each other in a tetrahedral arrangement. The
hybrids are
h¡ = !(s + p" + P. + P.)
h2 = !(s + p" - p. - P.)
(6.11)
h3 = !(s - p" + P, - P.)
h4 = !(s - p" - P. + P.)
These four hybrids are sketehed in Fig. 6.7(e).
The significance of the angular disposition of the hybrids is that it favours
bonding geometries which maximize the overlap of hybrids on neighbouring
atoms. For example, Sp2 hybridization favours planar structures in which
each atom is at the vertex of three bonds at 1200 to each other forming a
hexagonal network. This is the structure in the basal plane of graphite for
example. On the other hand, Sp3 hybridization favours tetrahedral bonding,
such as we find in the diamond cubic, wurtzite, and sphalerite structures.
We shall discuss this point in more detail in Chapter 10, where we shall also
consider hybridization involving d states.
It is very important to distinguish between the s and p content of hybrids
and the numbers of electrons occupying s and p states at an atomic site.
This distinction is often not made and can lead to a great deal of confusion.
The s and p content of a hybrid is defined simply by the particular linear
combination of s and p orbitals from which the hybrid is constructed. Thus,
the s content of sp, Sp2, and Sp3 hybrids is 1/2, 1/3, and 1/4 respectively.
However, this does not mean that 1/2, 1/3, or 1/4 of the electrons associated
with an atom that has formed sp, Sp2, or Sp3 hybrids respectively are
occupying an s state. The occupation of s and p states is determined by
solving the Schrodinger equation and filling the lowest eigenstates until the
required number of e1ectrons per atom is attained. The construction of
hybrids does not imply that the Schrodinger equation has been solved-it
is merely a linear transformation of the atomic basis.
The energy of a hybrid state
(6.12)
is given by
(6.13)
and is therefore a weighted average of the on-site energies e, and Ep' For
example, the energy of an Sp3 hybrid is (Es + 3E p )/4. If we occupy four Sp3
hybrids at a site with one electron each the on-site energy will be e, + 3E p •
On the other hand in the free atom of Si there are two electrons in the s
state and two electrons in p states. The energy associated with that is
28, + 2Ep. Therefore, in order to populate four Sp3 hybrids with one electron
each we must provide a promotion energy Ep - E, to promote an electron
from an s state into a p state. The cost of the promotion energy is offset by
the energy gain associated with the enhanced bonding energy provided by
having electrons occupying strongly overlapping Sp3 hybrids.1t is the balance
between the promotion energy and the bonding energy that determines the
occupations of the s and p states at each atomic site. If the promotion energy
is very large then the energy penalty of occupying Sp3 hybrids is too large
and we say that hybridization takes place to only a small extent. On the
other hand if the promotion energy is small then there is relatively Iittle cost
associated with occupying hybridized orbitals and we say that the orbitals
are strongly hybridized. A high degree of hybridization leads to strong,
directional bonds.
Simple models of the electronic structure of

tetrahedrally bonded silicon
In the previous section we have seen that Sp3 hybrids are tetrahedrally
arranged around an atom. This observation suggests that Sp3 hybrids are a
natural basis set of examining the electronic structure of tetrahedrally bonded
materials, such as silicon. In a nearest neighbour model we allow hopping
between atomic states on nearest neighbours but not between second nearest
neighbours and beyond. In Fig. 6.8 we show various interactions between
Sp3 hybrids on neighbouring atoms that a nearest neighbour Hamiltonian
allows. There are just four nonequivalent hopping integrals, which we have
labelled PI' P2' P3' and P4' The integral PI describes hopping between hybrids
pointing towards each other. Using eqn (6.3) it is not difficult to show that
PI is given by
PI = (sso') - 2(3 1 / 2 )(SpO') - 3(ppO'). (6.14)
4
Because the overlap between these hybrids is much larger than the overlaps
in the other three cases shown in Fig. 6.8 the integral PI is much greater in
magnitude than the other three. This suggests an approximation: we discard
P2' P3' and P4 and keep only PI' Now you can see the insight that introducing
hybrids has brought.
But what about on-site Hamiltonian matrix elements between hybrids on
the same atom? In eqn (6.13) we saw that the on-site Hamiltonian matrix
elements between the same hybrid are Sh = (s, + 3sp)/4. We may choose this
to be our zero of energy. The Hamiltonian matrix elements between different
hybrids on the same atom are easily shown'to be
(6.15)
We have already highlighted the importance of sp - s, in limiting the extent
to which hybridization occurs. If we keep only the hopping integral PI and
o
o
~l
~4
Fig. 6.8 The four types of hopping integral between Sp3 hybrids on neighbour-
ing atoms in a tetrahedral atomic environment.
Simple models of the electronic structure 123
the on-site Hamiltonian matrix elements /1 we shall capture the essential
features of bonding in tetrahedral Si, including the balance between the
promotion and bonding energies which controls the s and p state occcu-
pancies. This is known as the Weaire- Thorpe modelo
An even simpler model is to ignore /1. This is known as the molecular
model, or bond orbital model, because each hybrid is now coupled, through
PI> only to the hybrid it is pointing towards. Each pair of hybrids along a
bond is therefore just like the two atomic states of an Hz molecule. There
is a bonding state with energy PI and an antibonding state with energy - PI
for each pair of coupled hybrids. If there are N atoms, each of them
tetrahedrally bonded, then there are 2N bonds, giving 2N bonding states of
energy PI and 2N antibonding states of energy -PI. The 4N valence
electrons can all be accommodated in the 2N bonding states, leaving an
energy gap of 21Pd between the occupied and unoccupied states.
The success ofthe molecular model is in explaining the energy gap between
occupied and unoccupied states in Si. However, if we want to be more
realistic and broaden the molecular levels ±PI into bands we have to include
/1 as in the Weaire-Thorpe model. The Weaire-Thorpe model for diamond
cubic Si gives the density of states shown in Fig. 6.9. There are two bands,
separated by a gap El. The lower band consists of bonding states and is
called the 'valence band'. The upper band contains antibonding states and
is called the 'conduction band'. As shown in p. 221 there is a gap between
the valence and conduction bands provided
-PI + 3/1 > PI -/1
or
or, since PI and /1 are negative quantities,

IPII > 21/11· (6.16)
In going from the molecular model to the Weaire-Thorpe model we take
d(E)
~==~.==~~==~.~=7-E
Valence Conduction
band band
Fig. 6.9 The density of states for crystalline diamond cubic silicon in the
Weaire-Thorpe modelo
account of the coupling between different hybrids on the same atom. In the
absence of this coupling there is a gap of magnitude 21Pd. By switching on
the coupling the gap is reduced to 21Pd - 41AI. The presence or absence of
the gap is therefore a balance between the energy gained by bonding and
the promotion energy cost ep - e,. In diamond there is a wide gap (5.4 eV
at OK) indicating that the bond energy gained by occupying Sp3 hybrids
easily outweighs the energy of promoting an electron from an s state into a
p state. In silicon the gap is narrower (1.17 eV at O K) indicating that the
two terms are competing more evenly, but the bond energy is still dominant.
The energy gap continues to decrease as we proceed down group IV. This
is essentially a size effect. The larger the atom the greater the bond length
and hence the smaller IPII. The promotion energy also decreases as we go
down the group but not so rapidly as IPII. At the bottom of the group Sn
and Pb are metals.
The Weaire-Thorpe model provides a qualitative understanding of the
factors controlling the band gap in tetrahedrally bonded semiconductors. If
we want to make more subtle distinctions, such as the energy difference
between silicon in the diamond cubic and wurtzite structures, we have to
include the other interactions P2' P3' and P4 shown in Fig. 6.8. There is then
Iittle point retaining the transformation from atomic basis states into hybrids,
and in the next section we show how the band structure is solved with a
minimal atomic basis seto
The band structure of silicon in a minimal atomic

basis set
In this section we shall discuss the band structure of silicon by assuming a
minimal basis set consisting of one s valence state and three p valence sta tes
on each atom. We shall not make any simplifying assumptions about the
interactions between these states by invoking the approximations discussed
in the last section conceming hybrid states. Instead we shall allow an electron
in an s or p state on one atom to hop to an s or p state of any of its
four neighbours. We assume the atomic basis set is orthonormal. This is
called a nearest neighbour, orthonormal, tight binding model. The four
fundamental hopping integrals in silicon at the equilibrium density, were
found by Guo-Xin and Chadi (1987) to be
(ssu) = -1.9375 eV; (spu) = 1.745 ev;}
(6.17)
(Ppu) = 3.050 eV; (ppn) = -1.075 eVo
The on-site Hamiltonian matrix elements were given by Guo-Xin and Chadi
as follows
es = -5.25 eV; ep = 1.20eV. (6.18)
This Hamiltonian has been used extensively to study silicon. In this section
The band structure of silicon in a mini mal atomic basis set 125
:------c"""'\;I---!/".. ,"-,
é,j
fo--- a - - - - 1
(a) (b)
Fig. 6.10 The diamond cubic crystal structure consists ot two interpenetrating
t.c.c. lattices separated by 1/4(1 1 1). (b) The projection along (1 O O) ot the
structure shown in (a). From Harrison (1980).
we confine our attention to the band structure for the perfect crystal which
this Hamiltonian leads too
The diamond cubic structure consists of two interpenetrating f.c.c. crystal
lattices, which are separated by the vector a/4[11 1], where a is the lattice
parameter of the cubic unit cell. Each unit cell contains eight atoms, as shown
in Fig. 6.10, and each atom is surrounded by four neighbours at the vertices
of a tetrahedron. The primitive cell may be defined by the vectors a/2[1 1 O],
°
a/2[1 1], and a/2[0 1 1] and it contains two atoms, one at [O O] and
the other at 1/4[1 1 1]. Since we have a basis set of four atomic states at
°
each atomic site each primitive cell contains eight basis states. Tberefore,
there will be eight bands.
Let !mjO() denote the atomic state R .. + Tj where R .. is the mth lattice
°
vector of the f.c.c. lattice and TJ is the basis vector [O O] or a/4[1 1 1]. The
label O( denotes whether the state is an s, Px' p" or P. state. Let !nk) denote an
eigenstate of the crystal with band index n (1 =:;; n =:;; 8) and wave vector k.
We use Bloch's theorem to write !nk) as an expansion in the atomic states
as follows
!nk) = N!/2 L L eik.(Rm+·j)c}:)(k)!mjO(). (6.19)

.. Jo<
The factor of I/N 1 /2 is the usual normalization factor where N is the total
(inflnite) number of primitive cells in the crystal. F or each eigenstate there
are eight expansion coefficients, c}:)(k), corresponding to the two values of
j and the four atomic states that O( may refer too To find these coefficients
we put !nk) into the Schrodinger equation, H!nk) = E(n)(k)!nk) and project
the resulting equation onto the atomic state !O,j', 0(') by multiplying by the
bra (O,j', O('!
I Hj'a'ja(k)c}:)(k) = E(n)(k)c}?~,(k) (6.20)
ja
where
Hj'a'ja(k) =I eik'(R rn +<;-<j') <Oj'a'IH Imja). (6.21)
m
The Schrodinger equation reduces to an ordinary matrix eigenvalue problem,

eqn (6,20), where the matrix to be diagonalized is an 8 by 8 matrix at each
k. The elements of the matrix to be diagonalized are given in eqn (6.21) and
it is seen that they are expressed in terms of the matrix elements of the
Hamiltonian in the atomic basis set, which are found using eqn (6.17) and
eqn (6.18). Whenj = j' the matrix element <Oj'a'IHlmja) is zero unless m =
and a = a', Thus the site-diagonal e1ements, Hjaja(k), are either Os or 0P'
°
depending on whether a refers to an s or p state. For off-diagonal elements,
where j '# j', the sum in eqn (6.21) reduces to a sum ofjust four terms, one
term for each neighbour.
At each wave vector k within the Brillouin zone of the f,c,c, lattice we
obtain eight eigenvalues by diagonalizing the 8 x 8 matrix Hj'a'Ja(k). In this
way we obtain the band structure shown in Fig. 6.11. Because there are eight
bands at each value of k throughout the Brillouin zone we have to select
directions within the Brillouin zone along which to display the bands, In
Fig. 6.12 we show the Brillouin zone for f.c.c. lattices with high symmetry
points labelled. The bands in Fig, 6.11 have been shown along four directions:
L to r, r to X, X to K, and K to r. Notice that at points of high symmetry,
10,-----r------,-,,------, r - - - - - , - - - - - , - - - ,
L r X U,K r 0.5 1
d(E) I eV I atom
Fig. 6.11 The band structure (Ieft) and density of states (right) of diamond cubic
silicon derived from the Hamiltonian parameters of eqns (6,17-6.18), as com-
puted by Dr A. T, Paxton (1994), private communication ..
The bond order and bond energy 127
Fig. 6.12 The Brillouin zone for f.c.c. lattices with high symmetry points labelled.
The bands in Fig. 6.11 have been plotted along L to í. í to X. X to U. and K to í.
and along high symmetry directions, sorne of the bands have the same energy:
they are said to be degenerate.
On the right of Fig. 6.11 we have shown the density of states. The density
of sta tes is not zero at energy E provided this energy corresponds to at least
one of the eight bands E(n)(k). It is seen that the spectrum of the density oC
states consists of the valence and conduction bands separated by an energy
gap oC about 1.1 eVo The lower part of the density oC states is made up of
four valence bands. In our model, where we have associated a minimal basis
set with each atom, there are only four conduction bands. To obtain a more
realistic description oC the higher energy conduction band sta tes we would
have to include higher energy atomic states, such as d states. However,
inclusion of these higher energy sta tes has little effect on the four valence
bands, for reasons discussed on p. 105.
The four lowest bands are fully occupied by the eight electrons in each
primitive unit cell, leaving the conduction bands empty. The energy gap of
only 1.1 eV does enable sorne thermal excitation of electrons from the valence
band into the conduction bando Thus silicon is a semiconductor, and the elec-
trical conductivity increases with increasing temperature because more elec-
trons are excited into the conduction bando However, the band gap is indirect
because the wave vector at which the valence band is a maximum (r) does
not coincide with the wave vector where the conduction band is a minimum
(L)t. This has Car reaching implications for the optical properties of silicon.
The bond order and bond energy in silicon in a minimal

atomic basis set
The band structure, or 'spaghetti diagram', shown in Fig. 6.11, is not very
helpful in understanding the bond between adjacent atoms in silicon. For
t AClually Ihe conduclion band minimum is al X in silicon. The error in Fig. 6.11 arises from
Ibe use of a minimal basis sel.
128 s-p bonding-a case study in silieon
--\""-60~:.
,
"'-.::' ,,,
,,
,,
JJ
,,;á'
Fig. 8.13 A unit cell of the wurtzite crystal structure. Bonds (Iabelled 2) paraUel
to the e-axis (vertical) are not equivalent to the inclinad bonds labelled 1. AII
atoms are tetrahedrally coordinated as in the diamond cubic structure but the
stacking sequence of planes along <1 1 1> directions is different in the two
structures.
example, what is the degree of Sp3 hybridization7 Is the bonding eontribution

from ppn interaetions negligible? Why is the diamond eubie strueture for
silieon slightly more stable (by about 0.01 eV atom- 1 ) than the wurtzite
strueture, shown in Fig. 6.137 To address questions sueh as these we have
to abandon bands and think about bonds.
Consider a pair of neighbouring atoms, labelled 1 and 2, in the diamond
eubie strueture. Let the z-axis be a10ng the bond, as shown in Fig. 6.14. The
4 x 4 bloek of the Hamiltonian between the atomie states on atoms l and
2 is shown in Table 6.1.
The bond energy of the whole erystal is given by eqn (3.76), where the
subscripts i and j are now understood to be composite indices denoting
atomie site and type of atomie state. The energy of the bond between atoms
Fig. 8.14 A pair of silieon atoms with the bond axis along z. s, Px' Py' and P.
orbitals are shown on eaeh ato m, but the Py orbitals are seen in projection and
appear as small eireles.
The bond order and bond energy 129
Table 6.1 4 x 4 block of the Hamiltonian
between the atomic states on atom 1 and
atom 2
H 52 X2 Y2 Z2
S, (5501) O O (spu)
X, O (pplt) O O
y, O O (pPlt) O
Z, - (spu) O O (ppu)
1 and 2 is therefore given by

Ebond (1-2) = 2P",,(SSU) + 2(p.,,, - P.,.,)(spu) + 2p.,.,(ppu)
+ 2(PX'Xl + Pml)(ppn). (6.22)
There are four contributions to the bond energy arising from (ssu), (spu),
(Ppu), and (ppn) interactions. To evaluate and compare these contributions
we shall need the bond orders PS,." P.,.,,'p.,.,, P.,." PX,Xl' and Py,y,·
The two atomic sites in each primitive cell are equivalent because they
are related by the centre of inversion at the middle of each bond in the
diamond cubic structure. The number of electrons per atom in an s state
is therefore P.", (= P. ,.,), and the number of electrons per atom in
a p state is Px,x, + Py,y, + P.,., (= P"Xl + PYly, + P.,.,). The 's-p mixing' is
defined by
s-p mixing = Px,x, + Py,y, + P.,., (6.23)
PSt 8 !
and is equal to the ratio of the number of electrons occupying p states to

the number of electrons occupying s states. If the degree of Sp3 hybridization
were perfect the s-p mixing would be 3. But owing to the energy cost of
promoting an electron from an s state into a p state the s-p mixing will be
considerably less than 3. To evaluate the s-p mixing we need the on-site
occupancies, which are site diagonal elements of the density matrix.
The task of determining these off-diagonal and diagonal density matrix
elements was undertaken recently by Paxton (1988) using Chadi's
Hamiltonian for silicon (eqns (6.17), (6.18». The results for the bond energy
contributions in eV per bond are given in Table 6.2.
It is seen that the n bonding in the perfect diamond cubic phase of silicon
accounts for only 7 per cent of the total bond energy. The dominance of spu
and ppu over ssu bonding indicates that the bonding is highly directional
in character. The s-p mixing was found to be 1.711. Thus of the four valence
electrons associated with each atom 1.48 are in an s state and 2.52 are in a
130 s-p bonding-a casa study in silicon
Tabla 6.2 Bond energy contributions in eV per bond
55.,- SP"- pp.,- ppn Ebond (1-2)
-0.812 -2.479 -4.245 -0.603 -8.139
p state. The 'on-site energy' associated with these occupancies is given by

(6.24)
and using eqn (6.18) for Es and Ep we get Esite = -4.715 eVo If the sp mixing
were closer to 3 the on-site energy would increase more than the bond energy
would decrease. Adding the site energy to twice the energy of a single bond
(there are two bonds per atom) we get the band energy per atom, eqn (3.77),
which is the sum of the occupied eigenvalues in the crystal. This number is
-20.993eV per atom.
The comparison with the wurtzite phase of siJicon, shown in Fig. 6.13, is
very instructive. The four bonds to each atom are no longer equivalent, even
though all four bonds have the same length when the e/a ratio is ideal. There
is no symmetry operation relating the bond along the e-axis, labelled '2' in
Fig. 6.13, to the three bonds, labelled '1' in Fig. 6.13, inclined to the e-axis.
Therefore, it is necessary to consider the bond orders in bonds 1 and 2
separately, and the bond energies (eV) are shown in Table 6.3.
The bond energy per atom is therefore -(3/2) x 8.180 - 0.5 x 8.100 =
-16.32 eV which is larger in magnitude than the bond energy per atom in
the diamond cubic phase, -16.28 eVo So why does silicon assume the
diamond cubic structure? The answer is that the s-p mixing in the wurtzite
phase is 1.728, which gives rise to an on-site energy of -4.661 eVo The
band energy per atom in the wurtzite phase is therefore -16.32 - 4.661 =
-20.981 eV, which is approximately 0.01 eV atom -1 greater than in the
diamond cubic phase. Thus the more negative covalent bond energy in the
wurtzite phase is more than compensated by the more positive on-site energy,
thus rendering the wurtzite phase 0.01 eV atom -1 higher in energy. This
calculation is based on the assumption that the bond lengths are the same
in the two phases, which appears to be the case. It iIIustrates the delicate
Tabla 6.3 Bond energies in bonds 1 and 2
Bond ss.,- SP"- pp.,- ppn
1 -0.827 -2.485 -4.261 -0.607 -8.180

2 -0.821 -2.507 -4.211 -0.561 -8.100
References 131
balance between the on-site terms in the energy and the bond energy in
determining phase stability.
References
Cressoni, J. C. and Pettifor, D. G. (1991). J. Phys.: Condenso Matter, 3, 495.
Guo-Xin, Q. and Chadi, D. J. (1987). Phys. Rev. B, 35, 1288.
Harrison, W. A. (1980). E/ectronic structure and the properties 01 so/ida, p. 49. W. H.
Heme, V. (1980). So/id State Phys., 3S, pp. 47-57.
McWeeny, R. (1979). Cou/son's Va/erICe. Oxford Univcrsity Press.
Puton, A. T. (1988). Phi/o Mag. B, 511, 603.
7
Free electron theory
Introduction to free electron theory

Metals have quite different properties from insulators. Apart from their
much greater electrical conductivity, metals also have a greater thermal
conductivity, their surfaces, when c1ean, are lustrous and, on the whole,
metals are much more ductile. Understanding these fundamental differences
was the first challenge to he addressed by solid state theorists around the
turn of the century, and indeed it can be said that these questions brought
solid state physics into being. The first theory was put forward by Drude
and although it was entirely ciassical in nature it enjoyed sorne immediate
and dramatic successes. Acccording to this theory the electrons in the
outermost shell become completely detached and are free to wander through
the crystal, forming an e!ectron gas. However, the limitations of Drude's
model were soon appreciated and it was found that e!ectrons in metals (at
least at temperatures of up to about 60 000 K) obey the Schrodinger equation
rather than Newtonian mechanics. The key difference that this realization
led to was that the e!ectrons in the metal obey Fermi-Dirac statistics and
not Maxwell-Boltzmann. With this improvement it was possible to under-
stand the electronic contribution to the specific heat of metals, as well as
many other thermal properties which were not well described by Drude's
theory.
From a physical point of view the assumptions of free electron theory are
rather odd to say the least. When we say the e!ectrons in a metal are free
we mean that they do not fee! the e!ectrostatic potentials of the ion cores
they leave behind them (except their average potential which keeps the
electrons in the metal), nor do they feel each other. We have already
encountered Bloch's theorem on p. 49 which sta tes that provided the crystal
is perfect electrons can propagate through an infinite crystal indefinitely.
Thus, Bloch's theorem explains why e!ectrons are not continuously being
deviated by the ion cores. But the reasons why we can ignore electron-
electron interactions are more subtle, and indeed they break down under
certain conditions described. in Chapter 12. In this chapter we shall discuss
how this assumption has beenjustified only relatively recently. We shall also
relate the band structures we have discussed in earlier chapters, based on
linear combinations of atomic orbitals, to the free electron and nearly free
electron band structures. We shall see that they are mere!y different ways of
The free electron approximation 133
describing the electronic structure of solids, and although they appear to
start from rather different viewpoints there is a definite sense in which they
converge to the same physical picture.
The free electron approximation

In Drude's free electron model (Drude 1900) it is assumed that when atoms
of a metal are brought together the valence electrons become detached and
wander freely through the metal, while the positive ions they leave behind
remain intact and are immobile owing to their much greater mass. The
mobile valence electrons are often called conduction electrons in free electron
theory beca use they are responsible for electrical conduction. The other
electrons in the atom are core electrons and remain attached. Drude applied
Maxwell's highly successful kinetic theory of gases to the conduction
electrons ofthe metal (quantum theory had not yet been born). It is assumed
that the electrons travel in straight lines until they collide either with another
electron or, much more likely, with a positive ion. No forces are assumed to
act between the electrons or between the electrons and the ions, except
momentarily during a collision. In the presence of an externally applied
electric or magnetic field each electron responds according to Newton's
equations of motion for the electron in the presence of the external field
while neglecting the interactions with other electrons and ions in the metal.
The neglect of electron-electron electrostatic interactions between collisions
is known as the independent electron approximation.
The physical picture we have so far is of a gas of free electrons travelling
in straight lines until they bounce off the fixed ions. Drude introduced a
parameter , as the average time between successive collisions of an electron.
It is called the relaxation time or the collision time. It plays a central role
in the theory of the electrical conductivity which we shall discuss in the next
chapter. Since the electrons do not interact either with each other or with
the ions between collisions thermal equilibrium is assumed to be achieved
through the collisions with the ions.
Drude's model, despite its simplicity, can explain Ohm's law and provide
an estimate of the resistance, the Hall effect, and provide estimates for the
Hall coefficient and the magnetoresistance, the AC electrical conductivity of
a metal, the thermal conductivity of a metal and the Wiedemann-Franz
law. These properties will be discussed in the next chapter. The ductility of
metals can also be understood because metallic cohesion in Drude's model
comes from the attraction of the positive ions to the sea of electrons in which
they bathe. The attraction is not sensitive to the relative positions of
the ions provided the electrons remain truly free (i.e. there are no direc-
tional bonds and the conduction electron density remains uniform). Hence
metals (for which free electron theory is an adequate description) tend to
have small shear elastic constants compared to their bulk moduli. They tend
134 Free electron theory
to adopt close packed structures because of the absence of directional
bonds and they often fonn alloys with each other over a wide range of
compositions.
The most spectacular failure of Drude's model was its prediction that the
electronic specific heat should be 3k/2 per electron, where k is the Boltzmann
constant, whereas the observed electronic specific heat was much less. This
is the prediction that follows from assuming that the distribution of electron
velocities is the Maxwell-Boltzmann distribution. The discrepancy was
removed by Sommerfeld when it was realized that the conduction electrons
in the metal must obey Fermi-Dirac statistics because of the exclusion
principie. The exclusion principie plays a central role in many of the observed
properties of metals, as we shall see latero Sommerfeld's free electron model
is exactly the same as Drude's except for the replacement of the Maxwell-
Boltzmann distribution by the Fenni-Dirac distribution. The free electron
gas in Sommerfeld's model is sometimes called the free electron Fenni gas.
Why do the conduction electrons in a metal obey quantum laws rather
than classical laws? The reason is the high density of the electron gas and
the low electron mass. All particles (whether quantum or classical) must
obey the uncertainty relation
.1x.1p ~ h. (7.1)
.1x is the uncertainty in the position, .1p is the uncertainty in the momentum,
and h is the Planck constant (6.626 x 10- 34 Js). If there are N conduction
electrons per unit volume, the volume within which we are Iikely to find just
one electron is l/N. The uncertainty in the position, .1x, of any electron is
therefore 1/N 1/ 3 • From the uncertainty relation, eqn (7.1), the (minimum)
uncertainty in the momentum of any electron is therefore hN 1/ 3 • The electron
gas cannot obey classical equations of motion unless it is heated to such a
temperature that the available thermal energy exceeds the kinetic energy
arising from the minimum uncertainty in the particles' momenta. This kinetic
energy is just
(7.2)
Let us put sorne numbers into this. The mass, m, of a free electron is
9.1 x 10- 31 kg. In Cu we assume one conduction electron per atom,
and therefore N = 8.4 X 1028 m- 3 • Thus, (.1p)2/2m ~ 4.6 x 10- 18 J, and
therefore the electron gas behaves classically only if it is heated to
4.6 x 1O- 18/k ~ 300000 K, which is obviously much greater than the melt-
ing point! Therefore electrons in metals always obey quantum mechanical
equations of motion. Once this was realized theorists decided that a full
quantum mechanical description of electrons in metals was essential' and
this is what we now develop.
Electrons in a box
We shall now try to ca1culate the energy levels and wave functions for the
conduction electrons in a metal in the ground state, i.e. at OK, in the
independent electron approximation. Since the electrons are assumed not to
interact in this approximation we may ca1culate the allowed eigenvalues and
eigenfunctions for one conduction electron in the metal, as though the other
electrons were not there, and then populate these eigenvalues with electrons
according to Fermi-Dirac statistics,just as we have done in earlier chapters.
The assumption of T = O K is very good even at the melting point of the
metal because the melting point is only a very small fraction of the
temperature at which the electrons would behave classically. Consider a
cubic macroscopic lump of metal of side L. What shall we do with the positive
ions? In the free electron approximation we imagine they are smeared out
into a uniform positive charge distribution which compensates exactly the
negative charge distribution of the electron gas. The 'material' to which this
approximation corresponds is called jellium: a metallic jelly. Thus, the lattice
on which the ions sit is smeared out into a continuum of positive charge.
The role oí the smeared out positive charge is to provide an attractive
potential to stop the electrons escaping from the lump of metal. As shown
in Fig. 7.1, the electrons do spill over the edge of the positive charge
distribution at the sudace of the lump, thereby creating a surface dipole.
Although we shall not be concemed with surface effects it does have one
very important consequence for the whole specimen. The surface dipole
introduces a rigid shift, - Va, of the electrostatic potential throughout the
interior of the lump relative to the potential in the vacuum outside.
To avoid having to worry about the variation in the charge density near
the surface we apply the familiar trick of periodic boundary conditions to
l
the surfaces of the cubic lump. This leads to the following boundary
conditions on the wave function
'P(x, y, z + L) = 'P(x, y, z)
'P(x, y + L, z) = 'P(x, y, z) (7.3)
'P(x + L, y, z) = 'P(x, y, z).
_-810ad,0<I regíon
= positive continuum
./ E,lectron densíty
Fig.7.1 A finite lump of jellium: the shaded region is the smeared out positive
background charge density. The solid curve shows the electron density.
As we have already pointed out the potential inside the lump is Va below
that of the vacuum outside. We may take the potential inside the lump to
be zero, and the potential outside the lump to be + Vo. Our jellium model
with periodic boundary conditions thus boils down to a particle in a cubic
box oC side L in which the potential is zero, subject to the boundary
condition, eqn (7.3), on the wave function. The periodic boundary condition
effectively ensures that the electron never escapes Crom the box.
The SchrOdinger equation Cor an electron in the box is
_h 2
-2- V2 '1'(r) = E'I'(r) (7.4)
8n: m
which is just the equation for a free particle. It is readily shown that the
solutions to this equation are
Uf ( ) 1 ;k •• (7.5)
Tk r = V 1/ 2 e
where
(7.6)
and we have not yet imposed the boundary conditions, eqn (7.3). The
prefactor oC 1/V 1/ 2 in eqn (7.5) for the wave function is a normalization
constant to ensure that the probability oC finding the electron somewhere in
the box, the volume of which is V = L 3 , is one. Imposing the periodic
boundary conditions, eqn (7.3), we obtain
(7.7)
and thereCore the components oC the vector k must be oC the Corm
k = 2n:nx k = 2n:ny k = 2n:n. (7.8)

xL' y L' • L
where nx , ny, and n. are integers.
The vector k is the wave vector that was introduced on p. 50. Its physical
significance is that the momentum of the electron is directly proportional to
k. This may be seen by applying the momentum operator p = hV/2Tri to the
eigenstates, eqn (7.5), whereupon it is seen that the momentum is hk/2n:.
Thus the momentum and energy are quantized because the wave vector is
quantized, which in turn follows from the periodic boundary condition, eqn
(7.3). Notice the striking similarity with what we have already done in
Chapter 3 in connection with Bloch's theorem and linear combinations of
atomic orbitals.
Because L is very large compared with the atomic spacing the energy
difference between successive eigenvalues, in eqn (7.6), is vanishing1y small.
Therefore the quantization condition on k is not very significant and its main
use is that it enables us to keep track of the number of eigenstates in a region
of k-space that is large compared with 21t/L. For example, the volume of
k-space occupied by just one state is (21t/L)3 = 81t 3/V. Therefore the number
of states per unit volume of k-space is V/(81t 3). Into each state labelled by
the wave vector k we may place two electrons, one with spin up and one
with spin down. Thus, we fill up the available states with two electrons each,
starting with the lowest energy state and moving up in energy until we have
put in all the conduction electrons of the metal. The energy of the highest
occupied state is the Fermi energy. In eqn (7.6) we see that the energy of the
state with wave vector k is proportional to k 2 , and therefore all wave vectors
up to sorne maximum length are occupied. Thus in k-space the occupied
sta tes fall within a sphere and the radius of the sphere is denoted by kF where
h2k~
EF=--'
2
(7.9)
81t m
kF is the Fermi wave vector (introduced on p. 53), although it is only a scalar,
and the sphere Ikl ~ kF is called the Fermi sphere. The Fermi surface is the
surface of this sphere. The value of kF is determined by the condition that
the number of electrons contained within the Fermi sphere is equal to the
number of conduction electrons in the metal. Since there are V/(81t 3 ) states
per unit volume of k-space the number of electrons per unit volume of
k-space is V/( 41( 3 ). Thus the number of electrons contained in the Fermi
sphere is
41tk~ V
N = - - 3. (7.10)
3 41t
Equating N with the number of conduction electrons in the lump oC metal
ofvolume Vand letting n = NIVbe the electron density in the metal we have
(7.11)
Thus the Fermi wave vector increases as the cube root of the electron density.
Putting in typical conduction electron densities for metals we find that kF is
of the order of inverse angstroms. This means that the de Broglie wavelength
of the most energetic electrons in a metal is of the order of angstroms, which
is comparable to the interatomic spacing. Similarly, it is found, using the
relation for the Fermi momentum, PF = hk F/21t, and the Fermi velocity,
VF = PF 1m, that the Fermi velocity in metals is about 1 per cent of the speed
of the light. Using eqn (7.9) it is found tliat the Fermi energies of metals
vary between about 1.5 and 15 eV, which is much greater than the 0.025 eV
of thermal energy available at room temperature. The very high speed and
energy of electrons at the Fermi surface is entirely a consequence of the
exclusion principie which forbids the lower energy sta tes being occupied by
more than two e1ectrons. At O K all states up to the energy EF are occupied
by two electrons and all states aboye EF are unoccupied.
Density of states
The density of sta tes in two- and three-dimensional crystals was introduced
on p. 88, where we considered a simple s-band model. In the limit of small
band fillings the band structure for the s-band model reduces to the free
electron case, given by eqn (7.6). In this section we shall consider the density
of sta tes in the free electron approximation in one, two, and three dimensions.
One major difference compared with the densities of states in the s-band
model is that there is no upper bound on the energy spectra. In the s-band
model the E(k) relation looks free-electron-like at both the upper and lower
bounds of the energy spectrum. In the free electron case all that changes as
we go from one to two to three dimensions is that the wave vector is confined
to a line, aplane or three dimensions in k-s pace. The relation between the
energy and the wave vector remains eqn (7.6) in all three cases. In the
one-dimensional case the number of states per unit length of k-s pace is L/2n.
In the two-dimensional case the number of states per unit area of k-space
is (L/2n)2.
In the one-dimensional case the density of states in k-space is trivial. It is
D(k) dk = (L/2n) dk. From eqn (7.6) we have
2
dk = 4n m dE = n(2m)1 /2 dE (7.12)
2
h k hE l /2
and therefore the density of states D(E) becomes
L n(2m)1/ 2 Lm 1/2
D(E) = 2n hEl/2 = h(2E)1/2
(7.13)
This is shown in Fig. 7.2(a). As in the s-band model the density of states
diverges at the bottom of the energy spectrum. In the s-band model the
density of states diverges at the upper bound of the spectrum as well, where
the E(k) relation looksfree electron-like (see Fig. 3.13).
In the two-dimensional case a circle of radius k centred on the origin of
D(E) D(E) D(E)
E E E
(a) (b) (e)
Fig. 7.2 Densities of states in the free electron approximation: (a) one dimen-
sion, (b) two dimensions, and (c) three dimensions.
k-space is a line of constant energy. The area between this circle and a
concentric circle of radius k + dk is 2nk dk. The number of states contained
in this annulus is
D(k) dk = (~y 2nk dk. (7.14)
Since E is now a function ofthe vector k we have to consider dE = VkE-dk.

