Electronic and Optical Properties of D Band Perovskites

This page intentionally left blank
T . W O L F R A M A N D S . E L L İ A L T I O Ğ L U
Electronic and Optical Properties of d-band Perovskites
The perovskite family of oxides includes a vast array of insulators, metals, and
semiconductors. Current intense scientific interest stems from the large number of
diverse phenomena exhibited by these materials including pseudo-two-dimensional
electronic energy bands, high-temperature superconductivity, metal–insulator tran-
sitions, piezoelectricity, magnetism, photochromic and catalytic activity.
This book is the first text devoted to a comprehensive theory of the solid-state
properties of these fascinating materials. The text includes complete descriptions
of the important energy bands, photoemission, and surface states. The chapter on
high-temperature superconductors explores the electronic states in typical copper
oxide materials. Theoretical results are compared with experimental results and
discussed throughout the book.
With problem sets included, this is a unified, logical treatment of fundamen-
tal perovskite solid-state properties, which will appeal to graduate students and
researchers alike.
Thomas Wolfram is a retired professor and Chairman of the Department of

Physics and Astronomy at the University of Missouri-Columbia and retired Vice
President and General Manager of Amoco Laser Company, ATX.
Ş i̇ nas i̇ Ell i̇ altio ğ lu is a professor and former Chairman of the Department
of Physics at the Middle East Technical University, Ankara, Turkey.
P1: FXS/GQE P2: FXS
CUUK735-Wolfram August 9, 2006 2:57
ELECTRONIC AND OPTICAL

P R O P E R T I E S O F D- B A N D
PEROVSKITES
THOMAS WOLFRAM
Formerly of University of Missouri-Columbia
Ş İ N A S İ E L L İ A L T I O Ğ L U
Middle East Technical University, Ankara
iii
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521850537
© T. Wolfram and S. Ellialtioglu 2006
This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2006
ISBN-13 978-0-511-34856-3 eBook (EBL)

ISBN-10 0-511-34856-8 eBook (EBL)
ISBN-13 978-0-521-85053-7 hardback

ISBN-10 0-521-85053-3 hardback
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
Preface page ix
1 Introductory discussion of the perovskites 1
1.1 Introduction 1
1.2 The perovskite structure 3
1.3 Ionic model 4
1.4 Madelung and electrostatic potentials 6
1.5 Covalent mixing 9
1.6 Energy bands 12
1.7 Localized d electrons 16
1.8 Magnetism in the perovskites 17
1.9 Superconductivity 18
1.10 Some applications of perovskite materials 20
2 Review of the quantum mechanics
of N - electron systems 27
2.1 The Hamiltonian 27
2.2 The Slater determinant state 28
2.3 Koopman’s theorem 29
2.4 Hartree–Fock equations 30
2.5 Hartree–Fock potential 32
2.6 Approximate exchange potential 34
2.7 The LCAO method 35
2.8 Orthogonalized atomic orbitals 37
v
vi Contents
3 Empirical LCAO model 41

3.1 LCAO matrix elements 41
3.2 Slater–Koster model 42
3.3 Symmetry properties of the Löwdin orbitals 48
4 LCAO energy band model for cubic perovskites 53
4.1 The unit cell and Brillouin zone 53
4.2 LCAO matrix equation for an infinite lattice 56
4.3 LCAO matrix elements for the perovskite 57
4.4 LCAO eigenvalue equation for the cubic perovskites 61
4.5 Qualitative features of the energy bands 65
4.6 Summary of the chapter results 73
5 Analysis of bands at symmetry points 77
5.1 Energy bands at Γ 77
5.2 Energy bands at X 81
5.3 Energy bands at M 84
5.4 Energy bands at R 86
5.5 Cluster electronic states 90
6 Density of states 107
6.1 Definitions 107
6.2 DOS for the pi bands 109
6.3 DOS for the sigma bands 114
6.4 The Fermi surface and effective mass 123
7 Optical properties of the d-band perovskites 138
7.1 Review of semiclassical theory 139
7.2 Qualitative theory of ε2 (ω) 142
7.3 Interband transitions from non-bonding bands 153
7.4 Frequency dependence of ε2 (ω) for
insulating and semiconducting perovskites 157
7.5 Frequency dependence of ε2 (ω)
from σ 0 → π ∗ transitions 167
Contents vii

from π → π ∗ transitions 170
7.7 σ → π ∗ interband transitions 173
7.8 Summary 180
8 Photoemission from perovskites 182

8.1 Qualitative theory of photoemission 183
8.2 Partial density of states functions 187
8.3 The XPS spectrum of SrTiO3 189
8.4 Nax WO3 192
8.5 Many-body effects in XPS spectra 193
9 Surface states on d-band perovskites 199

9.1 Perturbations at a surface 199
9.2 Surface energy band concepts 202
9.3 Self-consistent solutions for the band-gap
surface states: SrTiO3 213
9.4 Surface–oxygen defect states 221
10 Distorted perovskites 231

10.1 Displacive distortions: cubic-to-tetragonal
phase transition 231
10.2 Octahedral tilting 240
11 High-temperature superconductors 249

11.1 Background 249
11.2 Band theory and quasiparticles 253
11.3 Effective Hamiltonians for low-energy excitations 256
11.4 Angle-resolved photoemission 257
11.5 Energy bands of the Cu–O2 layers 260
11.6 Chains in YBa2 Cu3 O6.95 277
11.7 Summary 279
viii Contents
Appendices
A Physical constants and the complete elliptic integral
of the first kind 285
A.1 Selected physical constants 285
A.2 The complete elliptic integral of the first kind 286
B The delta function 288
C Lattice Green’s function 291
C.1 Function Gε (0) 293
C.3 Lattice Green’s function for the pi bands 296
D Surface and bulk Madelung potentials
for the ABO3 structure 302
Index 305
Preface
Metal oxides having the cubic (or nearly cubic), ABO3 perovskite structure consti-
tute a wide class of compounds that display an amazing variety of interesting prop-
erties. The perovskite family encompasses insulators, piezoelectrics, ferroelectrics,
metals, semiconductors, magnetic, and superconducting materials. So broad and
varied is this class of materials that a comprehensive treatise is virtually impossible
and certainly beyond the scope of this introductory text. In this book we treat
only those materials that possess electronic states described by energy band the-
ory. However, a chapter is devoted to the quasiparticle-like excitations observed in
high-temperature superconducting metal oxides. Although principally dealing with
the cubic perovskites, tetragonal distortions and octahedral tilting are discussed in
the text. Strong electron correlation theories appropriate for the magnetic proper-
ties of the perovskites are not discussed. Discussions of the role of strong electron
correlation are frequent in the text, but the development of the many-electron the-
ory crucial for magnetic insulators and high-temperature superconductors is not
included.
This book is primarily intended as an introductory textbook. The purpose is
to provide the reader with a qualitative understanding of the physics and chem-
istry that underlies the properties of “d-band” perovskites. It employs simple linear
combinations of atomic orbitals (LCAO) models to describe perovskite materials
that possess energy bands derived primarily from the d orbitals of the metal ions
and the p orbitals of the oxygen ions. The results are usually obtained analytically
with relatively simple mathematical tools and are compared with experimental data
whenever possible.
The book is considered appropriate for science and electrical engineering grad-
uate students and advanced undergraduate seniors. It may be used as a primary
text for short courses or specialized topic seminars or it can serve as an auxiliary
text for courses in quantum mechanics, solid-state physics, solid-state chemistry,
materials science, or group theory. The reader will need a basic understanding of
quantum mechanics, and should have had an introductory course in solid-state
physics or solid-state chemistry. Knowledge of group theory is not required, but
some understanding of the role of symmetry in quantum mechanics would be help-
ful. The material covered is considered a prerequisite for understanding the results
of more complex models and numerical energy band calculations. Research scien-
tists seeking a qualitative understanding of the electronic and optical properties of
the perovskites will also find this book useful.
The theoretical results are derived in sufficient detail to allow a typical reader
with a calculus background to reproduce the formulae and derive independent re-
sults. Because most of the results are presented in analytic form, the relationships
ix
x Preface
among the physical variables are transparent and can easily be understood and ex-
plored. Using these analytical results the reader can obtain numerical results for the
electronic, optical, and surface properties of specific materials using nothing more
sophisticated than a programmable hand calculator or a desk computer equipped
with MS QuickBasic°c software.
Many of the topics discussed in the book were originally published by the
authors in research papers and were formulated in terms of Green’s functions. In
order to keep the material in this book as simple as possible the same results are
obtained here by more rudimentary mathematical methods.
For the most part our understanding of the properties of metals is derived from
various versions of the free-electron model (often with imposed periodic boundary
conditions). The simplicity of this model does not diminish its applicability, and in
many instances, particularly in the case of BCS (Bardeen–Cooper–Schrieffer) su-
perconductors, the results obtained are quantitatively correct. Of equal importance
is the pedagogical utility of the free-electron model, which permits scientists and
students alike to make simple calculations and to develop scientific concepts and a
useful intuition about the electronic and optical phenomena of metals.
In the case of compounds whose properties are dominated by the atomic or-
bitals of the constituent ions, the free-electron model is not particularly useful.
For compounds such as the perovskites the physical and chemical properties are
largely dependent upon the crystalline structure and the symmetry of the atomic
orbitals involved in the valence bands and the bands near the Fermi level. The
purpose of this book is to provide a relatively simple but complete description of
the d-band perovskites based on atomic-like orbitals. Models of this type were de-
veloped many years ago by chemists and physicists alike using LCAO and other
similar localized-orbital approaches. Later, such models were “put on the shelf”
as theoretical solid-state physicists moved almost exclusively into the realm of
momentum-space theories. Indeed, for some time it could be said with justifica-
tion that solid-state physicists were the Fourier transform of solid-state chemists.
With the recent discovery of high-temperature superconductivity (HTSC) in
the cuprate compounds interest in the science of the transition metal oxides has
grown enormously. Interestingly, solid-state theorists have returned to real-space
theories to look for an understanding of these materials. It is somewhat ironic that
the original migration to ~k-space was driven, to a large degree, by the success
of the BCS theory in explaining (low-temperature) superconductivity in terms of
a free-electron model. Now, HTSC is leading solid-state physicists back to real-
space approaches. Not withstanding the extreme importance of strong electron
correlations, renormalization effects, holons, and spinons, HTSC experimental data
are most often discussed in terms of local atomic-like orbitals, the symmetry of the
Preface xi
orbitals and the interactions between them. That is, the data are discussed in the
jargon characteristic of LCAO models.
Although high-temperature superconductors are not, strictly speaking, per-
ovskites, they share many structural and electronic features in common with the
perovskites. For that reason we have included a chapter on the low-lying quasipar-
ticle bands of these exciting, new materials.
1
Introductory discussion of the perovskites
1.1 Introduction
The mineral CaTiO3 was discovered in the Ural Mountains by geologist Gustav Rose
in 1839 and given the name perovskite in honor of the eminent Russian mineralogist,
Count Lev Alexevich von Perovski. The name perovskite is now used to refer to any
member of a very large family of compounds that has the formula ABC3 and for
which the B ion is surrounded by an octahedron of C ions. Perovskites (MgSiO3
and FeSiO3 ) are the most abundant compounds in the Earth’s crust.
The compounds with the formula ABO3 , with O = oxygen and B = a transition
metal ion, are a subclass of the transition metal oxides that belong to the perovskite
family. Table 1.1 provides a brief list of some well-studied ABO3 perovskites. Many
of the perovskites are cubic or nearly cubic, but they often undergo one or more
structural phase transitions, particularly at low temperatures.
The perovskite oxides are extremely interesting because of the enormous va-
riety of solid-state phenomena they exhibit. These materials include insulators,
semiconductors, metals, and superconductors. Some have delocalized energy-band
states, some have localized electrons, and others display transitions between these
Table 1.1. Some perovskite and related oxides.
Insulating Metallic Magnetic Superconducting

SrTiO3 (n-type)
WO3 ReO3 PbCrO3 Nax WO3 (t)
NaTaO3 NaWO3 LaCrO3 Kx WO3 (t)
SrTiO3 KMoO3 CaMnO3 Kx WO3 (h)
BaTiO3 SrNbO3 LaMnO3 Rbx WO3 (h)
KTaO3 LaTiO3 LaCoO3 Csx WO3 (h)
LiNbO3 LaWO3 LaFeO3 Lix WO3 (h)
t = tetragonal, h = hexagonal
1
2 Introductory discussion of the perovskites
two types of behavior. Many of the perovskites are magnetically ordered and a large
variety of magnetic structures can be found.
The electronic properties of the perovskites can be altered in a controlled man-
ner by substitution of ions into the A or B sites, or by departures from ideal stoi-
chiometry.
The electronic energy bands of the perovskites are very unusual in that they
exhibit two-dimensional behavior that leads to unique structure in properties such
as the density of states, Fermi surface, dielectric function, phonon spectra and the
photoemission spectra.
The perovskites are also important in numerous technological areas. They are
employed in photochromic, electrochromic, and image storage devices. Their ferro-
electric and piezoelectric properties are utilized in other device applications includ-
ing switching, filtering, and surface acoustic wave signal processing.
Many of the perovskites are catalytically active. Development of perovskite
catalyst systems for the oxidation of carbon monoxide and hydrocarbons, and the
reduction of the oxides of nitrogen have been proposed. The perovskites are also
employed in electrochemical applications including the photoelectrolysis of water
to produce hydrogen.
Scientific studies of the perovskites date back many years. The physical prop-
erties of the tungsten bronzes were investigated as early as 1823 [1]. However, it is
only in recent years that experimental and theoretical information on the electronic
structure has begun to become available. Energy band calculations [2], neutron
diffraction and inelastic scattering data [3], photoemission spectra [4], optical spec-
tra [5], and transport data [6] are now available for materials such as ReO3 , WO3 ,
NaWO3 , SrTiO3 , BaTiO3 , KMoO3 , KTaO3 , LaMnO3 , LaCoO3 , and a variety of
other perovskites.
Surface studies of single-crystal perovskites have been performed using pho-
toelectron spectroscopies that indicate that the surface properties are extremely
complex and interesting [7].
In this chapter we present brief discussions of some of the properties of the
perovskite oxides. The discussions are qualitative and intended only to give the
reader a general impression of the types of factors that must be considered. More
quantitative discussions are given in later chapters.
In Section 1.2 we describe the structural features of the perovskites. Sections
1.3 through 1.6 give a qualitative discussion of the electronic states starting from a
simple ionic model and then adding ligand field, covalency, and band effects. Section
1.7 deals briefly with localized d-electron states and why many perovskites do not
have conventional energy bands. In Section 1.8 we touch upon the multiplet config-
1.2 The perovskite structure 3
urations of localized d electrons and their role in determining the magnetic prop-
erties. In Section 1.9 we discuss briefly superconductivity among the perovskites.
The last section, 1.10, is a summary of some of the technological applications of
the perovskites.
1.2 The perovskite structure
The formula unit for the cubic perovskite oxides is ABO3 where A and B are metal
cations and O indicates an oxygen anion. The structure, illustrated in Fig. 1.1, is
simple cubic (O1h , P m3m) with five atoms per unit cell. The lattice constant, 2a, is
close to 4 Å for most of the perovskite oxides.
Figure 1.1. The crystal structure of perovskite oxides with ABO3 formula unit.
The B cation is a transition metal ion such as Ti, Ni, Fe, Co, or Mn. It is
located at the center of an octahedron of oxygen anions. The B site has the full
cubic (Oh ) point group symmetry. The A cation may be a monovalent, divalent,
or trivalent metal ion such as K, Na, Li; Sr, Ba, Ca; or La, Pr, Nd. The A ion is
surrounded by 12 equidistant oxygen ions. The A site also has the point group Oh .
The oxygen ions are not at sites of cubic point group symmetry. Focusing
attention on the oxygen ion marked with an “×” in Fig. 1.1 it may be seen that
the site symmetry is D4h . The B–O axis is a fourfold axis of symmetry and there
are several reflection planes; the yz-plane and planes passing through the edges
containing A sites. The transition metal ion (B site) will experience a cubic ligand
field that lifts the fivefold degeneracy of the d-orbital energies. The oxygen ions
experience an axial ligand field that splits the 2p-orbital energies into two groups.
These splittings are described in the next section.
Well-known examples of cubic perovskites are SrTiO3 , KTaO3 , and BaTiO3
(above the ferroelectric transition temperature). Many of the perovskites that we
shall want to include in our discussions are slightly distorted from the ideal cubic
structure. If the distortions are moderate the general features are not significantly
different from those of the cubic materials. BaTiO3 and SrTiO3 both have structural
transitions to a tetragonal symmetry at certain critical temperatures. Tetragonal
and orthorhombic distortions are very common among the perovskites.
Another class of compounds that we include in our discussions are the pseudo-
perovskites with the formula unit BO3 . Such compounds have the perovskite struc-
ture except that the A sites are empty. Examples of pseudo-perovskites are ReO3
and WO3 .
It is possible to form an intermediate class of perovskites from WO3 by adding
alkali ions to the empty A sites. These compounds, known as the tungsten bronzes,
have the formula unit Ax WO3 where x varies from 0 to 1 and A is H, Li, Na, K, Rb,
or Cs. The structure is often dependent upon the value of x. WO3 is tetragonally
distorted but becomes cubic for x > 0.5. NaWO3 is cubic.
In our discussions we shall also include substituted or mixed compounds of the
form (A1x A21−x )(By1 B1−y
2
)O3 and oxygen-deficient perovskites, ABO3−x . Including
distorted, substituted, and non-stoichiometric compounds, the class of materials
under consideration is very large. Within this broad class, examples may be found
that display almost any solid-state phenomena known.
1.3 Ionic model
The perovskite oxides are highly ionic, but they also possess a significant covalent
character. The ionic model is an oversimplified picture but it serves well as a starting
point for thinking about the electronic properties. The ionic model assumes that the
A and B cations lose electrons to the oxygen anions in sufficient numbers to produce
O2− ions. The usual chemical valence is assumed for the A cations; K+ , Ca2+ , and
La3+ , for example. The ionic state of the transition metal ion is determined by
charge neutrality. If the charge of the B ion is denoted by qB and that of the A ion
1.3 Ionic model 5
Table 1.2. Cations commonly found in perovskite-type oxides. In parentheses is

the coordination number, Z, if the radii given are not for 12 coordination; HS and
SL refer to high spin and low spin, respectively. The effective ionic radii (in Å) are
from Shannon [8].
Dodecahedral A site (Z = 12) Octahedral B site (Z = 6)
Ion Electrons Radius Ion Electrons Radius

Na+ 2p6 1.39 Li+ 1s2 0.76
K+ 3p6 1.64 Cu2+ 3d9 0.73
Rb+ 4p6 1.72 Mg2+ 2p6 0.72
Ag+ 2d10 1.28 (8) Zn2+ 3d10 0.74
Ca2+ 3p6 1.34 Ti3+ 3d1 0.67
Sr2+ 4p6 1.44 V3+ 3d2 0.64
Ba2+ 5p6 1.61 Cr3+ 3d3 0.615
Pb2+ 6s2 1.49 Mn3+ (LS) 3d4 0.58
La3+ 4d10 1.36 Mn3+ (HS) 3d4 0.645
Pr3+ 4f 2 1.18 (8) Fe3+ (LS) 3d5 0.55
Nd3+ 4f 3 1.27 Fe3+ (HS) 3d5 0.645
Bi3+ 6s2 1.17 (8) Co3+ (LS) 3d6 0.5456
Ce4+ 5p6 1.14 Co3+ (HS) 3d6 0.61
Th4+ 6p6 1.21 Ni3+ (LS) 3d7 0.56
Ni3+ (HS) 3d7 0.60
Rh3+ 4d6 0.665
Ti4+ 3p6 0.605
Mn4+ 3d3 0.53
Ru4+ 4d4 0.62
Pt4+ 5d6 0.625
Nb5+ 4p6 0.64
Ta5+ 5p6 0.64
Mo6+ 4p6 0.59
W6+ 5p6 0.60
by qA then qB = 6 − qA where the three oxygen ions contribute the factor of 6. A

list of the common A ions and their valence states is given in Table 1.2.
Once the charge state of the B ion is determined the number of d electrons
remaining is determined from the atomic electronic configuration (Table 1.2). For
example, for SrTiO3 we have Sr2+ and O2− so that the titanium ion is Ti4+ . The
electronic configuration of neutral titanium atom is [Ar] 3d2 4s2 . To form Ti4+ the
outer four electrons are removed leaving the closed-shell Ar core [Ar]. Since O2−
has the [Ne] configuration, all of the ions of SrTiO3 have closed-shell configurations.
The electronic configuration of W is [Xe] 5d4 6s2 . Thus in WO3 the W6+ ion has
a closed-shell [Xe] core; however, for NaWO3 the W5+ ion has a d1 configuration.
The electronic configurations of relevant transition metal ions are given in Table
1.2.
According to the ionic model when all of the ions have closed-shell configura-
tions the material is an insulator. If the B ion retains d electrons then the perovskite
may be a metallic conductor depending on other factors to be discussed. NaWO3
or ReO3 each have d1 configurations and are good metals. For compounds such
as Nax WO3 it is assumed that there will be x d electrons per unit cell. That is,
the Na donates its electron and the W ions donate the remaining electrons needed
to form O2− ions. One may imagine that there are (1 − x) W6+ and x W5+ ions
distributed at random or on an ordered array or that each tungsten ion has an
average valence of W(6−x)+ . The proper picture can not be decided from the ionic
model but depends on other considerations. For Nax WO3 experiments show that
metallic d bands are formed so that we may picture an average valency of (6 − x)+.
However, among the perovskites examples of ordered and random arrays of mixed
valence B ions can also be found.
1.4 Madelung and electrostatic potentials
Starting from the ionic model, other important effects that determine the electronic
properties can be added. The ionic model described above would apply to isolated or
free ions. The ions are, of course, not isolated but interact in several different ways.
One such interaction is through the electrostatic fields due to the charges on the
ions. The most important electrostatic effect is the Madelung potential. The A and
B ions are surrounded by negatively charged oxygen ions. The electrons orbiting
these ions therefore experience repulsive electrostatic (Madelung) potentials. Con-
versely, the electrons orbiting the oxygen ions are surrounded by positively charged
cations and they experience an attractive Madelung potential. The “site Madelung
potentials” are defined as the electrostatic potentials at the different lattice sites
due to all of the other ions. For example, the Madelung potential at a B site located
~ 0 is
at R B
X e2 |qO | X e2 |qA | X e2 |qB |
~B
VM (R 0
)= − − . (1.1)
~0 − R
|R ~ O| ~0 − R
|R ~ A| ~0 − R
|R ~ B|
~O
R B ~A
R B ~ B 6=R
R ~0 B
B
In (1.1), eqO , eqA , and eqB are the charges on the oxygen, A, and B ions, re-
spectively, and R~ O, R
~ A , and R
~ B are the vectors for the corresponding lattice sites.
The site Madelung potentials are very large for the perovskites because of the
large ionic charges. Typical Madelung potentials are 30–50 eV for the B site. For
1.4 Madelung and electrostatic potentials 7
A2+ B 4+ O2−
3 perovskites the (full ionic) site potentials [9] are: VM (B) = +45.6 eV,
VM (A) = +19.9 eV, and VM (O) = –23.8 eV. A table of Madelung potentials can be
found in Appendix D.
The stability of the perovskite structure is largely due to the energies associated
with the Madelung potentials. The attractive potential at the oxygen sites allows
the oxygen ions to bind a pair of electrons. In effect the site potential adds to the
electron affinity of the oxygen ion. The affinity of O− for the second electron is
actually positive. This means that the second electron would not be bound on a
free oxygen ion. O2− is stable in the lattice because of the attractive site Madelung
potential. Conversely, a d electron is bound to a Ti4+ ion with an (ionization)
energy of –43 eV. In the absence of the repulsive site Madelung potential, donation
of an electron from the Ti3+ to an O− ion in SrTiO3 would be energetically very
unfavorable. The site Madelung potential adds to the ionization energy so that the d
electron would have an effective binding energy of –43 + 45.6 = +2.6 eV (unbound)
for SrTiO3 with the full ionic charges.
Thus, it is seen that the Madelung potentials are responsible for the ionic
configurations.
An orbital centered on an ion has a finite radial extent so that an electron in
such an orbital would sample the electrostatic field over a distance comparable to
the ionic radius. In order to determine the complete effect of the electrostatic field
on the electron state we need to know the behavior of the field as a function of
position near each ion site. If we use the point ion model then,
e2 |qB |
V (~r) = − + Ves (~r) ,
|~r − R ~0 |
B
X e2 |qB | X e2 |qA | X e2 |qO |
Ves (~r) = − − + . (1.2)
|~r − R ~ B| ~ A|
|~r − R ~ O|
|~r − R
~ ~0
RB 6=RB RA ~ RO ~
The potential near R ~ 0 can be found by expanding Ves (~r) in terms of spherical
B
harmonics centered at R ~ 0 . The potential Ves (~r) then takes the form of an electric
B
multipole expansion. The monopole term is just the site Madelung potential. Thus,
as we have described, the site Madelung potential produces a shift in the energy of
an electron localized on the site.
The higher-order multipoles (dipole, quadrupole, etc.) create an electrostatic
field (with the point group symmetry of the site) which leads to a lifting of the
orbital degeneracies. The effect of the cubic electrostatic field at the B ion site is to
split the fivefold degenerate d states into two groups as shown in Fig. 1.2(c). The
eg group is doubly degenerate corresponding to the d orbitals having wavefunctions
with angular symmetry (x2 − y 2 )/r2 and (3z 2 − r2 )/r2 . The threefold degenerate
t2g group corresponds to the states (xy/r2 ), (xz/r2 ), and (yz/r2 ).
1
s (1) s (1)
O – ion
eg (2)
d (5)
t2g (3)
A – ion 6
Eg
p⊥ (2)
2p (3) ?
pk (1)
B – ion
(a) (b) (c)
Figure 1.2. Effect of the electrostatic potentials on the ion states: (a) free ions,
(b) Madelung potential, and (c) electrostatic splittings.
The oxygen 2p states are split by the axial electrostatic field into a doubly
degenerate level denoted by p⊥ and a non-degenerate pk state. The notation p⊥
and pk refer to 2p orbitals oriented perpendicular and parallel to a B–O axis,
respectively.
The lowest unoccupied state of the A ion is an s state. Its energy is shifted by
the monopole (Madelung potential) but unaffected by the other multipole terms
because it is a spatially non-degenerate function with spherical symmetry at a site
of cubic symmetry.
The particular level ordering shown in Fig. 1.2 may be understood by consid-
ering the orientation of orbitals relative to the charge distributions on neighboring
ions. The eg orbitals have lobes directed along the B–O axis and directly into the
negative charge clouds of oxygen ions. The t2g orbitals have lobes pointed perpen-
dicular to the B–O axis between the negative oxygen ions. As a result the eg states
experience a greater repulsion than the t2g states and consequently lie at a higher
energy. Similar reasoning suggests that the pk states lie below the p⊥ states when
it is noted that B ion cores appear as positively charged centers.
In insulating perovskites such as SrTiO3 the p states are completely filled while
the d states are completely empty. The energy difference, Eg , between the t2g and
p⊥ states is approximately equal to the energy gap. Metallic and semiconducting
materials have the d states partially filled. NaWO3 or ReO3 have a single electron
in a t2g state.
In most but not all cases the energy bands involving the s state of the A ion
are at energies much higher than the primary valence and conduction bands of a
perovskite and therefore these bands are unoccupied. As a result the s state of
the A ion usually does not play any significant role in determining the electronic
properties. This is not to say that the A ion is not important. The electrostatic
potentials of the A ions have a strong influence on the energy of the p–d valence
and conduction bands. Furthermore, the size of the A ion is a significant factor in
determining whether the crystal structure is distorted from the ideal cubic form.
Nevertheless, given a particular perovskite structure and the effective electrostatic
potentials acting on the B and O sites, the orbitals of the A ion may usually be
omitted from electronic structure calculations. This leads to a major conceptual
simplification because the electronic properties of the perovskites may be regarded
as arising solely from the BO3 part of the ABO3 structure. This implies, for exam-
ple, that the electronic structure of BaTiO3 and SrTiO3 should be essentially the
same. According to the same reasoning the electronic structure of Nax WO3 should
be independent of x. This does not mean that the properties are the same, but only
that the available electronic states are the same. Obviously, the properties of WO3
are completely different from those of NaWO3 ; the former is an insulator and the
latter is a metal. However, as a first approximation the only effect of the sodium is
to donate electrons which occupy the t2g states of the tungsten ion.
1.5 Covalent mixing
In addition to electrostatic interactions, the ions can interact because of the overlap
of the electron wavefunctions. This leads to hybridization between the p and d
orbitals and the formation of covalent bonds between the transition metal ions and
the oxygen ions. It is frequently assumed that the covalent mixing in insulating
materials such as SrTiO3 is negligible. This is not correct. Nearly all of the physical
and chemical properties of the perovskites are significantly affected by covalency.
To understand covalent mixing we consider a cluster of atoms consisting of a
transition metal ion and its octahedron of oxygen ions. The wavefunctions of the
cluster can be written in the form:

X X X (n)
ψ (n) (~r) = a(n) ~ i ),
α ϕdα (~
r) + biβ ϕpβ (~r − R (1.3)
α ~i
R β
where ψ (n) (~r) is the cluster wavefunction for the nth eigenstate. ϕdα (~r) is a d orbital
~ i ) is a p orbital centered
on the B ion of α-type (α = xy, xz, . . ., etc.) and ϕpβ (~r − R
at an oxygen ion located at R ~ i of the βth-type (β = x, y, or z). The coefficients
(n) (n)
aα and biβ are constants which specify the amplitudes of the different orbitals
which compose the nth eigenstate.
pz
z
pz pz
y y
d3z2 −r2
x dyz
pz
(a) (b)
Figure 1.3. Overlap between cation d orbitals and anion p orbitals. (a) Sigma overlap
and (b) pi overlap.
(n)
For the ionic model the wavefunctions are either pure d orbital (biβ = 0) or
(n)
pure p orbital (aα = 0). For the cluster the wavefunctions are still predominantly
d or p orbital in character but there is a significant covalent mixing between the two
(n) (n)
(both biβ and aα 6= 0). The mixing comes about because of the overlap between
d orbitals centered on the cation and the p orbitals on neighboring oxygen ions.
There are two types of p–d overlap. The first is overlap between the d orbitals of
the eg type with p orbitals of the pk type. This overlap is called “sigma” overlap.
The second type, “pi” overlap occurs between t2g -type d orbitals and p⊥ orbitals.
These two types of overlap are illustrated in Fig. 1.3. The overlap between t2g and
pk orbitals or between eg and p⊥ orbitals vanishes by symmetry. If only the p and
d orbitals are considered then there are 23 cluster states for a transition metal ion
and the octahedron of oxygen ions. These 23 cluster states arise from admixtures
3 eg (2)
d 10Dq
∆es
2 t2g (3)
1 t1g (3)
p
6 t1u (3)
1 t2u (3)
2 eg (2)
1 t2g (3)
6 a1g (1)
(a) (b) 5 t1u (3)
Figure 1.4. (a) BO6 cluster and (b) the cluster levels. The dashed levels are for the
electrostatic model. ∆es is the electrostatic splitting.
of the 23 basis states; 5 d orbitals and 18 p orbitals, three on each of the six oxygen
ions.
The cluster energy levels [10] are illustrated in Fig. 1.4. The labels given to
the cluster energy levels indicate the group theoretical irreducible representations
to which the wavefunctions belong. The prefix numbers are used to distinguish
different levels which have the same symmetry properties. The degeneracies of the
levels are indicated by the numbers in parentheses.
It is noted that the cation d orbitals are still split into the eg and t2g groups.
These, so-called “ligand-field states” differ from those of the electrostatic model
(Fig. 1.2) in two significant ways. First, the wavefunctions are no longer just d
orbitals. They are admixtures of p and d orbitals. A second difference is that the
splitting between the eg and t2g groups is much larger than for the electrostatic
model. The cluster ligand-field splitting denoted by 10Dq is due to both electrostatic
and covalent effects. The covalent contribution to 10Dq is usually much larger than
the electrostatic contribution, ∆es . Typically 10Dq is 2–3 eV in magnitude.
The ligand-field states, 3eg and 2t2g , have wavefunctions in which the d orbitals
combine out-of-phase with the p orbitals. The interference between the orbitals leads
to a depletion of charge between the B and O ions. For this reason these states are
called antibonding states. Bonding states are formed from in-phase combinations
of the d and p orbitals. These states have wavefunctions that correspond to an
accumulation of charge between the B and O ions. The bonding states are the 2eg
and 1t2g levels (shown in Fig. 1.4). These states have hybridized wavefunctions,
typically 70% p orbital and 30% d orbital. The percentage d-orbital admixture is a
measure of the covalent bonding.
The remaining cluster levels have wavefunctions that are combinations of p
orbitals located on the six oxygen ions. They do not hybridize with the d orbitals
and therefore they do not contribute to the metal–oxygen bonding. Such states are
called non-bonding states. Wavefunctions of the three types of cluster states are
illustrated in Fig. 1.5.
y y y
x x x
(a) 2t2g (b) 1t2g (c) 1t1g
Figure 1.5. Cluster states: (a) antibonding, (b) bonding, and (c) non-bonding.
It is important to note that electrons occupy d orbitals on the cation even when
the 3eg and 2t2g levels are unoccupied. This is because of the covalent mixing of the
d orbitals into the filled valence states below the 2t2g level. This covalency effect
is significant even for “ionic” insulators such as SrTiO3 . The ionic model implies
that the titanium ion is Ti4+ with a d0 configuration. Cluster models would give
an effective valence such as Ti3+ (d1 ).
1.6 Energy bands
In the preceding section we considered a cluster model for the perovskites in which
the transition metal ion interacts with the nearest-neighbor oxygen ions. The co-
valent mixing between the cation and anion wavefunctions leads to a partial oc-
cupation of d orbitals which, in the ionic model, were empty. A mechanistic in-
terpretation of the covalent mixing is that the overlap between cation and anion
wavefunctions provides a means of transferring electrons back and forth between
the ions. Clearly, for an extended crystal structure the same mechanism will al-
low electrons to be shared between cations in adjacent clusters. Each oxygen of a
given cluster is shared by adjacent cations. Cations can interact with each other
through the intervening oxygen ion. An electron on a cation may be transferred
1.6 Energy bands 13
to the oxygen ion and then from the oxygen ion to the second cation. When such
processes occur the electrons become delocalized and electron energy bands are
formed. It is important to note that the formation of d-electron bands requires two
independent electron transfer processes. The delocalization of d electrons therefore
is second order in the p–d overlap (or the probability of p to d electron transfer).
This is quite different from a typical monatomic metal where delocalization is first
order in the atomic overlap. For cubic perovskites the cation–cation separation is
nearly 4 Å. This is too large for a significant direct overlap between cation orbitals
and therefore band formation occurs by transfer of electrons between cations and
anions whose separation is only about 2 Å.
In considering the energy bands of a perovskite it is appropriate to divide the
crystal into unit cells each with the formula unit ABO3 . (The unit cell is shown
in Fig. 1.1.) As discussed previously, the s states of the A ion can be neglected.
Therefore, there will be 14 energy bands corresponding to the five d orbitals
and nine p orbitals of each unit cell. The wavefunctions of the band states are
characterized by a wavevector ~k and are of the form
XX ~ ~
XX ~ ~
Ψ~k (~r) = aα (~k) eik·Rd ϕdα (~r − R
~ d )+ bβ (~k) eik·Rp ϕpβ (~r − R
~ p ). (1.4)
~d
R α ~p
R β
~ ~ ~ ~
In (1.4), aα (~k) eik·Rd and bβ (~k) eik·Rp are respectively the amplitudes of the d and
p orbitals of symmetries α and β located at the lattice sites R ~ d and R
~ p.
An energy band diagram for a typical perovskite is shown in Fig. 1.6 for a
model which includes only the interactions between nearest-neighbor ions [11]. For
this simple model the energy bands divide into a set of sigma bands and a set of pi
bands. The sigma bands involve only the eg d orbitals and the pk oxygen orbitals.
The pi bands involve only the t2g d orbitals and the p⊥ oxygen orbitals.
The sigma bands have five branches: two distinct σ-type valence (bonding)
bands, two distinct σ ∗ -type conduction (antibonding) bands and a single σ 0 -type
non-bonding band. The pi bands have nine branches: three equivalent π-type va-
lence (bonding) bands, three equivalent π ∗ -type conduction (antibonding) bands,
and three equivalent π 0 -type non-bonding bands.
The bonding and antibonding (σ, σ ∗ , π, π ∗ ) bands have wavefunctions whose
p–d admixture varies as a function of the wavevector ~k. At Γ(~k = 0) in the first
Brillouin zone (see the inset in Fig. 1.6) the wavefunctions are pure p or pure d
orbital in composition. The states at Γ have no covalent character and therefore
correspond to the levels derived from the ionic model including the electrostatic
potentials (Fig. 1.2(c)). As ~k varies along Γ → X → M → R the covalent mixture
of the p and d orbitals increases. It is maximum at the point R, at the corner of
the Brillouin zone. The states at R are very similar to the “g” states of the cluster
σ∗
10Dq
Eσ∗
? 6
π∗
∆d
kz 6
Energy E
Eπ∗ R ?
6
Γ X-
Eg ky
+ M
kx
? ? π◦
Eπ
∆p σ◦
Eσ
6
Γ X M RM Γ R X
Figure 1.6. Energy bands for a typical perovskite showing the dispersion for ~k-vectors
along various lines in the Brillouin zone (inset) according to the LCAO model with nearest-
neighbor interactions. The lighter curves are the pi bands and the darker curves are the
sigma bands. The energies, Eg , 10Dq, ∆d , and ∆p are the band gap, total (cluster)
ligand-field splitting, d-orbital ligand-field splitting, and the p-orbital ligand-field split-
ting, respectively.
model (i.e., 2t2g , 3eg , etc.). Thus the ionic model underestimates the covalency and
the cluster model overestimates the covalency of the perovskites. The separation
between the σ ∗ and π ∗ bands at Γ, ∆es (d), corresponds to the electrostatic con-
tribution to the ligand-field splitting. The separation at R is the total ligand-field
band splitting and is approximately equal to 10Dq.
1.6 Energy bands 15
The non-bonding band states for σ 0 and π 0 involve only oxygen 2p orbitals
and therefore do not involve metal–oxygen covalent mixing. The band and cluster
models produce similar non-bonding states.
The energy separation between the π ∗ and π 0 bands at Γ is the fundamental
band gap, Eg . It varies between 1 and 4 eV and is largest for the insulating per-
ovskites. Covalent mixing decreases with increasing band gap. The magnitude of
the band gap is a measure of the ionicity of a perovskite. For example, the band
gap of SrTiO3 is 3.25 eV and that of ReO3 is about l eV. This means that SrTiO3
is much more ionic than ReO3 .
Insulating perovskites (e.g., SrTiO3 , BaTiO3 , or WO3 ) have filled valence
bands; that is, the σ, π, σ 0 , and π 0 bands are completely occupied with electrons.
The conduction bands (σ ∗ and π ∗ ) are empty. Metallic perovskites such as NaWO3
or ReO3 have one electron per unit cell in the π ∗ conduction band. Examples of
metallic compounds with two electrons in the π ∗ band are CaMoO3 , BaMoO3 , and
SrMoO3 . Perovskites with more than two d electrons tend to form localized-states
similar to those of the cluster model rather than delocalized band states.
Insulating perovskites can be rendered semiconducting or metallic by several
means. Reduction in a hydrogen atmosphere produces oxygen vacancies. The vacan-
cies act as donor centers; two electrons being donated by each vacancy (hydrogen
itself may also remain in the lattice and act as a donor). Electron concentrations
in the range of 1016 –1020 electrons/cm3 can be produced in this way. Reduced in-
sulating perovskites are n-type semiconductors with the Fermi level very near to
the bottom of the π ∗ conduction band. n-type SrTiO3 has been found to be a
superconductor at temperatures below 0.3 K [12].
Insulating perovskites can also be doped by substituting appropriate ions into
either the B or A sites. The tungsten bronzes Nax WO3 , Kx WO3 , Lix WO3 , and
Hx WO3 are special cases in which donor ions are substituted into the empty A
sites of insulating WO3 . Electron concentrations of the order of 1022 electrons/cm3
are obtained in this case. Many of the bronze compositions are superconductors.
One of the reasons perovskites are particularly valuable for research is that
the electronic properties can be varied in a controlled fashion to produce almost
any desired feature. The Fermi level in SrTiO3 can be varied over a 3 eV range by
going from cation- to anion-deficient compositions. The basic band structure does
not change appreciably so the properties of such compositions are easily understood
and interpreted in terms of a fixed band structure; that is the “rigid-band” approx-
imation is valid. The rigid-band model is also applicable to the tungsten bronzes,
(1) (2)
and mixed compounds of the Ax A1−x BO3 type where A(1) and A(2) are different
cations.
1.7 Localized d electrons
In the preceding section we indicated how the localized cluster states are delocal-
ized because of the overlap of wavefunctions between adjacent clusters. The d-band
formation is due to the transfer of electrons between cations via intervening oxygen
ions. These electrons become delocalized and have an equal probability (propor-
~ ~
tional to |eik·R |2 =1) of being found at any cation site. The band model neglects
any possible spatial correlation between d electrons. The potential experienced by
a given electron is assumed to be the same at every lattice site and equal to the av-
erage potential of the ion core and all other electrons. The usual one-electron band
model explicitly ignores the fact that at any given instant of time a non-average
number of electrons may be occupying the orbital of an ion. However, during the
lifetime of the “non-average” ionic state the electrons on the site will experience
a non-average potential. In particular, the intra-atomic Coulomb repulsion of an
electron on a non-average site will be different from that at an average site.
Consider the situation in which we start with two metal ions each having n
electrons. The electron–electron repulsion energy among the n electrons at each
site is 21 U n(n − 1) where U is the Coulomb integral. If we transfer an electron from
one site to the other there will be n − 1 electrons on one site and n + 1 on the
other. The electron–electron repulsion energy will be 21 U n(n + 1) on the site with
the extra electron and 21 U (n − 2)(n − 1) on the other site. There is a change in
the repulsion energy at one site of 12 U [n(n + 1) − n(n − 1)] = nU . At the other site
the change in energy is 12 U [(n − 2)(n − 1) − n(n − 1)] = −U n + U . Therefore, the
net change is an additional repulsive energy equal to U . Thus, there is a Coulomb
energy barrier to the creation of non-average ionic states.
Band formation is favorable because the delocalization of an electron reduces
its kinetic energy (provided that the electron can occupy a state near the bottom
of the band). For such a case the reduction in kinetic energy increases as the band
width increases.
It is clear from what has been said that energy band formation will only be
favorable if the reduction in kinetic energy is larger than the increase in the Coulomb
energy. A variety of models which include a form of the Coulomb correlation energy
have been used to find a criterion for the validity of the band model [13]. In general
it is found that band theory applies when W & U where W is the band width. For
W less than U , localized d-electron states are energetically favored. The precise
criterion is model-dependent.
The localized electron criterion leads to interesting possibilities for the per-
ovskites. The band width of the σ ∗ band is substantially larger than that of the
π ∗ band and consequently, for a number of perovskites, the t2g states are localized
1.8 Magnetism in the perovskites 17
while the eg states form σ and σ ∗ energy bands; LaNiO3 with filled t2g states and
a single electron in the σ ∗ band is an example [14].
1.8 Magnetism in the perovskites
The occurrence of magnetism in the perovskites is closely connected to the exis-

tence of localized d electrons. In almost all cases where magnetism exists the d
electrons are localized and possess localized spins. In such cases the local electronic
configuration becomes an important consideration. One must be concerned with
the multiplet structure. The tendency toward the formation of a multiplet configu-
ration with a net spin arises from intra-atomic exchange and correlation. In atomic
theory, Hund’s rule states that the lowest-energy configuration corresponds to the
state of maximum multiplicity or maximum spin and orbital angular momentum.
Hund’s rule is qualitatively applicable to the perovskites with localized d electrons.
There are, however, some significant differences between atomic theory and the
theory applicable to ions of the solid. The major differences between free ions and
the cations in a solid perovskite are:
(1) the fivefold degenerate d states are split into the eg and t2g groups with a splitting of
10Dq;
(2) the energy differences between different electronic configurations are not as widely
separated as for the free ions;
(3) there is significant covalent mixing between the d-ion orbitals and the neighboring
oxygen ion p orbitals.
As a consequence of (1) and (3) the electronic configuration of the cation

should be specified in terms of the one-electron cluster states 3eg and 2t2g . For
simplicity the numerical descriptors of these states may be omitted. The d-electron
configuration may then be specified by (tn2g em
g ), where n and m are the occupations
of the 2t2g and 3eg levels, respectively.
The effect of (2) is that different valence states and different electronic con-
figurations of the cation are closer in energy to each other than for the free ion.
This is a result of polarization and electron screening of the Coulomb interactions.
On applying Hund’s rule to a perovskite cation the ligand-field splitting must be
taken into account. When the number of d electrons, m+n, is between 4 and 7,
Hund’s rule can be violated if the ligand-field splitting is greater than the intra-
atomic exchange energy. Consider, for example, LaMnO3 which has Mn3+ ions
with four d electrons. The intra-atomic exchange favors the “high-spin” configura-
tion 5 Eg = (t2g ↑3 eg ↑). However, occupying the eg state involves a loss of binding
energy equal to the ligand-field splitting. Therefore, the “low-spin” configuration
3
T2g = (t2g ↑3 t2g ↓) is competitive. Assuming a constant exchange, J, between par-
allel spin electrons, the intra-atomic exchange involves

X
Eex = −J ~si · ~sj ,
i>j
where ~si and ~sj are the spins of the occupied states. The 5 Eg has an exchange
energy − 32 J while for the 3 T2g , Eex = − 34 J. However, the 3 T2g has a ligand-field
energy of 10Dq. Therefore, the difference in the energies of the two configurations
is
3
E(5 Eg ) − E(3 T2g ) = − J + 10Dq ≡ ∆E.
4
When ∆E < 0 the high spin state 5 Eg (spin = 2) is lower in energy than the low spin
state 3 T2g (spin = 1). If ∆E > 0 then the low spin state is favored. Experiments on
d4 ions in perovskites show that the low spin state is usually favored. This indicates
that the ligand-field splitting is larger than the intra-atomic exchange and Hund’s
rule does not apply.
When the cations possess localized spins, then long-range magnetic ordering
can occur. The principal mechanism of spin–spin interactions is superexchange.
Superexchange involves the antiferromagnetic coupling between nearest-neighbor
cations by exchange of electrons with the intervening oxygen ion.
Examples of magnetically ordered perovskites are LaCrO3 , PbCrO3 , CaMnO3 ,
LaFeO3 , and many others. Those named above form the simple G-type magnetic
cell in which the spins of nearest-neighbor cations are antiparallel. Many other types
of magnetic ordering also occur among the magnetic perovskites.
As a final comment on localized d electrons we mention the importance of the
Jahn–Teller effect. This effect is the spontaneous distortion of a cubic structure
such as that of perovskites. When the cation electronic configuration is orbitally
degenerate, the ground state will in some cases, be unstable to small distortional
displacements. This Jahn–Teller distortion occurs because the electronic energy
decreases linearly with displacement while the elastic energy increases as the square
of the displacement. A minimum in the total energy always occurs for a small but
finite distortional displacement.
1.9 Superconductivity
Superconductivity has been observed for n-type SrTiO3 and for many of the com-
positions of the tungsten bronzes: Lix WO3 , Nax WO3 , Kx WO3 , Rbx WO3 , and
Csx WO3 . The occurrence of superconductivity in compounds whose elements are
not superconducting and for which more than three-fifths of the atoms are oxygen
is truly remarkable.
1.9 Superconductivity 19
WO3 is an insulator with a tetragonally distorted perovskite structure. With

the addition of alkali ions to the empty A site a variety of metallic bronzes can
be formed. The tungsten bronzes occur with cubic, hexagonal, and two different
tetragonal perovskite-like structures [15]. For Nax WO3 the tetragonal I phase occurs
in the range 0.2 < x < 0.5. For values of x < 0.2 the tetragonal II phase exists. For
values of x > 0.5 the cubic perovskite structure is stable. Tetragonal I, Nax WO3
and Kx WO3 are superconducting with transition temperatures of 0.57 K [16] and
1.98 K [17], respectively. The cubic and tetragonal II phases are apparently not
superconducting. Except for Nax WO3 and Kx WO3 , superconductivity occurs for
the other bronzes when they are in the hexagonal phase [17–19]. The transition
temperatures of the hexagonal bronzes are close to 2 K.
It has been found that the transition temperature of the hexagonal bronzes
can be raised by a factor of 2 or 3 by etching in various acids [18]. The reasons for
this enhancement are not yet clear. The transition temperature of the tetragonal
I sodium tungsten bronze, Nax WO3 , increases rapidly as x approaches 0.2 [19].
This enhancement occurs as the composition approaches the tetragonal II phase
boundary, and is presumed to be associated with a lattice instability.
More recently various alloys of barium bismuthates have been studied exten-
sively. The highest recorded Tc for a non-layered metal oxide is about 30 K for the
alloy Ba1−x Kx BiO3 for x = 0.38. This superconducting material displays a tran-
sition to an insulating state at x < 0.38, but is a cubic, superconducting metal
for 0.38 < x < 0.6. The related compound BaPb1−x Bix O3 is also a superconduc-
tor with a maximum Tc of about 13 K. BaBiO3 itself is an insulator even though
according to conventional band theory it possesses a half-filled conduction band
(antibonding Bi 6s–O 2p sigma band).
In 1986 Bednorz and Müller [20] discovered a new class superconducting metal
oxides (La2 CuO4 doped with Ba2+ , Sr2+ , or Ca2+ ) one of which possessed a criti-
cal temperature, Tc , in excess of 30 K. Their discovery was followed by a worldwide
research effort that turned up many other cuprate superconducting materials with
even higher critical temperatures, the record high being around 166 K, a temper-
ature that is above the boiling point of liquid nitrogen. These “high-Tc cuprate
superconductors” are characterized by sets (one or more layers) of “immediately
adjacent” planes of copper ions surrounded by four oxygen ions. Each set of “im-
mediately adjacent” layers is separated from the next set by “isolation layers”
(La–O planes in the case of La2 CuO4 ) that are poorly conducting. Despite inten-
sive experimental and theoretical research efforts, the mechanisms underlying the
high-temperature superconductivity as well as the properties of the “normal” state
above Tc are not well understood. However, there seems to be agreement that the
two-dimensional character of the Cu–O bonding and the resulting large density of
states are important. The question of whether the mechanism of electron pairing
in these materials can be explained within the framework of “conventional”, BCS

(phonon-mediated) superconductivity theory [21] is still open and other significantly
different mechanisms have been put forward [22].
In general, these high-Tc superconductors do not have the perovskite structure.
However, they share an important characteristics with the metal-oxide perovskites.
Even though the perovskites are not layered in the way the high-Tc superconductors
are, the electronic coupling between the transition metal d orbitals and the oxygen
p orbitals is two-dimensional because of the symmetry of the interactions.
1.10 Some applications of perovskite materials
The technological uses of perovskite and perovskite-related materials are extensive

and we will not attempt to review the field. In this section we shall only briefly
mention some of the common applications. References are given to only a few
representative papers in the vast literature.
The piezoelectric insulating perovskites such as BaTiO3 , PbTiO3 , PbZrO3 ,
Pb(Zrx Ti1−x )O3 (PZT), and LiNbO3 have been employed extensively as solid-state
device materials. Some solid-state applications include switching devices, infrared
detectors, and a large variety of signal processing devices [23]. These materials are
employed as substrates for the generation of bulk and surface acoustic (elastic)
waves. Because of their piezoelectric properties, acoustic waves are accompanied
by an oscillating electric field [24, 25]. It is possible to generate acoustic waves
by applying an oscillating electric field to the substrate and conversely an acous-
tic wave may be detected by the electric field it generates. The coupling between
the elastic displacement and the electromagnetic field is nonlinear and produces
second-harmonic electric fields [26]. These properties have been employed to design
a number of acoustic wave signal processing devices, including time delay lines,
filters, acoustic wave image devices and nonlinear convolution and correlation de-
vices.
The nonlinear optical properties of the perovskite insulators are used for the
generation of second-harmonic optical waves. The second-harmonic generation co-
efficient of PbTiO3 is among the highest known [27]. Other applications of the in-
sulators include photochromic, electrochromic, image storage, and display devices
[27]. In the photochromic applications the transparent host materials are doped
with impurity transition metal ions or rare earth ions. The impurity ions have sev-
eral localized levels lying within the band gap of the substrate that correspond
to different valence states. The valence state of the impurity ion can be changed
by photoexcitation. Impurity ions are selected for which one valence state has an
absorption band in the visible while the second does not. Colored images can be
“written” by a light beam that causes photoexcitations of these metastable valence

states. The images can be erased by a second light beam of a different wavelength
which depopulates the metastable states. For electrochromic devices, the valence
states are changed by shifting the quasi-Fermi level or by reducing the material
electrochemically. A thin film of WO3 may be changed from transparent to a deep
blue color by electrochemically converting W6+ ions to W5+ ions.
There is also interest in the surface chemical properties of the perovskites.
Many are excellent gas-phase catalysts and in addition several are photocatalytic
and electrocatalytic.
Interest in the catalytic properties of the perovskites began with the suggestion
that the rare earth cobalt oxides, RCoO3 (R = rare earth ion) might prove useful
as substitutes for Pt-based automotive exhaust catalysts [24, 25]. Strong catalytic
activity and a high degree of selectivity have been found for a large variety of
perovskite materials [28].
Many energetically favorable (exothermic) gas-phase reactions do not occur
spontaneously but require a catalyst in order to occur. The detailed mechanisms
of catalytic action are not known in most cases, but some general features are
understood. There are two principal factors which inhibit an exothermic reaction.
The first is related to the symmetry of the reactant and product states. As two
molecules come together in a chemical reaction the orbitals of the complex must
evolve from those of the reactants to those of the products. One can imagine a con-
tinuously changing set of hybridized molecular orbitals for the reacting molecules.
It frequently happens that the occupied orbitals of the ground state of the product
can not evolve from a hybridization of the occupied orbitals of the reactants. An
example is the hydrogenation of ethylene to form ethane. The bonds associated
with the hydrogen atoms of ethane evolve from a hybridization of empty anti-
bonding states of the reacting hydrogen and ethylene molecules, and not from the
occupied bonding states. Electron flow from the occupied bonding states to the
empty antibonding states is forbidden by symmetry consideration. Reactions can
be classified as symmetry-“allowed” or symmetry-“forbidden” in much the same
way as optical translations [29]. Reactions that are “forbidden” are inhibited by a
symmetry-imposed barrier.
Another barrier encountered is associated with the charge transfer involved
in the reaction. As charge flows from the reactant states to the ground state of
the product, the molecules must often pass through a transient polar configuration.
Such polar configurations are usually energetically unfavorable because the electron
affinity of a molecule is small compared to its ionization energy. The inhibition to
charge transfer acts as an additional barrier to the chemical transformation.
The catalytic properties of the perovskites are directly related to the presence
of coordinatively unsaturated transition metal ions on the surface. The term coordi-
natively unsaturated, refers to the fact that an ion on the surface will often have less
than its normal complement of six oxygen ligands. Such ions provide active sites for
adsorption of reactant molecules because in this way the ion can attain its normal
number of ligands. The symmetry of the d orbitals is favorable for interaction with
both the bonding and antibonding states of most molecules.
It is generally believed that chemisorption of one or more of the reactant
molecules to form a surface complex is a precursor to a catalyzed reaction. The
role of the surface complex in catalysis is twofold: d orbitals can hybridize with the
reactant molecule orbitals in such a way as to provide a symmetry-allowed path
for the reactions [30]. In addition, the adsorption of the reaction species greatly
facilitates charge transfer processes. When molecules condense on a solid substrate
the ionization energy of the molecular levels is reduced due to a process known as
extra-atomic relaxation [31]. Furthermore the barrier to charge transfer is reduced
by the solid-state effects of polarization and electron screening. It is also possi-
ble for charge transfer to occur via the transition metal ion. The catalyst ion acts
as an intermediary to accept (donate) electrons from the reactants and to donate
(accept) electrons to the product. This process involves a valence fluctuation of
the cation. Such fluctuations are of low energy compared to fluctuations of charge
on free molecules. The energy required for a valence fluctuation can be minimized
in systems such as the mixed or non-stoichiometric perovskites since they already
contain mixed valence transition metal ions.
There are several factors that make the perovskites particularly attractive as
catalyst systems for research. One factor is that they form a large class of struc-
turally similar compounds whose electronic properties can be varied in a controlled
way. This permits a systematic study of the effects of variations in electronic pa-
rameters on catalytic rate, for example. (Pt is an excellent catalyst but there is
little that can be done to vary its electronic state and therefore to discover why
it is such a good catalyst.) A second factor making the perovskites important as
catalysts is that they are highly stable at high temperatures and in hostile chemical
environments.
Voorhoeve et al. [28] have reported extensive studies on a variety of perovskite
catalysts. Co, Mn, and Ru perovskites have been investigated as catalysts for the
oxidation of carbon monoxide and hydrocarbons and for the reduction of the ox-
ides of nitrogen. (Such catalytic conversions are important in removing pollutants
from auto exhaust.) Particular examples of perovskite catalysts investigated include
SrRuO3 , LaRuO3 , and the substituted system (Lax K1−x )(Ruy Mn1−y )O3 . The cat-
alysts are very active and highly selective in the reduction of nitrogen oxides. The
use of substituted systems permits a controlled variation of valence states for the
cations. The electronic properties can be tailored for a particular application.
Another important feature of perovskite catalysts is that they can be designed

to simultaneously catalyze reduction and oxidation reactions. An example of such a
catalyst is (La0.8 Sr0.2 )(Co0.9 Ru0.1 )O3 . Several perovskite compositions have been
found to be superior or comparable to commercial catalysts.
It is also noted that surface oxygen vacancies on a perovskite can serve as
catalytically active sites. NO is believed to dissociate into adsorbed N by reaction at
a vacancy site on manganite catalysts. The liberated oxygen from the NO molecule
can fill the empty surface oxygen site. The vacancy can be restored by reaction
with an reducing agent such as CO. Such sites are useful in catalysis systems where
both reduction and oxidation processes are desired.
The perovskites have not yet emerged as commercially competitive catalysts,
but have proved valuable in the study of the possible mechanisms of catalysis.
As a final topic in the applications of perovskites we mention their use as
electrochemical electrodes.
Materials such as LaCoO3 , n-type SrTiO3 and the tungsten bronzes have been
utilized as anode materials in electrochemical cells. They are particularly useful
because of their stability in an electrolyte. Nearly ideal reversible electrode behavior
has been shown for LaCoO3 and related compounds such as (La0.5 Sr0.5 )CoO3 [32].
A particularly interesting application is the photoelectrolysis of water. The
electrochemical cell consists of an n-type anode such as SrTiO3 , BaTiO3 or some
substituted perovskite and a Pt counterelectrode [33–36]. The electrolyte may be
either alkaline or acid aqueous solution. The electrodes are connected through an
external circuit and the electrolysis process is driven photocatalytically by photons
incident on the anode surface.
In alkaline aqueous solution the anode reaction is
1
2 p+ + 2 OH− → O2 + H 2 O ,
2
where p+ designates a hole. Electron–hole pairs are generated in the oxide anode
by absorption of incident photons with energies equal to or greater than the band
gap. The electrons and holes are separated by the internal electric field of the
oxide (due to band bending at the oxide–electrolyte interface). In n-type materials
such as SrTiO3 the band bending creates an electron depletion (hole accumulation)
region at the surface. The holes combine with adsorbed hydroxyl ions to produce
molecular oxygen and water as indicated by the anode reaction above. The electrons
are discharged at the cathode, producing hydrogen.
The feasibility of photoelectrolysis for the production of hydrogen has been
demonstrated using band-gap photons with ~ω ∼ 3 eV for several oxides. The first
such experiments utilized n-type TiO2 [36]. Later studies employed perovskites
such as SrTiO3 [33, 34]. These experiments have raised the exciting possibility
of developing a solar-driven electrolysis system for the production of hydrogen
fuel. The band-gap energy of SrTiO3 or TiO2 is too large for efficient solar-driven
devices and therefore interest has been stimulated to search for another oxide with
a smaller band gap. Methods of reducing the energy for creating electron–hole
pairs in large band-gap materials are also being considered. One such method is
the use of adsorbed sensitizing dye molecules. Surface states in the band-gap region
offer another way for the generation of electron–hole pairs with less-than-band-gap
radiation. Such surface states may also be involved in electrocatalyzing the anode
reaction.
Suggested additional reading material
General survey
J. B. Goodenough, Metallic oxides. Prog. Solid State Chem. 5, 145 (1971).
T. Wolfram, E. A. Kraut, and F. J. Morin, Qualitative discussion of energy bands.