But VkE is along the radius vector k and so is dk. Therefore eqn (7.12) still
applies. It follows that
D(E) = 2n~2m . (7.15)
h
which is a constant, and is shown in Fig. 7.2(b). Comparing this with the
density of states for a square lattice s-band model, Fig. 4.9, we see that the
s-band density of state cOQverges to a finite value at the upper and lower
bounds ofthe energy spectrum, where E(k) is free electron-like, and diverges
in the middle of the bando
In the three-dimensional case we consider two concentric spheres of radii
k and k + dk centred on the origin in k-s pace. The volume contained within
the spherical shell of thickness dk between them is 4nk 2 dk. The number of
states contained in this spherical shell is therefore
D(k) dk = (~r 4nP dk. (7.16)
Again we observe that VkE is along the radius vector k, as is dk, and therefore
D(E) = 4n~3m (2mE)1/2. (7.17)

h
This is sketched in Fig. 7.2(c). Comparing this with the density of states in
the simple cubic lattice s-model, Fig. 4.10, we see the free electron El/2 type
singularities at the upper and lower bounds of the energy spectrum. In
between these bounds the s-band density of states departs from free electron
behaviour, as expected from the functional form of E(k).
Using eqn (7.17) the average energy per electron in the three-dimensional
gas may be shown (see Problem 23) to be 3E F /5, the pressure due to the
electron gas is 2ET /(3V) where ET is the sum of the occupied eigenvalues
and the bulk modulus of the electron gas is 10ET /(9V). The shear elastic
constants of the free electron gas are zero.
Free electron bands and LCAO bands

The free electron 'band structure' is simply
h2 k2
E(k) = E(k) = -
2
. (7.6)
8n m
We may generate a free electron band structure for a material by using the
reduced zone scheme discussed on p. 77. As an iIlustrative example consider
a one-dimensional crystal with lattice constant a. The first Brillouin zone
lies in the range -n/a ~ k ~ n/a along the k-axis. But eqn (7.6) is defined
for all k from - 00 to + oo. In the reduced zone we fold the E(k) relation
for k outside the first Brillouin zone back inside, as shown in Fig. 7.3. In
a three-dimensional crystal we draw the Brillouin zone in k-space, e.g.
Fig. 4.2 or Fig. 4.3, and we plot eqn (7.6) along the high symmetry directions
as we did in Fig. 6.11 for silicon. When the Brillouin zone boundary is
reached we fold the free electron band back into the Brillouin zone as in
Fig.7.3.
To deD;lonstrate the construction of free electron bands for a lattice
consider a square lattice of spacing a. The reciprocallattice is also a square
lattice and the lattice constant is 2n/a. In the reduced zone scheme any wave
vector k' lying outside the first Brillouin zone is mapped onto a vector k
inside the Brillouin zone by
k=k' +G (7.18)
as shown in Fig. 7.4. The free electron band structure in the extended zone
scheme is given by
(7.19)
which is a paraboloid. In the reduced zone scheme this paraboloid is folded

into the first Brillouin zone through eqn (7.18). This means that for a vector
k lying within the first Brillouin zone not only is there an eigenvalue equal
E(k)
-""::::::-¡"-=--k
-'/t/a n/a
Fig. 7.3 The one-dimensional free electron bands in a redueed zone se heme.
This is obtained bV folding the parabala E(k) = h2~/8'/t2m into the first Brillouin
zone between -n/a and + n/a.
• • •
k
• •
• • •
Fig. 7.4 To show that a wave vector k' Iying outside the first Brillouin zone of
a two-dimensional square lattice is equivalent to a wave vector k inside the zone
by addition of a reciprocal lattice vector G.
to h2 k2 /(8rr: 2 m) but there is also an infinite set oC higher eigenvalues equal

to h2 1k + GF/(8rr: 2 m) where G is an arbitrary reciprocal lattice vector. In
Table 7.1 we show how the Cree electron bands along (k x • O) are generated
Cor the square lattice.
The six bands oC the table are plotted in Fig. 7.5.
Fig. 7.6 Free electron bands for a square lattice shown along the [1 O] direction.
Table 7.1 Generation of free electron bands along (kx ' O) for
the square lanice
Band (a/2n)G 8n2mE(0. O)/h' 8n2mE(kx .0)/h2
1 (O, O) O .tc;
2 (± 1, O) (2n/a)2 (k x ± 2n/a)2
3 (O. ± 1) (2n/a)2 .tc; + (2n/a)2
4 (±1. ±1) 2(2n/a)2 (k x ± 2n/a)2 + (2n/a)2
5 (±2, O) (4n/a)2 (kx ± 4n/a)2
6 (O. ±2) (4n/a)2 .tc; + (4n/a)2
It is interesting to ask how close the free electron bands are to the true
electron bands in a material, and also to the LCAO bands that we have
discussed in the first part of this book. In Fig. 7.6 we show the LCAO bands
fitted for Ge compared with the free electron bands. The two band structures
are surprisingly similar. The main difference is the appearance of energy gaps
in the LCAO band structure, but the surprise is that the gaps are relatively
T
] 3/8(h2¡mtf)
.g
~ 9/32(h2¡mtf)
1
í]
3¡16(h2¡mtf)
~
~ 3/32(h2¡mtf)
1 L r
(a)
x L r
(b)
x
Fig.7.8 Comparison between the LCAO bands far Ge, computed with an Sp3
basis, and the free electron bands. From Harrison (1980).
small compared with the bandwidths. Otherwise the shapes of the valence
bands are quite similar and in particular the valence bandwidth is quite c10se
to the corresponding free electron value. This observation suggests that since
the width of a band in the LCAO treatment scales with the hopping integral
and the width of a band in the free electron theory scales as k 2 , which scales
as 1/d 2 where d is the interatomic spacing, then the hopping integral in a
material like Ge should scale with bond length like 1/d 2 . This scaling
relationship for the hopping integrals in 'nearly free electron materials' is
used widely.
For the so-called 'simple' metals the free electron bands are a very
reasonable first approximation to the true electron bands. The simple metals
are metals of groups 1, 11, and 111, excluding the first periodo In Fig. 7.7(a)-(c)
we show the ca1culated band structures and densities of states for Na, Mg,
and Al. At low energies the band structures are very c10se to being free
0.60
0.40 r ....:~
~..
~ :
e¡¡ 0.20 r
~
0.0 V'"
....... \.,.", V,,/
-ü.20
r H N p r N
Na
2.50 0.70
<a)
"7'
-.:.- 2.00
0.60 T
~
~
....o
~
0.50 r"
~
1.50
.~
0.40 ~
VI
~ ~
-8=
~
0.30
1.00
~
i 0.20 ....o
~
j> 0.50
'5=
.s .... ..
0.10
8
0.00 0.0
-4 -3 -2 -1 O 2 3
Eneq¡y relative 10 Fermi ene'llr (eV)
Flg.7.7 Computed energy bands and densities of states for (a) Na. From
Moruzzi et al. (1978). Continued.
0.80,.----,....-----r-----..---r.......,
0.60 • ". ..•. ,.-.

~ ···0 .....
~ 0.40·~.:
E,
i 0.20
0.0
,."'-~./../
\ ...•••..
-0.20
W L r x w K
Mg
3.50 0.60
(b)
'7' 3.00 T
~
. 0.50 ~
t'.'"
.'a
M 2.50
0.40
.~ 2.00 M
. 0.30 en ~
.
.g 1.50 ... M.
. 0.20
. 'a
11.00
... ..
?;>
.~
j 0.50
....
0.10 e
¿¡
.' .'
0.0
-8
.'
-6 -2 -4 O 2 4
0.0
Energy relative to Fenni energy (e V)

Fig.7.7 (b) Mg. Continued.
electron·like and the densities of states are close to being parabolic, in

agreement with eqn (7.17). At higher energies additional structure is seen in
the densities of states which originates Crom the small band gaps at the
Brillouin zone boundaries in the band structures. Again, it is perhaps
surprising how small the band gaps at the Brillouin zone boundary are. This
is a point we shall develop in the next section.
The nearly free electron model

In this section we shall put the electrostatic potential oC the ions in the crystal
back into the Cree electron modelo Recall that we had smeared out the positive
potential oC the ions into a uniCorm background charge density. Now we
shall consider the effect that the discrete lattice oC positive ionic charges has
The nearly free electron model 145
0.80 ,......,,','--.---:--"""T----.--.:-.:-._'.. T",'.,...• ...,
... \ .... .
::: If-\-::\~-+:.---:,'
. .:.....--+--~. .:I-~-*.;~o: __: : I
§
j 0.20
0.0
.lo"
-0.20
-O.40~----!:__--_=_---~--=-~
w L r x W K
Al
4.--------------,0.50
(el T
~
0.40 -~
.,
.' 0.30 !
,
.. 0.20 !
t
~
~
0.10
..... ,.'
O L..L...:.1.._.L----L._.L----L._~~--' 0.0
-12 -10 -8 -6 -4 -2 O 2 4
Energy relative to Fermi energy (eV)
Fig. 7.7 (e) Al.
on the band structure of the conduction electrons. For simplicity we consider

a one-dimensional model. The generalization to two- and three-dimensional
crystals is straightforward.
In Fig. 7.8 we sketch the one-dimensional electrostatic potential U(x) felt
by an electron due to a one-dimensionallattice of positive ions, with spacing
a. At each ion core the potential diverges to - 00 because oC the l/x
singularity in the Coulomb potential. However, the potential is periodic and
thereCore it can be expanded in a Fourier series
<Xl
U(x) = L (7.20)
"'=-00
Fig. 7.8 The electronic potential U(x) for a linear chain of atoms.
The Schrodinger equation Cor a conduction electron in the presence oC this

periodic potential is
h 2 d 2 '1'
- -2- - -2 + U'I' = E'I'. (7.21)
8n m dx
Since the lattice is periodic the solutions of this differential equation must
satisfy aloch's theorem (eqn (3.34»
(3.34)
where p(x) is another periodic function with the periodicity of the lattice:
p(x + a) = p(x). But since p(x) is periodic it may also be expanded in a
Fourier series
ro
p(x) = ¿ c.(k) e 2 • i • x /. (7.22)
11= -00
where we have expressed the expansion coefficients, c.(k), as functions of k

in anticipation of the result that p(x) varies with k. Substituting 'l't(x) and
p(x) into the SchrOdinger equation we obtain a set of coupled equations for
the expansion coefficients c.(k)
h2 d 2
- - 2 - --2 ¿ co(k) ei(HG)x + ¿ UG, eiG'x ¿ cG(k) ei(k+G)x
8n mdx G G' G
= E(k) ¿ cG(k) ei(HG)x (7.23)
G
where G and G' are arbitrary reciprocailattice vectors ofthe one-dimensional

lattice. Carrying out the differentiation in the first term this equation becomes
h:
8n m G
¿ cG(k)(k + G)2 ei(HG)x + ¿
GG'
UG,cG(k) ei(HG+G')x
= E(k)¿ cG(k) ei(k+G)x. (7.24)

G
It we multiply throughout this equation by e-i(HG")x and integrate over x

from O to a we project out the following equation
h 2(k + G")2
2 CG,,(k) +¿ UG,cG"-G,(k) = E(k)cG,,(k). (7.25)
8n m G'
This is one of an infinite set of coupled equations for the expansion
coefficients. There are as many equations as there are coefficients. Con-
sequently for a given k there are as many solutions as there are expansion
coefficients, and these different solutions correspond to different bands at
the same value of k. They may be distinguished by a band index, n. All of
these equations may be combined into a single matrix eigenvalue problem
in which each eigenvector at a particular value of k consists of the expansion
coefficients for a particular band and the corresponding eigenvalue is the
energy of the band at that value of k. The matrix eigenvalue problem is
equivalent to the secular equation of eqn (3.4) in the LCAO case. In both
cases we express the wave function in sorne basis set and the Schriidinger
equation is then transformed into a matrix eigenvalue problem. In the LCAO
case the basis set is a set of atomic-like orbitals, which are functions that are
anchored to the atoms in the system. In the nearly free electron case the
basis set is a set of plane waves, which are complete1y independent of the
atomic positions. There would be absolutely no difference between the
solutions if the two basis sets were mathematically complete. lt would then
be entirely a matter of taste or computational convenience which basis set
one chose. In general one needs many more plane waves in which to expand
an eigenstate in a crystal than one needs atomic-like orbitals. But there are
competing advantages with plane waves as a basis set, such as the ease with
which the kinetic energy is ca1culated.
Recall that in the LCAO case band gaps arose for atomic/chemical
reasons, namely the presence of more than one atomic state in the unit cell
giving rise to bonding and antibonding states separated by an energy gap.
For example in a linear chain of atoms along x in which there is one s state
and one Px state per atom there are two bands separated by a gap, as shown
in Fig. 7.9. The upper band has a maximum at the Brillouin zone centre and
a minimum at the Bri1louin zone boundary and the reverse holds for the
,,,}Gap
----T-t-----t-----f-r'--~kx
Fig. 7.9 The energy bands for an infinite linear chain of atoms along x with s
and Px states on each atom.
lower bando The upper band is therefore primarily a p band and the lower
band primarily an s bando
Let us now try to understand the origin of band gaps at the Brillouin zone
boundary in the nearly free electron model. The key point is that Bragg
reflection takes place at the Brillouin zone boundary. This follows from the
Bragg condition Ikl = Ik + GJ because
P = (k + 6)2
:$ k'G = -G2 /2. (7.26)
By allowing G to range over the smallest reciprocallattice vectors such that
the surface defined by eqn (7.26) is c10sed we obtain the polyhedron defining
the first Brillouin zone. In one dimension, eqn (7.26) becomes k = - G/2, i.e.
k = ± me/a and therefore Bragg reflection occurs at the first Brillouin zone
boundaries k = ±n/a. Thus at k = n/a the travelling wave e iKX/' moving to
the right is Bragg reflected into the wave e-I~x/. travelling to the left. From
these two travelling waves we can construct two standing waves
'1'( +) = (eh'x/. + e-I~xfa) = 2 cos(nx/a) }
and (7.27)
'1'( -) = (e IKX/' - e-I~x/.) = 2i sin(nx/a).
Both standing waves are made up from equal parts of right and left travelling
waves.
The charge density is proportional to 1'I'(xW. For a travelling wave
!'I'(x)1 2 = e lb e-lb = 1, and therefore the charge density is constant. But the
charge densities for the standing waves are far from constant. The charge
density, p(+)(x), arising from '1'( +) is 4 cos 2 (nx/a), whereas the charge
density, p(-)(x), arising from '1'( -) is 4 sin2 (nx/a). These charge densities
are sketched in Fig. 7.10, where it is seen that p(+)(x) heaps the electronic
charge on the ions whereas p(-)(x) heaps the electronic charge between the
atoms. Therefore p(+)(x) resembles the charge density oC a set of atomic s
orbitals, whereas p(-)(x) resemblesthe charge density oC a set oC Px orbitals.
Moreover, the p(-)(x) charge density will be associated with a higher energy
Fig. 7.10 The charge densities p(+)(x) and p(-)(x) of standing waves set up by
Bragg reflection of free electron travelling waves at the Brillouin zone boundary
for a linear chain.
:,', E oc k 2 (free electron)
E.
,
, ,,
'1'(+)
Fig.7.11 The energy bands near the Brillouin zone boundary in the nearly free
electron approximation. At the boundary the 'P(+) standing wave state has a lower
energy than the free electron state while the 'P(-) standing wave has a higher
energy, leading to a band gap, Eg. Compare Fig. 7.9.
because the electrons are removed from the attractive electrostatic potential
of the ion cores. Thus the standing wave 'P( +) has a lower energy than
either travelling wave eib: or e-ib: whereas the standing wave '1'( -) has a
higher energy, and hence there is a band gap at the Brillouin zone boundary,
as sketched in Fig. 7.11. We see that the nearly free electron picture of the
origin of band gaps, in terms of Bragg reflection setting up standing waves,
is equivalent to the LCAO picture because the charge densities of the
standing waves resemble those of atomic states.
We can take the comparison of the nearly free electron and LCAO pictures
further. We have said that the valence band structures of highly covalent
materials such as Ge, Si, and diamond look quite free-electron like. In
particular the width ofthe valence band is quite well given by the free electron
result and the E(k) relation looks like the k 2 free electron band structure.
This appears very paradoxical at first because the eigenstates of a free
electron are plane waves, e ik ", which have a constant charge density. On
the other hand, in diamond the valence electron charge density is far from
constant, being concentrated along the bonds between neighbouring atoms
and almost zero in the tetrahedral holes of the crystal structure. How is it
possible that diamond can have a nearly free electron valence band structure
with such a nonuniform charge density?
The resolution of this paradox is that the kinetic energies of the standing
waves is the same as the kinetic energy of the travelling waves from which
they are made. Consider again the one-dimensional nearly free electron
model. The kinetic energies of the travelling waves at the Brillouin zone
boundaries are h 2 /(8ma 2 ). The normalized standing wave '1'( +) is
2)1/2 nx
'1'( + ) = ( ~ cos -; (7.28)
and its kinetic energy is given by
--2-
h
2
fa <'1'(+ )1-
d
2
1'1'( +» dx =
h2
--2' (7.29)
8n m o dx 2 8ma
The same result is obtained for the kinetic energy associated with '1'( -). The
important point to note here is that the standing waves that are set up by
Bragg reflection at the Brillouin zone boundaries lead to a nonuniform
charge density as we saw aboye, but they do not alter the kinetic energy of
the free electron eigenstates from which they are made. The band structure
therefore continues to look free electron-like except for gaps that are opened
at the Brillouin zone boundaries which are due to electrostatic interactions
with the ion cores.
In many metals the Fermi surface reaches certain parts of the Brillouin
zone boundary and standing waves are formed from those occupied travel-
ling wa ves at the Brillouin zone boundary and their Bragg reflected
counterparts. In such metals it is therefore quite possible for sorne degree of
directional bonding to exist. On the other hand a spherical free electron-Iike
Fermi surface indicates an absence of directional bonding.
To estimate the magnitude ofthe band gap at the Brillouin zone boundary
consider the one-dimensional periodic potential
U(x) = Uo cos(2nx/a) (7.30)
where Uo < O. If IUol is small then we may use first order perturbation theory
to find the change of energy of the normalized standing wave state '1'( + )
M( +) = <'1'( + )IUI'I'( +»
= 2Uo fa cos 2 (nx/a)cos(1.nx/a) dx

a o
Uo
(7.31)
=2'
Similarly ~E( -) = <'1'( - )1 U 1'1'( -» =- Uo/2. Therefore, the band gap is
given by
(7.32)
We see that the band gap is given by the magnitude ofthe Fourier component
of the crystal potential. In fact this is an exact result for wave functions given
by eqn (7.27) and the one-dimensional potential of eqn (7.30). The relatively
small band gaps in many materials, compared with the band widths, indicate
The pseudopotential 151
that the potential in those materials must be relatively weak. But how can
we possibly maintain that the potential felt by the electron due to the ion
core is weak, when the electrostatic potential diverges to - ro at each atomic
nucIeus? This was a long-standing paradox in the nearly free eIectron model
and it was resolved only with the introduction of the pseudopotential.
The pseudopotential
The pseudopotential was introduced to explain why the effective potential
felt by an itinerant electron in the material is weak compared with the
Coulomb potential of the ion cores. Before we outline the quantum
mechanical origin of the pseudopotential let us first give a cIassical picture.
We are concerned with the change in the total (kinetic + potential) energy
of an electron moving past an ion core, as sketched in Fig. 7.12. The electron
' - . / - Electron
Fig. 7.12 An electro n moving past an ion coreo
accelerates towards the positive charge of the ion coreo The gain in kinetic
energy cancels the decrease in the potential energy and the total energy remains
constant throughout the process. If we transform the increase in kinetic
energy into an effective repulsive potential we see that this effective potential
cancels the real electrostatic potential. In a quantum mechanical picture the
cancellation of the effective repulsive and real attractive potentials is not
complete, and the residual, weak potential is called the pseudopotential.
The quantum mechanical picture is as follows. Consider a free electron
plane wave state. The excIusion principIe requires that the free electron state
be orthogonal to the atomic core states. The core sta tes are highly localized
orbitals and the orthogonality requirement introduces many new oscillations
into the plane wa ve state in the region of the atomic core, as shown in
Fig. 7.13. Since the kinetic energy of a state is proportional to the second
derivative of the wave function the kinetic energy of the free electron state
is increased in the vicinity of the atomic core, as in the cIassical picture.
Consider now the charge density of the free electro n sta te. Far from the
atomic core the wave function is aplane wave and therefore the charge
density is constant. The rapid oscillations in the wave function at the atomic
eore depletes the charge density there, essentially because the average value
oC sin 2 x is 1/2. Thus there is an 'orthogonality hole' in the charge density
oC the free electron state in the vicinity of the eore, as shown in Fig. 7.13.
This is the excIusion principie in action: it has literally excIuded the free
electron state from the atomic coreo
'1'
fWave function)
8 p(Charge
I:V:I
Orthogonality
density)
hole
Fig. 7.13 The wave function, '1', and the charge density, p, of an electron moving
past an ion coreo The rapid oscillations in the electron wave function near the
ion core result in a reduction in the electronic charge density called the
orthogonality hole.
We show in Problem 25 how the orthogonality requirement leads to an

etrective repulsive potential, which when added to the real attractive potential
results in a residual weak potential that is the pseudopotential. In many
cases the cancellation between the real attractive potential and the etrective
repulsive potential is almost exact and this leads to the empty core
pseudopotential shown in Fig. 7.14. Beyond sorne core radius, re' the
potential due to the ion is the ionic Coulomb potential. But within the core
radius the cancellation of the Coulomb potential by the etrective repulsive
potential is assumed to be exact. The wave functions that are obtained by
solving the SchrOdinger equation in which the pseudopotential appears
rather than the true potential are not the true valence wave functions but
pseudowave functions. The true wave function oscillates in the atomic
core (because the attractive Coulomb potential of the nucleus increases
the electronic kinetic energy), whereas the pseudowave function is much
smoother. But outside the core the true and pseudowave functions converge
to the same values. In order to understand bond formation it is necessary
only to have an accurate representation of the wave function outside the
atomic coreo For this purpose the pseudowave functións are quite adequate,
and therefore we can work with the pseudopotential rather than the true
potential.
vps(r)
-------,------r-----~-------r
Ze-
--
Fig. 7.14 An empty core pseudopotential. Inside fe there is an exact cancellation

between the true attractive Coulomb potential of the ion, -Zej41tBof, and the
effective repulsive potential arising from the requirement that the state is
orthogonal to the core states.
Screening 153
The cancellationof the real potential of the ionic core depends on the
existence of appropriate states in the core which the free electron state has
to be orthogonalized too Mathematically, aplane wave can be decomposed
into components which have different angular momentum about some
axis. Each angular momentum component of the plane wave can be
orthogonalized to an atomic core state of the same angular momentum.
For example, consider Ge. The electronic configuration of the core is
ls22s22p6 3s23p6 3dlO. Thus, the core contains states with s, p, and d angular
momenta. Therefore the s, p, and d angular momentum components of a
plane wave in Ge may be orthogonalized to the atomic core states. The same
does not hold for carbono In that case the electronic configuration of the
core is ts 2. Thus the p and d components of aplane wave in diamond cannot
be orthogona1ized to the core and therefore the p and d angular momentum
components of the pseudopotential felt by a valence electron state in
diamond are strong. In diamond only the s component ofthe pseudopotential
is weak. The much greater strength ofthe p component is significant because
the valence states in diamond comprise Sp3 hybrids, and therefore they
contain p components. There are two points to note here. The first is that
there are different pseudopotentials for different angular momentum com-
ponents of the valence state. The second is that the pseudopotential for a
particular angular momentum component is strong if the core does not
contain any states with that angular momentum. It is for this reason that
clements of the first row of the periodic table and the 3d transition metal
series are characterized by strong pseudopotentials.
Screening
The Coulomb potential of a point charge q in a vacuum is well known
q
V(r)=-. (7.33)
41t6or
Suppose q is positive and we introduce it into a lump of jellium. What is the
clectrostatic potential ofthis'point charge injellium? We can see immediately
that it is not the same as it is in a vacuum because the free electrons of the
jcllium crowd around the positive intruder. The positive point charge
together with the c10ud of electrons attracted to it has an electrostatic field
that is very similar to that of a neutral atom once we are some distance from
the point chargc. We say that the point charge is screened by the electrons.
The electrostatic potential ofthe point charge decays much more rapidly with
distance r than eqn (7.33).
We can give a (very) crude estimate of the decay length of the potential
by noting that the Fermi wave vector defines a minimum wavelength for
clectronic fluctuations in the system. Since the Fermi wave vector is of order
154 Free electro n theory
inverse angstroms we therefore expect the decay length of the electrostatic
potential of the point charge in jellium to be of the order of angstroms, i.e.
comparable to a nearest neighbour bond length. This expectation is borne
out by more accurate and much more complicated analyses.
The simplest analytic treatment of screening is the Thomas-Fermi
approximation. This results in the following form for the screened potential
of the point charge q
v.: (r) = _q_e- tTF ' (7.34)

se 4nt:oT '
where k-rF is the Thomas-Fermi wave vector and is defined as follows

(7.35)
We see that the bare Coulomb potential, q/41t8or, is damped exponentially,

with a decay length of l/kTF • The reason why the density of states at the
Fermi energy appears in this expression is that the screening is effected by
occupied states just below the Fermi energy being excited into formerly
unoccupied states just aboye the Fermi energy by the bare Coulomb
potential energy. Lower Iying electron states cannot participate in the
screening beca use the exclusion principie forbids the occupation of sta tes by
more than two electrons.
Substituting eqn (7.17) for the density of states into eqn (7.35) (and taking
L 3 in eqn (7.11) to be unit volume) we obtain
2ém
k?F = - - kF (7.36)
h280
and therefore
k - 2.95 A-1 (7.37)
TF -
r, ao )1/2
(/
where r, is the radius of a sphere containing one e1ectron of the jellium on

average and ao is the Bohr radius. Since r./ao is .between 2 and 6 for metallic
densities we see that k:ri is between 1 and 2 A and is therefore comparable
to a nearest neighbour distance.
The aboye result demonstrates the extraordinary efficiency with which free
electrons in a metal screen a charge. Physically the result means that atoms
in a metal are electrically neutral. Each ion core carries around with it its own
screening electron c1oud, and the combination of the ion core + screening
c10ud is called a pseudoatom. Concepts such as ionic bonding are therefore
completely inapplicable to metals.
Screening manifests itself in many other properties of metals. In the next
section we shall see how it explains the success of the independent electron
approximation, which we have made throughout this book.
Exchange and correlation 155
Exchange and correlation