Phys. Rev. B 7, 1677 (1973).
Reference texts
S. Sugano, Y. Tanabe, and H. Kamimura, Multiplets of transition metal ions (New
York, Academic Press, 1970).
References
[1] F. Wohler, Ann. Chem. Phys. 29, 43 (1823).
[2] L. F. Mattheiss, Phys. Rev. 6, 4718 (1972).
[3] E. O. Wollan and W. C. Koehler, Phys. Rev. 100, 545 (1955).
[4] F. L. Battye, H. Höchst, and A. Goldman, Solid State Commun. 19, 269 (1976).
[5] M. Cardona, Phys. Rev. 140, A651 (1965).
[6] P. A. Lightsey, Phys. Rev. B 8, 3586 (1973).
[7] V. E. Henrich, G. Dresselhaus, and H. J. Zeiger, Bull. Am. Phys. Soc. 22, 364
(1977).
[8] R. D. Shannon, Acta Cryst. A 32, 751 (1976).
[9] E. A. Kraut, T. Wolfram, and W. E. Hall, Phys. Rev. B 6, 1499 (1972).
[10] T. Wolfram, R. A. Hurst, and F. J. Morin, Phys. Rev. B 15, 1151, (1977).
[11] T. Wolfram, E. A. Kraut, and F. J. Morin, Phys. Rev. B 7, 1677 (1973).
[12] J. F. Schooley, W. R. Hosler, E. Ambler, J. H. Becker, M. L. Cohen, and O. S.
Koonce, Phys. Rev. Lett. 14, 305 (1965).
[13] J. Hubbard, Proc. Roy. Soc. (London) A 276, 238 (1963).
Problems for Chapter 1 25
[14] J. B. Goodenough and P. M. Roccah, J. Appl. Phys. 36, 1031 (1965).

[15] A. D. Wadsley, Non-stoichiometric compounds, ed. L. Mandalcorn (New York,
Academic Press, 1964).
[16] Ch. J. Raub, A. R. Sweedler, M. A. Jensen, S. Broadston, and B. T. Matthias,
Phys. Rev. Lett. 13, 746 (1964).
[17] A. R. Sweedler, Ch. J. Raub, and B. T. Matthias, Phys. Lett. 15, 108 (1965).
[18] J. P. Remeika, T. H. Geballe, B. T. Matthias, A. B. Copper, G. W. Hull, and
E. M. Kelley, Phys. Lett. 24A, 565 (1967).
[19] H. R. Shanks, Solid State Commun. 15, 753 (1974).
[20] J. G. Bednorz and K. A. Müller, Z. Phys. B 64, 189 (1986).
[21] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957).
[22] See for example, P. W. Anderson, The theory of superconductivity in the high-Tc
cuprates (Princeton, NJ, Princeton University Press, 1997).
[23] A collection of papers on the applications may be found in the IEEE Trans.
Sonics Ultrason. 19 (2) (1972).
[24] L. A. Pedersen and W. E. Libby, Science 176, 1355 (1972).
[25] D. B. Meadowcraft, Nature 226, 847 (1970).
[26] T. C. Lim, E. A. Kraut, and R. B. Thompson, Appl. Phys. Lett. 20, 127 (1972).
[27] Z. J. Kiss, Photochromics. Phys. Today 23, 42 (1970).
[28] R. J. H. Voorhoeve, D. W. Johnson, Jr., J. P. Remeika, and P. K. Gallagher,
Science 195, 827 (1977).
[29] R. B. Woodward and R. Hoffman, The conservation of orbital symmetry (Wein-
heim/Bergstrasse, Verlag-Chemic, 1970).
[30] T. Wolfram and F. J. Morin, Appl. Phys. 8, 125 (1970).
[31] D. A. Shirley, J. Vac. Sci. Technol. 12, 280 (1975).
[32] A. C. C. Tseung and H. L. Beven, J. Electroanal. Chem. Interfacial Elec-
trochem. 45, 429 (1973).
[33] J. G. Mavrodies, D. I. Tchernev, J. A. Kafalas, and D. F. Kolesar, Mater. Res.
Bull. 10, 1023 (1975).
[34] M. S. Wrighton, A. B. Ellis, P. T. Wolczanski, D. L. Morse, H. B. Abrahamson,
and D. S. Ginley, J. Amer. Chem. Soc. 98, 2774 (1976).
[35] R. D. Nasby and R. K. Quinn, Mater. Res. Bull. 11, 985 (1976).
[36] A. Fujishima and K. Honda, Nature 238, 37 (1972).
Problems for Chapter 1
1. The perovskite structure is often depicted as an array of octahedra. Each octahedron

consists of the B ion at the center and six oxygen ions, one at each corner (shared
by adjacent octahedra), along the ±x, ±y, and ±z axes. The A ions are located in
the spaces between the octahedra. Make a sketch of the perovskite structure using an
array of octahedra and indicate the positions of the different ions.
2. Using the ionic model discuss the electronic structure expected for KTaO3 . What are
the electronic configurations of the ions? Would you expect this material to be metallic
or insulating?
3. For the perovskites why are the electronic states derived from the A ion usually less
important than those of the B and O ions?
4. The notation, M:ABO3 , indicates an ABO3 perovskite doped with M ions. Classify
the following materials as n-type or p-type semiconductors: Nb:SrTiO3 , La:BaTiO3 ,
Na:WO3 , KTaO3−δ (oxygen deficient).
5. Using information contained in the chapter what would you expect for the ionic
energy gap in eV between the p- and d-levels in BaTiO3 ? Assume covalency reduces
the effective charges to 80% of their full ionic charges and that the electron “affin-
ity” for adding an electron to the O ion is a repulsive energy of 9 eV. What effect
would you expect ligand-field splitting to have on the energy gap? Explain your answer.
6. The notation “g” and “u” for the levels of the BO6 cluster comes from the German
words “gerade” and “ungerade” meaning even and odd or symmetric and unsymmetric.
For a cluster with cubic symmetry the states must be either “g” or “u”, and “g” and
“u” functions can not be combined. The “g” cluster states are symmetric with respect
to inversion through the center of the B ion. The d orbitals are all symmetric under
inversion. Specify the combinations of neighboring p orbitals that will covalently mix
with the d orbitals to form “g” states.
~
7. Energy band states are electron waves that vary in phase as eik·~r where ~k is
the wavevector for the state. For ~k = 0 the phase of an orbital is the same in
all unit cells. Thus the oxygen orbitals have the same phase on either side of
the B ion. Explain why the BO6 cluster can never have a wavefunction involving
p and d orbitals for which the p orbitals have the same phase on either side of the B ion.
8. The energy bands for the ABO3 structure are illustrated in Fig. 1.6. Discuss why the
parameter 10Dq is shown as the energy difference between the π ∗ and σ ∗ bands at
M or R rather than at Γ. The components of the wavevectors for Γ, M, and R are
(0, 0, 0), (1, 1, 0), and (1, 1, 1), respectively, in units of π/2a.
9. The density of states, ρ(E), is defined to be the number of electronic states in the
energy range between E and E + dE. In Fig. 1.6, the energy bands have flat bands
along various symmetry directions. What happens to ρ(E) at an energy for which one
of the bands is flat?
2
Review of the quantum mechanics
of N - electron systems
This chapter is intended as a brief review of the quantum theory of N - electron sys-
tems. It also serves to introduce the linear combinations of atomic orbitals (LCAO)
method.
2.1 The Hamiltonian
The total N - electron Hamiltonian, HT , for an atomic, molecular, or solid system

is of the form:
( )
X ~2 X ZA ZB e2
2
HT = − ∇ +
2MA A ~ ~
A B<A |RA − RB |
 
X  ~2 X e 2  X X Z e2
A
+ − ∇2i + − , (2.1)
 2m |~ri − ~rj |  ~ A|
|~ri − R
i j<i A i
where R~ A and R~ B are the nuclear positions and ~ri and ~rj are the electron coordi-
nates. The terms in the first set of brackets are the kinetic energies of the nuclei
(having mass MA ) and the Coulomb repulsions among them. The terms in the
second brackets are the kinetic energies of electrons and the electron–electron re-
pulsions. The last term is the electron–nuclear attractions. eZA is the charge of the
nucleus located at R~ A , −e is the charge of an electron and m is the electron mass.
For our purposes we consider the nuclei fixed at their equilibrium positions
in the crystal and seek the solutions of Schrödinger’s equation for the electronic
wavefunction (Born–Oppenheimer approximation [1]). The electronic wavefunctions
satisfy the equation
HΨ(τ ) = EΨ(τ ) (2.2)
X ~2 XX e2 X X ZA e 2
H=− ∇2i + − (2.3)
2m |~ri − ~rj | ~ A|
i i j<i i |~
A ri − R
27
28 Review of the quantum mechanicsof N - electron systems
where Ψ(τ ) is the N -electron wavefunction with τ representing the N electron

spatial and spin coordinates. The electronic energy, E, is determined by
R ∗
Ψ (τ ) H Ψ(τ ) dτ
E= R ∗ . (2.4)
Ψ (τ ) Ψ(τ ) dτ
2.2 The Slater determinant state
The N -electron wavefunction Ψ(τ ),

Ψ(τ ) ≡ Ψ(~r1 , ~r2 , ~r3 , . . . , ~rN ; s1 , s2 , s3 , . . . , sN ), (2.5)
is required to be antisymmetric under the set of operations which interchange the
space and spin coordinates of any pair of electrons.
The Slater determinant wavefunction is an approximate N -electron wavefunc-
tion, constructed from one-electron spin orbitals, that satisfies the antisymmetric
property. Let φi (~r), i = 1, 2, . . . be a complete set of orthonormal one-electron func-
tions and χi (s) the spin functions corresponding to either “spin up” or “spin down”.
The spin orbitals ψi (τ ) are the product states
ψi (τ ) ≡ φi (~r) χi (s) (2.6)
where τ represents the space and spin coordinates of an electron. The orthogonality
condition is
Z
ψi∗ (τ ) ψj (τ ) dτ = δij . (2.7)
The Slater determinant wavefunction for an N -electron system is

¯ ¯
¯ ψ1 (1) ψ2 (1) · · · ψN (1) ¯
¯ ¯
N 1 ¯¯ ψ1 (2) ψ2 (2) · · · ψN (2) ¯¯
∆ν = √ ¯ . ¯ (2.8)
N ! ¯¯ .. ¯
¯
¯ ψ (N ) ψ (N ) · · · ψ (N ) ¯
1 2 N
where ν symbolizes
√ the set of N spin orbitals used in constructing ∆N ν .
The
R factor 1/ N ! ensures that the wavefunction is normalized so that
(∆N ∗ N
ν ) ∆ν dτ1 dτ2 · · · dτN = 1. A different Slater determinant can be constructed
from each different set of N spin orbitals. The wavefunction of an N -electron system
can be approximated by a linear combination of such Slater determinants
X
Ψ= aν ∆N
ν (2.9)
ν
where aν are the constant coefficients specifying the amplitude of different “con-
figurations” comprising the total wavefunction, Ψ. Use of more than one slater de-
2.3 Koopman’s theorem 29
terminant is referred to as configuration interaction. In many applications to solids

and molecules, configuration mixing is omitted and Ψ is approximated by a single
Slater determinant. We shall limit this discussion to a single Slater determinant
wavefunction.
The electronic Hamiltonian, H, includes one-electron operators f (i),
~2 2 X ZA e2
f (i) = − ∇ − , (2.10)
2m i ~ A|
|~ri − R A
and two-electron operators gij ,
e2
gij = . (2.11)
|~ri − ~rj |
The expectation or average value of the energy of a particular N -electron Slater

determinant wavefunction is
Z
hEνN i = (∆N ∗
ν ) H∆ν dτ
N
XZ
= φi (1)∗ f (1) φi (1) d~r1
i
ZZ
1X
+ ψi (1)∗ ψj (2)∗ g12 (1 − P12 ) ψj (2) ψi (1) dτ1 dτ2 (2.12)
2 ij
where the sums are over all N states appearing in the set ν. The operator, P12 is
the exchange operator defined by
P12 ψj (2) ψi (1) = ψj (1) ψi (2). (2.13)
2.3 Koopman’s theorem

N −1
Consider two Slater determinant wavefunctions ∆N ν and ∆ν(k) where the set of
(N – 1) spin orbitals in ν(k) is the same as those of ν except that the state ψk does
not appear.
We want to compare the average energy of these two approximate wavefunc-
tions. The energy difference, according to (2.12), is
N −1
εN N
k ≡ hEν i − hEν(k) i
Z
= φk (1)∗ f (1) φk (1) d~r1
XZ Z
+ ψk (1)∗ ψj (2)∗ g12 (1 − P12 ) ψj (2) ψk (1) dτ1 dτ2 . (2.14)
j
εN
k is the energy difference between the N and (N – 1)-electron states having (N – 1)
common spin orbitals. As we shall see εNk is closely related to, but not equal to, the
binding energy of an electron of the N -electron system.
2.4 Hartree–Fock equations
The Hartree–Fock equations are obtained by application of the variational princi-

ple to a Slater N -electron wavefunction. Given an approximate wavefunction con-
structed from one-electron spin orbitals, the best choices of basis orbitals are those
that minimize the average energy. We therefore seek the minimum value of hEνN i
with variations in the spin-orbital functions. The variations are constrained by the
requirement that the states remain orthonormal:
Z
dτ ψi (τ )∗ ψj (τ ) = δij .
This constraint may be imposed by the method of Lagrange multipliers. Thus

we minimize the function F , where
XX ½Z ¾
N N ∗
F ≡ hEν i − λij dτ ψi (τ ) ψj (τ ) − δij . (2.15)
i j<i
The constants, λN
ij , are the Lagrange multipliers. We require
δF
=0
δφ∗k
and find
½ ·XZ ¸¾ X
∗
f (1) + dτ2 ψj (2) g12 (1 − P12 ) ψj (2) ψi (1) = λN
ij ψj (1) . (2.16)
j j
Equations (2.16) are known as the Hartree–Fock equations and it is convenient

to transform them to a representation in which λN ij is diagonal. The set of numbers
λN
ij form a matrix that can be diagonalized by a unitary transformation:
X
εNi δij =
∗
Cki λNk` C`j . (2.17)
k`
The unitary properties of the transformation are expressed by the relations:

+ ∗
Cik = Cki , (2.18)
X
∗
Cki Ckj = δij . (2.19)
k
2.4 Hartree–Fock equations 31
We define the transformation of the spatial part of the spin orbitals by

X
φi (1) = Cik φ0k (1), (2.20)
k
X X
+ 0
φi (1)∗ = Cik φk (1)∗ = ∗
Cki φ0k (1)∗ . (2.21)
k k
We note that the term

XZ XZ
∗
dτ2 ψj (2) g12 (1 − P12 ) ψj (2) = dτ2 ψj0 (2) g12 (1 − P12 ) ψj0 (2)
j j
(2.22)
where
ψj0 (2) = φ0j (r2 ) χj (s2 ) .
Substituting (2.20) and (2.21) into (2.16) gives
X ½ ·XZ ¸¾
Cik f (1) + dτ2 ψj (2) g12 (1 − P12 ) ψj (2) ψk0 (1)
0 ∗ 0
k j
X
= Cjk λN 0
ij ψk (1). (2.23)
kj
∗
Multiplying (2.23) by Cì and summing over all i gives
½ ·XZ ¸¾
f (1) + dτ2 ψj (2) g12 (1 − P12 ) ψj (2) ψ`0 (1) = εN
0 ∗ 0 0
` ψ` (1). (2.24)
j
Equation (2.24) is the diagonal form of the Hartree–Fock equations. If we

multiply (2.24) by ψ`0 (1)∗ and integrate over τ1 then we obtain
Z
dτ1 ψ`0 (1)∗ f (1) ψ`0 (1)
XZ Z
+ dτ1 dτ2 ψ`0 (1)∗ ψj0 (2)∗ g12 (1 − P12 ) ψj0 (2) ψ`0 (1) = εN
` . (2.25)
j
Comparison of (2.25) with (2.14) shows that

(N −1)
εN N
` = hEν i − hEν(`) i. (2.26)
Thus the eigenvalues of the Hartree–Fock equations are, by Koopman’s theorem,
equal to the difference between the energies of the N and N –1 electron systems
having (N –1) common spin orbitals. There are several reasons why εN ` is not the
true binding energy of an electron in the state ψ`0 . First, we note that the Hartree–
Fock equations are approximate. However, even if the Hartree–Fock equations were
exact, εN
` would still not be the correct binding energy because the set of spin
orbitals {ν(k)} are not those which minimize the energy of the (N –1)-electron
system. If we solved the Hartree–Fock equations for the N and (N –1)-electron
systems separately we would find that the spin orbitals which result are different.
For small systems such as atoms and molecules this difference can be large. For
larger systems such as solids, εN
` is a better approximation to the binding energy.
We shall return to the question of the relation of εN` to the binding energy in a
later chapter where we discuss the interpretation of photoemission experiments.
2.5 Hartree–Fock potential
The Hartree–Fock equations, (2.24), may be written in the form

· 2 ¸ Z
−~ 2 (i)
∇i + VN (~r) + VC (~r) φi (~r) − Vex (~r, ~r 0 ) φi (~r 0 ) d~r 0 = εN
i φi (~
r)
2m
(2.27)
where VN (~r) is the nuclear attraction potential, VC (~r) is the direct Coulomb poten-
(i)
tial and Vex (~r, ~r 0 ) is the non-local exchange potential. These are defined by
X ZA e2
VN (~r) = − , (2.28)
~ A|
|~r − R
~A
R
X Z
φj (~r 0 )∗ φj (~r 0 )
VC (~r) = e2 d~r 0 , (2.29)
j
|~r − ~r 0 |
and
X φj (~r 0 )∗ φj (~r)
(i)
Vex (~r, ~r 0 ) = e2 χ(sj ) χ(si ) . (2.30)
j
|~r − ~r 0 |
The charge density at ~r 0 due to an electron in the orbital φj is given by

e φj (~r 0 )∗ φj (~r 0 ). The total charge density, ρ(~r 0 ), due to all occupied states is the
sum of the individual charge densities. The Coulomb potential may therefore be
written as
Z
ρ(~r 0 )
VC (~r) = e d~r 0 (2.31)
|~r − ~r 0 |
which is recognized as the potential arising from a continuous charge distribution.
The Coulomb potential entering (2.27) (defined in (2.29)) includes the interaction
of an electron with itself. This self-interaction, however, is exactly canceled by one
of the terms in the exchange potential.
(i)
The exchange potential, Vex (~r, ~r 0 ), is seen to be a non-local potential acting
between electrons with parallel spins. The dependence on the superscript “i” is
2.5 Hartree–Fock potential 33
only that the states contributing to the exchange must have the same spin state
as χi . The potential is therefore the same for all states having the same spin and
is not dependent on the spatial orbital. The exchange potential may be written in
the form of a local potential,
Z
(i)
Vex (~r, ~r 0 ) φi (~r 0 ) d~r 0 = vex
(i)
(~r) φi (~r) (2.32)
where
XZ ½ ¾
(i)
φ∗j (~r 0 ) φj (~r) φi (~r 0 )
vex (~r) ≡ d~r 0 (χ(sj ), χ(si )) . (2.33)
j
|~r − ~r 0 | φi (~r)
(i)
The local potential, vex (~r), depends explicitly on the spatial orbital φi (~r).
For parallel spins the exchange reduces the Coulomb repulsion between two
electrons. This comes about because the antisymmetric wavefunction must vanish
whenever two electrons with parallel spins are at the same point. This may be seen
by noting that when the two coordinates of two electrons ~rn and ~rm are equal then
two of the rows of the Slater determinant are equal and the determinant vanishes.
This effect is expressed by the Pauli exclusion principle which prevents parallel spin
electrons from occupying the same point in space. The exchange potential is a form
of electron–electron correlation. The probability that one electron is at ~rn and a
second electron is at ~rm is
Z
Γ (τn , τm ) = dτ 0 ∆N ∗ N
ν ∆ν (2.34)
where the integration over dτ 0 is over all τ except τn and τm . For parallel spin
electrons the probability is found to be
1 XX© ª
Γp (~rn , ~rm ) = |φi (~rn )|2 |φj (~rm )|2 − φ∗i (~rn ) φ∗j (~rm ) φi (~rm ) φj (~rn ) . (2.35)
N! i
j6=i
For antiparallel spin electrons

1 XX
Γa (~rn , ~rm ) = |φi (~rn )|2 |φj (~rm )|2 . (2.36)
N! i
j6=i
It is seen that the probability of antiparallel spin electrons at ~rn and ~rm is equal
to the product of the individual probabilities. Thus there is no correlation between
antiparallel electrons. However, the parallel spin electron probability has an inter-
ference due to exchange-correlation. When ~rm = ~rn , the probability of parallel spin
electrons vanishes. If we fix one electron at ~rn then the probability of finding an-
other (parallel spin) electron near ~rn is small. The depletion of the probability due
to the second term in (2.35) is called the “exchange hole”. As an electron moves
through space it is always surrounded by the exchange hole which is the result of
the correlated motion of same spin electrons as they avoid occupying the same point
in space.
Returning to the Hartree–Fock equations, (2.27), it is clear that the potential
is not known a priori. To construct the potential one must have the orbitals, but to
obtain the orbitals one must have the potential. Some type of self-consistent proce-
dure is required in order to obtain the solutions of the Hartree–Fock equations. In
practice the equations are solved iteratively. A starting potential V 0 is assumed and
the orbitals are determined by solution of the eigenvalue equation. These orbitals
are then used to construct a new potential, V 1 . The process is iterated until V n is
(sufficiently close) equal to V n+1 . It is assumed that such self-consistent solutions
are unique.
2.6 Approximate exchange potential
The Hartree–Fock equations are difficult to solve self-consistently because of the

complex, non-local exchange potential. As a result an approximate form of the ex-
change potential is desirable. One approximation, known as the “Xα” approxima-
tion, has proved to be particularly convenient and it has been employed extensively
in solid-state calculations. The essential idea comes from the Fermi–Thomas [1–3]
model where the exchange energy is found to be proportional to the cube root of
the charge density. Slater [4] suggested that a semiempirical potential, VXα , be used
in place of the Hartree–Fock exchange potential, where
µ ¶1/3
9 3ρ(~r)
VXα (~r) = α . (2.37)
2 8π
Here the quantity α is a scaling parameter which is generally in the range 23 ≤ α ≤ 1.
The X in Xα is shorthand for exchange. The Xα approximation replaces the non-
local Hartree–Fock exchange potential by a local potential that is proportional to
the cube root of the charge density. The form in (2.37) is used for non-magnetic
systems. For magnetic systems spin dependent forms are used which are defined
so that their sum is equal to (2.37) when the number of spin-up and spin-down
electrons is equal. The form used is
µ ¶1/3
(s) 9 3ρs (~r)
VXα (~r) = α (2.38)
2 4π
where ρs (~r) is the charge density for s = ↑ or s = ↓. When self-consistent solu-
tions are determined, the degree of spin polarization must also be determined self-
consistently. The optimum scaling parameter, α, appearing in (2.38) is usually not
the same as that in (2.37).
There are two points worth noting with regard to the use of the Xα approx-
2.7 The LCAO method 35
imation. The first is that the exact cancelation mentioned in the preceding sec-
tion between the Coulomb and exchange self-interaction terms is lost. The second
point is that Koopman’s theorem no longer applies to the eigenvalues. That is, the
eigenvalues εN` of the Hartree–Fock equations with Vex replaced by VXα do not
correspond to the energy difference between the Slater determinant states ∆N ν and
(N −1)
∆ν(`) . On the other hand, the Hartree–Fock eigenvalues do not correspond to the
true binding energy of an electron because of the neglect of the relaxation of the
orbitals of the (N –1) electron system. In many cases the eigenvalues found for the
Xα approximation compare reasonably well with experimental ionization energies.
Orbital relaxation effects are sometimes included by means of what is called
the “transition state” approximation. With this method, Xα solutions are obtained
with one half of an electron assigned to the orbital whose ionization energy is sought.
The ionization energy is therefore approximated by the eigenvalues of a system with
(N – 12 ) electrons.
The difference between an N -electron (ground state) eigenvalue εN ` and the
ionization energy I` is often found to be nearly independent of `. In such a case,
ground state energy differences, εN N
` − εk , compare well with ionization energy dif-
ferences, I` − Ik .
2.7 The LCAO method
The Hartree–Fock equations with or without the Xα approximation are the basis
for many current electronic structure calculations. Many different methods are em-
ployed in the solution of Hartree–Fock equations. Each method has its advantages
and disadvantages. The LCAO method for finding the solutions is particularly valu-
able because it provides a very simple and intuitive interpretation of the electronic
structure. With the LCAO method the spatial parts of the spin orbitals comprising
the Slater determinant are expressed as linear combinations of atomic orbitals. The
Hartree–Fock equations are then transformed to matrix equations which determine
the amplitudes of the atomic orbitals that make up an eigenstate.
The orbitals, φi (~r), which form the basis of the Slater determinant, ∆ν , are
written in the form
XX
φi (~r) = (i)
Cnα ~ n)
ϕα (~r − R (2.39)
~n
R α
where ϕα (~r − R~ n ) is an atomic orbital for an atom located at R ~ n . The subscript

α labels the different atomic states; α = 1s, 2s, 2px , 2py , 2pz , . . . . The coefficients,
(i)
Cnα , specify the amplitudes of the atomic orbitals for the state whose eigenvalue is
εN
i ≡ εi . The Hartree–Fock equations with the local exchange potential VXα are of
the form
H φi (~r) = εi φi (~r) . (2.40)
Using (2.39) we have

X ½ ¾
(i) −~2 2 ~ n)
Cnα ∇ + VN (~r) + VC (~r) − VXα (~r) ϕα (~r − R
nα
2m
X
= εi (i)
Cnα ~ n) .
ϕα (~r − R
nα
~ m ) and integrating over ~r leads to the result

Multiplying (2.40) by ϕ∗β (~r − R
Xn o
(i)
Hmβ,nα − εi Smβ,nα Cnα = 0, (2.41)
nα
where the matrix elements are

Z · 2 ¸
~ m ) −~ ∇2 + VN (~r) + VC (~r)−VXα (~r) ϕα (~r − R
Hmβ,nα = d~r ϕ∗β (~r − R ~ n)
2m
(2.42)
and
Z
Smβ,nα = ~ m )∗ ϕα (~r − R
d~r ϕβ (~r − R ~ n) . (2.43)
The atomic orbitals belonging to a given atom are orthonormal, but orbitals on
different atoms have overlap that is specified by Smβ,nα . If we denote the matrix
of the Hamiltonian whose elements are Hmβ,nα by H and the overlap matrix by S
then (2.41) is
~ (i) = 0
(H − εi S) C (2.44)
(i)
~ (i) are the Cnα . Because of the overlap matrix,
where the components of the vector C
S, the eigenvectors of (2.44) are not orthogonal. Instead they satisfy the condition
XX (j)
(C~ (i) , S C
~ (j) ) = (i)
Cnα Snα,mβ Cmβ = δij . (2.45)
nα mβ
The diagonal elements of H are of the order of the ionization energy of the cor-
responding atomic state. The off-diagonal matrix elements are called transfer or
resonance integrals or simply LCAO integrals.
In the use of the LCAO-Xα approach, the atomic orbitals are usually taken
from prior calculations which are available for most atoms. The “Herman–Skillman”
orbitals [5] are frequently used. To find self-consistent solutions some type of iter-
ative procedure has to be employed. Very often, the charge density of the non-
interacting atoms is used to generate initial potentials for VC and VXα . Each itera-
2.8 Orthogonalized atomic orbitals 37
(i)
tion produces a set of Cnα ’s which may be used to calculate a new charge density
for the interacting atoms.
The charge density is given by
X XXX
ρ(~r) = Ψ~∗k Ψ~k = ∗
Ciα ~ i ) ϕβ (~r − R
Cjβ ϕ∗α (~r − R ~ j)
~
k ~
k iα jβ
Z X
d~r ρ(~r) = ~ (k) , S C
(C ~ (k) ) = N (2.46)
~
k(occ)
where (occ) means that the sum is over the occupied eigenstates. For an N -electron
system the N lowest-energy eigenstates are the occupied states.
The LCAO-Xα self-consistent method is as rigorous as any other method em-
ploying the Xα approximation. The advantages of the method are twofold. First, the
wavefunctions are very easy to conceptualize and the interpretation of the results
in terms of elementary chemical concepts is immediate. A second advantage is the
absence of the necessity to employ artificial boundary conditions such as those em-
ployed in most other methods. For example, the “multiple scattering” [6] or linear
combinations of “muffin-tin” orbitals (LCMTO) [7] methods currently employed
use artificial spherical boundaries about the atoms and surrounding the molecule
itself. These methods are not well suited for systems such as planar molecules.
The disadvantage of the LCAO-Xα method seems to be principally the amount
of computational time required to obtain accurate numerical solutions. On the other
hand, and of primary importance in our discussions, is the fact that the LCAO-Xα
method is ideally suited as a basis for the development of simpler, empirical models.
The LCAO method provides a rigorous solution to the self-consistent problem
only if the atomic orbital basis set includes all of the atomic states. In practice, the
set of atomic states employed is finite and restricted to only a few atomic states
beyond those occupied for a free atom. This introduces an error which is difficult
to assess. All methods which express the orbitals φi (~r) in terms of a finite set of
basis states suffer from this type of “truncation error”.
2.8 Orthogonalized atomic orbitals
It is often convenient to work with localized orbitals that are orthogonal in order
to eliminate the overlap between orbitals localized on different atomic sites. This is
accomplished by the transformation:
~ (k) = S1/2 C
D ~ (k) , (2.47)
H 0 = S−1/2 H S−1/2 , (2.48)

which leads to the new matrix equation
~ (k) = 0 .
[ H 0 − εk I ] D (2.49)
It is always possible to make this transformation because the eigenvalues of S

are real, positive numbers and therefore S1/2 and S−1/2 can always be constructed.
~ (k) ,
Equation (2.49) is in the standard eigenvalue form and the eigenvectors, D
form an orthonormal set
~ (k) , D
(D ~ (`) ) = δk` . (2.50)
The new localized orbitals corresponding to the transformation (2.47) are called
Löwdin orbitals [8]. They are related to the atomic orbitals by the relation
X
~ k) =
ξα (~r − R ~ j) .
(S−1/2 )jβ,kα ϕβ (~r − R (2.51)
jβ
~ k ), is localized near R
The Löwdin orbital, ξα (~r − R ~ k but it is somewhat more
extended than the atomic orbital ϕα (~r − R ~ k ).
An important property of the transformation (2.51) is that it preserves the

symmetry properties. That is, ξα (~r − R~ k ) has symmetry transformation properties
~ k ). Proof of this property will be given in
that are identical to those of ϕα (~r − R
Chapter 3.
Suggested reference texts
C. J. Ballhausen and H. B. Gray, Molecular orbital theory (New York, W. A.

Benjamin Inc., 1965).
J. C. Slater, The self-consistent field for molecules and solids, Vol. 4, Quantum
theory of molecules and solids (New York, McGraw-Hill, 1974).
G. A. Segal, ed., Semiempirical methods of electronic structure, Part A: Methods

(New York, Plenum Press, 1977).
References
[1] L. H. Thomas, Proc. Camb. Phil. Soc. 23, 542 (1927).
[2] E. Fermi, Z. Physik 48, 73 (1928).
[3] P. A. M. Dirac, Proc. Camb. Phil. Soc. 26, 376 (1930).
[4] J. C. Slater, Phys. Rev. 81, 385 (1951).
[5] F. Herman and S. Skillman, Atomic structure calculations (Englewood Cliffs,
NJ, Prentice-Hall, 1963).
[6] K. H. Johnson, J. Chem. Phys. 45, 3085 (1966), Int. J. Quantum Chem. 51,
361 (1967); ibid 52, 223 (1968).
[7] O. K. Andersen, Phys. Rev. B 12, 3060 (1975).
[8] P.-O. Löwdin, J. Chem. Phys. 18, 365 (1950).
1. Show that the Slater determinant for an N -electron system vanishes if ψi = ψj for any
i 6= j. Explain how this result is related to the Pauli exclusion principle.
2. In x-ray photoelectron spectroscopy a photon impinges on a solid surface. An electron

in an initial state of energy Ei absorbs the photon and is ejected from the solid into
a high-energy electronic state Ef . Discuss the connection between the Hartree–Fock
energies and Koopman’s theorem in this photoexcitation process? Under what
conditions is the Born–Oppenheimer approximation valid?
3. Explain the following terms:

(a) overlap integral
(b) transfer or resonance integral
(c) Xα approximation
(d) exchange hole.
4. Consider two Slater determinant wavefunctions, ∆N N

a and ∆b , with different configu-
N
rations. Assume the orthogonal basis orbitals for ∆a are (Ψa , Ψ2 , · · · , ΨN ) and those
for ∆N N N
b are (Ψb , Ψ2 , · · · , ΨN ), with Ψb 6= Ψa . Show that ∆a is orthogonal ∆b .
5. Empirical LCAO models use adjustable parameters for the diagonal and two-center
interaction integrals between neighboring atoms lying within some cutoff radius R0 .
Interactions beyond R0 are assumed to be negligible. In addition, overlap integrals are
often ignored. Give reasons why both of these assumptions may be approximately valid.
6. Consider two empirical models, model I and model II. Assume I and II use orbitals
with the same symmetry properties. Model I assumes the overlap integrals between
orbitals on different sites vanish. Model II uses the overlap integrals between different
sites as adjustable parameters. As a result, model II has many more adjustable
parameters than model I. Which model is capable of the most accurate representation
of the electronic states?
3
Empirical LCAO model
The LCAO method described in the previous chapter forms the basis for a num-
ber of empirical or qualitative models. In such models the LCAO matrix elements
are treated as “fitting” parameters to be determined from experiment or in some
empirical way. Such models have provided a great deal of physical insight into the
electronic properties of molecules and solids.
One of the first and simplest LCAO models was used by Hückel [1] to discuss
the general qualitative features of conjugated molecules. Later, Slater and Koster
[2] introduced an LCAO method for the analysis of the energy bands of solids. The
Slater–Koster LCAO model has been used extensively as an interpolation scheme.
The LCAO parameters are determined by choosing the model parameters to
give results that approximate those of more accurate numerical energy band calcu-
lations at a few points in the Brillouin zone. Once the parameters are determined
the LCAO model gives approximate energies at any point in the Brillouin zone.
LCAO models have been remarkably useful for ordered solids and molecules
having a high degree of symmetry. The reason for this is that in many cases the
electronic structure is qualitatively determined by symmetry or group theoretical
considerations. The group theoretical properties of a system are preserved in LCAO
models and therefore they are able to correctly represent the general features of the
electronic states.
3.1 LCAO matrix elements
The LCAO matrix elements (see (2.42)) were derived in Chapter 2. They are of the
form:
Z
Hkα,jβ ≡ ϕα (~r − R ~ k )∗ H(~r) ϕβ (~r − R
~ j ) d~r , (3.1)
~2 2
H(~r) = − ∇ + V T (~r) , (3.2)
2m
41
42 Empirical LCAO model
where ϕα may be taken as an atomic orbital or a Löwdin orbital and where V T (~r)
consists of the nuclear attraction, Coulomb, and exchange potentials. V T may be
expressed in terms of a sum of potentials localized at each atomic site,
X
V T (~r) = ~ .
v(~r − R) (3.3)
~
R
Using (3.2), the LCAO matrix element may be decomposed into a kinetic
energy matrix element and a potential energy matrix element:
T
Hkα,jβ = Tkα,jβ + Vkα,jβ , (3.4)
Z µ ¶
~ ∗ ~2 2 ~ j ) d~r ,
Tkα,jβ = ϕα (~r − Rk ) − ∇ ϕβ (~r − R (3.5)
2m
Z
T
Vkα,jβ = ϕα (~r − R~ k )∗ V T (r) ϕβ (~r − R
~ j ) d~r . (3.6)
T
Using (3.3) Vkα,jβ takes the form
XZ
T
Vkα,jβ = ~ k )∗ v(~r − R)
ϕα (~r − R ~ ϕβ (~r − R
~ j ) d~r . (3.7)
R
The kinetic energy matrix elements in (3.5) have two types of integrals. The matrix
elements for R ~k = R~ j are “one-center” integrals, while for R ~ k 6= R~ j they are “two-
center” integrals. The potential energy matrix elements have three possible types of
integrals; one-center integrals for R ~k = R ~ =R~ j , two-center integrals when two of the
position vectors are the same, and three-center integrals for R ~ k 6= R
~ 6= R
~ j . Since
the amplitudes of atomic orbitals decrease exponentially with distance from the
nucleus, it is clear that the integrals involved in the matrix elements will decrease
rapidly with increasing |R ~k − R~ j | and may therefore be neglected beyond some
cutoff distance. In a similar fashion, the localized potential v(~r − R) ~ will decrease
~
with distance away from R. Therefore the integrals also decrease rapidly as either
~ −R
|R ~ k | or |R
~ −R~ j | increases. In general then, one expects the one-center integrals
to be the largest, followed by the two-center integrals and with the three-center
integrals being the smallest. However, there are many three-center contributions
and for accurate calculations they must be retained.
3.2 Slater–Koster model
In order to reduce the number of integrals involved in calculating the LCAO matrix
elements certain approximations must be employed. One approximation already
mentioned is to neglect matrix elements for which |R ~ k − R|,
~ |R~ j − R|,
~ or |R
~k − R
~ j|
exceed some chosen distance.
3.2 Slater–Koster model 1 43
~j
R xj
θj
k
R~
yj
j −
R~
~r ′
φj
k
R~
j −
R~
O ~k
R θk
xk ~r
yk
φk =φj
Figure 3.1. Coordinate system with the z-axis along the internuclear axis of atoms lo-
~ j and R
cated at R ~ k showing that φk = φj .
Slater and Koster suggested that the three-center contributions be neglected

altogether for the purpose of establishing a workable empirical model. As a further
simplification they assumed the local potentials to be spherically symmetric about
the atomic center so that
~ = v(|~r − R|)
v(~r − R) ~ . (3.8)
This approximation is not accurate because the charge density around a given
ion in a molecule or solid is seldom spherical. Nevertheless it is a reasonable start-
ing point. Use of this approximation does not mean that the final, resulting charge
density will be spherically symmetric about the atomic sites. Instead, the charge
density calculated from the resulting wavefunctions will reflect the bonding symme-
try between the neighboring atomic orbitals. Therefore, this approximation can be
considered as the first step of a self-consistent procedure. With these assumptions
we may express the matrix elements of H with R ~ k 6= R
~ j as
X Z
Hkα,jβ = Tkα,jβ + ~ k )∗ v(~r − R)ϕ
ϕα (~r − R ~ β (~r − R~ j ) d~r
R
Z · 2 ¸
~ ∗ −~ 2 ~ ~ ~ j ) d~r . (3.9)
' ϕα (~r − Rk ) ∇ +v(|~r − Rk |)+v(|~r − Rj |) ϕβ (~r − R
2m
The atomic orbitals will be of the form

~ k ) = Rn (r) P mα (cos θk ) eimα φk
ϕα (~r − R (3.10)
α `α
where nα , `α , and mα are the principal, orbital, and magnetic quantum numbers,
respectively, Rnα (r) is the radial part of the wavefunction, and the angular part,
P`m
α
α
(cos θk ), is an associated Legendre polynomial. If the z-axis is chosen along
the line joining two atoms, R ~j − R
~ k , as shown in Fig. 3.1, then it is clear that the
φk is equal to φj . Therefore, the integral of (3.9) contains a part,
Z 2π
e−i(mα −mβ )φ dφ = δmα mβ .
0
This means that the only non-vanishing two-center integrals are those for which the
orbitals have the same symmetry about the internuclear axis. That is, the matrix
elements are non-zero only if there is a non-vanishing overlap between the two
orbitals involved. Some pictorial examples are shown in Fig. 3.2 to illustrate this
principle. In practice it is a trivial matter to sketch the angular parts of the orbitals
and to determine whether the matrix element vanishes by symmetry.
A useful nomenclature for describing the nature of the overlap of two atomic
orbitals has been developed. Non-zero overlap exists when mα = mβ = m. For these
cases the overlap is called σ, π, and δ for m = 0, l, or 2, respectively. Examples
of these are illustrated in Fig. 3.3. The atomic orbitals generally employed are
chosen to be real by taking linear combinations of atomic states. Thus px ∝ cos φ
and py ∝ sin φ are employed instead of functions involving eimφ . In all cases, the
functions combined have the same |m|. Thus, when we use these real atomic orbitals,
the rule for deciding whether the overlap vanishes becomes |mα | = |mβ |, where |mα |
is the magnitude of the magnetic quantum number associated with the functions
that make up the real atomic orbital ϕα .
The overlap integrals are denoted by S(βαt). For example, S(pdπ) represents
the overlap between a p orbital with a d orbital where each has |m| = 1. The inter-
action matrix elements themselves are represented by (βαt) alone. Both overlap and
interaction matrix elements are defined by convention for the specific configurations
y y
s s
z z
pz
py
x x
(a) (b)
Figure 3.2. (a) Overlap between an s orbital and a p orbital. The overlap vanishes by
symmetry ms = 0, mpy = 1. (b) Non-zero σ overlap between an s orbital and a pz orbital:
ms = mpz = 0.
ϕα ϕβ
pz s d3z2 −r2 s
s s
z z z
(ssσ) (spσ) (sdσ)
pz pz d3z2 −r2 pz
z z
(ppσ) (pdσ)
y y
py py dyz py z
dyz dyz
z z y
(pdπ) (ddπ)
(ppπ)
dxy dxy dx2 −y2 dx2 −y2
z z
x
x
y (ddδ) y (ddδ)
and derived from the above definitions by symmetry:
y y
dx2 −y2 dx2 −y2
s px
x x
√ √
3 3
2 (sdσ) 2 (pdσ)
Figure 3.3. Fundamental types of overlaps (or matrix elements).

as shown in Fig. 3.3. Other configurations are related to these basic interactions.
For example, a (pdπ) overlap changes sign if the positions of the orbitals are in-
terchanged as shown in Fig. 3.4. The relative signs of particular interactions are
usually obvious. The relation, between the definitions in Fig. 3.3 and the various
basic interactions is
Z
~ i )∗ H ϕβ (~r − R
(βαt) ≡ ϕα (~r − R ~ j ) d~r (3.11)
Z
~ i )∗ ϕβ (~r − R
S(βαt) ≡ ϕα (~r − R ~ j ) d~r (3.12)
(t = σ, π, or δ)
where R ~j – R
~ i is a vector with components along the positive z-axis. The matrix
element for dx2 −y2 with s or p, shown in Fig. 3.3, have been deduced from the
basic matrix elements by rotating the coordinate system and then re-expressing
the transformed orbitals in terms of those which have basic definitions. To do this
one must pay attention to the fact that the normalization of the angular part of the
wavefunctions for different orbitals is sometimes different. Table 3.1 lists the forms
of the orbitals. For compactness we shall often use the notation dx2 for dx2 −y2 and
dz2 for d3z2 −r2 .
(pdπ) – (pdπ)
Figure 3.4. Definition of (pdπ) overlap and interaction integral.
Consider the overlap of the dx2 −y2 with the s orbital as shown in Fig. 3.3. We
can relabel the x-axis as z 0 and y = y 0 . Then the relevant angular part becomes
r µ ¶
15 x2 − y 2
dx2 =
16π r2
r µ ¶
15 z 02 − y 02
→
16π r02
r · µ ¶ µ ¶¸
15 1 3z 02 − r02 1 x02 − y 02
= +
16π 2 r02 2 r02
√
3 1
= dz02 + dx02 .
2 2
The s orbital has no angular variation and so it remains unchanged by the coordi-
nate transformation. Then

Z √ Z Z √
∗ 3 ∗ 1 ∗ 3
dx2 s d~r = dz02 s d~r + dx02 s d~r = (sdσ) . (3.13)
2 2 2
The second integral on the right-hand side of (3.13) vanishes by symmetry.
Table 3.1. Forms of atomic orbitals.
Orbital q# Angular function Radial function

q
1 (n)
ns 4π Rs (r)
q ¡x¢
3 (n)
npx 4π r rRp (r)
q ¡y¢
3 (n)
npy 4π r rRp (r)
q ¡z¢
3 (n)
npz 4π r rRp (r)
q ³ ´ 
5 3z 2 −r 2 (n)
ndz2 16π r2 r2 Rd (r) 
q ³ ´ eg
x2 −y 2
r2 Rd (r) 
15 (n)
ndx2 16π r2
q ¡ xy ¢ 
15 (n)
ndxy 4π r2 r2 Rd (r) 



q ¡ yz ¢ 
15 2 (n)
ndyz 4π r2 r Rd (r) t2g


q ¡ zx ¢ 

15 (n) 
ndzx 4π r2 r2 Rd (r)
In more general cases we shall have LCAO integrals between orbitals displaced
from one another in an arbitrary direction. It becomes tedious to perform the trans-
formations of the orbitals. To ease this discomfort, Slater and Koster worked out a
table which gives matrix elements for displaced orbitals in terms of the fundamen-
tal integrals. Their results are given in Table 3.2. To use the table to evaluate the
interaction integral of the form:
Z
Eβα = ϕα (~r − R ~ i )∗ H ϕβ (~r − R
~ j ) d~r , (3.14)
~ ji = R
one need only calculate the direction cosines `, m, and n of the vector R ~j – R
~ i;
xj − xi yj − yi zj − zi
`= , m= , n= . (3.15)
~ ji |
|R ~ ji |
|R ~ ji |
|R
~ j ) is a px orbital with R
For example, if ϕβ (~r − R ~ j = a(1, −1, 1) and ϕα (~r − R
~ i ) is
~ ~
a dz2 orbital at Ri = a(−1, 1, 2) then |Rji | = 3a, ` = + 3 , m = − 3 , and n = − 13 .
2 2
In Table 3.2 we find the line labeled Ex,3z2 −r2 and the integral is
Z · ¸
1 2 √
dz2 H px d~r = ` n − (` + m ) (pdσ) − 3`n2 (pdπ)
2 2
2
· ¸
2 1 1³4 4´ √ ³2 1´
= − + (pdσ) − 3 · (pdπ)
3 9 2 9 9 3 9
· ¸
2 (pdπ)
= − (pdσ) + √ . (3.16)
9 3
3.3 Symmetry properties of the Löwdin orbitals
When using the LCAO method as an empirical model it is often convenient to use
the orthogonalized Löwdin orbitals [3] discussed in Section 2.8. With the Löwdin
orbital basis the overlap matrix elements between orbitals on different sites vanish
and the interaction matrix elements are between Löwdin orbitals rather than atomic
orbitals. The formalism described in the preceding section can then be used without
change. This is allowed only because the symmetry properties of the Löwdin orbitals
are identical to those of the corresponding atomic orbitals. In the remainder of this
section we give a proof of this.
We want to show that the Löwdin orbital ξα (~r − R ~ ` ) has precisely the same
~ ` ). They are related by
symmetry properties as the atomic orbital, ϕα (~r − R
X
~ `) =
ξα (~r − R (S−1/2 )mν,`α ϕν (~r − R ~ m) (3.17)
mν
where S is the overlap matrix

Z
Smν,`α = ~ m )∗ ϕα (~r − R
ϕν (~r − R ~ ` ) d~r . (3.18)
The symmetry properties of ϕ are completely determined by their transformation

properties under a symmetry operation of the group corresponding to the system
in question. If O is a symmetry operator of the group then from group theory we
know that
X
~ `) =
O ϕα (~r − R ~ k)
Γ (O)kβ,`α ϕβ (~r − R (3.19)
kβ
where Γ (O) is a unitary representation matrix for O; Γ (O)−1 ∗

`α,kβ = Γ (O)kβ,`α . For
the Löwdin orbitals to have precisely the same symmetry properties as the atomic
~ i and R
Table 3.2. LCAO two-center integrals [2] for orbitals centered at R ~ j . The
~ ~
variables `, m, and n are the direction cosines of (Rj − Ri ).
Es,s (ssσ)
Es,x `(spσ)
Ex,x `2 (ppσ) + (1 − `2 )(ppπ)

Ex,y `m(ppσ) − `m(ppπ)
Ex,z `n(ppσ) − `n(ppπ)
√
Es,xy 3`m(sdσ)
√ 2
Es,x2 −y2 1
2 3(` − m2 )(sdσ)
Es,3z2 −r2 [n2 − 1
2 (`
2
+ m2 )](sdσ)
√ 2
Ex,xy 3` m(pdσ) + m(1 − 2`2 )(pdπ)
√
Ex,yz 3`mn(pdσ) − 2`mn(pdπ)
√ 2
Ex,zx 3` n(pdσ) + n(1 − 2`2 )(pdπ)
√
Ex,x2 −y2 1
2 3`(`2 − m2 )(pdσ) + `(1 − `2 + m2 )(pdπ)
√
Ey,x2 −y2 1
2 3m(`2 − m2 )(pdσ) − m(1 + `2 − m2 )(pdπ)
√
Ez,x2 −y2 1
2 3n(`2 − m2 )(pdσ) − n(`2 − m2 )(pdπ)
√
Ex,3z2 −r2 `[n2 −1 2 2
2 (` + m )](pdσ) − 3`n2 (pdπ)
√
Ey,3z2 −r2 m[n − 2 (` + m )](pdσ) − 3mn2 (pdπ)
2 1 2 2
√
Ez,3z2 −r2 n[n2 − 21 (`2 + m2 )](pdσ) + 3n(`2 + m2 )(pdπ)
Exy,xy 3`2 m2 (ddσ) + (`2 + m2 − 4`2 m2 )(ddπ) + (n2 + `2 m2 )(ddδ)

Exy,yz 3`m2 n(ddσ) + `n(1 − 4m2 )(ddπ) + `n(m2 − 1)(ddδ)
Exy,xz 3`2 mn(ddσ) + mn(1 − 4`2 )(ddπ) + mn(`2 − 1)(ddδ)
3 2 2 2 2 1 2 2
Exy,x2 −y2 2 `m(` − m )(ddσ) + 2`m(m − ` )(ddπ) + 2 `m(` − m )(ddδ)
3 2 2 2 2 1 2 2
Eyz,x2 −y2 2 mn(` − m )(ddσ) − mn[1 + 2(` − m )](ddπ) + mn[1 + 2 (` − m )](ddδ)
3 2 2 2 2 1 2 2
Ezx,x2 −y2 2 n`(` − m )(ddσ) + n`[1 − 2(` − m )](ddπ) − n`[1 − 2 (` − m )](ddδ)
√ √ √
Exy,3z2 −r2 3`m[n2 − 12 (`2 + m2 )](ddσ) − 2 3`mn2 (ddπ) + 12 3`m(1 + n2 )(ddδ)
√ √
Eyz,3z2 −r2 3mn[n2 − 12 (`2 + m2 )(ddσ) + 3mn(`2 + m2 − n2 )(ddπ)
√
− 12 3mn(`2 + m2 )(ddδ)
√ √
Ezx,3z2 −r2 3`n[n − 12 (`2 + m2 )(ddσ) − 2 3`n(`2 + m2 − n2 )(ddπ)
2
1
√ 2 2
+ 2 3`n(` + m )(ddδ)
3 2 2 2 2 2 2 2 2 2 1 2 2 2
Ex2 −y2 ,x2 −y2 4 (` − m ) (ddσ) + [` + m − (` − m ) ](ddπ) + [n + 4 (` − m ) ](ddδ)
1
√ 2 2 2 1 2 2
√ 2 2 2
Ex2 −y2 ,3z2 −r2 3(` − m )[n − 2 (` + m )](ddσ) + 3n (m − ` )(ddπ)
2 √
+ 14 3(1 + n2 )(`2 − m2 )(ddδ)
E3z2 −r2 ,3z2 −r2 2
[n − 2 (` + m2 )](ddσ) + 3n2 (`2 + m2 )(ddπ) + 43 (`2 + m2 )2 (ddδ)
1 2
orbitals, we require that
X
~ `) =
O ξα (~r − R ~ k ).
Γ (O)kβ,`α ξβ (~r − R (3.20)
kβ
However, from (3.17) we know

X
~ `) =
O ξα (~r − R ~ m)
(S−1/2 )mν,`α O ϕν (~r − R
mν
X X
= (S−1/2 )mν,`α ~ k)
Γ (O)kβ,mν ϕβ (~r − R
mν kβ
Xh i
= Γ (O)S −1/2 ~ k) .
ϕβ (~r − R (3.21)
kβ,`α
kβ
On the other hand, substitution of (3.17) into (3.20) gives

Ã !
X X
~
O ξα (~r − R` ) = Γ (O)kβ,`α (S −1/2 ~
)mν,kβ ϕν (~r − Rm )
kβ mν
Xh i
= S −1/2
Γ (O) ~ m)
ϕν (~r − R
mν,`α
mν
Xh i
= S−1/2 Γ (O) ~ k ).
ϕβ (~r − R (3.22)
kβ,`α
kβ
Now comparison of (3.21) with (3.22) shows that the equations are compatible only
if
S−1/2 Γ (O) = Γ (O)S−1/2
or
Γ (O)−1 S −1/2 Γ (O) = S−1/2 . (3.23)
According to (3.23) the Löwdin orbitals can possess the same symmetry as the
atomic orbitals only if S−1/2 is invariant under the unitary transformation, Γ . We
shall prove that this is true for physical systems.
The overlap matrix S has the form,
S=I+∆ (3.24)
where I is the unit matrix and ∆ has zero diagonal elements. This form results
because the atomic orbitals are normalized and therefore the overlap of the orbital
with itself is unity. The non-vanishing off-diagonal elements of ∆ correspond to the
overlap between atomic orbitals centered on different atoms. These overlap integrals
are not necessarily small but they are necessarily less than the diagonal overlap for
any atomic orbitals. That is,
|∆kβ,`α | < 1 . (3.25)
Therefore, it is possible to expand S−1/2 in a convergent power-series as follows:

1 3 5
S−1/2 = (I + ∆)−1/2 = I − ∆ + ∆2 − ∆3 + · · · .
2 8 16
If we can prove that every power of ∆ is invariant under transformation by

Γ (O) then it will follow that S−1/2 is also invariant. The proof is easy:
£ −1 ¤ XX
Γ (O)∆Γ (O) kα,jβ = Γ −1 (O)kα,mν ∆mν,nγ Γ (O)nγ,jβ
mν nγ
Z Ã !Ã !
X X
= d~r Γ −1
(O)kα,mν ϕ∗ν (~r ~
− Rm ) ~
Γ (O)nγ,jβ ϕγ (~r − Rn )
mν nγ
Z
£ ¤ £ ¤
= ~ k ) ∗ Oϕβ (~r − R
Oϕα (~r − R ~ j ) d~r
Z
= ~ k )∗ ϕβ (~r − R
ϕα (~r − R ~ j ) d~r = ∆kα,jβ . (3.26)
Equation (3.26) shows that ∆ is invariant, and therefore so is every power of ∆, since
£ ¤£ ¤ £ ¤
Γ −1 (O)∆N Γ (O) = Γ −1 (O)∆Γ (O) Γ −1 (O)∆Γ (O) · · · Γ −1 (O)∆Γ (O)
= ∆N .
This completes the proof.
A few comments on the use of the LCAO method for building an empirical
model are now in order. In particular, we note that it is not necessary to employ
atomic orbitals as the basis functions. Any set of orbitals that possess the symmetry
of the atomic orbitals can be used. If the interactions are to be treated as empirical
parameters, the results are independent of the actual basis orbitals (assuming the
same number of basis orbitals are employed). The symmetry types and degeneracies
of the electronic states do not depend on the actual basis orbitals, provided they
possess the same transformation properties as the atomic orbitals. For example, s–p
hybrids are often used when deriving molecular electronic states. The symmetry and
degeneracies of the resulting electronic states are no different from those obtained
when one employs atomic p orbitals or Löwdin orbitals. Including overlap integrals
in an empirical model does not improve the results even though it appears there are
more empirical parameters (the overlap integrals). The two models (one without
overlap integrals and one with overlap integrals) are equivalent because they are
related by a unitary transformation.
Suggested reference texts
C. J. Ballhausen and H. B. Gray, Molecular Orbital Theory (New York, W. A. Ben-

jamin, 1965).
W. G. Richards and J. A. Horsley, Ab initio Molecular Orbital Calculations for
Chemists (Oxford, Clarendon Press, 1970).
References
[1] E. Hückel, Z. Physik 70, 204 (1931); ibid 72, 310 (1932); ibid 76, 628 (1932).
[2] J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954).
[3] P.-O. Löwdin, J. Chem. Phys. 18, 365 (1950).
1. Construct the matrix eigenvalue equation for a dx2 −y2 orbital at (0, 0, 0) interacting
with p orbitals at (a, 0, 0) and (−a, 0, 0).
(a) Find the eigenvalues and eigenvectors. Classify the resulting states as bonding,
non-bonding, or antibonding. Use Ed and Ep for the diagonal energies.
(b) Determine the amount of covalent mixing, that is, the ratio of the d to p amplitudes
squared for the various states.
(c) Using Ep = −9 eV, Ed = −5 eV and (pdσ) = 1 eV, calculate the eigenvalues and d
to p ratios of the amplitudes squared.
2. Show that the angular function for ndxy in Table 3.1 is properly normalized.
3. Using the definitions of the angular functions in Table 3.1 express the orbital function
3y 2 − r2 as a linear combination of the two orbitals dz2 and dx2 .
4. Using Table 3.2 express the interactions of the following orbitals in terms of the Slater–
Koster parameters:
(a) px orbital located at (0, a, 0) with a py orbital located at (a, 0, 0),
(b) px orbital at (0, a, 0) with a dx2 −y2 orbital at (0, 0, 0),
(c) dxz at (0, 0, 0) with a dxz at (0, a, 0).
5. Derive the Slater–Koster formula for Ex,x2 −y2 shown in Table 3.2.
4
LCAO energy band model for cubic perovskites
In Chapter 1 a qualitative description of the energy bands of perovskites was devel-

oped starting from a simple ionic model. It was argued that the essential electronic
structure of the perovskites is derived from the BO3 ions of the ABO3 compound.
The A ion was shown to be important in determining the ionic state of the B
ion. Also, the A ion contributes to the electrostatic potentials. However, the energy
bands associated with the outer s orbitals of the A ion were found to be far removed
in energy from the lowest empty d bands and were unoccupied. As a result of these
considerations the electronic structure of the A ion can be neglected in discussing
the principal features of perovskite energy bands.
The energy bands that are electronically and chemically active are derived from
the d orbitals of the transition metal (B) cation and the 2p orbitals of the oxygen
anions. The deep core states of the ions produce atomic-like levels at energies far
below the valence bands and may also be omitted in our discussion.
One of the first energy band calculations for perovskites was carried out by
Kahn and Leyendecker [1] for SrTiO3 . They employed a semiempirical approach
based on the method of Slater and Koster [2] (described in Section 3.2). The model
presented in this chapter follows closely the work of Kahn and Leyendecker.
4.1 The unit cell and Brillouin zone
The BO3 unit cell is shown in Fig. 4.1. The B ion is located at the origin and the
three oxygen ions are located at a distance, a, along the three coordinate axes. The
A ion (not shown in Fig. 4.1) is located at (a, a, a).
Let ~ex , ~ey , and ~ez represent unit vectors along the x, y, and z axes, respectively,
then the perovskite lattice can be described as a single cubic lattice of unit cells
with lattice vectors
~ B (n̂) = 2a(nx~ex + ny ~ey + nz ~ez ) .

R (4.1)
53
54 LCAO energy band model for cubic perovskites
z 6
(0, 0, a)
(0, a, 0)
(0, 0, 0)
B
-
y
(a, 0, 0)
x
+
Figure 4.1. BO3 unit cell for a perovskite.
The symbol n̂ represents the three integers nx , ny , and nz , which may be positive,
negative, or zero. The oxygen ions are located at
~ j (n̂) = R
R ~ B (n̂) + a~ej (j = x, y, or z) (4.2)
O
where ~ej is one of the unit vectors, ~ex , ~ey , or ~ez . The A ions are located at the
positions
~ A (n̂) = R
R ~ B (n̂) + (a~ex + a~ey + a~ez ) . (4.3)
For each unit cell we shall consider 14 basis states; five d orbitals centered at
~ B (n̂) and three 2p orbitals centered on each of the three oxygen ions. These 14
R
basis states produce 14 energy bands. Each energy band has N states, where N is
the number of unit cells in the solid. Each of the 14N energy band states is specified
by a band index, ν, and a wavevector, ~k. The wavevector is chosen to lie in the first
Brillouin zone of the perovskite structure. This Brillouin zone is a cube in ~k-space
as shown in Fig. 4.2.
We can assume that the ~k-vectors corresponding to the N states of each band
lie on a cubic lattice obtained by dividing the first Brillouin zone into N equal sized
cubes.
In the limit as N → ∞, the spacing between these ~k vectors becomes arbitrarily
small and ~k may be treated as a continuous variable.
The points of high symmetry in the zone are shown in Fig. 4.2(a). These points
4.1 The unit cell and Brillouin zone 155
6
kz
R M R
M X M
kz 6
R M R R
M
X
Γ - Γ -
X X ky ky
M M
+ M
+ R kx
kx
R M R
(a) (b)
Figure 4.2. (a) Brillouin zone for a simple cubic perovskite showing the points of high
symmetry, (b) 1/48 segment of the Brillouin zone.
have the coordinates in ~k-space as follows:

³π´
Γ= (0, 0, 0)
³ 2a
π ´ ³π´ ³π´
X= (±1, 0, 0); (0, ±1, 0); (0, 0, ±1)
³ 2a
π ´ ³ 2a ´
π ³ 2a
π ´
M= (±1, ±1, 0); (±1, 0, ±1); (0, ±1, ±1)
2a
³π´ 2a 2a
R= (±1, ±1, ±1) . (4.4)
2a
Figure 4.2(b) shows a representative segment of the first Brillouin zone having 1/48
of the volume of the full zone. The entire zone can be constructed from 48 of these
segments joined in an appropriate manner. If the 48 operations of the cubic point
group (Oh ) are performed on a segment with Γ as the origin, then the segment
will be rotated, reflected, or inverted in such a way as to map out the entire first
Brillouin zone. This means that every ~k-vector point inside a segment is related
to 47 equivalent points in the 47 other segments by a (cubic) symmetry operation.
Consequently, it is necessary to find the energy band solutions only for the ~k-vectors
which lie inside and on the surface of one of the segments of the Brillouin zone.
Solutions for symmetry-equivalent ~k-vectors can then be obtained by performing
the symmetry operation of Oh on the wavefunctions from the segment. The energies
of symmetry-equivalent ~k-vector states are the same.
4.2 LCAO matrix equation for an infinite lattice
In Section 2.8 we obtained the LCAO matrix eigenvalue equation:

~ (i) = 0
(H0 − Ei I )D (4.5)
where the matrix elements of H0 are integrals of a one-electron Hamiltonian between
Löwdin (orthogonalized) orbital basis states.
The Löwdin orbitals are specified by a symmetry index, α, and a position
vector. For compounds such as the perovskites it is convenient to specify the atomic
position by a lattice vector R~ m which locates the unit cell and a vector ~τj which
locates the position of the jth atom within the unit cell. Thus, we write the Löwdin
orbitals as
~ m − ~τj ) = ξα (~r − R
ξα (~r − R ~ mj ) , (4.6)
where
~ mj ≡ R
R ~ m + ~τj .
For a solid having N unit cells, each unit cell having ns basis orbitals, (4.5) is
an ns N × ns N matrix equation. The eigenvector D ~ (i) has ns N components, d(i) ,
mjα
which specify the amplitudes of the Löwdin orbitals comprising the ith eigenstate.
(i)
In an infinite, periodic solid the modulus |dmjα |2 must be the same for every equiv-
alent atomic position. Therefore the amplitudes for a particular symmetry type
orbital of equivalent atoms can differ at most by a phase factor. According to
Bloch’s theorem for periodic systems the amplitudes can be taken in the form
1 ~ ~
dmjα = √ eik·Rm djα (~k, ν)
(i)
(i = ~k, ν) . (4.7)
N
For convenience we also introduce a phase factor within the unit cell and write
~
djα (~k, ν) = eik·~τj ajα . The eigenstates are characterized by the wavevector, ~k, and
a band index ν and therefore on the right-hand side of (4.7) we have replaced the
eigenstate index i by (~k, ν).
The LCAO Bloch wavefunction then assumes the form
1 X X i~k·R~ mj
ψ~kν (~r) = √ e ajα (~k, ν)ξα (~r − R
~ mj ) . (4.8)
N m jα
For a periodic solid the matrix elements of the Hamiltonian between a given pair of
orbital types depend only upon the difference in the position vectors locating the
orbitals so that
~n − R
[H]mjβ,niα = Hjβ,iα (R ~ m) . (4.9)
Equation (4.5) may be reduced to a matrix of ns × ns dimensions by using

(4.7) and (4.9), which reflect the translational invariance of the solid. One has for
the matrix eigenvalue equation:
1 XXn ~n − R
o
~ m ) − E~ δαβ δnm δij ei~k·R~ ni aiα (~k, ν) = 0 .
√ Hjβ,iα (R kν
N n iα
(4.10)
~ ~ ~ m to obtain
We multiply (4.10) by √1 e−ik·Rmj and sum over all R
N
½ ¾
1 X XX ~ ~ ~
√ Hjβ,iα (Rn − Rm ) − E~kν δαβ δnm δij eik·(Rni −Rmj ) aiα (~k, ν)
~ ~
N iα m n
Xn o
= hjβ,iα (~k) − E~kν δαβ δij aiα (~k, ν) = 0 . (4.11)
iα
In obtaining (4.11) we have defined

X ~ ~ ~
hjβ,iα (~k) ≡ ~ p)
eik·Rp eik·(~τi −~τj ) Hjβ,iα (R (4.12)
p
and used the relations

XX ~ ~ ~ X ~ ~
~n − R
eik·(Rn −Rm ) Hjβ,iα (R ~ m) = N ~ p) ,
eik·(Rp ) Hjβ,iα (R
n m p
XX
i~ ~ n −R
k·(R ~ m)
e δnm = N ,
n m
i~
k·(~
τi −~
τj )
δij e = δij .
Equation (4.11) is the desired result. It shows that the energies and wavefunctions
are determined by an ns × ns secular equation. In this form the matrix elements,
hjβ,iα (~k) are the lattice Fourier transforms of the LCAO integrals.
4.3 LCAO matrix elements for the perovskite
In order to solve (4.11), we must specify the elements hjβ,iα (~k) and hence
~ p ). The lattice-space matrix elements can be parameterized by using the
Hjβ,iα (R
Slater–Koster method described in Section 3.2. We must consider the matrix ele-
ments between the 14 basis states within a unit cell and in neighboring unit cells.
For the perovskites an excellent model is obtained if only matrix elements
between first and second nearest neighbors are retained. With this approxima-
tion, cation–anion (nearest-neighbor) interactions and anion–anion (second-nearest-
neighbor) interactions between adjacent oxygen ions are retained and all other in-
teractions are neglected.
The Hamiltonian is of the form

~2 2
H(~r) = − ∇ + veff (~r) , (4.13)
2m
where veff includes the nuclear, Coulomb, and exchange potentials. For our purpose
here it is not necessary to know the explicit form of veff . We need only know that
it possesses certain symmetry properties. In particular, veff (~r) must be invariant
under the operations of the cubic point group Oh when the origin for ~r is at a B-
ion site. Similarly, when the origin is at an oxygen site then veff (~r) must be invariant
to the operations of the point group C4v .
(a) Diagonal LCAO matrix elements
The symmetry properties of the Löwdin orbitals are identical to those of the cor-
responding atomic orbitals as was proved in Section 3.3. As a consequence, the
forms of the atomic wavefunctions listed in Table 3.1 will also be the forms of the
Löwdin orbitals. Table 3.2, for the LCAO two-center integrals, may therefore be
used without change for either atomic or Löwdin basis orbitals.
The functions listed in Table 3.1 are the linear combinations of the spherical
harmonics appropriate for cubic symmetry. According to group theory, the eg and
t2g type d orbitals belong to different irreducible representations of the Oh point
group. This means that the matrix elements of a Hamiltonian (which is invariant
under Oh ) between eg and t2g orbitals centered on the same B ion site must vanish.
Furthermore, the two eg orbitals (three t2g orbitals) belong to different rows of
the same irreducible representation. This means that the matrix elements of the
Hamiltonian between different eg (t2g ) orbitals centered on the same site must also
vanish. No such symmetry restrictions apply to matrix elements between orbitals
centered on different sites.
Similar symmetry considerations show that the matrix elements of the Hamil-
tonian between the different p orbitals centered on the same oxygen ion must also
vanish.
The above discussion shows that the only non-vanishing LCAO integrals be-
tween orbitals centered on the same atomic site are the diagonal matrix elements.
The diagonal matrix e1ements are of the form
Z
Hαα (0) = ϕα (~r)∗ H(~r)ϕα (~r) d~r . (4.14)
From the discussion given in Section 1.3 we know that these integrals will be
approximately the sum of an ionization energy plus a Madelung potential plus an
electrostatic splitting resulting from the non-spherical part of veff (~r). For the d
orbitals we define these elements as
Ed + VM (B) + η(j)∆(d) (j = eg or t2g ). (4.15)
The energy Ed is approximately equal to the ionization energy of a d electron

for the free transition metal ion having the appropriate charge state (e.g., Ti3+ →
Ti4+ for SrTiO3 ). The term VM (B) is the Madelung potential at the B-ion site
and η(j) is a numerical parameter which determines how the electrostatic splitting,
∆(d), is shared between the eg and t2g levels. According to group theory the total
energy of a manifold of degenerate levels is unchanged by the non-spherical part of
veff (~r), that is the “center of gravity” of the levels is unchanged by the electrostatic
splitting of the d states. This requires that
2 η(eg ) + 3 η(t2g ) = 0 ,
∆(d) η(eg ) − ∆(d) η(t2g ) = ∆(d) . (4.16)
3
Thus we find that η(eg ) = 5 and η(t2g ) = − 25 .
In a similar manner, we define the diagonal matrix elements for the 2p oxygen
orbitals by
Ep + VM (O) + η(j) ∆(p) (j = p⊥ , pk ) , (4.17)
where Ep is the (fictitious) electron affinity of the O− ion to form O2− , VM (O)
is the oxygen-site Madelung potential and ∆(p) is the electrostatic splitting due
to the non-spherical part of veff (~r). Invariance of the center of gravity of the 2p
manifold requires that η(p⊥ ) = 31 , while η(pk ) = − 23 where p⊥ and pk refer to 2p
orbitals oriented perpendicular and parallel, respectively, to the B–O internuclear
axis.
The consequence of (4.15) and (4.17) is that we can characterize the LCAO
diagonal matrix elements in terms of four parameters Ee , Et , E⊥ , and Ek which
are defined as follows:
Ee = Ed + VM (B) + η(eg ) ∆(d) ,
Et = Ed + VM (B) + η(t2g ) ∆(d) ,
E⊥ = Ep + VM (O) + η(p⊥ ) ∆(p) ,
Ek = Ep + VM (O) + η(pk ) ∆(p) . (4.18)
(b) Off-diagonal matrix elements
For the model we are considering, which neglects the third nearest-neighbor interac-
tions, there are two types of off-diagonal matrix elements; cation–anion interactions
and anion–anion interactions. The cation–anion matrix elements are of the type,
Z
Hβα (±a~ej ) = ϕα (~r)∗ H ϕβ (~r ± a~ej ) d~r (j = x, y, or z) . (4.19)
The oxygen–oxygen off-diagonal matrix elements are of the type,

Z
Hβα (±a~ei ± a~ej ) = ϕα (~r ± a~ei )∗ H ϕβ (~r ± a~ej ) d~r (4.20)
(i 6= j; i, j = x, y, or z) .
In (4.19), for the p–d matrix elements, ϕα is a d(p) orbital and ϕβ is a p(d) orbital.
The p–d matrix elements can be calculated in terms of the (pdσ) and (pdπ) integrals
(shown in Section 3.2) with the help of Table 3.2. The t2g d orbitals interact only
with the p⊥ -type orbitals; the t2g -pk LCAO integrals vanish by symmetry. Similarly,
the eg -type d orbitals have non-vanishing interactions only with the pk -type orbitals.
For the t2g -p⊥ type matrix elements one finds
Z
dαβ (~r)∗ H pα (~r ∓ a~eβ ) d~r = ±(pdπ) , (4.21)
Z
pα (~r)∗ H dαβ (~r ± a~eβ ) d~r = ∓(pdπ) (αβ = xy, xz, or yz) . (4.22)
For the eg -pk type LCAO matrix elements we have

Z
1
d3z2 −r2 (~r)∗ H pα (~r ∓ a~eα ) d~r = ∓ (pdσ) (α = x or y) , (4.23)
2
Z
d3z2 −r2 (~r)∗ H pz (~r ∓ a~ez ) d~r = ±(pdσ) , (4.24)
Z √
∗ 3
dx2 −y2 (~r) H px (~r ∓ a~ex ) d~r = ± (pdσ) , (4.25)
2
Z √
3
dx2 −y2 (~r)∗ H py (~r ∓ a~ey ) d~r = ∓ (pdσ) . (4.26)
2
The remaining possibilities are determined by noting that
Z Z
p(~r)∗ H d(~r ± a~ej ) d~r = − d(~r)∗ H p(~r ± a~ej ) d~r
for any p and d orbitals. It is noted that all of the possible p–d interactions are
described in terms of only two LCAO integrals: (pdπ) and (pdσ).
Next, we consider the interactions between p orbitals on adjacent oxygen ions.
These interactions can be expressed in terms of the two LCAO integrals (ppπ) and
(ppσ). There are three types of integrals that give non-vanishing matrix elements.
They are
Z
1
pα (~r ∓ a~eα )∗ H pβ (~r ∓ a~eβ ) d~r = −(∓)(∓) [(ppπ) − (ppσ)] (4.27)
2
4.4 LCAO eigenvalue equation for the cubic perovskites 61
(α 6= β; α, β = x, y, or z) ,
where (∓)(∓) means the product of the signs occurring in the arguments of the
orbitals,
Z
1
pα (~r ∓ a~eα )∗ H pα (~r ∓ a~eβ ) d~r = [(ppπ) + (ppσ)] , (4.28)
2
(α 6= β; α, β = x, y, or z) ,
Z
pγ (~r ± a~eα )∗ H pγ (~r ± a~eβ ) d~r = (ppπ) , (4.29)
(α 6= β 6= γ; α, β, γ = x, y, or z) .
It is helpful to visualize the various types of interaction described by (4.21)–(4.29).

Figure 4.3 shows schematic representations of some of these interactions. Equations
(4.21)–(4.29) give all of the possible matrix elements for interactions between first
and second nearest-neighbor ions. They are characterized entirely in terms of four
basic LCAO integrals: (pdσ), (pdπ), (ppσ), and (ppπ). There are also four diagonal
integrals: Ee , Et , Ek , and E⊥ . For this approximate model the electronic structure
of the cubic perovskites is determined entirely by these eight parameters.
4.4 LCAO eigenvalue equation for the cubic perovskites
In the preceding section we determined the forms of all of the matrix elements
which enter the model. To find the energy bands and wavefunctions we must solve
the 14×14 matrix eigenvalue equation corresponding to (4.11).
At this point, it is convenient to make a choice of the labels for the rows and
columns of the 14×14 matrix (see Table 4.1). We make the following correspon-
dence:
dz2 (~r) ⇒ 1; pz (~r − a~ez ) ⇒ 2; dx2 −y2 (~r) ⇒ 3;
px (~r − a~ex ) ⇒ 4; py (~r − a~ey ) ⇒ 5
dxy (~r) ⇒ 6; px (~r − a~ey ) ⇒ 7; py (~r − a~ex ) ⇒ 8 (4.30)
dxz (~r) ⇒ 9; px (~r − a~ez ) ⇒ 10; pz (~r − a~ex ) ⇒ 11
dyz (~r) ⇒ 12; py (~r − a~ez ) ⇒ 13; pz (~r − a~ey ) ⇒ 14 .
This choice is suggested by the fact that in the absence of oxygen–oxygen in-
teractions, the 14×14 matrix block-diagonalizes into a 5×5 and three 3×3 matrices
where rows and columns 1–5 form the 5×5, rows and columns 6–8, 9–11 and 12–14
form the three 3×3 matrices. Since the oxygen–oxygen LCAO integrals are small
this choice of labels places the largest matrix elements in the diagonal blocks.
Next, we determine the matrix elements, hjβ,iα (~k), which enter (4.11). For the
α
(a)
R
β dαβ (~r)∗ H pα (~r − a~eβ ) d~r = (pdπ) (4.21)
α
(b)
R
β pα (~r)∗ H dαβ (~r − a~eβ ) d~r = −(pdπ) (4.22)
(c) z
R
d3z2 −r2 (~r)∗ H pα (~r − a~eα ) d~r = − 12 (pdσ) (4.23)
α
y
(d)
R
d3z2 −r2 (~r)∗ H pz (~r − a~ez ) d~r = (pdσ) (4.24)
x
(e) y
R √
3
dx2 −y2 (~r)∗ H px (~r − a~ex ) d~r = (pdσ) (4.25)
x 2
y
(f)
R √
3
dx2 −y2 (~r)∗ H py (~r − a~ey ) d~r = − (pdσ) (4.26)
x 2
(g) α R
pα (~r − a~eα )∗ H pβ (~r − a~eβ ) d~r
= − 12 [(ppσ) − (ppπ)] (4.27)
β
(h) α
R
pα (~r − a~eα )∗ H pα (~r − a~eβ ) d~r
= 21 [(ppσ) + (ppπ)] (4.28)
β
(i) γ
R
pγ (~r − a~eα )∗ H pγ (~r − a~eβ ) d~r = (ppπ) (4.29)
β
α
Figure 4.3. Schematics of the two-center integrals.

Table 4.1. The 14×14 matrix (H − E~kν I) with vanishing determinant.
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 Ee−E~kν 2i(pdσ)Sz 0 −i(pdσ)Sx −i(pdσ)Sy 0 0 0 0 0 0 0 0 0
2 Ek−E~kν 0 −2bSx Sz −2bSy Sz 0 0 0 0 0 2cCx Cz 0 0 2cCy Cz

√ √
3 Ee−E~kν 3i(pdσ)Sx − 3i(pdσ)Sy 0 0 0 0 0 0 0 0 0
4 Ek−E~kν −2bSx Sy 0 2cCx Cy 0 0 2cCz Cx 0 0 0 0
5 Ek−E~kν 0 0 2cCx Cy 0 0 0 0 2cCy Cz 0
6 Et−E~kν 2i(pdπ)Sy 2i(pdπ)Sx 0 0 0 0 0 0
7 E⊥−E~kν −2bSx Sy 0 4(ppπ)Cy Cz 0 0 0 0
63
8 E⊥−E~kν 0 0 0 0 4(ppπ)Cz Cx 0
9 Et−E~kν 2i(pdπ)Sz 2i(pdπ)Sx 0 0 0

∗
10 Hji = Hij E⊥−E~kν −2bSx Sz 0 0 0
11 E⊥−E~kν 0 0 4(ppπ)Cx Cy
12 Et−E~kν 2i(pdπ)Sz 2i(pdπ)Sy
13 E⊥−E~kν −2bSy Sz
14 E⊥−E~kν
√
Parameters used are b ≡ (ppσ) − (ppπ), c ≡ (ppσ) + (ppπ), Sα ≡ sin kα a and Cα ≡ cos kα a. i ≡ −1.
–2
(a) (b)
–4 σ∗
Eσ ∗ π∗
–6
Energy E (eV)
Eπ ∗
6
–8
Em Eg
–10 Eπ ? π0
Eσ σ0
π
–12
σ
–14
Γ X M R Γ X M R
Figure 4.4. (a) Energy bands for the parameters (all in eV): Ee = – 5.8, Et = – 6.4, E⊥ =
– 10.0, Ek = – 10.5, (pdσ) = 2.1, (pdπ) = 0.8, (ppσ) = – 0.2, (ppπ) = – 0.1. (b) Energy bands
for the same parameters as in (a) except (ppσ) =(ppπ) = 0.
model considered here we need only retain the terms for which
√
~ p + ~τi − ~τj | ≤ 2a ' 2.76 Å .
|R (4.31)
Using the results of (4.18) and (4.21)–(4.29) we obtain the matrix shown in Table
4.1. The eigenvalues, E~k , are determined by the matrix eigenvalue equation
det(H − E~k I) = 0 , (4.32)
where I is a ns × ns unit matrix.

Before discussing the details of how solutions for the energy bands may be
obtained, it is helpful to have a picture of the general structure of the energy bands
[3]. This is best accomplished by displaying a typical set of bands obtained from
(4.32).
Figure 4.4(a) shows the energy bands using the eight LCAO parameters for
SrTiO3 [4, 5]. Figure 4.4(b) shows a similar set of bands for the same parameters
as used in calculating the bands in Fig. 4.4(a) except that the oxygen–oxygen
interaction parameters, (ppπ) and (ppσ), are taken to be zero.
The basic structure of the bands is most easily explained by beginning with
Fig. 4.4(b). There are basically three groups of bands. The first group consists of the
bands labeled σ ∗ and π ∗ which lie between –7 and –2 eV. These are the d-electron
conduction bands. A second group, σ and π, are the mirror images of the σ ∗ and
π ∗ bands, respectively. They are oxygen valence bands. The last group consists of
flat bands labeled σ 0 and π 0 . These are non-bonding oxygen valence bands. The
wavefunctions for the σ and σ ∗ bands involve only the LCAO parameter (pdσ) and
those of the π and π ∗ involve only (pdπ). The widths of the σ and σ ∗ bands are
determined by (pdσ). For the π and π ∗ bands the widths are determined by (pdπ).
The wavefunctions of the σ, σ ∗ , π, and π ∗ bands are admixtures of p and d orbitals.
The non-bonding σ 0 and π 0 bands have wavefunctions that are entirely composed
of oxygen 2p orbitals.
When the oxygen–oxygen interactions, (ppπ) and (ppσ), are non-zero (as for the
bands in Fig. 4.4(a)) the non-bonding bands are no longer flat. They are broadened
into bands whose widths are controlled by (ppπ) and (ppσ). There is also some
minor changes in the σ and π valence bands, because of interaction with the non-
bonding bands near crossing points. However, the general structure of Fig. 4.4(a)
is not qualitatively different from that of Fig. 4.4(b) and the σ ∗ and π ∗ bands are
essentially the same in both figures.
The above analysis suggests that the qualitative features of the energy bands of
the perovskites are determined by nearest-neighbor cation–anion interactions and
that the effects of the anion–anion interactions are small. In the remainder of this
chapter we investigate the analytic solution of (4.32) when (ppπ) and (ppσ) vanish.
Discussions and solutions of the secular matrix equation including the effects of the
oxygen–oxygen interactions are given in Chapter 5.
4.5 Qualitative features of the energy bands
In this section we obtain and discuss the solutions of (4.32) in the absence of
oxygen–oxygen interactions.
Inspection of the matrix in Table 4.1 shows that the matrix equation block-
diagonalizes into a 5×5 and three equivalent 3×3 blocks when (ppπ) and (ppσ) are
set to zero.
(a) Pi bands
The 3×3 blocks involve only Et and the (pdπ) two-center integral and therefore we
refer to these bonds as the “pi bands”. Consider the 3×3 block obtained from the
rows and columns 6–8. The other two 3×3 blocks are equivalent since they may be
obtained from the first by permutation of the coordinate axis labels; substitution
of z for y in the first 3×3 block gives the second and substitution of z for x gives
the last 3×3 block.
The secular equation for the 3×3 blocks is of the form
  
(Et − E~kν ) 2i(pdπ)Sβ 2i(pdπ)Sα aαβ
 −2i(pdπ)Sβ (E⊥ − E~kν ) 0   aα  = 0 (4.33)
−2i(pdπ)Sα 0 (E⊥ − E~kν ) aβ
where αβ = xy, xz, or yz. The coefficients aαβ , aα , and aβ specify the amplitudes
of the orbitals dαβ (~r), pα (~r − a~eβ ), and pβ (~r − a~eα ) making up the eigenstates.
Requiring the determinant of the coefficients to vanish gives the eigenvalue condition
(secular equation),
(Et − E~kν )(E⊥ − E~kν )2 − 4(pdπ)2 (Sα2 + Sβ2 )(E⊥ − E~kν ) = 0 . (4.34)
Since the term (E⊥ − E~kν ) can be factored out from (4.34), one eigenvalue is
E~kν = E⊥ = E~kπ0 . (4.35)
The eigenvector for this energy is easily seen to be
 
0
1  Sα  .
q (4.36)
Sα2 + Sβ2 −Sβ
The real-space wavefunction corresponding to (4.36) is

1 Xn
Ψ~kπ0 (~r) = q ~ m −a~eβ ) ei~k·(R~ m +a~eβ )
Sα pα (~r − R
N (Sα2 + Sβ2 ) m
o
~ m −a~eα ) ei~k·(R~ m +a~eα )
−Sβ pβ (~r − R (4.37)
where the sum is over the lattice vectors, R ~ m , for the locations of the unit cells
only. It is clear from (4.37) that the wavefunction involves only the oxygen orbitals
and contains no d-orbital mixture. This solution correspond to the three symmetry
equivalent π 0 non-bonding bands. Their energy, E~kν = E⊥ is independent of the
~k-vector so the bands are “flat”, that is without dispersion.
The remaining two eigenvalues of (4.34) are

s· ¸2
αβ 1 1
¡ ¢
E~ π∗ = (E⊥ + Et ) ± (Et − E⊥ ) + 4(pdπ)2 (Sα2 + Sβ2 ) . (4.38)
k π 2 2
For ~k = 0 (at Γ), E~kπ∗ = Et and E~kπ = E⊥ . Therefore, we see that the energy gap
between the π (or π 0 ) valence band and the π ∗ conduction band is
Eg0 (Γ) = Et − E⊥ , (4.39)
0
and the “mid-gap” energy, Em (Γ), is
0 1
Em (Γ) = (Et + E⊥ ) . (4.40)
2
The energies, Eg0 (Γ) and Em0
(Γ), shown in Fig. 4.4(b), allow the energies of the π
∗
and π bands to be expressed in a more physical form; namely,
s· ¸2
1 0
E~αβπ∗ = Em 0
(Γ) ± Eg (Γ) + 4(pdπ)2 (Sα2 + Sβ2 ) . (4.41)
k( π ) 2
For simplicity, in the remainder of this book, we shall use the notation, Eg and
Em for Eg0 (Γ) and Em0
(Γ), respectively. The center of gravity of the two bands is at
the mid-gap and the bands are mirror reflections of one another with the “mirror”
located at mid-gap. It is also to be noted that these bands have a two-dimensional
character in that each band depends only on two components of ~k. Thus E~αβ∗ and
kπ
E~αβ are flat along the γ direction in the Brillouin zone, where γ 6= α 6= β.
kπ
The eigenvectors for these bands are
 
(E⊥ − E~αβ )
 kν 
C~kν  2i(pdπ)Sβ  ; (ν = π ∗ or π) (4.42)
2i(pdπ)Sα
where q
C~−1 = (E⊥ − E~αβ )2 + 4(pdπ)2 (Sα2 + Sβ2 ) ,
kν kν
p
= 2(E − Em )(E − E⊥ ) . (4.43)
The validity of (4.42) is easily verified by substitution into (4.33) and use of (4.34).
The real-space wavefunctions are given by
C~ X i~k·R~ m n ~ m)
Ψ~αβ (~r) = √kν e (E⊥ − E~αβ ) dαβ (~r − R
kν N m kν
~ m −a~eβ ) eikβ a
+2i(pdπ)Sβ pα (~r − R
o
~ m −a~eα ) eikα a .
+2i(pdπ)Sα pβ (~r − R (4.44)
Except at Γ the π ∗ and π band wavefunctions are admixtures of d and p

orbitals. The admixture varies with ~k. At Γ, E~kπ∗ = Et and E~kπ = E⊥ . Equation
(4.42) shows that the eigenvector for the π ∗ band is pure d orbital since Sα = Sβ =
0. For the π bands we must determine eigenvectors components by a limiting process
because (E⊥ − E~kπ ) → 0 as ~k → 0, so all the components of (4.42) are tending to
zero. It is easily shown that (E⊥ − E~kπ ) tends to zero quadratically for small |~k|
and thus the eigenvector tends to the form

   
0 0
1  2iSα  −−−→  cos φ  ,
q (4.45)
4(Sα2 + Sβ2 ) 2iSβ ~k → 0
sin φ
q
where cos φ = kα / kα2 + kβ2 . Therefore, the π bands become pure p orbital at Γ.
We conclude that there is no mixing between the p and d orbitals at Γ. This
turns out to be a general result (as we shall show later) and is not associated
with the approximations made in the secular equation. The eigenvalues of the three
symmetry-equivalent blocks are the same so each eigenvalue is triply degenerate.
As a result, arbitrary linear combinations of symmetry-equivalent eigenvectors are
also eigenvectors.
At other points in the Brillouin zone the wavefunctions are admixtures of p
and d orbitals. The mixing increases as |~k| increases and is maximum at the point
R in the Brillouin zone.
The admixture, rd , is easily calculated. For the π valence band the d-orbital
probability rd (E), at E~αβ = E is according to (4.42) given by
kν
|aαβ |2 (E⊥ − E)2

rd (E) = =
|aαβ |2 + |aα |2 + |aβ |2 (E⊥ − E)2 + 4(pdπ)2 (Sα2 + Sβ2 )
(E − E⊥ )
= (4.46)
2(E − Em )
where Em = 12 (Et + E⊥ ). If we define E 0 ≡ E − Em the result is
1 Eg
rd (E 0 ) = + . (4.47)
2 4E 0
Equation (4.47) is valid for energies
p within the range of the valence band, that is for
−EB ≤ E 0 ≤ −Eg /2 with EB = (Eg /2)2 + 8(pdπ)2 . The result shows that for a
given E 0 the amount of mixing of the d orbitals into the valence band is dependent
only on the band gap, Eg , and the p–d interaction, (pdπ).
A rough measure of the d-orbital probability averaged over the valence band,
hrd i, can be obtained if the density of states is taken to be constant (see Chapter
6 for detailed calculations of the density of states). For this approximation,
Z E⊥
1 (E − E⊥ )
hrd i ≡ dE (4.48)
(E⊥ − Em + EB ) Em −EB 2(E − Em )
Z −Eg /2 ³
1 1 Eg ´
= dE 0 + (4.49)
(−Eg /2 + EB ) −EB 2 4E 0
1 1 η Eg
= + ln η, with η= . (4.50)
2 2 1−η 2EB
hrd i depends only on the ratio of the band gap, Eg , to EB , a measure of band
width. The ratio, η, lies between 0 and 1. Small values of η correspond to strong
covalent mixing. The behavior of hrd i versus Eg for several values of (pdπ) is shown
in Fig. 4.5(a). Figure 4.5(b) shows hrd i as a function of (pdπ) for several values of
Eg .
50 50
Eg = 1.0 eV
(pdπ) = 2.0 eV
2.0 eV
1.5 eV 3.0 eV
40 1.0 eV 40 4.0 eV
0.5 eV
+ +
30 30
hrd i%
hrd i%
20 20
+ +
10 10
0 0
0 2 4 6 8 10 0 1 2 3
(a) Eg (eV) (b) (pdπ) (eV)
Figure 4.5. The average d-orbital mixing in the π valence band as functions of Eg and
(pdπ). The two examples solved in the text are marked on the graphs.
For perovskites such as SrTiO3 or BaTiO3 , Eg is approximately 3 eV and (pdπ)

about 1 eV so that
q
EB = (Eg /2)2 + 8(pdπ)2 = 3.2016 eV (4.51)
Eg
η= = 0.4685 (4.52)
2EB
1 1 η
hrd i = + ln η = 0.1658. (4.53)
2 2 1−η
This result means that the average valence-band wavefunction (for SrTiO3 or
BaTiO3 ) is about 17% d orbital and 83% p orbital in content. For the average
π ∗ conduction-band wavefunction it is just the reverse: 17% p orbital and 83% d
orbital in content.
For the metallic perovskite ReO3 , the band gap is only about 1 eV while
(pdπ) ' 1.5 eV. In this case, EB = 4.2720 eV, η = 0.1170, and hrd i = 0.3578. Thus
the average valence-band wavefunction has about 36% d orbital and 64% p orbital.
Clearly, ReO3 is much more covalent than SrTiO3 or BaTiO3 .
(b) Sigma bands
The upper 5×5 block in (4.32) determines the sigma bands. With (ppσ) = (ppπ) = 0
the remaining three parameters involved are Ee , Ek , and (pdσ). The matrix equation
is
  
(Ee − E~kν ) 2i(pdσ)Sz 0 −i(pdσ)Sx −i(pdσ)Sy az 2
 −2i(pdσ)Sz (Ek − E~ ) 0 0  az 
 kν √ √ 0  
 3i(pdσ)Sx − 3i(pdσ)Sy   
 0 0 √ e − E~kν )
(E  ax2  = 0.
 i(pdσ)Sx 0 −√ 3i(pdσ)Sx (Ek − E~kν ) 0  ax 
i(pdσ)Sy 0 3i(pdσ)Sy 0 (Ek − E~kν ) ay
(4.54)
The coefficients specify the amplitudes of the orbitals as follows:
az2 ⇒ dz2 (~r)

az ⇒ pz (~r − a~ez )
ax2 ⇒ dx2 −y2 (~r) (4.55)
ax ⇒ px (~r − a~ex )
ay ⇒ py (~r − a~ey ) .
This 5×5 matrix equation can be solved exactly. The eigenvalue equation ob-
tained from the determinant of the matrix is
©
(Ee − E~kν )2 (Ek − E~kν )2 − 4(pdσ)2 (Ee − E~kν )(Ek − E~kν )(Sx2 + Sy2 + Sz2 )
ª
+12(pdσ)4 (Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 ) (Ek − E~kν ) = 0 . (4.56)
Since the term (Ek − E~kν ) can be factored from the result one obvious eigenvalue
is
E~kν = Ek . (4.57)
This flat, non-bonding sigma band will be denoted by E~kσ0 . The eigenvector is
easily found to be
 
0
 Sx Sy 
1  
q  0 . (4.58)
 
Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2  Sy Sz 
Sz Sx
The corresponding real-space wavefunction is
1 X n
~ ~ ~ m −a~ez ) eikz a
Ψ~kσ0 (~r) = q eik·Rm Sx Sy pz (~r − R
N (Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 ) m
o
~ m −a~ex ) eikx a + Sz Sx py (~r − R
+Sy Sz px (~r − R ~ m −a~ey ) eiky a . (4.59)
Equation (4.59) shows that the wavefunction for the σ 0 band is composed entirely
of p orbitals.
Returning to the eigenvalue equation (4.56), we see that after factoring out
the term (Ek − E~kν ) the remaining expression is quadratic in the variab1e (Ee −
E~kν )(Ek − E~kν ). This allows an immediate solution with the result that
n¡ ¢
(Ee − E~kν )(Ek − E~kν ) = 2(pdσ)2 Sx2 + Sy2 + Sz2
q¡ ¢ ¡ ¢o
± Sx4 + Sy4 + Sz4 − Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 . (4.60)
Equation (4.60) is quadratic in the variable E~kν so it can be solved to give the
eigenvalues. They are:
rh
(±) 1£ ¤ 1¡ ¢i2 h¡ ¢ i
E~ ∗ = Ee + Ek + Ee −Ek + 2(pdσ)2 Sx2 + Sy2 + Sz2 ±S 2 (4.61)
kσ 2 2
rh
(±) 1¡ ¢ 1¡ ¢i2 h¡ ¢ i
E~ = Ee + Ek − Ee −Ek + 2(pdσ)2 Sx2 + Sy2 + Sz2 ±S 2 (4.62)
kσ 2 2
q
S 2 ≡ (Sx4 + Sy4 + Sz4 )−(Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 ) . (4.63)
(±) (±)
depend on all three components of ~k. However, E~ ∗
(−)
In general E~ and E~
kσ ∗ kσ kσ
(−)
and E~ are flat along any Γ to X direction. To see this let ~k = kα~eα , where α =
kσ
(−) (−)
x, y, or z, then S 2 = Sα2 and E~ = Ek ; E~ ∗ = Ee independent of the magnitude
kσ kσ
of ~k.
The eigenvectors are:

 
(Ek − E~kν )Xν /(pdσ)
 2iSz Xν 
 √ 
C~kν − 3(Ek − E~kν )(Sx − Sy )/(pdσ)
 2 2
, (4.64)
 −iSx [Xν + 3(Sx2 − Sy2 )] 
2 2
−iSy [Xν − 3(Sx − Sy )]
with ν = σ ∗ (±) or σ(±) and
Xν = (Ek − E~kν )(Ee − E~kν )/(pdσ)2 − 3(Sx2 + Sy2 ) ,

= 2Sz2 − Sx2 − Sy2 ± 2S 2 ,
where the factor, C~kν , the normalization coefficient is equal to the inverse square
root of the sum of the squares of the components in (4.64), namely
n o−1/2
C~kν = Xν2 [ηk2 + Sx2 + Sy2 + 4Sz2 ] + (Sx2 − Sy2 )2 [6Xν + 3ηk2 + 9(Sx2 + Sy2 )] (4.65)
where ηk = (Ek − E~kν )/(pdσ). The real-space wavefunction is
C~ X i~k·R~ n ~ m ) + az eikz a pz (~r − R

~ m −a~ez )
Ψ~kν (~r) = √kν em az2 dz2 (~r − R
N m
o
~ m −a~ex ) + ay eiky a py (~r − R
+ax eikx a px (~r − R ~ m −a~ey ) . (4.66)
The form of the eigenvector given by (4.64) is not convenient to use when
only one of the components of ~k is non-zero (that is along Γ–X) because all of the
components vanish and a limiting process must be used. It is more convenient to
return to the original matrix equation, (4.54) and recalculate the eigenvectors.
If kx = ky = 0 and kz 6= 0, then (4.54) becomes
  
(Ee − Ekz ν ) 2i(pdσ)Sz 0 0 0 az2
 −2i(pdσ)Sz (Ek − Ekz ν ) 0 0 0  az 
  
  ax2
 0 0 (Ee − Ekz ν ) 0 0   = 0.
 0 0 0 (Ek − Ekz ν ) 0  ax 
0 0 0 0 (Ek − Ekz ν ) ay
(4.67)
It is immediately seen that two of the valence-band states have eigenvalues Ek and
   
0 0
0 0
   
that 0 and 0 are the eigenvectors. Similarly, one of the conduction bands
1 0
0 1
4.6 Summary of the chapter results 73
 