Throughout this book we have assumed that although the electrons are
charged particies we can essentially ignore this fact and regard their motion
as independent. If we acknowledged the Coulomb interactions between the
electrons we would not be able to discuss the electronic structure of the
material in terms of single electron states, as we have done, because each
state of the system would now depend on all the electrons in the system. So
how do we get away with it?
The answer is that the electrons screen each other! Each electron in the
system repels the other electrons in the system for two reasons. The first is
the ciassical electrostatic repulsion interaction. The second is the exchange
interaction which arises from the exciusion principie: e1ectrons with paralle1
spins tend to avoid each other. The result of this repulsion is that there is a
depletion in the probability of finding an electron near to a given electron,
as shown in Fig. 7.15. This depletion is called an 'exchange-correlation hole'.
In effect each electron digs a hole for itself in the electron gas charge density.
The hole contains minus one electron. So viewed from a distance greater
than the diameter of the hole the electron and its hole appear electrically
neutral. This is the reason why the electrons behave as though the Coulomb
interactions between them have been switched off.
Since the electron and its exchange-correlation hole are inseparable they
move through the system as a single entity which is called a quasiparticie.
It is the behaviour of these quasiparticies that we are really concerned with
when we talk about the electronic energy levels of a material. How big are
these quasiparticies? It should come as no surprise (after our discussion of
screening) that they are about an angstrom in diameter, which is of course
much bigger than the size of the electron.
A useful way of thinking about the concept of an exchange-correlation
hole is to imagine what happens if we bring an electron in a vacuum up to
the surface of a lump of jellium. Suppose the electron is at a distance x from
the edge of the jellium. We know from elementary electrostatics that the
p(r)
Electron
/
------~~---.~--~-------.r
Exchange-correlation hole
Fig. 7.15 The reduction in the charge density surrounding each electron is called
the exchange-correlation hole.
Metal Vacuun
........-x
-q
x
q
1
oc - -
X
- 00
Fig. 7.16 An electron of charge q at a distance x from a metal surface induces

an image charge, - q, at a distance x below the surface. The classical energy of
interaction is proportional to -1 Ix at all separations and diverges as x -+ O.
-
external electron induces positive charges on the surface of the jellium, whose
combined effect can be represented by a positive image charge at a distance
x beneatb tbe jellium edge. See Fig. 7.16. Tbe classical electrostatic energy
of attraction between tbe electron and the jellium is - e 2 /( 161tEox). (Note
that it is half of -e 2/(81tE oX) because of tbe repulsion between the induced
charges on the surface of the metal.) In fact tbis classical electrostatic result
is also the result of a fuIl quantum mechanical analysis of the interaction at
large values of x. But as tbe electron approaches the jellium edge something
must happen because the classical resuIt diverges as x -+ O. What happens
is that the wave function of the external electron starts to overlap with tbe
wave functions of the electrons in tbe jellium at separations of 1-2 A and
the classical l/x variation then starts to deviate from the true quantum
mechanical variation. There are two aspects to this. The first is that as tbe
externa! electron enters the jellium tbe classical image cbarge becomes a
depletion in the electron density surrounding the electron: this is the
correlation boleo In effect the lower limit on the wavele~gth of electron
density variations set by the Fermi wavelength ensures that x does not reach
zero, and therefore the singularity in l/x is avoided. The second aspect is
tbat the external electron becomes indistinguishable from the other electrons
in the jellium and therefore there is an exchange interaction which tends to
keep electrons of the same spin as the incoming electron away: this is the
exchange hole. The energy of the electron surrounded by its exchange-
correlation hole matches smoothly onto tbe classical image charge inter-
action outside the metal, as shown in Fig. 7.17. If a positron (a positively
cbarged electron) were brought up to the surface of the jellium tbere would
be no exchange interaction witb the electrons of the jellium beca use the
positron is a distinguishable particle. The classical negative image charge
would become tbe screening electron density surrounding tbe positron inside
the jellium.
References 157
Metal V aCUllO
1
-----oc --
X
1-2Á~
Fig. 7.17 The quantum mechanical energy of interaction between an electron

of charge q at X from a metal surface is proportional to -1 Ix at large separations
and does not diverge as x -+ O. As the electron enters the metal the 'image charge'
becomes the exchange-correlation hole surrounding the electron.
References
Drude, P. (1900). Anna/. Phys., 1, 566 and 3, 369.
Harrison, W. A. (1980). E/ectronic structure and lhe properties of so/ids. W. H.
Moruzzi, V. L., Janak, J. F., and Williarns, A. R. (1978). Ca/cu/ated e/ectronic
properties of meta/s. Pergamon, Oxford.
8
Properties of free electron metals
Fermi-Dirac statistics
We noted in the previous chapter that Drude's free electron model of a metal
failed in one principal aspect: it regarded the electrons as c1assical partic1es
obeying Maxwell-Boltzmann statistics. Sommerfeld's free electron model
recognized that the electrons obey quantum mechanicallaws. Owing to the
electron spin of 1/2, electrons are fermions and they obey Fenni-Dirac
statistics. Each quantum state of the system can be occupied by at most two
electrons, one with spin up and the other with spin down. The probability
of a state of energy E being occupied at a temperature T is given by the
Fermi-Dirac distribution formula
f(E) = 1 (8.1)
1 + exp«E - EF )/kT)
where EF is the Fermi energy. Since E F, relative to the bottom of the band,
ín metals is of order 10 eV whereas kT at room temperature is -1/40 eV
the function f(E) has the appearance of a step function at E = EF with
slightly rounded corners, as seen in Fig. 8.1. When E« EF then f(E) = 1
and the state is occupied by two electrons. For E» EF then f(E) = O and
the state is empty. The effect of temperature is to blur the occupation
probability slightly near the Fermi energy. Thus, the sharply defined
spherical Fermi surface of the free electron model at T = O K becomes only
slightly fuzzy at temperatures up to the melting point of the metal. The main
¡(E)
~------------~---'E
o EF
Fig. 8.1 The Fermi-Dirac distribution, f (E). The solid curve is for a finite
temperature and the broken line shows the step-like form that applies at O K.
Contact potential 159
point to grasp here is that it is only those states within a Cew kT oC the Fermi
energy whose occupations can be altered by variations in the temperature.
The vast majority oC occupied states are complete1y unaffected by changes
in the temperature: they remain oecupied. This is the key to understanding
the thermal properties of electrons in metals. It is related to the fact that
only electrons near the Fermi energy can take part in screening as we
discussed in the last chapter. If the exclusion principIe did not apply to the
electrons in metals they would not obey Fermi-Dirac statistics and many
oC the properties of metals would be completely different from what we know
them to be.
Contact potential
The work Cunction oC a metal i~ the mínimum energy required to move an
electron Crom deep inside the metal into the vacuum Car Crom the metal
surface. We have already noted in the previous chapter that there is a dipole
at the surCace oC a metal caused by the electron density oC the metal spilling
into the vacuum a little. This dipole lowers the potential oC electrons in the
metal relative to the vacuum level outside. The work Cunction is the total
energy difference between an e1ectron in the vacuum and an electron at the
Fermi energy in the metal, as shown in Fig. 8.2(a). The difference in total
energy includes the potential energy shift due to the surface dipole and the
exchange-correlation energy arising from the interaction with its exchange-
correlation hole inside the metal.
When two pieces of metal are joined together there is, in general, an
imbalance in their work functions. In Fig. 8.2(b) the Fermi energy, EFe , oC
metal B, is initially aboye E FA . Electrons ftow Crom metal B into metal A.
At equilibrium, Fig. 8.2(c), the Fermi energy oC the metals is constant and
a dipole is set up at the junction. The dipole effects the alignment oC the
Fermi energies by introducing a relative shift in e1ectrostatic potential which
Vacunn - r
+-
w B +- A
+-
+
+-
+-
+-
+-
+
+
+-
+-
(a) (b) (e)
Fig. 8.2 (a) To iIIustrate the work function. W. of a metal. (b) Two metals A
and B before they are brought ioto contaet: the difference in Fermi energies.
relative to the (common) vacuum level. gives rise to a contact potential
(EFB - EFA)/ e. (e) After contact a dipole is created at the interface to equalize
the Fermi levels.
160 Properties of free electron metals
is equal and opposite to (EFB - EFA)/e. The quantity (EFB - EFA)/e is called
the contact potential. Dipoles of the kind shown in Fig. 8.2(c) always occur
at interfaces between dissimilar materials.
Electronic specific heat

Maxwell-Boltzmann statístics predicts that the specific heat per electron is
3k/2. Thus if there are N electrons per unit volume then the predicted
electronic specific heat is 3Nk/2 per unit volume, which is the same as the
specific heat of a classical monatomic gas with N atoms per unit volume.
But the observed electronic contribution to the specific heat at room
temperature is usually 1 per cent of this value.
The solution to this problem was not found until it was recognized that
electrons in the metal obey Fermi-Dirac statistics. The key point is that only
those electrons - kT from the Fermi energy can be excited thermally because
all lower Iying states are filled to their maximum capacity. The strict
requirement of the exclusion principIe prevents lower Iying electrons from
being thermally excited because there are no states that are partially
occupied, and no state may hold more than two electrons. If we introduce
the Fermi temperature, TF, which is defined by kTF = E F , then only a fraction
- T /TF of alI the electrons in the metal can contribute to the specific heat.
For example, in aluminium the Fermi energy is 8.4 eV and therefore
TF = 97 500 K. Thus, at room temperature T /TF - 0.3 per cent.
Each of these -NT/TF electrons has a thermal energy of order kT.
Therefore, the total electroníc thermal energy, U, is
U ~NT kT. (8.2)

TF
Therefore, the electronic specific heat is given by
dU Nk
Ce. = dT ~ T T, (8.3)
F
which we see is linearly proportional to temperature. An equivalent estímate

is obtained by saying that the number of electrons contributing to the
electronic specific heat per unít volume is -D(EF)kT (where D(E F) is the
density of states at the Fermi energy) and each of them has a specific heat
of -k, so
(8.4)
In practice the electroníc contribution to the specific heat of a metal is seen
only at low temperatures because the phonon contribution dominates at
higher temperatures. At low temperatures the phonon contribution varies
as - T 3 , and therefore is smaller than the electronic contribution at
sufficiently low temperatures.
Electronic specific heat 181
Our 'back-of-an-envelopc' estimate of the electronic specific heat can be
done more respectably by writing down the exact expression for U and
differentiating it with respect to temperature. The internal energy, Ey, of the
electron gas is given by
Ey = LO ED(E)f(E) dE (8.5)
where f(E) is given by eqn (8.1) and D(E) is the total density of states. The
only quantity in eqn (8.5) that varies with temperature is the Fenni-Dirac
distribution function, f(E). After a little manipulation we get
(8.6)
which is shown in Fig. 8.3 as a function of energy. The function (of/oT) is

symmetric about the Fermi energy. The shape of (of/oT) as a function of E
is clear: electrons that were occupying states just below EF are thermally
excited into states just aboye E F. Substituting this expression for (of/oT)
into the electronic specific heat Ce. = (oEyjoT)y we obtain
2
Ce. = roo ED(E) (E - E p )j(kT ) dE. (8.7)
Jo 4COSh 2 (E - EF)
2kT
Since (ofjoT) varies significantIy with E only near E = EF and since D(E)
varies only slowly with E as E1/2; we may replace D(E) by D(E F) and take it
outside the integral. Introducing the change of variable x = (E - EF )/(2kT)
al
aT EF
,
,
:"
,
,,
,
v,,
,
Fig. 8.3 The partial derivative of the Fermi-Dirac distribution function with
respect to temperature, plotted as a function of energy.
we can express Ce. as follows
a) X
Ce. = D(EF)k + E F) dx.
f
_ EF/(2kT)
(2kTx
cosh
2
X
(8.8)
This integral can be simplified straight away by noting two points. The first
is that E F/(2kT) is a number of order 100. Because of the cosh 2 x in the
denominator, which rapidly tends to infinity as x increases, we may replace
the lower limit of the integral by - OC) with almost no error. The second
point is that x/cosh 2 x is an odd function and therefore its integral from
- OC) to + OC) is zero. We are left with
(8.9)
and using the standard integral result
(8.10)
we finall y obtain
(8.11)
This result vindicates our back-of-an-envelope estímate, eqn (8.4), which

difIers from the 'exact' result by only a factor of rc 2 /3.
Note that Ce. is proportional to the density of states at the Fetmi energy.
This is the only material parameter in eqn (8.11). In those materials, such
as Bi, which have a low D(E F ) the electronic contribution to the specific heat
is very small. In a nonmetal the electronic contribution to the specific heat
decreases exponentially as the band gap increases. In most semiconductors
and insulators it is negligible.
Electrical conductivity of metals

One of the great successes of Drude's free electron model was the derivation
of Ohm's law for a metal. In eqn (4.21) we derived the equation of motion
for the wave vector k in the presence of an electric field ~
hdk
-e~=-- (4.21)
2rc dt
which can be solved as follows
k(t) _ k(O) = _ 2rce~t . (8.12)

h
Electrical conductivity of metals 163
Since the right hand side is independent of k we see that all states, k, in the
system evolve in time in precisely the same way. This means that in k-space
the Fermi sphere moves rigidly in the direction of the electric field ~. But of
course this is an oversimplified treatment because it ignores the fact that
electrons are scattered by defects in the metal (vacancies, interstitials,
impurities, dislocations, grain boundaries, etc.) and by thermal vibrations of
the crystal structure. Drude recognized this and introduced the average time,
t, between collisions which Drude thought were collisions between the
electrons and the ion cores. In that case the average change in wave vectors,
Dk, between successive collisions is obtained by setting t = t in eqn (8.12).
The velocity of the electron is just
Mk e~r
11=--= --o (8.13)
21tm m
If there are n e1ectrons per unit volume then the current density is
ne 2 r
j= -nev=-~. (8.14)
m
The proportionality between the current density and the electric field is
Ohm's law. The electrical conductivity, (1, defined by j = (1~, is given by
(8.15)
In pure copper at 4 K the collision time, r, is about 2 ns. The mean free path
between collisions is - vFr, where V F is the velocity of electrons at the Fermi
energy. (The reason we use VF here is that it is only electrons near the Fermi
energy which can be scattered into other states owing to the exc1usion
principie again.) Now VF in Cu is about 1.56 x 106 m S-l. Therefore the
mean free path at 4 K is about 3 mm. This means that each electron travels
approximately 10 7 lattice constants at 4 K in pure Cu before it is scattered.
Therefore, unless the electrons are accurately aligned between the rows of
atoms in the crystal, the scattering mechanism cannot be off the ion cores
as assumed by Drude. Indeed we have already seen that in a perfect crystal
Bloch's theorem ensures that an electron may propagate indefinitely with no
scattering taking place. The source of the scattering at 4 K in the pure copper
is the presence of a small number of defects and occasional thermal
vibrations. Indeed, at 300 K the mean free path in pure Cu is much reduced
by thermal vibrations to about 300 A, i.e. 10- 3 what it was at 4 K. Thus
although Drude's model correctly predicted Ohm's law the nature of the
scattering process which gives rise to the finite conductivity is quite different
from what Drude envisaged.
Thermal conductivity of metals

From the kinetic theory of gases we know that the thermal conductivity of
a system of independent particles may be expressed as follows
K = tCvl, (8.16)
where C is the specific heat per unit volume, vis the average particle velocity,
and I is the mean free path of the particles between collisions. The basis of
this formula is that heat is exchanged between particles only during
collisions. We may apply this formula to the independent electrons in the
metal with a few minor modifications to take account of the fact that the
exclusion principie applies to the electrons. We shall use the electronic
specific heat, C.I , which we derived in eqn (8.11). For the velocity v we shall
use the Fermi velocity because it is only electrons near the Fermi energy
that may change their state (i.e. be scattered) and absorb or release energy.
For I we shall of course take the mean free path of the electrons at the Fermi
energy, which is VFT. Therefore the electronic thermal conductivity becomes
(8.17)
Using EF = mvU2, and eqn (7.17) for the density of states, eqn (8.17) can
be rewritten as
(8.18)
where n is the electron density.

In pure metals the electronic contribution to the thermal conductivity
dominates over the contribution from lattice vibrations at all temperatures.
But in disordered alloys where the mean free path is much reduced the
electronic contribution is comparable to the lattice vibration contribution.
The Wiedemann-Franz law

Perhaps the crowning achievement of Drude's free electron theory was to
explain the Wiedemann-Franz law. This empiricallaw states that the ratio
of the thermal conductivity to the electrical conductivity is proportional to
temperature. Furthermore, the constant of proportionality, which is called
the Lorentz number, L, is independent of the type of metal.
Using eqn (8.15) for the electrical conductivity, (J, and eqn (8.18) for the
thermal conductivity, K. I , we have
K.I
2 2
= 1[ nk TT/3m =
2
1[2 (~)2 T = LT (8.19)
(J ne T/m 3 e
The Hall effect 165
where
(8.20)
is the Lorentz number. Experimental Lorentz numbers for elemental metals

vary between about 2.30 x 10- 8 and 3.20 x 10- 8 W ohm K -2 for tempera-
tures between O and 100°C. Note that in eqn (8.19) we have assumed that
the collision times for the e1ectrical and therrnal conductivities are the same.
This assumption does not hold at low temperatures and the Lorentz number
for pure copper near 15 K is an order of magnitude smaller than eqn (8.20).
The Hall effect

If an electric current is passed along the x-direction through a conductor in
which there is a magnetic field along the z-direction then an electric field is
set up along the y-axis, at right angles to both the applied electric field and
the magnetic field. This is the Hall effect. To understand this effect we use
the equations of motion of the electron in an electric field (eqn (4.21» and
a magnetic field (eqn (4.31». The equation of motion is as follows
m( -do + o)
- = -e(, + o x B) (8.21)
dt t
where , is the total electric field and B is the magnetic field. The terrn o/t
is a 'frictional force' due to the collisions. With B along the z-axis, and in
the steady state where do/dt = O, eqn (8.21) becomes the following set of
three equations
mv,,/t = -e,,, + eV'B)
mv,/t = -e" + ev"B (8.22)
mv./t = -e,•.
If current cannot fiow along the y-direction then there must be an electric
field component " such that v, = o. This is the Hall field and it is given by
the second of eqn (8.22)
" = v"B. (8.23)
But from the first of eqn (8.22), and v, = O, we have
v" = -(er/m),,,, (8.24)

and therefore
";e,, = -eBt/m. (8.25)
Introducing the Hall coefficient, R H , as "/(j,,B), and using j" = (nét/m),,,,

we obtain
RH = (eBT/m)~x =-L (8.26)
(néT/m)~xB ne
Thus the Hall coefficient is negative for free electrons. The use of the Hall
effect is that it pro vides a means of measuring the carrier density n and
ascertaining whether the carriers are electrons or holes. If the carriers are
holes, as they are in Al and In, then the Hall coefficient is positive. As the
carrier concentration decreases so the Hall coefficient rises. In •semimetals'
such as Bi, As, and Sb the density of states at the Fermi energy is very low and
therefore the Hall coefficient is very large.
The cohesive energy of simple metals and its volume

dependence
The sp-bonded metals of groups 1, 11, and 111 are usually called simple metals.
They are nearly free electron metals with free electron-like band structures
and small gaps at the Brillouin zone boundaries. Their pseudopotentials are
weak (except for those elements in the first row of the periodic table) and
therefore we should be able to treat the effect of the ion cores on the cohesive
properties of the metals through perturbation theory. This is indeed the case,
but the story is a rather long and tortuous one starting from free electron
theory. Nevertheless, it is a story well worth telling, if only in outline, beca use
it shows how difficult it is to make quantitative predictions of cohesive
energies and the volume dependence of the energy of the metal. Throughout
this section we ignore the effects of temperature and assume T = OK.
The story starts with free electron theory applied to a jellium model of
the metal. The average kinetic energy is obtained by integrating the kinetic
energy times the density of states up to the Fermi energy, and multiplying
by 2 because each state is occupied by two electrons. Using eqn (7.17) for
the density of sta tes we obtain
(8.27)
where N is the total number of electrons and V is the volume of the

metal. Thus the average kinetic energy is repulsive at all volumes: it favours
an increase in volume beca use the kinetic energy of the electrons will
decrease.
We have already noted in Chapter 7 that each electron in the jellium digs
a hole in the electron density due to Coulomb repulsion and the exchange
interaction between electrons with parallel spin. Since this •exchange-
correlation' hole reduces the amount of time electrons spend very near to
each other it reduces the Coulomb repulsion between the electrons on
The cohesive energy of simple metals 167
2.23Á
~r--+--~~------~~
Fig. 8.4 Eog, given byeqn (8.29), plotted as a function of ' •.
average. Therefore the exchange and correlation contributions to the total

energy of the electron gas are negative.
It is convenient to switch to a new variable, r" which is the radius of a
sphere containing one electron on average
41t 3 V
-r =- (8.28)
3' N'
It is also convenient to work in atomic units (the unit of length is the Bohr
radius (0.529 A), the unit of energy is the Rydberg (1 ry = 13.6 eV)). The
total energy per electron in the electron gas of the jellium is then
E.s = 2.21/r; - 0.916/r, - (0.115 - 0.0313 In r,). (8.29)
The first term represents the average kinetic energy, eqn (8.27). The second
term is the exchange energy and the final term is the correlation energy.
Equation (8.29) is plotted in Fig. 8.4. It is seen that there is a minimum at
r, = 4.2 au = 2.23 A with a binding energy per electron of 0.16 ry or 2.2 eVo
To distinguish between the metals, or altematively to account for why they
have ditrerent values of r., we must proceed to the next level of approximation
in which the uniform positive background of the jellium is replaced by the
ion cores.
Each ion core is now assumed to be a point positive charge of Ze. The
electrostatic interaction between the point ions and the electron gas intro-
duces an attractive term, E. pl ' To estimate the magnitude of this term we
use the atomic sphere approximation, which tums out to be an excellent
approximation in close packed metallic structures. A sphere is centred on
each point ion with radius equal to the Wigner-Seitz radius, rws, given by
eqn (1.1). (Recall that the volume of the Wigner-Seitz sphere is equal to the
atomic volume.) Comparing eqn (1.1) and eqn (8.28) we see that
•
'ws - Zl/3r.
- S' (8.30)
Eaeh sphere is eleetrieally neutral beeause it eontains an ionie eharge of + Z
and Z electrons. The spheres may overlap a little, but that is ignored. The
eleetrostatie energy per atom is thus the eleetrostatic energy contained within
each sphere, which is as follows
E i
3Z
= ---
frws 41tr 2 Ze 2
--
3Z 2 e 2
dr = - - - - (8.31 )
.p 41tr~s o 41t~or 81t~orws
In atomic units this formula becomes
(8.32)
The eleetrons within each sphere repel each other and this gives rise to a
positive contribution, E•• , which is as follows
1.2Z 2
E •• = - - (8.33)
rws
The trouble with the point ion model is that it does not account for the
repulsion between the core and valence electrons, which gives rise to the
orthogonality hole in the valence electron charge density. The point ion
model is therefore an overestimate of the cohesive energy. In the next stage
of the caIculation the point ion model is replaced by a pseudopotential which
describes the effect that the point ion model has overlooked. The simplest
is the empty core pseudopotential, shown in Fig. 7.14. According to this
model there is an exact cancellation between the attractive Coulomb
potentiaI and the repulsive core orthogonality contribution within sorne core
radius re' for all angular momentum components of the valence wave-
function. In atomic units the pseudopotential has the following form
O for r < re
V. ()
r = { (8.34)
p. -2Z/r for r > re'
The electron-ion attractive interaction is now reduced from E. pi to E.i
given by
(8.35)
The cohesive energy per atom is obtained by adding Z times the electron
gas contribution per electron, eqn (8.29), to the electron-electron repulsion
per atom, eqn (8.33), and the electron-ion electrostatic attraction per atom,
eqn (8.35)
(8.36)
The equilibrium atomic radius, rws , is determined by minimizing the cohesive
energy with respect to rws , which yields an equation relating rws and the
Structural energy differences 169
core radius re
2 1 0.102 0.OO35rws 0.491
(re/rWS) =- +~ + - 1/3 • (8.37)
5 Z Z Z rws
This equation can be used to determine an effective empty core radius re
using the experimentally known values of rws. When this is done it is found
that the core radii increase with atomic number in a given group of the
periodic table, as we would expect. It is interesting to note that only Na has
a value of r, close to the free electron gas value of 4.2 a.u. This indicates that
although the pseudopotential is weak in simple metals its effect on the
cohesive energy and the equilibrium volume is crucial.
The expression we have derived for the cohesive energy of the metal, eqn
(8.36), depends only on the density of the material, for a given core radius.
Therefore, it gives zero elastic shear constants and it does not distinguish
between the energies of different crystal structures of the material at the same
density. The vast majority of the cohesive energy of the metal is completely
independent of the crystal structure it adopts, for a given density. This sad
fact means we have to work even harder to discriminate between the energies
of rival crystal structures for a metal (f.c.c., b.c.c., h.c.p., etc.) at constant
density. It is these energy differences which determine the energies of
important defects in a metal, such as stacking faults. How this is done is
outlined briefly in the next section.
Structural energy differences

In a simple metal the pseudopotential of the ion cores is weak. We may
apply perturbation theory by regarding the uniform electron gas of the
appropriate density as the unperturbed system and the ionic pseudo-
potentials as the perturbing potential. In the previous section we carried this
out to first order in the pseudopotential. To distinguish between the energies
of different atomic arrangements of the metal at constant density we must
extend the analysis to second order in the pseudopotential. When this is
done the cohesive energy per atom may be written in the following physically
transparent form
1
Eeoh = E 1(r,) + - L tjJ(R, r,). (8.38)
2R .. O
The first term, E 1 (r,), is density dependent and structure independent and
accounts for more than 90 per cent of the cohesive energy of the metal. The
second term is a sum of pair potentials, and it is this term that distinguishes
between the energies of different crystal structures at the same density. The
pair potentials are also dependent on the density. In the section on screening
of the previous chapter we discussed the concept of a pseudoatom: the
combination of each ion core and its own screening electron c10ud behaving
as a neutral object. If an ion is displaced in the metal then the resulting
charge redistribution is simply obtained by a rigid displacement of the whole
pseudoatom. The energy change accompanying such a displacement is the
change in the electrostatic energy between the displaced pseudoatom and aH
the other fixed ions. This is the energy change that is described by the sum
of pair potentials in eqn (8.38). We identify each pair potential as the
potential energy of the ion in the electrostatic field of a pseudoatom at R.
It consists of the Coulomb repulsion between the ions, plus the attraction
of the first ion to the screening c10ud of the second. Since the screening
efficiency is dependent on the electron density we find that the pair potentials
are density dependent.
The pair potentiaIs in eqn (8.38) look quite unlike Morse or Lennard-
Jones potentials. Since most of the cohesion of the metal is provided by the
structure independent term there is no need even for the pair potentials to
be attractive at first neighbours. Pettifor and Ward (1984) derived analytic
forms for the pair potentiaI
2z2 3
cf¡(R, r.) = - L A. cos(k.R + IX.) e- K
•
R (8.39)
R .=1
where k., IX., and K. are density dependent parameters. Potentials for Na,
Mg, and Al are shown in Fig. 8.5. It is seén that the potentials are osciHatory
and that the osci1lations are exponenually damped. The oseillatory
Na
I 2 3
V kFRI1t
10 Mg
~ 5
~
-e-
O /"'00
2 3 kFRI1t
IV
-5
Al
10
~ 5
~-e- O
I
\ 2""" 3 kFRI1t
-5
Fig. 8.6 Pettifor-Ward pair potentials for (a) Na, (b) Mg, and (e) Al, plotted as
a function of kFR/1t. From Pettifor and Ward (1984).
Reference 171
behaviour is a wave mechanical effect stemming from the interference of
waves scattered by the ion cores. They are called Friedel oscillations. Note
that in Al the first neighbours are on a repulsive part of the potential. The
variation of tP(R, r,) across the periodic table is determined by just two
parameters: the electron density, which determines r, and the effective radius
of the ion core re' The valence Z scales the amplitude of the potential but
does not affect its shape.
Potentials such as those described in eqn (8.39) have been used to model
defects in simple metals by computer simulation. Their use is limited,
however, to situations where the defect is not associated with a large change
in the local electron density. For example, these potentials cannot be used
to model the atomic structures of surfaces or the vacancy formation energy.
At these defects the electron density changes dramatically and therefore the
perturbation which these defects represent to the electron gas in the metal
is large and cannot be treated adequate1y by second order perturbation
theory: higher order terms are important. In these situations we are obliged
to use Finnis-Sinclair type potentials, which are described in the next
chapter.
Reference
Pettifor, D. G. and Ward, M. A. (1984). Solid State Commun., 49, 291.
9
The transition metals
The transition metals

The transition metals are not simple metals. Their densities of states are
characterized by a partially mled narrow d band, superimposed on a broad
free electron-like s-p band, as shown schematically in Fig. 9.1. As we move
across the series the d band is gradually being filled, until at the noble metals
Cu, Ag, and Au it is mled. We shallsee in this chapter that most of the
properties of the transition metals are characterized by the filling of the d
band, and that for many purposes we may neglect the sp electrons as a first
approximation. This was first recognized by- Friedel in his simple d-band
model which we describe in the next section.
The width of the sp band is in marked contrast to the narrowness of the
d bando The narrowness of the d band, especially in the 3d series, is a
consequence of the relative constriction of the d orbitals compared with the
outer s and p orbitals, as discussed on p. 12. Because the d orbitals are
constricted they do not overlap very much with orbitals on neighbouring
atoms and therefore the hopping integral between d orbitals is small.
This indicates that an LCAO description of the d band is more appropriate
than a free electron description because it would take many plane waves
to describe the spatially constricted d orbitals. Thus we shall return
to the LCAO formulation that we discussed in Chapter 2-6 to discuss the
d bando
The 3d series is complicated by the appearance of magnetismo We shall
not say any more about magnetism in this book except to point out now
why it arises. In a partly-filled shell in a free atom the exchange interaction
D(E)
..... d band
s-p band
E
Fig.9.1 Schematic density of states for a transition metal. showing a bread. free
electron-like. sp band with a superimposed narrew d bando
Tha transition matals 173
between electrons favours the parallel alignment of electron spins. This is
the basis of one of H und's rules. Within the free atom the exc1usion principie
requires that electrons with parallel spin keep out of each other's way, which
reduces the Coulomb repulsion between them. (The energy gain is of the
order of an electron volt.) But in the solid these electrons may enter giant
molecular sta tes and thereby lower their kinetic energies. If the electron spins
remain aligned in the solid the exc1usion principie forbids more than one
electron from occupying each molecular state. This forces electrons to occupy
higher energy states than would have been occupied if the spins were not
parallel. There is therefore a trade-off between the kinetic energy of the
electrons, which favours no spin alignment, and the exchange interaction
which favours spin alignment and hence magnetismo In most materials the
kinetic energy dominates, but if the band is narrow the energy cost in
occupying higher Iying states can be compensated by the energy gain from
the exchange interaction favouring spin alignment. Therefore, the occurrence
of magnetism in the 3d series is a consequence of the narrowness of the 3d
bando
One of the central issues we shall be concerned with in this chapter is the
explanation of trends in properties across the transition metal series.
In particular we would like to understand the approximately parabolic
variations of the cohesive energy, the bulk modulus and the equilibrium
atomic volume across the series (see Fig. 9.5). These empirical trends are
particularly clear in the 4d and 5d series where magnetism does not
complicate the picture. Similarly in the 4d and 5d series the crystal structures
of the elements change in remarkably well defined ways as the d band is
fi1led (see Table 9.1).
We would like to understand these variations in crystal structure, or at
least the factors controlling them. M uch of this chapter is based on Pettifor
(1987).
Tabla 9.1 Crystal structures of the 4d and 5d transition metals

4d series
y Zr Nb Mo Te Ru Rh Pd Ag Cd
h.e.p. h.e.p. b.e.e. b.e.e. h.e.p. h.e.p. t.e.e. t.e.e. t.e.e. h.e.p.
5d series
La Ht Ta W Re Os Ir Pt Au Hg
hex. h.e.p. b.e.e. b.e.e. h.e.p. h.e.p. t.e.e. t.e.e. t.e.e. rhomb.
174 The transition metals
The Friedel model

Let us assume that the trends in the cohesive energy across the transition
metal series are due to the filling of the d bando Friedel's model for the d
band is to represent the density of states by a rectangular forro shown in
Fig. 9.2. The rectangular density of states has a width W, a centre of gravity
Ed, and since it must contain five states the height of the rectangle is 5jW.
The bond energy (see eqn (3.71» of the d band is then
E bond = 2 fEF (E - Ed) ~ dE (9.1)

",-W/2 W
where the factor of 2 is for spin degeneracy (i.e. each state is occupied by
two electrons). The evaluation of this integral is straightforward
(9.2)
where Nd is the number of d electrons per atom

5 10
f
EF
Nd =2 - dE =- (E F - Ed + Wj2). (9.3)
",-W/2 W W
The bond energy, eqn (9.2), is plotted in Fig. 9.3. It is seen that it varies
D(E)
Fig. 9.2 Friedel's rectangular density of states, D(E), for the d band of a
transition metal.
Ebond
Fig. 9.3 The bond energy in Friedel's model as a function of the number, N d ,
of electrons occupying the d bando
The Friedel model in the second moment approximation 175
parabolically across the transition metal series, in agreement with the
variation of the cohesive energy shown in Fig. 9.5. This result certainly
supports the assumption that the trends in the cohesive energy across the
transition metal series are associated with the filling of the d bando At the
ends of the series, where the d band is empty or full, cohesion is provided
by electrons in sp states. However, we have not said anything about the
repulsive contribution to the binding energy of the solid, which is necessary
in order to maintain equilibrium at sorne volume per atom. This we do in
the next section.
Tha Friadal modal in the sacond moment approximation

In order to predict an equilibrium volume for the solid and an equilibrium
cohesive energy there must be a balance between the attractive bond energy
contribution and a short range Jepulsion. If two atoms are brought very
close together they repel each other because the exclusion principie keeps
electrons in the same equilibrium state apart and because the electrons repel
each other Coulombically. Since this is a short range interaction, involving
primarily the two atoms concemed, we may represent it by a short range
pair potential. The model we now have of the cohesive energy of a transition
metal consists of the attractive bond energy, represented by eqn (3.71) and
a sum, E,op, of repulsive pair potentials acting between the atoms
(9.4)
For the bond energy we use eqn (9.2). To relate the width, W, of the band
to the local atomic environment we use the second moment of the density
of states, as described on p. 68. The second moment per d orbital, p.(2), is
given by the moments theorem and it is equal to zp 2 , assuming nearest-
neighbour hopping only, where z is the coordination number and p is the
average hopping integral between d states. On the other hand the second
moment of the rectangular density of states, per d orbital, is given by
1 W2
f
"+W'2
p.(2) = (E - ed)2 -dE = - . (9.5)
•.,-W/2 W 12
Therefore,
(9.6)
As noted on p. 69 we see that the bandwidth is proportional to the square
root of the coordination number and the magnitude of the hopping integral.
We now need a functional form for the hopping integral P as a function of
the interatomic separation, R. Since atomic orbital s decay exponentially we
expectp(R) - exp( -KR) where K is an unknown constant. We also assume
that P is proportional to Nd , and we write P(R) in the following form
(9.7)
where b is an unknown constant. Since the repulsive potential originates
Crom the overlap oC the charge densities on adjacent atoms it will vary as
the square oC p. ThereCore we express the repulsive pair potential in the
Collowing form
(9.8)
where a is another unknown constant. The cohesive energy per atom becomes
Z 2 -2.R W
E.oh = "2aNd e - 20 Nd(lO - Nd) (9.9)
and substituting eqn (9.6) and eqn (9.7) Cor the bandwidth W we obtain
(9.10)
The cohesive energy is plotted in Fig. 9.4 as a Cunction oC the bond length
R. The minimum in this curve determines the equilibrium bond length and
hence the equilibrium volume per atom for a given coordination number.
In Problem 29 you are asked to show that the equilibrium bond length is
given by
10a(z)1/2 }
R =K- 1 1n { (9.11)
o 31/2b(10 - Nd )
the equilibrium cohesive energy is given by

2
E~oh = 3b [Nd(lO _ N d)]2 (9.12)
200a
the bandwidth is given by
3b 2
W= -Nd(lO- Nd ) (9.13)
Sa
Fig. 9.4 The cohesive energy, given by eqn (9.10) plotted as a function of the
bond length R.
Tha Friadal modal in tha sacond momant approximation 177
and the bulk moduIus is given by
2(2 1/2 ),,2
B= E~oh' (9.14)
9R o
The experimentally observed equiIibrium atomic voIumes of the 4d series
are fitted by eqn (9.11) if vaIues of the inverse decay Iength " vary almost
linearly across the series
(9.15)
The ratio b2/a = 2/3 eV is obtained by fitting eqn (9.13) to the experimental
bandwidth of approximateIy 10 eV for Mo, for which Nd = 5. Fitting the
observed Wigner-Seitz radius for Mo of 1.545 A in eqn (9.11) gives
a/b = 18.0. Thus we obtain a = 216 eV and b = 12 eV for the 4d series.
Having thus fitted a, b, and " we pIot the predicted and experimental vaIues
of the bandwidth, Wigner-Seitz radius, cohesive energy, and buIk moduIus
in Fig. 9.5. The 'experimental' values of the bandwidth are taken from first
principIes caIculations.
y Zr NbMoTcRuRhPd Y ZrNbMoTcRuRhPd
eTheory 4
x Expl
l';><., 'lo- ___ .....,