0
0
(−)  
is flat with energy Ekz σ∗ = Ee and eigenvector 1. The two orbitals dz2 (~r) and
0
0
pz (~r − a~ez ) are mixed. The 2×2 matrix may be diagonalized to obtain:
rh i2
(+) 1 1 4(pdπ)2 (kz a)2
Ekz σ∗ = (Ee + Ek ) + (Ee − Ek ) + 4(pdσ)2 Sz2 −−k−z−→
−−→ Ee +
0 ,
2 2 Eg
(4.68a)
rh i
(+) 1 1 2 4(pdπ) (kz a)2
2
Ekz σ = (Ee + Ek ) − (Ee − Ek ) + 4(pdσ)2 Sz2 −−k−z−→
−−→ Ek −
0 .
2 2 Eg
(4.68b)
The amplitudes az2 and az are the only non-zero components of the eigenvectors.
For the conduction band and the valence band the eigenvectors, respectively, are
 (+)
  
(Ek − Ekz σ∗ ) 1
   0
 2i(pdσ)S z 
1   −−−−−−→  
q  0  kz → 0  0 

 , (4.69a)
(+)   
(Ek − Ekz σ∗ )2 + 4(pdσ)2 Sz2  0  0
0 0
   
2i(pdσ)Sz 0
 (+)  1
 −(Ee − Ekz σ ) 
1   −−−−−−→  
q  0  kz → 0  0 . (4.69b)
(+)   0
(Ee − Ekz σ )2 + 4(pdσ)2 Sz2  0 
0 0
It follows from (4.68) and (4.69) that the mixing between the p and d orbitals
vanishes at Γ and increases linearly with |~k| away from Γ along the vector (0, 0, kz ).
Similar results can easily be obtained for other symmetry directions.
4.6 Summary of the chapter results
In the preceding sections of this chapter we formulated a 14×14 matrix energy band
problem for a model which includes cation–anion interactions (nearest-neighbor)
and anion–anion (second-nearest-neighbor) interactions. The cation–anion inter-
actions involved the LCAO two-center integrals (pdσ) and (pdπ). For the per-
ovskites, (pdπ) is usually in the range of 1.0–1.5 eV and (pdσ) is usually two to
three times larger than (pdπ). The anion–anion interactions involve the LCAO pa-
rameters (ppσ) and (ppπ). These parameters are usually 5–10 times smaller than
the p–d two-center integrals. Because of this, the qualitative features of the energy
bands can still be obtained even when (ppσ) and (ppπ) are neglected.1 This point
is demonstrated in Fig. 4.4 which shows that the major effect of (ppσ) and (ppπ)
is to broaden the flat non-bonding oxygen bands into bands of narrow widths.
The LCAO energy bands model was solved analytically for the approximation
(ppσ) = (ppπ) = 0. For this approximation the matrix equation is block-diagonal. A
5×5 block involves only the eg -type d orbitals and the 2p orbitals oriented parallel
to the B–O axis. These orbitals interact only through LCAO integral (pdσ). The
five bands resulting from the 5×5 block were designated as the “sigma bands”.
One band, a flat non-bonding band called σ 0 , involves only the 2p orbitals. Two
conduction bands, termed σ ∗ bands, are formed whose width depends on (pdσ).
Two valence bands, called σ bands, are also formed. The σ bands are the mirror
reflection of the σ ∗ bands with the mirror located at the mean energy (Ee + Ek )/2.
The mixing between the p and d orbitals was shown to vanish at Γ, and to
increase with |~k| near Γ. The maximum covalent mixing occurs at R in the Brillouin
zone. The remaining 9×9 of the secular matrix is block-diagonal and consists of
three equivalent 3×3 blocks. Each block involves one of the three t2g -type d orbitals,
dαβ , and the p orbitals which interact through the LCAO (pdπ) parameter. These
bands were named the “pi bands”. Each 3×3 yields one (π ∗ ) conduction band, one
(π) valence band and one flat non-bonding (π 0 ) valence band. The π and π ∗ bands
are mirror reflections of one another with the mirror located at (Et + E⊥ )/2. The
π and π ∗ bands were found to possess two-dimensional character. The dispersion
(E~kν versus ~k) depends only on two components of the three-dimensional ~k. This
results in flat bands along several lines in the Brillouin zone. Similar features were
found for one of the σ and σ ∗ bands. Such flat regions have significant effects
on the density of states and, as will be shown in subsequent chapters, produce
characteristics structure in the optical, photoemission spectra.
The p–d mixing of the eigenvectors of the π and π ∗ bands also vanishes at Γ
and also increases with increasing |~k| away from Γ. Typical mixing ratios at R in
the Brillouin zone range from 65% to 85% d and 35% to 15% p for the conduction
bands and conversely for the valence bands. There is no d-orbital component for
the non-bonding (σ 0 and π 0 ) bands for the approximation (ppσ) = (ppπ) = 0. Even
when (ppσ) and (ppπ) are finite there is only a very small d component so that the
non-bonding bands may be regarded as having pure p-orbital wavefunction for all
practical purposes.
In the next chapter we look at some analytical solutions of the energy
bands with (ppσ) and (ppπ) 6= 0 and determine in more detail the effect of the
oxygen–oxygen interactions.
1
An important exception to this rule occurs for the high-temperature superconductors. For these mate-
rials the effective oxygen–oxygen interactions may be comparable to the p–d interactions due to strong
electron–electron correlation effects. The d bands of the cuprate HTSCs are discussed in Chapter 11.
References for Chapter 4 75
Suggested references on “Group theory and ligand fields”
S. Sugano, Y. Tanabe and H. Kamimura, Multiplets of transition metal ions in

crystals. In Pure and Applied Physics, Vol. 33 (New York and London, Academic
Press, 1970).
M. Tinkham, Group theory and quantum mechanics, International Series in Pure

and Applied Physics (New York, McGraw-Hill, 1964).
References
[1] A. H. Kahn and A. J. Leyendecker, Phys. Rev. A 135, 1321 (1964).
[2] J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954).
[4] T. Wolfram, Phys. Rev. Lett. 29, 1383 (1972).
[5] L. F. Mattheiss, Phys. Rev. B 6, 4718 (1972).
1. Calculate all of the matrix elements of the forth row of the 14×14 matrix given in
Table 4.1.
2. Show by direct matrix multiplication that the eigenvector of (4.42) satisfies (4.33).
3. Calculate the covalent mixing ratio,

|az2 |2 + |ax2 |2
rm =
|az |2 + |ax |2 + |ay |2
for the σ ∗ band at the symmetry point M = (π/2a)(1, 1, 0).
4. Find an analytic expression for the band width of the π ∗ band and show that it is
dependent only on the energy gap Eg = Et − E⊥ and (pdπ). (The band width is
defined as the difference between the highest and lowest energy of the band.) Assuming
a band gap of 3 eV and (pdπ) = 1 eV, calculate the π ∗ band width.
5. Find an analytic expression for the band width of the σ ∗ band and show that it is
dependent only on Egσ = (Ee − Ek ) and (pdσ). Assuming Eg = 5 eV and (pdσ) =
3 eV, calculate the σ ∗ band width.
6. The average, hsin2 (kx a)i, is defined as

µ ¶Z
1
d~k sin2 (kx a)
VBZ VBZ
where the integral is over the entire first Brillouin zone and VBZ is the volume of the
first Brillouin zone. For the pi bands show that
1 (E − Em )2 − (Eg /2)2
hsin2 (kx a)i = .
2 4(pdπ)2
(Hint: Use symmetry arguments.)
7. Using the parameters of Fig. 4.4(b) to calculate the π ∗ -band energies corresponding to
the wavevectors at Γ, X, and M.
5
Analysis of bands at symmetry points
In Chapter 4 the general features of the energy bands of the perovskites were
determined for the approximation in which (ppσ) = (ppπ) = 0. In this chapter we
examine the solutions of the energy bands for ~k-vectors at points of high symmetry
in the Brillouin zone with (ppσ) and (ppπ) 6= 0. From these solutions the role of the
oxygen–oxygen interactions in determining the band gap and energy band structure
can be assessed.
It is possible to diagonalize the 14×14 energy band matrix exactly at all of
the points of high symmetry in the Brillouin zone including Γ, X, M, and R. In
what follows we present tables which give the eigenvalues, eigenvectors, and the
real-space wavefunctions for each of the 14 energy bands.
5.1 Energy bands at Γ
At Γ in the first Brillouin zone, ~k = (0, 0, 0) and many of the matrix elements of
secular matrix (4.32) vanish. The 14×14 matrix can be block-diagonalized by rear-
ranging rows and columns (equivalent to a unitary transformation).
Table 5.1 summarizes the results. The first column specifies a (arbitrary) nu-
merical label for each of the 14 states. The second column gives the rows and
columns of the 14×14 matrix of (4.32) involved in the block which determines en-
ergy band states. These numbers also specify the basis orbitals involved according
to the notation adopted in Chapter 4. The notation is given by (4.30). By rearrang-
ing the columns and rows of (4.32) in the order 1, 3, 6, 9, 12, 2, 11, 14, 4, 7, 10, 5, 8,
13 the secular matrix assumes a block-diagonal form. Rows and columns 1, 3, 6, 9,
12 correspond to 1×1 blocks. Rows and columns 2, 11, 14; 4, 7, 10; and 5, 8, 13 each
form 3×3 blocks. The 3×3 blocks are symmetry equivalent. The dimensionality of
the blocks can be inferred from column two of Table 5.1.
Column three of Table 5.1 specifies the type of energy band state in terms
of the “pi” and “sigma” notation discussed in Chapter 4. An entry such as π + σ
77
78 Analysis of bands at symmetry points
indicates a valence band involving both pi and sigma basis orbitals and π ∗ + σ ∗
would indicate a conduction band involving pi and sigma basis orbitals.
Column four of Table 5.1 gives the energy of the band state, EΓν . The eigenvec-
tor for the state (within the unit cell) is given in column five, where |ni represents
the orbital specified by n in (4.30). The total real-space wavefunction can be con-
structed by summing the local eigenvectors |n, mi over all unit cells, taking into
~ ~
account the phase factors eik·(Rm +~τj ) of each orbital, and properly normalizing the
total state. For instance, the total real-space wavefunction for the 14th state in the
last row of Table 5.1 will read as follows:
1 X i~k·R~ mn ~
h
~ ~
io
√ e −4c|5, mi eik·~τ5 + (Ek − EΓ14 ) |8, mi eik·~τ8 + |13, mi eik·~τ13 .
NS m
(5.1)
~ ~ ~ ~ ~
Since ~k = 0 at Γ, the phase factors eik·Rm , eik·~τ5 , eik·~τ8 , and eik·~τ13 are all unity. The
total wavefunction reduces to
1 Xn £ ¤o
√ −4c|5, mi + (Ek − EΓ14 ) |8, mi + |13, mi . (5.2)
NS m
Here the notation is
|1, mi = dz2 (~r − R~ m ), ~ m − a~ez ),
|2, mi = pz (~r − R
|3, mi = dx2 −y2 (~r − R~ m ), ~ m − a~ex ),
|4, mi = px (~r − R
~ m − a~ey ),
|5, mi = py (~r − R
~ m ),
|6, mi = dxy (~r − R ~ m − a~ey ),
|7, mi = px (~r − R
~ m − a~ex ),
|8, mi = py (~r − R (5.3)
~ m ),
|9, mi = dxz (~r − R ~ m − a~ez ),
|10, mi = px (~r − R
~ m − a~ex ),
|11, mi = pz (~r − R
~ m ),
|12, mi = dyz (~r − R ~ m − a~ez ),
|13, mi = py (~r − R
~
|14, mi = pz (~r − Rm − a~ey )
~ m = 2a[nx (m), ny (m), nz (m)].
where R
The last column gives the symmetry labels of the irreducible representations
of the group of the wavevector in the notation of Bouckaert, Smoluchowski, and
Wigner [1].
Table 5.1 shows that the conduction bands, σ ∗ and π ∗ , have wavefunctions
that involve only d orbitals. There is no mixing between the p and d orbitals at
Γ. The energies at Γ of these states are just the diagonal matrix elements which
correspond to the ionic model energies including the electrostatic splittings.
The nine valence bands consist of three sets of equivalent states. The π 0 state
Table 5.1. Energy bands at Γ, ~k=(0, 0, 0).
State Rows Type Energy (EΓν ) Eigenvector Wavefunction Sym.

P
Γ1 1 σ∗ Ee |1i √1 |1, mi Γ12
N
∗
P
Γ2 3 σ Ee |3i √1 |3, mi Γ12
N
P
Γ3 6 π∗ Et |6i √1 |6, mi Γ250
N
∗
P
Γ4 9 π Et |9i √1 |9, mi Γ250
N
P
Γ5 12 π∗ Et |12i √1 |12, mi Γ250
N
 1
P 1
Γ6 π0 E⊥ − 4(ppπ) √ [|11i − |14i] √1 √ [|11, mi − |14, mi] Γ25
 2 N 2
 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ7 2, 11, 14 π+σ EA + EDc √1 −4c|2i + Ek − EΓ7 (|11i + |14i] √1 −4c|2, mi + Ek − EΓ7 [|11, mi + |14, mi] Γ15
S NS

79
 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ8 π+σ EA − EDc √1 −4c|2i + Ek − EΓ8 (|11i + |14i] √1 −4c|2, mi + Ek − EΓ8 (|11, mi + |14, mi] Γ15
S NS
 1
P 1
Γ9 π0 Same as Γ6 √ [|7i − |10i] √1 √ [|7, mi − |10, mi] Γ25
 2 N 2
 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ10 4, 7, 10 π+σ Same as Γ7 √1 −4c|4i + Ek − EΓ10 [|7i + |10i] √1 −4c|4, mi + Ek − EΓ10 [|7, mi + |10, mi] Γ15
S NS

 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ11 π+σ Same as Γ8 √1 −4c|4i + Ek − EΓ11 [|7i + |10i] √1 −4c|4, mi + Ek − EΓ11 [|7, mi + |10, mi] Γ15
S NS
 0 1
P 1
Γ12 π Same as Γ6 √ [|8i − |13i] √1 √ [|8, mi − |13, mi] Γ25
 2 N 2
 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ13 5, 8, 13 π+σ Same as Γ7 √1 −4c|5i + Ek − EΓ13 [|8i + |13i] √1 −4c|5, mi + Ek − EΓ13 [|8, mi + |13, mi] Γ15
S NS

 £ ¡ ¢ ¤ P£ ¡ ¢ ¤
Γ14 π+σ Same as Γ8 √1 −4c|5i + Ek − EΓ14 [|8i + |13i] √1 −4c|5, mi + Ek − EΓ14 [|8, mi + |13, mi] Γ15
S NS
1
EA = 2 (Ek + E⊥ ) + 2(ppπ), c = (ppσ) + (ppπ), S = 2(Ek − EΓν )2 + 16c2 .
1
p
ED = − E⊥ ) − 2(ppπ), EDc = 2 + 8c2 , Rm = 2a[nx (m), ny (m), nz (m)].
2 (Ek ED
Figure 5.1. Schematic representation of the eigenstates at Γ. Symmetry equivalent states

can be obtained by switching axes. Γ8 in (c) and Γ14 in (d) are presented as an example
of such an operation. The difference in relative sizes of the planar and apical orbitals stem
from the difference in the magnitudes of (Ek − EΓ8 ) and (Ek − EΓ14 ) as compared to −4c
(see Table 5.1 and Fig. 5.3).
is a non-bonding oxygen p-orbital band. Its energy is shifted by the oxygen–oxygen

interaction from E⊥ to E⊥ − 4(ppπ). There are three degenerate π 0 bands. The
wavefunction for a π 0 band is illustrated schematically in Fig. 5.1(a). The wave-
function is repeated throughout the lattice. The π + σ states, 7, 8; 10, 11; and 13,
14 are admixtures of p⊥ and pk orbitals. Schematic representations of these states
are given in Figs 5.1(b) and (c).
The effect of the oxygen–oxygen interactions on the band-gap energies at Γ
is shown in Fig. 5.2. For the usual case of |(ppπ)| < |c|, the mixed states Γ7 (or
Γ10 or Γ13) lie above the non-bonding π 0 (Γ6, Γ9, or Γ12) states when (ppπ) and
(ppσ) 6= 0, so that the true band gap, Eg (Γ), is somewhat smaller than the ionic
gap Eg0 (Γ). This feature is illustrated in Fig. 5.2. For example, for SrTiO3 the
ionic gap is approximately 3.6 eV while Eg (Γ) = 3.24 eV. Thus the oxygen–oxygen
Ee (σ ∗ ) Ee (σ ∗ )
Et (π ∗ ) Et (π ∗ )
6 6
Eg (Γ)
Eg0 (Γ)
E(Γ7) (π + σ) ?
E⊥ (π 0 , π) ?
E⊥ − 4(pdπ) (π 0 )
Ek (σ 0 , σ) E(Γ8) (π + σ)
(ppσ) = (ppπ) = 0 c ≫ (ppπ) > 0
Figure 5.2. Splitting of the states at Γ by the oxygen–oxygen interactions.
interactions cause a 10% reduction in the ionic band gap at Γ. The band-gap energy
at Γ can be written in terms of ∆(p) = E⊥ − Ek as
Eg (Γ) = Eg0 (Γ) − δEg ,

h1 i rh 1 i2
δEg = − ∆(p) − 2(ppπ) + ∆(p) + 2(ppπ) + 8c2 . (5.4)
2 2
The energies at the various symmetry points are shown in Fig. 5.3. It is noted
that there is no mixing between the pi- and sigma-type (t2g and eg ) d orbitals but
the pi- and sigma-type p orbitals are mixed by the oxygen–oxygen interactions.
5.2 Energy bands at X
The X points in the Brillouin zone are at ~k = 2a

π π
(±1, 0, 0); 2a π
(0, ±1, 0); 2a (0, 0, ±1).
~ π
We consider the X point at k = 2a (1, 0, 0) for which Sx = Cy = Cz = 1 and Cx =
Sy = Sz = 0 where Sα = sin kα a and Cα = cos kα a.
The results of the analysis of the secular equation are summarized in Table 5.2.
The total real-space wavefunction can be constructed by summing the local eigen-
~ ~
vectors, |n, mi, over all unit cells, taking into account the phase factors eik·(Rm +~τj )
of each orbital and normalizing the total state. For instance, the 10th state in Table
π
Table 5.2. Energy bands at X, ~k = 2a (1, 0, 0).
State Rows Type Energy (EXν ) Eigenvector Wavefunction Sym.

P
X1 12 π∗ Et |12i √1 ηm |12, mi X3
N
¾ P
X2 π0 E⊥ + 4(ppπ) √
1 [|7i − |10i] √1 ηm [|7, mi − |10, mi] X 0
2 2N 4
7, 10 P
X3 π0 E⊥ − 4(ppπ) √
1 [|7i + |10i] √1 ηm [|7, mi + |10, mi] X 0
2 2N 3
) p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X4 π∗ E1 + E 2 + 4(pdπ)2 p1 −2i(pdπ)|6i + Et − EX4 |8i p1 ηm −2i(pdπ)|6, mi + Et − EX4 |8, mi X5
2
6, 8 S2 N S2
p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X5 π E1 − E 2 + 4(pdπ)2 p1 −2i(pdπ)|6i + Et − EX5 |8i p1 ηm −2i(pdπ)|6, mi + Et − EX5 |8, mi X5
2
S2 N S2
) £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X6 π∗ Same as X4 p1 −2i(pdπ)|9i + Et − EX6 |11i p1 ηm −2i(pdπ)|9, mi + Et − EX6 |11, mi X5
9, 11 S2 N S2
£ ¡ ¢ ¤ P £ ¡ ¢ ¤
X7 π Same as X5 p1 −2i(pdπ)|9i + Et − EX7 |11i p1 ηm −2i(pdπ)|9, mi + Et − EX7 |11, mi X5
S2 N S2
) p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X8 π+σ EA + E 2 + 4c2 p1 −2c|2i + Ek − EX8 |14i p1 ηm −2c|2, mi + Ek − EX8 |14, mi X 0
D 5
82
2, 14 S1 N S1
p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X9 π+σ EA − E 2 + 4c2 p1 −2c|2i + Ek − EX9 |14i p1 ηm −2c|2, mi + Ek − EX9 |14, mi X 0
D 5
S1 N S1
) £ ¡ ¢ ¤ P £ ¡ ¢ ¤
X10 π+σ Same as X8 p1 −2c|5i + Ek − EX10 |13i p1 ηm −2c|5, mi + Ek − EX10 |13, mi X 0
5
5, 13 S1 N S1
£ ¡ ¢ ¤ P £ ¡ ¢ ¤
X11 π+σ Same as X9 p1 −2c|5i + Ek − EX11 |13i p1 ηm −2c|5, mi + Ek − EX11 |13, mi X 0
5
S1 N S1
 1
£√ ¤ P £√ ¤
X12 σ∗ Ee √ 3|1i + |3i √1 ηm 3|1, mi + |3, mi X2
 2 2N

 h h i
p £ √
¤ ¡ ¢ i P £ √
¤ ¡ ¢
X13 1, 3, 4 σ∗ E3 + E 2 + 4(pdσ)2 p1 (pdσ) |1i − 3|3i −i Ee − EX13 |4i p1 ηm (pdσ) |1i − 3|3, mi + Ee − EX13 |4, mi X1
4
 S3 N S3

 p h £ ¤ ¡ ¢ i P h £ ¤ ¡ ¢ i
√ √
X14 σ E3 − E 2 + 4(pdσ)2 p1 (pdσ) |1i − 3|3i −i Ee − EX14 |4i p1 ηm (pdσ) |1i − 3|3, mi + Ee − EX14 |4, mi X1
4
S3 N S3
1 1
EA = 2 (Ek + E⊥ ), ED = 2 (Ek − E⊥ ), S1 = (Ek − EXν )2 + 4c2 , c = (ppσ) + (ppπ).
1 1 2 2
E1 = 2 (Et + E⊥ ), E2 = 2 (Et − E⊥ ), S2 = (Et − EXν ) + 4(pdπ) ηm = (−1)nx (m) .
1 1 2 2
E3 = 2 (Ee + Ek ), E4 = 2 (Ee − Ek ), S3 = (Ee − EXν ) + 4(pdσ) , R
~ m = 2a[nx (m)~
ex + ny (m)~
ey + nz (m)~
ez ].
M11
R11,R13
X13
M13
M9 R2,R5,R8
X12
Γ2,Γ1
X4,X6 M4,M6
Γ5,Γ4,Γ3
X3
Γ13,Γ10,Γ7 X2 R10
M2,M3
Γ12,Γ9,Γ6 X8,X10
X1 M8 R1,R4,R7
Γ14,Γ11,Γ8 X9,X11 M1
X5,X6 M5,M7
M10 R3,R6,R9
M14
X14
R12,R14
M12
Γ X M R
Figure 5.3. The symmetry states listed in Tables 5.1–5.4 indicated on the energy bands
diagram calculated from (4.32) with the parameters (all in eV): Et = –6.5, Ee = –5.5,
E⊥ = –10.0, Ek = –10.5, (pdσ) = –2.0, (pdπ) = 1.0, (ppσ) = –0.15, and (ppπ) = 0.05.
5.2 will read as follows:
1 X i~k·R~ m £ ~ ~ ¤
√ e −2c|5, mi eik·~τ5 + (Ek − EX10 )|13, mi eik·~τ13 (5.5)
N S1 m
Since ~k at X is equal to π
2a ~
ex , we have
~ ~ π
eik·Rm = ei 2a [2anx (m)] = eiπnx (m) = (−1)nx (m) ,
where nx (m) is the integer nx in R ~ m = 2a(nx~ex + ny ~ey + nz ~ez ). Since ~τ5 = a~ey and
i~
k·~
τ5 ~
~τ13 = a~ez we find that e = 1, and eik·~τ13 = 1. The resulting total wavefunction is
1 X £ ¤
√ (−1)nx (m) −2c|5, mi + (Ek − EX10 )|13, mi . (5.6)
N S1 m
Using the 14th state in Table 5.2 we find the total real-space wavefunction to be
1 X © £ √ ¤ ª
√ (−1)nx (m) (pdσ) |1, mi − 3|3, mi + (Ee − EX14 )|4, mi . (5.7)
N S3 m
~ ~ ~
In obtaining (5.7) we have used the fact that at X, eik·~τ1 = 1, eik·~τ3 = 1, and eik·~τ4 =
π π
ei 2a ~ex ·(a~ex ) = ei 2 = i. Thus we have
~
−i(Ee − EX14 )|4, mieik·~τ4 = +(Ee − EX14 )|4, mi.
The energies at the X point are indicated in Fig. 5.3 for a typical set of energy
bands. The highest valence-band state depends upon the values of (ppπ) and (ppσ).
Usually X2 is the upper valence band when |(ppπ)| ¿ |(ppσ)|. In such a case the
band gap at X, Eg (X) is Et –E(X2). From Table 5.1 it can be seen that E(Γ7) >
E(X2) and therefore the energy gap at Γ is smaller than at X. There is some
controversy about whether the minimum band gap is the direct gap between Γ7
and Γ5 or an indirect gap between Γ7 and X3. The model considered here gives
the same energy for either gap since E(Γ5) = E(X3) = Et (because one of the π ∗
bands is flat along Γ to X). Other effects, such as the spin–orbit interaction and
more distant neighbor interactions may alter the equality of E(Γ5) and E(X3). It
is noted that there is no mixing between the t2g and eg orbitals, however, the pi
and sigma p orbitals are mixed by the oxygen–oxygen interactions.
5.3 Energy bands at M
The M points in the Brillouin zone are at ~k = 2a π

(±1, ±1, 0); π
2a (±1, 0, ±1);
π ~ π
2a (0, ±1, ±1). We consider the M point at k = 2a (1, 1, 0).
The secular matrix equation is block-diagonal as follows: three 1×1 blocks cor-
responding to non-bonding oxygen states, two equivalent 2×2 blocks corresponding
to pi conduction and valence bands, one 3×3 which yields a π 0 non-bonding valence
band, a π ∗ conduction band and a π valence band, and one 4×4 involving only the
sigma-type orbitals which gives two σ ∗ conduction bands and two σ valence bands.
There is no mixing between pi- and sigma-type orbitals. (See Table 5.3.)
π
Table 5.3. Energy bands at M, ~k = 2a (1, 1, 0).
State Rows Type Energy (EM ν ) Eigenvector Wavefunction Sym.

P
M1 2 σ0 Ek |2i √1 ξm |2, mi M 0
N 4
P
M2 10 π0 E⊥ |10i √1 ξm |10, mi M 0
N 5
P
M3 13 π0 E⊥ |13i √1 ξm |13, mi M 0
N 5
) p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M4 π∗ E1 + E 2 + 4(pdπ)2 p1 2(pdπ)|9i + i Et − EM 4 |11i p1 ξm 2(pdπ)|9, mi − Et − EM 4 |11, mi M5
2
9, 11 S1 N S1
p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M5 π E1 − E 2 + 4(pdπ)2 p1 2(pdπ)|9i + i Et − EM 5 |11i p1 ξm 2(pdπ)|9, mi − Et − EM 5 |11, mi M5
2
S1 N S1
) £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M6 π∗ Same as M4 p1 2(pdπ)|12i + i Et − EM 6 |14i p1 ξm 2(pdπ)|12, mi − Et − EM 6 |14, mi M5
12, 14 S1 N S1
£ ¡ ¢ ¤ P £ ¡ ¢ ¤
M7 π Same as M5 p1 2(pdπ)|12i + i Et − EM 7 |14i p1 ξm 2(pdπ)|12, mi − Et − EM 7 |14, mi M5
S1 N S1
 P
M8 π0 E⊥ + 2b √
1 [|7i − |8i] √1 ξm [|7, mi − |8, mi] M4
 2 2N
85

 p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M9 6, 7, 8 π∗ (E1 − b) + (E2 + b)2 + 8(pdπ)2 p1 4(pdπ)|6i + i Et − EM 9 [|7i + |8i] p1 ξm 4(pdπ)|6, mi − Et − EM 9 [|7, mi + |8, mi] M3
 S2 N S2

 p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M10 π (E1 − b) − (E2 + b)2 + 8(pdπ)2 p1 4(pdπ)|6i + i Et − EM 10 [|7i + |8i] p1 ξm 4(pdπ)|6, mi − Et − EM 10 [|7, mi + |8, mi] M3
S2 N S2
 p £ √
¡ ¢ ¤ P £ √
¡ ¢ ¤
M11 σ∗ (E3 + b) + (E4 − b)2 + 6(pdσ)2 p1 2 3(pdσ)|3i + i Ee − EM 11 [|4i − |5i] p1 ξm 2 3(pdσ)|3, mi − Ee − EM 11 [|4, mi − |5, mi] M2

 S3 N S3

 p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
 √ √
M12 σ (E3 + b) − (E4 − b)2 + 6(pdσ)2 p1 2 3(pdσ)|3i + i Ee − EM 12 [|4i − |5i] p1 ξm 2 3(pdσ)|3, mi − Ee − EM 12 [|4, mi − |5, mi] M2
1, 3, 4, 5 S3 N S3
 p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M13 
 σ∗ (E3 − b) + (E4 + b)2 + 2(pdσ)2 p1 2(pdσ)|1i − i Ee − EM 13 [|4i + |5i] p1 ξm 2(pdσ)|1, mi + Ee − EM 13 [|4, mi + |5, mi] M1

 S4 N S4
 p £ ¡ ¢ ¤ P £ ¡ ¢ ¤
M14 σ (E3 − b) − (E4 + b)2 + 2(pdσ)2 p1 2(pdσ)|1i − i Ee − EM 14 [|4i + |5i] p1 ξm 2(pdσ)|1, mi + Ee − EM 14 [|4, mi + |5, mi] M1
S4 N S4
1 1
E1 = 2 (Et + E⊥ ), E2 = 2 (Et − E⊥ ), S1 = (Et − EM ν )2 + 4(pdπ)2 , S3 = 2(Ee − EM ν )2 + 12(pdσ)2 .
1 1
E3 = 2 (Ee + Ek ), E4 = 2 (Ee − Ek ), S2 = 2(Et − EM ν )2 + 16(pdπ)2 , S4 = 2(Ee − EM ν )2 + 4(pdσ)2 .
mx +my
b = (ppσ) − (ppπ), ξm = (−1) , Rm = 2a(mx , my , mz ).
The total real-space wavefunction, for example, for the 14th state is
1 X £ ¡ ¢¤
√ (−1)nx (m)+ny (m) 2(pdσ)|1, mi + (Ee − EM 14 ) |4, mi + |5, mi .
N S4 m
(5.8)
In obtaining (5.8) we used the following:

~ ~ π
eik·Rm = ei 2a (~ex +~ey )·2a(nx~ey +ny ~ex +nz ~ez ) = eiπ[nx (m)+ny (m)]
= (−1)nx (m)+ny (m) ,
~
eik·~τ1 = 1,
~ π
eik·~τ4 = ei 2a (~ex +~ey )·a~ex = eiπ/2 = i,
~ π
eik·~τ5 = ei 2a (~ex +~ey )·a~ey = eiπ/2 = i.
The energies at the M point are shown in Fig. 5.3 for a typical set of energy
bands, and the analytical results are given in Table 5.3.
5.4 Energy bands at R

π
The R points in the Brillouin zone are at 2a (±1, ±1, ±1). We consider the R point
given by ~k = 2a (1, 1, 1). The secular equation is then block-diagonal and consists of
π
a 5×5 and three equivalent 3×3 blocks. The 5×5 block involves only the sigma-type
orbitals and the 3×3 blocks involve only the pi-type orbitals. There is no mixing
between the pi and sigma orbitals at R.
The energies and eigenvectors at the R point are listed in Table 5.4 and the
total real-space wavefunction can be constructed by summing the local eigenvectors
|n, mi over all unit cells and properly normalizing. The total real-space wavefunction
for the 14th state in Table 5.4 is, for example,
1 X
√ (−1)nx (m)+ny (m)+nz (m)
N S3
£ √ ¡ ¢¤
−2 3(pdσ)|3, mi + (Ee − ER14 ) |4, mi − |5, mi . (5.9)
In obtaining (5.9) we used the following results:

~ ~
eik·Rm = (−1)nx (m)+ny (m)+nz (m) ,
~
eik·~τ3 = 1,
~ ~
eik·~τ4 = eik·~τ5 = eiπ/2 = i.
Schematic representations of the wavefunctions are shown in Fig. 5.4(a)–5.4(g).

The splitting of the d bands at R is of interest since this quantity is the band-
π
Table 5.4. Energy bands at R, ~k = 2a (1, 1, 1).
State Rows Type Energy (ERν ) Eigenvector Wavefunction Sym.

 P
R1 π0 E⊥ + 2b √
1 [|7i − |8i] √1 γm [|7, mi − |8, mi] R 0
 2 2N 15
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R2 6, 7, 8 π∗ E1 + E5 p1 4(pdπ)|6i + i Et − ER2 [|7i + |8i] p1 γm 4(pdπ)|6, mi + Et − ER2 [|7, mi + |8, mi] R
250
 S1 N S1
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R3 π E1 − E 5 p1 4(pdπ)|6i + i Et − ER3 [|7i + |8i] p1 γm 4(pdπ)|6, mi + Et − ER3 [|7, mi + |8, mi] R
250
S1 N S1
 P
R4 π0 Same as R1 √
1 [|10i − |11i] √1 γm [|10, mi − |11, mi] R 0
 2 2N 15
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R5 9, 10, 11 π∗ Same as R2 p1 4(pdπ)|9i + i Et − ER5 [|10i + |11i] p1 γm 4(pdπ)|9, mi + Et − ER5 [|10, mi + |11, mi] R
250
 S1 N S1
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R6 π Same as R3 p1 4(pdπ)|9i + i Et − ER6 [|10i + |11i] p1 γm 4(pdπ)|9, mi + Et − ER6 [|10, mi + |11, mi] R
250
S1 N S1
 P
R7 π0 Same as R1 √
1 [|13i − |14i] √1 γm [|13, mi − |14, mi] R 0
 2 2N 15
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R8 12, 13, 14 π∗ Same as R2 4(pdπ)|12i + i [|13i + |14i] 4(pdπ)|12, mi + [|13, mi + |14, mi] R
87
p1 Et − ER8 p1 γm Et − ER8
250
 S1 N S1
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R9 π Same as R3 p1 4(pdπ)|12i + i Et − ER9 [|13i + |14i] p1 γm 4(pdπ)|12, mi + Et − ER9 [|13, mi + |14, mi] R
250
S1 N S1
 P
R10  σ0 Ek − 4b √
1 [|2i + |4i + |5i] √1 γm [|2, mi + |4, mi + |5, mi] R1
 3 3N

 £ ¡ ¢ ¤ P £ ¡ ¢ ¤

R11  σ∗ E3 + E6 p1 6(pdσ)|1i + i Ee − ER11 [2|2i − |4i − |5i] p1 γm 6(pdσ)|1, mi − Ee − ER11 [2|2, mi − |4, mi − |5, mi] R12

 3S2 3N S2
 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
R12 1, 2, 3, 4, 5 σ E3 − E 6 p1 6(pdσ)|1i + i Ee − ER12 [2|2i − |4i − |5i] p1 γm 6(pdσ)|1, mi − Ee − ER12 [2|2, mi − |4, mi − |5, mi] R12
 3S2 3N S2

 £ √
¡ ¢ ¤ P £ √
¡ ¢ ¤
R13 
 σ∗ Same as R11 p1 −2 3(pdσ)|3i − i Ee − ER13 [|4i − |5i] p1 γm −2 3(pdσ)|3, mi + Ee − ER13 [|4, mi − |5, mi] R12

 S2 N S2

 £ ¡ ¢ ¤ P £ ¡ ¢ ¤
√ √
R14 σ Same as R12 p1 −2 3(pdσ)|3i − i Ee − ER14 [|4i − |5i] p1 γm −2 3(pdσ)|3, mi + Ee − ER14 [|4, mi − |5, mi] R12
S2 N S2
1 1
E1 = 2 (Et + E⊥ ) − b, E2 = 2 (Et − E⊥ ) + b, S1 = 2(Et − ERν )2 + 16(pdπ)2 , γm = (−1)nx (m)+ny (m)+nz (m) .
1 1
E3 = 2 (Ee + Ek ) + b, E4 = 2 (Ee − Ek ) − b, S2 = 2(Ee − ERν )2 + 12(pdσ)2 , Rm = 2a[nx (m) + ny (m) + nz (m)].
E52 = E22 + 8(pdπ)2 , E62 = E42 + 6(pdσ)2 , b = (ppσ) − (ppπ).
Figure 5.4. Schematic representation of the real-space wavefunctions of the energy band
states at R in the Brillouin zone. To display the local symmetry, parts of the wavefunction
in neighboring unit cells are shown as well.
5.4 Energy bands at R 89
0
SrTiO3
–2
–4
Energy (eV)
–6
–8
–10
–12
–14
Γ X M R
Figure 5.5. Comparison of LCAO energy bands (thick lines) with LDA calculations [2]
(thin lines) for SrTiO3 . The LCAO parameters used for the fit are Et = –6.52, Ee = –4.52,
E⊥ = –10.95, Ek = –12.10, (pdσ) = –2.35, (pdπ) = 1.60, (ppσ) = –0.05, and (ppπ) = 0.50 (all
in eV) .
theory analog of the ligand-field splitting of d ions. The splitting between the R11
and R2 energies is given by
1³ ´ rh 1 ¡ ¢ i2
∆R (d) = Ee −Et + Ek −E⊥ + 4b + Ee −Ek − 2b +6(pdσ)2
2 2
rh i
1 2
+ (Et − E⊥ + 2b) + 8(pdπ)2 . (5.10)
2
It is readily seen that ∆R (d) depends on the ionic splittings, ∆(d) and ∆(p), and
on the two-center LCAO integrals (pdσ) and (pdπ), which lead to covalent mix-
ing between the cation and the oxygen ions. As mentioned earlier, the splitting
at R is larger than at Γ. For SrTiO3 , ∆(d) is about 2.4 eV and ∆R (d) is about
3.5 eV. For KTaO3 , ∆(d) ' 3.4 eV and ∆R (d) ' 6 eV, and for ReO3 , ∆(d) ' 4 eV
and ∆R (d) ' 7 eV.
In Fig. 5.5, the energy bands for SrTiO3 obtained from LCAO model (in thick
lines) are shown. The band parameters are determined by fitting the important
bands to the results [2] of ab initio density functional calculations in local density
approximation (in thin lines). The π ∗ conduction band, lower part of σ ∗ conduction
band, valence-band width are in good agreement with the local density approxima-
tion (LDA) results.
5.5 Cluster electronic states
Calculations for a small cluster of atoms are used to explore the properties of small
particles. They are also sometimes used to infer the electronic and optical properties
of a solid with similar constituents. It is therefore useful to understand the electronic
states of a cluster and how they relate to the energy band states discussed in the
previous sections.
The local characters of band wavefunctions, when extended to the neighboring
unit cells (Figs 5.1 and 5.4), have a close resemblance to the localized wavefunctions
of a cluster of atoms composed of a B ion surrounded by an octahedron of oxygen
ions.
Figure 5.6. BO6 cluster.
In order to analyze the cluster states, we consider the BO6 cluster shown in
Fig. 5.6. There are 23 basis orbitals for the cluster: three p orbitals on each of the
six oxygen ions and five d orbitals on the central B ion. The orbitals are labeled as
follows:
|1i = dz2 (~r), |2i = pz (~r − a~ez ), |20 i = pz (~r + a~ez ),
|3i = dx2 −y2 (~r), |4i = px (~r − a~ex ), |40 i = px (~r + a~ex ),
|5i = py (~r − a~ey ), |50 i = py (~r + a~ey ),
|6i = dxy (~r), |7i = px (~r − a~ey ), |70 i = px (~r + a~ey ),
|8i = py (~r − a~ex ), |80 i = py (~r + a~ex ), (5.11)
0
|9i = dxz (~r), |10i = px (~r − a~ez ), |10 i = px (~r + a~ez ),
|11i = pz (~r − a~ex ), |110 i = pz (~r + a~ex ),
|12i = dyz (~r), |13i = py (~r − a~ez ), |130 i = py (~r + a~ez ),
|14i = pz (~r − a~ey ), |140 i = pz (~r + a~ey ).
The LCAO matrix elements may be calculated as in the preceding section

with the neglect of interactions beyond second nearest neighbors. The resulting
Hamiltonian, shown in Table 5.5, is a 23×23 matrix.
The matrix of Table 5.5 possesses many zero elements, but straightforward
diagonalization is extremely tedious. The complexity of the problem can be greatly
reduced using well-known group theory methods [3, 4]. By transforming the Hamil-
tonian to a symmetry-coordinates representation the matrix can be reduced to a
block-diagonal form in which the largest blocks are 2×2s. We will not go into all
the details of the group theory method here, but a brief outline of the procedure is
described in the next subsection. Readers unfamiliar with group theory may wish
to skip the next subsection and go directly to Subsection (b).
A more detailed analysis of the BO6 cluster states that includes overlap in-
tegrals as well as the B-ion s orbitals may be found in [5]. The results are not
qualitatively different from those presented here.
(a) Block-diagonalizing the Hamiltonian using symmetry coordinates
In this section we give a brief description of how group theory can be used to
block-diagonalize the matrix of Table 5.5. A central result of group representation
theory is that the matrix elements of a Hamiltonian vanish between functions that
transform according to different IRs (irreducible representations) of the symmetry
group or between functions which transform according to different rows of the same
IR. Thus, if we transform the Hamiltonian matrix to a representation labeled by
basis functions for the IRs of the point group of the cluster, non-zero matrix element
will occur only between the functions which transform according to the same row
of the same IR.
The BO6 cluster possesses Oh symmetry. To determine the nature of the block-
Table 5.5. Hamiltonian matrix for the BO6 cluster.

dz2 dx2 dxy dxz dyz 2 20 4 40 5 50 7 70 8 80 10 100 11 110 13 130 14 140
dz 2 Ee 2s −2s −s s −s s
dy 2 Ee t −t −t t
dxy Et d −d d −d
dxz Et d −d d −d
dyz Et d −d d −d
2 2s Es −v v −v v u u u u
20 −2s Es v −v v −v u u u u
4 −s t −v v Es −v v u u u u
40 s −t v −v Es v −v u u u u
5 −s −t −v v −v v Es u u u u
50 s t v −v v −v Es u u u u
7 d u u Ep −v v p p
70 −d u u Ep v −v p p
8 d u u −v v Ep p p
0
8 −d u u v −v Ep p p
10 d u u p p Ep −v v
100 −d u u p p Ep v −v
11 d u u −v v Ep p p
110 −d u u v −v Ep p p
13 d u u p p Ep −v v
130 −d u u p p Ep v −v
14 d u u p p −v v Ep
140 −d u u p p v −v Ep
Ep = E⊥ , d = (pdπ), s = (pdσ)/2, u = [(ppσ) + (ppπ)]/2 = c/2.

√
Es = Ek , p = (ppπ), t = 3(pdσ)/2, v = [(ppσ) − (ppπ)]/2 = b/2.
diagonal form of the Hamiltonian we need only know the numbers and types of IRs
that will occur in the solution of the cluster problem. That information can be
determined by decomposing the representation based on the 23 cluster orbitals (5 d
orbitals plus 18 p orbitals) into the IRs of Oh point group. Using the character table
for the Oh group and the calculated characters for the cluster–orbital representation
we arrive at the results shown in (5.12a-e) and in the table that follows:
Γ (orbital) = Γ (B) + Γ (oxygens) (5.12a)

Γ (B) = eg + t2g (5.12b)
Γ (oxygens) = a1g + eg + t1g + t2g + t2u + 2t1u (5.12c)
Γ (orbital) = a1g + 2eg + t1g + 2t2g + t2u + 2t1u . (5.12d)
In the equations (5.12), Γ (B) is the representation based on the five d orbitals of
the B ion, Γ (oxygens) is the representation based on the 18 p orbitals of the six
a1g eg t1g t2g t2u t1u

Number of states
that transform 1 4 3 6 3 6
as the IR (5.12e)
Dimensions of the
block-diagonalized (1×1) 2(2×2) 3(1×1) 3(2×2) 3(1×1) 3(2×2)
Hamiltonian
oxygens. The Oh irreducible representations (IRs) have the following dimensions:

a1g is one-dimensional, eg is two-dimensional, and all others in (5.12c) are three-
dimensional representations. The subscript ‘g’ (gerade) indicates basis functions
that are invariant under inversion through the origin (the center of the cluster) and
‘u’ (ungerade) indicates functions that are antisymmetric. Since there can be no
non-zero matrix elements of the Hamiltonian between different rows of the same
IRs, the dimensionalities of the blocks in the block-diagonalized Hamiltonian will
be determined solely by the decomposition coefficients (i.e., the number of times an
IR is contained in the orbital representation). The number and size of the blocks
in the transformed Hamiltonian are shown in (5.12e). For example, eg is a two-
dimensional representation. It occurs twice in the decomposition formula, once in
Γ (B) and once in Γ (oxygens). Therefore, the block-diagonalized matrix will have
a 2×2 which involves the first-row eg d orbitals with the first-row eg p orbital
combination. The second 2×2 involves the second-row IR basis functions for the p
and the d orbitals. Each of the 2×2 blocks will yield the same two eigenvalues, E
and E 0 , even though the eigenvectors of the 2×2 blocks are different. Thus, the eg
solutions consist of two pairs of doubly degenerate states. That is, two states with
eigenvalues E and two states with eigenvalues E 0 .
To calculate the block-diagonalized Hamiltonian we need to find combinations
of the cluster orbitals that form bases for the IRs of Oh that appear in the de-
composition. Group theory provides a method for doing this using that are called
projection operators [3, 4]. A brief description of the process is as follows. A set of
cluster orbitals is selected as a “trial” function. To generate a function which can
be used as the basis for the nth row of the jth IR, the trial function is subjected
to an operation of the group (a rotation, for example) and the resulting function
multiplied by the (n, n) diagonal element of the jth IR matrix representing that
operation. The sum obtained by repeating this procedure for all of the operations
of the group is either a null function or a function that transforms according to the
nth row of the jth IR. (Although it is usually not necessary, the trial function can
be taken as the sum of all of the cluster orbitals and that guarantees that a null
function will not be generated for any IR contained in the orbital representation.)
The orbital basis functions generated in this way are often referred to as symmetry
coordinates or symmetry functions. A particular eigenfunction of the Hamiltonian

will always be a linear combination of symmetry coordinates which belong to the
same row of the same IR. The collection of symmetry coordinates, when normalized,
can be used to form a unitary matrix, U, which will block-diagonalize the cluster
Hamiltonian matrix; that is,
U† HU = H0 (a block diagonalized matrix).
The decomposition (5.12b) shows that the d orbitals are invariant under in-
version. Thus, the antisymmetric u states (ungerade) are composed entirely of p
orbitals. This means that the u states are non-bonding. That is, states that do not
contribute to the B–O chemical bond. Also, it is clear that the bonding and anti-
bonding states will have eg or t2g symmetries because these are the only symmetries
that can produce an eigenstate with both p and d orbitals. Furthermore, the basis
functions of the IRs (a1g , t1g , and t2u ) which occur with a coefficient of unity in the
decomposition equation of (5.12c) must be cluster eigenfunctions since they can
have no matrix elements with any other symmetry coordinate. Thus, group theory
gives the eigenfunctions for seven of the 23 cluster states. These eigenfunctions are
independent on the values of the parameters that enter into the Hamiltonian matrix
elements appearing in Table 5.5.
The 23 symmetry coordinates for the BO6 cluster are given in Table 5.6.
(b) Calculation of cluster states
The Hamiltonian matrix in Table 5.5 can be block-diagonalized by transforming it

to the representation based on the symmetry coordinates listed in Table 5.6. The
matrix, U, which accomplishes this is given in Table 5.7.
(c) Types of cluster states
The block-diagonalized Hamiltonian is shown in Tables 5.8 and 5.9. The types
of states for the BO6 cluster that result are briefly discussed in this section. For
convenience we denote a cluster-state eigenfunction by CN (N = 1, 2, . . ., 23) and
the corresponding eigenvalue by ECN .
The a1g state
The a1g state is a totally symmetric combination of pk orbitals that form a non-
bonding cluster state. The symmetry coordinate, S(1), is the eigenfunction of the
Table 5.6. Symmetry coordinates for the BO6 cluster.
Name IR Row Symmetry coordinates
S(1) a1g 1 √1 [(|2i − |20 i) + (|4i − |40 i) + (|5i − |50 i)]∗

6
S(2) eg 1 |1i
1
S(3) eg 1 √
2 3
[2(|2i − |20 i) − (|4i − |40 i) − (|5i − |50 i)]
S(4) eg 2 |3i
1
S(5) eg 2 2 [(|4i − |40 i) − (|5i − |50 i)]
S(6) t2g 1 |6i

1
S(7) t2g 1 2 [(|7i − |70 i) + (|8i − |80 i)]
S(8) t2g 2 |9i
1
S(9) t2g 2 2 [(10i − |100 i) + (|11i − |110 i)]
S(10) t2g 3 |12i
1
S(11) t2g 3 2 [(|13i − |130 i) + (|14i − |140 i)]
1
S(12) t1g 1 2 [(|11i − |110 i) − (|10i − |100 i)]∗
1
S(13) t1g 2 2 [(|14i − |140 i) − (|13i − |130 i)]∗
1
S(14) t1g 3 2 [(|8i − 80 i) − (|7i − |70 i)]∗
S(15) t1u 1 √1 [|2i + |20 i]

2
1
S(16) t1u 1 2 [(|11i + |110 i) + (|14i + |140 i)]
S(17) t1u 2 √1 [(|4i + |40 i)]
2
1 0
S(18) t1u 2 2 [(|7i + |7 i) + (|10i + |100 i)]
1
S(19) t1u 3 2 [|5i + |50 i]
1
S(20) t1u 3 2 [(|8i + |80 i) + (|13i + |130 i)]
1
S(21) t2u 1 2 [(|7i + |70 i) − (|10i + |100 i)]∗
1
S(22) t2u 2 2 [(|13i + |130 i) − (|8i + |80 i)]∗
1
S(23) t2u 3 2 [(|14i + |140 i) − (|11i + |110 i)]∗
∗
Indicates that the symmetry function is also an eigenfunction.
Table 5.7. Symmetry coordinate transformation matrix, U.

dz 2 dx2 dxy dxz dyz 2 20 4 40 5 50 7 70 8 80 10 100 11 110 13 130 14 140
dz 2 1
dy 2 1
dxy 1
dxz 1
dyz 1
2 q 2r s
20 −q −2r s
4 q −r t s
40 −q r −t s
5 q −r −t s
50 −q r t s
7 t −t t t
70 −t t t t
8 t t t −t
80 −t −t t −t
10 t −t t −t
100 −t t t −t
11 t t t −t
110 −t −t t −t
13 t −t t t
130 −t t t t
14 t t t t
140 −t −t t t
1 1 1 1
q= √ , r= √ , s= √ , t= 2.
6 2 3 2
Hamiltonian with energy, EC1 :
EC1 = Ek − 2b,
C1 = S(1). (5.13)
The eg states
The eg states are bonding and antibonding states involving admixtures of eg -type
d orbitals with pk orbitals. The two 2×2 blocks in Table 5.8 yield identical pairs of
energies, EC2 and EC3 where
rh i2
1 1
EC2 = (Ee + Ek + b) + (Ee − Ek − b) + 3(pdσ)2 ,
2 2
rh i2
1 1
EC3 = (Ee + Ek + b) − (Ee − Ek − b) + 3(pdσ)2 , (5.14)
2 2
EC4,5 = EC2,3 .
Table 5.8. Block-diagonalized g-cluster states.
a1g Ek − 2b
 √

 Ee 3(ppσ)

 √

 3(ppσ) Ek + b


eg
 √

 Ee 3(ppσ)

 √

 3(ppσ) Ek + b



 Et 2(pdπ)

97



 2(pdπ) E⊥ + b






 Et 2(pdπ)
t2g

 2(pdπ) E⊥ + b







 Et 2(pdπ)



 2(pdπ) E⊥ + b


 E⊥ + b





t1g E⊥ + b





 E⊥ + b

Table 5.9. Block-diagonalized u-cluster states.
 √

 Ek 2c



 √

 2c E⊥ +2(ppπ)



 √
Ek 2c
t1u √

 2c E⊥ +2(ppπ)



 √

 Ek 2c



 √
2c E⊥ +2(ppπ)


 E −2(ppπ)
 ⊥
t2u E⊥ −2(ppπ)


 E⊥ −2(ppπ)
C2 and C4 are antibonding states and C3 and C5 are bonding states. The
eigenfunctions are given below:
(row 1 eigenfunctions):
n√ o
C2, 3 = 3(pdσ)S(2) − (Ee − EC2,3 )S(3) /NC2,3 (5.15)
q
NC2,3 = (Ee − EC2,3 )2 + 3(pdσ)2 (5.16)
(row 2 eigenfunctions):
n√ o
C4, 5 = 3(pdσ)S(4) − (Ee − EC2,3 )S(5) /NC2,3 . (5.17)
The t2g states
The t2g states are bonding and antibonding states involving admixtures of t2g -type
d orbitals with p⊥ orbitals. The three identical 2×2 matrices in Table 5.8 yield a
pair of triply degenerate energies, EC6 and EC7 , where
rh i2
1 1
EC6,7 = (Et + E⊥ − b)± (Et − E⊥ + b) + 4(pdπ)2 , (5.18)
2 2
EC8,9 = EC10,11 = EC6,7 . (5.19)
The six eigenfunctions are given by

n 1 o
C6, 7 = 2(pdπ)S(6) + (Et − EC6,7 )S(7) /NC6,7 (5.20)
2
n 1 o
C8, 9 = 2(pdπ)S(8) + (Et − EC6,7 )S(9) /NC6,7 (5.21)
2
n 1 o
C10, 11 = 2(pdπ)S(10) + (Et − EC6,7 )S(11) /NC6,7 (5.22)
rh 2
1 i2
NC6,7 = (Et − EC6,7 ) + 4(pdπ)2 . (5.23)
2
The t1g states
The t1g states are non-bonding combinations of p⊥ orbitals. They form a triply
degenerate set with energy,
EC12,13,14 = E⊥ + b, (5.24)
C12 = S(12), (5.25)
C13 = S(13), (5.26)
C14 = S(14). (5.27)
The eigenfunctions are the same as the symmetry coordinates.
The t1u states
The t1u states are non-bonding combinations of pk orbitals, and p⊥ orbitals. They
form a pair of triply degenerate states (Table 5.9). The energies are given by
1£ ¤ q£ ¤2
EC15,16 = Ek + E⊥ + 2(ppπ) ± Ek − E⊥ − 2(ppπ) + 2c2 . (5.28)
2
The eigenfunctions are:

n√ o
C15, 16 = 2cS(15) − (Ek − EC15,16 )S(16) /NC15,16 (5.29)
n√ o
C17, 18 = 2cS(17) − (Ek − E15,16 )S(18) /NC15,16 (5.30)
n√ o
C19, 20 = 2cS(19) − (Ek − E15,16 )S(20) /NC15,16 (5.31)
q
NC15,16 = 2c2 + (Ek − EC15,16 )2 . (5.32)
The t2u states
The t2u states are non-bonding combinations of p⊥ orbitals. They form a triply
degenerate set with energy,
EC21,22,23 = E⊥ − 2(ppπ), (5.33)

C21 = S(21), (5.34)
C22 = S(22), (5.35)
C23 = S(23). (5.36)
The t2u eigenfunctions are the same as the symmetry coordinates.