I "\
I ,
I ,
i ,
le
o~~~~~~~~~~~~~~~~o
3 S 7 3 S 7
Nd Nd
Fig. 9.6 Predicted variations in properties across the 4d transition metal series
basad on the Friedel model in the second moment approximation compared with
experiment (from Pettifor 1987). Top left: the bandwidth, W (eqn (9.13». Top
right: the equilibrium bond length, Ro in Bohr (eqn (9.11». Bottom left: the
cohesive energy, ~oh (eqn (9.12». Bottom right: the bulk modulus, B, in 1011 Pa
(eqn (9.14».
The trends aeross the series are certainly well reproduced by this seeond
moment mode\. However, the predieted eohesive energies are about 2 eV too
small. This is primarily beeause we have ignored the sp band and the faet
that hybridization between the sp band and the d band ean inerease the
eohesive energy. But sinee this hybridization eontribution is almost eonstant
aeross the series it may be neglected Cor our present purpose oC predieting
trends.
A surprising feature oC this seeond moment model is that the predieted
eohesive energy, eqn (9.12), is independent oC the eoordination number, z.
This means the predieted eohesive energies are independent oC the erystal
strueture, and the predicted shear elastic constants are zero! We shall see
later in this ehapter that this result is eonsistent with the faet that the
energy differences between different crystal structures, in a nearest-neighbour
hopping model, are determined by higher moments than the second. In
practice this embarrassment (for such it is) is overcome by extending the
range oC the hopping integrals to beyond first neighbours, and introducing
wiggles into their Cunctional forms at appropriate places. But this Cudges the
issue because it does not address the physics of what controls the structural
energy differences, namely the higher moments.
For a given number, Nd , of d electrons, eqn (9.11) predicts that the
equilibrium bond length increases with the coordination number, Z. Since
the bond energy per atom is proportional to Z1!2 (see eqn (9.2» the bond
order is proportional to l/z'!2. On the other hand, the repulsive pair potential
contribution is independent of the coordination number. It follows that
as the coordination number increases the cohesive contribution to each
bond energy decreases and the repulsive contribution then becomes more
dominant. Thus the bond length increases. If nb is the number oC electrons
contributed by each atom per bond then
(9.16)
If nb = 1 then each bond would be 'saturated' beca use it would contain two
e1ectrons, one Crom each atom. Let R, be the corresponding length oC the
saturated bond. Then, setting z = Nd in eqn (9.11) we have
R - K
-1 1n { lOa(Nd )1!2 }
(9.17)
,- 3 112 b(10 - Nd )
and eqn (9.11) may be rewritten as follows

1
Ro = R, - -ln(nb) (9.18)
2K
which is identical in Corm to an empirical relationship first found by Pauling.
Since the coordination number in transition metals is high, nb is always less
than 1 and therefore the bond length is always greater than the saturated
Finnis-Sinclair potentials for computer simulations 179
value R,. At a free surface the coordination number is reduced and therefore
the equilibrium bond length is reduced compared with the bulk value. This
explains why there is a contraction in the spacing of layers parallel to a free
surface of a transition metal. Within the surface layer the atomic spacings
are imposed by the bulk, but since the bonds within the surface layer would
actually like to be shorter, they are in tension. This is the physical origin of
surface stress.
At the risk of labouring the point let us emphasize the physical import of
eqn (9.18). The dependence ofthe equilibrium bond length on the coordina-
tion number arises from the unsaturated nature of the metallic bond. By
contrast, chemists can talk about the length of a C-C single bond (this is a
saturated bond) as a transferable quantity from one molecule to another.
For example the C-C bond length in diamond is almost the same as it is
in ethane (CZH6). However, even the C-C bond length is not independent
oC the local environment oC the bond. For example, chemists talk about
single, double, and triple bonds between C atoms, depending on the number
of neighbours each carbon atom has. So when we saya C-C bond has a
certain length and energy, we are reCerring to a particular bonding configura-
tion , such as tetrahedral, planar or linear.
In a metal there are not enough electrons available to saturate all the
bonds. It is not possible to isolate one bond in the metal and regard its
energy and length as transCerable to other environments. In other words,
whereas the properties oC saturated bonds are transferable from one saturated
environment to another similarly saturated environment, the properties oC
unsaturated (i.e. metallic) bonds are not. The equivalent realization of this
basic feature oC metallic bonding in simple metals is the purely density
dependent contribution to the cohesive energy in eqn (8.38). This dominant
contribution depends on the arrangement of all atoms in the system and it
is not possible to break it down into terms that may be assigned to individual
bonds. These considerations are at the heart of the metallic bond.
Finnis-Sinclair potentials for computer simulations of

transition metals
The second moment approximation we described in the last section has been
used by Finnis and Sinclair (1984) to suggest a Cunctional form for
interatomic potentials in transition metals that are then fitted to empirical
data such as cohesive energies and elastic constants. By extending the range
oC the potentials to beyond first neighbours it is possible to stabilize a
desired crystal structure. These potentials are simple to understand and
simple to use and thereCore they have been used widely by computer
modellers. A huge range of computer simulations of the structures and
properties oC deCects and processes in transition metals has been carried out
with these potentials, or with the closely related 'embedded atom potentials'
of Daw and Baskes (1984). As the field of atomistic simulations is now one
of the major areas of growth in materials science, not least because of the
increased availability and power of computers, we should give a brief account
of these popular potentials.
The goal of any interatomic potential is to represent the potential energy
of a collection of atoms as a function of the atomic coordinates. By
differentiating the potential energy with respect to the coordinates of a given
atom we can obtain the force acting on it due to alI the others. We can then
minimize the potential energy of the system by displacing the atoms
incrementally until the forces on all atoms are zero. In that way we obtain
the relaxed atomic structure of a defect (at O K), such as a surface
or dislocation and its energy. Altematively we can introduce temperature
into the system by giving the atoms kinetic energy and letting the positions
and velocities evolve according to Newtonian equations of motion for each
atom. In this way we could simulate, forexample, a phase transformation
or the formation of a radiation cascade or diffusional transport or dislocation
emission from a loaded crack tip, a11 at the atomic scale. With current (1992)
desktop computers it is possible to model up to a million atoms. By the tum
of the century this number could increase by 1-2 orders of magnitude.
The starting point for Finnis-Sinclair (FS) potentials is eqn (9.4) for the
cohesive energy
(9.4)
The repulsive energy, Erep , is again represented by a sum of repulsive pair

potentials. The bond energy is again given by eqn (3.71), but now we express
the total density of states, D(E), as a sum of local densities of states
Ebond = 2 tfEF
(E - 1l¡)d¡(E) dE. (9.19)
Here the centre of gravity of the local density of states, d¡(E), at site i is Il¡.
The reason we use local densities of states is simply that we wish to describe
imperfect crystals, or even amorphous solids or liquids, where the atomic
environments differ from one atom to the next. The local environment
dependence is contained in the local density of states.
To capture the dependence ofthe local density of states on the local atomic
environment we use the moments theorem. The zeroth moment of d¡(E) is
always 5 because every local (d band) density of states must contain five
states. The first moment of d¡(E) is always zero
(9.20)
Finnis-Sinclair potantials for computar simulations 181
The second moment is given by
JI!2) = f"", (E - 8Yd¡(E) dE = 5 j~¡ p2(r¡j) (9.21)
where p2(r¡j) is an average square of the hopping integrals between atomic

d sta tes on atoms i and j
2
p2(r¡j) = ddu (r¡j) + 2ddn;(r¡j) + 2dd(P(r¡j) (9.22)
and ddu, ddn, and ddo are hopping integrals between d states as will be
explained below.
In the second moment approximation we do not use any higher moments
of the local densities of states. The question now is what form do we choose
for the local density of states d¡(E) which has JlIO) = 5, JI?) = 0, and
Jl1 2) = 5 Lj,.¡ p2(rij)'? There is an infinity of possible forms for d¡(E) but it
tums out that ir does not matter which one we choose. For the purposes oC
illustration we choose the following Gaussian form
d¡(E) =
5
(2) 1/2 exp -
(E - 8¡)2)
(2) (9.23)
(2nJl¡ ) 2J1¡
which is sketched in Fig. 9.6. The centre of gravity oC the Gaussian is e" as
required by JlP), and the width oC the Gaussian scales with (JlI 2)1/2. Putting
eqn (9.23) into eqn (9.19) Cor the bond energy the contribution to the total
bond energy from atom i is as Collows
d,(E)
--~~----~------~~E
fig. 9.6 The Gaussian local density of states described by eqn (9.23). The centre
of gravity is t, and the width is proportional to (1I)2)} 1/2.
So far so good. But we have not yet talcen into account the fact that we are
talking about a metal and that each atom must be charge neutral because
screening is exceedingly efficient in a metal. This is achieved by adjusting
the centre of gravity, S¡, of each local density of states. The physical picture
behind this is the following. Suppose atom j has more than its quota of
electrons, i.e. more than Nd e1ectrons. The excess electronic charge on atom
j makes it a repulsive site for further electrons. This means the electrostatic
potential at site j is higher than other sites which have fewer e1ectrons. The
increase in electrostatic potential of site j raises the on-site energy sj' which
results in electronic charge spilling off site j. The on-site energies are adjusted,
either upwards or downwards, on all sites until each atom has Nd electrons.
Of course in the perfect crystal all electrons are automatically neutral but
in a defective environment sites become nonequivalent and then there is a
tendency for charge to be piled up on sorne atoms and removed from others.
The adjustment of the on-site energies is then essential to ensure that all
atoms are restored to charge neutrality.
Each atom is electrically neutral if
(9.25)
and substituting eqn (9.23) for d¡(E) we obtain, after sorne manipulation,
(9.26)
where
EF - S¡
L·=--=~~ (9.27)
• (2JLI 2)1 /2 .
It follows from these two equations that in order for each atom to be
electrically neutral the upper limit, L¡, on each integral must be constant. In
other words, Si adjusts on each site i in such a way that (E F - s¡)j(2JL)2)1/2
is the same on a11 sites. Therefore, the only quantity in eqn (9.24) for the
bond energy which varies from one site to the next is the (JLI 2)1 /2 factor
before the exponential. Thus, the contribution to the bond energy from site
i may be written in the following form
bond = constant x (1I!2)1/2

E(i) ,-,. (9.28)
Recall from eqn (9.21) that JL)2) is a sum of hopping integrals between atom
i and its neighbours. Thus, eqn (9.28) says that the bond energy per atom
can be represented by the square root of a sum of pairpotentials between
the atom and its neighbours. This is the functional form of a Finnis-Sinclair
potential. An illustrative example is given in Problem 30.
d-d bonding 183
d-d bonding
On p. 114 we considered the kinds of bonds that may be formed from s and
p atomic states on neighbouring atoms. There were only four kinds of bond
that could be formed which conserved the angular momentum about the
bond axis. We also saw that the magnitudes of the hopping integrals
increased as the overlap between the atomic sta tes increased. If we consider
two neighbouring atoms withthe five d states described on p. 14 on each
atom, we find that there are just three types of bond that may be formed
which preserve the angular momentum about the bond axis. They are called
(ddu) (m = O), (ddn) (m = 1), and (dd.5) (m = 2), and they are shown in
Fig. 9.7. They are the three fundamental hopping integrals involving d
orbitals, and they appeared in eqn (9.22).
From their appearance in Fig. 9.7 we expect the (ddu) hopping integral
to be largest in magnitude because the lobes of the d-orbitals are pointing
towards each other and therefore overlap the most. Similarly, we expect the
(dd.5) hopping integral to be the smallest in magnitude because the lobes of
the d orbitals are parallel to each other and therefore there is little overlap
between them. We expect the (ddu) hopping integral to be negative because
the lobes that overlap have the same signo The same goes for the (dd.5)
hopping integral. But the (ddn) integral involves overlap of lobes of opposite
sign and therefore the integral is positive. The relative values of the three
integrals are as follows
(ddu):(ddn):(ddo) = - 2.4: + 1.6: -0.4. (9.29)
Let two atoms be labelled 1 and 2 and choose the z-axis to be along the
bond between them. Hopping integrals between the five d states on each
atom are shown in Table 9.2.
ddlt (m = ±I) ddS (m = ±2)
Fig. 9.7 The three fundamental hopping integrals between d orbitals on neigh-
bouring atoms. (ddu) has no angular momentum about the bond. (ddn) has
angular momentum m = ± 1 about the bond. (ddb) has angular momentum·
m = ±2 about the bond.
184 Tha transition matals
Tabla 9.2 Hopping integrals between d states on neighbouring atoms
(xy), (yz), (zx), (x2 - y-l), (3z2 - "),
(xy), (dda)
(yz), . (dd1t)
(zx), (dd1t)
(x2 - y-l), (dda)
(3r -"h (ddu)
The angular variation of hopping integrals as one atom is rotated around

the other is quite complicated for d-d interactions, and was tabulated by
Slater and Koster (1954). Their table is incJuded in Table 9.3. In fact this
table gives the angular variation of the hopping integrals between a1J
combinations of s, p, and d orbitals, and it is of fundamental importance in
understanding the angular dependence of bonding. To illustrate the use oC
this table consider the entry for the interaction between a d~y orbital on
atom 1 and a d~,_y' orbital on atom 2. The bond is along [1, m, n] where
1, m, n are direction cosines. In the table we see the hopping integral is listed as
E(~,.~,_y,) = !lm(1 2 - m2)"dd. + 21m(m2 _12)"dd. + tlm(1 2 - m2)"dd6
(9.30)
and "dd., "dd., and "dd6 are the fundamental hopping integrals that we have
caJled (ddu), (ddlt), and (dda). We see that as the bond axis varies, the
hopping integral between the d~ orbital on atom 1 and a d~, _y' orbital on
atom 2 is a mixture of the three fundamental integrals.
On pp. 127-131 we showed how the relative contributions of ssu, spu,
ppq, and pplt bonding in siJicon could be assessed by breaking down the
energy of a bond into these components. In Fig. 9.8 we show the result of
a similar breakdown of d-d bond energies in b.c.c., f.c.c., and h.c.p. crystal
structures as a function of the fiJling of the d bando There are several features
about these curves that are worth noting. The first is that in aJl three crystal
structures the ddo bonding contribution is negligible. Bonding in transition
metals is dominated by ddu and ddlt interactions. Secondly, even though the
second neighbour distance in b.c.c. crystals is only 15 per cent greater than
the first neighbour distance the bond energies oC the second neighbours are
much less than those of the first neighbours. This is because the hopping
integrals for d-d interactions scale as the inverse fifth power of the bond
length (compared with the inverse square oC the bond length for s-p hopping
integrals). Therefore the hopping integrals to the second neighbours are 49
per cent of those to the first neighbours. Third1y, in the h.c.p. structure we
see that the ddlt bonds between neighbours in the basal plane are weaker
than those out of the basal plane in the region where the number oC electrons
d-d bonding 185
Tabla 9.3 Hopping integrals between s, p, and d atomic states as a
function of the direction cosines 1, m, and n, of the vector from the left
state to the right state. From Slater and Koster (1954)
Es.s = V...,
E.,x = IVsp•
Ex,x = ¡2vpp• + (1 _/2 ) VpP'
Ex,y = ImVpp• - ImVpp•
Ex.z = InVppo -lnVpp"
Es,xy = 3' /2 /mVsd •
Es,x2-y2 = ;3'/2(¡2 - m 2)VSd'
Es,3z2-r' = [n 2 - W2 + m 2 )] Vad.
Ex,xy = 3'/2¡2mVpd. + m(1 - 2¡2) Vpd.
/2
Ex,yz = 3' /mnVpdO - 2/mnVpd.
2
Ex,zx = 3' /2 / nVpdO + n(1 - 2/2) Vpd•
Ex ,x2 _y2 = 23'/2/(/2 - m 2)VpdO + 1(1 _/ 2 + m 2)Vpd.
2 2 2
Ey,x2_y2 = ;3' /2m(¡2 - m ) Vpdo - m(1 + 1 - m ) Vpd.
Ez ,x2 _y2 = ;3' /2 n(l2 - m 2) Vpdo - n(/2 - m 2) Vpd.
2 2
E x ,3z2-r' = l[n - ~(/2 + m )]Vpdo - 3'!2ln Vpd.
2
2 2
E ,3z2-r' = m[n - ;(12 + m )]Vpdo - 3'/2mn2Vpd.
y
E z ,3z2-r' = n[n2 - ;(12 + m 2)]Vpd. + 3' /2 n(¡2 + m 2) Vpd.
2 2 2 2 2 2 2
Exy,xy = 31 m Vddo + (12 + m - 4¡2m ) Vdd • + (n + 1 m ) Vdd6
2 2
Exy,yz = 3lm2nV~do + In(1 - 4m ) Vdd • + In(m - 1) Vdd6
2
Exy,zx = 3/ mnVdd • + mn(1 - 4/2) Vdd • + mn(P - 1) Vdd6
2 2 2
E xy ,x'-y2 = ~/m(/2 - m ) Vddo + 2/m(m2 -/2)Vdd • + ;lm(/ - m ) Vdd6
2
E yz.x'-y2 = ~mn(¡2 - m ) Vdd • ~ mn[1 + 2(¡2 - m 2)]Vdd •
+ mn[1 + ;(¡2- m 2)]Vdd6
2
E zx ,x'-y2 = ~nl(¡2 - m ) Vddo + nl[1 - 2(¡2 - m2)]Vdd •
- nl[1 - W2 - m 2)]Vdd6
E Xy ,3z2-r' = 3'/2/m[n2 - ;(12 + m 2)]Vddo - 3'/22Imn2Vdd.
+ ;3' /2 /m(1 + n 2) Vdd6
2 2 2 2
E yz,3z2-r' = 3' /2 mn[n - ;(12 + m )]Vdd • + 3' /2 mn(l2 + m - n )Vdd •
2
- ;3' /2mn(¡2 + m ) Vdd6
E zx,3z2-r' = 3'/2In[n2 - ;(12 + m 2)]Vddo + 3'/21n(l2 + m 2 - n 2)Vdd •
- 23'/2In(/2 + m 2) Vdd6
2
Ex'- y2.x'-y2 = i(/2 - m )2Vddo + [12 + m 2 - (12 - m 2)2]Vdd •
+ [n + 1(/2 - m 2)2]Vdd6
2
2
Ex'- y2,3z2-r' = ;3'/2(12 - m2)[n - 2(¡2 + m2)]Vddo + 3' /2 n 2(m 2 - ¡2) Vdd •
2
+ 13'/ (1 + n )(¡2 - m 2) Vdd6
2
2 2 2 2 2
E 3z2 -r',3z2-r' = [n - !(¡2 + m )]2Vddo + 3n (¡2 + m ) Vdd • + i(¡2 + m )2Vdd6
(a)
(b)
(c)
(electroDS atom-1)
2 -=3:'~:f4"''''5--6'''''(-'7'''''' 10
6J. 62
~.l
~.2
~.3
Fig. 9.8 ddu, ddlt, and ddt5 eontributions to the bond energy as a funetion of
the number of eleetrons per atom in d-bonded erystals, normalized in sueh a way
that their sum gives the bond energy per atom. The unit of energy is the
bandwidth, W. (a) b.e.e. strueture. 1 and 2 label 1st and 2nd neighbours.
(b) f.e.e. strueture (first neighbour interaetions only). (e) An ideal h.e.p. strueture
(first neighbour interaetions only). 1 and 2 label inequivalent neighbours in the
basal plane and out of the basal plane respectively. From Sutton et al. (1988).
Changas in crystal structura 187
per atom is between 1 and 2. In the transition metal series this d band filling
eorresponds to Ti(3d), Zr(4d), and Hf(5d) whieh are all h.c.p. metals with
Iess than ideal ela ratios. The nonideal ela ratios are a consequence of the
stronger bonds out of the basal plane than within it. Once again we see the
vaIue of a chemical picture of bonding in solids.
Changes in crystal structure across the transition

metal series
The second moment model, which led to the development of Finnis-Sinclair
potentials, has captured sorne of the essential physics of the metallic bond,
namely the environment dependence of the bond order and the bond energy.
But we saw in eqn (9.12) that it is not able, in a nearest neighbour model,
to discriminate between the energies of different crystal structures. In this
section we consider what it is that determines the changes in crystal structure
across the 4d and 5d series. This is an important issue because the factors
giving rise to the structural energy differences also control the structures and
energies of deJects in the transition metals. F or example if the difference in
energy between. f.c.c. and h.c.p. crystal structures is very small we would
expect the stacking fault energy on close packed planes to be small and
therefore the crystallattice dislocations on those planes would be dissociated,
which has a profound bearing on the mechanical properties of the metal.
Figure 9.9 shows the variation of the calculated d-bond energy across the
4d and 5d transition metal series for the f.c.c., ideal h.c.p., and b.c.c. crystal
structures. The solid curve shows the d-bond energy of the b.c.c. crystal
relative to that of the f.c.c. crystal and the dotted curve shows the d-bond
energy of the ideal h.c.p. crystal relative to that of the f.c.c. crystal. These
10
h.c.p.-f.c.c.
-. b.c.c. -f.c.c.
-15
Fig. 9.9 The ealeulated variation 01 the d-bond energy aeross the 4d and 5d
transition metal series lor the I.e.e., ideal h.e.p., and b.e.e. erystal struetures. The
solid curve shows the d-bond energy 01 the b.e.c. erystal relative to that 01 the
I.e.c. crystal and the dotted curve shows the d-bond energy 01 the ideal h.c.p.
crystal relative to that 01 the I.e.c. crystal. From Pettilor (1972).
curves predict the change in crystal structure from h.c.p. -+ b.c.c. -+ h.c.p. -+
f.c.c. -+ b.c.c. as Nd increases from 1 to 10. This trend agrees with Table 9.1
except at the end of the series where the noble metals have the f.c.c. crystal
structure rather than the b.c.c. The error arises from the neglect of the outer
s electrons, which are important in the noble metals. We conclude from this
comparison that the changes in crystal structure across the 4d and 5d
transition metal series are driven by the filling of the d bando
So, if the changes in crystal structure are driven by the filling of the d
band, and the second moment of the density of states of the d band is not
able to discriminate between the energies of the crystal structures, then the
factors determining the structural stability must be in higher moments of the
d-band density of states. That this is the case was first proved in a theorem
by Ducastelle and Cyrot-Lackmann (1970). The theorem goes as follows: let
the first n moments (i.e. /leo), /l(t), /l(2), ••. , /l(ft-t) of a function p(E) be zero.
Here p(E) is a real function of energy which is nonzero in sorne energy band
a < E < b. Then the function p(E) crosses the energy axis in a < E < b at
least n times. If p(E) is now a dilference, t5d(E), between two densities of
states (for example the dilference between the densities of states in the f.c.c.
and h.c.p. crystals) and the first n moments of p(E) are zero then t5N(EF )
has n - 1 zeros and t5U(EF ) has n - 2 zeros as EF varies through the band,
where
(9.31)
and
t5U(EF ) = 2 r~ (E - EF ) t5d(E) dE (9.32)
t5N(EF ) is the dilference in the number of electrons in the two bands up to

energy EF and t5U(EF ) is the dilference in the bond energies in the two bands
up to energy EF •
Let us now apply this theorem to the ideal h.c.p.-f.c.c. curve in Fig. 9.9.
The curve shown in Fig. 9.9. is a plot of t5U(Nd ) but since EF increases
monotonically with Nd in each band the number of zeros t5U(Nd ) is the same
as the number of zeros in t5U(E F ). In the h.c.p. and f.c.c. crystal structures
the first four moments are identical (in a nearest neighbour model). In
principie /l(4) could be dilferent in f.c.c. and ideal h.c.p. crystal structures, but
in practice it is almost the same. The first moment to show a significant
dilference between f.c.c. and ideal h.c.p. is /les), and it is /les) and /l(6) which
determine the relative stabilities of the two crystal structures. The theorem
predicts that the bond energy dilference between these two crystal structures
is zero at least three times between the band limits. This is in agreement
with Fig. 9.9. where it is seen that the h.c.p.-f.c.c. curve crosses the axis more
than three times between the band limits. In an s-band model the densities
of states in the f.c.c. and ideal h.c.p. crystal structures are identical (this is
D(E)
~+---------~--.E
Fig. 9.10 A schematic density of states for a b.c.c. transition metal showing a
pseudogap.
not difficult to believe but it is tricky to pro ve!). Therefore the origin of the
relative stability in the f.c.c. and h.c.p. transition metals is the directional d
bonding leading to different moments Jl(S) and Jl(6) and higher.
Tbe reason why the b.c.c. crystal structure is stable in the middle of the
series is that the d-band density of states for the b.c.c. lattice has a bimodal
form, with a pseudogap in tbe middle of the band as shown in Fig. 9.10.
Tberefore, the density of states at the Fermi level is relatively low for the
b.c.c. structure compared witb a density of states of unimodal form when
tbe Fermi level is in the middle of the bando Tbis leads to a more negative
bond energy when the band is approximately half filled for the b.c.c. crystal
structure. Using eqn (3.89) we see that the condition for a bimodal density
of states is tbat [Jl(4) Jl(2) - (Jl(J)2]f(Jl(2)3 is less than 2. Therefore it is the
balance between tbe third and fourth moments which controls the stability
of tbe b.c.c. structure.
Bonding in metallic alloys

In tbis section our aim is to give a eomparison of bonding in metallie alloys
and bonding in ionie compounds. In the diseussion of eontaet potentials on
p. 159 we pointed out that when two different pieces of metal are brought
into contact there is in general a fiow of eharge from one to the other, to
set up a dipole at tbe interface between tbem, whieb equalizes tbe Fermi
level tbrougbout botb metals. Sometbing similar to tbis bappens wben we
make an alloy of two elements A and B. Let us first eonsider tbe limit of
ionie bonding.
In Fig. 9.11(a) we sbow a sebematie density of states plot for making an
AB ionie eompound. Element B bas a bigber on-site energy, E~, tban tbat
oC element A, E~. The pure A and pure B bands do not overlap. Wben A
and B are mixed in a binary alloy cbarge fiows from tbe bigher energy states
associated witb atom B to tbe lower energy states on atom A. Tbe result is
<a)
E E E E
EO EO
a a
Ea EO
A A
na nA na nA ntot
B + A AB
Metallic bond
(b)
E E E E
EOa I 1
W
¡ + W
I- WAB
¡ I
na nA na nA n,."
Fig.9.11 Schematic illustrations of the local and total densities of states during
the formation of (a) an ionic AB compound and (b) a metallic AB alloy. From
Pettifor (1987).
that the lower band is filled and the upper band is emptied and there is a
gap between them. Thus we have an insulator and the lower band is
associated with the anion sites and the upper band with the cation sites. This
is the rigid-band model, so called because the alloy density oC states is assumed
to be a rigid superposition oC the elemental densities oC states. It is a
reasonable description oC a material such as NaCl.
This band picture oC an ionie eompound may seem somewhat removed
from the more Camiliar Bom model, but in Caet they are equivalent, as we
now show. In the Bom model the energy oC the ionie solid compared with
that oC Cree atoms is given, per ion pair, by
Eionlc = 1 + A + UMad (9.33)
where 1 is the ionization energy oC the eation, A is the electron affinity oC
the anion, and UMad is the Madelung energy oC the ionie lattice. If we identify
the ionization energy with the free atomie energy level E3 (i.e. 1 = - E3)
and the eleetron affinity with E2 + ~, where ~ describes the Coulomb
repulsion on the anion site due to the faet that we are adding an eleetron
to an atom that has already got an almost filled shell of eleetrons, then we
can rewrite E,o. as follows
E'on = -(E~ - E2) + (UM ' d + ~). (9.34)
We find that for the alkali halides the seeond term in braekets is generally
small eompared to the first beeause the negative interatomie Madelung
energy is almost eaneelled by the positive intra-atomie Coulomb energy. For
example in NaCI UM • d + ~ = -8.9 + 10.3 = 1.4 eV per atom pair, whieh is
onIy 16 per eent of the first term, -8.8 eVo Consequently the ionie binding
energy of the alkali halides may be approximated by
E'on ~ -(E~ - E2)· (9.35)
This is just the predietion of the rigid band model for making AB ionie
moleeules from isolated A and B atoms.
It is perhaps worth elarifying a possible source of eonfusion. While it is
tfUe that the Madelung energy is largely eancelled by the intra-atomie
Coulomb repnIsion, ~, it is also tfUe that the Madelung energy ehanges as
ions are moved inside the solid whereas the inter-atomie Coulomb repulsion
does no!. Therefore if we are interested in modelling the energeties of defeets
in ionie crystals, where we are concemed with energy differences between
different ionic configurations, the Madelung energy remains of central
importance.
To summarize, the ionie bond is eharaeterized by bringing together
elements with widely separated densities of states, with electronic eharge
Dowing from the higher to the lower energy band, resulting in a filled band
located on the anion sites and an empty band on the cation sites.
The metallic alloy bond eould not be more different! Now the energy
bands of the A and B elements are too close in energy for the rigid band
model to apply and a new common band is formed, as shown schematieally
in Fig. 9.11(b). New eommon states are formed in the alloy from bonding
and antibonding combinations of states on the A and B atoms. The local
densities of states on the A and B atoms in the alloy thus have a common
width, Jv..B' The centre of gravity, E2B' of the eommon density of states is
the average of the centres of gravity of the A and B local densities of states,
which remain E2 and E~
(9.36)
The width of the eommon band, WAB ' in a rectangular band model is given
by eqn (9.5) as
WiB = 12Jl~á (9.37)
where Jll.'J is the second moment of the common band
JI~ = Loooo (dA (E) ; d.(E» (E - E2.)2 dE (9.38)
where dA(E) and d.(E) are the local densities ofstates on the A and B atoms
in the alloy. Using the moments theorem, eqn (3.85), we may write this as
follows
Jll.'J = Z{J2 + !«E2 - E~B)2 + (E~ - E~.)2) (9.39)
where Z{J2 = Jll.') = Jlll) = W 2/12 and W is the bandwidth of pure A and
pure B (see eqn (9.6». It follows from eqn (9.37) and eqn (9.39) that
AE)2)1/2
WAB = W ( 1 +3 ( W (9.40)
where AE = E~ - E~. Thus the width of the common band increases as AE

increases as we would expect.
The local densities of states dA(E) and d.(E) are constrained to have the
same common bandwidth WAB • We have also noted that their centres of
gravity remain E~ and E~ respectively. To satisfy these two constraints
the local densities of states dA(E) and d.(E) are skewed, as shown in
Fig. 9.11(b). If the skew-reetangular density of states dA(E) takes the values
5(1 + <X)/Jv,.. and 5(1 - <x)fWA • at the bottom and top of the band respect-
ively, then demanding that the centre of gravity is E~ leads to the result that
3AE
<X=-- (9.41)
WA •
provided <X :S 1. The local density of states d.(E) skews in the opposite sense
by the same amount. The electron eharge transfer from A to B is caleulated
by fiUing up the bands to the Fermi energy. It is given by
q = tAN. + to(N.)(1O - (N.» AE (9.42)

WA •
where (N.) = (N~ + N:)/2 and AN. = N: - N~. The position and label-
ling of the bands in Fig. 9.11(b) eorrespond to AE > O and AN. < O. The
first term in eqn (9.42) drives eharge from the higb valence A atom to tbe
low valence B atom, whereas the second term takes eharge in tbe opposite
direction from the higher to the lower atomie energy leve!.
If we stopped bere we would have a partially ioníe solid due to the ebarge
transfer deseribed by eqn (9.42). But we have already emphasized several
times that sereening in metals is praetieaUy perfeet! This means !hat
eaeh atom, together with its sereening eloud, is eharge neutral. This is
aebieved in our LCAO model by adjusting the on-site energy dilference
AE AB = E. - EA.1t follows from eqn (9.40) and eqn (9.42) tbat local charge
References 193
neutrality is achieved iC
I'1E - - WI'1Nd (943)

AH - (n[ (Nd )(10 - (Nd ) )]2 - 3(I'1Nd )2)1/2 .
There is ample evidence Crom accurate calculations oC the density oC states

that the common band description oC metallic alloys is valido The skewing
of the local densities of states is also found in accord with eqn (9.41). We
conclude that the ionic model is completely inapplicable to metallic alloys,
and the concept of 'charge transCer' should never arise when we are
discussing metallic alloys.
Rafarancas
Daw, M. S. and Baskes, M. I. (1984). Phys. Rev. B, 29, 1285.
Ducastelle, F. and Cyrot-Lackmann, F. (1970). J. Phys. Chem. Solids, 31, 1295.
Finnis, M. W. and Sinc1air, J. E. (1984). Phi/o Mag. A, 50, 45.
Pcttifor, D. G. (1972). In Meta/lurgical Chemistry (oo. O. Kubaschewski), p. 191.
HMSO, London.
Pcttifor, D. G. (1987). Solid State Physics, 40, 43.
Slater, 1. C. and Koster, G. F. (1954). Phys. Rev., 94, 1498.
Sutton, A. P., Finnis, M. W., Pettifor, D. G., and Ohta, Y. (1988). J. Phys. C: Sol.
State Phys., 21, 35.
10
Structural stability of compounds
Hybridization and crystal structural stability

The notion that particular atomic configurations of molecules and crystals
are stabilized by the formation of bonds between particular hybrid states
assumes the bonds are saturated, that is, occupied by two electrons each. F or
example, in our earlier discussion of sp hybrids we assumed that each bond
was occupied by two electrons. A saturated single e-e bond has a
characteristic length and energy which is almost independent of the types of
other atoms bonded to the two e atoms. The properties of a e-e bond
depend on the configuration of atoms in the local environment because the
configuration determines whether the e-e bond is a single, double, or triple
bond. But for a given number and angular disposition of bonds to each e
atom the e-e bond is less sensitive to whether other e or H or O, etc.
atoms are bonded to the two e atoms. This is the basis of' covalent radii'
for atoms in characteristic bonding topologies. It is as if each saturated bond
has a personalized local wave function and properties. In valence bond
theory this is known as the 'perfect pairing approximation'. In molecular
orbital theory it is known as the bond orbital approximation, which we
discussed on p. 123 in the context of Si.
If the bonds are not saturated then we have a partially filled band and
metallic bonding. In that case it is still possible to construct hybrid states,
but there are not enough electrons to saturate all the bonds. In valence bond
parlance the electrons then resonate between the bonds and the properties
of a metallic bond are not separable and transferable from one environment
to another in the way that those of a saturated bond are. The concept of
directional hybrids stabilizing particular crystal structures is not applicable
to metallic bonding.
In our discussion of s-p hybridization we could have incJuded d states
and even higher energy states into the hybrids. There are two reasons why
we did not. Firstly, is the obvious one that for s-p bonded elements and
compounds the d states have energies that are so much higher than those
of the s and p states that the energy of promotion inhibits the electronic
occupation of such hybrids in practice. Secondly, in s-p bonded materials
the d orbitals are rather dilfuse on the scale of the observed bond lengths.
It is not easy to form localized, directed bonds between neighbours involving
such states.
Hybridization and crystal structural stability 196
In heavier elements, however, the energy difference between the outer s,
p, and d orbitaIs is smaller. For example in S the electronic configuration
oC the Cree atom is Is22s22p 6 3s 23p· and the 3d shell is not much higher in
energy than the 3s and 3p shell. By contrast in O, where the electronic
configuration is Is 22s 22p· there is no 2d shell and the 3d shell is much higher
in energy. As another example the 3d, 4s, and 4p levels in Ni lie within about
4 eV oC each other. The principies on which s-p-d hybrids are constructed
are the same as those on p. 117, namely maximum overlap oC hybrids along
a bond and orthogonality between hybrid states on the same atom. By
suitable combinations oC such orbitals very strongly directed hybrids can be
Cormed giving coordination numbers and bond angles quite different Crom
those expected using s, p, and d atomic orbitals separately. The main types
oC s-p-d hybridization and the coordination numbers and geometries that
they lead to are shown in Table 10.1.
Notice that the maximum coordination number is 6. In metallic structures
the coordination number is usually greater than 6 and this underlines our
earlier point that we should not expect the predicted configurations in the
right hand column oC the table to apply to metals.
To illustrate an s-p-d hybrid consider the tetragonal planar hybrids
(dsp2). Ifwe mix together dXl_yl, s, Px, and Py orbitals we obtain the Collowing
Table 10.1 Coordination numbers and bonding geometries

for the main types of s-p-d hybridization
Coordination number Atomic orbitals

of hybrids used Resulting hybrids
s sp Linear
dp Linear
sd Bent
3 Sp2 Trigonal plane
dp2 Trigonal plane
d 2s Trigonal plane
d 2p Trigonal pyramid
4 Sp3 Tetrahedral
d 3s Tetrahedral
dSp2 Tetragonal plane
5 dSp3 Trigonal bipyramid
d 3sp Trigonal bipyramid
d4 s Tetragonal pyramid
6 d 2sp3 Octahedral
d4 sp Trigonal prism
196 Structural stability of compounds
set of four square planar hybrids along ±x, ± y
h±x = t(s ± 21/2px + d '-y2)
X
(10.1)
h±y = t(s ± 21/2py + d x2_ y2). (10.2)
The formation of two h±x hybrids is illustrated in Fig. 10.1. These four
hybrids are equivalent and they point in the ± x and ± y directions. They
arise in certain transition metal oxides and sulphides, e.g. PdO, PtO, PdS,
and PtS where each oxygen is tetrahedrally coordinated by metal atoms but
the metal atoms themselves are surrounded by four oxygens in aplane at
the corners of a square, as shown in Fig. 10.2. The metal atoms are therefore
dSp2 hybridized.
x " z +
/;9,'
x ",-:- ", ......... y
Fig.10.1 Combining d x 2_y" s, and Px atomic orbitals to lorm two dsp hybrids,
called h.x in eqn (10.1). From McWeeny (1979).
i
0 --- -----:
--------0
_......
,
,, '
,
--y
. : Pd,Pt; O: o,S
fig. 10.2 A unit cell 01 PdO, PtO, PdS, and PtS.

Atomic factors influencing the structures of compounds 197
Atomic factors influancing tha structuras of compounds

The structures oC tens oC thousands oC binary, temary, and quatemary
compounds have been detennined since the advent oC X-ray crystallography
in 1910. More than 2000 different types oC crystal structure have been
identified posing a daunting challenge to theory to (i) order the data base
and (ii) understand it. Four Cactors have been identified that are thought to
be important in determining the type oC crystal structure. The first three are
related to the Hume-Rothery rules of metallic alloy formation.
1. Atomic size. For example, the structure oC an ionic compound depends
on the relative sizes oC the cations and anions. The Madelung energy, which
is the term that distinguishes between the energies of different crystal
structures oC ionic compounds, Cavours small separations between the anions
and cations. But Cor each crystal structure, when the ratio oC the size oC the
cation to the size oCthe anion, R+/R_, is less than a critical value the anions
come into contact and the lattice parameter oC the crystal does not decrease
when R+/R_ is lowered Curther. Another crystal structure then becomes
Cavourable. This is ilIustrated in Fig. 10.3, where the lowest energy structure
changes Crom the CsCl, to the NaCl to the ZnS as R+/R_ decreases. If
the atomic diameters oC a metallic alloy differ by more than 15 per cent the
primary solid solubility is small. The size Cactor is also important in the
stability oC the Laves phases (Cor example MgCu2) and the u phases (Cor
example FeCr), as discussed by Frank and Kaspar (1958).
2. Electronegativity. This determines the covalent vs ionic content oC a
bond in an insulatoT. Covalent bonds are directional whereas ionic bonding
leads to closed shells with spherical charge densities and therefore no
directionality. As discussed at the end oC the last chapter, the role oC
, ,
..,
b'"
¡¡ CsCl
i . ,,.,, ,,"
'i " ,,'/
:::;
-- - -
NaCI
Zns ,"
,""
- -,'- -
------¡- I .1 ,
o 0.2 0.4 0.6 0.8 1.0
R.IR_
Fig. 10.3 The Madelung energy in ionic compounds as a function of the ratio
of ionic radii, assuming the ions are hard spheres and the packing density is the
maximum attainable for the given crystal structure and ionic radii ratio.
electronegativity in metal/ic compounds is more subtle because screening
ensures that each atom remains charge neutral; for a fuIl discussion see
Pettifor (1987).
3. Number of valence electrons (band filling). In Chapter 9 we saw that
the changes in crystal structure across the transition metal series are
determined by the filling of the d bando In covalent solids of groups IV-VII
the number of nearest neighbours is determined by the '8 - N rule'. Each
atom has 8 - N nearest neighbours, where N is the ordinal number of the
periodic group (i.e. 4 for C, 5 for N, 6 for 0, 7 for F and so on). This is a
statement of the octet rule of molecular chemistry. In metallic aHoys the
electron to atom ratio can also have a decisive influence on the stable
structure of the aHoy, but unfortunately there is no simple explanation for
this which stands up to a critical examination, although there are more
advanced theories (see CottreH (1988».
4. Angular character of valence electrons. This is important in aH
materials, because it alrects the density of sta tes in metals and the nature
and angular disposition of hybrid bonds in insulators. For example Ti and
Pb both have four valence electrons per atom and the same electronegativity
but the h.c.p. structure of Ti is due to s-d bonding while the f.c.c. structure
of Pb is due to s-p bonding.
Structure maps
The purpose of a structure map is to order the vast empirical database of
crystal structures by representing compounds with a given stoichiometry on
two- or three-dimensional plots in such a way that aH compounds with a
given structure type are located in weH defined domains, which are separated
from domains containing other structure types. Their use is to enable
predictions to be made of other possible compounds with a desired crystal
structure, and to suggest possible aHoying elements to modify the crystal
structure of a known compound in sorne desired way.
The central problem is the choice ofaxes for the two- or three-dimensional
plot. There is a considerable history of attempts to use combinations of the
four factors listed in the previous section as the axes. F or example, one type
of three-dimensional structure map used the dilrerence in electronegativity,
AX, between the constituent elements as one axis, the dilrerence in atomic
sizes, AR, as a second axis and the average number, e/a, of valence electrons
per atom as the third axis. But when this scheme is applied to the AB binary
compounds it is found that the fifth most common structure type (the NiAs
structure) cannot be separated from the other crystal structures and has to
be omitted. Moreover, the scheme cannot separate the structures of the
elements, e.g. Ti and Si have AX = 0, AR = 0, and e/a = 4, yet their crystal
structures are certainly not the same! The problem with this selection of the
Structure maps 199
axes is that it ignores the angular character of the bonding. However, it
would ciearly be problematie (!) to use Jour axes for a strueture map.
Pettifor introduced a single parameter to eharaeterize eaeh element, whieh
is entirely phenomenological and which leads to good struetural separation
in binary and pseudobinary eompounds in two-dimensional plots. The
thinking is that we should use the periodie table of the elements as the basis
for ordering the elements along eaeh of the two axes of the strueture map.
In principie the periodie table captures aH the variations in atomic size,
eleetronegativity, eleetron-to-atom ratio, and the angular eharaeter of the
bonding. This suggests that we use the atomie number, but that inereases
along the periods, whereas ehemieal behaviour is often determined by the
group in whieh an elernent appears. The atomie numbers of elements in the
same group differ by considerable amounts and therefore the atomie number
is not an appropriate parameter. Instead the elements of the periodie table
are ordered in a somewhat eurious way, but it is sueh that it has been found,
by a process of trial and error, to lead to the best struetural separation in
the structure map. The success in separating the different erystal struetures
is the only justifieation for the empirical ordering of the elements. The
ordering of the elements may be visualized by running a string through the
periodie table as shown in Fig. 10.4. It is seen that this string has a number
of kinks and turns in it, and H is at the end of the string, rather than the
beginning, and the rare gases appear on the left of the table rather than the
right. It is stressed that these adjustments were made empirically in order
to aehieve good struetural separation in the subsequent strueture maps.
Pulling the ends of the string apart then orders the elements along a
one-dimensional axis, and their sequential order is ealled the Mendeleev
number M.
Struetural separation of all binary eompounds with a given stoiehiometry
is aehieved by using the Mendeleev numbers of the eonstituent elements
along the horizontal and vertical axes. Strueture maps for the following
stoiehiometries have been plotted: AB, AB 2 , AB 3, AB 4, AB" AB 6, AB!!,
AB!2' AB!3' A2 B3, A2B" A2 B!7' A3B4' A3B" A3B7' A4B" A6B23' The
Mendeleev number also successfully demareates the struetural domains of
moleeules displaying, for example, the regions where AB 2 trimers are bent
or linear.
The AB strueture map was shown in Fig. 1.6. The bare patehes eonespond
to a positive heat of formation where eompounds do not formo The diagonal
line MA = Mo is a minor plane in the map. There are 52 AB strueture types
whieh have more than one representative eaeh and they are all weH separated
on this two-dimensional plot. The lines drawn on this map are a guide to
the eye to show the struetural separation. As expeeted the CsCI domains
(represented by open eircles) adjoin the line MA = Mo where elemental b.e.e.
struetures are found. Similarly CuAu domains (represented by eircles with
a dot in the middle) adjoin the line where f.e.e. erystal struetures are stable,
o lA nB 1nB IVB VB VIB VllB
,
1 ,
12 TI
I
86
I
95
I
100 10\ 102
:~ ~:~i N~
He ti
I I
2 11
, 1
~ ~ ~
i ji P
11A IDA
q
\VA VA
P P P
V1A V11A VllIa V111b VllIc
q. q.
lB I
'-'"/
Mg
lit I I
3 \O 16 19 51 54 57 60 61 64 67 72 76 81 84 89 93 98
I I • I I I I r--1 I I I • I I I I I
Al K Ca Sen V Ce Mo Fe Co Ni Cu la Ga GeAsSe Br
I
4
I
I
9
I
I
15
I
~495~ ~! 6~ 6~ 69 JI 7~
, I I I I I I I
79
I
83J8
1
I I
I I97
92
I I
Kr Rb Sr y 'h Nb Mo 11: Ru Rb Pd As Cd In So Sb 11: 1
X.
I
5
"
8
I
,
Ca
I
14
I
Ba
~ 5~ 5~ 5~ ~3 ~ +\ ~o 7~ ~8 8~ 8~
HfTa W Re Os Ir PI Au Ha TI PbBi
I 96I
91
I
Po
I
Al
I I I
,
6 7
, 13
I
Rn-Fr Ra
17 18 33 32 31 30 29 28 27 26 24 23 22 21 20
I I
Yb-Ea La-c..- Pr- Nd -Pro-Srn.f(E.)L Gel - 'Ib- --Dy-Ho- Ec-TmJ(Yb)L Lu
48 47 46 45 44 43 42 41 40 39 38 37 36 35 34
Ac-'lb-Pa -U-Np-Po-Am-Cm-Bk-Cf-Es- Fm-Md-No- u
Fig. 10.4 The Mendeleev number is defined by running a string through a modified periodic table. When
the string is pulled straight all 103 elements are ordered along a one-dimensional axis, From Pettifor
(1988).
Applications of structure maps 201
and the cubic ZnS domains (represented by an open diamond) adjoin the
line where the elemental diamond cubic structure occurs.
The AB 2 structure map is shown in Fig. 10.5. AlI 84 AB 2 structure types
with more than one representative are well separated. The map is no longer
symmetric about MA = MB • This reflects, for example, the importance of the
relative sizes of the atoms in the occurrence of Laves phases whose main
domains are found for MA > MB •
Applications of structure maps

One of the principal uses of structure maps is to guide the alloy designer in
the selection of ternary and quaternary additions to a binary alloy in order
to effect a desired change of crystal structure. Such use rests on the important
discovery that ternary and quaternary alloys, with binary AB or AB3
structure types, can be treated as pseudobinary alloys characterized by
average Mendeleev numbers <MA >and <MB >on the A and B sublattices.
This means that if <MA >and <MB >of pseudobinary AB alloys are used as
the coordinates of a structure map then the structural domains are very
similar to those of the pure binaries.
As an example of the use of structure maps consider AI3 Ti which has a
tetragonal crystal structure. This intermetallic alloy has high strength and
high creep resistance at high temperature, its density is low and its oxidation
resistance is high. It would therefore be a good candidate for high tempera-
ture applications, such as a turbine blade in a jet engine, if it were not for
the fact that it is extremely brittle at room temperature. It was thought that
if a suitable ternary alloying addition could be found such that the tetragonal
crystal structure were transformed into the cubic Cu 3Au structure, then the
availability of five independent slip systems should lead to greater ductility
at low temperature. In other words the alloy would be ductilized. The
question facing the alloy designer is which ternary element should be added?
Figure 10.6 shows the relevant part of the AB3 structure map. We see that
the intermetallics Al 3Hf, Al3 Ti, AI3 Ta, AI 3Nb, and AI3 V, with Mendeleev
numbers ranging from MA = 50 lo 53 respectively all lie in a tetragonal
(D022 ) domain (open triangles). A cubic Cu 3Au (Ll 2 ) domain (open circ\es)
lies below these alloys at lower values of MB • Thus if AI3 Ti is alloyed with
an element which goes onto the B sublattice, and which reduces the average
Mendeleev number <MB >, then the crystal structure might transform into
the CU3Au structure. It turns out that this works if the ternary addition is
Cu, Ni, or Fe. However, it also turns out that the cubic alloys remain brittle
and fail by transgranular c\eavage, for reasons that are not understood!
But it does not always work and that is why the 'might' was emphasized
aboye. For example, attempts to stabilize AI3Nb in the cubic structure by
replacing sorne of the Al by Cu, Ni, or Fe have failed. The reasons for this
failure are understood and they are associated with the quantum mechanical
.JIA
o 10 20 30 40 50 60 70 80 90 lOO
lA DA EuLu!lo lb Pm Ce No Ea Cm Np lb !VA VA VIA VIIA VIlla VIDb VIIk m UD
F H
lOO N O
VUB
e
vm
90 VB
B
IVB
80 nm
UB
m
70
vme
VIIIb
vme
60 VllA
VIA
.JI D VA
50 !VA
Th
Np
Cm
40
Es
No
30
Ce
Prn I
Th ¡
Ho
/
20 Lu /
Bu /
•
~
lO I
lA
¡
J~o//1
/
/
o
e CaF, 11 Cu,Sb n CS2 d MgZn, • CoSb, " PtHg2 ' CaSi, A CeCu, o MoS~
el CaC, 11 CaCI, u Pd,As V Cu,Mg , RuB, " V2N rlCdCI, V ZrSb, e Al,Cu
m TI02 !I VO, ~ DySe, <> MgNi, .. FeS,(p) < N~ L,As,Ge "'HfGa, ~ Ni,ln
B CdIz ~ CuCl, e GdSi, t>Calnz • FeS2(m) > CaSb, ~ Fe,N .. irSe, e Fe,P
11 Ag20 " zr02 H MoS, .. Cd,Ce ~ MoCl, z CuMg2 , FeSb, O ZrSi,
s Li2Sb • crSi,
11Ta,P T HoSb
2 .. SiS2 ~ OsGe
2 '" EuS 2 + PdSe, ~ TISi, 11 La,Sb
lO ThH 11 NbTe, i Li Ga ~ OaCI, ~ AUzV N AUzNd .... BaSi2 ~ TbSi 2
2 2 • M0l'1
11 BaS, " SmSb2 ~ Zt{Ja, A PdCI, ~ CuP, ~ NbS ... TbC, ~ AlB o SrBr¡
2 2
• 112Ni 11 Fe,C -t FeSi 2 ~ ZnP mLi2Sb , PbO, 11 Co Si
2 2 • SrIz
w ReB, " Pt,Oa • Other • Ag,1O
• NbSb2
AB,
Fig. 10.5 The AB, structure map. From Pettifor (1988),

References 203
TiNb
.
B
IVB
•
I
80 Al IlB
t
1IB
70
Cu lb
Ni vme /
.ita VIllb
Fe Vma
60
VIlA
VIA
VA O Ll2
A 00 22
50 IVA
.00 19
Tb /
Fig. 10.6 Relevant part of the AB3 structure map showing the transition metal
tri-aluminides. From Pettifor and Aoki (1991).
nature of the bonding in the interrnetallic alloys. But this is not a failure of
the structure map since the map displays inforrnation only about existing
crystal structures. Rather, it is a faHure of over-interpretation of the
structure map, and a blind application of empirical rules without regard for
fundamental theory.
References
Cottrell, A. H. (1988). Introduction to the modern theory 01 meta/s, Chapo 8. The
Institute of Metals, London.
Frank, F. C. and Kaspar, J. S. (1958). Acta Cryst., 11, 184.
McWeeny, R. (1979). Cou/son's Va/ence. Oxford University Press.
Pettifor, D. G. (1987). Solid State Phys., 40, 43.
Pettifor, D. G. (1988). Mat. Sci. Techno/., 4, 675.
Pettifor, D. G. and Aoki, M. (1991). Phi/o Trans. R. SOCo LoOO. A, 334, 439.
11
Introduction to modern
quantitative theory
Modern quantitative predictions of crystal structure

and stability
So far in this book we have used simple models to gain insight and basic
understanding about the electronic structure and properties of materials.
These models have all involved a number of approximations and while they
give valuable insight and can even predict trends in properties, they contain
parameters which must be fit either to experiment or to accurate calculations.
But in order to make confident predictions about the properties of materials
it is necessary to work with theories that contain nothing that has to be fit
to experiment, that is the calculation should be done from first principies
with as few approximations as possible.
Since about 1980 it has become possible to make quantitative predictions
of a wide range of materials' properties essentially from first principies. These
calculations are almost exact solutions of the Schrodinger equation for
the material. They entail massive amounts of number-crunching and the
'problems of current interest' are always at the edge of what the latest
super-computers can do. The purpose of such a calculation might be to
predict the elastic constants of a new intermetallic alloy in the confident
knowledge that if the elastic constants were measured they would agree with
the calculated values to within experimental error. It is an exciting fact 01
current computational modelling that such a calculation can be done (see
Fu (1990». The key theoretical breakthrough carne in 1964 and 1965 with
the advent of density functional theory. But it was not unti1 the mid to late
1970s that computers became sufficiently fast to enable the theory to be
exploited. A further technical advance carne in 1985 with the development
of the Car-Parrinello method (Car and Parrinello 1985) and related
techniques on which sorne of the most ambitious modern calculations are
based. In this chapter our aim is not to provide a discussion of the
technicalities of the calculations (although these are important because they
enable the calculations to be done and they are being improved constantly),
but rather to give an elementary account of density functional theory on
which the calculations are based. We shall also illustrate the application 01
this theory to problems in materials science. But first we describe the
Born-Oppenheimer approximation, which we have been using implicitly
throughout this book and which is also used in these calculations.
The Born-Oppenheimer approximation

This approximation amounts to saying that we can separate the electronic
and nuclear degrees of freedom. Because the electronic mass is so much
smaller than that of the nuclei the e1ectrons respond almost instantaneously
to changes in the positions of the nuclei. It is then a good approximation to
say the electrons are always in their ground state as the atoms of the solid
vibrate thermalIy. This means that the positions of the nuclei are parameters
that appear in the potential of the Schrodinger equation defining the wave
functions of the electrons. You may be wondering whether this approxi-
mation ever fails! In fact it does break down sometimes, for example in many
chemical reactions an electron jumps from one energy surface to another in
a nonadiabatic way. It also fails in nonradiative transitions in solids where
an electron falIs from a high energy surface to a low energy surface not by
emitting a photon but by emitting phonons. We shalI assurne the Born-
Oppenheirner approximation is valido
Outline of density functional theory

In order to appreciate the significance of the breakthrough that the advent
of density functional theory represents let us first describe the status of the
'many-body problern' prior to 1964. To sirnplify rnatters consider jelliurn.
AII the electrons in jelliurn repel each other electrostaticalIy. Therefore the
electrons are not independent in the sense that the motion of any one of
thern is determined by the distribution of all the others. If we want to write
down an exact wave function for the jelIiurn we have to specify the positions
of alI the electrons in the systern. If there are N electrons in the jelliurn then
the wave function of the systern is a function of 3N variables, which is about
1023 variables in a cubic centimetre of a typical metal. This is the many-body
problern. It is a cornpletely hopeless task!
We have already discussed the independent electron approxirnation on
p. 155. We gave physical argurnents for its validity on the basis of the
exchange and correlation hole that each electron clothes itself with. The
rigorous justification for the independent electron approxirnation carne only
with the developrnent of density functional theory. In 1964 Hohenberg and
Kohn (1964) proved the folIowing theorern. All aspects of the electronic
structure of a systern of interacting electrons, in the ground state, in an
'externa!' potential v(r), are cornpletely determined by the electronic charge
density p(r). Note that the charge density is a function of only three variables!
The 'external' potential is the set of nuclear potentials. Note also that the
theory is a theory of the ground sta te: it cannot address excited states of the
208 Introduction to modern quantitative theory
system. The theorem says that aH we need to know for the ground state
properties of the system is the electronic charge density. Thus the charge
density becomes the central quantity that must be found, in place of the
many-electron wave function.
Hohenberg and Kohn also proved that the ground state energy of the
interacting electron gas is a unique functionalt of p(r). Unfortunately, the
functional is not known! But whatever. the functional is, it acquires a
minimum value when the charge density is the correct (true) charge density.
MathematicaHy this means there is a variational principie for finding the
charge density.
The next step carne in 1965 when Kohn and Sham (1965) derived a system
of one-particle equations for the description of the electronic ground state.
The interacting N-electron problem was thus mapped exactly onto N single
particle equations. This means each electron is moving independently of the
other electrons, but it experiences an effective potential which emulates aH
the interactions with the other electrons. Here we have the rigorous
justification for the independent electron approximation that we have been
using aH along. Kohn and Sham further showed that the etTective potential
is a unique functional of the charge density. The one particle equations are
known as the Kohn-Sham equations.
Before we get into the Kohn-Sham equations let us discuss a simpler case,
namely the Hartree approximation. This will enable us to introduce the idea
of self-consistency between the charge density and the etTective potential in
an intuitive way. In the Hartree approximation each electron moves
independently in the mean electrostatic field of the other electrons and the
nuclear electrostatic potentials. The electrostatic potential due to the elec-
tronic charge density p(r) is given by Poisson's equation
(11.1)
or
VH(r) = f p(r')
4nEolr - r'l
dr' (11.2)
where VH(r) is caHed the Hartree potential. The etTective potential felt by
each independent electron is then
V.rr(r) = VH(r) + VN(r) (11.3)
where ~(r) is the electrostatic potential due to the nuclei
(11.4)
t A functional is a function oC a function.

and Z¡e is the charge on the ith nuc1eus and R¡ is its position. The
Schrodinger equation for the wave function of the j th independent electron is
h2
--2- V 2 'Pir ) + V,ff(r)'Pj(r) = Bj'Pj(r). (11.5)
8n m
Once this equation is solved the eigenstates are populated with electrons (at
OK). A new charge density is then calculated from the occupied states as
follows
p(r) = L 'Pir )'P1(r). (11.6)
j occupied
We see that the wave functions 'Pj(r) are being defined by an effective
potential in eqn (11.5) which is a functional of the charge density, through
eqn (11.2), which in tum is defined by the wave functions in eqn (11.6)! This
is called a self-consistent field problem. When it is solved the output charge
density is the same as the input charge density. The way this would be done
on a computer is shown schematically in Fig. 11.1.
Guessinput p(r) e.g. a superposition of atomic charge densities
r
,..----to-----1 Construct VH(r) =J4 neoP(l T-
') '1 dr'
T
Construct V'If(r) =VH(r) + VN(r)
Construct output p(r) = r

'I'j(r)'I'j(r)
joccupied
Is oulput p(r) !he same as input p(r)?
Fig. 11.1 Flow chart of a self-consistent calculation in the Hartree approxi-

mation.
Another way of setting up the Hartree problem is to start with the total
energy of the system as a functional of the charge density and then to
minimize the energy with respect to the charge density. This is the variational
problem. In this way of doing things the ground state totalenergy is
expressed as follows
Eo[p(r)] = T[p(r)] + f p(r)VN(r) dr +~ f

p(r)VH(r) dr (11.7)
where T[p(r)] is the kinetic energy of the electrons. Minimization of this

energy, subject to the constraint that the total number of electrons is
conserved,
f p(r) dr = N, (11.8)
is equivalent to solving eqns (11.3), (11.5), and (11.6).

It tums out the Hartree approximation gives very poor answers. The
reason is that the Hartree energy,! Jp(r) VH(r) dr, overestimates the Coulomb
repulsion between the electrons. For example, the electron is interacting with
itself in the Hartree energy, which is obviously wrong. In addition, the
interaction between each electron and all other electrons in the system is
also overestimated in the Hartree energy. The Hartree approximation ignores
the exchange-correlation hole that each electron dresses itself with. The
amount of missing electronic charge in the hole is precisely one electron and
this corrects for the e1ectron self-interaction, and for the overestimation of
the electron-electron interactions in the Hartree energy.
Kohn and Sham (1965) showed that the exact ground state energy
functional could be written in a form similar to eqn (11.7)
Eo[p(r)] = T[p(r)] + f p(r) VN(r) dr +~ f p(r) VH(r) dr + Exc[p(r)] (11.9)
where the new term is the exchange energy Exc[P(r)], which is the energy
correction arising from the exchange-correlation holeo Minimization of this
energy, subject again to the constraint of charge conservation, eqn (11.8),
leads to the Kohn-Sham equations
2
h 2
--2- V 'I'ir ) + v,ffer)'I'¡(r) = tiKS 'I'j(r) (11.10)
81r m
V,rrer) = VH(r) + VN(r) + VxcCr ) (11.11)
p(r) = I 'I'¡(r)'I'j(r) (11.12)
j occupied
where Vxc(r) is the exchange-correlation potential (equal to the functional

derivative of the exchange-correlation energy ¿;Exc[p(r)]/¿;p(r». As in the
Applications 209
Hartree equations these equations have to be solved iteratively until
self-consistency is achieved between the input and output charge densities.
As they stand the Kohn-Sham equations, eqns (11.10)-(11.12), are an
exact solution of the interacting electron problem. The only difficulty is that
the exchange-correlation energy functional, Exc, is not known for a spatially
varying charge density. In a solid (as in an atom) the charge density is a
rapidly varying function of position and therefore this is no mean obstac1e.
However, the exchange-correlation energy can be computed to any required
accuracy for a jellium of uniform electron density. This ca1culation can be
repeated for a range of uniform electron densities. Thus we obtain the
exchange-correlation energy as a function of arbitrary uniform electron
density. To apply this Exc to a nonuniform electron density we make what
is called the local density approximation. In this approximation we pick a
volume element in the solid and measure the charge density there and find
that it is sorne value, which we call Po' The exchange-correlation energy we
assign to this volume element is then approximated as the exchange-
correlation energy of a volume element in a uniform electron gas of the same
density as Po' This is an approximation because it ignores the fact that the
charge density in the solid is varying from one volume element to the next.
Until extensive computations were carried out it was thought that the local
density approximation would apply only if the charge density were a slowly
varying function of position, and that the rapid variations in an atom or in
the solid would prec1ude its use. But it worked extraordinarily well even in
atoms and solids, and then the challenge was to explain its success! This was
finally solved when it was shown that the exchange-correlation hole in the
local density approximation still contains minus one electron, as it should,
and that all that was being approximated was the shape of the hole in the
inhomogeneous electron gas. The local density approximation is made in
virtually all density functional calculations.
Applications
Table 11.1, from Srivastava and Weaire (1987), gives a comparison of
predictions from first principies, using density functional theory, of lattice
parameters ao, bulk moduli B, and cohesive energies Ecoh ' for a selection of
materials with their experimental values, the latter being shown in brackets.
Notice that the range of materials spans metals, semiconductors, and
insulators.
It is seen that the agreement is very good overall. However, the cohesive
energies are consistentJy overestimated in the local density approximation,
which is most probably due to errors made in the ca1culations of the total
energy of free atoms.
In the 3d transition metals (only one of which is shown in the table) the
lattice parameter and the bulk modulus are consistently underestimated and
Table 11.1 Comparison of lattice parameters, 80' bulk
moduli, B, and cohesive energies, Ecoh ' for a selection of
materials (experimental values in brackets). From Srivastava
and Weaire (1987)
Material 80 (A) 8 (Mbar) Ecoh (eV)
C 3.60 (3.57) 4.41 (4.43) 7.57 (7.35)