A schematic diagram of the cluster states is shown in Fig. 5.7. The ordering of
the states is dependent upon the values of the LCAO parameters.
0
eg (C2)
–2
t2g (C6)
–4
d
–6
Energy (eV)
–8
t2u (C21)
t1u (C15)
–10
t1g (C12)
p
a1g (C1)
–12 t2g (C7)
t1u (C16)
–14
eg (C3)
–16
Figure 5.7. Cluster states for the following parameters (in eV): Ee = −4.0, Et = −6.0,
Ek = −12.0, E⊥ = −10.0, (pdπ) = −2.0, (pdσ) = −4.0, (ppπ) = −0.5, (ppσ) = −1.0. The
levels are at: a1g = −11.0, eg = −1.02 and −15.48, t2g = −3.38 and −12.12, t1g = −10.5,
t1u = −9.32 and −13.68, t2u = −9.0, all in eV.
In some clusters or analogs of perovskite metals the states are completely

occupied below the t2g levels, but have only one or two electrons in the triply de-
generate t2g levels. In this case the cluster ground state is electronically degenerate
since several arrangements of the electrons in the t2g levels have the same energy.
According to the Jahn–Teller theorem [6] if a nonlinear molecule’s ground state
is electronically degenerate the molecule must distort to lower its symmetry and
energy. For example, a compression along the z-axis of the BO3 cluster would lower
the symmetry from cubic to tetragonal. It would also raise the energies of the xz
and yz levels, but lower the xy level. The xy level could then accommodate one
or two electrons. The total energy of the completely filled states below t2g is often
nearly unchanged by the perturbation. Some levels move up, and some down in
such a way that “energy center of gravity” is nearly unchanged. Therefore, the new
ground state has a lower total energy than that for the cubic case.
(d) Comparison of cluster states with energy band states
A reasonable correspondence between an LCAO energy band states and the BO6
cluster states can be established by inspecting the symmetry and composition of
the eigenfunctions. In Fig. 5.4 the local symmetries of the band wavefunctions are
made evident by showing the orbitals in a unit cell and also those in neighboring
cells which would belong to a BO6 cluster. We shall refer to this combination of
orbitals as the local band function. In what follows we shall show that there is a
one-to-one correspondence between the cluster eigenfunctions and the local band
functions.
The energy-band wavefunctions possess full Oh symmetry at Γ and R and
therefore we look for correlations with the cluster states at these points in the
Brillouin zone. In general, the u-cluster states will be correlated with band states
at Γ and the g-cluster states correlated with band states at R. To understand this
assignment consider a p orbital, p(~r −a~eα ), involved in an energy-band wavefunction
in a particular unit cell. Its partner orbital, p(~r +a~eα ), will be located in an adjacent
unit cell and therefore the amplitude will differ by a phase factor of ei(2kα a) . This
phase factor is +1 at Γ and –1 at R. For the u-cluster states p(~r −a~eα ) is always
combined with +p(~r +a~eα ) to form a function that is antisymmetric under inversion.
For the g states, p(~r −a~eα ) is always combined with –p(~r +a~eα ) to form a function
that is symmetric under inversion. Therefore the u states will correlate with energy-
band states at Γ and the g-cluster states will correlate with energy-band states at
R.
The eigenfunctions of the seven cluster states, C1, C12, C13, C14, C21, C22,
and C23 are the symmetry coordinates S(1), S(12), S(13), S(14), S(21), S(22), and
S(23), respectively. To illustrate how a correlation with a band state can be estab-
lished consider the energy band state in Table 5.4 labeled R10. The wavefunction
for the unit cell is proportional to [|2i + |4i + |5i]. The partner p orbitals (the p0
orbitals), which lie in adjacent unit cells, will have the negative of these amplitudes
because the phase factor ei(2kα a) = −1 at R. Combining unit-cell orbitals with their
partner orbitals yields the local band function for R10 that is given by,
[|2i − |20 i + |4i − |40 i + |5i − |50 i] (R10 local band function) . (5.37)
The function in (5.37) is proportional to the symmetry coordinate S(1). Accord-

ing to (5.14) the eigenfunction for the C1 cluster state is S(1) (a1g symmetry).
Therefore, except for a constant normalization factor, the C1 cluster-state eigen-
function is identical to the local band function for the R10 energy band state. In a
similar fashion we can show that the local band functions for R4, R7, and R1 are
proportional to the symmetry coordinates S(12), S(13), and S(14) and hence are,
within a constant factor, the same as the eigenfunctions for the cluster states (t1g
symmetry) C12, C13, and C14, respectively.
Next, consider the band state, Γ9 . The unit-cell eigenfunction is proportional
to |7i − |10i so the local band function is
|7i + |70 i − |10i − |100 i, (5.38)
since the phase factor ei(2kα a) = +1 at Γ. The function in (5.38) is proportional to

the t2u symmetry coordinate, S(21). Thus, the local band function for the energy
band state Γ9 is, apart from a constant, the same as the eigenfunction for the cluster
state C21. Similarly, Γ12 , and Γ6 have local band functions that are proportional
to the symmetry coordinates S(22) and S(23), respectively. That is, the local band
functions for Γ12 , and Γ6 are the same as the eigenfunctions for the cluster states
C22 and C23 (t2u symmetry).
The remaining cluster states have wavefunctions that are admixtures of two
different functions that transform according to the same row of the IR representa-
tion. As a last example, consider the band states in Table 5.4 labeled R11 and R12.
The local band functions are constructed as before except that we need to include
an internal phase factor ei(2kα a) = i for the p orbitals relative to the d orbital. The
orbitals of the eigenfunction in the unit cell take the form
6(pdσ)|1i − (Ee − ER11,12 ) [2|2i − |4i − |5i] ,
so the local band function is

√
6(pdσ)|1i − 2 3(Ee − ER11,12 ) [2|2i − 2|20 i − |4i + |40 i − |5i + |50 i]
√
∝ 3(pdσ)S(2) − (Ee − ER11,12 )S(3). (5.39)
Comparison of this result with the eg cluster eigenvectors in (5.15) and (5.17) shows
that the band states R11 and R12 are identical in form to the cluster states C2
and C3. They differ only by the eigenvalues that enter into the coefficient of S(3).
The band state R11 (R12) involves ER11 (ER12 ) and the cluster state C2 (C3)
involves the energy EC2 (EC3 ). In general, the band energies are not the same as
the cluster energies so the eigenfunctions are different. However, the wavefunctions
are identical with respect to their symmetry properties. We discuss the energy
differences between the cluster states and the energy band states at the end of this
section.
The correlations between the energy-band states and the cluster states is sum-
marized as follows:
C1 (a1g ) → R10 ( σ 0 valence band at R)
C2, 4 (eg ) → R11, 13 ( σ ∗ conduction band at R)
C3, 5 (eg ) → R12, 14 ( σ valence band at R)
C6, 8, 10 (t2g ) → R2, 5, 8 ( π ∗ conduction band at R)
C7, 9, 11 (t2g ) → R3, 6, 9 ( π valence band at R) (5.40)
0
C12, 13, 14 (t1g ) → R1, 4, 7 ( π valence band at R)
C15, 17, 19 (t1u ) → Γ7, 10, 13 ( π + σ valence band at Γ)
C16, 18, 20 (t1u ) → Γ8, 11, 14 ( π + σ valence band at Γ)
C21, 22, 23 (t2u ) → Γ6, 9, 12 ( π 0 valence band at Γ)
where the arrow indicates that the states have identical symmetry properties.
Figure 5.8 shows the locations of the energy-band states which correlate with
the cluster state symmetries.
As we have seen, the cluster eigenstates can be correlated with energy-band
wavefunctions at Γ or R in the Brillouin zone. The eigenvalues of the cluster states
are also closely related to the energy band eigenvalues, but they are clearly different.
For example, the eigenvalue of the cluster state, C1, is Ek − 2b, but the energy
of the correlated band state, R10, is Ek − 4b. Similarly, the energy of ER11 (see
Table 5.4) involves a term 6(ppσ)2 , whereas for the correlated cluster state, C2,
the corresponding factor is 3(ppσ)2 . These differences are due to the fact that an
oxygen ion in the cluster interacts with only one B ion while in the solid it interacts
with two B ions. Also each oxygen interacts with twice as many other neighboring
oxygen ions in the solid as in the cluster. If the following replacements
(ppπ) → 2(ppπ),
(ppσ) → 2(ppσ),
√
(pdσ) → 2(pdσ),
√
(pdπ) → 2(pdπ)
are made in the expressions for the cluster energies to correct for the difference in the
R11, R13 (eg )
R2, R5, R8 (t2g )
(t1u ) Γ13, Γ10, Γ7 R10 (a1g )

R1, R4, R7 (t1g )
(t2u ) Γ12, Γ9, Γ6
(t1u ) Γ14, Γ11, Γ8
R3, R6, R9 (t2g )
R12, R14 (eg )
Γ X M R
Figure 5.8. Symmetries of energy band states and of corresponding cluster states (in
parentheses).
numbers of interactions, the cluster energies become identical to the corresponding

band energies.
In summary, there are remarkable similarities between the electronic states of
the BO6 cluster and those of the corresponding perovskite solid. However, there are
also significant differences that must be borne in mind. For an insulating perovskite
(unoccupied conduction bands) the usual, direct band gap is the energy difference
between the top of the valence band at Γ (t1u in the Fig. 5.8 or t2u in Fig. 5.7) and
the bottom of the π ∗ conduction band at Γ. However, the cluster model does not
have a state corresponding to the bottom of the π ∗ (or σ ∗ ) conduction band at Γ.
The closest analog to the band gap for the cluster is the energy difference between
the t2g state a R and the t1u state at Γ. Consequently, the analogs of the electronic
transitions involved to the onset of optical absorption are not represented in the
cluster model. Furthermore, the π, π ∗ , σ, and σ ∗ band states at Γ are composed
entirely of d orbitals. There are no analogs of these states in the cluster model. It
is sometimes argued that the cluster levels should occur roughly at the “center of
gravity” of the corresponding energy bands of the solid. This suggestion is certainly
not valid for the perovskites. As will be discussed in Chapter 6, the energy bands of
the perovskites possess critical singularities in the electronic density of states that
are responsible for structure in the optical and electronic properties. These impor-
tant characteristics can not be accounted for with a cluster model. One conclusion
of the above analysis is the cluster model can not give even a rough idea of the
electronic properties of the solid. Conversely, the band model can not give even a
rough idea of the electronic properties of an actual cluster particle. Clearly, such
particles can possess electronic and optical properties that are quite different from
those of the corresponding solid.
References
[1] L. B. Bouckaert, R. Smoluchowski, and E. P. Wigner, Phys. Rev. 50, 65 (1936).
[2] E. Mete, R. Shaltaf, and Ş. Ellialtıoğlu, Phys. Rev. B 68, 035119 (2003).
[3] E. P. Wigner, Group theory and its application to the quantum mechanics of
atomic spectra (New York and London, Academic Press, 1959).
[4] D. C. Harris and M. D. Bertolucci, Symmetry and spectroscopy (New York,
Dover, 1989).
[5] T. Wolfram, R. Hurst, and F. J. Morin, Phys. Rev. B 15, 1151 (1977).
[6] H. A. Jahn and E. Teller, Proc. Roy. Soc. A 161, 220 (1937).
1. Derive the energies and wavefunctions for the π and π ∗ bands at Γ and R given in
Tables 5.1 and 5.4.
Hints: (i) Rearrange rows and columns of the eigenvalue matrix to achieve a block-
diagonal form. (ii) Use unit-cell symmetry coordinates to transform the 5×5 to a 1×1
and two 2×2s.
2. Calculate the orbital amplitudes for the eigenvector of the state R11 , using the LCAO
parameters given in Fig. 5.7.
3. (a) Explain why there are no BO6 cluster states corresponding to the pure d-orbital
energy band states at Γ. (Hint: refer to Problem 7 in Chapter 1.)
(b) What symmetry does the infinite cubic array have that the BO6 cluster lacks?
4. Using the symmetry coordinates of Table 5.6 derive the block-diagonals shown in
Table 5.8 for the a1g and t2g states.
5. (a) Calculate the energies of the t2g (C6) and t2u (C21) states using the parameters of
Fig. 5.7.
(b) Compare the energy difference [E(C6) − E(C1)] with (Eπ∗ − Eπ ) at Γ.
6
Density of states
The density of states (DOS) of a solid is of fundamental importance in determining

its electronic, optical, and transport properties. The DOS of an ideal perovskite is
unusual in that it possesses peculiar structures called van Hove singularities which
are associated with the two-dimensional character of the pi bands and portions of
the Brillouin zone where the sigma bands are flat. The effects of these singularities
show up in the optical reflectivity and absorption and in the photoelectron energy
distributions observed for perovskites.
In this chapter we determine the DOS functions and use them to calculate the
Fermi surface and effective masses for the d-band perovskites.
6.1 Definitions
A solid consisting of N unit cells will have ns × N electronic states, where ns is

the number of basis orbitals per unit cell. The electronic states are characterized
by a wavevector ~ki (i = 1, 2, . . . , N ) and a band index ν (ν = 1, 2, . . . , ns ). For an
infinite crystal (N → ∞) the ~ki ’s form a dense set. One simple way of choosing
the set of ~ki is to divide the Brillouin zone into N cells of equal volume and take
the ~k-vectors which lie at the center of each of the N cells. As N → ∞, each cell
becomes a differential volume element in ~k-space with volume d~k = Ω/N where Ω
is the volume of the first Brillouin zone. In this limit we may use the relation
N Z
1 X 1
f~ =⇒ d~k fν (~k)
N i=1 ki ν Ω BZ
N →∞
where fν (~k) is a continuous function with value f~ki ν at ~k = ~ki and the integral is
over the first Brillouin zone.
We define the density of (electronic) states, ρ(E) by the condition that the
quantity ρ(E)∆E is the number of states per unit cell in the energy range between
107
108 Density of states
E and E + ∆E. If we consider one of the N cells of the first Brillouin zone with
wavevector ~ki then the number of energy band eigenvalues corresponding to ~ki with
energy less than E + ∆E is given by
ns
X ³ ´
Θ E + ∆E − E~ki ν , (6.1)
ν=1
where Θ(x) is the unit step function; Θ(x) = 1 for x > 0 and is zero otherwise.
Similarly, the number of eigenvalues with energy less than E is
ns
X ³ ´
Θ E − E~ki ν . (6.2)
ν=1
It is then obvious that the number of eigenvalues corresponding to ~ki which lie
in the energy range between E and E + ∆E is just the difference between (6.1)
and (6.2). If we sum this difference over all of the ~ki -vectors we will have the total
number of states with energies between E and E + ∆E. Thus,
ns X n ³ ´ ³ ó
1 X
ρ(E) ∆E = Θ E + ∆E − E~ki ν − Θ E − E~ki ν , (6.3)
N ν=1 i
where the 1/N factor is introduced to give the number of states per unit cell.
In the limit as ∆E → 0 the difference term on the right-hand side of (6.3) can
be replaced by
d n o
Θ(E − E~ki ν ) ∆E, (6.4)
dE
so that
ns X
1 X d
ρ(E) = Θ(E − E~ki ν ) . (6.5)
N ν=1 i dE
The step function Θ may be written as

Z E
Θ(E − E~ki ν ) = δ(t − E~ki ν ) dt (6.6)
−∞
where δ(t − E~ki ν ) is the delta function with its singularity at t = E~ki ν . (A detailed
discussion of the delta function and properties of the DOS functions is given in
Appendix B.) From (6.6) it is apparent that
d
Θ(E − E~ki ν ) = δ(E − E~ki ν ) . (6.7)
dE
Thus, we arrive at the result that
ns X
1 X
ρ(E) = δ(E − E~ki ν ). (6.8)
N ν=1 i
Next, we need to pass to the limit as N → ∞ and convert the sum over ~ki into an
integral. This requires that we weigh each term by the volume it represents in ~k-
space. Each ~ki represents a volume Ω/N = ∆~k, where ∆~k is the differential volume
element in ~k-space. Thus,
ns Z
1 X
ρ(E) = d~k δ(E − E~kν ), (6.9)
Ω ν=1
where the integral is over the volume of the first Brillouin zone. It is clear from
(6.9) that
Z ∞
ρ(E) dE = ns . (6.10)
−∞
It is often convenient to work with other density functions. For example, if

f (E) is a single-valued function of the energy, then
Z
1
ρ(f )ν = d~k δ[f (E) − f (E~kν )] (6.11)
Ω
ns
X
ρ(f ) = ρ(f )ν .
ν=1
The relation between these densities is

¯ ¯
¯ df ¯
¯
ρ(E) = ρ(f ) ¯ ¯. (6.12)
dE ¯
In utilizing (6.9) or (6.11) the δ function can be represented as follows:

½ ¾
1 1
δ(x − x0 ) = − Im
π x − x0 + i0+
where 0+ is a positive infinitesimal and Im{· · ·} indicates the imaginary part of the
expression within the brackets. Thus, we have that
Z ½ ¾
1 1 1
ρ(f )ν = − Im d~k , (6.13a)
π Ω f (E) − f (E~kν ) + i0+
Z ½ ¾
1 |df /dE| 1
ρ(E)ν = − Im d~k . (6.13b)
π Ω f (E) − f (E~kν ) + i0+
6.2 DOS for the pi bands
In Chapter 4 we derived simple analytical expressions for the 14 energy bands of

a d-band perovskite for the approximation (ppσ) = (ppπ) = 0. In this section we
make use of these formulae for the energy bands to derive the DOS functions.
The energy bands are classified as pi and sigma bands and DOS expressions
can be derived for each type separately. We begin by considering the pi bands.
There are nine pi bands described by
E~kπ0 (αβ) = E⊥ (non−bonding bands), (6.14)

rh i2 ³ ´
1 1
E~kπ? (αβ) = (Et + E⊥ )+ (Et − E⊥ ) + 4(pdπ)2 Sα2 + Sβ2 (6.15)
2 2
rh i2 ³ ´
1 1
E~kπ(αβ) = (Et + E⊥ )− (Et − E⊥ ) + 4(pdπ)2 Sα2 + Sβ2 (6.16)
2 2
(αβ = xy, xz, or yz).
We can express both equations (6.15) and (6.16) in the dimensionless form:
£ ¤2 £ ¤2
E~kν − 21 (Et + E⊥ ) − 21 (Et − E⊥ )
− 2 = − (C2α + C2β ) (6.17)
2(pdπ)2
where C2α ≡ cos( 2kα a) and C2β ≡ cos( 2kβ a). If (6.17) is solved for E~kν , the two
solutions are E~kπ∗ and E~kπ . This strange form turns out to be a mathematically
convenient function for investigating the density of states of the π ∗ and π bands.
Rather than working with the actual energies, E~kπ∗ and E~kπ , it is much easier to
calculate the density of state of the function on the left-hand side of (6.17). To do
this we define the dimensionless function
£ ¤2 £ ¤2
~ E~kν − 21 (Et + E⊥ ) − 12 (Et − E⊥ )
επ (k) ≡ −2 (6.18)
2(pdπ)2
so that (6.17) is given by
επ (~k) = − (C2α + C2β ) . (6.19)
Now using (6.13a) we calculate the density, ρ(επ ), defined by

Z
1 1 d~k
ρ(επ ) = − Im
π Ω επ − επ (~k) + i0+
µ ¶3 Z Z Z π/2a
1 2a dkα dkβ dkγ
= − Im +
(6.20)
π 2π −π/2a ε π + (C 2α + C2β ) + i0
where we have used the fact that Ω = (2π/2a)3 for the cubic perovskites, with a
being the B–O distance.
The integration over dkγ may be performed immediately and then introducing
the variables x = 2kα a and y = 2kβ a we find
µ ¶2 Z π Z π
4 1 1
ρ(επ ) = − Im dx dy . (6.21)
π 2π 0 0 ε π + Cx + Cy + i0+
We proceed as follows:
Z π Z π
dy
Im +
= −π dy δ[(επ + Cx ) + Cy ] (6.22)
0 επ + Cx + Cy + i0 0
Z π
d(− Cy )
= −π δ[(επ +Cx )−(− Cy )]
0 Sy
Θ[1 − (επ + Cx )2 ]
= −π p . (6.23)
1 − (επ + Cx )2
The Θ function ensures that the value of επ is such that the δ function is satisfied
in the integration range. Thus, we find that
Z π
1 Θ[1 − (επ + Cx )2 ]
ρ(επ ) = 2 p dx . (6.24)
π 0 1 − (επ + Cx )2
The Θ function is non-zero only when επ lies within the pi bands; that is
when |επ (k)| ≤ 2. From (6.24) it is apparent that when |επ | > 2, then Θ[1 − (επ +
Cx )2 ] = 0 at all points in the range of the integration. Thus, the density of states
vanishes for επ outside of the range of the π or π ∗ bands, as of course it must. We
may then confine our attention to the range |επ | ≤ 2. We make the substitution
z = επ + Cx and obtain
Z b
1 dz
ρ(επ ) = 2 p , (6.25)
π c (z − a)(z − b)(z − c)(z − d)
where a > b > c > d. For −2 ≤ επ < 0, we have a = 1, b = 1 + επ , c = −1, and
d = επ − 1. For 0 ≤ επ < 2, we have a = 1 + επ , b = 1, c = επ − 1, and d = −1. The
required integral is given by Gradshteyn and Ryzhik [1] (see 3.147.4) with the result
that
1 ³p ´
ρ(επ ) = 2 K 1 − (επ /2)2 Θ[1 − (επ /2)2 ] (6.26)
π
where K(k) is the complete elliptic integral of the first kind:
Z π/2
dθ
K(k) = p . (6.27)
0 1 − k 2 sin2 θ
A graph of ρ(επ ) is shown in Fig. 6.1(a). It is seen that ρ(επ ) has jump discontinuities
(designated as P0 and P2 critical points) at the band edges (επ = ±2). The peak
at επ = 0 is a logarithmic singularity (a P1 critical point). The discontinuities arise
from the fact that the π ∗ and π bands depend only on two components kα and kβ ,
of the three-dimensional k-vector. They therefore behave as two-dimensional energy
bands and have constant energy with variation of kγ . The logarithmic singularity
occurs at επ = 0 and arises from the saddle points in the energy band dispersion
near the X-points in the Brillouin zone.
(a) (b)
2 P2
1
επ (~k)
0
P1
–1
–2 P0
Γ X M R 1/2π ρ(επ )
(c) (d)
π ∗
Et
Energy
E⊥
Γ X M R ρπ (E)
Figure 6.1. Density of states functions for the pi bands. (a) The three reduced bands
επ (~k) for α, β, γ = x, y or, z; α 6= β, and (b) the corresponding reduced density of states
ρ(επ ) with P0 , P1 , and P2 being van Hove singularities related to the critical points in 2D.
(c) Energy bands Eπ (~k) and (d) the corresponding density of states ρπ (E) in E-space.
The function επ (~k) defined by (6.17) and (6.18) describe both the π ∗ conduction
bands and the π valence bands.
To convert to the density of states in energy space, ρ(E), we employ (6.13b)
1
and (6.18) to find that

¯ ¯
1 ¯ E − 12 (Et + E⊥ ) ¯¯ ³p ´
ρπ (E) = 2 ¯¯ 2 ¯ K 1 − (επ (E)/2)2 Θ[1 − (επ (E)/2)2 ].
π (pdπ)
(6.28)
The function ρ(E) has two regions where it is non-vanishing as illustrated in Fig.
6.1(b). The upper region is the DOS for a π ∗ conduction band and the lower region
is the DOS for a π valence band.
The DOS for a π 0 non-bonding band is simply a δ function since these bands
are flat. The broadening of the bands due to the oxygen–oxygen interactions can
be included approximately by replacing the δ function with a Gaussian shape of
the form:
r
∼ λπ0 −λπ0 (E−Eπ0 )2
ρπ (E) =
0 e , (6.29)
π
where λ is a parameter that determines the band width and Eπ0 is the position of
the center of the π 0 band. These parameters can be estimated from the results of
Chapter 5. An interpolation formula for the π 0 bands is
E~kπ0 ' E⊥ − 4(ppπ)Cα2 Cβ2 + 2bSα2 Sβ2 , (6.30)
where b = (ppσ) − (ppπ). At Γ, (6.30) gives a triply degenerate energy, Eπ0 (Γ) =
E⊥ − 4(ppπ) and at M it gives Eπ0 (M) = E⊥ + 2b, and a doubly degenerate energy
E⊥ . At R (6.30) gives a triply degenerate energy Eπ0 (R) = E⊥ + 2b.
Thus (6.30) give exact energies at the Γ, M and R points. However, at X it gives
two energies at E⊥ , and one at E⊥ − 4(ppπ). The exact solution at X gives E =
E⊥ + 4(ppπ) and E⊥ − 4(ppπ) for the two π 0 energies. Despite this shortcoming,
(6.30) is a reasonably good approximation from which the band width and DOS
of the π 0 bands can be estimated. If (ppπ) and b have the same sign then the π 0
bands extend from E⊥ − 4(ppπ) to E⊥ + 2b and the band width is |2b + 4(ppπ)| =
|2(ppσ) + 2(ppπ)| = 2|c|. If (ppπ) and b differ in sign then the band extends from
E⊥ to E⊥ + 2b, and the band width is 2|b|. Since (ppπ) is generally smaller than
(ppσ) the band width will be approximately equal to 2|(ppσ)|.
If we define the band width to be the full-width at half-maximum of the
Gaussian of (6.29) then λπ0 = ln 2/(ppσ)2 . For SrTiO3 , (ppσ) ' 0.2 eV so one has
λ ' 17 eV−2 .
An analytical formula [2] for the DOS for π 0 bands corresponding to the inter-
polation formula (6.30) can be obtained in terms of complete elliptic functions of
the first kind. The resultant DOS contains jump discontinuities and a logarithmic
singularity, shown in Fig. 6.2.
The total density of states arising from the nine pi bands can finally be written
ρπ0 (ε)
ε1 ε2 ε3 ε4
Figure 6.2. Density of states [2] for the π 0 bands given by (6.30) with parameters b and
(ppπ) opposite in sign. ε1 = E⊥ , ε2 = E⊥ − 4(ppπ) and ε4 = E⊥ +2b.
as
ρpi (E) = 3ρπ0 (E) + 3ρπ (E) . (6.31)
The units of ρpi are electron states per unit cell per spin per unit energy or simply
states/(spin-cell-energy).
Z ∞ Z E⊥
ρpi (E) dE = 9 and ρpi (E) dE = 6 . (6.32)
−∞ −∞
Including spin there are 18 states. The 12 valence-band states are occupied in a
perovskite.
6.3 DOS for the sigma bands
According to (4.57) the eigenvalue equation for the sigma bands is

n
(Ee − E)(Ek − E) = 2(pdσ)2 (Sx2 + Sy2 + Sz2 )
q o
± (Sx4 + Sy4 + Sz4 )−(Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 ) . (6.33)
It is convenient to introduce the dimensionless variable, εσ (E), defined by
εσ ≡ [(Ee − E)(Ek − E)/(pdσ)2 ] − 3. (6.34)

Equation (6.33) can be expressed as, εσ (E) − ε± ~

σ (k) = 0 where
ε± ~
σ (k) = −(C2x + C2y + C2z )
q
± C2x2 + C2 + C2 − C C
2y 2z 2x 2y − C2y C2z − C2z C2x (6.35)
with C2α = cos 2kα a. The function εσ (~k) describes two branches ε+ ~ − ~
σ (k) and εσ (k)
as shown in Fig. 6.3.
Along the line ΓX, C2y = C2z = 1 and

p
ε± ~ (1 − C2x )2 ,
σ (k) = −(C2x + 2) ± (6.36)
ε+ ~
σ (k) = −1 − 2C2x , (6.37)
ε− (~k) = −3 .
σ (6.38)
Along the line XM, C2x = −1 and C2z = +1 so

q
ε± ~ 2 ,
σ (k) = −C2y ± 3 + C2y (6.39)
q
ε+ ~ 2 ,
σ (k) = −C2y + 3 + C2y (6.40)
q
ε− ~ 2 .
σ (k) = −C2y − 3 + C2y (6.41)
Along the line MR, C2x = C2y = −1 and

p
ε± ~ (C2z + 1)2 ,
σ (k) = 2 − C2z ± (6.42)
+ ~
εσ (k) = 3 , (6.43)
ε− (~k) = 1 − 2C2z .
σ (6.44)
According to (6.38) ε− ~ + ~
σ (k) is flat along ΓX and according to (6.43) εσ (k) is flat
along MR. As we shall show shortly, these lines of constant energy produce jump
discontinuities in the DOS at the bottom (εσ = −3) and top (εσ = +3) of the σ ∗
and σ bands as shown in Fig. 6.3. Such jumps are designated as P0 and P2 critical
points [3].
In order to determine the nature of the critical points we investigate the an-
alytic properties of the sigma energy bands near particular energies for which the
derivative of the density of states is infinite or discontinuous. The form of ρ(εσ )
near εσ = ±3 and ±1 can be determined by using a power-series expansion of the
band energies near these special points. We consider first the branch ε− σ (k) near
εσ = −3. Along ΓX the band is flat. Consider a cylinder with axis along ΓX. The
contribution to ρ(εσ ) from this cylinder can be obtained by expanding the energy
ε− ~
σ (k) in powers of ky and kz near the ΓX line along the kx -axis.
3 3 3 2
ε− 2 2
σ = −3 + (2ky a) + (2kz a) = −3 + r ,
4 4 4
r2 ≡ (2ky a)2 + (2kz a)2 . (6.45)
Then the contribution ∆ρ(εσ ) to ρ is

µ ¶3 Z π/2a ZZ
1 2a dky dkz
∆ρ(εσ ) = − Im dkx
π 2π −π/2a cylinder εσ + 3 − 34 r2 + i0+
Z 2π Z r0
1 1 r dr
=− Im dφ
π (2π)2 0 0 (εσ + 3) − 34 r2 + i0+
Z r02
1 4/3 dx
= − Im with ξ ≡ (εσ + 3)
π 4π 0 43 ξ − x + i0+
Z r02 µ ¶
1 4 1
= dx δ ξ−x = . (6.46)
3π 0 3 3π
p
The result of 1/(3π) is valid provided the cylinder radius r0 > 4ξ/3, where ξ =
(εσ + 3). We are interested in the result as εσ → −3 so that ξ → 0, therefore the
result is valid and independent of r0 so long as r0 is not identically zero. Thus, as
εσ varies from –3+0− to –3+0+ there is a jump in the density equal to 1/(3π).
Since there are three equivalent ΓX lines the total jump is 1/π. Symmetry requires
that the same result holds for the ε+ σ behavior of the density near εσ = +3 along
the MR lines along the edges of the first Brillouin zone. Each MR line is shared by
four zones so that the weight of each line is 1/4. Thus the 12 MR lines have a total
weight of 3 and the total jump in ρ(εσ ) is again 1/π.
The ε+σ branch is quadratic near εσ = −3 and contributes nothing to the jump
in ρ(εσ ) as εσ → −3. Similarly, ε−σ does not contribute to the jump in ρ(εσ ) as
εσ → +3.
Next, we consider the behavior of the branches near εσ = ±1, where the ε+
σ
branch has the expansion
1 2
ε+ 2
σ =1−α + r (6.47)
4
with α = (π − 2kx a) and r2 = (2ky a)2 + (2kz a)2 . Equation (6.47) describes a sad-
dle point region of the energy dispersion and the associated singular points are
designated as M1 and M2 [3].
Consider a cylinder with radius r0 extending from α = 0 to α0 . The contribu-
tion to the DOS from this cylinder is
µ ¶3 Z α0 Z r0 Z 2π
1 1 r dr
∆ρ(εσ ) = − Im dα dφ
π 2π 0 0 εσ − 1 + α2 − 14 r2 0
Z Z r02 /4
1 2(2π) α0 dx
= − Im dα
π (2π)3 0 0 ε σ − 1 + α2 − x
Z α0 Z r02 /4
1
= dα δ(εσ − 1 + α2 − x) dx
2π 2 0 0
Z √r02 /4−(εσ −1)
1
= dα , (6.48)
2π 2 L1
where

0 if εσ − 1 > 0,
L1 = (6.49)
√
1 − εσ if εσ − 1 < 0.
Thus, apart from the numerical factor 1/(2π 2 )

p 2
 r0 /4 − (εσ − 1) if εσ > 1,
∆ρ(εσ ) → p p (6.50)

r02 /4 + (1 − εσ ) − (1 − εσ ) if εσ < 1.
In the limit of εσ → 1+ ,
1 (εσ − 1)
∆ρ(εσ ) → r0 − (6.51)
2 r0
which shows that ∆ρ(εσ ) contains a contribution that decreases linearly as εσ

increases from 1. On the other hand, as εσ → 1−
1 (1 − εσ ) p
∆ρ(εσ ) → r0 + − (1 − εσ ) . (6.52)
2 r0
In this limit as εσ approaches 1 from below, the second term on the right-hand
side of (6.52) is negligible compared to the square root term. Thus, we see that
∆ρ(εσ ) increases as εσ increases towards 1 from below. More significantly, the
derivative d(∆ρ)/dεσ becomes infinite (tends to − ∞) as εσ → 1− . At εσ = 1+ or
1− , ∆ρ = r0 /(4π 2 ) and therefore the DOS near εσ = −1 is the mirror reflection of
the analytic behavior near εσ = +1.
A similar analysis of the ε−
σ branch near εσ = −1 yields the result that:
1
 2 r0 + (εσ + 1)/r0 if εσ < −1,
∆ρ(εσ ) → (6.53)
1 √
2 r0 + (εσ + 1)/r0 − 1 + εσ if εσ > −1.
A schematic of ρ(εσ ) is shown in Fig. 6.3(b). The DOS in εσ -space is symmetric

about εσ = 0.
An exact result for ρ(εσ ) has not yet been obtained, but knowing the analytical
behavior at the critical points suggests a simple approximation. We express ρ(εσ )
3 P0
1 M2
k)
εσ (~
–1 M1
–3 P0
Γ X M R 0 1/π ρ(εσ )
Figure 6.3. Schematic of ρ(εσ ).
as a linear combination of terms which have the correct analytical behavior:

( ρ (ε ) for |εσ | ≤ 1,
1 σ
ρ(εσ ) = (6.54)
ρ2 (εσ ) for 1 ≤ |εσ | ≤ 3,
with
p
ρ1 (εσ ) = A + B 1 − ε2σ + F (1 − |εσ |) |εσ | , (6.55)
√
ρ2 (εσ ) = C + Dx2 + F (1 − x) x , (6.56)
x ≡ (3 − |εσ |)/2.
To determine the constants A, B, C, D, and F , we use the following conditions:
ρ2 (±3) = 1/π (jump at the band edges), (6.57)

ρ1 (±1) = ρ2 (±1) (continuity), (6.58)
Z 3
ρ(εσ ) dεσ = 2 (two sigma bands are represented), (6.59)
−3
ρ1 (±1) = ρNum (±1), (6.60)
where ρNum (1) ≈ 0.432 is the value of ρ1 (1) obtained by numerical evaluation of
theoretical
0.6
numerical
0.4
ρ(εσ )
0.2
0
–3 –2 –1 0 1 2 3
εσ
Figure 6.4. The function ρ(εσ ) of (6.54) compared with numerical calculations of the
exact DOS.
the density of states. Imposing equations (6.57)–(6.60) yields the following results:
A = ρNum (±1) ≈ 0.432

B = −0.1646
1
C = = 0.3183 (6.61)
π
D = A − C ≈ 0.1136
F = −0.0151.
Figure 6.4 shows the sigma density of states computed from (6.54) and (6.61)
compared with results obtained by numerical integration. The approximation for
ρ(εσ ) given by (6.55) and (6.56) is remarkably good [4], being within 1% of the
numerical value for all values of εσ . A comparison of the approximate analytic
expression for ρ(ε) with numerical calculations is shown in Fig. 6.4.
To calculate the DOS in E-space we need only multiply the function in (6.54)
by |∂εσ /∂E|. This gives
¯ ¯
¯ E − 1 (E + E ) ¯ ¡ ¢
¯ 2 e k ¯
ρσ (E) = ¯ ¯ ρ εσ (E) . (6.62)
¯ (pdσ) 2 ¯
The factor 21 (Ee + Ek ) is the mid-gap energy between the σ and σ ∗ band edges at
Γ. Thus, the conversion factor |E − 21 (Ee + Ek )|/(pdσ)2 introduces a distortion of
the DOS which enhances the DOS at the bottom of the σ valence band and the top
of the σ ∗ conduction bands. A sketch of ρσ (E) is shown in Fig. 6.5.
|E−Em |
(pdσ)2
ρσ (E)
Em E
Figure 6.5. Sigma-band DOS as a function of E.
The non-bonding σ 0 band may be treated in the same manner as the π 0 bands.
Thus
r
∼ λσ0 −λσ0 (E−Eσ0 )2
ρσ0 (E) = e (6.63)
π
where λσ0 is a parameter that determines the band width of the σ 0 band and Eσ0
is the center of the band.
The width of the σ 0 band can be calculated approximately using first-order
perturbation theory to include the effects of the (ppπ) and the (ppσ) interactions.
We can write the Hamiltonian, H (discussed in Chapter 4), as H = H 0 + H 0 , where
H 0 is H with (ppπ) and (ppσ) set equal to zero, and H 0 is the part of H which
contains matrix elements involving (ppπ) and/or (ppσ). Using the matrix elements
described in Chapter 4 we find that
12 b Sx2 Sy2 Sz2

E~kσ0 ∼
= Ek + hσ 0 |H 0 |σ 0 i = Ek − . (6.64)
(Sx2 Sy2 + Sy2 Sz2 + Sx2 Sz2 )
The approximate band width is therefore 4b, which is about twice the band width
of the π 0 bands. If we equate 4b to the full-width at half-maximum of the function
in (6.63) then we find that
ln 2
λσ 0 ' . (6.65)
4b2
An analytical formula [2] for the σ 0 DOS corresponding to the approximate

expression (6.64) gives the structure shown in Fig. 6.6 which has a base width of 4b.
With the analytic forms of the “pi” and “sigma” DOS functions we are now able to
give a simple expression for the total density of states including the contributions
from all of the 14 primary energy bands,
· ³ ´ ³ ´¸
ρtotal (E) = 2 3 ρπ0 (E) + ρpi (E) + ρσ0 (E) + ρsigma (E) , (6.66)
1
ρσ0 (ε)
–1 0 1
0
Figure 6.6. Density of states [2] for the σ band given by (6.64) with parameters ε = 1
corresponding to E = Ek , and ε = −1 to E = Ek − 4b.
where the factor of 2 accounts for the two spin states. We note that
Z ∞
ρtotal (E) dE = 28 , (6.67)
−∞
Z E⊥
ρtotal (E) dE = 18 . (6.68)
−∞
Equation (6.68) expresses the fact that there are 18 electrons per unit cell in the
occupied valence bands. The valence-band wavefunctions, however, are admixtures
of p and d orbitals so that d orbitals are partially occupied as a result of the filled
valence-band states.
Mattheiss [5] has carried out augmented plane wave (APW) calculations of
the energy bands of several of the perovskites. Using the numerical results he has
constructed histograms of the DOS for SrTiO3 , KTaO3 , and ReO3 . In Figs 6.7,
comparisons of the results of Mattheiss with the analytical result of (6.66) are
presented. Also shown is a similar comparison for NaWO3 with the result of Kopp
et al. [6]. The model parameters have been selected to produce the prominent
structures in the same locations as in the histograms. It is seen that the analytical
density of states ρtotal (E) reproduces the DOS with considerable success.
6 6
(a) SrTiO3 (c) ReO3
5 5
4 4
3 3
LCAO model
2 2 APW model
APW model
LCAO model
N (E) (states/spin/cell/eV)
1 ?
1
?
? ?
0 0
–6 –4 –2 0 2 4 6 8 10 12 14 –6 –4 –2 0 2 4 6 8 10 12
E − EV (eV) E − EV (eV)
6 6
(b) KTaO3 (d) NaWO3
122
5 5
4 4
3 3
APW model KKR model
2 2
LCAO model LCAO model
1 1 ?
? ?
?
0 0
–6 –4 –2 0 2 4 6 8 10 12 14 –6 –4 –2 0 2 4 6 8 10 12
E − EV (eV) E − EV (eV)
Figure 6.7. Comparison of the LCAO model DOS functions with APW numerical histograms for (a) SrTiO3 [5], (b) KTaO3 [5], and
(c) ReO3 [7]. (d) Comparison of the LCAO model DOS function with KKR numerical histogram for Nax WO3 [6].
6.4 The Fermi surface and effective mass
(a) Semiconducting perovskites
The Fermi surface is an important feature of a metallic or semiconducting solid that

plays a vital role in determining the physical and chemical properties of the material.
The Fermi surface of a crystal is defined as the surface of constant energy in ~k-space
that separates the occupied states from the unoccupied states. At absolute zero
(T = 0 K) the lowest energy states are occupied up to a particular energy called the
Fermi energy, EF , and the states lying above EF are unoccupied. Hence at T = 0,
the surface is sharply defined. At finite temperatures the surface will be fuzzy with
a few “holes” below EF and a few electrons in states above EF . The fuzzyness is
described by the Fermi distribution function, f (E, T ) = [1 + e(E−EF )/kB T ]−1 , where
kB is Boltzmann’s constant. In this section we discuss the shape of the Fermi surface,
the density of states at the Fermi energy, and the effective mass of electrons in the
conduction bands (π ∗ bands) of a semiconducting cubic perovskite.
Let Eαβ (~k) (αβ = xy, xz, or yz) denote a particular π ∗ band of an insulating
perovskite. The crystal can be made conducting or semiconducting by any of a
number of “doping” methods. It can be converted to an n-type semiconductor by
heating in a hydrogen atmosphere, introducing oxygen vacancies, or adding cation
impurities. For example, SrTiO3 becomes n-type when doped with Nb atoms. For
an n-type material, the donated electrons occupy the lowest energy states of the π ∗
bands. Typical n-type materials usually have electron concentration of the order of
1018 –1020 electrons/cm3 . For a lattice parameter of 3.9 Å this gives a concentration
of 6 × 10−5 to 6 × 10−3 electrons per unit cell. Therefore, the occupied states will
be confined to a very small volume in ~k-space near Γ.
According to (6.28), at the bottom of the π ∗ bands (E = Et ) the total density
of states is
1 Eg0
ρtotal (Et ) = 2(spins) × 3(equivalent bands) × . (6.69)
4π (pdπ)2
Assuming a typical band gap, Eg0 , of 3.2 eV, and (pdπ) of 1.2 eV, we find that
ρtotal (Et ) = 1.06 states/(unit cell eV). (6.70)
For typical donor concentrations ρπ (E) can be taken to be constant for the small
range of occupied states near Et with [ρπ∗ (EF )]total = [ρπ∗ (Et )]total ≡ ρ0 . We have
that ∆F ρ0 = ncell where ∆F = (EF − Et ) and ncell is the donor concentration per
unit cell. Thus,
∆F = 0.943 ncell (in eV) . (6.71)

Figure 6.8. Surfaces of constant energy for E near Et . For E = EF any wavevector on
the surface of the “jack” corresponds to a state of energy EF . ~k-vectors inside and on
the surface of the “jack” correspond to filled states while those outside the “jack” are
unoccupied at T = 0 K. (a) For δ = 0: the surfaces are those of three circular cylinders
aligned along the three principal axes. Each cylinder extends between equivalent X points
of symmetry on the opposite faces of the Brillouin zone. The axes of the three cylinders
intersect at Γ. (b) For δ > 0: the arms of the “jack” become slender “cigars”. (c) For δ < 0:
the arms of the “jack” flare out, forming six “trumpets”.
For a concentration of 1020 electrons/cm3 (ncell = 6 × 10−3 electrons/cell) we find

that ∆F = 5.6 × 10−3 eV.
For E near Et we may write the energy of the αβ-type π ∗ band as
∗
π ~ 4(pdπ)2 2
Eαβ (k) − Et ≈ a [(kα )2 + (kβ )2 ] (αβ = xy, xz, or yz) (6.72)
Eg0
From (6.72) it follows that for the π ∗ (xy) band the surface of constant energy in
~k-space is that of a circular cylinder with its axis along kz . Similarly, for π ∗ (xz)
and π ∗ (yz) the surfaces of constant energy are those of circular cylinders oriented
along the ky and kx axes, respectively. The three cylinders form an object that
resembles a “jack” as shown in Fig. 6.8(a). The shape of the Fermi surface for a
free-electron model and for the energy bands of many materials is the surface of a
sphere or spheroid in ~k-space. Clearly, for the cubic perovskite the Fermi surface is
quite different.
The transport properties of the electrons (holes) in a crystal can often be
expressed in terms of the effective mass of the carriers. It can be shown that in an
external electric field an electron acquires an acceleration, ~a given by
1 d2 E(~k) ~
~a = q E, (6.73)
~2 dk 2
where ~ is Planck’s constant divided by 2π, q is the electron charge, and E is the
external electric field strength. The effective mass, m∗ , is defined so that a Newton-
like equation, (1/m∗ )F~ = ~a, applies to the motion of an electron in a solid. For
electronic bands with the energy dependent only on the square of the magnitude
of the wavevector, E(~k) = E(k 2 ), the surfaces of constant energy are spherical in
~k-space and one needs only a single number, m∗ , given by
1 1 ∂ 2 E(~k)
= .
m∗ ~2 ∂k 2
For more complex energy bands a tensor is required to describe the electron’s
dynamics in a magnetic or electric field. The inverse mass tensor is defined so that
Xh 1 i
aα = Fβ ,
m∗ αβ
where
h 1 i 1 ∂ 2 E(~k)
= . (6.74)
m∗ αβ ~2 ∂kα ∂kβ
Using (6.74) we find that for an n-type perovskite,

h 1 i h 1 i 8(pdπ)2 a2
= = for π ∗ (αβ) band, (6.75)
m∗ αα m∗ ββ ~2 Eg0
h 1 i h 1 i h 1 i
= = = 0 for π ∗ (αβ) band, (6.76)
m∗ γγ m∗ αβ m∗ αγ
where Eg0 = (Et − E⊥ ) is the energy gap at Γ and “a” is half the lattice constant.
With Eg0 = 3.2 eV, (pdπ) = 1.2 eV, and a = 1.95 Å, numbers appropriate for SrTiO3 ,
we find m∗ = 0.56 m0 , where m0 is the free-electron mass, 9.11×10−28 g. This result
means, for example, that an electron in the bottom of the π ∗ (αβ) band moves under
the influence of an electric field oriented along the α- or β-axis as if it had only
56% of its free-electron mass. Conversely, in a field oriented along the γ-direction
the electron behaves as if it had an infinite mass.
Of course, there are three symmetry-equivalent π ∗ (αβ) bands so the effective
mass is isotropic for cubic perovskites. It should also be mentioned that different
average effective masses are employed in the analysis of different types of experi-
ments. In the case of conductivity, for example, the quantity hm∗ icond is employed,
where
D 1 E 1³ 1 1 1 ´
∗
= ∗
+ ∗ + ∗ . (6.77)
m cond 3 mαα mββ mγγ
3
For the example above, this would yield hm∗ icond = 2 m∗αα .
As can be seen from (6.75), the effective mass increases linearly with increas-
ing energy gap, Eg0 , but decreases quadratically with (pdπ). Therefore the result
is strongly dependent on the LCAO parameters employed. This is not true for
all quantities. For example, ∆F /[1/m∗ ]αα = ∆F m∗αα = π~2 ncell /(12a2 ) involves only
physical constants and the electron concentration and is therefore independent of
the LCAO parameters, (pdπ), and Eg0 .
The interesting shape of the Fermi surface (Fig. 6.8) is a direct result of the
two-dimensional character of the π and π ∗ bands. In a more elaborate theory (with
more distant interactions, additional orbitals, or spin-orbit effects), the π ∗ (αβ) and
π(αβ) bands would have at least a weak dependence on kγ . The dispersion would
be expected to take the form
~ 4(pdπ)2 2 £ ¤
Eπαβ
∗ (k) − Et ≈
0
a (kα )2 + (kβ )2 + δ(kγ )2 (6.78)
Eg
where δ is a dimensionless number that is small compared to 1. The effect of the

δ-term would be to convert the circular cylinders in Fig. 6.8 into thin, cigar-shaped
rods (Fig. 6.8(b) for δ > 0) or flare the cylinders into trumpet-shaped objects
(Fig. 6.8(c) for δ < 0). Using (6.75) we find for n-type SrTiO3 that
h 1 i h 1 i ³ 1 ´ 0.56
∗
=δ ∗
=δ or m∗γγ = . (6.79)
m γγ m αα 0.56 δ
Recent numerical calculations for SrTiO3 reported by Marques et al. [8] give
m∗αα = 0.408 m0 (light mass) and m∗γγ = 7.357 m0 (heavy mass),1 while older cal-
culations by Kahn [9] give “bare”2 values of m∗αα = 0.96 m0 and m∗γγ = 4.7 m0 and
Frederikse et al. [10] deduced values of 1.5 m0 and 6.0 m0 from magnetoresistance
measurements. The scatter of these results is great, but they have in common the
existence of a “light” and “heavy” effective mass which is predicted by the simple
LCAO model to result from the two-dimensional character of the π ∗ bands. Refer-
ring to (6.79) it is seen that a small value of δ, for example δ = 0.1, would bring the
heavy mass into the range of the numbers quoted above.
It should be mentioned that m∗αα (m∗γγ ) is often referred to as the transverse
(longitudinal ) effective mass, m∗t (m∗l ) because it applies to motion transverse (lon-
gitudinal) to the long axis of the Fermi surface.
Since the π valence bands are the mirror reflection of the π ∗ conduction bands,
the results for the effective mass tensor of a hole at the top of the π valence band
are the same as for the π ∗ band (but opposite in sign because the hole acts as a
1
Marques et al. [8] reports both relativistic and non-relativistic values for the effective masses. The
results quoted above are their non-relativistic masses. Relativistic effects have only a very small effect
on the light mass, but greatly reduce the heavy mass according to these authors.
2
The effective mass tensor components derived here are called the “bare” effective masses because they
do not include the effects of phonons or polarons. Such lattice effects can in some cases “clothe” an
electron, greatly increasing its effective mass. This is particularly important when the crystal has a
“soft” phonon mode as is the case for crystals that become ferroelectric at lower temperatures.
positive charge). However, the non-bonding bands usually lie near to or above the
top of the π valence bands. Consequently, the effective mass at the Fermi surface
will likely be determined by the curvature of the non-bonding bands. Since the
non-bonding bands have little curvature, heavy hole masses are expected.
(b) Metallic perovskites
Next, we consider a metallic perovskite with a significant density of electrons in the

π ∗ conduction band. In this case the Fermi energy is well above the conduction-band
edge and the two-dimensional cross-section of the Fermi surface will depart from
circularity. Figure 6.9 shows the cross-section of the constant energy surfaces in ~k-
space for any one of the three π ∗ conduction bands. The contours would correspond
to the Fermi surface cross-section as the number of electrons per band increases from
kβ
6
π
2 X
C
B
A
X -k
X π α
2
X
Figure 6.9. Cross-section of the constant energy surfaces for the π ∗ (αβ) band for a
cubic perovskite. The figure is a plot of the curves Sα2 + Sβ2 = Ω for various values of Ω.
For small Ω (curves inside A) the cross-section is nearly circular. The square cross-section
(B) occurs for Ω = 1 and corresponds to the cross-section of the Fermi surface for a half-
filled band. For Ω increasing beyond 1 the surfaces curve around the four X points in the
Brillouin zone. For a nearly filled band, (C) the cross-section in the first Brillouin zone
consists of four, quarter-round rods. In an extended Brillouin-zone view the Fermi surface
would consist of four circular rods each centered on one of the X points and extending
perpendicular to the plane of the diagram.
0 up to 1 (2 including spin). Using the definition (6.74) we find that the effective
mass is given by
h 1 i · ¸
4(pdπ)2 a2 (pdπ)2 sin2 (2kα a)
= cos(2kα a) − , (6.80)
m∗ αα ~2 (E − Em ) (E − Em )2
for the π(αβ) band. An equivalent expression holds for [1/m∗ ]ββ with kβ replacing
kα in (6.80) and as before [1/m∗ ]γγ = 0. In contrast to the results of the previous
section for low electron densities, it is clear that the effective mass components
depend upon the magnitude of the wavevector and will therefore vary on the Fermi
surface. In addition, it can be seen from (6.80) that the sign of [1/m∗ ]αα may
change from positive to negative for a value of kα a sufficiently large (but still
< π/2). In such a case an electron would be accelerated by an electric field as if it
had a positive charge (hole-like behavior). The effective mass will change sign at
the point for which [1/m∗ ]αα = 0 on the Fermi surface. The condition for this is
(E − Em )2
cos2 (2kα a) + 2β cos(2kα a) − 1 = 0, where 2β = . (6.81)
(pdπ)2
The solution to (6.81) is

1 £ p ¤
kc a = arccos − β + β 2 + 1 , (6.82)
2
where kc is the “critical” value of kα for which [1/m∗ ]αα = 0.

As a concrete example consider the compound NaWO3 , a metallic, cubic per-
ovskite at room temperature. For this compound we have W6+ , Na1+ , and O2−
ions so that there is one electron in the π ∗ conduction bands (1/6 of an electron in
each of the six degenerate π ∗ bands including the two spin states). To determine
the Fermi energy, EF , we make use of the DOS function and write,
Z εF
1
ρπ∗ (ε) dε = , (6.83)
−2 6
where ρπ∗ (ε) is given by equation (6.26) and ε is defined by (6.19). For the
case of NaWO3 the relevant parameters are, 2a = 3.87 Å, |(pdπ)| = 1.54 eV and
Eg = 3.72 eV. Using these values we find with the help of (6.26) that εF = –1.086.
Figure 6.10(a) shows the density of states and the number of electrons as a function
of ε. The position of εF is shown in the figure. Using the definition in (6.19) we can
also write,
q
EF − Em = (Eg /2)2 + 2(pdπ)2 (εF + 2) = 2.79 eV . (6.84)
For the π(xy) band, we define ΩF to be the projection of the Fermi surface on
6
Total density of states/cell
Total electron states/cell
5
(a)
4
TDOS/cell
TES/cell
3
εF = −1.086
2
?
1 - 1
0
2 1 0 –1 –2 ε
0.7
(b) Fermi surface
0.6 20
?
0.5
m∗ /m0
0.4 0
6
ky a
0.3
0.2 m∗ /m0 –20
0.1
0 –40
0 0.1 0.2 0.3 0.4 0.5 0.6 6
0.7
kx a kc a
3.0
(c)
2.5
2.0
m∗ /m0
1.5
1.0
0.5
kc a
0.0 ?
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
kx a
Figure 6.10. (a) Total density of states/cell for the π ∗ bands and the total number of
electron states/cell as a function of ε. The arrow indicates the value of the Fermi energy,
εF , for one electron occupying the bands. (b) The upper curve is the projection of the
Fermi surface on the kx–ky plane, ΩF . The lower curve is the effective mass, m∗ , as a
function of kx for wavevectors on ΩF . kc is indicated. (c) Expanded plot of m∗ > 0.
the kx–ky plane. For NaWO3 , εF is defined by the equation
(εF + 2)
Sx2 + Sy2 = = 0.457, (6.85)
2
which gives an upper limit of 0.742 radians for kα a and (6.82) gives kc a = 0.643
radians. The pairs of wavevectors (kx and ky ) lying on ΩF can easily be found.
For example, one can pick a value for kx then use (6.85) to calculate ky with the
requirement that ky must be real. Once the values of the two-dimensional vectors,
(kx , ky ) lying on ΩF are determined the effective mass can be calculated from (6.80)
as a function of kx or ky .
Figure 6.10(b) shows the right-hand, upper quadrant of the Brillouin zone with
the values of m∗ (the inverse of [1/m∗ ]xx ) as a function of the value of kx a on ΩF
for the π ∗ (xy) conduction band of NaWO3 .
The singularity in m∗ occurs when kx a = kc a = 0.643. At that point
[1/m∗ ]xx = 0 so m∗ diverges to plus infinity on one side and minus infinity on
the other. Both m∗ and ΩF are independent of kz and do not depend upon the sign
of kx or ky , so the results for the entire kx–ky plane of the Brillouin zone can be
obtained by reflecting the results in the quadrant through the kx and ky axes. From
Fig. 6.10(b) it can be seen that the effective mass is positive for kx a less than kc a
and negative in the range kc a < kx a < 0.742. The upper limit is the largest value
of kx a which lies on ΩF . The average effective mass, h1/m∗ i−1 , averaged over the
Fermi surface is 1.66 m0 for NaWO3 .
It should be noted that as the number of electrons in the π ∗ bands is decreased
the critical value of kx a given by (6.82) will eventually be larger than the maximum
value of kx a on ΩF . When this happens, the electron effective mass will be positive
everywhere on ΩF .
(c) Electronic properties of Nax WO3
In this section we discuss some of the electronic properties of the sodium–tungsten

bronze alloys, Nax WO3 , as a function of x. In particular, we use our LCAO model
to calculate the x dependence of the effective mass, the electronic specific heat, and
the magnetic susceptibility.
The properties of Nax WO3 depend strongly upon the value of x. To a very
good approximation each Na atom contributes one electron to the π ∗ conduction
bands. Thus, the conduction-band electron concentration is equal to x. For x < 0.24,
Nax WO3 is believed to be a Mott-type insulator [11–13]. For 0.24 < x < 0.49, it
is a superconductor with a tetragonal structure [14–16]. In the range 0.5 ≤ x ≤ 1,
Nax WO3 is metallic and has the aristotype, cubic structure [17, 18]. We confine
our discussions to the latter range of x values where the material is cubic.
Specific heat
The specific heat due to the conduction electrons, Cel , is defined as the rate of
change of the average total electronic energy with temperature per unit cell at
constant volume:
Z Z
d df (E, T )
Cel = E ρ(E) f (E, T ) dE = E ρ(E) dE (6.86)
dT dT
where T is the temperature, E is the electronic energy, ρ(E) is the electronic den-
sity of states, and f (E, T ) is the Fermi distribution function. Since df (E, T )/dT is
negligible except in a small range of thermal energies about the Fermi energy, EF ,
the limits of the integral can be taken arbitrarily as E1 and E2 as long as they are
far from the Fermi energy. In the small thermal range where df /dT is significant,
the density of states is nearly constant and hence it may replaced by ρ(EF ). It is
convenient to measure all of the energies from the Fermi energy so that
Z (E2 −EF )
2 [(E − EF )/kB T ]2 exp[(E − EF )/kB T ]d[(E − EF )/kB T ]
Cel = kB ρ(0)
−(E1 −EF ) [1 + exp[(E − EF )/kB T ]2
Z L2 2
2 α exp(α) dα
= kB T ρ(0) ≡ γ T, (6.87)
−L1 [1 + exp(α)]2
1
γ = π 2 kB 2
ρ(0) = 2.36 ρ(EF ) in mJ/mole K2 . (6.88)
3
In (6.87) α = (E − EF )/kB T , ρ(0) = ρ(EF ), L1 = (E1 − EF )/kB T , and L2 = (E2 −
EF )/kB T . The specific heat coefficient, γ, given in (6.88) is the well-known result
for metals, but care must be used when employing it. The integrand in (6.87)
vanishes as α2 e−|α| when α → ±∞. Therefore, if L1 and L2 are sufficiently large
the actual limits become unimportant provided there are no energy gaps near EF ;
that is, provided |EF − Eedge | & 10kB T , where Eedge is the edge of an energy band
beyond which there is an energy gap. At room temperature kB T ≈ 0.026 eV so
that the criterion would require |EF − Eedge | & 0.25 eV. At higher temperatures
the requirement is more demanding. In the previous section on doped insulators
we found that |EF − Eedge | ≈ 0.006 eV and therefore (6.86) can not be used for the
electronic specific heat of these materials.
Specific heat of Nax WO3 as a function of x
The Fermi energy of Nax WO3 lies in the π ∗ conduction bands and for x ' 0.5, EF −
Et ' 0.55 eV so the above criterion, |EF − Eedge | = |EF − Et | & 10kB T is satisfied
for T ≤ 600 K. The application of (6.88) requires ρ(EF ) as a function of x. One
approximation often made is the so-called “rigid-band model” (RBM). According

to this model the band structure itself is independent of x. That is, the energy
bands are “rigid” and as x increases the added electrons donated by the Na atoms
simply occupy the fixed (in energy) π ∗ states. This is not a good approximation for
Nax WO3 [19]. Optical experiments [20] show that the band gap at Γ for x = 0.5 is
about 0.5 eV smaller than for x = 1.0. In addition, the lattice constant 2a increases
by 0.04 Å [18] over this same range.
To calculate γ we make use of the results for the density of states for the π ∗
bands,
6 ³p ´
ρπ (ε) = 2 K 1 − (ε/2)2 , (6.89)
π
Z εF (x)
number of conduction electrons = x = ρπ (ε) dε, (6.90)
−2
p
(Eg /2)2 + 2(pdπ)2 (εF + 2)
ρπ (EF ) = ρπ (εF ) . (6.91)
(pdπ)2
Equation (6.89) gives the total DOS in terms of the dimensionless variable, ε. The
prefactor 6 accounts for the three degenerate bands and two spin states. Equation
(6.90) expresses the number of conduction electrons, x, in terms of the integral over
the dimensionless density of states. This expression determines the dimensionless
Fermi energy, εF (x). The DOS, ρπ (EF ) (in real energy) is then obtained by use of
(6.91). It should be noted that the result for ρπ (EF ) (and hence γ) depends only on
two energies, Eg and (pdπ). Specific energies such as Et , and E⊥ are not required.
For the specific heat coefficient in units of mJ/(mole K2 ) we have:
p
(Eg (x)/2)2 + 2(pdπ(x))2 (εF (x) + 2)
γ(x) = 2.36 ρπ (εF (x)), (6.92)
(pdπ(x))2
The calculation can be simplified by making use of approximate formulae for the
x dependence of the quantities. The following approximate relationships have been
shown to agree well with a wide range of experiments [19]:
Eg (x) = 2.85 + 0.84 x, (in eV) (6.93a)

(pdπ(x)) = 2.44 − 0.86 x, (in eV) (6.93b)
1
a(x) = (3.785 + 0.0818 x), (W−O distance in Å) (6.93c)
2
εF (x) = −2 + (2 − 1.086) x, (in eV) (6.93d)
6
ρπ (x) = (1 + 0.326 x), (total density of states/cell). (6.93e)
2π
The results obtained from equations (6.92) and (6.93) for γ(x) are shown in
Fig. 6.11 where they are compared with experimental data [21–23]. The dashed lines
3.5
3.0
▽
2.5 RBM x=0.8
γ (mJ/mole K2 )
▽
2.0
▽N RBM x=0.6
1.5 ▽
N LCAO
N N
1.0 NNN
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Na concentration x
Figure 6.11. Electronic specific heat coefficient, γ, as a function of x for Nax WO3 . The
solid line gives the result of (6.92). The dashed lines show the results for the rigid-band
model fits at x = 0.6 and 0.8. The experimental data are from Höchst et al. [21] (O), Vest
et al. [22] (◦), and Zumsteg [23] (N).
marked ‘RBM’ result from applying the rigid-band approximation [21] where the
band structure is assumed to be independent of x. The x-dependent model clearly
fits the experimental data better than the RBM.
Magnetic susceptibility and effective mass of Nax WO3

as functions of x
The magnetic susceptibility consists of a paramagnetic contribution from the elec-

tron spins (Pauli paramagnetic term) and a diamagnetic contribution due to the
orbital motion of the electrons. An approximate expression that combines these two
terms (in units of emu/mole), is
³ 1 D m0 E2 ´
χ = 4.0424 × 10−6 a2 1 − Nπ (EF ), (6.94)
3 m∗
where Nπ (EF ) is the total density of states at EF and m0 is the electron rest mass.
To calculate χ the inverse effective mass averaged over the Fermi surface, hm0 /m∗ i,
is required. This quantity can be obtained from the expression given for the inverse
of the effective mass in (6.80). The averages, hcos(2kα a)i, and hsin2 (2kα a)i are given
by line integrals around the curve that is the projection of the Fermi surface on the
αβ-plane:
Z
1 `
hcos(2kα a)i = cos[2kα (`)a] d` , (6.95)
` 0
Z
1 ` 2
hsin2 (2kα a)i = sin [2kα (`)a] d` , (6.96)
` 0
where ` is the length of the Fermi-surface curve in the αβ-plane. The average,
hcos(2kα a)i, can be obtained by symmetry arguments. The Fermi surface for our
LCAO energy bands is defined by the equation, Sα2 + Sβ2 = ΩF (εF ) = 12 (εF + 2), a
constant for kα a and kβ a on the Fermi surface. Since cos(2kα a) = 1 − 2 sin2 (kα a)
we have that
hcos(2kα a)i = 1 − h2 sin2 (kα a)i. (6.97)
Whatever shape the Fermi-surface curve has on the αβ-plane it must be symmetric
in the variables along α and β because these “directions” are physically equivalent
and indistinguishable. Thus, the average hsin2 (kα a)i must be equal to the average
hsin2 (kβ a)i. That being the case, (6.97) may be written as
1
hcos(2kα a)i = 1 − hsin2 (kα a) + sin2 (kβ a)i = 1 − ΩF (εF ) = − εF . (6.98)
2
The other average, hsin2 (2kα a)i, can not be obtained analytically, but a simple
interpolation formula that errors by less than 0.3% over the cubic range is given by
hsin2 (2kα a)i ≈ ΩF (x)[ 2 − 1.49 ΩF (x)]. (6.99)
Using (6.99), (6.80), and (6.84) yields

D m E 4m0 (pdπ(x))2 a(x)2
0
=
m∗ (x) ~2 [(Eg (x)/2)2 + 4(pdπ(x))2 ΩF (x)]1/2
½ ¾
£ ¤ (pdπ(x))2 ΩF (x)[2 − 1.49 ΩF (x)]
× 1 − ΩF (x) − . (6.100)
[(Eg (x)/2)2 + 4(pdπ(x))2 ΩF (x)]
The x-dependent quantities in (6.100) are given by (6.93) and (6.99). The results
for hm∗ (x)/m0 i, shown in Fig. 6.12, agree well with the data of Camagni et al. [24]
but lie well above the data of Owen et al. [20].
Use of equation (6.100) in (6.94) yields an expression for the magnetic suscep-
tibility as a function of x:
³ 1 D m0 E2 ´
χ(x) = 4.0424 × 10−6 a(x)2 1 − Nπ (x), (6.101a)
3 m∗ (x)
(EF − Em )
Nπ (x) = ρπ (εF (x)) (6.101b)
(pdπ(x))2
p
(Eg (x)/2)2 + 2(pdπ(x))2 (εF (x) + 2)
= ρπ (x) (6.101c)
(pdπ(x))2
where the x dependence of the parameters on the right-hand side of (6.102) is given
by equations (6.93a-e). The results of (6.101) are shown in Fig. 6.13 and compared
1.8
1.6
1.4
m∗eff /m0 1.2

N
1.0 N
0.8
N
0.6
0.5 0.6 0.7 0.8 0.9 1.0
Na concentration x
Figure 6.12. Effective mass averaged over the Fermi surface as a function of x for
Nax WO3 . The solid curve is obtained from the results of equation (6.100). The Dashed
curves show the RBM results fixed at x=0.6 and 0.8. The experimental data are from
Camagni et al. [24] (N) and Owen et al. [20] ( ). ◦
with experimental data. The theoretical curve is somewhat higher than the exper-
imental data but follows closely the general shape. Also shown in Fig. 6.13 is the
susceptibility calculated from the effective mass data of Owen et al. [20]. The data
of Greiner et al. lie about equally between the LCAO model results and those from
Owen et al. The above results show that the simple empirical, nearest neighbor,
LCAO model is able to predict reasonable values for the electronic properties of the
cubic perovskites including the effective mass, specific heat, and magnetic suscepti-
16

14
χ(10−6 emu/mole K2 )
12
H
10
H
H
8
6 H H
H
4
H
2
0.5 0.6 0.7 0.8 0.9 1.0

Na concentration x
Figure 6.13. Contribution of the conduction electrons to the magnetic susceptibility, χ,

as a function of x for Nax WO3 . The solid curve gives the result of (6.101). The dashed
curve is the result obtained using the data for the effective mass from Owen et al. [20]
(H). The other data are from Kupka et al. [25] (¤) and Greiner et al. [26] ( ). ◦
bility. In addition, the model shows that these electronic properties are principally
determined by two parameters, the (pdπ) integral and the band gap Eg . The den-
sity of states, ρ(ε), is a dimensionless function determined entirely by symmetry. It
is a universal function in that it applies to all cubic perovskites and is independent
of the values of the empirical parameters.
References
[1] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products, 4th
edn (New York, Academic Press, 1965).
[2] T. Wolfram and Ş. Ellialtıoğlu, Phys. Rev. B 25, 2697 (1982).
[3] L. van Hove, Phys. Rev. 89, 1189 (1953); J. C. Phillips, Phys. Rev. 104, 1263
(1956).
[4] Ş. Ellialtıoğlu and T. Wolfram, Phys. Rev. B 15, 5909 (1977).
[5] L. F. Mattheiss, Phys. Rev. B 6, 4718 (1972).
[6] L. Kopp, B. N. Harmon, and S. H. Liu, Solid State Commun. 22, 677 (1977).
[7] L. F. Mattheiss, Phys. Rev. 181, 987 (1969).
[8] M. Marques, L. K. Teles, V. Anjos, L. M. R. Scolfaro, J. R. Leite, V. N. Freire,
G. A. Farias, and E. F. da Silva, Jr., Appl. Phys. Lett. 82, 3074 (2003).
[9] A. H. Kahn, Phys. Rev. 172, 813 (1968).
[10] H. P. R. Frederikse, W. R. Hosler, W. R. Thurber, J. Babiskin, and P. G. Sieben-
mann, Phys. Rev. 159, 775 (1967); H. P. R. Frederikse, W. R. Hosler, and W. R.
Thurber, Phys. Rev. 143, 648 (1966); H. P. R. Frederikse and G. A. Candela,
Phys. Rev. 147, 583 (1966).
[11] J. B. Goodenough, Prog. Solid State Chem. 5, 195 (1971).
[12] A. Ferreti, D. B. Rogers, and J. B. Goodenough, J. Phys. Chem. Solids 26,
2007 (1965).
[13] N. F. Mott, Philos. Mag. 35, 111 (1977).
[14] C. J. Raub, A. R. Sweedler, M. A. Jensen, S. Broadston, and B. T. Matthias,
Phys. Rev. Lett. 13, 746 (1964).
[15] H. R. Shanks, Solid State Commun. 15, 753 (1974).
[16] K. L. Ngai and R. Silberglitt, Phys. Rev. B 13, 1032 (1976).
[17] G. Hagg, Z. Phys. Chem. Abt. B 29, 192 (1935).
[18] B. W. Brown and E. Banks, J. Am. Chem. Soc. 76, 963 (1954).
[19] T. Wolfram and L. Sutcu, Phys. Rev. B 31, 7680, (1985).
[20] J. F. Owen, K. J. Teegarden, and H. R. Shanks, Phys. Rev. B 18, 3827 (1978).
[21] H. Höchst, R. D. Bringans, and H. R. Shanks, Phys. Rev. B 26, 1702 (1982).
[22] R. W. Vest, M. Griffel, and J. F. Smith, J. Chem. Phys. 28, 293 (1958).
[23] C. Zumsteg, Phys. Rev. B 14, 1401 (1976).

[24] P. Camagni, A. Manara, G. Campagnoli, A. Gustinetti, and A. Stella, Phys.
Rev. B 15, 4623 (1977).
[25] F. Kupka and M. J. Sienko, J. Chem. Phys. 18, 1296 (1950).
[26] J. D. Greiner, H. R. Shanks, and D. C. Wallace, J. Chem. Phys. 36, 772 (1962).
1. Find expressions for the DOS, ρ(E), for the following energy bands whose dispersion
is given by: p
(a) E = E0 + E12 + E22 Sx2 , where Sx = sin(kx a), and − π2 < kx a < π2 .
(b) E = E0 + E1 Sx .
(Hint: Make use of the expression ρ(E) = ρ(f (E))|df (E)/dE|.)
(c) What is the energy dependence of ρ(E) for
E = E0 + E1 [sin2 (kx a) + sin2 (ky a) + sin2 (kz a)]
when E is near E0 ?
2. Discuss the nature of the singularities (if any) of the DOS for (a), (b), and (c) in
Problem 1.
3. A saddle point is said to exist at E0 if the dispersion takes the form E = E0 + αkx2 − βky2
near E0 , where α and β are real positive numbers. Show that this form leads to a
logarithmic singularity at as E → E0 .
4. Show that the π and π ∗ energy bands have saddle points at the X points in the
Brillouin zone. (For simplicity ignore the oxygen–oxygen interactions.)
5. Show that for Ω = Sx2 + Sy2 ≤ 1, the minimum value of the dimensionless
Fermi penergy, ²F , for which 1/m∗ has a zero, is determined by the condition
−β + β 2 + 1 = −(²F + 1), where β = [(ωg /2)2 + 2(²F + 2)]/2, and ωg = Eg /(pdπ).
6. For Nax WO3 the total number of electrons in the π ∗ bands, n, is described accurately
by the expression n ≈ (²F + 2)/(1.828) for 0 < n < 1. Use this result and the result of
Problem 5 find an expression for the minimum total number of electrons per unit cell,
nmin , for which 1/m∗ has a zero. Using Eg = 3.72 eV and (pdπ) = 1.54 eV find the value
of nmin .
7
Optical properties of the d-band perovskites
Light interacts with the electrons of a solid through the electromagnetic field as-
sociated with the light wave. The electric field exerts an oscillating force on the
electrons and ions which produces electronic transitions and other excitations in
the solid.
There are several different types of optical adsorption mechanisms for ionic
solids such as the perovskites. In the infrared region the electromagnetic field of the
photons is strongly coupled to the polarization field of the vibrating ions and “op-
tical” phonons can be created. If the solid is magnetic then adsorption of light can
occur due to the excitation of spin waves or magnons. absorption by free electrons
(or holes) is also important in the infrared optical region for metallic or semicon-
ducting perovskites. Another important source of absorption is the excitation of
plasmons. In doped semiconducting perovskites, plasmon absorption may occur in
the infrared region while for metallic materials it is in the visible to ultraviolet
region.
Photons with energy greater than the electronic band gap between the highest
occupied and the lowest unoccupied bands can cause interband transitions. An
interband transition involves the excitation of an electron from a filled valence
band to an unoccupied state in another band. For an insulating material interband
transitions can occur for photon energy ~ω > Eg , where Eg is the fundamental band
gap. The optical properties of insulating perovskites in the visible and ultraviolet
regions are mainly determined by such interband transitions. This chapter deals
principally with the nature of interband transitions in the insulating perovskites.
There are many other mechanisms of importance to the optical properties
such as absorption by excitons and indirect interband transitions, which involve an
electronic transition accompanied by the simultaneous creation or absorption of a
phonon or magnon. These topics will not be discussed.
The optical response of a solid to higher-energy photons (~ω of the order of
138
103 eV) such as are employed in x-ray photoelectron spectroscopy (XPS) will be
discussed in Chapter 8.
7.1 Review of semiclassical theory
In this section we briefly review the semiclassical theory of the optical properties
of solids. More detailed discussions may be found in the references given at the end
of this chapter.
In the presence of an electromagnetic field the kinetic energy operator for an
electron, p2 /2m, must be replaced by the new operator
1 h e~ i2
p~ − A(~
r, t) , (7.1)
2m c
~ A(~
where p~ = −i~∇, ~ r, t) is the vector potential of the electromagnetic field, e is the
magnitude of the electron charge, and c is the velocity of light. The electromagnetic
field may be described in the Coulomb gauge where ∇ ~ ·A~ = 0.
The vector potential is usually a small perturbation on the electrons of a solid

~ r, t)2 in (7.1) may be neglected and the term linear in A(~
so that the term in A(~ ~ r, t)
may be treated by means of first-order time-dependent perturbation theory.
The one-electron Hamiltonian for the solid is written as
H = H 0 + H 0 (~r, t) (7.2)
p2
H0 = + V (~r)
³2me ´
H 0 (~r, t) = ~ r, t) · p~ .
A(~
mc
The vector potential for a field of angular frequency ω is
~ r, t) = A ~a0 e[i(~q·~r−ωt] + c.c.

A(~ (7.3)
where A is a scalar amplitude, ~a0 is a unit vector (the polarization vector) perpen-
dicular to the propagation vector ~q, and “c.c.” means the complex conjugate. The
~ r, t) is related to A(~
electric field strength, E(~ ~ r, t) by the equation
~ r, t) = − 1 d A(~
E(~ ~ r, t)
c dt
³ iω ´
= A ~a0 ei(~q·~r−ωt) + c.c. (7.4)
c
~ r, t) = E~1 (~r, t) + E~1 (~r, t)∗ . The electric displacement field
It is useful to write E(~
140 Optical properties of the d-band perovskites
~ r, t) can then be written as

D(~
~ r, t) = ε(~q, ω) E~1 (~r, t) + ε(~q, ω)∗ E~1 (~r, t)∗
D(~ (7.5)
where ε(~q, ω), the frequency- and wavevector-dependent dielectric function, is a

3 × 3 tensor whose elements are complex:
ε(~q, ω)αβ = ε1 (~q, ω)αβ + i ε2 (~q, ω)αβ . (7.6)
Since we are concerned here with cubic crystals we may assume that ε is a multiple
of the unit tensor and consequently it may be treated as a scalar quantity.
According to electromagnetic theory, the average rate of loss of energy density
(energy/volume/s) from an electromagnetic field in a medium with a dielectric
function ε(~q, ω) is
¿ ~À
1 ~ dD
E· , (7.7)
4π dt
where h · · · i means the time average over a period of oscillation, T = 2π/ω. Using
(7.4) and (7.5) one finds the total energy loss per second
¿ ~À Z 2π/ω µ ~¶
1 ~ dD ω ~ dD
E· = E· dt
4π dt 2π 0 dt
1 |A|2 ω 3
= ε2 (~q, ω). (7.8)
2π c2
Equation (7.8) provides a means of relating macroscopic electromagnetic the-

ory to microscopic quantum theory. The approach is to find the rate of electronic
transitions, dW/dt, due to the perturbation H 0 of (7.2). Each transition corresponds
to the absorption of a photon of energy ~ω and therefore to a loss rate, (dW/dt)~ω,
of energy from the electromagnetic field. By equating this loss to that given by (7.8)
one can find an expression for ε2 in terms of the quantum states of the solid.
According to first-order time-dependent perturbation theory, an oscillating per-
turbation of the form H 0 (~r)e−iωt , where H 0 (~r) is a constant or operator function
of ~r, will cause transitions between the unperturbed states of the system at a rate
(transitions per second) given by
dWif (~ω) 2π
= |hf |H 0 (~r)|ii|2 δ(Ef − Ei − ~ω) δSi ,Sf . (7.9)
dt ~
In (7.9), |ii and hf | designate the initial and final states, respectively, Ei and Ef are
the corresponding energies and Si and Sf indicate the spin states. Equation (7.9)
shows that the strength of the transition is proportional to the absolute square of
the matrix element and the δ function ensures conservation of energy in the process.
Using (7.2), the rate of energy loss per unit volume is

µ ¶ XX µ ¶
1 dW 1 dWif
~ω = ~ω f (Ei ) [1 − f (Ef )]
N (2a)3 dt N (2a)3 dt
if Si Sf
~ω 4π ³ e ´2 X
= 3
|A|2 |hf | ei~q·~r ~a0 · p~ |ii|2
N (2a) ~ mc
if
× δ(Ef − Ei − ~ω) f (Ei ) [1 − f (Ef )] , (7.10)
where f (E) is the Fermi distribution function. The term f (Ei ) is the probability
that the initial state was occupied and the term [1 − f (Ef )], is the probability that
the final state is empty.
Then equating (7.10) to the result of (7.8) and solving for ε2 (~q, ω) one has
8π 2 ³ e ´2 X
ε2 (~q, ω) = |hf | ei~q·~r ~a0 · p~ |ii|2
N (2a)3 mω
if
× δ(Ef − Ei − ~ω) f (Ei ) [1 − f (Ef )]. (7.11)
Equation (7.11) is the fundamental relation between macroscopic electromag-

netic theory and the energy band states of quantum theory. Once ε2 is known, all
of the optical constants can be derived from it.
The real part of the dielectric function ε1 is calculated from ε2 by means of
Kramers–Kronig dispersion relation:
Z ∞ 0
2 ω ε2 (~q, ω 0 ) 0
ε1 (~q, ω) = 1 + P dω , (7.12)
π 0 ω 02 − ω 2
where P indicates the principal value.
The complex index of refraction N (~q, ω) is related to ε = ε1 + i ε2 by
N 2 (~q, ω) = ε(~q, ω) (7.13)

N (~q, ω) = n(~q, ω) + i κ(~q, ω) (7.14)
where n and κ being the index of refraction and the extinction coefficient, respec-
tively. The absorption coefficient, α, is given by
(ω/c) ε2 (~q, ω)
α(~q, ω) = . (7.15)
n(~q, ω)
Finally, we note that for radiation incident normal to the solid surface the
reflectance is
¯p ¯
¯ ε(~q, ω) − 1 ¯2
¯ ¯
R(~q, ω) = ¯ p ¯ . (7.16)
¯ ε(~q, ω) + 1 ¯
7.2 Qualitative theory of ε2 (ω)
In the preceding section we saw that the dielectric function depends upon the
transition matrix elements Mif where
Mif = hf | ei~q·~r ~a0 · p~ |ii. (7.17)
The initial and final states are to be energy band states which are of the form
1 X ~ ~
ψ~kν (~r) = √ ajα (~k, ν) eik·Rmj ξα (~r − R
~ mj ) (7.18)
N mjα
where ajα (~k, ν) are the eigenvector components of the band state with wavevector
~k and band index ν. The quantity R ~ mj = R ~ m + ~τj , where R
~ m is the position vector
of the mth unit cell and ~τj locates the jth atom relative to the origin of the mth
~ mj ) is a Löwdin orbital with symmetry index α,
unit cell. The function ξα (~r − R
~
centered at Rmj .
The matrix element Mif ≡ M~kν,~k 0 ν 0 is
Z
M~kν,~k 0 ν 0 = ψk~0 ν 0 (~r)∗ ei~q·~r (~a0 · p~) ψ~kν (~r) d~r . (7.19)
Using (7.18) it is easily shown that M~kν,~k0 ν 0 involves a factor

X 1 ~ ~0 ~
X
ei 2 [k−k +~q]·R = N δ~k+~q−k~0, G
~ (7.20)
~
R ~
G
where G ~ is any reciprocal lattice vector including zero. Thus, M~ ~ 0 0 vanishes un-
kν,k ν
less k~ 0 = ~k + ~q ± G.
~ The wavevector of the light, ~q, is in most cases small compared
with any G ~ 6= 0. For example, the |~q| for a 1-eV photon is about 5 × 104 cm−1 , while
|G| = 2π/a ' 2 × 108 cm−1 . Thus M~ ~ 0 0 will be small unless k~ 0 ∼
~
kν,k ν = ~k ± G.
~ How-
ever, if both k~ 0 and ~k are confined to the interior of the first Brillouin zone, then
their vector difference or sum can not be equal to a (non-zero) reciprocal lattice vec-
tor. Thus the only non-zero matrix elements are those for which k~ 0 ∼ = ~k and G ~ =0
in (7.20). A schematic of the situation is shown in Fig. 7.1. It is apparent that the
interband transitions are essentially “vertical” on an energy band diagram such as
that of Fig. 7.1. On the other hand, the intraband transition is nearly “horizontal”.
If ~ω = (E~k 0 ν 0 − E~kν ) and is large compared with (E~k 0 ν − E~kν ) or (E~k 0 ν 0 − E~kν 0 )
then the intraband transition is far from “resonance” and can not occur. In the case
of insulating perovskites the energy gap between the filled valence bands and the
empty conduction bands is about 3 eV. Interband transitions can occur then only
for ~ω & 3 eV. Intraband transitions are not possible because there are no empty
final states in the valence bands and no occupied initial states in the conduction
bands.
intraband
transition E~k′ ν ′
U ?
6 6
(E~k′ ν ′ − E~kν ′ )
(E~k′ ν ′ − E~kν ) interband

Energy
transition

- q E~kν
~k ~k ′ ~k
Figure 7.1. Interband and intraband transitions.
For metallic perovskites such as alkali tungsten bronzes or ReO3 or for n-type
semiconducting materials the conduction band is partially occupied. This allows the
possibility of intraband transitions at low photon energies; namely, ~ω ≈ |E~k+~q,ν −
E~k,ν |. Such intraband transitions lead to what is called “free-carrier” absorption.
In addition, in metals, collective excitations of the conduction electrons in
the form of plasmons can also absorb energy. For Nax WO3 or ReO3 , the plasmon
energy is about 2 eV while for n-type SrTiO3 with 1019 electrons/cm3 it is only a
few hundredths of an eV.
In view of the preceding discussion it is clear that for ~ω & 1 eV the optical
properties of the insulating perovskites are determined principally by interband
transitions. The same is true for the metallic systems except that plasmon absorp-
tion must also be considered. In either case, the magnitude of the photon wavevector
may be neglected when ~ω . 100 eV and the interband transitions are essentially
“vertical”. Consequently, the transition matrix elements that are required are
h~k, ν 0 | ~a0 · p~ |~k, νi ν 6= ν 0 . (7.21)
The expression for the dielectric function due to interband transition takes the
form
1 ³ e ´2 ³ 2π ´3 1 X ~ 0
ε2 (ω) ≡ ε2 (0, ω) = |hkν | ~a0 · p~ |~kνi|2
π mω 2a N
~
kνν 0
× δ(E~kν 0 − E~kν − ~ω) f (E~kν ) [1 − f (E~kν 0 )]. (7.22)
In the following subsection we shall show that the function, ε2 (ω), for the per-
ovskites has pronounced structure which reflects the nature of the van Hove singu-
larities in the electronic density of states.
(a) Joint density of states
Before launching into a detailed calculation of the matrix elements needed for ε2 (ω)
we shall explore a simple, but frequently employed model for the purpose of demon-
strating how the structure of the electronic density of states manifests itself in the
optical properties. We begin by expressing the matrix elements of the momentum
in terms of those of ~r by making use of the relation
im
hf |~
p|ii = (Ef − Ei ) hf |~r|ii . (7.23)
~
Then we write (7.22) in the form
1 ³ 2π ´3 1 X
ε2 (ω) = |Mνν 0 (~k)|2 (E~kν 0 − E~kν )2
(~ω)2 2a N
~
kνν 0
× δ(E~kν 0 − E~kν −~ω) f (E~kν )[1− f (E~kν 0 )] (7.24)
with
e2 ~ 0
|Mνν 0 (~k)|2 = |hk, ν |~a0 · ~r|~k, νi|2 . (7.25)
π
In (7.24) we have used Ef − Ei = E~kν 0 − E~kν . Because of the delta function in
(7.24) we may replace (E~kν 0 − E~kν )2 by (~ω)2 and obtain
³ 2π ´3 1 X
ε2 (ω) = |Mνν 0 (~k)|2 δ(E~kν 0 − E~kν − ~ω) f (E~kν ) [1 − f (E~kν 0 )] . (7.26)
2a N
~
kνν 0
Now consider an insulating perovskite for which the valence bands are filled
and the conduction bands are empty. In this case the only allowed transitions
are between an initial state in one of the valence bands to one of the unoccupied
conduction bands. With the convention that ν refers to a valence band and ν 0 to a
conduction band, f (E~kν ) = [1 − f (E~kν 0 )] = 1 and
³ 2π ´3 1 X
ε2 (ω) = |Mνν 0 (~k)|2 δ(E~kν 0 − E~kν − ~ω). (7.27)
2a N
~
kνν 0
In order to proceed further we need to know the transition matrix elements. The
~k dependence of Mνν 0 (~k) is discussed later in this chapter. For our purpose in this
section we shall simply replace the matrix element by its average value over the
Brillouin zone, hMνν 0 i. With this approximation we have
nX o³ 2π ´3 1 X
ε2 (ω) ' h|Mνν 0 |2 i δ(E~kν 0 − E~kν − ~ω). (7.28)
0
2a N
νν ~
k
P
The quantity N1 k δ[~ω − (E~kν 0 − E~kν )] is similar to the density of states function
defined in the preceding chapter by (6.8). However, in this case we have two energies,
E~kν 0 and E~kν rather than a single energy. The function is called the joint density of
states, abbreviated by “JDOS” and denoted by [J(ω)]νν 0 . It specifies the number of
pairs of valence- and conduction-band states with an energy difference in the range
between ~ω and ~ω + d(~ω).
We now show that the JDOS has the same van Hove singularities that the DOS
has and therefore that these structures are expected to be reflected in the optical
properties of the cubic perovskites.
The JDOS is easily calculated for interband transitions from non-bonding
bands to conduction bands. The non-bonding state energy is constant (indepen-
dent of ~k) for the model developed in Chapter 4. The initial and final state energies
to be considered are
E~kν = Ek for ν = σ 0 (sigma non−bonding bands),

0
= E⊥ for ν = π (pi non−bonding bands),
0 ∗
E~kν 0 = E~kσ∗ for ν = σ (sigma conduction bands),
0 ∗
= E~kπ∗ for ν = π (pi conduction bands). (7.29)
The JDOS for such transitions is

1 X 1 X
[J(ω)]α0 β ∗ = δ[E~kν 0 − E~kν − ~ω] = δ[~ω − (E~kν 0 − E~kν )]
N N
~
k ~
k
= ρβ∗ (~ω + Eα0 ) = ρβ (~ω + Eα0 ) (7.30)
where α0 = π 0 or σ 0 and β ∗ = π ∗ or σ ∗ . In particular, we have the JDOS for four

interband transitions:
[J(ω)]σ0 σ∗ = ρσ (~ω + Ek ) (threshold for σ 0 → σ ∗: ~ω = Ee − Ek ),

[J(ω)]σ0 π∗ = ρπ (~ω + Ek ) (threshold for σ 0 → π ∗: ~ω = Et − Ek ),
[J(ω)]π0 σ∗ = ρσ (~ω + E⊥ ) (threshold for π 0 → σ ∗: ~ω = Ee − E⊥ ),
[J(ω)]π0 π∗ = ρπ (~ω + E⊥ ) (threshold for π 0 → π ∗: ~ω = Et − E⊥ ).
(7.31)
The functions ρπ and ρσ are the DOS functions given in Chapter 6 by (6.28)
and (6.62), respectively.
Equation (7.31) shows that in the case of transitions from the non-bonding
bands, the JDOS degenerates into a DOS function and therefore possesses the
same critical structure as the DOS. This result is not limited to transitions from
the narrow, non-bonding bands, but is also true for transitions from the σ or π
valence bands as well. For example, the JDOS for interband transitions from the
π(αβ) valence band to a π ∗ (αβ) conduction band (αβ = xy, xz, or yz) is
1 X
[J αβ (ω)]ππ∗ = δ[~ω − (E~αβ∗ − E~αβ )]
N kπ kπ
~
k
1 1 X 1
= − Im αβ
. (7.32)
π N
~
~ω − (E~ ∗ − E~αβ ) + i0+
k kπ kπ
The calculation is greatly simplified if we add a null term to (7.32); namely,

1 1 X 1
− Im . (7.33)
π N
~
~ω + (E~ ∗ − E~αβ ) + i0+
αβ
k kπ kπ
The term of (7.33) is zero for ~ω > 0, because the denominator can not vanish.
Now, combining (7.32) and (7.33) and using
(E~αβ∗ − E~αβ )2 = (Et − E⊥ )2 + 16(pdπ)2 (Sα2 + Sβ2 ), (7.34)

kπ kπ
we find
1 1 X 2~ω
[J αβ (ω)]ππ∗ = − Im
π N (~ω) − (Et − E⊥ ) − 16(pdπ)2 (Sα2 + Sβ2 ) + i0+
2 2
~k
1 1
( 2 ~ω) 1 X 1
=− Im (7.35)
π 2(pdπ)2 N W + C2α + C2β + i0+
~
k
where
( 12 ~ω)2 − [ 12 (Et − E⊥ )]2
W ≡ − 2. (7.36)
2(pdπ)2
The sum in (7.35) converges to an integral for large N and is easily evaluated by
using (6.20) and (6.26). The result is
µr ³ W ´2 ¶ µ ³ W ´2 ¶
αβ ( 21 ~ω) 1
[J (ω)]ππ =
∗ K 1− Θ 1− . (7.37)
2(pdπ)2 π 2 2 2
It will be recalled from Chapter 6 that the complete elliptic integral K(x) has a
jump discontinuity of π/2 at x = 0 and a logarithmic infinity at x = 1. For the K
function in (7.37) this means that the jump discontinuity occurs at W/2 = ±1 and
the logarithmic singularity at W/2 = 0. For W/2 = ±1 we have

p
~ω = 2 [(Et − E⊥ )/2]2 + 8(pdπ)2 (for the plus sign),
~ω = (Et − E⊥ ) (for the minus sign). (7.38)
The first energy corresponds to the energy separation of the π and π ∗ band at
R and the second energy of (7.38) to the energy difference between the π and
π ∗ band at Γ. That is, at the
p top and the bottom of the bands. The singularity at
W/2 = 0 occurs for ~ω = 2 [(Et − E⊥ )/2]2 + 4(pdπ)2 . This corresponds to vertical
transitions from π(αβ) to the π ∗ (αβ) band states whose wavevector is such that
Sα2 + Sβ2 = 1. Wavevectors satisfying this condition arise from the surface of a rod
of square cross-section oriented along kγ with corners at the X symmetry points in
the kα –kβ plane. Since the π(αβ) and π ∗ (αβ) bands are mirror image of each other
with respect to the mid-gap, the shape of the JDOS is similar to the DOS functions,
but twice as wide. A similar result can be obtained for the JDOS corresponding
to the σ → σ ∗ transition. Figure 7.2(a) shows the JDOS for all the possible 45
interband transitions and the total of all transitions is indicated by the dashed line.
Figure 7.2(b) shows a comparison of the (total JDOS)/ω 2 with the experimental
result of ε2 (ω) for SrTiO3 [1].
JDOS functions can be derived for the many different types of interband tran-
sitions and each contribution to ε2 reflects a convolution of the critical structure
of the bands involved. However, it should be remembered that the approximation
which leads to the JDOS result does not take into account the ~k dependence of the
transition matrix elements. Imbedded in the transition matrix elements are selec-
tion rules and shapes that can possibly modulate and smooth out sharp structure
that occurs in the JDOS. In the next section we develop a more detailed model
that takes the ~k dependence into account.
(b) LCAO transition matrix elements
The optical response of a cubic crystal is isotropic and consequently we may choose a
particular direction of the polarization of the electric field without loss of generality.
We assume that the polarization of A ~ and E~ is along the x-axis:
~ = A ~ex .
A
The interband matrix elements of concern are then

D ¯ ¯ E X X
~k, ν 0 ¯¯ − i~ ∂ ¯¯~k, ν = [aj 0 α0 (~k, ν 0 )]∗ ajα (~k, ν)
~ ~ ~
eik·(Rmj −Rm0 j 0 )
∂x
jj 0 αα 0 ~ m ,R
R ~ 0
m
Z µ ¶
∂ £ ¤
× d~r ξα∗ 0 (~r) −i~ ξα ~r − (R ~m − R ~ m0 ) − (~τj − ~τj 0 ) . (7.39)
∂x
(a) (1) π0 → π∗
10 (2) σ0 → π∗
(3) π → π∗
(4) π0 → σ∗
8 (5) σ0 → σ∗
JDOS(ω)
(6) σ → σ∗
6 (7) all others
1 (8) total × 12
4
8
2 2 4 7
5
3 6
0
(b)
10
8
JDOS(ω)/ω 2
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Photon energy (eV)
Figure 7.2. (a) Contributions to JDOS. (b) Comparison of JDOS/ω 2 with the experi-
mental result for ε2 (ω) (dashed line) for SrTiO3 [1].
The integral on the right-hand side of (7.39) is the matrix element of the x-
component of the momentum operator between Löwdin orbitals centered at the
origin and at (R ~ mj − R
~ m0 j 0 ). The integral depends on the vector difference R
~s =
~ ~ ~ ~
(Rm − Rm0 ) but not on the individual unit-cell vectors, Rm and Rm0 . The symbol
P ~ ~
~ s denotes a sum over Rm − Rm0 . It is convenient to define the matrix elements
R
of the momentum operator between Löwdin orbitals by
Z µ ¶
~ s) ≡ ∂ ~ s − (~τj − ~τj 0 )] .
Pαx0 j 0 ,αj (R d~r ξα∗ 0 (~r) −i~ ξα [~r − R (7.40)
∂x
Then, the matrix element of (7.39) takes the form

D ¯ ¯ E X h (~kν 0 ) i∗ (~k,ν) X ~ ~
~kν 0 ¯¯ − i~ ∂ ¯¯~kν = a j 0 α0 ajα ~ s) .
eik·[Rs −(~τj −~τj0 )] Pjx0 α0 ,jα (R (7.41)
∂x 0 0
jj αα ~s
R
According to (7.39), the interband transition matrix element is a sum of contribu-

tions of electron transitions between localized Löwdin orbitals. The optical excita-
tion of an electron between the Löwdin orbitals is, according to (7.40), dependent
upon the overlap of the final state orbital with the derivative of the initial state
orbital. Because of the localized character of the Löwdin orbitals one expects that
this overlap will decrease rapidly with increasing distance between the two orbitals.
Therefore the optical transitions are determined by electron excitation between
Löwdin orbitals centered on sites near one another.
A qualitative LCAO theory similar to that employed for energy bands in Chap-
ter 4 will be developed in the remainder of this chapter. This model considers only
the optically induced transition between Löwdin orbitals centered on the same site
and those between cation and anion orbitals on adjacent atomic sites.
There are two types of electron excitations to consider; site-diagonal and
nearest-neighbor transitions. The site-diagonal transitions are between Löwdin or-
bitals on the same atomic site and involve the integrals
Z µ ¶
∗ ∂
d~r ξα0 (~r) −i~ ξα (~r) (7.42)
∂x
while the nearest-neighbor transitions, involving p to d (or d to p) transitions, are
of the form
Z µ ¶
∂
d~r ξα∗ 0 (~r) −i~ ξα (~r ± a~ej ) (7.43)
∂x
where ~ej is a unit vector along any of the coordinate axes.
The simple LCAO band model we have been employing in previous chapters
includes only the d orbitals on the B-ion sites and the p orbitals on the anion sites.
Consequently, the only types of site-diagonal transitions that can occur in our model
are “d to d” or “p to p” transitions. It can easily be shown by symmetry arguments
that the matrix elements of (7.42) for such transitions vanish. If other orbitals,
representing core states or higher-lying atomic states, were included in the basis
set for the energy bands then site-diagonal transitions would occur. For example,
the matrix elements for nd to (n + 1)p or (n − 1)p to nd orbital transitions on a
B-cation site do not vanish. Such transitions are indeed important for the optical
properties of the perovskites; however, the photon energy at which they occur is
usually higher than for the p to d transitions between adjacent ions. The effect
of neglecting these higher-energy site-diagonal transitions is to limit the validity
of our description of ε2 (ω) to a photon energy range below the threshold for such
transitions. The method we shall use in the remainder of this chapter to discuss
interband transitions can easily be generalized to include site-diagonal transitions,
but we shall not include them in our discussion.
We now focus our attention on the p to d (or d to p) transitions between
nearest-neighbor ions. To determine the character of these matrix elements we first
consider their symmetry properties. The Löwdin orbitals (as in Table 3.1) are of
the form:
r
3 ³ rα ´
ξp(α) (~r) = rRp (r)
4π r
r
5 √ ³ rα rβ ´ 2
ξd(αβ) (~r) = 3 r Rd (r) (t2g orbitals) (7.44)
4π r2
r µ ¶ 
5 1 3z 2 − r2 
ξd(z2 ) (~r) = 2
r2 Rd (r) 

4π 2 r 
r √ µ 2 ¶ (eg orbitals)
5 3 x − y2 

ξd(x2 ) (~r) = 2
r Rd (r) 

4π 2 r2
where rα is the αth Cartesian component of ~r, r = |~r| and
R Rp (r) and Rd (r) are
spherically symmetric radial functions normalized so that ξ ∗ ξ d~r = 1.
We are concerned with integrals of the type
Z µ ¶
∗ ∂
d~r ξd (~r) −i~ ξp (~r ± a~ej ) , (7.45)
∂x
Z µ ¶
∗ ∂
d~r ξp (~r) −i~ ξd (~r ± a~ej ) , (7.46)
∂x
where ξd and ξp are any of the d or p orbitals, respectively.
Integration by parts shows that the matrix elements are Hermitian so that
Z µ ¶ ·Z µ ¶ ¸∗
∗ ∂ ∗ ∂
d~r ξp (~r) −i~ ξd (~r ± a~ej ) = d~r ξd (~r ± a~ej ) −i~ ξp (~r) . (7.47)
∂x ∂x
Also by a shift of the origin:
Z µ ¶ Z µ ¶
∗ ∂ ∗ ∂
d~r ξd (~r) −i~ ξp (~r ± a~ej ) = d~r ξd (~r ∓ a~ej ) −i~ ξp (~r) , (7.48)
∂x ∂x
and again by symmetry conditions
Z µ ¶ Z µ ¶
∂ ∂
d~r ξd∗ (~r + a~ej ) −i~ ξp (~r) = d~r ξd∗ (~r − a~ej ) −i~ ξp (~r) . (7.49)
∂x ∂x
Therefore we need only investigate the symmetry properties of the integral on the
right-hand side of (7.48). To proceed further in the analysis it is convenient to define
a fictitious set of localized orbitals which have the angular symmetries of atomic
orbitals. We define a set of “bar” orbitals with s and d atomic-orbital symmetries

by
√
∂ 5
R̄p (r) ≡ r Rp (r) ≡ R̄d (r) (7.50)
∂r 2
in terms of which
r
3 ³ 1 ´
s̄(~r) ≡ Rp (r) + R̄p (r) (7.51)
4π 3
r
3 rα rβ
d¯αβ (~r) ≡ 2 2 R̄p (r) (7.52)
4π r
r
3 1 (3z 2 − r2 )
d¯z2 (~r) ≡ √ R̄p (r) (7.53)
4π 3 r2
r
3 (x2 − y 2 )
d¯x2 (~r) ≡ R̄p (r) . (7.54)
4π r2
Now consider the derivative of the Löwdin orbital;
r ³ ´
∂ 3 xrα
ξp(α) (~r) = Rp (r) δαx + 2 R̄p (r) . (7.55)
∂x 4π r
Then using the bar orbitals of (7.51)–(7.54) we have
∂ 1 1
ξp(x) (~r) = s̄(~r) − √ d¯z2 (~r) + d¯x2 (~r) (7.56)
∂x 2 3 2
∂ 1¯
ξp(y) (~r) = dxy (~r) (7.57)
∂x 2
∂ 1¯
ξp(z) (~r) = dxz (~r) . (7.58)
∂x 2
Equations (7.56)–(7.58) show that the derivatives of the ξp(α) orbitals have the
angular symmetries of s and d atomic orbitals. Consequently, we can use the Slater–
Koster results in Table 3.2 to determine which matrix elements vanish by symmetry
and also to express the various non-vanishing elements in terms of a minimum set
of parameters. To do this we need to define four parameters which are analogous
to the Slater–Koster overlap parameters. We define
Z
1 ∗
(s̄dσ) ≡ ξd(z 2 ) (~
r ± a~ez ) s̄(~r) d~r , (7.59)
a
Z
1 ¯
(ddσ) ≡ ξd(z∗
r ± a~ez ) d¯z2 (~r) d~r ,
2 ) (~ (7.60)
a
Z
1 ¯ ∗
(ddπ) ≡ ξd(xy) (~r ± a~ex ) d¯xy (~r) d~r , (7.61)
a
Z
1 ¯ ∗
(ddδ) ≡ ξd(xy) (~r ± a~ez ) d¯xy (~r) d~r . (7.62)
a
All of the integrals of the type in (7.48) can be expressed in terms of the four
parameters (s̄dσ), (ddσ),¯ ¯
(ddπ), ¯
and (ddδ), with the help of Table 3.2. As examples,
Z µ ¶ Z
∂ 1
∗
ξd(xz) (~r ± a~ex ) −i~ ξp(z) (~r) d~r = −i~ ξd(xz)∗
(~r ± a~ex ) d¯xz (~r) d~r
∂x 2
−i~ ¯
= (ddπ)
2a
Z µ ¶ Z
∗ ∂ ∗
ξd(z2 ) (~r ± a~ez ) −i~ ξp(x) (~r) d~r = −i~ ξd(z 2 ) (~
r ± a~ez )
∂x
· ¸
1 ¯ 1¯
× s̄(~r) − √ dz2 (~r) + dx2 (~r) d~r
2 3 2
· ¸
i~ 1 ¯
=− (s̄dσ) − √ (ddσ)
a 2 3
Z µ ¶ Z
∂
∗
ξd(x2 ) (~r ± a~ey ) −i~ ξp(y) (~r)d~r = −i~ ξd(x ∗
r ± a~ey ) d¯xy (~r) d~r = 0 .
2 ) (~
∂x
Proceeding in this manner it is easily found that there are only four non-zero
types of integrals:
Z µ ¶
∗ ∂
ξd(xy) (~r ± a~ej 0 ) −i~ ξp(y) (~r) d~r , (7.63)
∂x
Z µ ¶
∗ ∂
ξd(xz) (~r ± a~ej 0 ) −i~ ξp(z) (~r) d~r , (7.64)
∂x
Z µ ¶
∗ ∂
ξd(z 2) (~r ± a~
e j 0 ) −i~ ξp(x) (~r) d~r , (7.65)
∂x
Z µ ¶
∗ ∂
ξd(x 2 ) (~
r ± a~ej 0 ) −i~ ξp(x) (~r) d~r . (7.66)
∂x
The integrals of (7.63)–(7.66) involve a d orbital of (α0 β 0 ) type located at