Si 5.45 (5.43) 0.98 (0.99) 4.67 (4.63)
GaAs 5.57 (5.65) 0.73 (0.75)
GaP 5.34 (5.45) 0.897 (0.887)
NaCI 5.56 (5.60) 0.284 (0.266)
Al 4.01 (4.02) 0.715 (0.722) 3.646 (3.401)
V 2.97 (3.03) 2.0 (1.62) 5.83 (5.30)
overestimated respectively in the local density approximation. Figure 11.2

shows the energy-volume curves (from Paxton et al. 1990) calculated for the
3d transition metals in the local density approximation for a variety of real
and hypothetical crystal structures of the 3d series. These calculations did
not take into account spin polarization (which leads to magnetism in sorne
of the 3d elements), and that is one reason why the predicted lowest energy
crystal structures are not always the experimenta11y observed structures.
Another reason is the use of the local density approximation itself. For
example, in Fig. 11.2 Fe is predicted to have the h.c.p. crystal structure at
OK. However, the crystal structures of a11 the nonmagnetic elements are
predicted correctly and sorne interesting new features are seen. F or example
in Cu the energy difference between the f.c.c. and b.c.c. crystal structures is
very sma11. In Co, even in the absence of a proper account of spin polarization
in the calculation, it is seen that the f.c.c. and h.c.p. crystal structures have
very similar energies, which is consistent with the existence of a martensitic
transformation between these structures in Co at 400°C.
Figure 11.3 (from Yin and Cohen 1982) shows the computed structural
energy curves for Si and Ge in the local density approximation again for
observed and hypothetical crystal structures. In both cases the predicted
crystal structures at O K are the diamond cubic structure, which agrees with
experiment. It is also seen that the hexagonal diamond structure is only
sJightly higher in energy, which is consistent with the relatively low stacking
fault energies in Si and Ge a110wing partial dislocations to exist. But the
most interesting feature of the curves is the prediction that under pressure
both Si and Ge will transform to the p-tin crystal structure. Since these
calculations were done this prediction has been experimenta11y confirmed.
Since the charge density is the central quantity in density functional theory
it is important to test whether the predicted charge density agrees with
experimento This has been done (Yin and Cohen 1982) in Si and Ge
Se TI v
-2.S -3.0 -2.5
~c -3.5
~ubie
-3.0 -3.0
.-'
'i
~
-3.S
s.cubic
-5.0 -4.5
\
e ~s.cubic
-d.eubie
~'" -4.0
-5.5 -5.0 ~s.hex.
~f.e.e.
s.hex.
-4.5 -{í.0 -5.5
~b.e.e.
~f.c.c. Vb,C,C.
b.e.p. -{í.5 c....,~~~=....:::0.~9.l1:::J'0
0.8 1.0 1.2 1.4 0.8 1.4 1.6 0.8 1.0 1.2 1.4 1.6
ntOo ntOo
-1.5 ~-~~o--------,
Mn -3.5 Fe
d.eubie ' " Cr -2.5
~ie
-2.0 "-....~~..O~
-3.0
-3.0
~
~s.hex. ~
J>s.eubie
-3.5
-3.5
-5.0
s.hex. -4.3
\. b.e.p. -4.0
-4.0 ~f.c.c. -4.4
-5.5
~ b.~.c.
~b.c.c. -4.5 ~ -4.5
-4.5 h.c.p. -4.6 L.,,~"""""'i
0.8 1.0 1.2 1.4 1.6 0.8 1.6 0.8 1.0 1.2 1.4
ntOo OJn"
-3.5~-----~ -1.5 ~---------,
-3.0 \ Ni
-4.0 ~ -2.0
'i -4.5
d.cubic
d.cubic
~ -5.0 s.cubic
~.. .......~~s.hex.
-3.7
-5.5
-5.0 \ --5.2 b.e.e.
-6.0 r.c.c. ~ f.c.c ...... _...vl b.c.c.
f.e.e. -5.3 "=,~"=--1 -4.0
0.8 0.9 1.0 -5.5 ú--"-~L..J.-"0::::.9.....:;1.0::...w 1.0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
ntOo ntOo ntOo
Fig. 11.2 The binding energy (eV) vs volume curves tor the 3d transition
metals computed in the local density tunctional approximation (without spin
polarisation) by Paxton et al. (1990). In each case the t.c.c., b.c.c., h.c.p.,
simple hexagonal, simple cubic, and diamond cubic crystal structures are con-
sidered. The voluma íl is normalizad to the experimentally observed equilibrium
volume, ílo.
-7.84 -7.82
Si Ge
-7.86 -7.84
~
le i;
~ .
la
~ -7.88 ~ -7.86
J -7.90
J -7.88
Diamond I Diamond I
-7.92 -7.90
0.6 0.7 0.8 0.9 1.0 1.1 0.6 0.7 0.8 0.9 1.0 1.1
Volume Volume
(al (bl
Fig.11.3 The total energy (in ryatom-') 01 (a) Si and (b) Ge in seven crystal
structures as a lunction 01 volume, normalized to the experimental volume. The
diamond cubic, hexagonal diamond (wurtzite), p-tin, simple cubic, t.c.c., b.c.c.,
and h.c.p. crystal structures are considerad. The broken line is the common
tangent lor the diamond cubic and p-tin phases. From Yin and eohen (1982).
by comparing the predicted X-ray scattering factors with experimentally

measured structure factors. The agreement is within 1-2 per cent for
reftections up to (4 4 O) that are not forbidden. A particularly demanding
test is the calcuiation of the (2 2 2) structure factor, which is forbidden in a
simple superposition of atomic charge densities. The reason why this
structure factor is not zero in reality is the charge density in the bonds
between the atoms. In Si the caiculated value is 0.34, compared with the
experimental value oC 0.38. In Ge the calculated value is 0.28, compared with
the experimental value oC 0.27.
Local density calcuiations have been carried out recently (Fu 1990) oC
elastic constants and ideal anti-phase domain boundary and surface energies
in AI3Sc and AI3 Ti intermetallic a1loys. The calcuiated elastic constants
are within a few per cent of available experimental measurements. These are
extraordinarily demanding calculations since the accuracy has to be within
0.01 eV atom -1 and one has to cope with the low symmetry oC a distorted
crystal structures. The self-consistent charge density reveals directional
bonding between the transition metal d sta tes and the aluminium p
states, which explains why earlier attempts to develop second moment
Appl ications 213
Finnis-Sinclair type models (which ignore directional bonding) for these
alloys failed.
Sorne of the most exciting applications of local density calculations
have been the simulation of relaxed extended defects and dynamical
processes within solids, and at their surfaces. For example, Stich et al.
(1992) and Brommer et al. (1992) independently carried out a full relaxation
of the (1 1 1) surface of Si in the (7 x 7) Takayanagi reconstruction.
These simulations demonstrate that it is now (end of 1992) possible to
consider problems up to 1000 atoms, in which every atom is relaxed
to its equilibrium positions with forces determined by the solution of the
Schrodinger equation in the local density approximation. It is quite
conceivable that, with the imminent arrival of computers with teraflop
ratings (i.e. capable of performing 10 12 floating point operations per second),
simulations involving tens of thousands of atoms will be possible using these
first principies methods. An example of a dynamical simulation is the
diffusion of hydrogen through silicon in real time by molecular dynamics,
again using forces determined by local density functional theory (Buda et al.
1989).
A particularly interesting application of these first principies methods in
materials science is the simulation of the 90° Shockley partial dislocation
in silicón by Bigger et al. (1992). This edge dislocation is one of the
eommonest found in silicon, and its atomic and electronic structures have
been studied intensively both experimentally and theoretically. A key
question is whether there are electronic sta tes in the band gap associated
with the dislocation coreo Such states are important because they can trap
and recombine electrons and holes in the material, and thereby reduce the
current that can flow through the material. Earlier simulations suffered from
the uncertainty associated with using empirically constructed interatomic
potentials, or inadequate (and in sorne cases incorrect) boundary conditions.
By showing how periodic boundary conditions should be used to model
dislocations correctly, and by relaxing all atomic positions using a first
principies local density functional method, Bigger et al. (1992) were able to
provide the dejinitive solution for the atomic and electronic structures of this
dislocation. They showed that the band gap remains essentially clear of
states, and they found only one stable atomic structure for the dislocation
eore, which is stabilized by purely electronic factors. The importanee of this
result is that it rules out the possibility that the core of the 90° partial
dislocation in silicon is associated with electronic sta tes in the gap. If sta tes
are observed experimentally at this dislocation then their origin must be
something else, such as segregated impurity atoms, kinks or the long range
elastic field of the dislocation.
Examples such as these demonstrate that we are at the beginning of a new
era of predictive modelling in materials science. Future progress will be
limited only by computer power.
References
Bigger, J. R. K., McInnes, D. A., Sutton, A. P., Payne, M. C., Stich, l., King-Smith,
R. D., Bird, D. M., and Clarke, L. J. (1992). Phys. Rev. Lett., 69, 2224.
Brommer, K. D., Needels, M., Larson, B. E., and Joannopoulos, J. D. (1992). Phys.
Rev. Lett., 68, 1355.
Buda, F., Chiarotti, G. L., Car, R., and Parrinello, M. (1989). Phys. Rev. Lett., 62, 555.
Car, R. and Parrinello, M. (1985), Phys. Rev. Lett., 55, 2471.
Fu, C. L. (1990). J. Mater. Res., 5, 971.
Hohenberg, P. and Kohn, W. (1964). Phys. Rev., 136, 864.
Kohn, W. and Sham, L. J. (1965). Phys. Rev., 140, 1133.
Paxton, A. T., Methfessel, M., and Polatoglou, H. M. (1990). Phys. Rev. B, 41, 8127.
Srivastava, G. P. and Weaire, D. L. (1987). Adv. Phys., 36, 463.
Stich, l., Payne, M. c., King-Smith, R. D., Lin, J.-S., and Clarke, L. J. (1992). Phys.
Rev. Lett., 68, 1351.
Yin, M. T. and Cohen, M. L. (1982). Phys. Rev. B, 26, 5668.
12
Where band theory breaks down
Electrons in noncrystalline materials

In this book we have tried to keep a balance between a real space, chemical
picture, and a reciprocal space band theory picture of the electronic structure
of a solid. In an amorphous solid the reciprocal space approach breaks down
completely. There is no translational symmetry with which to define a
reciprocal lattice or Brillouin zone and we are forced to think in terms of
chemical bonds. One could argue that this remark applies to a crystalline
solid containing a defect but that is not entirely fair beca use it is possible to
think of the defect as a perturbation to the crystal and to solve for its
electronic structure using perturbation theory. But in an amorphous material
that is not possible because the whole specimen is 'perturbed' in the sense
that there is no crystalline material anywhere.
For many years the optical transparency of g1ass was considered a mystery.
The fact that light can pass through g1ass implies that there must be an
energy gap in the spectrum of electronic energy levels in the material. The
magnitude of the energy gap must be greater than hv where v is the frequency
ofthe Iight. But if, as had been shown in the 1930s, band gaps in crystalline
materials were due to Bragg reflection of travelling electron waves at
Brillouin zone boundaries how could there possibly be a band gap in glass
where the concepts of a Brillouin zone and Bragg reflection are inapplicable?
The same question applies to the optical transparency of Iiquids Iike water.
We saw on p. 148 that the explanation of band gaps in crystalline materials
in terms of Bragg reflection is entirely consistent with the atomicjchemical
origin of band gaps that we described in Chapter 5. In amorphous materials
the chemicaljatomic explanation of band gaps is still valid and we shall
develop it in the next section. We will see that general arguments may be
given for the existence of a band gap that have nothing to do with
translational symmetry. In simple terms a band gap exists in glass because
of the finite energy required to excite an electron in a bonding state into an
antibonding state.
Although there is no long range translational order in g1ass or amorphous
silicon there is short range order and this leads to the important concept of
a continuous random network. Consider the structure of amorphous Si.
The radial distribution function, as measured by X-ray diffraction, in an
evaporated amorphous film of Si is shown in Fig. 12.1 (broken line). When
216 Where band theory breaks down
lOO Á Si film
- - - As deposited
--Crysta1lized
20.0 r- I
I
I
I
I
15.0 r-
,,
~ .¡
..... 'r~
" "- 10.0
"~
5.0 -
,
I
I
J J
, J
V
I
,,
J
\j
0.0 L
1.0 2.0 3.0 4.0 5.0 6.0
Radial distance r (Á)
Fig.12.1 Radial distribution lunction 01 an evaporated amorphous thin lilm

(broken line) and after it is crystallized (sol id line). From Moss and Graczyk
(1970).
the film is crystallized the solid line is obtained. The first peak is at almost
the same position and height in both the amorphous and crystalline forms.
The positions of the second peaks are also almost identical. The first
significant difference comes in the third peak, where the peak in tbe
crystalline case is almost absent in the amorphous case. This observation
tells us that each Si atom in the amorphous specimen has four nearest
neighbours with bond lengths within about 1 per cent of the crystalline
value and approximately 12 second-nearest neighbours at about the same
separation as in the crystalline case. But the short range order breaks down
beyond second neighbours in the amorphous film. The conclusion is that in
the amorphous film each Si atom is tetrahedrally coordinated just as it is in
the crystalline state, but there is no long range order. This is the continuous
random network (CRN).
Electrons in noncrystalline materials 217
Fig. 12.2 A computer generated model of amorphous silicon showing a con-

tinuous random network (CRN) . From Wooten and Weaire (1987).
Ball and stick models have been eonstructed of CRNs and one is shown
in Fig. 12.2. In this model each Si atom that is not on the surfaee of the
cluster, has four nearest neighbours in an approximately tetrahedral arrange-
ment. The bond angles cannot be exactly the same as they are in the
crystalline state. When the atomic coordinates are entered into a computer
and the strain energy of the cluster minimized it is found that small
adjustments take place of bond lengths and angles but the coordination
remains the same. After relaxation the bond lengths are within 1 per cent of
the crystalline value and bond angles are within 10° of the crystalline value.
The pronounced short range order of the CRN enables us to define an
electronic defect in the network. lf one neighbour of an atom is missing then
we have an unsatisfied bond: a 'dangling bond'. A ball and stick model of
a dangling model in a CRN is shown in Fig. 12.3. Vacant sites in the network
may also exist and there are then four dangling bonds pointing into the
Fig . 12.3 A dangling bond in a random network simulating the strueture 01

amorphous germanium or silieon. From Mott and Davis (1979).
vacant site. The dangling bond may contain zero, one, or two electrons. In
its neutral state the dangling bond contains one electron. This electron has
a spin which can be detected experimentally by electro n spin resonance
(ESR) measurements. The measurements reveal 10 18 _10 20 cm- 3 dangling
bond centres in amorphous Si. The dangling bond may capture a second
electron and the excIusion principie requires that its spin is opposite to that
of the first electron. The negatively charged dangling bond is no! associated
with a net spin and therefore there is no ESR signal. The energy level of the
second electron in the dangling bond is higher than that of the neutral
dangling bond beca use of the Coulomb repulsion between the two electrons.
Finally, the neutral dangling bond may give up its electron and become
positively charged. This is equivalent to saying that the dangling bond
captures a hole. Again there is no ESR signal.
The energy gap in amorphous silicon 219
The Cact that dangling bonds in amorphous Si may capture electrons or
holes means that these deCects may seriously impair the operatíon oC
semiconductor devices such as solar cells. In a solar cell electrons and holes
are generated by optical excitation. But iC these carriers are captured by
dangling bonds they may be recombined at those centres and lost, thus
reducing the output oC the solar cell. The common solution to this problem
is to introduce hydrogen into the amorphous silicon. Hydrogen, which
diffuses rapidly owing to its small size, has a valence oC one. It seeks out
and saturates the dangling bonds in the silicon. This is called hydrogen
passivation.
The energy gap in amorphous silicon

In this section we shall use the Weaire-Thorpe model that we introduced
on p. 123 to address the question oC why there is an energy gap in amorphous
silicon. The approach we shall take is rather intuitive and builds on our
experience oC the nature oC bonding and antibonding states in solids.
However, the condition we derive Cor the existence oC a gap is the same as
those derived more rigorously and it is identical to eqn (6.16). The point
about the model is that it shows that provided one has strong short-range
order, as in a CRN, then the condition Cor an energy gap is no different Crom
what it is in a crystalline solido The argument is entírely in real space and
Collows Heine (1971).
Recall that in the Weaire-Thorpe model we set up 4 Sp3 hybrids on each
atom. If there were no short-range order in the amorphous Si it would not
be valid to assume Sp3 hybridization leading to tetrahedral bonding. But we
have seen that in the CRN this remains a very good approximation. We
now transCorm this basis oC Sp3 hybrids by Corming bonding and antibonding
combinations oC them on neighbouring atoms. Let Iij) denote the ith hybrid
(i = 1,2,3, or 4) on atomj. Let lipjp) be its partner pointing directly towards
lü) Crom the neighbouring atom, as shown in Fig. 12.4(a). We construct
bonding, lB, ü), and antibonding, lA, ij), combinations oC these hybrids
(12.1)
(12.2)
The Hamiltonian matrix elements in the new basis may be expressed in terms
oC the matrix eleÍllents PI and /j. introduced on p. 122. The only nonzero
matrix elements between bonding sta tes are as follows
(12.3)
~
lij) liojo)
(a)
(b)
(e)
Fig. 12.4 (a) A bonding state formed from the ith hybrid on atom i interaeting
with the ipth hybrid on atom ip.(b) In the most bonding molecular state all
bonding states between neighbouring pairs of atoms are in phase. (e) In the
most antibonding molecular state bonding states between neighbouring pairs of
atoms are 180' out of phase.
and
(B, ijIHIB, i'j) = !«ijl + (i.j.I)H(li'j) + li~j~» = 1l/2 (12.4)
The matrix element (B, ijIHIB, i'j) is between different hybrids, i and i', on
the same atomj. Ifwe now think ofthe bonds as the 'sites', rather than the
atoms, then (B, ijIHIB, ij) is an on-site energy and (B, ijIHIB, i'j) is a
hopping integral between neighbouring sites.
For the moment let us ignore the antibonding states and consider only
the bonding states {lB, ij)}. What are the maximum and minimum eigen-
values of the Hamiltonian matrix in the subspace of bonding states? This is
where we use a bit of intuition. The eigenstates of the system, no matter
whether it is crystalline or amorphous, are linear combinations of the basis
states {lB, ij) }
'1' = L cijlB, ij). (12.5)
ij
The lowest eigenvalue corresponds to the most bonding combination of

states {lB, ij)}, which arises when all the expansion coefficients, cIJ ' are equal.
This most bonding state is shown in Fig. 12.4(b). The energy oC this state is
easily obtained.1t is equal to the on-site energy, p, (eqn (12.3», oC a bonding
state plus the sum oC the hopping integrals between this state and the three
Electron localization 221
bonding states at either end of the bond: 6 x 1'1./2 (eqn (12.4». Thus the
energy of the most bonding state is PI + 31'1. (remember that PI and 1'1. are
negative quantities). The highest eigenvalue corresponds to the most anti-
bonding combination of the states {lB, ij)}. This arises when aH the cijs have
the same magnitude but two of the cijs on each atom are positive and two
of them are negative, as shown in Fig. 12.4(c). In that case there are two
hopping integrals between sta tes of the same sign and four hopping integrals
between states of opposite sign for the central bond: (2 - 4) x 1'1./2 = -1'1..
Therefore the energy oC the most antibonding state is PI - 1'1.. The width of
the valence band is thereCore 411\1.
Now consider the subspace of the antibonding states {lA, if)}. The only
nonzero Hamiltonian matrix elements between these states are as foHows
(A, ijIHIA, ij) = !«ijl- (ipjpl)H(lij) -lipjp» = -PI (12.6)
and
(A, ijIHlA, i'j) = y(ijl- (ipjpI)H(li'j) -li~j~» = ±A/2. (12.7)
Applying the same arguments as before to the {lA, ij)} subspace oC the
Hamiltonian we find that the lowest eigenvalue is - PI + 31\ and the highest
eigenvalue is -PI - 1\. Thus the conduction band is also ofwidth 411\1, and
there is an energy gap between the highest eigenvalue of the valence band,
PI - 1'1., and the lowest eigenvalue of the conduction band, - PI + 3i\, if
IPII > 211'1.1, which is eqn (6.16).
In this simple analysis we have ignored the coupling between the subspaces
{lB, ij)} and {lA, ij)}. But this coupling will only increase the bonding-
antibonding splitting and widen the energy gap. Our conciusion, that there
is an energy gap if eqn (6.16) holds, thereCore remains valido
Electron localization
The electronic states we have considered up to now have aH been delocalized
over the N atoms in the molecule or solido We argued in Chapter 2 that for
any finite value of the hopping integral there is a corresponding finite
probability, per unit time, that the electron will hop (tunnel) Crom one site
to the next. Thus the probabiJity of finding the electron at any one site is
proportional to l/N Cor such a delocalized state. But there are instances
where this simple picture breaks down and the probability oC finding the
electron outside sorne region in the solid is exponentiaHy smaH. In such cases
the electron is said to be in a localized state. Localization oC the electron can
be viewed, at the simplest level, as a competition between an electrostatic
potential term tending to localize the electron and the kinetic energy oC the
electron tending to delocalize it. When the electron is in a localized sta te
the electrostatic term dominates. Localization arises in many crystalline
and noncrystaHine materials. One of its most dramatic maniCestations is
Anderson localization where it is possible for aH states in the system to
become localized as a consequence of disorder. In that case the material
becomes an insulator at OK because the electrons cannot escape from the
regions in which they become localized. But there are many less extreme
examples as well. For example, as discussed in Chapter 11, localized states
in crystalline semiconductors are often associated with defects such as
free surfaces, impurities, grain boundaries, dislocations, and so on. These
localized states can trap electrons and holes in the material and recombine
them. There is therefore a lot of interest in the existence of localized states
at defects in semiconductors.
Polarons
In an ionic crystal an electron or hole polarizes and distorts the crystal
lattice. If the e1ectron is to move through the crystal then the lattice distortion
it induces must move along with it. Thus the effective mass of the electron
or hole is greater than the band theory estimate. In an extreme case the
carrier can become trapped by its distortion field: it then becomes localized.
The inseparable combination of the electron and the polarizationjdistortion
field it induces is called a polaron.
There are large and small polarons. In the case of a large polaron the
lattice polarization is small and the electron moves in a band with a slightly
enhanced mass. The lattice polarization at a small polaron is much greater
and the electron is trapped most of the time at a single lattice site. At high
temperatures the electron moves by thermally-activated hopping from site
to site. At low temperatures the electron tunnels through the crystal slowly,
as ir in a band of large effective mass.
Holes are self-trapped in all the alkali and silver halides. The hole is a
missing electron on the halide ion (for example chlorine). That ion and a
neighbouring CI- ion move together to form a CI, molecular ion pair as
shown in Fig. 12.5. To move the hole requires rearrangement of the ions.
The rate at which a self-trapped hole can tunnel to neighbouring sites is on
the scale at which ions rather than electrons tunnel through potential
barriers.
Suppose we place an electron in an unfilled orbital on a particular ion in
the solido The electrostatic polarization of the lattice that the electron induces
favours localization of the electron. But then the kinetic energy of the e1ectron
is high owing to the exclusion principie (the electron is in a small box!).
Therefore the kinetic energy favours delocalization. Let us try to make this
more quantitative. The particJe in a box estimate of the kinetic energy of the
electron is
(12.8)
where R is the radius in which the electron is confined and m* is the effective
Polarons 223
Fig. 12.5 A hale on a CI- ion in AgCI. The absence 01 the electron on the CI-
ion attracts a neighbouring CI- ion to form a CI, ion. The hale is then trapped
by the lattice distortion. From Harrison (1980).
mass of the electron given by eqn (4.24). To estimate the potential energy
we ealeulate the energy ehange in moving a eharged sphere of radius R from
a vaeuum into the solid
M=-~(I-~)
8neoR e,
(12.9)
where e, is the statie relative permittivity of the solid. The polarization of

the solid whieh eontributes to e, has two eomponents: eleetronie and ionie.
We are eoncerned only with the ionie polarization energy beeause the
eleetronie polarization energy is present whether the eleetron is loealized or
not. The eleetronie polarization energy is obtained by using the high
frequeney limit of the relative permittivity whieh is eaned the optiea! relative
permittivity, eopt
2
/'J.E.= - -e - ( 1 -1-) . (12.10)
8neoR eopt
The ionie eontribution to the polarization energy is the differenee
2
e (1 ¡:1).
/'J.E¡ = - 8ne R e
o opt
-
(12.11)
Adding E kin to /'J.E¡ gives the total energy of the loealized state of radius R.
Differentiating this with respeet to R and setting the derivative to zero we find
8ne oh 2
R= --:----='----- (12.12)
m*e 2 (1/eoPt - l/e,)
Therefore a small radius R is favoured by a large ionie polarizability (giving
a large value of e, - eopt ) ane a large effeetive mass m*. Recall from
eqn (4.24) that a large effective mass implies a narrow band and therefore a
small hopping integral.
Anderson localization
Anderson localization refers to the localization of states by the presence oC
disorder in the system. It is a wave mechanical phenomenon that arises in
other areas of wave propagation science as well, such as acoustics. The
phenomenon is best iIIustrated by considering the following problem.
Consider a simple cubic s-band model in which the hopping integrals
between neighbouring sites are Pand zero otherwise. If aI1 the on-site energies
were equal to B say then the eigenvalues of the system would be
E(k) = B + 2P(cos kxa + cos kya + cos kza) (12.13)
where k is the wave vector of the state. This result follows immediately from
Bloch's theorem as described on p. 78.
Suppose we now allow the on-site energies to assume random values in
some range from - V to + V. What happens to the eigenstates of the system?
It is cIear that we can no longer appeal to Bloch's theorem because the
random variation in the on-site energy has destroyed the translational
symmetry, even though the atoms are still arranged on a simple cubic
structure. Let the on-site energy at site i be B" where - V:5; B, :5; V. For large
variations in Si and small hopping, p, the electrons will become trapped
(Iocalized) where the potential is locally attractive, i.e. where Si is large and
negative. In order for the electron to escape it must tunnel (i.e. hop) to
another state of the same energy. The probability of whether such an escape
is possible depends on how far the e1ectron has to go before it finds another
state of the same energy; The further it has to go the smaller the hopping
integral. The larger V is the less likely it is that a state of the same energy
exists nearby. States that are cIose in energy are then far apart in space, so
that their overlap is exponentially small. Anderson showed that when zIPl/V
is less than some critical value of order unity (where z is the coordination
number, i.e. 6 in our case) then all states become localized. The wave
functions of the localized states have an exponential envelope centred on
particular sites and the average characteristic decay of the envelope is called
the localization length. In the limit of strong disorder the localization length
is comparable to the atomic spacing. From a wave mechanical point of view
the disorder leads to destructive interference of a11 states in the Iimit of very
strong disorder.
At intermediate levels of disorder there are both localized and delocalized
sta tes in a bando The sta tes that are most susceptible to localization are those
at the band edges beca use they involve correlations of the electron phase
throughout the specimen. Thus the first states to become localized are at the
band edges, as shown in Fig. 12.6. Mott showed tnat it is not possible for
Metal-insulator transitions. oro what is a metal? 225
D(E)
Delocalized
states
Mobility edge
--.E
Fig. 12.6 A schematic density of states in an amorphous material, showing
mobility edges separating localized and delocalized stateso
localized and delocalized states to coexist at the same energyo The energy
separating localized and delocalized sta tes is ca11ed a mobility edge. In energy
regions where the states are localized the electronic conductance is zero at
O K. At finite temperatures electrons can move from one localized state lo
another by thermally-activated hopping. Thermal activation is necessary
because in order for the wave functions of two localized states lo have any
significant overlap, which is necessary for a finite tunneling probability, they
must be quite close spatia11y. Bul if they are close spatially they are unlikely
to have the same energy and therefore a phonon must be involved in the
hopping process. The involvement of a phonon requires thermal energy. As
the disorder increases so the mobility edges at the band edges move inwards
towards the centre of the bando When the Anderson criterion is reached the
mobility edges meet and a11 states are then localized. The material is then
an insulator regardless of where the Fermi energy lies in the bando
When a mobility edge passes Ihrough the Fermi energy the material is
transformed from a metal into an insulator. This is ca11ed the Anderson
transition and it is one example of a metal-insulator Iransition which we
tum to next.
Metal-insulator transitions. oro what is a metal?

On p. 20 we gave a working definition of a metal as a material that has a
nonzero electronic conductivity at O K. We refined this definition a little by
bringing in the temperature dependence of the conductivityo In the applica-
tion of band theory to crystalline materials we have seen that a metal is a
material in which there is no energy gap between occupied and unoccupied
states. If a band is fu11 there is no way of upsetting the balance between
sta tes moving in the + k and - k directions, and hence there is no current.
On the other hand a partia11y full band is always predicted by band theory
'10 be a metal: a metal is a material with a Fermi surface (p. 87).
E E
Unoccupied
Occupied
+--====----.. k + - - - - - -.. k
f----------
Insulator ... Metal
Fig. 12.7 A metal-insulator transition caused by the removal 01 band overlap.
On the right we have the metallic state with overlapping bands. Under pressure
the bands move apart giving rise to nonoverlapping bands separated by a gap:
an insulator at O K.
The simplest form of metal-insulator transition in crystals is from a metal

in which the conduction and valence bands overlap to an insulator in which
they do no!. This is sketched in Fig. 12.7. The metals of group HA would
be insulators if it were not for the overlap between the valence and
conduction bands. U nder pressure the overlap between the bands in Ba, Sr,
and Ca diminishes and a metal-insulator transition is expected when the
overlap ceases. Such a transition has been observed in another divalent meta~
Yb. If there were no interactions between electrons the transition would be
continuous; the separation between the valence and conduction bands would
decrease uniformly to zero at which point an infinitesimal number of
electrons and holes would appear and then increase smoothly. However,
Mott showed that interaction between electrons and holes invalida tes this
conclusion and at a transition of this kind there must be a discontinuous
change in the number of current carriers. The argument was that the first
electrons and holes to be formed when the bands cross are not free to conduct
electricity beca use they are strongly attracted to each other. A bound
electron-hole pair is called an exciton. Physically it means that the electron
that is now in the conduction band, having just left the valence band where
it was before the band crossing occurred, is still attracted electrostatically to
the hole it has left behind. In order to conduct electricity the electron and
hole have to break free of each other, i.e. the exciton has to dissociate. As
more excitons are created by further band crossing screening of the
electron-hole attraction becomes possible and this liberates further electrons
and holes which can contribute to more screening. This cooperative process
therefore leads to a discontinuous change in the density of free carriers at a
critical degree of overlap of the two bands.
Approximately half the known binary compounds are predicted to be
metallic by band theory when they are in fact insulating. A good example
is NiO. Local density functional band theory ca1culations for NiO place the
Fermi energy through the d band of the Ni ions: since the d band is partially
occupied the material is predicted to be a metal. In fact NiO is a
semiconductor. What has gone wrong?
In band theory we work with an average potential felt by a typical electron
in the solido Each electron moves in an effective potential defined by the
mean electrostatic field of aH the other electrons and the nuclei. To see that
this kind of averaging can break down consider the following thought
experiment. Suppose we take a perfect Cs crystal, which has one electron
per atom in the conduction band and is therefore metallic. Imagine we dilate
the lattice parameter of the crystal from its equilibrium value to 1 m! We
would agree that at sorne point in this dilation process the crystal must
become an insulator beca use certainly when the atoms are 1 m apart they
are not interacting. But band theory says that the crystal remains a metal
because at all dilations the energy difference between occupied and unoccupied
states is vanishingly small. Now look at this thought experiment from the
other way. Why is the crystal with a lattice parameter of 1 m an insulator?
Because to transfer an electron from one atom to another we have to supply
an ionization energy, 1, to remove the electron and then we recover the
electron affinity, A, when we add the electron to a neutral Cs atom. The
energy cost in this process is
U=l-A (12.14)
which is electronvolts of energy. Band theory ignores terms such as these.
In NiO the energy U is the energy to transfer an electron from one Ni 2 +
ion to another
(12.15)
AH metal-insulator transitions may be thought of as a competition
between the electronic kinetic energy favouring delocalization of the electron
and metaHic behaviour, and a potential energy favouring localization of the
electron and insulating behaviour. If an electron is to become itinerant it
has to overcome an electrostatic energy U tending to localize it. But once
the electron is free it can gain energy, relative to the energy of an atomic
sta te, by entering an unoccupied low Iying Bloch state at the bottom of the
bando This is the energy of bond formation and it corresponds, crudely, to
the reduction in the kinetic energy of the electron through delocalization.
We now try to quantify these considerations.
The characteristic length scale for an electron in an atomic system is the
radius of a hydrogenic type orbital in the medium of dielectric constant e
and effective mass m*
a* = (m/m*)w o (12.16)
where ao = (6 oh2 )/(nme2) is the Bohr radius (0.529 Á). The characteristic
potential energy for the e1ectron in this sytem is e2/4n6ow*. The kinetic
energy of the electron gas, with density n, is proportional to h2n 2/3 j(4n 2m*).
Therefore for metallic behaviour we require
h2n 2/3 j(4n 2m*);;:: Ce2j(4neoea*) (12.17)
where C is an unknown constant. Therefore,
a*n 2 / 3 > C e2 m*n (12.18)

- eoeh 2
or, using ejm* = a* j(ma o) and ao = (e oh2 )j(nme 2 ),
(a*n '13 )2;;:: C
(12.19)
n•1t3a*
H =026
•
lO'
~
-<
~ Ge:Sb.
• Ge:P Ge:As
"
~
CdS:el
CdS:ln
:a" Si:Sb Si:P
!! HMPA:Li Si:As
~ lO GaP:Zn GaAs:Mn
'y
I\l WSe,:Ta
NH3:U eNH,:U
es Ar:Na
Rb Li
Na•
1 W03:Na
Ar:eU\
14 16 18 20 22
3
Log [critica! concentration, n.] (cm- )
Fig. 12.8 A logarithmic plot 01 the effective Bohr-orbit radius, a', vs the critical
concentration lor the transition to the metallic state, ne , in a variety 01 doped
semiconductor systems. Open circles represent experimental data lor supercritical
vapours 01 alkali metals. The solid line is a*n~13 = 0.26. From Edwards and Sienko
(1981).
where D = e '/2 is another constan!. Equation (12.19) is the form found for
all criteria of metal-insulator transitions based on a wide range of physical
models. The best empirical estimate for the constant D is ~O.26. As shown
in Fig. 12.8 the criterion works over 8 orders of magnitude for the electro n
density. Physically the criterion asserts that anything becomes metallic if it
is compressed enough. Under sufficient compression the kinetic energy cost
of localizing electrons is prohibitive and the material transforms from an
insulator into a conductor.
One of the first physical models to be given for a metal-insulator
transition was the M ott transition. This is illustrated in Fig. 12.9 for P doping
of Si. P has five valence electrons and when it sits at a substitutional site in
Si only four of these electrons are taken up in bonds to the neighbouring Si
atoms. At low temperatures the fifth electron enters a hydrogenic like orbit
of radius a* given in eqn (12.16). In Si a* is about 25 A and the binding
\03
.-
JO'
•
I
.----
./
•
O'min------
10
'eu
,
i
€
b
!
Insulator Metal
10-1
10-2 ,
O~----~2~----~4------~6~--~8
Donor concentration (1018cm-3)
Fig. 12.9 The zero-temperature electrical conductivities 01 phosphorus-doped

silicon samples across the metal-insulator transition. From Rosenbaum el al.
(1980).
230 Where band theorv breaks down
energy oC the fiCth electron to each P + core is about 0.045 eV. Thus at room
temperature the P ions are ionized and they are said to have donated their
fiCth electrons to the conduction band oC the Si. This is the basis oC 'n-type'
doping oC Si by group V elements. But at very low temperatures the electrons
are bound to the P + cores. In that case the material continues to be an
insulator at low temperatures. Figure 12.9 shows the electrical conductivity
oC P-doped Si at very low temperatures as a Cunction oC the P donor
concentration. lt is seen that at a critical concentration oC about 3.8 x 10 18
cm - 3 there is an abrupt transition Crom insulating to metallic behaviour.
At the high donor concentration the hydrogenic orbits oC neighbouring P
atoms overlap. Electrons can then tunnel Crom one ion to the next and screen
the interaction between the orbiting electron and the P + ion coreo In this
way the screening liberates the electrons and the more that are liberated the
greater the screening and so on. This is the same kind oC cooperative process
leading to an abrupt transition that we discussed aboye in connection with
band crossing transitions. Once the screening gets underway it takes off in
an avalanche and the material becomes a metal. Mott gave the Collowing
simple quantitative analysis.
The screened potential oC each ion core in the Thomas-Fermi approxi-
mation (see p. 154) is
(12.20)
where kii is the Thomas-Fermi screening length
2 -
4n l / 3
kT F---· (12.21)
a*
The condition Cor metallic behaviour is that the screening length is less than
a*. ThereCore,
TF <
I
k- - a* (12.22)
or
(12.23)
which should be compared with eqn (12.19).