~r = ±a~ej 0 and a p orbital of the γ type at the origin, where (α0 β 0 ) = xy, xz, x2, or
z 2 and γ = x, y, or z. Therefore, dropping also the superscript x of P x , the notation
for the momentum matrix elements can be simplified as follows:
Z µ ¶
∗ ∂
ξd(α 0 β0 ) (~
r − a~
e j 0 ) −i~ ξp(γ) (~r) d~r ≡ P(α0 β 0 )j 0 ,γ . (7.67)
∂x
The phase factors appearing in (7.41) are simply e±ikj 0 a since we are including
only the nearest-neighbor interactions. Finally, the transition matrix elements of
(7.41) may be written as
D ¯³ ´¯ E X h (~kν 0 ) i∗ (~kν)
~kν 0 ¯¯ − i~ ∂ ¯¯~kν = a(α0 β 0 ) a(j 0 γ) P(α0 β 0 )j 0 ,γ 2 cos(kj 0 a) . (7.68)
∂x 0 0 0
(α β )j ,γ
Table 7.1. Momentum operator matrix elements P(α0 β 0 )j 0 ,γ (defined in (7.67)) for
nearest-neighbor Löwdin orbitals, in units of (−i~/2a).
(α0 β 0 ) γ j0 = x j0 = y j0 = z
(xy) y ¯
(ddπ) ¯
(ddπ) ¯
(ddδ)
(xz) z ¯
(ddπ) ¯
(ddδ) ¯
(ddπ)
√
(z 2 ) x −(s̄dσ) − 1 ¯
(ddσ) −(s̄dσ) − 1 ¯
(ddσ) − 3 ¯ 1 ¯
2 (ddδ) 2(s̄dσ) − (ddσ)
√ √ √
4 3 4 3 3
√ √
(x2 ) x ¯
3(s̄dσ) + (ddσ) ¯
− 3(s̄dσ) + (ddσ) + 1 ¯ ¯
2 (ddδ) (ddδ)
In order to proceed with the analysis it is necessary to specify the bands in-
volved in the interband transition. There are a large number of possible transitions
corresponding to just the p to d and d to p transitions. For example, for an insu-
lating perovskite the nine valence bands (three π, two σ, three π 0 , and one σ 0 ) are
occupied and the five conduction bands (three π ∗ and two σ ∗ ) are empty. There are
45 possible interband transitions; transitions from any of the nine valence bands to
any of the five conduction bands. The contributions to ε2 (ω) due to some interband
transitions are shown in Fig. 7.3. For a given polarization of the electric field many
of the 45 transitions are forbidden and many are symmetry equivalent so that the
number of transition matrix elements that must be considered is greatly reduced.
Nevertheless, the number of distinct transitions is still quite substantial.
In the following sections we consider the character of different interband tran-
sitions.
7.3 Interband transitions from non-bonding bands
There are four types of interband transitions from the non-bonding bands to the
conduction bands: π 0 → π ∗ , π 0 → σ ∗ , σ 0 → π ∗ , and σ 0 → σ ∗ .
(a) π 0 → π ∗ interband transition
There are three π 0 and three π ∗ bands; π 0 (αβ) and π ∗ (αβ) where αβ = xy, xz,
or yz. Thus there are nine possible π 0 → π ∗ transitions; arising from π 0 (αβ) and
π ∗ (αγ). For an x-polarized electric field, Table 7.1 shows that non-zero matrix
elements occur only for transitions between an dxy orbital and a py orbital and
between an dxz orbital and a pz orbital. Consequently, there are only four allowed
Label : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
6 6 6 66
6 6
6 6 6
Ee
6 6
66
66
6
Et 6 6
E⊥
Ek
π0 , σ0 → π∗ , σ∗ π → π∗ , σ∗ σ → π∗ , σ∗
Γ X M R
Figure 7.3. Major contributions to optical transitions.
π 0 → π ∗ transitions:
π 0 (xy) → π ∗ (xy) (7.69a)

0 ∗
π (xz) → π (xz) (7.69b)
0 ∗
π (yz) → π (xy) (7.69c)
0 ∗
π (yz) → π (xz) . (7.69d)
The restriction of the transitions to those in (7.69) shows that selection rules
for optical transitions are contained in the transition matrix elements.
Transitions (a) and (b) of (7.69) are “unmixed” transitions in that they involve
initial and final band states derived from the same 3×3 block of the Hamiltonian
(see Chapter 4). Transitions (7.69c) and (7.69d) are “mixed” transitions that in-
volve initial and final states from different 3×3 blocks of the Hamiltonian. The
unmixed transitions, (7.69a) and (7.69b), are symmetry-equivalent and make iden-
tical contributions to ε2 (ω). Similarly, the mixed transitions (7.69c) and (7.69d) are
also symmetry-equivalent. Consequently, it is necessary only to consider (a) and (c).
The π 0 band wavefunctions have zero amplitudes for the d orbitals. Therefore
the only non-vanishing matrix element between Löwdin orbitals that contributes
x
to (a) is P(xy)x,y . According to (7.68) the transition matrix element for (a) is
D ¯³ ´¯ E h i∗
~kπ ∗ (xy)¯¯ − i~ ∂ ¯¯~kπ 0 (xy) = a[~kπ∗ (xy)] a[~kπ0 (xy)] P(xy)x,y 2 Cx . (7.70)
xy y
∂x
For the mixed interband transitions, (c), we find

D ¯³ ´¯ E h i∗
~kπ ∗ (xy)¯¯ − i~ ∂ ¯¯~kπ 0 (yz) = a[~kπ∗ (xy)] a[~kπ0 (yz)] P(xy)z,y 2 Cz . (7.71)
xy y
∂x
In calculating ε2 (ω) the matrix elements of (7.70) and (7.71) should be multi-
plied by 2 in order to account for the equivalent transitions (7.69b) and (7.69d).
(b) π 0 → σ ∗ interband transitions
There are six possible π 0 → σ ∗ interband transitions; π 0 (αβ) → σ ∗ (±), where

σ ∗ (+) and σ ∗ (−) are the two distinct branches of the σ ∗ conduction bands.
The wavefunctions for the σ ∗ band states involve the x2 and z 2 type of d
orbitals and hence, according to Table 7.1, have matrix elements only with the x
type of p orbital. This limits the allowed transitions to those between π 0 (xy) or
π 0 (xz) and σ ∗ (±). The four allowed transitions are:
π 0 (xy) → σ ∗ (±) (7.72a)

0 ∗
π (xz) → σ (±). (7.72b)
The two types of transitions represented by (7.72a) and (7.72b) are equivalent
and hence we need only consider (7.72a). The interband transition matrix elements
are:
D ¯³ ´¯ E
~kσ ∗ (±)¯¯ − i~ ∂ ¯¯~kπ 0 (xy)
½h ∂x ¾
i∗ h ~ ∗ i∗
[~
kσ ∗ (±)] [kσ (±)] ~ 0
= az 2 P(z2 )x,x + ax2 P(x2 )x,x a[xkπ (xy)] 2 Cx . (7.73)
In calculating ε2 (ω) these matrix elements should be multiplied by 2 to account

for the symmetry-equivalent transitions of (7.72b).
(c) Tabulation of interband transition matrix elements
Proceeding in the manner described above, the matrix elements for all of the inter-
band transitions can easily be found. The results for all of the distinct transitions
are given in Table 7.2. The first two columns specify the initial and final bands.
Table 7.2. Interband transition matrix elements. Labels refer to the assignments in Fig. 7.3. ~k dependence of amplitudes are
dropped for short-hand.
∂ ~
ν ν0 Label Weight h~
kν 0 | − i~ ∂x |kνi
¡ ∗ (xy)
¢∗ π
¡ 0 (xy) ¢
π 0 (xy) π ∗ (xy) 1 2 aπ
xy P(xy)x,y ay 2 Cx
¡ ¢
π ∗ (xy) ∗
¡ π 0 (yz)
¢
π 0 (yz) π ∗ (xy) 1 2 axy P(xy)z,y ay 2 Cz
n³ ´∗ ³ ´∗ o³ ´
σ ∗ (±) σ ∗ (±) 0 (xy)
π 0 (xy) σ ∗ (±) 3, 2 2 a P(z2 )x,x + a P(x2 )x,x aπ
x 2 Cx
z2 x2
¡ ¢
π ∗ (xy) ∗
¡ σ0
¢
σ0 π ∗ (xy) 4 2 axy P(xy)y,y ay 2 Cy
n³ ´∗ ³ ´∗ o¡ ¢
σ ∗ (±) σ ∗ (±) 0
σ0 σ ∗ (±) 6, 5 1 a P(z2 )x,x + a P(x2 )x,x aσ
x 2 Cx
z2 x2
n¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢∗ o
156
π ∗ (xy) ∗ π(xy) π(xy) ∗ π ∗ (xy)
π(xy) π ∗ (xy) 9 2 axy P(xy)x,y ay + axy P(xy)x,y ay 2 Cx
¡ π(xy)
¢ ∗
¡ ∗ (yz) ¢∗
π(xy) π ∗ (yz) 7 2 axy P(xy)z,y aπ
y 2 Cz
¡ ∗ (xy) ¢∗ ¡ π(yz)
¢
π(yz) π ∗ (xy) 8 2 aπ
xy P(xy)z,y ay 2 Cz
n³ ´∗ ³ ´∗ o¡ ¢ ¡ ¢ ¡ ¢∗
σ ∗ (±) σ ∗ (±) ∗ ∗ (±)
π(xy) σ ∗ (±) 13, 12 1 a P(z2 )y,x + a P(x2 )y,x aπ(xy)
x 2 Cy + aπ(xy)
xy P(xy)y,y aσ
y 2 Cy
z2 x2
n³ ´∗ ³ ´∗ o¡ ¢ ¡ ¢ ¡ ¢
σ ∗ (±) σ ∗ (±) π(xz) π(xz) ∗ σ ∗ (±) ∗
π(xz) σ ∗ (±) 11, 10 1 a P(z2 )z,x + a P(x2 )z,x ax 2 Cz + axz P(xz)z,z az 2 Cz
z2 x2
n³ ´ ³ ´ o¡ ¢∗ ¡ ¢∗ ¡ ¢
σ(±) ∗ σ(±) ∗ ∗ (xy) (xy) ∗
σ(±) π ∗ (xy) 15, 14 1 a P(z 2 )y,x + a P(x 2 )y,x aπ
x 2 Cy + aπ
xy P(xy)y,y aσ(±)
y 2 Cy
z2 x2
n³ ´∗ ³ ´∗ o¡ ¢ n¡ ¢ ¡ ¢ o¡ ¢∗
σ ∗ (±) σ ∗ (±) σ(±) ∗ σ(±) ∗ ∗ (±)
σ(±) σ ∗ (±) 16–19 1 a P(z2 )x,x + a P(x2 )x,x aσ(±)
x 2 Cx + a P(z 2 )x,x + a P(x2 )x,x aσ
x 2 Cx
z2 x2 z2 x2
7.4 Frequency dependence of ε2 (ω) for insulating and semiconducting perovskites 157
The third column labels the inequivalent transitions and the forth column, labeled
“weight”, gives the number of symmetry-equivalent transitions. In calculating ε2 (ω)
the matrix element should be multiplied by the weight factor. In Table 7.2 it should
be noted that the row corresponding to π 0 (xy) → σ ∗ (±) specifies two inequivalent
transitions and the row corresponding to σ(±) → σ ∗ (±) specifies four inequivalent
transitions.
7.4 Frequency dependence of ε2 (ω) for

insulating and semiconducting perovskites
In the preceding section we obtained the forms of the matrix elements for interband
¯
transitions in terms of the parameters (s̄dσ), (ddσ), ¯
(ddδ), ¯
and (ddπ). The matrix
elements may be used to calculate ε2 (ω) from (7.22). However, it should be kept in
mind that the model we are employing includes only matrix elements between 2p
and nd orbitals located on adjacent cation and anion sites.1
The goal of this section is to describe the dominant optical properties of the in-
sulating and semiconducting cubic perovskites. For photon energies less than about
10 eV these properties are dominated by interband transitions. That is, transitions
in which an electron in one of the valence-band states absorbs a photon and is pro-
moted to one of the conduction-band states. There are numerous other processes
that contribute to the optical properties. These include exciton, defect, impurity,
and free-carrier absorption. In addition, processes in which an electron and a col-
lective excitation such as a phonon are simultaneously excited. However, for the
range of photon energies being considered these other processes only add fine detail
or tend to broaden and smear out sharp structure associated with the interband
transitions. We do not discuss these other processes here.
The band-gap energy, Eg , for typical insulating perovskites (e.g., SrTiO3 ,
BaTiO3 or KTaO3 ) is 3–3.5 eV. The lowest-energy interband transitions are the
π 0 → π ∗ and σ 0 → π ∗ type transitions. These occur for a photon energy of ~ω in
the range Eg ≤ ~ω ≤ Eg + Wπ , where Wπ is the π ∗ -band width; 3–4 eV. Interband
transitions from the top of the π and σ bands to the π ∗ band also occur in this
range. In addition, transitions from π 0 and σ 0 to the bottom of the σ ∗ band may
1
The reader is cautioned not to take the language of this description literally. The model does not imply
that the actual optical transitions consist of transferring an electron from one ion to its neighbor. The
optical transitions involve electrons making transitions from one delocalized energy band state to
another delocalized energy band state. The method of evaluating the momentum integrals in terms of
nearest-neighbor Löwdin orbitals is a mathematical convenience which leads to a real-space description
of the contributions to the transition matrix elements. The same can be said for the band model.
Even though only nearest-neighbor matrix elements of the Hamiltonian are included, the resulting
states are delocalized energy band states. By contrast, in a strong correlation model, hopping of a
localized electrons between neighboring ions involves a substantial Coulomb repulsion energy and a
consideration of the orbital and spin quantum numbers of the initial and final atomic states.
occur in the upper portion of the previously mentioned photon energy range. Some
of these transitions are illustrated in Fig. 7.2.
At higher photon energies, Eg + Wπ ≤ ~ω ≤ Eg + W2 , transitions from the
bottom of the π or σ bands to the π ∗ and σ ∗ bands become possible, where W2
is 10–12 eV. In this range, however, other types of transitions not included in our
model can also occur. For example, interband transitions from the π 0 and σ 0 bands
into bands associated with the (n + 1)p and (n + 1)s states of the B ions and the
s states of the A ions can occur. Also, at even higher ~ω, transitions from the
core states of the oxygen and B ions become possible. For SrTiO3 these transitions
would be into the energy bands derived from the Ti(4p), Ti(4s), and Sr(4s) orbitals.
At lower energies, ~ω < Eg , there are a number of other processes that can
contribute to optical absorption, including excitonic state absorption, defect and
impurity state absorption, and low-frequency plasmon absorption. These processes
will add additional fine structure to the absorption spectrum, particularly in the
energy gap region.
In the following sections we shall derive approximate results for the frequency
dependence of ε2 (ω) using our model based on the 2p–nd transitions. The effect
of neglecting other types of transitions is to limit the validity of our description of
ε2 (ω) to values of ~ω less than the threshold for these other processes. The 2p–nd
transitions will dominate the optical properties for ~ω ≤ 10 eV. For perovskite met-
als, such as Nax WO3 or ReO3 , interband transitions can initiate from the partially
filled π ∗ bands.
(a) Band-edge behavior of ε2 (ω)
The onset of strong optical absorption in perovskite insulators or semiconductors

for ~ω & Eg is due to interband transitions from the non-bonding bands into the
empty π ∗ conduction bands. Usually the π 0 band is higher than the σ 0 band so that
for ~ω just above Eg the absorption is due to the π 0 → π ∗ interband transitions. In
this section we shall show that ε2 (ω) has an abrupt rise from zero to a large value
at ~ω = Eg . We shall also show that the band edge value of ε2 (ω) is determined by
¯
the parameter (ddπ).
With the help of (7.22) and Table 7.1 we find for the (unmixed) π 0 (xy) →
π ∗ (xy) transitions per spin state that
2 ³ e ´2 ³ 2π ´3 1
[ε2 (ω)]π0 (xy)→π∗ (xy) =
π mω 2a N
X ¯¯D ¯µ
¯
¶
∂ ¯¯~ 0 E¯¯2
× ¯ ~ ∗ ¯
¯ k, π (xy)¯ −i~ ∂x ¯k, π (xy) ¯
~
k
h i
× δ E~k,π∗ (xy) − E~k,π0 (xy) − ~ω
Z ¯ ¯
2 ³ e ´2 ¯ ~k,π∗ (xy)] [~k,π0 (xy)] ¯2
= d~k ¯a[xy ay ¯
π mω BZ
³ ~ ´2 h i
× ¯ 2 (2 Cx )2 δ E~ ∗
(ddπ) − (E ⊥ + ~ω) , (7.74)
2a k,π (xy)
where the integral is over the first Brillouin zone. In obtaining the final result in
(7.74) we have set
µ ¶3 Z
¡ ¢ ¡ ¢ 1 X 2a
f E~k,π0 (xy) = 1 − f E~k,π∗ (xy) = 1, = d~k,
N 2π BZ ~
k
and have used the flat band approximation E~k,π0 = E⊥ described in Chapter 4. The
initial factor of 2 on the right-hand side of (7.74) accounts for the two symmetry-
equivalent transitions and the factor of (~/2a)2 results from the units employed in
Table 7.1.
The required wavefunction amplitudes were determined in Chapter 4. They
are
£ ¤
~
k,π ∗ (xy)]
E⊥ − E~k,π∗ (xy)
a[xy = q£ ¤2 , (7.75)
E⊥ − E~k,π∗ (xy) + 4(pdπ)2 (Sx2 + Sy2 )
~ 0 Sy
a[yk,π (xy)]
= −q ; (Sα = sinkα a). (7.76)
Sx2 + Sy2
~
k,π ∗ (xy)
Because of the δ function, the amplitude axy may be evaluated at E~k,π∗ (xy) =
E⊥ + ~ω in the integrand of (7.74). Then we obtain the result
2 ³ e ´2 ~2 ¯ 2
[ε2 (ω)]π0 (xy)→π∗ (xy) = (ddπ)
π mω a2
Z h i
× d~k gxy,xy (~k, ω) δ E~k,π∗ (xy) − (E⊥ + ~ω) (7.77)
BZ
where
· ¸
Cx2 Sy2 (~ω)2
gxy,xy (~k, ω) ≡ . (7.78)
(Sx2 + Sy2 ) (~ω)2 + 4(pdπ)2 (Sx2 + Sy2 )
The integral of (7.77) is similar to those encountered in calculating the DOS

(density of states) functions in Chapter 6. For example, the DOS function for the
π ∗ band is
µ ¶3 Z h i
2a
ρπ∗ (E) = d~k δ E~k,π∗ − E . (7.79)
2π BZ
For E > E⊥ , ρπ∗ (E) is the same as the function ρπ (E) given by (6.28). It is evident
from (7.77) that gxy,xy (~k, ω) contains the k dependence of the interband transition
matrix elements and as may be seen from (7.78), it is not constant over the Brillouin
zone. On the other hand, gxy,xy (~k, ω) is a very smoothly varying function of ~k and
consequently any sharp structure in ρπ (E) will be replicated in ε2 (ω). As was shown
in Chapter 6, ρπ (E) has very pronounced structure due to the two-dimensional
behavior of the π bands. In particular, ρπ (E) possesses jump discontinuities at the
edges of the π bands and a logarithmic singularity at the center of the bands. These
van Hove singularities will also appear in ε2 (ω).
Let us consider in detail the behavior of ε2 (ω) for ~ω very near to the band-
gap energy, Eg = Et − E⊥ . Under this condition, the only possible π 0 (xy) → π ∗ (xy)
transitions are those which arise from a small cylinder oriented along the kz -axis
from the Γ to the X point in the Brillouin zone. Within this small cylinder, kx a
and ky a are small so that
Sy2
gxy,xy (~k, ω) ' . (7.80)
Sx2 + Sy2
The integral of (7.77) is unchanged if kx a and ky a are interchanged so that we

may make the substitution
Ã !
Sy2 1 Sx2 Sy2 1
2 2
→ 2 2
+ 2 2
= . (7.81)
Sx + Sy 2 Sx + Sy Sx + Sy 2
This gives
µ ¶3 4 2 ¯ 2
2π ~ e (ddπ)
[ε2 (ω)]π0 (xy)→π∗ (xy) →
2a πm2 a2 (~ω)2
½µ ¶3 Z ¾
2a £ ¤
× ~
dk δ E~kπ∗ − (E⊥ + ~ω) . (7.82)
2π BZ
In (7.82) the arrow signifies an equality in the limit as ~ω tends to Eg . Using (7.79)
we obtain
µ ¶3 4 2 ¯ 2
2π ~ e (ddπ)
[ε2 (ω)]π0 (xy)→π∗ (xy) → 2 2
ρπ∗ (E⊥ + ~ω) ,
2a πm a Eg2
1 ¯ 2 Θ(~ω − Eg ) ,
= χπ (ddπ) (7.83)
2
with
π~4 (e2 /a)
χπ ≡ (7.84)
2m2 a4 Eg (pdπ)2
where we have used (6.28) to evaluate ρπ∗ :

1 12 Eg
ρπ∗ (E⊥ + ~ω) = ρπ (E⊥ + ~ω) → K(0) Θ(~ω − Eg )
π 2 (pdπ)2
Eg
= Θ(~ω − Eg ) .
4π(pdπ)2
Equation (7.83) shows that ε2 (ω) due to the unmixed π 0 → π ∗ interband tran-
sition possesses a jump at ~ω = Eg , the magnitude of which is determined by the
¯
optical parameter (ddπ) and by the energy band parameters Eg and (pdπ). The
jump in ε2 (ω) decreases with increasing band gap. This is the opposite of the be-
havior of the JDOS, which increases with increasing band gap. On the other hand,
ε2 (ω) decreases as the square of (pdπ) which is the same dependence that the JDOS
has. Thus, we see that matrix-element effects are significant [2].
To obtain the total jump in ε2 (ω) at the band edge we must also include
the contributions from the mixed π 0 → π ∗ transitions. A calculation similar to that
described above, using (7.71) and Tables 7.1 and 7.2, gives for the mixed transitions
2 ³ e ´2 ~ 2 ¯ 2
[ε2 (ω)]π0 (yz)→π∗ (xy) = (ddδ)
π mω a2
Z h i
× d~k gyz,xy (~k, ω) δ E~kπ∗ (xy) − (E⊥ + ~ω) (7.85)
BZ
with
· ¸
Sy2 Cz2 (~ω)2
gyz,xy (~k, ω) = . (7.86)
(Sy2 + Sz2 ) (~ω)2 + 4(pdπ)2 (Sx2 + Sy2 )
For ~ω very near to the band-gap energy, Eg , the transitions arise from the same
cylindrical region described above. In that region Sx and Sy are small
Sy2 Cz2
gyz,xy (~k, ω) → .
(Sy2 + Sz2 )
In the integral of (7.85) only gyz,xy (~k, ω) contains a kz dependence and

Z π/2a Z π/2a
dkz cos2 (kz a)
dkz gyz,xy (~k, ω) = Sy2 2 2
−π/2a −π/2a Sy + sin (kz a)
π hq 4 i
= Sy + Sy2 − Sy2 . (7.87)
a
As Sy tends to zero, the result of (7.87) vanishes and hence the mixed transitions
make no contribution to ε2 (ω) as ~ω tends to Eg . Therefore, the contributions of
all of the π 0 → π ∗ interband transitions at the band gap, ~ω = Eg , produce a jump
in ε2 (ω) given by (7.83).
Later in this chapter (see (7.132)) we show that the contribution to ε2 (ω)
due to the π → π ∗ transitions at ~ω = Eg is exactly one-half of the contribution
from π 0 → π ∗ . Moreover, if E⊥ and Ek are close to each other then the contri-
butions by the σ 0 → π ∗ (see (7.113)) and σ(−) → π ∗ (see (7.139)) transitions at
~ω = Eg0 ' Eg should also be added. Therefore the total jump at ~ω = Eg will be
2.5 – 3 times that given by (7.83). It is of interest to estimate the jump in ε2 (ω). Us-
ing (e2 /a) ' 12 (e2 /2aH ) = (13.6/2) eV, Eg = 3.85 eV, (pdπ) = 1.15 eV, Egσ = 5.7 eV,
(pdσ) = −2.6 eV, m = 9.11 × 10−28 g, and ~ = 1.055 × 10−27 erg s we obtain a total
jump (∼ 2.78 times that of (7.83), and an extra factor of two for the two spin states)
at the band edge,
¯
¯ 2.
[ε2 (ω)]π,π0 →π∗ ¯~ω=Eg ' 21.1 (ddπ) (7.88)
To obtain a numerical value we need to estimate (ddπ). ¯ This dimensionless

parameter is the overlap integral between a Löwdin d orbital and a neighboring
fictitious d orbital derived from the derivative of a Löwdin p orbital. Therefore,
it should be a number between zero and one. Experimentally, Cardona [1] has
observed a very sharp rise in the optical absorption and reflectivity of BaTiO3 and
SrTiO3 . His result for ε2 (ω) is shown as dashed line in Fig. 7.4(b). The behavior of
ε2 (ω) for these two perovskites is quite similar, in agreement with our simple model.
There is an abrupt rise in ε2 (ω) for ~ω between 3 and 4 eV. It is not actually a
jump discontinuity because the π 0 bands are not actually dispersionless (flat) as we
have assumed here. The magnitude of ε2 (ω) at 4 eV is about 6 for either BaTiO3 or
¯
SrTiO3 . According to (7.88) this implies that (ddπ) is about 0.5. This value is large
for a nearest-neighbor overlap. However, the Löwdin orbitals are more extended
than atomic orbitals and the empirical parameter (ddπ) ¯ has to compensate for the
interactions of more distant neighbors that have been neglected in our model.
Experimentally, the index of refraction (real part) is about 2.5 at ~ω = Eg for
BaTiO3 or SrTiO3 [1]. According to (7.15) the absorption coefficient α is given by
ω ε2 (ω)
α(ω) = .
c n(ω)
If we set ~ω = Eg , n(Eg /~) ' 2.5 and use the value ε2 (Eg /~) ' 6 we find that
α ' 4.67 × 105 cm−1 which is also in good agreement with Fig. 7 of [1].
(b) Frequency dependence of ε2 (ω) from π 0 → π ∗ transitions
In the preceding section we obtained the contribution of the π 0 → π ∗ transition to

ε2 (ω) for ~ω ' Eg . To obtain ε2 as a function of ω we must evaluate the integrals in
(7.77) and (7.85) including the functions gxy,xy (~k, ω) and gyz,xy (~k, ω) which contain
the ~k dependence of the transition matrix elements. Consider first the unmixed
6 3
(a) π 0 (xy) → π ∗ (xy) (c) σ 0 → π ∗ (xy) (d) π(xy) → π ∗ (xy)
(b) π 0 (xy) → π ∗ (xz) (e) π(xy) → π ∗ (yz)
5 (f) π(yz) → π ∗ (xy) 2
4 d 1
Arbitrary units
e+f
3 0
Eg Eg +2Wπ
c
a
2 2
(g) σ(−) → π ∗ (xy)
(h) σ(+) → π ∗ (xy)
1 g 1
b
h
0 ′ ′ ′ ′ 0
Eg Eg +Wπ Eg Eg +Wπ Eg Eg +Wπ +Wσ
10
8
Arbitrary units
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Photon energy (eV)
Figure 7.4. (a) Major contributions to ε2 (ω) and (b) comparison with the experimental
result (dashed line) for SrTiO3 [1].
π 0 → π ∗ transitions. Referring to (7.77) one sees that the integrand is evaluated

only on the surface in ~k-space where ~ω = E~kπ∗ (xy) − E⊥ . On this surface,
~ω(~ω − Eg )
Sx2 + Sy2 = ≡ Ω2 (ω) (7.89)
4(pdπ)2
so that
· ¸
2 ³ e ´2 ~2 ¯ 2 (~ω)2 1
[ε2 (ω)] π 0 (xy)→π ∗ (xy) = (ddπ)
π mω a2 (~ω)2 + 4(pdπ)2 Ω2 (ω) Ω2 (ω)
Z h i
× d~k Cx2 Sy2 δ E~kπ∗ (xy) − (E⊥ + ~ω) . (7.90)
BZ
The factor, Cx2 Sy2 in the integrand may be written as

1 1 2
Cx2 Sy2 = 2Sy2 − (Sx2 + Sy2 )Sy2 − + C
4 4 2y
³ ´ 1 1 2
= 2 − Ω2 (ω) Sy2 − + C2y (7.91)
4 4
where Sα = sin kα a, Cα = cos kα a, and C2α = cos 2kα a. From symmetry consider-
ations, it follows that Sy2 may be replaced by 12 (Sx2 + Sy2 ) in the integrand of (7.90),
thus we find the integral of (7.90) is
½ ³ ´ ¾Z h i
1 1
2 2
2 − Ω (ω) Ω (ω) − d~k δ E~kπ∗ (xy) − (E⊥ + ~ω)
2 4 BZ
Z h i
1
+ d~k C2y
2
δ E~kπ∗ (xy) − (E⊥ + ~ω) . (7.92)
4 BZ
Using the expression
~ω(~ω − Eg ) £ ¤
λ(ω) ≡ 2
− 2 = 2 Ω2 (ω) − 1 , (7.93)
2(pdπ)
which is obtained from (6.19) at (~ω + E⊥ ), the first term of (7.92) may be imme-
diately evaluated from the definition of the density of states, ρπ∗ (E), to give
½ · ¸¾ µ ¶3
1 λ(ω)2 2π
1− ρπ∗ (E⊥ + ~ω) . (7.94)
4 2 2a
Calculation of the remaining term of (7.92) must be evaluated directly. The calcu-
lation is simplified by employing a mathematical manipulation.
£ ¤
We note that for ~ω > 0 the quantity δ E~kπ(xy) − (E⊥ + ~ω) vanishes since
the valence-band states all have energy E~kπ(xy) ≤ E⊥ . Therefore, we may add this
null term to the integrand of the second term of (7.92). We then write:
Z h i
d~k C2y
2
δ E~kπ∗ (xy) − (E⊥ + ~ω)
BZ
Z n h i h io
= d~k C2y
2
δ E~kπ∗ (xy) − (E⊥ + ~ω) + δ E~kπ(xy) − (E⊥ + ~ω)
BZ
½ Z Z π ¾
2(~ω − Eg /2) π cos2 r
= −Im dr dt
(2a)3 (pdπ)2 −π −π λ(ω) + cos t + cos r + i0+
Z π Z π
8π(~ω − Eg /2) £ ¤
= dr dt cos2 r δ λ(ω) + cos t + cos r . (7.95)
(2a)3 (pdπ)2 0 0
In obtaining the result of (7.95) we have employed the expressions for E~kπ∗ (xy) and
E~kπ(xy) , and defined r = 2ky a, t = 2kx a. The integration over t in (7.95) can be
performed with the result that
Z h i
d~k C2y
2
δ E~kπ∗ (xy) − (E⊥ + ~ω)
BZ
Z 1
8π(~ω − Eg /2) µ2 dµ Θ[1 − (λ + µ)2 ]
= p . (7.96)
(2a)3 (pdπ)2 −1 (1 − µ2 )[1 − (λ + µ)2 ]
Because of the Θ function the integral of (7.96) is non-vanishing only in the
region q
−2 < λ(ω) < 2. This range is equivalent to Eg < ~ω < Eg + Wπ , where
Wπ = ( 12 Eg )2 + 8(pdπ)2 − 12 Eg is the π ∗ -band width.
The integral may be evaluated in terms of complete elliptic functions and for
|λ(ω)| < 2 we find
Z h i
d~k C2y
2
δ E~kπ∗ (xy) − (E⊥ + ~ω)
BZ
³ 2π ´3 (~ω − E /2) ½· λ2 (ω)
¸ ¾
g
= 1+ K(k) − 2E(k) , (7.97)
2a (pdπ)2 π 2 2
where k 2 = 1 − λ2 (ω)/4. The functions K(k) and E(k) are the complete elliptic
integrals of the first and second kind, respectively,
Z π/2
dα
K(k) = p , (7.98)
0 1 − k 2 sin2 α
Z π/2 p
E(k) = dα 1 − k 2 sin2 α . (7.99)
0
Combining the results of (7.94) and (7.97) in (7.90), we obtain:

1 ¯ 2
Eg (ddπ)
[ε2 (ω)]π0 (xy)→π∗ (xy) = χπ
2 (~ω)2 + 4(pdπ)2 Ω2 (ω)
· ¸
E(k) 4π(pdπ)2
× 1− ρπ∗ (E⊥ + ~ω) . (7.100)
K(k) Ω2 (ω)
As ~ω → Eg , Ω2 (ω) → 0 and
· ¸
E(k) 1
1− →1
K(k) Ω2 (ω)
so that [ε2 (ω)]π0 (xy)→π∗ (xy) correctly tends to the result of (7.83). Approaching the
p
center of the absorption band where ~ω → Eg /2 + (Eg /2)2 + 4(pdπ)2 ≡ ~ωc , we
have Ω2 (ω) = 1, k 2 = 1, E(k)/K(k) = 0 and
1 ¯ 2 Eg
(ddπ)
[ε2 (ω)]π0 (xy)→π∗ (xy) = χπ ρπ∗ (E⊥ + ~ωc ) . (7.101)
2 (~ωc )(2~ωc − Eg )
The function ρπ∗ (E⊥ + ~ω) produces a logarithmic infinity in [ε2 (ω)]π0 (xy)→π∗ (xy)
at ~ω(~ω − Eg ) = 4(pdπ)2 .
p
At the top of the absorption band, ~ω → Eg /2 + (Eg /2)2 + 8(pdπ)2 = Eg +
Wπ , for which Ω2 (ω) = 2, k 2 = 0, E(k)/K(k) = 1, so that [ε2 (ω)]π0 (xy)→π∗ (xy) = 0.
Therefore the jump discontinuity in the DOS function at the top of the π ∗ band does
not manifest itself in [ε2 (ω)]π0 (xy)→π∗ (xy) . The reason for this is that the momentum
matrix element, which varies as Cx2 , vanishes for transitions to the top of the π ∗
band.
Next, we consider the contribution of the mixed π 0 (xz) → π ∗ (xy) transition
to ε2 (ω). Using (7.85)–(7.87), and the definitions (7.89) and (7.93), we have
2 ~2 ¯ 2 ³ e ´2 (~ω)2
[ε2 (ω)]π0 (yz)→π∗ (xy) = (ddδ)
π a2 mω (~ω)2 + 4(pdπ)2 Ω2 (ω)
Z Z π/2a hq i
π π/2a
× dkx dky Sy4 + Sy2 − Sy2
a −π/2a −π/2a
h i
× δ E~kπ∗ (xy) − (E⊥ + ~ω) (7.102)
where the Brillouin zone integral in the second and third lines can be expressed as
¡ ¢ µ ¶2 Z π Z π
π ~ω − Eg /2 2 £ ¤
2
dr dt δ λ(ω) + cos r + cos t
a (pdπ) 2a 0 0
n1 p 1 o
× (3 − cos r)(1 − cos r) − Ω2 (ω)
2 2
with r = 2ky a and t = 2kx a. This expression can be put into form
½¡ ¢ Z 1 p £ ¤ ¾
π3 ~ω − Eg /2 (3 − µ) Θ 1 − (λ + µ)2 2
dµ q £ ¤ − Ω (ω) ρ π ∗ (E⊥ + ~ω) .
2a3 π 2 (pdπ)2 −1 (1 + µ) 1 − (λ + µ)2
(7.103)
The integral in the first term of (7.103) can be solved in terms of incomplete elliptic
integrals of the first and third kind as
Z 1 p £ ¤
(3 − µ) Θ 1 − (λ + µ)2 1 h 2
i
Re dµ q £ ¤ = √2 (λ + 4)F (ϕ, k0 ) − λ Π (ϕ, α , k0 )
−1 (1 + µ) 1 − (λ + µ)2
7.5 Frequency dependence of ε2 (ω) from σ 0 → π ∗ transitions 167
p
with α2 = 1 − λ/2, k02 = α2 (1 + λ/4), and sin ϕ = (1 + λ/2)/(1 + λ/4) within
the interval −2 < λ < 0, and sin ϕ = 1 for 0 < λ < 2. Hence, combining this with
(7.103) we get
¯ 2 2π(pdπ)2 Eg
[ε2 (ω)]π0 (yz)→π∗ (xy) = χπ (ddδ)
(~ω)2 + 4(pdπ)2 Ω2 (ω)
½¡ ¢
~ω − Eg /2 h i
× √ (λ + 4)F (ϕ, k0 ) − λ Π (ϕ, α2 , k0 )
2π 2 (pdπ)2
¾
− Ω2 (ω) ρπ∗ (E⊥ + ~ω) . (7.104)
As ~ω → Eg , Ω2 (ω) → 0, λ → −2, ϕ → 0, F (0, k0 ) → 0, and Π (0, α2 , k0 ) → 0.

Therefore, the transition from π 0 (yz) to π ∗ (xy) does not contribute any jump at
~ω = Eg .
As ~ω → Eg + Wπ , i.e., transition to the top of the π ∗ band, Ω2 (ω) → 2, λ → 2,
ϕ = π/2, F (π/2, 0) = Π (π/2, 0, 0) = π/2, and a jump of
¡√ ¢ ³ Eg ´
¯ 2
2 − 1 χπ (ddδ) (7.105)
Eg + Wπ
occurs at ~ω = Eg + Wπ (see Table 7.3).

from σ 0 → π ∗ transitions
The σ 0 band lies at E = Ek according to our simple energy band model. Usually
Ek is about the same as E⊥ or 1 eV or so lower in energy. Thus the σ 0 → π ∗
interband transitions are important in contributing to ε2 (ω) near the band-gap
energy, ~ω = Et − E⊥ , or within an eV of it. In this section we derive expressions
for the contributions of the σ 0 → π ∗ transitions to ε2 (ω).
We begin by calculating the contribution at the threshold energy ~ω = Eg0 ≡
Et − Ek . From Table 7.2 we have
[ε2 (ω)]σ0 →π∗ = 2 [ε2 (ω)]σ0 →π∗ (xy)
2 ³ e ´2 ³ 2π ´3 1 X ¯¯ [~k,π∗ (xy)] [~k,σ0 ] ¯¯2 ³ ~ ´2 ¯ 2

= ¯axy ay ¯ (ddπ) (2Cy )2
π mω 2a N 2a
~
k
h i
× δ E~kπ∗ (xy) − (Ek + ~ω) . (7.106)
Making use of (4.42) and (4.58) for the wavefunction amplitudes and using Table
7.1, we can rewrite (7.106) as

Z ¡ ¢2
2 ³ e ´2 E~kπ∗ (xy) − E⊥
[ε2 (ω)]σ0 →π∗ = d~k ¡ ¢2
π mω E~ + 4(pdπ)2 (Sx2 + Sy2 )
kπ ∗ (xy) − E⊥
BZ
µ ¶µ 2 ¶
Sx2 Sz2 ~ ¯ 2 2
× ( ddπ) C y
Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 a2
h i
× δ E~kπ∗ (xy) − (Ek + ~ω) . (7.107)
Because of the δ function we may make the following replacements:
E~kπ∗ (xy) → Ek + ~ω = E⊥ + ~ω̄ , (7.108)

~ω̄(~ω̄ − Eg )
(Sx2 + Sy2 ) → Ω2 (ω̄) = , (7.109)
4(pdπ)2
~ω̄ ≡ ~ω + Ek − E⊥ , (7.110)
and by symmetry,
1 2 1 1
Cy2 → (C + Cx2 ) = 1 − (Sy2 + Sx2 ) = 1 − Ω2 (ω̄) . (7.111)
2 y 2 2
Only the factor F ≡ Sx2 Sz2 /[Sx2 Sy2 + Sz2 (Sx2 + Sy2 )] = 21 Sz2 Ω2 /[Sx2 Sy2 + Sz2 Ω2 ] has
R p
a dependence on kz and its integral IF = dkz a F = π/2{1 − A/(1 + A)} with
A = Sx2 Sy2 /Ω2 .
The analytical evaluation of (7.107) can be accomplished but is rather tedious.
Therefore, let us first consider an approximation for IF that simplifies the calcula-
tion.
For the integration in (7.107) over kx –ky plane, the values of kx and ky are
constrained to lie on an curve on kx –ky plane defined by the δ function. This
energy surface is determined by Ω2 and therefore we look for an approximation
for IF in terms of√Ω2 . We √note that IF → π/2 for near Γ and X, and near M
it tends to (π/2)( 3 − 1)/ √3. An approximation that matches these results is
IF ∼= (π/2)[1 + (Ω2 − Ω4 )/(2 3)]. With this approximation the integral of (7.107)
is easily evaluated as
¯ 2·
π 2 ~4 (e2 /a)(ddπ) (~ω̄)2
¸
[ε2 (ω)]σ0 →π∗ ∼
=
m2 a4 (~ω)2 (~ω̄)2 + 4(pdπ)2 Ω2 (ω̄)
h · ¸
1 2 i Ω2 − Ω4
× 1 − Ω (ω̄) 1 + √ ρπ∗(~ω̄ +E⊥ ). (7.112)
2 2 3
Just above threshold, as ~ω̄ → Eg , ~ω → Eg0 , and Ω2 (ω̄) = 0. This leads to a jump
7.5 Frequency dependence of ε2 (ω) from σ 0 → π ∗ transitions 169
discontinuity contribution to ε2 of
¯ 2 ³
π~4 (e2 /a)(ddπ) Eg ´2
[ε2 (ω)]σ0 →π∗ →
4m2 a4 Eg (pdπ)2 Eg + E⊥ − Ek
1 ³ ´2
= ¯ 2 Eg .
χπ (ddπ) (7.113)
2 Eg0
p
At ~ω = Em + (Eg /2)2 + 8(pdπ)2 − Ek = Eg0 + Wπ , corresponding to the
transition from σ 0 to the top of the π ∗ band, the contribution vanishes (see Table
7.3) because the factor Cy2 in (7.107) vanishes at that energy.
Returning to (7.107), the analytical solution can be found as follows:
· ¸
2 ~2 ¯ 2 ³ e ´2 (~ω̄)2
[ε2 (ω)]σ0 →π∗ = (ddπ)
π a2 mω (~ω̄)2 + 4(pdπ)2 Ω2 (ω̄)
Z 2 2 2 h i
Cy S x S z
× d~k 2 2 δ E ~
kπ ∗ (xy) − (E k + ~ω) . (7.114)
BZ Sx Sy + Sz2 Ω2
Integration over the variable kz gives

Z π/2a · ¸
Cy2 Sx2 Sz2 π Cy2 Sx2 Sx2 Sy2
dkz 2 2 = 1 − q . (7.115)
−π/2a Sx Sy + Sz2 Ω2 a Ω2 Ω2 Sx2 Sy2 + Sx4 Sy4
Therefore the contribution of the first term to ε2 (ω) can be obtained using the same
procedure in Subsection 7.4(b) as
¯ 2·
π 2 ~4 (e2 /a)(ddπ) (~ω̄)2
¸
[ε2 (ω)]σ0 →π∗ (xy) =
m2 a4 (~ω)2 (~ω̄)2 + 4(pdπ)2 Ω2 (ω̄)
· ¸
E(k) 1
× 1− 2
ρπ∗ (E⊥ + ~ω̄) (7.116)
K(k) Ω (ω̄)
with k 2 = 1 − λ2 /4 and
(~ω̄)(~ω̄ − Eg ) £ ¤
λ(ω̄) = − 2 = 2 Ω2 (ω̄) − 1 .
2(pdπ)2
The second term of (7.115) can be put into form

Z Z
π π/2a π/2a Cy2 Sy2 Sx4 £ ¤
dkx dky q δ E~kπ∗ (xy) − (E⊥ + ~ω̄)
aΩ2 −π/2a −π/2a Ω2 Sx2 Sy2 + Sx4 Sy4
2π 1 (~ω̄ − Eg /2)
= I(λ)
(2a)3 Ω2 (ω̄) (pdπ)2
where
Z 1 £ ¤ t2 + (λ − 2)t + (1 − λ)
I(λ) = dt Θ 1 + (λ + t)2 p (7.117)
−1 (1 + t)(1 − λ − t)(r − t)(λ + r + t)
with
λ 1p
r=− + (λ + 2)(λ + 10) .
2 2
The integral in (7.117) can be solved in terms of the incomplete elliptic integrals of
the first and second kind to give
£ ¤
I(λ) = 2 F 0 (ϕ, k0 ) − E 0 (ϕ, k0 ) + (λ + 2) cos ϕ
£ ¤
+ (1 − r) E 0 (ϕ, k0 ) − k0 F 0 (ϕ, k0 ) (7.118)
with k0 = (λ + r − 1)/(1 + r), sin ϕ = 1 within the interval 0 < λ < 2, and
(λ + 2)(1 + r)
sin ϕ = for −2 < λ < 0 .
2(λ + r − 1) − λ(r − 1)
Therefore, the total contribution of the σ 0 → π ∗ (xy) transition to ε2 (ω) can be
written as
¯ 2 ·
(ddπ) (~ω̄)2
¸
2π(pdπ)2 Eg
[ε2 (ω)]σ0 →π∗ (xy) = χπ
(~ω)2 (~ω̄)2 + 4(pdπ)2 Ω2 (ω̄) Ω2 (ω̄)
½· ¸ ¾
E(k) (~ω̄ − Eg /2)
× 1− ρπ∗ (E⊥ + ~ω̄) + I(λ) . (7.119)
K(k) 2π 2 (pdπ)2
As ~ω̄ → Eg , Ω2 (ω̄) → 0, λ(ω̄) → −2, k → 0, k 2 /Ω2 (ω̄) → 2, [1 − E/K]/Ω2 (ω̄) →

1, r → 1, ϕ → 0, k0 → 1, and I(−2) = 0. Hence, as before (in (7.113)) we get a
jump of
µ ¶2
1 ¯ 2 Eg
[ε2 (ω)]σ0 →π∗ (xy) → χπ (ddπ) per spin state. (7.120)
2 Eg0
As ~ω̄ → Eg + Wπ , Ω2 (ω̄) → 2, λ(ω̄) → 2, k → 0, k 2 /Ω2 (ω̄) → 2, [1 − E/K]/Ω2 (ω̄)

→ 1, r → 1, ϕ → π/2, k0 → 1, and I(2) = 0. Hence, we get no jump from the tran-
sitions to the top of the π ∗ band. The results for the two limiting cases are the
same as those obtained previously using the approximation for IF (see (7.113)).

from π → π ∗ transitions
The top of the π valence band coincides with the π 0 band at E = E⊥ . Conse-
quently, transitions from the top of the π band to the bottom of the π ∗ band will
7.6 Frequency dependence of ε2 (ω) from π → π ∗ transitions 171
contribute to ε2 (ω) for ~ω near to the band-gap energy. In this section we consider
the contributions of the π → π ∗ interband transitions to ε2 (ω).
According to Table 7.2, there are three types of such interband transitions:
π(xy) → π ∗ (xy), π(xy) → π ∗ (yz), and π(yz) → π ∗ (xy). For the unmixed transi-
tions, π(xy) → π ∗ (xy), we find from (7.22) and Tables 7.1 and 7.2 that
Z
2 ³ e ´2 ~2 ¯ 2 h
~k C 2 δ E~ ∗
i
[ε2 (ω)]π(xy)→π∗ (xy) = (ddπ) d x kπ (xy) − E~
kπ(xy) − ~ω
π mω a2 BZ
¯³ ´ ³ ´ ∗ ¯2
¯ [~k,π∗ (xy)] ∗ [~k,π(xy)] ~ ~ ∗ ¯
× ¯ axy ay − a[xy
k,π(xy)]
a[yk,π (xy)] ¯ . (7.121)
The minus sign within the absolute signs comes from the fact that P ∗ = −P , since
[~
k,π ∗ (xy)]
the Löwdin orbitals, ξ, are real. The amplitude, axy is given by (7.75) and
[~
k,π(xy)]
axy is obtained by substituting E~kπ(xy) for E~kπ∗ (xy) in that equation. The
remaining amplitudes are determined from
~ 2i(pdπ)Sx
a[yk,ν] = q (7.122)
(E⊥ − E~kν )2 + 4(pdπ)2 (Sx2 + Sy2 )
for ν = π(xy) or π ∗ (xy). Using (7.75) and (7.122) together with the expressions for
E~kπ∗ (xy) and E~kπ(xy) gives
· ¸
2 ³ e ´2 ~2 ¯ 2
[ε2 (ω)]π(xy)→π∗ (xy) = (ddπ) 4(pdπ)2 Eg2
π mω a2
Z
S2 C2
× d~k £ x x ¤
BZ 4(pdπ)2 (Sx2 + Sy2 ) Eg2 + 16(pdπ)2 (Sx2 + Sy2 )
£ ¤
× δ E~kπ∗ (xy) − E~kπ(xy) − ~ω . (7.123)
The δ function is satisfied when

r³
1 ´2 ~ω
Eg + 4(pdπ)2 (Sx2 + Sy2 ) = (7.124)
2 2
or
(~ω)2 − Eg2
= (Sx2 + Sy2 ) ≡ Ω2 (ω) . (7.125)
16(pdπ)2
These relations allow the integrand to be simplified considerably and the result is
2 ³ e ´2 ~2 ¯ 2 Eg2
[ε2 (ω)]π(xy)→π∗ (xy) = (ddπ)
π mω a2 (~ω)[(~ω)2 − Eg2 ]
Z
£ ¤
× d~k (2 Cx Sx )2 δ λ(ω) + C2x + C2y , (7.126)
BZ
where
(~ω)2 − Eg2 £ ¤
λ(ω) ≡ 2
− 2 = 2 Ω2 (ω) − 1 . (7.127)
8(pdπ)
In obtaining (7.126) we have employed the relation
~ω £ ¤ £ ¤
δ λ(ω) + C2x + C2y = δ E~kπ∗ (xy) − E~kπ(xy) − ~ω , (7.128)
4(pdπ)2
which is valid for ~ω > 0. Another useful relation is
µ ¶ ³ ´ Z
~ω 1 2a 3 ~ω £ ¤
ρπ ∗ + (Et + E⊥ ) = d~k δ λ(ω) + C2x + C2y
2 2 2π 2(pdπ)2 BZ
1 ~ω
= K(k) Θ(k) (7.129)
π 2(pdπ)2
2
with k 2 ≡ 1 − λ2 (ω)/4. Use of (7.129) gives