The Cs thought experiment that we discussed aboye can be analysed in
the same way, and the transition is again discontinuous. The transition Crom
metallic to insulating behaviour as we dilate the Cs crystal is due to the
reduction in the screening oC the interaction between the electrons and the
Cs + ions. Eventually, at a critical dilation it is energetically preCerable Cor
the valence electrons to condense onto the ion cores and Corm an insulating
state. At that point the potential wins over the kinetic energy and the valence
electrons are bound to the ion cores. Clearly we do not expect to find sorne
Cs atoms ionized and sorne neutral in the dilated crystal. They are either all
ionized or al1 neutral. Thus at the transition there is an abrupt change in the
density of the free electrons: it is a first order phase change. Such a phase
change may be accompanied by a change in crystal structure of the material.
Another physical rnodel was proposed by Hubbard and is cal1ed the
Hubbard transition. Consider the reaction of eqn (12.15) in NiO again. The
energy U associated with this reaction is cal1ed the Hubbard U. If the two
ions are isolated ions in space the Hubbard U is the energy required to
remove an electron from a Ni2+ ion (making it Ni 3 +) and transferring it to
another isolated Ni2+ ion (making it Ni +). This energy is shown at the left
of Fig. 12.10. In a crystal an electron on a Ni + ion can travel to another
Ni 2 + ion and there is a band of states in which the electron can move.
Similarly the hole on the NP + ion can move to another Ni2+ ion in the
crystal and there is a band of states in which the hole can move. These
electron and hole bands are known as Hubbard bands. They are not the
Electron band
Ni+
t
1...
~
~ Ni3+
Holeband
msulator Metal w
Fig. 12.10 Schematic illustration of the electron and hole bands in NiO as a
function of the band width W. At W = O we have isolated Ni+ and Ni 3 + ions
separated by U = I - A. As the band width. W, increases the gap, U - W,
between the electron and hole band decreases until at the broken line the bands
overlap and the material becomes a metal.
normal bands of band theory. The Hubbard bands have a width Wand the
energy gap between them is U - W. When this gap is zero the material
becomes a metal. Therefore metallic behaviour sets in when
W~U. (12.24)
For a hydrogenic-like state it can be shown that
5é
U".,- (12.25)
8a*
and the bandwidth in a simple cubic lattice is
W = 121PI = 121PoI e- R /.· (12.26)
where Pis the hopping integral and R is the first-neighbour separation. For
a hydrogenic-like wave function
IPol = !(1 + R/a* + (R/a*)2) é . (12.27)

6 a*
Therefore the condition for metallic behaviour, eqn (12.24), becomes
R/a* ,,; 5.8 (12.28)
or
(12.29)
which should be compared with eqn (12.19). It is remarkable how similar
this condition is to Mott's condition, eqn (12.23), C'onsidering how different
the physical pictures are. But there is one important difference. The Hubbard
transition is expected to occur continuously as sorne parameter, like the
lattice constant, is varied, in contrast to the discontinuous change of the
Mott transition.
The final model we shall consider for the insulator-metal transition is the
polarization catastrophe which was proposed before the advent of quantum
theory by Goldhammer in 1913 and Herzfeld in 1927 (see Logan and
Edwards 1985). The physical basis of the criterion is somewhat controversial
but it has two points in its favour. The first is that it separates almost all
the elements into metals and nonmetals correctly. The second is that it yields
a criterion of the same form as eqn (12.19).
The criterion is based on the Clausius-Mossotti relation of c/assical
electrostatics. This relates the atomic polarizability, tX, the atomic volume, v,
and the relative permittivity 6
2(47ttX/3v) + 1
6 = -:'--'-:-,'-:-:---:- (12.30)
1 - (47ttX/3v)
or
6 - 1 47ttX
--=- (12.31)
6 +2 3v
Nb Mo n: 80s
oLi B e Ru Ir Pt
'Zz e oW Re BRh SPd R!: oAl
ln~08(.)
Oy BHf 1'0 o Mo Fe Co . 6 8'" 08(1)
.~ ~. "",-",-~"h,
o o o o Ca Cd R
......
ONI
~"""O"~(am)
Os. °n v Sn(jl)
1.0 °Rb
-
------
k
R.
Ou.
-------
-- --
-
_______
Nonmetals
-----n'
....
, ,~.
I.
el)
....
....
o p (w)
p (w)
_ __
(y)
.'"
Br,m
~
Rn
CI,mo
N,m<: oKr
e 0,0) >Ar(')
0.1
011,0) F,m\
oNe
olle
0.01
IAllAllIBIVBVBVlBVIlBVIll VIll lB IIB UIA IVA VA VIA VllA O
Fig.12.11 A plot of R/V, eqns (12.32), (12.33), for elements 01 the periodic table, under ambient
conditions. R is the molar refractivity and V is the molar volume. The lilled circles represent elements for
which both R and Vare known experimentally. The open circles are lor elements for which only Vis known
experimentally, and R is calculated. From Edwards and Sienko (1983).
In order for this relation to apply, the material must be an insulator. But if
the atomic volume is decreased, by applying a pressure for example, we can
reach a situation in which
41tCl -> 1. (12.32)
3v
There is then a 'polarization catastrophe' and & -> oo. The material then
becomes a metal. Avogadro's number times 41tClj3 is called the 'molar
refractivity', R. Avogadro's number times the atomic volume is the molar
volume, V. Thus
R>v=>metal}
(12.33)
R < V => insulator.
The remarkable (and controversial) feature about this criterion is that it
involves the molar refractivity of the gaseous atomic state and the molar
volume of the condensed state. The success of the criterion may be seen,
Insulator
1~ __L -__~__~~~__~__~__~~
o 2 4 6
Donor concentration(1018cm-3)
Fig. 12.12 The left hall shows the divergence in the dielectric susceptibility as
the phosphorus-dopant concentration in silicon approaches the critical value lor
the insulator-metal transition (seen in Fig. 12.9). The right hall shows the
conductivity at O K rising abruptly Irom zero at the transition. From Hess el al.
(1982) .
References 235
however, by examination of Fig. 12.11 (p.233). Apart from a few marginal
cases it is successful for all the elements of the periodic table!
The polarization catastrophe that takes place in P-doped Si at the critical
dopant concentration is shown in Fig. 12.12. For hydrogenic-like atomic
states the atomic polarizability is given by
(12.34)
Putting this in the condition for the catastrophe, eqn (12.33), we obtain
n'/3 a· 2': 0.37 (12.35)
as the condition for metallic behaviour. Again this should be compared with
eqn (12.19).
References
Edwards, P. P. and Sienko, M. J. (1981). J. Am. Chem. Soc., 103, 2967.
Edwards, P. P. and Sienko, M. J. (1983). Chemistry in Britain, 19, 39.
Harrison, W. A. (1980). Electronic structure and the properties of solids. W. H.
Heine, V. (1971). J. Phys. C: Sol. State Phys., 4, L221.
Hess, H. F., De Conde, K., Rosenbaum, T. F., and Thomas, G. A. (1982). Phys. Rev.
B, 25, 5578.
Logan, D. E. and Edwards, P. P. (1985). In The metal/ic and nonmetal/ic states of
matter (ed. P. P. Edwards and C. N. R. Rao), Chapo 3. TayJor & Francis, London.
Moss, S. e. and Graczyk, J. F. (1970). Proc. Int. Con! on the Physics ofSemiconductors,
Cambridge, MA (ed. S. P. Keller, J. C. HenseJ, and F. Stem), p. 658. United States
Atomic Energy Commission, Washington, D.e.
Mott, N. F. and Davis, E. A. (1979). Electronic processes in non-crystalline materials.
CJarendon Press.
Rosenbaum, T. F., Andres, K., Thomas, G. A., and Bhatt, R. N. (1980). Phys. Rev.
Lett., 43, 1723.
Wooten, F. and Weaire, D. L. (1987). Solid State Phys., 40, 1.
Problems
1. Equation oJ state at o K. At °K the equation of state is given by

P = -(aE./aV).
Using the universal binding energy curve
E. = IE~lg(a)
where
g(a) = -(1 + a + O.05a') e-a
a = (rws - rw,)/I
1=( IE~I )1 1'

121tr~sBe
and rY" is the Wigner-Seitz radius, IE~I is the cohesive energy and B' is the bulk
modulus all at the minimum in the binding energy curve E.(rw,), show that
~ = 3«V/Vo)II' - 1) (1 _ O.l5a + O.05a') e-a

B' (V/Vo)'I'
where Vo is the equilibrium volume at zero pressure. This is the equation of state of
°
the material at K. Sketch P as a function of V.
Using the definition of the bulk modulus
B = - V(op/aVh
show that the bulk modulus varies with pressure as follows
2.3
(aB/aPh = 1 + -~,
3
where ~ is the anharmonicity parameter ro,s/I.
2. Melting point oJ metals. We can use the universal equation of state to deduce
a simple relation for the melting point, TM , of a metal if we assume a simple criterion
for melting. The criterion we adopt (Guinea, F., et al. (1984). Appl. Phys. Leu., 44,1)
is th.t melting occurs when the amplitude of therma! vibrations exceeds the inflection
point in the universal binding energy curve. Beyond this distance the restoring force
decreases with increasing separation. In the Debye model, above the Debye tempera-
ture, the rms amplitude of vibration is given by
'11' 1
(u > = (o.827kT)1 2
B rws
Problems 237
where k is the Boltzmann eonstant, T is temperature, r~s is the equilibrium
Wigner-Seitz radius, and B is the bulk modulus. The inflection point in the universal
binding energy curve is given approximately by
rws - r~s = 1
where I is the sealing length given by
_ (
1-
IE~I )1/2 .
121tr~sB
Here, IE~I is the eohesive energy at equilibrium. Show that

TM ~ O.032IE~I/k.
Note that the melting point is directly proportional to the eohesive energy. ·An
empirieal eorrelation between eohesive energy and melting point has long been
known. Comment on the physieal validity of the eriterion for melting that we have
used (i.e. do you believe it is right regardless of the agreement with experiment?).
Compare it with' the Lindemann eriterion whieh asserts that melting occurs when
the amplitude of vibrations exceeds a certain fraetion (about 15 per eent) of r~s,
rather than the infteetion point l. Show that the Lindemann eriterion predicts
where " ~ r~s/I is a measure of the anharmonicity in the crystal and varies by over
a factor of 2 between the metals.
3. Consider a metal that can exist in either the f.e.e. or the h.e.p. erystal strueture
with the same equilibrium density at zero pressure. Using the universal equation of
state whieh of the following statements is (are) true at zero pressure:
(a) the eohesive energies in the f.e.e and h.e.p. phase are the same
(b) the bulk moduli in the f.e.e. and the h.e.p. phases are the same
(e) the Wigner-Seitz radii in the f.e.e. and h.e.p. phases are the same
(d) the ratio ofthe eohesive energy to the bulk modulus is the same in the two phases.
J ustify your answers.
4. Radial solutions of the hydrogen alOm. The Sehriidinger equation for the
hydrogen atom is:
This is sol ved by exploiting the spherical symmetry of the atom to express the solution
in a separated formo The solutions are as follows:
'I'.'m(r, (J, t/J) ~ N"F,,(~) l'ím«(J, t/J)

naB
238 Problems
where N.,is a normalization constant to ensure that the integral of 1'I'.'m(r, (J, </>W
over all space is unity:
2 ( (n - / - 1)! )1/2
N., = n 2 [a.(n + /)!]3 '
and a. is the Bohr.radius = (41t6 0 h2)/me'. Y¡m«(J, </» is a spherical harmonic and is a
function of only the angular variables (J and </>. The radial part of the solution is
g.,(r) = N.,F.,(2r/na.) and is normalized so as to satisfy
Loo r 2(g.,(r»2 dr = 1.
The functions F.,(p) are given by the following expressions:

F.,(p) = e-pI2p'L:~j l(p)
where
2r
p=-
na.
and L:~j l(p) are assocíated Laguerre polynomials given by
.-'-1 [( + /),]2 •
L:~jl(p)= L (-lr2' n. p .
'=0 (n - / - 1 - k)! (2/ + 1 + k)!k!
Using these formulae obtain the Is, 2s, 3s, 3p, and 3d normalized radial solutions,
and sketch them. Comment on the numbers of nodes.
Armed with these wave functions we can derive the mean radial position and mean
square radial position of an electron. These are defined as follows:
(r) = falllpace
rl'l'.'m(r, (J, </>W dr
and
(r 2 ) = faU.pace
r 2 1'1'.'m(r, (J, </»1 2 dr
The results are as follows
(r) = a B [3n 2 - /(/ + 1)]

2
and
n2
(r 2 ) = a~ - [5n 2 + 1- 3/(/ + 1)].
2
Notice that for a given n the means decrease as the angular momentum increases.
Does Ihis accord with your sketches? Compare the value of (r) for the 3d, 4s, and
4p orbitals.
Retuming to the SchrOdinger equation, we can understand why in a given shell
(Le. a given n) the orbitals are more contracled the higher the angular momenlum
by the following argument. The functions F.,
are solutions of Ihe radial part of the
Problems 239
Schrodinger equation
h2 1d
-----(rF¡)
2 2
2
+ (h 2(1(12+ 21» - -e2-) F¡ = EF¡.
8n m r dr 8n mr 4n80r
The total energy of the electron, E, consists of its kinetic energy mv 2/2 and its
potential energy V(r). The total energy is conserved. The kinetic energy can be
resolved into two components. The lirst component arises from radial oscillations of
the electron through the nueleus and equals mv: /2, where v, is the radial velocity.
The second component arises from the cireulating motion of the electron around the
nuc1eus with angular momentum L. The kinetic energy associated with tbis is
L 2/2mr 2 • Tbe total energy and tbe angular momentum are conserved
E = mv:/2 + V(r) + L'/2mr 2.
This equation is true both for a elassical partic1e and a quantum partic1e. By
expressing L 2 as h2/(1 + 1)/4n2, and mv;/2 as tbe radial part of _h 2V 2/8n 2m obtain
the aboye radial SehrOdinger equation. Using the conservation of tbe total energy
and the independence of the total energy on / give a qualitative argument for why
the orbitals contraet as / inereases.
5. The angular so/utions of the hydrogen atom. The eigenfunctions of the angular
dependent part of the SchrOdinger equation for the hydrogen atom are the spherical
harmonics lí... The spherieal harmonics for / = O, 1, 2 are as follows
1
/= O
Yoo = (4n)112
3)112
Yi-I =(8n sinOe- l•
3)112 cos ()
Yio = (4n /= 1
1';-2 = (-
15 )112 sin 2
(} e- 2i,;
32n
15)1/2 . .
lí-I (8n smOcosOe-'·
=
lío = (-
5 )112 (3 cos2() - 1) / = 2.
16n
15)112 sin Oeos Oel.
líl = - (8n
240 Problems
The normalization factors ensure that the integral of llí.. (/i. 4>)1' over the surface of
a sphere is unity. These solutions are not convenient for solid state applications
because (i) they are complex and (ii) they are expressed in spherica1 polars rather
than Cartesian coordinates. Show that. by taking linear combinations of the functions
ror each 1, the Px' P" P., dxJ" dyZ', dxz ' d3z 1- r2, dX 2_y2 orbitals may be obtained. For
each of these orbitals give the angular momentum about the z-axis. m. from which
they are derived.
6. Let eh e,. e3 be three orthonormal vectors. Let " •• "2' "3 be another set of three
orthonormal vectors. Write down the components of lhe matrix L that relates
el' e 2 , el lo "1' "2' "3' i.e.
(
::) = (~:: ~:: ~::) (::).
e3 L3. L 3, L33 "3
7. Consider a three-dimensional Hilbert space with three orthonormal basis states

le.). le,). and le3)' Let 1".). la,). and la3) be another set ofthree orthonormal basis
states. Write down the transformation that relates le.). le,). and le3) to la.). la,).
and la3) in terms of bras and kets.
8. In"ariance of the trace of an operator. Consider an N-dimensional Hilbert

space. Let two alternative basis sets for the space be 11).12).13)•...• IN) and
11').12').13')•...• IN'). Let the matrix elements of an operator H in the two basis
sets be denoted by H,! = (iIHIj) and H,.}' = (¡'IHlr). Show that
H" + H" + H33 + ... + HNN = H., •. + H"" + H3·3, + ... + HN·N·.
This result is used in Chapter 2 and elsewhere in the book. It means that the trace
of an operator is independent of the basis se!. (The trace is the sum of the diagonal
elements in sorne representation.)
9. Hydrogen molecule with overlap. Consider the ground state of the H, molecule.
Let 11) and 12) be atomic 1s states on the hydrogen atoms and assume they form a
basis in which to expand the molecular state 1")
1") = c.11) + c,12).
We assume the basis states are normalized so that (111) = (212) = 1. and we let the
overlap between 11) and 12) be S: (112) = (211) = S. Let the Hamiltonian matrix
elements be (1IHll) = (2IHI2) = Ea and (1IHI2) = (2IHI1) = p.
Calculate the energies of the bonding and antibonding states and the normalized
state vectors to which they correspondo Compare your results with those obtained
in eqn (2.29) for the case that S = O.
10. A trimer. A molecule consisting of three identical atoms may exist in two
alternative forms: either as an equilateral triangle or as a linear chain. The bond
lengths are constant and the Hamiltonian matrix elements are aH zero except those
between nearest neighbour atoms which are aH -1131. Determine the molecular
Problema 241
configuration with the lower electronic energy as a function of the total number of
electrons in the molecule from zero to six.
11. Bond order in the AB molecule. Let the basis set for the molecular state vectors
consist of one atomic state on each atom and assume the basis set is orthonormal.
Let the on-site Hamiltonian matrix elements for the A and B atoms be EA and E B
and let the hopping integral be p. Calculate the bond order of the AB bond when
there are O, 1, 2, 3, and 4 e1ectrons in the molecule. How are your results affected
when EA = E B ?
12. Bond order in a trimer. Consider again the two alternative configurations of
the trimer discussed in Problem lO. In the linear chain the two bonds are equivalent.
Similarly in the equilateral triangle the three bonds are equivalent. Calculate the
bond orders in representative bonds of the linear chain and the equilateral triangle
for the cases where there are O, 1, 2, 3, 4, 5, and 6 electrons in the molecule and
compare your answers graphically. What is the ratio of the bond order in the linear
chain to the bond order in the equilateral triangle when the total number of electrons
in the molecule is the value you computed in Problem 10 where the total electronic
energies are equal? Could you have deduced this ratio without computing the bond
orders explicitly?
13. Group velocity. Consider an infinite linear chain with periodic boundary
conditions applied to the ends ofthe chain. The band structure E(k) was derived in
eqn (3.19) as
E(k) = el + 2p cos ka
where k is a wave vector lying in the first BriUouin zone between -nla and nla and
a is the spacing of atoms in the chain. el is the on-site Hamiltonian matrix element
and p is the nearest-neighbour hopping integral and a1l other Hamiltonian matrix
elements are zero. The group velocity of the electron in the eigenstate '1', is defined by
where
h d
p = 2m dx
is the momentum operator and m is the electron mass at rest. Using Bloch's theorem
'I',(x) = eltxu(x)
where u(x) = u(x + a) and the Schrodinger equation for 'I',(x) show that
v,(x) = 2n dE(k).
h dk
Hint. Put Bloch's form of the eigenstate 'I'.(x) into the SchrOdinger equation
h 2 d2~ (x)
--2- .. + V(x)'I'.(x) = E(k)'I'.(x)
8n m dx
242 Problems
and differentiate the resulting expression with respect to k. Explore the relationship
between what you get and l/m<'I'.lpl'l',). Hence, show that eigenstates labelled by
k and - k have equal and opposite velocities. Show that the group velocity is zero
at the Brillouin zone boundaries. Calculate the maximum group velocity for a = 1 A
and IPI = 1 eVo
14. Electron current. We can use the time dependent Schrodinger equation to get
a more physical picture of the meaning of the group velocity as follows. Consider
again the infinite linear chain of the previous problem. The eigenstates are
1 N
1'1'.) = N /2 I c}'lU)
j.,
'
where N tends to infinity. In the kth occupied eigenstate the amount of electronic
charge associated with atom m at time t is - 2elc!:,)(t)1 2 , where e is the electronic
charge, and the factor of 2 is because the state is occupied by 2 electrons. This quantity
does not change with time as may be seen from the expression for c!:,)(t)
c~)(t) = e"'" e-((2.1E(·)')/·)/N ' / 2
from which it follows that - 2elc!:')(tW = - 2e/N. In any time interval as much
electronic charge flows onto atom m from atom m - 1 in the kth occupied eigenstate
as leaves atom m and flows onto atom m + 1. To calculate the electron current flowing
in the kth eigenstate through atom m we must look at the current flowing between
atoms m - 1 and m or between atoms m and m + 1. We will see Ihat this current is
proportional to the group velocity v,.
Using the time dependent Schrodinger equation show that the derivative of
2e1c!:,)(t)1 2 with respect to time (which we know is zero) may be written as follows
d 41te
-2e -d Ic~)(t)l' = --:-h I c~)(t)*H_jcj')(t) - c!:,)(t)Hj_cj')(t)*
t , j
where the sum is taken over atomic states U) which are coupled to the atomic state
1m) through the Hamiltonian matrix elements H..j and Hj... Each tenn in the
summation on j describes electron current flowing into or out of site m from site j.
Thus, the current flowing between sites m and m + 1 is obtained by selecting the
tennj = m + 1 in the summation. Show that Ihis current is -(81tep/hN) sin(ka) and
that it is proportional to the group velocity for the kth eigenstate.
15. Local densities of states and bond orders in an infinite linear chain AH al/oy.
Consider an infinite linear chain of A and B atoms in which there is nearest-neighbour
hopping only, with hopping integral p, and where the on-site Hamiltonian matrix
elements on A and B atoms are eA and eB' The A-B bond length is a. Calculate and
sketch the band structure and show that there are two bands and that the Brillouin
zone may be defined to Iie between -1t/2a to + 1t/2a. By examining the corresponding
eigenstates show that states in the lower band are bonding and those in the upper
band are antibonding. Sketch the total density of states and the local densities of
states dA (E) and dB(E). For a given Fenni energy would you expect the amount of
e1ectronic charge associated with an A atom to be the same as that associated with
a B atom? If no!, how would you calculate the difference? Sketch the bond order
Problems 243
between neighbouring atoms as a function of the Fermi energy varying from the
bottom of the valence band to the top of the conduction bando (Note that when 1
say 'sketch' 1 am not asking you to derive the functional forms (which are pretty
awful!) but merely to draw what you think the function should look like based on
sound reasoning).
16. Bond energy in injinite linear chain. For the inflnite linear chain of Problem 13
calculate the bond energy per atom in two ways:
(a) using the bond order
2 sin kF(m - n)a
P... =-
1t m-n
(b) using the local density of states
1 1
d(E) = ;;"'"(4""::p2;;-_--:-(E=--_-"':-:;)2::-;)17.0¡2
17. Interatomicforces. Assuming tbat the on-site Hamiltonian matrix elernents do

not change when an atom is displaced, and that the hopping integrals depend only
on the positions of the two atoms they connect, show that the derivative of the band
energy with respect to the x-coordinate of atom m is given by
Give a physical interpretation of this formula. This formula is known as the

Hellmann-Feynman theorem.
18. Bloch oscillations. Consider a simple cubic lattice s-band model with hopping
integral P between nearest neighbours. A constant electric fleld (e .. o, O) is applied
to the crystal where the x, y, z axes are aligned along the crystal axes. Show that
electrons in the crystal perform harmonic oscillations along the x-axis and that the
distance, s, travelled by each electron in a half period is 4p/ee. where eis the electronic
charge. If IPI = 1 eV and e.
= 1 V rn -1 show that s = 4 m. If the band is partially
occupied the crystal is metallic. But if the electrons are performing harmonic
oscillations under the influencc of a DC electric fleld how does the crystal conduct
electricity?
19. s-band modelfor fc.c. crystal. Consider an f.c.c. crystal with one atom per lattice
site and lattice constant a. Let each atom be associated with an s state and let the
on-site Hamiltonian matrix elernents be '" and the hopping integrals between nearest
neighbours be P(P < O). AlI other Hamiltonian matrix elements are zero. Show that
the band structure is given by
E(k) = IX + 4P(cos(k.a/2) cos(k,a/2) + cos(k,a/2) cos(k,a/2)

+ cos(k,a/2) cos(k.a/2».
244 Problems
In the limit of small band Ii!lings show that E(k) becomes
E(k) = ce + 12P - pa 2 k 2 •
What is the functional form of the density of states in the limit of smal! band fillings?
As the Bril!ouin zone is lil!ed with occupied states where are the last parts of the
zone to be lil!ed?
What are the maximum and minimum values of E(k)?
Using the moments theorem, calculate the lirst three moments of the local density
of states and sketch how you think the local density of states curve should look
throughout the entire energy range of states. Pay particular attention to the positions
of the band edges, how the density of states tends to zero at the band edges and the
symmetry of the density of states curve about the centre of the bando (N.B. You are
not being asked to evaluate the entire density of states analytical!y, but to use the
information available from the moments and the limits of the band strueture to sketch
the function.)
20. Band gap in s-band model of N aCI structure. Consider an AB crystal with the
NaCl structure. An atomic s state of energy EA is associated with each A atom and
an atomic s state of energy EB with each B atom. Assume EA > EB • The Hamiltonian
matrix elements between nearest neighbours are assigned the value p ( < O) and all
other Hamiltonian matrix elements are assumed to be zero. The atomic basis set is
also assumed to be orthonormal.
Usong Bloch's theorem show that the eigenstates of the AB crystal are as follows
1'I'(')(k) = N~/2 L (c~)(k)IRm' A) + cJr)(k)IRm +~, B») e"' Rm

3 m
where n is the band index (n = 1 or n = 2 because there will be only two bands), N,
is the (infinite) number of lattice sites in the crystal, Rm is the position vector of the
mth lattice site and ~ is the position of the B atom relative to the A atom in each
primitive unit cell. We may take ~ to be a/2[1 O O] where a is the lattice parameter.
Project the SchrOdinger equation HI'I'(')(k) = E(')(k)/'I'(')(k) onto lO, A) and I~, B)
in tum to obtain tbe following secular equations
(eA - E(')(k))c~)(k) + f(k) cJr) (k) = O
and
where
k,a k,a k,a)
f(k) = 2p ( cos-+ cos- + cos- .
2 2 2
Show tbat there are two roots to these equations
E(1)(k) = e - (d 2 + f2(k))'/2
and
E(2)(k) = e + (d 2 + f2(k))'/2
where e = (EA + E B )/2 and d = (EA - EB )/2. What are the energy ranges for these
two bands? What is the value of the energy gap? Sketch the total density of states
(do not try to calculate it!).
Problems 245
21. Weaire- Thorpe Hamiltonian parameters. Derive eqn (6.14) and eqn (6.15).
22. Hydrogen in metals. Hydrogen dissolves interstitially in many metals. Using

the uncertainty relation discuss whether hydrogen is expected to behave as a quantum
partide or a classical particle at room temperature. Assume that the interstitial hole
that the hydrogen sits in is about 1 A in diameter. Calculate the temperature, To, at
which the thermal energy of the partide, - kT, is equal to the kinetic energy of the
partide arising from the uncertainty in its momentum. How does hydrogen diffuse
through the solid at temperatures below and aboye To? What is To for an Fe atom
in b.c.c. Fe?
23. Free electron gas. For a free electron gas with a fixed number of electrons
occupying a volume Vand with a Fermi energy EF show that
. (JEF 2EF
(1) av = - 3V'
(ii) the average electronic kinetic energy is 3EF /5,
(iii) the pressure of the electron gas is 2ET/(3 V), where ET is the total electronic
kinetic energy,
(iv) the bulk modulus of the electron gas is IOET /(9 V).
Assume that the temperature is O K.
24. Free electron band. for f.e.c. lattice. Construct the first six free electron bands
for an f.c.c. lattice along the [1, O, O] direction in the Brillouin zone, ie. from
k = [O, O, O] to k = (2,,/a)[I, O, O] where a is the lattice parameter of the f.c.c. lattice.
25. Phillips-Kleinman pseadopotential. The pseudopotential arises from the require-

ment that the valence electron state must be orthogonal to atomic core states. A
smooth nodeless wave function e(r) is made orthogonal to atomic core states as
follows. We consider a new state 'I'(r) which looks like e(r) as well as having sorne
components of core states
(1)
,
where </>, is a normalized core state, a, is a mixing coefficient to be determined and
the sum is taken over all atomic core states. Show that the requirement that '1' is
orthogonal to each core state leads to the result that
(2)
The functions '1' are good approximations to the valence state wave functions.
Substitute'l' i.nto the SchrOdinger equation HI'I') = EI'I') and obtain
(3)
,
where HI</>,) = E,I</>,) has been used. Writing H as a,sum of a kinetic energy, T, and
246 Problems
a potential energy, V, show that eqn (3) may be rewritten as
(4)
The quantity in round brackets is the pseudopotential. The attractive negative real
potential Vis reduced by the positive potential
(why is it positive?). Notice that L (E - E,)Icf>,) <cf>,1 has the form of a sum of
projection operators. Each core state has a particular angular momentum and
therefore each projection operator in the sum acts on a particular angular momentum
component of 10. The eigenstates of eqn (4) are the pseudowave functions. The
remarkable feature about eqn (4) is that tbe eigenvalues, E, of this pseudowave
function equation are identical to the eigenvalues of the true wave function equation,
HI'I') = EI'I'). Demonstrate that this is so by multiplying both sides of eqn (4) by
<'1'1. Notice that the same result would have been obtained if the terms (E - E,) had
been replaced by anything else! Thus, the pseudopotential and the pseudowave
functions are not unique. This feature is exploited in modern 'norm-conserving'
pseudopotentials.
26. Free energy of an electron gas. Let f. denote the Fermi-Dirac occupation
probability that state n, of energy E.. is occupied, where
1
f. = ,
1 + exp«E. - EF)/kT)
where EF is the Fermi energy and k is the Boltzmann constant. The entropy of tbe
electron gas (in the independent electron approximation) is given by
s= -k ¿ f.ln f. + (1 - f.) In(1 - f.),

•
where the sum is taken over all states of the system and 'In' denotes the natural
logarithm. The Helmholtz free energy, A, is defined by A = U - TS, where U is the
internal energy
U = ¿f.E•.
•
Show that A is given by
A = NE F - kT¿ In(1 + exp«EF - E.)/kT»

•
where N is the total number of electrons in the system. All the thermodynamic
properties of the electron gas at a finite temperature may be derived from this
function. If you are feeling brave you can try to calculate the pressure in the electron
gas at a finite temperature byevaluating -(ilA/ilVh. (Remember though that N is
fixed and therefore EF changes when the volume is changed.)
Problema 247
27. Frequency dependence 01 electrical conductivity. Using the equation of motion
mdv/dt + mv/r = - e{
for an electron in an electric field { = {o el~' show that the conductivity at frequency
W is given by
<T(W) = <T(O)( 1 - iW,,)
1 + (wr)
where <T(O) = ne2 r/m. What is the physical meaning of the imaginary part of
<T(w)?
28. A materials processing problem. A materials engineer is trying to design a

composite of a dielectric and small Ag metal partides for laser applications. The
electrons in the metal particles are confined and the potential barrier at the surface
of the partides may be regarded as infinite. The partides are only about 100 A in
diameter and the quantization of the electron energy levels within them is significant.
The engineer has discovered that she can alter the shape of the metal partides, and
in particular she can make them cubic or spherical. But she knows that the shape
of the partides must have an influence on the distribution of energy levels within
them and therefore on the wavelength of the light at which the material will lase.
She comes to you for help. The question she poses is the following: if the partic1es
have the same volume what is the distributiQll of energy levels within them if they are
(a) cubic or (b) spherical?
We may treat the electrons in the metal as particles in a box, which is either cubic
or sphericaJ. The potential inside the partide we set to zero and outside the partide
the potential is set to + oo. The Schrodinger equation for electrons in the partide is
that of a free partide
___h2 V2 '1' = E'I'. (1)

8,,2 m
If the side of the cubic form of the particle is a, then show that the energy levels
within the partide are as follows
2
h 2
E, = - - 2 (n, + n,2 + n,)
2
(2)
8ma,
where nx , n" nz ~ 1 are integers.
Show that the radius of the spherical form of the particle is given by
3
a, = (:,,)'/ a,. (3)
To determine the distribution of energy levels within the sphere we have to solve
eqn (1) in spherical polar coordinates. Using the spherical syrnmetry of the partide
show that eqn (1) can be written in the following form
d 2R 2 dR 1(1 + 1) R +R= O
- +- - (4)
dp2 p dp p2
248 Problema
where p = kr and
k 2 = 8,,2 mE
(5)
h2
and I is the angular momentum (=0, 1,2, ...). Equation (4) has to be solved subject
to tbe boundary condition that R = O at r = a,. Tbe solutions oC eqn (4) are called
spherical Bessel functions, which are oscillatory functions of p. There is one solution
for each angular momentum I
R = Ai,(P), (6)
where A is a nonnalization constan!. The eigenvalues E are detennined by imposing
the boundary condition
j,(ka,) = o. (7)
Because each j,(p) is an oscillatory function of p there is more than one solution of
eqn (7) for each l. These solutions are labelled by n = 1,2,3, ... in order of increasing
p. The 11 lowest solutions are shown in Table 1.
Tabla 1 Eleven lowest solutions for eqn (7)
1= O 1= 1 1= 2 1= 3 1= 4 1= 5
n =1 3.14 4.49 5.76 6.99 8.18 9.36

n=2 6.28 7.73 9.10 10.42
n=3 9.42
Using this infonnation tabulate the 11 lowest eigenvaIues in the cube and the
sphere in units of h2 /8ma:. Notice that the shape ofthe partic\e does indeed have an
inlluence on the distribution of eigenvalues. This question arose reccntly during sorne
materials rescarch.
29. Friedel model in the second moment approximation. Derive eqns (9.11)-(9.14)
from eqn (9.1 O).
30. Finnis-Sinclair potential. A Finnis-Sinclair potential has the following fonn
Eco. = '[L J'"L V(r¡j) - eL pt l2 ]

J i
(1)
wbere • is a parameter with the dimensions of energy, e is a dimensionless parameter,

i and j are atomic sites, V(r) is a pair potential, r'j is the distance between sites i and
j, and p, is a sum of pair potentials, q,(r), from site i
p, = L t/l(r¡j). (2)
j
Let the pair potentials V(r) and q,(r) be inverse powers
V(r) = (;)" (3)

Problems 249
wbere a is a parameter witb tbe dimensions of lengtb and m and n are positive
integers, sucb tbat n > m ~ 4.
Let us apply tbis potential to an f.c.c. metal sucb as Cu or Au. We are free to
cboose tbe parameter a to be wbatever lengtb we like: all it does is fix tbe units of
lengtb in wbicb r is measured in !he crystal. Cboose a to be tbe f.c.c. lattice parameter.
Tbe condition for equilibrium of tbe f.c.c. crystal is tbat tbe derivative of tbe energy
witb respect to tbe lattice parameter is zero (otberwise tbe lattice parameter would
cbange spontaneously until tbe derivative were zero). Sbow tbat tbis condition fixes
e to be tbe following
c=
nS'le
(4)
m(S~)'/2
wbere S! is tbe following lattice sum
S' =
11
L (~)"
J#<f rl} .
(5)
Sbow tbat at equilibrium tbe cobesive energy per atom is given by

eS'
E:. h = - " [2n - m]. (6)
2m
Sbow tbat at equilibrium tbe bulk modulus is given by
B = (2n - m)mS!
(7)
360
wbere íl is tbe atomic volume (a 3/4).
Tbe square root in tbe functional form of tbe Finnis-Sinclair potential is wbat
distinguisbes it from an ordinary pair potential. But tbere are instances wben it is
possible to derive an effective pair potential for tbe Finnis-Sinclair formo For example
if we are dealing witb a vacancy in tbe crystal we see tbat tbe neigbbours of tbe
vacancy bave lost only one of tbeir 12 first neigbbours. Tbis is a relatively small
cbange, so we may expand tbe square root term in a Taylor series about tbe value
of P in tbe perfect crystal. Let tbe value of P in !he perfect crystal be Po. Tben
expanding p'/2 about p = Po we get
1
P'/2 "" (p o)'12 + (p _ P ). (8)
2(Po)'/2 o
Sbow bow tbis expansion leads to tbe following effective pair potential for describing
energy differences
(9)
Tbis bas tbe form of a Lennard-Jones potential. Comment on whether you tbink
tbis expansion is valid at (i) a free surface of the metal, (ii) an internal defect sucb
as a grain boundary, stacking fault, or dislocation coreo
31. Metal/ic al/oy bonding. Derive eqns (9.41)-(9.43).

250 Problems
32. Energy gap in random substitutional al/oy. Consider an AB random crystaUine
alloy in which the coordination number is z. In an s-band model of the alloy assume
that the nearest neighbour hopping integrals are always fJ (i.e. regardless of the types
of atom that the electron is hopping between) and that the on-site Hamiltonian
matrix elements at the A and B atomic sites are tA and tH, where tH> t A. Prove that
an energy gap exists if
(1)
This is how you can do it. The eigenstates of the system are linear combinations of
the atomic basis states
1'1') = L alli) (2)
I
where Ii) is the atomic state at site i and al is the expansion coefficient. Put this
eigenstate into the Schriidinger equation, HI'I') = EI'I'), and show that
~ la.!
lE - t¡1 ~ t... -lfJl (3)
.... ¡ la¡1
where j is an arbitrary atomic site. Choose j to be the site where the amplitude is a
maximum, i.e. la)1 2: la.! for all i. Hence show that
lE - t¡1 ~ zlfJl· (4)
By letting site j be an A site and then a B site show that there are two bands
tA - zlfJl ~ E :$; t A+ zlfJl (5)
and
tH - zlfJl :$; E :$; tH+ zlfJl. (6)
Hence show that eqn (1) is the condition for a band gap. Notice that the proof of
the existence ofa band gap has nothing to do with long-range order, but that
short-range order is assumed through the fixed coordination number z and the fixed
hopping integral {J.
Sample examination questions
QUESTIONS labelled with an asterisk are intended to be more advanced and should
take about 1 hour to answer. The other questions should take about 30 minutes.
1. Explain the significance oC the existence of a universal binding energy curve
Es = IE~lg(a)
where
g(a) = -(1 + a + 0.05a') e-o
a = (rws - r~s)/l
1= ( IE~I )1/2
12"r~B'
and r~s is the equilibrium Wigner-Seitz radius, IE:I is the equilibrium cohesive
energy, and B' is !he equilibrium bulk modulus. To which type of materials do you
think this 'universal' curve applies?
Consider two materials, which obey !he universal binding energy curve and which
have the same volume per atom. Show that the ratio of the cohesive energy to !he
bulk modulus, at equilibrium, is the same in the two materials.
2.· Consider the universal binding energy curve
where
g(a) = -(1 + a) e-O
a = (rws - r~s)/l
1= ( IE'I )'/2
12"r~sB'
and r~ is the equilibrium Wigner-Seitz radius, IE.I is the equilibrium cohesive

energy, and B' is the equilibrium bulk modulus. Discuss the applicability of this
relation to metallic and nonmetallic materials.
Show that a volume change of + or - 1 per cent of and from the equilibrium
volume, gives rise to a pressure oC - or + 1 per cent oC the equilibrium bulk modulus,
Cor all materials satisfying the universal binding energy curve.
3. Discuss the concepts of covalency and ionicity in binary compounds. What do

you understand by the term 'electronegativity'?
252 Sample examination questions
Consider a heteronudear diatomie AB moleeule with a basis set eomprising the
atomie state lA) on atom A and the atomie state lB) on atom B. Let the Hamiltonian
matrix elements be defined as follows
(AIHIA) = O (AIHIB) = (BIHIA) = P (BIHIB) = 2ti
where p is negative and ti is positive. The states lA) and lB) are normalized and for
simplicity assume that the overlap (AIB) = (BIA) = O. Show that the bonding and
anti-bonding molecular states have the following energies
bonding state: ti _ (ti 2 + P2)I/2

antibonding sta te: ti + (ti 2 + P2)I/2
Sketch your results on an energy level diagram, showing the levels of both atomic
states and both molecular sta tes.
4* Consider a heteronudear diatomie AB moleeule with a basis set eomprising

the atomie state lA) on atom A and the atomie state lB) on atom B. Let tbe
Hamiltonian matrix elements be defined as follows:
(AIHIA) = O (AIHIB) = (BIHIA) = P (BIHIB) = 2ti
where Pis negative and ti is positive. The states lA) and lB) are normalized and tbe
overlap matrix elements (AIB) = (BIA) = S:> O. Show that the energies of the
bonding and antibonding molecular sta tes are as follows
ti - pS - (ti 2 + P(P - 2S!i))I/2
bonding sta te:
1 - S2
ti - pS + (ti 2 + P(P - 2S!i))I/2

antibonding sta te:
1 - S2
If instead of working with the above atomic states lA) and lB) we worked with new
states IX) and lB') sueh that
2ti - pS
(XIHIX) = -~
1 - S2
(B'IHIB') = ----'o;-
1 - S2
(XIHIB') = (B'IHIX) = (P({l - 2Sti))I/2

1 - S2
and
(XIX) = (B'IB') = 1 (XIB') = (B'IX) = O
show that the same bonding and antibonding energy levels are obtained. How do
you ¡nterpret this result?
5. What do you understand by the eharge density and the density matrix?
Explain the meaning of the bond order. What is the relationship between the bond
order and the eoneepts of single, double, and triple bonds in the chemistry of
hydrocarbons?
6.· Explain earefully the dilferenee between band energy and bond energy.
Consider an infinite, perfeet linear chain of atoms, each of which is associated with
one atomic state. Let Ij> denote the atomic state associated with atom j. Assume
that !he atomie states form a complete orthonormal se!. How would you define the
local density of states, d¡(E), associated with site j, where E is energy? Prove that
where e¡ is the Hamiltonian matrix element H¡¡ and H,¡ = (iIHU>, EF is the Fermi
energy and Pi¡ is the density matrix element between states i and j. What is the
significanee of this result?
Let all Hamiltonian matrix elements for the linear ehain be zero, exeept (i) those
between neighbouring pairs of atoms and let those be P (P < O) and (ii) on-site
elements and let those be <x. Ir the atomic spacing is a, show that the energy band is
E(k) = <X + 2p eos ka,
where - "la ~ k ~ "la. Calculate the local density of states. Caleulate the band
energy and the bond energy for 2p ~ EF ~ - 2p.
7. State Bloch's theorem and discuss why it is useful. When is it not useful? What
is a Brillouin zone and why is it important in the electronie structure of crystals?
Consider a two-dimensional hexagonal erystallattiee, and let the distanee between
neighbouring lattice points be a. Construet the reciprocallattiee and the first Brillouin
zone.
Is the Brillouin zone for the diamond eubie erystal strueture the same as that for
the faee centred eubic crystal structure if the erystallattiee parameters are the same,
and why?
8.· Explain the concept of the effective mass in band theory. Whal does it mean
to say that an electron has a negative effective mass? What is the group velocity of
an electron and explain why it tends to zero as the wave vector approaehes a Brillouin
zone boundary. Explain the relationship between this phenomenon and Bragg
reftection. Discuss why scattering of electrons is both essential to electronie condue-
tion in metals and also the limiting factor for the eleetronic conductivity.
9. Consider a face-eentred eubie erystal with one atomie s state per atom. Let these
atomie states form a complete orthonormal se!. Let all matrix elements of the
Hamiltonian in this basis be zero except those between nearest neighbour pairs, and
let those by p. The nth moment of the local density of states, d,(E) associated with
atomie state i is defined by
I'i"· = f
.... bolc band
E"d,(E) dE .
Using the moments theorem calculate the zeroth, first, and seeond moments of the
local density of states for (a) an atom in the bulk of a perfeet crystal, (b) an atom
on a ftat (1 OO) surface, (e) an atom on a ftat (1 I 1) surface, and (d) an atom on a
ftat (1 1 O) surface. Ir the local density of states in eaeh case is approximated by a
Gaussian form, sketch how it looks in the four cases. If all bond lengths remained
ideal which surface energy would be highest and which lowest if the band is half full?
10.· Consider a body-centred cubic crystal with one atomic s state per atom. Let
these atomic states form a complete orthonormal sel. Let all intersite matrix elements
of the Hamiltonian in this basis be zero except (i) those between the eight
nearest-neighbour pairs, and let those be {J, and (ii) those between the six second
nearest-neighbour pairs, and let those be (J2' Let the on-site matrix element of the
Hamiltonian <iIHli) be 8,. The nth moment of the local density of states, d,(E)
associated with atomic state i is defined by
1'1") = fwholc band

(E - 8,)"d,(E) dE.
Using the moments theorem calculate the zeroth, first, and second moments of the
density of states for (a) an atom in the bulk of the perfect crystal and (b) an atom
in a defect which has z, first neighbours and Z2 second neighbours.
Assume the local density of states d,(E) may be represented by a Gaussian function
as follows
d,(E) = ~2)
(2,,1', )
'/2 exp _ (E _(::)2)
21',
where 1'1') is tbe second momenl.lfthe Fermi energy is EF show that the bond energy
associated with site i is given by
_(~~J/2 exp - «EF - 8,)2/21'1 2).
Assume that the on-site energies adjust in order to maintain local charge neutrality.
Show that this requires that (E F - 8,)2/21'1') is a constan!. Hence prove that the bond
energy associated with site i is proportional to (1'1')'/2.
11. Discuss the origins of electronic band gaps in sp-bonding systems from the
point of view of (i) nearly free electron theory and (ii) linear combination of atomic
orbitals theory. Show how these two viewpoints are equivalent by analysing the
eigenstates at the Brillouin zone boundary in a one-dimensional model.
12.· Discuss the origins of electronic band gaps in crystalline and noncrystalline
materials.
Consider an infinite linear chain of atoms along x, bonded together by sp hybrids.
Let the atomic s state at sÍle j be denoted by U, s) and the P. state be denoted by
U, x). Assume all Hamiltonian matrix elements are zero except
(a) all on-site matrix elements <i, slHli, s) are 8, and all on-site matrix elements
<i, xlHli, x) are 8,.
(b) the intersite matrix elements <i, slHli ± 1, s) = (ssu), <i, xlHli ± 1, x) = (ppu),
and <i, slHli ± 1, x) = ±(spu).
Calculate and sketch the energy bands in the first Brillouin zone, bearing in mind
that (ssu) < O, (Ppu) > O, and (spu) > O. What are the band gaps at the centre and
edge of the Brillouin zone?
13. Why must quantum theory be used to describe the motion of electrons in a
metal at room temperature? Is quantum theory necessary to describe the motion of
iron atoms in b.c.c. iron at room temperature? (Density of iron = 7.87 x 103 kg m - '.
lattice parameter = 0.287 nm.)
A new composite material is made in which cubic a1uminium particles, of side
20 nm, are embedded in an insulating solid matrix. The particles crystallize in the
f.c.c. crystal structure, with lattice constant 0.4 nm. Assuming there are three valence
electrons per atom estimate the energy of the highest occupied electronic state at
OK. Estimate the smallest frequency of electromagnetic radiation at which the
particles will absorb energy strongly.
14.· Describe qualitatively the process of screening by the electrons of a metal,

and give examples of properties or phenomena where screening plays an important
role.
Consider an inteñace between a sodium chloride crystal and a gold crystal. Show
that a lattice of image charges is induced in the gold by the sodium chloride lattice.
What are the consequences of this phenomenon for adhesion at the inteñace?
15. Why does the cohesion of the Iransition metals have to be treated differently
from that of the nearly free electron metals? Sketch the density of states for a typical
transition metal and compare it with that of a nearly free electron metal. Why does
the cohesive energy of the 4d and 5d series vary roughly parabolically across the
series, with the maximum cohesion al the middle of the series?
16.· Consider the following model of the binding energy in a face-centred cubic
transition metal
E = .[ ~ j~, V(ril) - e ~ (p,)'/2 ] (1)
where B is a parameter with the dimensions of energy, e is a dimensionless parameter,

V(r) is a pair potential, r'l is the distance between sites i and j, and Pi is a sum of
pair potentials, <f¡(r), from site i
Pi = L <f¡(ril )· (2)
j
Give a physical interpretation of the terms in eqns (1) and (2).

Let the pair potentials V(r) and <J¡(r) be the exponential exp( - 2kr) where k is a
constant. If interactions beyond first neighbours are ignored show that the condition
for the peñect f.c.c. crystal to be in equilibrium, with the first-neighbour bond length
equal to R, requires that e = 2(12)1/2 e-u and that the cohesive energy per atom of
the peñect crystal is then equal to 118 e- 2kR •
Calculate the bulk modulus, B = - V(d 2 E/dV 2 ), at equilibrium.
Show that if the local coordination number at a defect in the crystal is z ( < 12),
and the first neighbour separations remain R, then the energy of each of the z bonds
is greater by a factor of (2(12/z)I/2 - 1) than the energy of the bonds in the peñect
crystal. Comment on the consequences of this result for the expected atomic
relaxation al a free suñace of the metal.
17. What do you understand by 'the nearly free electron approximation' and
explain why it is that metals such as sodium and aluminium may be modeUed
successfully using it. Why can't a transition metal like niobium or molybdenum be
described as a nearly free electron metal? What differences exist in the valence charge
density distribution in nearly free electron metals compared with transition metals,
and what consequences follow from this for their elastic properties?
18." What is a metal?
19. What do you understand by the term 'metaJ-insulator transition'? Give an

argument for why the critical electron density, n, must satisfy a"nl!3 ~ e for metallic
behaviour, where a" is a characteristic Bohr radius of a hydrogenic electron orbital
and e is a constant. Given lhat e = 0.26 calculate lhe critical density of phosphorus
donor atoms in silícon, for which a" = 1.7 nm, at which the material becomes a metal
at OK.
20." Compare the faetors lhat determine lhe struetures of binary ionic, covalent,
and metallic compounds.
Index
amorphous materials 215-21 Bloch function 47-8

Anderson locaIization 224-5 Bloch oscillations 82, 243
Anderson transition 225 Bloch's thcorem 48-9, 75, 132, 163
angular momentum 11 bond charge densily 29, 54, 212
about bond axis IS, 114, 183 bond energy 35-7
anharmonicity 6,9, 236 in a binary alloy 242-3
antibonding state and bond order 65-6
in chain of atoms 46 and density of states 64-6
in hydrogen moleeule 27-9 in a diatomic moleeule 35-7
atomic eigenotates 15-20 and Finnis-Sinclair potentials 180-2
se. also hydrogen atom in an infinite linear chain 243
'aulbau' principIe 15-16 in a simple cubic cryslal 95-6
in Iransition melals 175-6, 184-9
bond length 178-9
band crossing tranoition 226 bond orbital model 122-3, 194
band energy bond order 35-7
definition 63 and atomic environment 179
real space vs k-space approaches 66 and bond strength 96-7
band gap in diatomic molecule 35-7,241
in amorphous mal.risls 21S-21 in infinile linear chain S2-7, 242-3
in amorphous silicon 219-21 as real space concept SI-2
atomic/chemicai origins 101-11 in second moment approximation 178
and Bragg reftection 147-9 in simple cubic cryslal 92-7
direcl vs indirect 105-6, 127 al surfaces 179
in NaCl structure 244 in trimer 241
and Peierls dislortion 107-9 and total charge density 96-7
in random substitutional alloy 250 bonding state
and screening 111 in chain oC atoms 46
in Weaire-Thorpe model 122-4,219-21 in hydrogen molecule 27-9
band index 103 Boro model 190-1
band slructure Boro-Oppenheimer approximation 205
for f.c.c. crystal 243-4 bra 21
as Ihe Fourier transCorm oC Ihe real-space Bragg reftection 82-3
Hamiltonian 49 and band gaps 147-9
Cree eleetron compared wilh LCAO Brillouin zone 51
139-43 for b.c.c. lattice 79-80
Cor linear chain 38-47 for C.c.c. lattice 78-9, 126-7
in NaCl slruclure 244 for linear lattice 51
oC silicon 124-7 Cor simple cubic laltice 78
Cor square lattice 76 Cor square latticc 77-8
bandwidth 67,175-7,191-2 bulk moduluo 5, 236
basis sel 21 oC Iransition metalo 177
basis state 21
basis veelor 21
binary alloy
chain 72-3, 106,242-3 charge densily 29, 32, 52-4, 92-4, 205
NaCl structure 244 charge neutraJily in metalo 182, 192-3
binding energy 2 charge transfer 32-5
258 Index
cohesive energy embedded atom potential 179-80
examples in density funetional energy band 42
tbeory 209-12 equation of motion
second moment approximation 175-7 of electron 80-4
transition metals 174-9 of hole 86
volume dependence for simple equation oC state
metals 166-9 for ftuids 2
see abo binding energy for metals 1-9, 236-7
collision time 133, 163 exchange-correlation energy 208-9
eonduction band 123 exchange-correlation hole 155-7,208-9
contact potential 159-60 exchange-correlation potential 208-9
continuous random network 215-19 exciton 226
core electrons 16 exclusion principIe
correlation bole 155-7 and 'autbau' principie 15-16
covaleney of a bond 33 and exchange interaction 155-7
covalent bond 194 and Fermi momentum 137
erystal momentum 81 and Fermi-Dirac statistics 158-9
and hydrogen atom 12
andlocalization 222-4
dangling bond 217-19 and occupation of molecular states 52
d-d bonding 183-7 and paeudopotential (orthogonality
delta function 61 bole) 151-3
density funetional tbeory 205-14 and sbort range repulsion 64
density matrix 54 extended zone scheme 77
density of states
bimodal vs unimodal form 71
in 1-dimensional crystal 58-9,242-3 Fermi energy 53, 137
in 2-dimensional crystal 89-91 Fermi momentum 137
in 3-dimensional crystal 91-2 Fermi sphere 137
and Finnis-Sinclair potentials 180--2 Fermi surface 87-8, 137
in free electron tbeory 138-9 Fermi wave vector 53, 137
Gaussian form 181 Fermi-Dirac statistics 158
in k-space SO, 89, 91, 137 Finnis-Sinclair potentia! 179-82, 248-9
local 59-63 free electron tbeory 132-71,245,246,247-8
moments 66-71 Friedel model for transition metals 174-9
relation between sbape and moments 67
in simple metals 143-4
total 58, 61-2 glass 215
in transition metals 172 group velocity 80,86,241-2
in Weaire-Tborpe model 123
diatomie molecuJe
heteronuclear 32-3,35-7,241 Hall eft'ect 165-6
homonuclear 25-31,240 Hamiltonian matrix 26
Dirae notation 21-5 Hartree approximation 206-7
dislocation in siticon 213 Hartree potential 206
Drude's free electron model 132-4 Hel1mann-Feynman theorem 243
Hilbert space 22
Hohenberg-Kohn tbeorem 205
eft'ective mass 81-3,86,222-4 holes 84-6,222-4,226,231-2
electrieal conductivity hopping integral 31, 113-16, 183-5
and Bragg reftection 83-4 Hubbard transition 231-2
in free electron approximation 162-3 Hubbard U 227
and scallering 83-4, 162-3 Hume-Rothery rules 197-8
table for the elements 19 bybrid energy 121
electron eurrent 242 bybridization 102-6,117-21, 194-6
electron spin 11 bydrogen atom 9-15
electron spin resonance 218 angular components oC the
electronegativity 34-5, 197-8 eigenfunctions 12-15,239-40
Index 259
d orbitals 13-15 in f.e.e. erystal 243-4
orthogonality of eigenfunetions 11-12 and structural stability of transition
p orbitals 12-13 metals 187-9
quantum numbers 11 substitutional impurity 98-100
radial components of the theorem 66-71
eigenfunetions 11-12, 237-9 in two- and three-dimensional crystals
s orbitals 12 97-100
hydrogen molecule 25-31 Mott transition 229-31
identity operator 22, 23 nearly free electron model 144-51

image eharge interacrion 155-7 nodes 12, 15,42-4
independent electron approximation 133,
205-7
ionie bond 32-3, 189-93, 197 Ohm'.law 162-3
on-site energy 130
on-site Hamiltonian matrix elements 16, 18,
jellium 135 20,26
operator 22-4
orthogonality hole 151-3
ket 21 orthogonality principie for hybrids 118
Kohn-Sham equations 206, 208-9
k-space 50
pair potentials for simple metals 169-71
Peierls distortion 38, 107-9
Lindemann criterion for melting point 237 perfect pairing approximation 194
linear ebain periodie boundary conditions 44, 135-6
finite 38-44 polarity of a bond 33
infinite 42-6 polarization catastrophe 232-5
local density approximation 209 polaron 222-4
localization 221-5 principal quantum number 11
Lorentz number 164-5 promotian energy 121
pseudopotential 151-3
empty eore 152
Madelung energy 190-1,197 Phillips-Kleinman 245-6
magnetie quantum number 11
magnetism 172-3
many body problem 205 quantum state 21-2
matrix eIernent 24 quasipartieles 155
maximum overiap principie for
hybrids 117-18
melting point of metals 236-7 rectangular d band model 174-9
Mendeleev number 199-200 reduced zone scheme 77
metal-insulator transition 225-35 relaxation time 133, 163
Anderson transition 225 resonance integra~ see hopping integral
band crossing transition 226 rigid band model 189-91
and breakdown of band theory 227 ring of atoms
Hubbard transition 231-2 finite 44-5
Mott transition 229-31 infinite 45-6
and Peierls distortion 107-9
metallie bond 110-11,179,189-93,194
metallie state screening
and band theory 109-10 in atoms 15
definitions 20, 87 and band crossing transition 226
mobility edge 224-5 and exehange and correlation 155-7
molecular model 122-3 in metallie alloys 192-3, 197-8
moments 66-71 in metals vs insulators 111
and binary alloy ehain 72-3 and Mott transition 229-30
260 Index
screcniog (continued) specificheat 134, 160-2
in second moment approximation 182 state vector 22
in Thomas-Fermi approximation 153-4 struetural energy differenceo
and valeney 16 for smcon and germanium 210, 212
second moment approximation for simple metalo 169-71
and band width 175 for transition metalo 178, 187-9, 210-11
and cohesive energy 175-7 otruetural stability of compounds 194-203
and Friede1 mnde1 175-9 otrueture maps 9, lO, 198-203
see also Finnis-Sinclair potentials oubstitutional impurity 98-100
secular determinant 27 surfaces of silicon 213
secular equations 26-7, 39 thermal conduetivity of metalo 164
se1f-ronsistent calculation 206-8 thermalJy activated hopping 222, 225
shell 11 Thomas-Fermi approximation 153-4
short range order 215-19 time reversal symmetry 46
siücon 122-31 time-dependent SchrOdinger equation 29
arnorphous 215-21 transfer integral, se. hopping integral
band energy 130 transition metalo 172-93
band structure with mjnjmal basis trimer 240-1
set 124-7 tunnelling in bonding 31
bond energy 129-31
bond order 127-9
Chadi Hamiltonian 124
defects 213 uncertainty relation 134, 245
degree of sp' hybridization 127-9
doping 229-30
s-p mixing 129
structuraI stabiüty 130-1, 210, 212 valence band 123
Slater-Koster table 185 valence electrons 16
solar cells 219 valency 16
s-p bonding 112-31 variational principie oC density functional
angular dependence of p-p theory 206, 208
interactions 116-17,185
angular dependence of s-p
interactions 116, 185
fundamental hopping integraIs 113-16 wave vector 46, SO, 75, 86, 136
occurrence in elements 112 Weaire-Thorpe model 123
silicon 122-31 amorphous oilicon 219-21
sp hybrids 119 crystaI1ine silicon 123
sp' hybrids 119-20 Wiedemann-Franz law 164-5
sp' hybrids .120 Wigner-Seitz radius 2

Sutton - Electronic Structure of Materials PDF

Uploaded by

Copyright:

Available Formats

Sutton - Electronic Structure of Materials 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

Sutton - Electronic Structure of Materials PDF

Uploaded by

Copyright:

Available Formats

Electronic Structure of

CLARENDON PRESS . OXFORD

Oxford is a trade mark of Oxford Universily Press

Published in Ihe U niled Slales

© Adrian P. Surron, 1993

Firsl published 1993

AII righrs reserved. No parl of Ihis publication may be

This book is sold subjecl lo the condition that it shall nat,

Library of Congress Cataloging in Publicalion Data

Printed in Greal Brilain by Bookcraft (Bath) LId

Modem materials science is just as much concemed with deCects in crystals

1 am grateful to the following authors for permission to reproduce figures

2. The diatomic molecule 21

3. From the finite to the infinite 38

4. Into two and three dimensions 74

6. s-p bonding-a case study in silicon 112

7. Free electron theory 132

8. Properties of free electron metals 158

10. Structural stability of oompounds 194

11. Introduction to modern quantitative theory 204

12. Where band theory breaks down 215

Sample examination questions 250

Aims of the book

The 'universal' equation of state for metals

different equilibrium values of the Wigner-Seitz radius and the binding

Fig. 1.3 The definition of L in eqn (1.5).

which crosses the length axis when

Thus, we define a scaled length a as foIlows

1= ( IE~I )1/2 = ( IE~I )1/2 (1.9)

- Tbeorv Stainless 1.0 Rb

1.0 0.8 0.6 0.4 0.8 0.6 0.4

The hydrogen atom

f 'I'.'m(r, 9, q¡)'I'.·"m,(r, 9, q¡) dr =O (1.19)

Fig. 1.9 The 1 s atomic orbital for the hydrogen atom.

of them if we prefer. It turns out that it is usually more convenient to work

Fig. 1.10 Px' Py• and Pz atomic orbitals.

Fig. 1.11 dxyo d yz • d zx • d .. _,... and d.. atomic orbitals.

This is significant because a small vaIue of Ep - E, will enable s and p

Metals, semiconductors, and insulators

Review of bras, kets, and all that

It is important to realize the distinction we have made between a state

('1'10 = ('1'1110 = <'1'1 I It/J)<t/JIIO = I ('I'It/J) (t/JIO· (2.5)

A possibly more familiar representation of the overlap <'1' I~) is in terms of

<'1'10 = fan,pace 'I'*(r)~(r) dr = I.n,pace <'I'lr)<rl~) dr. (2.7)

<4>1'1') = <4>111'1') = L <4>li)<il'l') (2.9)

which relates the expansion coefficients <4>1'1') to the set of <il'l').

<jlO = <jILI'I') = GILI L li)<ill'l') = L GILli)<il'l'). (2.12)

where we have used different dummy variables i and J in our double

This equation allows us to transform the matrix elements of an operator

The chain of N billiard balls thus spans an N-dimensional Hilbert space. As

A homonuclear diatomic molecule: the hydrogen

t Although the potentials oC both atoms are contained in H.

Eo - E H12 l' = O. (2.27)

E2 - 2EoE + E~ - H l2 H21 = O. (2.28)

The normalized state vector corresponding to the eigenvalue Eb is

while the normalized state vector corresponding to the eigenvalue E. is

Now <11 - WV 2 j8n2 m) + V¡(r)12) is zero because

and therefore 01- (h 2 V 2 j8n 2 m) + V¡(r)12) = Er(112) = O. Thus H 12 =

If the charge density in the molecule were simply a superposition of the