( µ ¶
π 2 ~4 (e2 /a) ¯ 2 Eg2 ~ω 1
[ε2 (ω)]π(xy)→π∗ (xy) = (ddπ) ρπ ∗ + (Et + E⊥ )
4m2 a4 Ω2 (ω) (~ω)4 2 2
Z )
1
~ω µ2 Θ[1 − (λ + µ)2 ] dµ
− 2 q ¤ . (7.130)
2π (pdπ)2 −1 (1 − µ2 )[1 − (λ + µ)2
The second term of (7.130) can be immediately evaluated by using the result of
(7.97). The final result for ε2 (ω) is
1 4π(pdπ)2 Eg3
[ε2 (ω)]π(xy)→π∗ (xy) = ¯ 2
χπ (ddπ)
4 (~ω)4 Ω2 (ω)
· ¸ µ ¶
E(k) λ2 (ω) ~ω 1
× − ρπ ∗ + (Et + E⊥ ) . (7.131)
K(k) 4 2 2
In the limit as ~ω → Eg , the quantity
· ¸
E(k) λ2 (ω)
− → Ω2 (ω)
K(k) 4
and
µ ¶
~ω 1 Eg
ρπ ∗ + (Et + E⊥ ) →
2 2 4π(pdπ)2
so that,
¯ 1 ¯ 2 Θ(~ω − Eg ) .
[ε2 (ω)]π(xy)→π∗ (xy) ¯~ω→E = χπ (ddπ) (7.132)
g 4
It is exactly one-half of the π 0 (xy) → π ∗ (xy) band-edge contribution (per spin
q
state) given by (7.83). As ~ω → 2 ( 12 Eg )2 + 8(pdπ)2 = 2W1 , namely the top of
the absorption band, one has
λ(ω) → 2 − α ,
k2 → α ,
· ¸
E(k) λ2 (ω) α
− → ,
K(k) 4 2
where
W1 (2W1 − ~ω)
α= .
(pdπ)2
One finds that apart from a numerical factor

³ E ´2 E (2W − ~ω)
¯ 2 g g 1
[ε2 (ω)]π(xy)→π∗ (xy) ∝ χπ (ddπ) . (7.133)
2W1 2(pdπ)2
Thus, the contribution to ε2 (ω) vanishes linearly in this limit. This behavior of
ε2 (ω) for transitions to the top of the band is not what is expected on the basis of
the constant-matrix-elements approximation.
¯ 2 , which is much
All of the mixed transitions have coefficients in terms of (ddδ)
¯ 2 2 2
smaller relative to (ddπ) ¿ (pdπ) < (pdσ) . Therefore they do not lead to con-
siderable contributions, especially at the band edge.
7.7 σ → π ∗ interband transitions
Calculation of the contributions of σ → π ∗ transitions to ε2 (ω) are difficult to treat

analytically because they involve complicated transition matrix elements. From
Table 7.2 one may see that the σ(±) → π ∗ (xy) transition involves three terms
in the matrix element. In addition, the σ band amplitudes are themselves quite
complex as may be seen from Subsection 4.5(b) in Chapter 4.
Fortunately, to obtain the threshold behaviors at Γ and R we¯ need only the ¯
¯ π∗ (xy) σ(±) ¯2
dominant term of the transition matrix element, the term involving ¯axy ay ¯ .
The other two terms in Table 7.2 are negligible at the thresholds. The dominant
term of the transition matrix element for σ(±) → π ∗ (xy) is
¯ ∗ ¯2 ~2 ¯ 2 2 ¯¯ π∗ (xy) σ(±) ¯¯2
¯ π (xy) ¯
¯axy P(xy)y,y 2 Cy aσ(±)
y ¯ = 2 (ddπ) Cy ¯axy ay ¯ . (7.134)
a
π ∗ (xy) σ(±)
Both axy and ay are finite as ~k → 0. For the two transitions we have approx-
Table 7.3. Band-edge behavior of interband contributions to ε2 (ω) including a

factor of 2 for the two spin states. [ε2 (ω)]ν→ν 0 = χπ Tνν 0 f (~ω).
ν ν0 Edge (E0 ) Tνν 0 f (~ω)
π 0 (xy) π ∗ (xy) Eg ¯
(ddπ)2
Θ(~ω − E0 )
¡ Eg ¢ ¡ W1 (E0 −~ω) ¢
E g + Wπ ¯
(ddπ)2
E0 (pdπ)2
¡ (~ω−E0 )E0 ¢1/2

π 0 (yz) π ∗ (xy) Eg ¯ 2
(ddδ)
2(pdπ)2
√ ¡E ¢
E g + Wπ ¯ 2
2( 2 − 1) Eg (ddδ) Θ(E0 − ~ω)
0
¡ Eg ¢2
σ0 π ∗ (xy) Eg0 ¯
(ddπ)2
Θ(~ω − E0 )
E0
Eg (Eg +Wπ ) ¯
¡ W1 (E0 −~ω) ¢
Eg0 + Wπ (ddπ)2
E2 2(pdπ)2
0
1 ¯
π(xy) π ∗ (xy) Eg 2 (ddπ)
2
Θ(~ω − E0 )
¡ Eg ¢2 ¡ Eg (E0 −~ω) ¢
Eg + 2Wπ ¯
(ddπ) 2
E0 2(pdπ)2
4 ¯
¡ (~ω−E0 )E0 ¢5/2
π(xy) π ∗ (yz) Eg ( pdπ
Eg ) (ddδ)
2
2(pdπ)2
(pdπ)2
¡ Eg ¢ ¡ E0 (E0 −~ω) ¢3/2
Eg + 2Wπ ¯ 2
(ddδ)
W2 E0 2(pdπ)2
1
¡ (~ω−E0 )E0 ¢1/2

π(yz) π ∗ (xy) Eg ¯ 2
(ddδ)
2(pdπ)2
(Eg +Wπ )(pdπ)2
¡ Eg ¢ ¡ E0 (E0 −~ω) ¢3/2
Eg + 2Wπ ¯ 2
(ddδ)
Wπ W 2 E0 2(pdπ)2
1
¡ Eg ¢2
σ(−) π ∗ (xy) Eg0 4ξ ¯
(ddπ)2
Θ(~ω − E0 )
(4ξ+3) E0
¡W ¢3/2
Eg0 + Wπ + Wσ ¯
C1 η Jf (η) (ddπ) 2 2 (E0 −~ω)
(pdσ)2
¡ Eg ¢2 ¡ (~ω−E0 )Egσ ¢1/2

σ(+) π ∗ (xy) Eg0 2 ξ If (ξ) ¯
(ddπ) 2
E0 2
2 C1 η
¡ W 2(pdπ)
(E −~ω)
¢
Eg0 + Wπ + Wσ ¯
(ddπ)2 2 0
(4η+3)2 (pdσ)2
p
(a) χπ = π~4 (e2 /a)/[2m2 a4 Eg (pdπ)2 ], (b) Eg = Et − E⊥ , (c) W1 = ( 1 E )2 + 8(pdπ)2 ,
p 2 g
(d) Wπ = W1 − 1
2 Eg , (e) Eg0 = Et − Ek , (f) Egσ = Ee − Ek , (g) W2 = ( 21 Egσ )2 + 6(pdσ)2 ,
(h) Wσ = W2 − 12 Egσ , (i) The ratios ξ, η, C1 and the functions If (ξ), Jf (η) are defined in the text.
(j) Tνν 0 are exact only for the cases with jumps. For the cases with power law behavior, the numerical
factors that vary according to the shape of the small volume chosen for integration, are not included.
imately
· ¸
1 ³ e ´2 ~2 ¯ 2
[ε2 (ω)]σ(±)→π∗ (xy) ' ( ddπ)
π mω a2
Z ¯ ∗ ¯2 h i
¯ ¯ 2
× d~k ¯aπxy (xy) aσ(±)
y ¯ Cy δ E~kπ∗ (xy) −E~kσ(±) −~ω . (7.135)
BZ
The σ(±) energy bands are given by

rh i2
1 1 £ ¤
E~kσ(±) = (Ee + Ek ) − (Ee −Ek ) + 2(pdσ)2 Sx2 + Sy2 + Sz2 ± S 2 ,
2 2
(7.136)
q
S2 = (Sx4 + Sy4 + Sz4 ) − (Sx2 Sy2 + Sy2 Sz2 + Sz2 Sx2 ). (7.137)
Equation (7.136) shows that the σ(−) branch is flat along the kx , ky or kz -axis.
Since E~kπ∗ (xy) is also flat along the kz -axis there will be a jump in ε2 (ω) at ~ω =
Eg0 ≡ Et − Ek arising from transitions where the ~k-vectors are located in a small
cylinder about the kz -axis. To calculate the jump contribution, we write, for ~k near
the kz -axis:
4(pdπ)2
E~kπ∗ (xy) → Et + [(kx a)2 + (ky a)2 ] ,
Eg
1
S 2 → Sz2 − [(kx a)2 + (ky a)2 ] ,
2
3(pdσ)2
E~kσ(−) → Ek − [(kx a)2 + (ky a)2 ] ,
Egσ
Egσ ≡ Ee − Ek , Eg0 ≡ Et − Ek ,
¯ ∗ ¯ ¯ ¯
¯ π (xy) ¯2 ¯ σ(±) ¯2
¯axy ¯ →1, ¯ay ¯ → 1 , Cy2 → 1 .
It then follows that

Z h i
d~k Cy2 δ E~kπ∗ (xy) − E~kσ(−) − ~ω
BZ
Z 2π Z ∞ ½ · ¸ ¾
π 4(pdπ)2 3(pdσ)2 2
→ dφ ρ dρ δ (~ω − Eg0 ) − + ρ
a3 0 0 Eg Egσ
· ¸−1
π 2 4(pdπ)2 3(pdσ)2
= 3
+ Θ(~ω − Eg0 ) , (7.138)
a Eg Egσ
and hence
¯ ³ 2ξ ´µ E ¶2
¯ 2 g
[ε2 (ω)]σ(−)→π∗ (xy) ¯~ω→E 0 = χπ (ddπ) Θ(~ω − Eg0 )
g 4ξ + 3 Eg0
(pdπ)2 Egσ
with ξ≡ . (7.139)
(pdσ)2 Eg
The σ(+) branch behaves quite differently; the energy E~kσ(+) depends on all
three components of ~k. As |~k| → 0,
2(pdσ)2
E~kσ(+) → Ek − [(kx a)2 + (ky a)2 + (kz a)2 + S 2 ] . (7.140)
Egσ
The contribution to ε2 (ω) is

Z Z 2π Z π
1 ³ e ´2 ~2 ¯ 2 ∞ 2
[ε2 (ω)]σ(+)→π∗ (xy) → ( ddπ) r dr dϕ sin θ dθ
π mω a5 0 0 0
½ µ ¶ ¾
2(pdσ)2
× δ (~ω − Eg0 ) − r2 f (θ, ϕ) (7.141)
Egσ
where we have used spherical coordinates defined by
kx a → x = r sin θ cos ϕ
ky a → y = r sin θ sin ϕ
kz a → z = r cos θ (7.142)
and
q
f (θ, ϕ) = 2ξ sin2 θ + 1+ 1−3 sin2 θ cos2 θ−3 sin4 θ cos2 ϕ sin2 ϕ .
One then obtains

µ ¶2 · ¸1/2
¯ Eg (~ω − Eg0 )Egσ
¯ 2
[ε2 (ω)]σ(+)→π∗ (xy) ¯~ω→E 0 → χπ (ddπ) ξ If (ξ)
g Eg0 2(pdπ)2
(7.143)
where,
Z 2π Z π
1 sin θ dθ
If (ξ) = dϕ . (7.144)
2π 2 0 0 f (θ, ϕ)3/2
The integral, If (ξ), can not be evaluated analytically but is approximately given
by
4.01/π
If (ξ) ' ¡ ¢3/2 , (7.145)
ξ + 2.4869
or a quadratic fit will give
If (ξ) ' 0.05526 ξ 2 − 0.1834 ξ + 0.3247 . (7.146)
The analytical behavior of the contributions to ε2 (ω) for energies near the pi
and sigma band gaps are summarized in Table 7.3. Inspection of the table shows
that the contributions to ε2 (ω) for ~ω near a particular threshold (edge), E0 = ~ω0 ,
obey a power law of the form |E0 − ~ω|p , where p = 0 (jump), an integer, or n/2
(n = odd integer). We can understand how these power laws arise. A particular
contribution to ε2 (ω) has the form
Z
¯ ¯2 ¯£ ~ 0 ¤∗ ~kν ¯2
ε2 (ω) ∝ ¯P(α0 β 0 )j 0 ,γ ¯ d~k ¯ akν ¯ 2
(α0 β 0 ) aj 0 γ cos (kj 0 a) δ[E~
k(α0 β 0 ) − E~
kγ − ~ω]
(7.147)
where the ‘P -factor’ and the ‘a-factors’ are given in Table 7.2. The ‘P -factor’ is
a constant and therefore may be ignored for our purpose here. The power laws of
Table 7.3 apply to the thresholds at Γ or R. Each of the factors of the integrand of
(7.147) are functions of sin2 (kj a) and cos2 (kj a), (j = x, y, and z). For a threshold
at Γ, cos2 (kj a) → 1 and sin2 (kj a) → (kj a)2 . Therefore each of the factors will be a
non-zero constant or expressible as a power series in the small parameters, (kj a)2 .
If the integrand involves all three components of the wavevector, ~k, the integrand
is conveniently expressed in spherical coordinates defined by (7.142). The delta
function will then assume the form δ[~ω − E0 − A(θ, ϕ) r2 ] and the remainder of
the integrand can be written as B(θ, ϕ) rs so that,
Z Z Z
© ª
ε2 (ω) ∝ sin θ dθ dϕ r2 dr B(θ, ϕ) rs δ[~ω − E0 − A(θ, ϕ) r2 ]
Z Z ¸Z · · ¸
B(θ, ϕ) (~ω − E0 )
= sin θ dθ dϕ dr rs+2 δ − r2
A(θ, ϕ) A(θ, ϕ)
½Z π Z 2π · ¸¾
B(θ, ϕ)
= sin θ dθ dϕ (~ω − E0 )(s+1)/2
0 0 A(θ, ϕ)(s+3)/2
= constant × (~ω − E0 )(s+1)/2 , (as ~ω → E0 ) . (7.148)
For a threshold at R in the Brillouin zone a similar procedure can be used by writing
a power series for the integrand factors in terms of the small parameters, x, y, z,
where kx a = π/2 − x, ky a = π/2 − y, kz a = π/2 − z. In this case the delta function
assumes the form δ[E0 − ~ω − A(θ, ϕ)r2 ] and the result is
ε2 (ω) ∝ constant × (E0 − ~ω)(s+1)/2 , (as ~ω → E0 ) . (7.149)
When [E~k(α0 β 0 ) − E~kγ ] depends on only two components of the wavevector, ~k, cylin-
drical coordinates may be used to arrive at similar results. However, in this case
the power-law exponent will be an integer. It may also be zero, meaning that there
is a jump in the contribution to ε2 (ω) as ~ω → E0 . The transition from σ(−) to
π ∗ (xy) for E0 = Eg0 (at Γ) is an example for such a singular case (see (7.139)).
The above general rule can be applied easily to the behavior of the contributions
by the σ(±) → π ∗ transitions to ε2 (ω) at R, namely, as ~ω → E0 = Wπ + Wσ + Eg0 .
At this limit one has

1 2(pdπ)2 2
E~kπ∗ (xy) → (Et + E⊥ ) + W1 − (x + y 2 ) ,
2 W1
1 (pdσ)2 2
E~kσ(±) → (Ee + Ek ) − W2 + [x + y 2 + z 2 ∓ S 2 ] ,
2 W2
p
S2 → (x4 + y 4 + z 4 ) − (x2 y 2 + y 2 z 2 + z 2 x2 ) ,
Egσ ≡ Ee − Ek , Cy2 → y 2 = r2 sin2 θ sin2 ϕ , (s = 2) ,

¯ ∗ ¯ E g + Wπ ¯ ¯ Egσ + Wσ
¯ π (xy) ¯2 ¯ σ(±) ¯2
¯axy ¯ → , ¯ay ¯ → .
2W1 2W2
The contribution to ε2 (ω) is
Z Z 2π Z π
1 ³ e ´2 ~2 ¯ 2 ∞ 2
[ε2 (ω)]σ(±)→π∗ (xy) → (ddπ) r dr dϕ sin θ dθ
π mω a5 0 0 0
³E + W ´ ³E + W ´
g π gσ σ
× r2 sin2 θ sin2 ϕ
2W1 2W2
½ µ 2
¶ ¾
2 (pdσ)
× δ (E0 − ~ω) − r f (θ, ϕ) (7.150)
W2
where
q
f (±) (θ, ϕ) = 2η sin2 θ + 1 ∓ 1−3 sin2 θ cos2 θ−3 sin4 θ cos2 ϕ sin2 ϕ ,
(pdπ)2 W2
with η≡ . (7.151)
(pdσ)2 W1
First, let us define
µ ¶
(pdσ)2
A(θ, ϕ) = f (−) (θ, ϕ) (7.152)
W2
and
µ ¶µ ¶
E g + Wπ Egσ + Wσ
B(θ, ϕ) = sin2 θ sin2 ϕ . (7.153)
2W1 2W2
One then obtains
· ¸3/2
¯ C1 W2 (E0 − ~ω)
¯
[ε2 (ω)]σ(−)→π∗ (xy) ~ω→E → ¯ 2
χπ (ddπ) η Jf (η) (7.154)
0 2 (pdσ)2
where
Eg (Eg + Wπ )(Egσ + Wσ )
C1 = , (7.155)
E02 W2
Z 2π Z π
1 sin2 θ sin2 ϕ
Jf (η) = dϕ sin θ dθ . (7.156)
2π 2 0 0 f (−) (θ, ϕ)5/2
The integral, Jf (η), can be evaluated numerically and an approximate expression

is
0.669/π
Jf (η) ' ¡ ¢5/2 , (7.157)
η + 2.042
or a quadratic fit will give
Jf (η) ' 0.0163 η 2 − 0.0389 η + 0.0356 . (7.158)
For the case of σ(+) → π ∗ (xy) transition near R, the function f (+) (θ, ϕ) = 0
when θ = 0 and this will cause the integral over the spherical angles to diverge. To
treat this case we can use a small cylinder around the kz -axis at R instead of a
sphere around R. Near R this gives
1 2
S 2 → 1 − Sz2 − ρ ,
2
1 3(pdσ)2 2
E~kσ(+) → (Ee + Ek ) − W2 + ρ .
2 2W2
The BZ integral will then be

Z h i
d~k Cy2 δ E~kπ∗ (xy) − E~kσ(+) − ~ω
BZ
Z 2π Z ∞
π
→ dφ ρ dρ ρ2 sin2 φ
a3 0 0
½ · ¸ ¾
2(pdπ)2 3(pdσ)2 2
× δ (Wπ + Wσ + Eg0 − ~ω) − + ρ
W1 2W2
· ¸−2
π 2 2(pdπ)2 3(pdσ)2
= + (Wπ + Wσ + Eg0 − ~ω) , (7.159)
2a3 W1 2W2
and hence the behavior is linear as ~ω → E0 = Wπ + Wσ + Eg0

· ¸
¯
¯ 2 ¡ C1 η ¢
[ε2 (ω)]σ(+)→π∗ (xy) ¯~ω→E = χπ (ddπ)
W2 (E0 − ~ω)
(7.160)
0 2 (pdσ)2
4η + 3
where the parameters η and C1 are defined in (7.152) and (7.156), respectively.
7.8 Summary
Light interacts with the electrons of a solid through its electromagnetic field causing
a variety of electronic processes. absorption of photons by the solid is accomplished
by direct and indirect transitions between electronic band states, excitation of op-
tical phonons, magnons, and plasmons. For photon energies between 1 and 10 eV,
absorption of light by insulating, cubic, d-band perovskites is determined princi-
pally by interband electronic transitions of electrons from occupied valence-band
states to unoccupied conduction-band states.
In this chapter a semiclassical theory of the optical properties was described in
which classical electromagnetic theory was joined with quantum theory by equat-
ing the classical energy loss to the quantum mechanical rate of transitions between
quantum states. The resulting theory provided a description of the dielectric func-
tion, ε2 (ω), in terms of the electronic transition between energy band states of the
solid. The frequency-dependent optical constants were then calculated from the
dielectric function.
A qualitative theory of ε2 (ω) based on replacing the transition matrix elements
by their average value led to a description of the optical properties in terms of joint
density of states (JDOS) functions. It was shown that the JDOS possessed the
same characteristic structures as the DOS, including jump discontinuities at the
band edges and a logarithmic singularity in the center of the band. As a result
these structures are apparent in the frequency-dependent dielectric function, ε2 (ω).
A method based on the LCAO approximation was developed to calculate the
matrix element for interband transitions. This method involved the use of overlap
integrals between fictitious localized orbitals and led to a description of the transi-
tion matrix elements in terms of three fundamental parameters: (ddπ), ¯ ¯
(ddδ), and
(s̄dσ). Detailed calculations were carried out for the contributions to ε2 (ω) from
the various interband transitions. It was shown that the DOS singularities show up
in ε2 (ω) even when the frequency dependence of the matrix elements is included. It
was also shown that matrix-element effects lead to significant changes in ε2 (ω) from
what was calculated assuming a constant matrix element. Explicit results for ε2 (ω)
and the absorption coefficient for BaTiO3 and SrTiO3 for ~ω = Eg (the band-gap
energy) were found to be in reasonable agreement with the experimental observed
values.
References
[1] M. Cardona, Phys. Rev. A 140, 651 (1964).
[2] T. Wolfram and Ş. Ellialtıoğlu, Appl. Phys. 22, 11 (1980).
1. The theory presented in this chapter is called a semiclassical theory, meaning that it
combines classical electromagnetic theory with quantum theory. Identify the key steps
that connect the classical theory with the quantum theory.
2. Prove (7.23).
3. Derive the results of (7.55)–(7.58).
4. Explain why the optical properties of insulators or doped insulators are determined
principally by interband transitions for ~ω ≥ Eg . For infrared light what new processes
might be expected to play an important role?
5. Let the initial state belong to the energy band described by Ei (~k) = α(kx a)2 , and the
final state belong to the band described by Ef (~k) = E0 + β(ky a)2 . Calculate the joint
density of states, J(ω), defined by
µ ¶3 Z π/2a Z π/2a Z π/2a ³ ´
2a
J(ω) = dkx dky dkz δ ~ω − [(Ef (~k) − Ei (~k)] .
π −π/2a −π/2a −π/2a
Evaluate the behavior of J(ω) for ~ω → E0+ and ~ω → E0 + β(π/2)2 .
x
6. Derive the result for P(xy)z,y shown in Table 7.1.
7. If a complex function f (z) = f1 (z) + if2 (z) is analytic in the upper half-plane, the real
and imaginary parts of f (z) are related by
Z ∞
1 fβ (z 0 ) dz 0
fα (z) = P (α, β = 1, 2),
π −∞ z − z0
given that f (−z) = f ∗ (z) show that
Z ∞
2 ω 0 f2 (ω 0 ) dω 0
f1 (ω) = P ,
π 0 ω 2 − ω 02
Z ∞
2ω f1 (ω 0 ) dω 0
f2 (ω) = − P .
π 0 ω 2 − ω 02
8. Show that the contribution to ε2 (ω) due to the transition π(xy) → π ∗ (yz) obeys the
power law (~ω − E0 )5/2 in the limit as ~ω → E0 = Eg , and (E0 − ~ω)3/2 in the limit
as ~ω → E0 = 2W1 (see Table 7.3).
8
Photoemission from perovskites
In Chapter 7 we discussed interband transitions in which an electron occupying

an energy band state was excited to a final state in a higher band by a photon of
energy ~ω. The relation between the electronic structure of the solid and optical
properties such as reflectivity or absorption was established through the optical
dielectric function.
If the photon energy is sufficiently large, the final state of an excited electron
may be above the vacuum level of the solid. That is, the final state may be an
unbound or continuum state in which the electron can escape from the solid. Such
processes are called electron photoemission.
The kinetic energy of a photoemitted electron, Ekin , is
Ekin = ~ω − |Ei | − Φ , (8.1)
where Ei is the initial (bound) state energy and Φ is the work function of the
solid. A considerable amount of information about the electronic states of the solid
can be obtained by analyzing the kinetic energy distribution of the photoemitted
electrons. Consequently, photoelectron spectroscopy has become a very important
method for studying the electronic structure of solids and solid surfaces.
Recently, high-resolution electron-energy analyzers have become available
which allow finer detail of the emitted electrons to be measured in using ultravio-
let photoelectron spectroscopy (UPS). This advance coupled with the development
of tunable, polarized synchrotron radiation sources have made it possible to track
both the kinetic energy and the angle of photoemitted electrons. Angle-resolved
photoemission spectroscopy or ARPES is now routinely used to determine the ini-
tial state energy and the wavevector with some precision. An energy resolution of
about 2 meV and an angular resolution of 0.2◦ (about 1% of the Brillouin zone
reciprocal lattice vector) are obtained experimentally.
The application of ARPES to the study of high-temperature superconductors
is one of the most important methods of probing the electronic structures of the
182
superconducting cuprates. Detailed features of the superconducting gap and the

Fermi surface have been mapped out for many of the high-Tc materials [1] (see
Chapter 11).
The range of electronic states that can be probed by photoemission depends
upon the energy of the incident photon. UPS experiments make use of photon
energies in the range 5–50 eV. The energy bands associated with the outer (or
valence) electrons of the solid can be investigated by UPS.
To probe the deep core levels of the solid the photon energy must be in the
x-ray region. X-ray photoelectron spectroscopy (XPS) studies employ x-ray photon
sources such as the aluminum Kα line at 1486 eV. XPS experiments can probe both
the valence electron energy bands and the deep core levels of the solid.
In this chapter we shall be mainly concerned with the interpretation of pho-
toemission data for the valence electron energy bands.
8.1 Qualitative theory of photoemission
In photoemission the excitation of an electron proceeds through the same mecha-

nism as in ordinary optical excitation. According to (7.10) the rate at which elec-
trons make transitions due to photons of energy ~ω is
dW 4π 2 ³ e ´2 X
= |A| |hf |ei~q·~r~a0 · p~ |ii|2 δ(Ef − Ei − ~ω)f (Ei )[1 − f (Ef )] ,
dt ~ mc
i,f
(8.2)
where A is the amplitude of the vector potential, i and f denote initial and final
states, ~q is the wavevector of the photon and f (E) is the Fermi distribution function.
In the usual optical experiment one is interested in how the rate of transitions
varies with the photon energy ~ω. On the other hand, in a typical photoemission
experiment, the photon energy is held constant and the energies of the photoemit-
ted electrons are analyzed. Thus in photoemission, one measures the number of
electrons with kinetic energy in a range between E and E + dE (per second). It is
seen from (8.1) that the kinetic energy of a photoemitted electron is directly related
to the initial state energy, Ei . Photoemission data are usually presented in the form
of an electron energy distribution, I(E), where E is the energy of the initial state
from which the electron was emitted.
The rate of transitions from an initial state of energy E to a final state of
energy E + ~ω is
X dWif (~ω)
I(E) = δ(E − Ei ) , (8.3)
dt
if
184 Photoemission from perovskites
where the δ function, δ(E − Ei ), projects out transitions from the initial states with
energy Ei [2]. The emission rate per unit cell, I(E), is
1 XX
I(E) = 2 |Mνν 0 (~k, ~k 0 )|2 δ(~ω − E~k0 ν 0 + E~kν )δ(E − E~kν )f (E), (8.4)
N 0
~
k~k0 νν
where we have set 1 − f (E~k0 ν 0 ) = 1, since the final states are unoccupied prior to
emission and the matrix element is
4π 2 ³ e ´2 0 0 i~q·~r
|Mνν 0 (~k, ~k 0 )|2 = |A| |hk ν |e ~a0 · p~ |~kνi|2 . (8.5)
~ mc
If we make the constant matrix element approximation (CMEA) so that
|Mνν 0 (~k, ~k 0 )|2 is independent of ~k and ~k 0 , then,
X ½ X ¾
1 X 1
I(E) ' f (E) h|Mνν 0 |2 i δ(E − E~kν ) δ[(~ω + E~kν ) − E~k0 ν 0 ] . (8.6)
0
N N
νν ~
k ~
k0
The term in curly brackets may immediately be identified as the final state DOS
function, ρν 0 (~ω + E~kν ). The dependence of this function on E~kν is very weak. The
reason for this is that ~ω ' 1.5 × 103 eV (the Al Kα line, for example, is 1486 eV)
while |E~kν | . 1.5 × 10 eV for a typical valence-band width. Thus, ~ω + E~kν varies
by only about 1% as the initial states range over the valence bands. Furthermore,
since the final states have E~kν ' ~ω they are continuum states which may be ap-
proximately described as plane waves that are slightly distorted by the periodic
potential of the solid. Since plane-wave states will have a DOS function that varies
as the square root of the energy it follows that
q √
ρν 0 (~ω + E~kν ) ∝ (~ω + E~kν ) ' ~ω . (8.7)
According to (8.7) ρν 0 is approximately a constant that is independent of E~kν .

Therefore, writing ρν 0 (~ω + E~kν ) = ρν 0 (~ω) we obtain from (8.6),
XµX ¶ X
1
I(E) ' ρν 0 (~ω)h|Mνν 0 |2 i δ(E − E~kν )f (E)
ν
N
ν0 ~
k
X
= Mν2 ρν (E)f (E) (8.8)
ν
P P
where Mν2 ≡ ν0 h|Mνν 0 |2 i ρν 0 (~ω) and ρν = 1/N δ(E − E~kν ) is the initial DOS.
Equation (8.8) shows that CMEA leads to the result that the energy distri-
bution of photoelectrons in an XPS experiment is proportional to the density of
filled states, ρν (E)f (E). This conclusion appears to be approximately valid for a
number of materials. A detailed comparison of I(E) with theoretical DOS functions
for the d bands of gold has been carried out by Shirley [3] and Freeouf et al. [4].
The agreement is remarkably good and suggests that the constant matrix element
approximation is valid for gold.
The results of the CMEA are reasonably good for monatomic materials and
for bands whose wavefunctions are composed of orbitals predominantly of one type;
d orbitals for example in the case of gold. By contrast, application of (8.8) to the
analysis of the XPS spectra of compounds such as the perovskites is not successful
[5]. The principal difficulty is that the probabilities of exciting electrons from p
and d orbitals are substantially different; at XPS energies the d-orbital probability
appears to be 3–10 times larger than the p-orbital probability [6]. In order to take
this effect into account it is necessary to formulate the CMEA in a different way.
An alternative approximation is to assume that the energy distribution is a
sum of contributions arising from transitions from each of the basis orbitals which
enter the wavefunctions. Then,
X X
I(E) = njαν (E) , (8.9)
ν jα
C XX ~
njαν (E) = 2 |hk, ν; jα| e−i~q·~r (~a0 · p~)|~k 0 ν 0 i|2
N 0 ~
k~k0 ν
×f (E) δ(E − E~kν ) δ(~ω − E~k0 ν 0 + E~kν ) (8.10)
where C is a collection of constants and h~k, ν; jα| denotes the amplitude of the
(jα)th orbital in the wavefunction ψ~kν . For the LCAO wavefunctions the (jα)th
component is
1 X i~k·R~ mj
√ e ajα (~k, ν) ξα (~r − R
~ mj ) . (8.11)
N m
If (8.11) is employed in (8.10) and the final DOS taken to be constant, then one
finds that
µX ¶ X
1
njαν (E) ' 2
ρν 0 (~ω)h|Mjαν 0 | i |ajα (~k, ν)|2 δ(E − E~kν )f (E) . (8.12)
0
N
ν ~
k
For this model h|Mjαν 0 |2 i is an empirical parameter representing the average value
of the square of the matrix element
Z
1 X i~k·R~ mj
Mjαν 0 (~k, ~k 0 ) = √ e ~ mj ) e−i~q·~r (~a0 · p~) ψ~ 0 0 (~r) . (8.13)
d3 r ξα (~r − R k ν
N m
The quantity,
X
ρν 0 (~ω) h|Mjαν 0 |2 i ≡ σjα (~ω) , (8.14)
ν0
is the effective cross-section for emission from the (jα)th type orbital into the
final continuum states. There are two cross-sections for the perovskites, σp (~ω) and
σd (~ω) for the p and d orbitals, respectively. The partial energy distributions of
photoelectrons of (8.12) can now be expressed in the form
njαν (E) ' C σjα (~ω) ρjαν (E) (8.15)
where the quantity

1 X
ρjαν (E) = |ajα (~k, ν)|2 δ(E − ε~kν ) , (8.16)
N
~
k
is the partial density of states function, abbreviated as PDOS. It specifies the

portion of the DOS of the νth band that is associated with the Löwdin orbitals of
the (jα)th type. The PDOS functions satisfy the relations
X
ρν (E) = ρjαν (E) , (8.17)
jα
X
ρ(E) = ρν (E) , (8.18)
ν
where ρν is the DOS for the ν band and ρ(E) is the total density of states. The
sum of the PDOS functions for the d-symmetry orbitals (dxy , dyz , dzx , dx2 −y2 , and
d3z2 −r2 ) is defined as ρdν while the sum of the PDOS functions for the p-symmetry
orbitals (x, y, z functions on each of the three oxygen sites of a unit cell) is desig-
nated by ρpν :
X
ρdν (E) = ρjαν (E) ,
d−symmetry
X
ρpν (E) = ρjαν (E) . (8.19)
p−symmetry
With these definitions the energy distribution of photoelectrons can be written in

the convenient form:
½ X X ¾
I(E) ' C σd (~ω) ρdν (E) + σp (~ω) ρpν (E) . (8.20)
ν ν
Equation (8.20) indicates that the electron distribution can be expressed as a

weighted sum of the PDOS functions where the weighting factors are the effec-
tive photoionization cross-sections for the p and d orbitals.
The validity of (8.20) depends on, among other things, the assumptions im-
plicit in writing (8.9); that there is no interference between emission from the dif-
ferent basis orbitals. However, when the squared matrix element is summed over
the Brillouin zone (over all initial states) the cross-terms are subject to destructive
interference while the diagonal terms always add constructively. This suggests that
the diagonal contributions will dominate the cross-terms and therefore (8.9) is a
reasonable approximation. In the latter portion of this chapter we shall show that
8.2 Partial density of states functions 187
XPS experiments on perovskites appear to be in good agreement with the form of

(8.20).
8.2 Partial density of states functions
In the preceding section it was found that the photoelectron energy distribution
could be expressed approximately in terms of the PDOS functions. The PDOS
functions can easily be obtained from the DOS functions given in Chapter 6.
The PDOS, ρpπ0 or ρpσ0 associated with the oxygen orbitals of the non-bonding
bands are the same as the corresponding DOS functions since there is no d-orbital
component:
ρpπ0 (E) = ρπ0 (E) (see (6.29)) , (8.21)

ρpσ0 (E) = ρσ0 (E) (see (6.63)) . (8.22)
For the π (or π ∗ ) bands we have from (8.16) that
1 X (E⊥ − E~αβ )2 δ(E − E~kν )

kν
ρdπ (E) = αβ 2
(8.23)
N (E⊥ − E~ ) + 4(pdπ)2 (Sα2 + Sβ2 )
~ k kν
1 X 4(pdπ)2 (Sα2 + Sβ2 ) δ(E − E~kν )

ρpπ (E) = . (8.24)
N
~
(E⊥ − E~αβ )2 + 4(pdπ)2 (Sα2 + Sβ2 )
k kν
These PDOS can be obtained easily by using (4.34) and replacing E~kν by E in the
coefficients of δ(E − E~kν ) in (8.23) and (8.24). The result is that
1
2 (E − E⊥ )
ρdπ (E) = ρπ (E) (8.25)
[E − 12 (Et + E⊥ )]
1
2 (E − Et )
ρpπ (E) = ρπ (E) , (8.26)
[E − 12 (Et + E⊥ )]
where ρπ (E) is given by (6.28).

Similarly for the sigma bands we find
1 X (Ek − E~kσ )2 Xσ2

ρdσ (E) = δ(E − E~kν )
N Cσ2 (pdσ)2
~
k
1 X [3(Ek − E~kσ )(Sx2 − Sy2 )]2

+ δ(E − E~kν )
N Cσ2 (pdσ)2
~
k
1
2 (E − Ek )
= 1 ρσ (E) (8.27)
[E − 2 (Ee + Ek )]
1 X 4Sz2 Xσ2
ρpσ (E) = δ(E − E~kν )
N Cσ2
~
k
1 X Sx2 [Xσ − 3(Sy2 − Sx2 )]2

+ δ(E − E~kν )
N Cσ2
~
k
1 X Sy2 [Xσ + 3(Sy2 − Sx2 )]2

+ δ(E − E~kν )
N Cσ2
~
k
1
2 (E − Ee )
= 1 ρσ (E) . (8.28)
[E − 2 (Ee + Ek )]
The quantities E~kσ , Xσ , and ρσ (E) are defined by (4.62)–(4.64), (6.54), and (6.62),
respectively. Cσ is the normalization coefficient for the eigenstates given by (4.65).
The PDOS functions are more conveniently written in terms of the dimension-
less DOS functions (see Sections 6.2 and 6.3):
(E − E⊥ )
ρdπ (E) = ρπ (επ ) ,
2(pdπ)2
(E − Et )
ρpπ (E) = ρπ (επ ) ,
2(pdπ)2
(E − Ek )
ρdσ (E) = ρσ (εσ ) ,
(pdσ)2
(E − Ee )
ρpσ (E) = ρσ (εσ ) , (8.29)
(pdσ)2
where
1 ¡p ¢ ¡ ¢
ρπ (επ ) = ρπ (επ (E)) = K 1 − [επ (E)/2]2 Θ 1 − [επ (E)/2]2 ,
π2
ρσ (εσ ) = ρσ (εσ (E))
£ √ ¤ ¡ ¢
= 0.3183 + 0.1136 x2 − 0.0151 (1 − x) x Θ 1 − [εσ (E)/3]2
h p i ¡ ¢
+ 0.432 − 0.1646 1 − ε2σ − 0.0151(1 − |εσ |) |εσ | Θ 1 − εσ (E)2 ,
with
x ≡ (3 − |εσ |)/2 ,
(E − Et )(E − E⊥ )
επ (E) = − 2,
2(pdπ)2
(E − Ee )(E − Ek )
εσ (E) = − 3. (8.30)
(pdσ)2
Finally, we can write the total (all-bands) PDOS functions as
ρd (E) = 3ρdπ (E) + ρdσ (E) , (8.31)

ρp (E) = 3ρpπ (E) + ρpσ (E) + 3ρπ0 (E) + ρσ0 (E) . (8.32)
(Note that the function ρdσ gives the DOS of both sigma bands.)
8.3 The XPS spectrum of SrTiO3
The XPS spectrum of SrTiO3 has been reported by Battye et al. [7] (and also by
Kowalczyk et al. [8] as well as Sarma et al. [9]) and analyzed by Wolfram and
Ellialtıoğlu [6]. The result of Battye et al. [7] is shown in Fig. 8.1 (dotted curve).
Since there are no electrons occupying the conduction bands the emission arises
entirely from the valence bands. The data represent I(E) the rate of emission of
electrons from initial state of energy E. Also shown in Fig. 8.1 (solid curve) is
the theoretical I(E) curve obtained from (8.20) with σp (~ω)/σd (~ω) = 1/3. The
theoretical I(E) possesses several features (labeled 1 through 6) which are related
to the energy band structure. The peak centered at about – 1.7 eV and labeled 1, is
due to the π 0 and σ 0 non-bonding bands. The edges labeled 2 and 4 are the top and
bottom of the π valence bands, respectively. Feature 3 is the logarithmic singularity
SrTiO3 XPS (1486 eV)

I(E) (arbitrary units)
1
5
4
2
6
–7 –6 –5 –4 –3 –2 –1 0 1 2
Energy E − EV (eV)
Figure 8.1. XPS photoelectron energy distribution I(E) for SrTiO3 . The initial state
energy is measured from the top of the valence band, EV . (~ω = 1486 eV for Al Kα line.)
The theoretical I(E) is indicated by the solid curve. The XPS data is from [7].
in ρπ (E) at the center of the π band. The top and bottom of the σ valence bands
correspond to features 5 and 6, respectively.
In order to make a direct comparison between I(E) and the XPS data the the-
oretical curve should be broadened to account for the experimental resolution. The
broadened curve, denoted by hI(E)i, may be obtained from I(E) in the following
manner:
Z ∞ n ³ E − E 0 ´2 o
1
hI(E)i = √ I(E 0 ) exp − dE 0 , (8.33)
R π −∞ R
where the parameter R determines the resolution. The resolution decreases as R

increases. The FWHM (full-width at half-maximum)
√
FWHM = 2 R ln 2 = 1.665 R . (8.34)
The instrumental resolution for most XPS data is about 0.55 eV so that Rinst. =
0.33 eV. Figure 8.2 shows hI(E)i (thin dashed line) compared with the experimental
data for SrTiO3 using R = Rinst. = 0.33 eV. The solid curve passing through the
data is hI(E)i with a resolution of 1.34 eV (R = 0.8 eV). The agreement between
the data and hI(E)i with R = 0.8 eV is excellent and suggests that the effective
experimental resolution is less (is not as good) than the instrumental resolution.
SrTiO3 XPS (1486 eV)

hI(E)i (arbitrary units)
–7 –6 –5 –4 –3 –2 –1 0 1 2
Figure 8.2. XPS photoelectron energy distributions (dots [7] and little circles [9])
compared with hI(E)i for resolution parameters of R = 0.33 eV (FWHM 0.55 eV, thin
dashed line) and R = 0.8 eV (FWHM 1.34 eV, thick solid line). The cross-section ratio is
σp /σd ' 1/3.
The theoretical fit, hI(E)i in Fig. 8.2 indicates that the cross-section ratio is
σp (1486 eV)/σd (1486 eV) ' 1/3. The cross-sections are dependent on the energy of
the photons used in the photoemission experiment. For most XPS experiments with
103 . ~ω . 3 × 103 eV the cross-sections probably do not vary much. However, UPS
photoemission is performed with much lower photon energies (typically 5–50 eV)
and σp /σd can differ substantially from that found in XPS. As an example of this
effect consider the UPS spectrum of SrTiO3 for ~ω = 21.2 eV. The spectrum (open
circles) reported by Henrich et al. [10] is shown in Fig. 8.3. The solid curve is hI(E)i
calculated for the same parameters as used for the (solid curve) hI(E)i in Fig. 8.2
except that σp /σd = 1. The agreement between theory and experiment is essentially
exact. Thus, it appears that σp (21.2 eV)/σd (21.2 eV) = 1 and therefore the 21.2 eV
UPS spectrum closely resembles the total valence-band DOS of SrTiO3 .
SrTiO3 UPS (21.2 eV)

hI(E)i (arbitrary units)
–7 –6 –5 –4 –3 –2 –1 0 1 2
Figure 8.3. UPS spectrum of SrTiO3 from [10] (open circles) compared with hI(E)i for
σp /σd = 1 and R = 0.8 eV.
Some further comment on the UPS analysis is needed since 21.2 eV is not
sufficiently large to use the arguments employed in Section 8.1. In particular, it can
not be argued that the initial state energy is small compared to ~ω, since ~ω =
21.2 eV is comparable to the valence-band width.
In addition, modulation of the spectrum by varying matrix elements can also
be expected. The reason that the partial DOS model still applies is that the UPS
final states are energy bands derived from the Ti(4p), Ti(4s), and Sr(4s) orbitals.
These bands are presumably very broad and produce an approximately constant
final state DOS. The similarity of the UPS and XPS spectra tend to support this
conclusion. Further evidence comes from the studies of Powell and Spicer [11],
Derbenwick [12] and Henrich et al. [10] which suggest that the spectrum is not
changing rapidly with ~ω for 12 ≤ ~ω . 21 eV.
8.4 Nax WO3
In Chapter 6, Section 6.4 we discussed the x dependence of the electronic proper-

ties of Nax WO3 . These properties also influence the photoelectron distributions. In
addition to the x dependence the electronic parameters, the effective photoioniza-
tion cross-section ratio, σd /σp is approximately 12, for high-energy XPS and of the
order of 1 for UPS experiments [6].
For convenience the x-dependent parameters described in Section 6.4 (see
(6.93)) are given below:
Eg (x) = 2.85 + 0.84 x, (in eV) (8.35a)

(pdπ(x)) = 2.44 − 0.86 x, (in eV) (8.35b)
1
a(x) = (3.785 + 0.0818 x), (W−O distance in Å) (8.35c)
2
εF (x) = −2 + (2 − 1.086) x, (in eV) (8.35d)
6
ρπ (x) = (1 + 0.326 x). (8.35e)
2π
Using (8.20), (8.33), and (8.35) the XPS photoelectron distributions as a function of
x can be calculated. The results for I(E) are shown in Fig. 8.4 for several values of
x [13] and compared with the UPS experimental results of Hollinger et al. [14]. The
scale factor, C, of (8.20) was chosen so that the theoretical peak intensity matched
the experimental peak for x = 0.4. All other factors are known. The photo-emitted
electron distributions arise from electrons occupying the π ∗ bands. The energies of
these bands relative to the top of the valence band and the band widths change with
x because of the x dependence of Eg and (pdπ). The shifting of the peak intensity
toward higher energy as x increases is quite evident in Fig. 8.4.
The agreement between theory for XPS and the experimental UPS results
is quite remarkable considering that strong transition matrix elements effects can
modulate the UPS intensity curves. The areas under the theoretical curves compare
well with the areas of the experimental curves even though no adjustment have been
made to normalize the areas. Plasmon effects which contribute to the intensity tail in
the band-gap region on the low-energy side are not included in the theoretical curves
of Fig. 8.4. In addition, σd /σp = 12 for the theoretical curves while for UPS energies
it should be approximately 1. However, this difference does not have a strong effect
1.0
Nax WO3
0.9
0.8
Intensity (arbitrary units) 0.7 x = 0.83
0.6
0.5 x = 0.73
0.4
0.3 x = 0.62
0.2
0.1 x = 0.40
0.0
–6.0 –5.0 –4.0 –3.0
Energy (eV)
Figure 8.4. Comparison of the theoretical (solid curves) [13] and experimental (dashed
curves) [14] photoemission energy distribution curves for several values of x.
on the Nax WO3 theoretical curves because the amount of p-orbital mixing into the
band states of the lower part of the π ∗ bands is small. The contribution of the
plasmons is discussed in the next section.
8.5 Many-body effects in XPS spectra
When an electron is photo-ejected from a metallic solid, there is simultaneously

created a “hole” in the Fermi sea of electrons. The hole produces a number of
effects which are observed in the photoemission spectra of a solid.
If an electron is instantaneously ejected from a solid the state of the remaining
(N – 1)-electron system will not in general be the ground state. If we define the
binding energy Ebind of the emitted electron as the difference between the ground
state energies of the N - and (N – 1)-electron system, then it is clear that the energy,
E, deduced from the kinetic energy of the emitted electron is not Ebind . In fact,
E ≤ Ebind because the (N – 1)-electron system is left in an excited state after pho-
toemission process. The difference Er = |Ebind − E| is called the relaxation energy.
The quantity Er is the decrease in the energy of the (N – 1)-electron system after
it “relaxes” to its ground state. Relaxation shifts of electrons emitted from core
levels are often observed in photoemission experiments and have been the topic of
theoretical discussions [15–18].
The particular manner in which a solid manifests hole relaxation in photoe-
mission depends upon the photon energy, the electronic structure of the solid, and
the state from which the electron is emitted. If ~ω is near to the photoionization
threshold the kinetic energy of the photoelectron will be small. It will move slowly
away from the ion core and consequently its dynamics will be strongly influenced
by the attractive potential of the hole it leaves behind. If the emission were slow
enough to justify adiabatic relaxation of the (N – 1)-electron system then the ki-
netic energy of the emitted electron at the onset of photoemission would approach
Ebind − Φ. (However, this energy would not be that calculated by the usual one-
electron energy band theory; that is, it is not the eigenvalue based on Koopman’s
theorem which was discussed in Section 2.3.) For XPS experiments the photoelec-
tron kinetic energy is large and the “sudden approximation” is nearly valid. The
hole potential appears to be suddenly switched-on. According to the sudden ap-
proximation of perturbation theory the (N – 1)-electron system may be described
as a superposition of the eigenstates of the new Hamiltonian; the original Hamilto-
nian for the N -electron system plus the hole potential. Therefore, there is a specific
probability for each excited state of the (N – 1)-electron system. If Eα is the excited
state energy and E0 is the ground state energy then there will be a distribution of
peaks in the kinetic energies, Ekin,α , of the emitted electrons at
Ekin,α = ~ω − Φ − (Eα − E0 ) . (8.36)
For emission from core levels these series of peaks of (8.36) are called “shake-
up” peaks. In addition, during relaxation from the αth excited state to the ground
state there is a probability of a second electron being ejected. The latter spectrum
is called a shake-off satellite.
In a metallic material the conduction electrons will move rapidly to neutralize
a photohole. The adjustment of the conduction electrons to the hole has two im-
portant effects. First, electrons with energies near the Fermi energy will relax by
making transitions from states just below the Fermi level to excited single-particle
states above the Fermi level. A large number of low-energy electron–hole pairs can
be generated and their effect is to produce a “tail” on the low-energy side of any
characteristic peak in the XPS spectrum [19].
In addition to single-particle excitations, collective excitations in the form of
plasma oscillations can also be stimulated by hole relaxation of the conduction elec-
trons [20]. The excitation of plasmons produces satellite lines and structure in both
core-level and valence-band spectra. These plasmon effects are particularly strong
in metallic perovskites such as ReO3 and alkali tungsten bronzes. For example,
in Nax WO3 and Hx WO3 intense core-level satellite lines and band-gap emission
associated with plasmon creation have been observed [21, 22].
In order to analyze the XPS spectra of metallic perovskites it is necessary to
include the effect of plasmon creation on the photoelectron distribution. When hole
relaxation is accompanied by the creation of a plasmon an electron emitted from
an initial state of energy E will appear to have originated from a state at E − Epl ,
where Epl is the plasmon energy. This effect can be included in the theoretical model
of I(E) by using an apparent distributions I 0 (E) which includes the plasmon-shifted
electrons. We define
Z ∞ ½ µ ¶2 ¾
0 β 0 E − E0
I (E) = (1 − β) I(E) + √ dE exp − I(E 0 + Epl ) ,
Γ π −∞ Γ
(8.37)
where β is the plasmon creation probability. The factor (1 − β) is the fraction of

emitted electrons not accompanied by plasmon creation. The second term accounts
for the photoelectrons that are down-shifted in energy because of the plasmon effect.
The plasmon band is represented
√ by a Gaussian distribution centered at Epl . The
band width FWHM= 2 Γ ln 2. In the limit as Γ → 0, since then
I 0 (E) → (1 − β) I(E) + β I(E + Epl ) (8.38)
which states that the apparent number of photoelectrons from initial states at E is
the sum of contributions due to unshifted photoelectrons from states at E plus the
number from states at E + Epl which were down-shifted in energy due to plasmon
creation.
In comparing I 0 (E) with experiment the distribution must be convolved with
an experimental resolution function precisely as in (8.33) to produce the function
hI 0 (E)i.
A theoretical analysis of the XPS spectrum of Nax WO3 [21, 22] has been carried
out by Wolfram and Ellialtıoğlu [6] using (8.37). Their results are shown in Fig. 8.5.
The data is the dotted curve, the function I 0 (E) is the dashed curve and hI 0 (E)i
is the solid curve. The analysis provides a simple interpretation of the XPS data.
The peak near the Fermi level, EF , is due to electrons emitted from the partially
filled π ∗ bands, which contain 0.8 electrons per unit cell. The small peak in the
band-gap region ( ∼ – 1 to 3 eV) is due to conduction-band electrons shifted down
in energy due to plasmon creation associated with hole relaxation. The plasmon
energy Epl = 2.0 eV [21, 22]. The peak in I 0 (E) (the shoulder in the data) near
– 4.8 eV is due to emission of electrons from the non-bonding (π 0 and σ 0 ) bands.
The large central peak is produced by the logarithmic peak in the π valence-band
DOS. The lowest peak, near – 10 eV, arises from the jump discontinuity in the DOS
Na0.8 WO3 XPS (1486 eV)

hI ′ (E)i (arbitrary units)
–12 –10 –8 –6 –4 –2 0
Energy E − EF (eV)
Figure 8.5. Comparison of the XPS valence-band spectrum of Na0.8 WO3 with theory.
Data is the dotted curve [22], function I 0 (E) is the dashed curve and hI 0 (E)i is the solid
curve.
at the bottom of the σ valence band. The tail from about – 10 to – 12 eV is due to σ
band electrons shifted down in energy by the plasmon effect. The analysis indicates
that σp /σd (1486 eV) is about 1/12 and that β, the probability of plasmon creation,
is 0.2. Similar results are obtained from the theoretical analysis of the XPS spectra
of ReO3 and Hx WO3 [5, 21, 22].
The PDOS model, (8.20), appears to provide a useful method for analyzing
the XPS spectra of the perovskites. Application of the model indicates σp /σd
varies roughly between 0.3 and 0.1 for many of the perovskites. For the metallic
perovskites the many-body plasmon excitation probability is about 0.2 for both
core-level and valence-band emission.
References
[1] A. Damascelli, Z. Hussain, and Z.-X. Shen, Rev. Mod. Phys. 75, 473 (2003),
(A. Damascelli, Z.-X. Shen, and Z. Hussain, arXiv:cond-mat/0208504 v1 27
Aug 2002).
[2] N. V. Smith, Phys. Rev. B 3, 1862 (1971).
[3] D. A. Shirley, Phys. Rev. B 5, 4709 (1972).
[4] J. Freeouf, M. Erbudak, and D. E. Eastman, Solid State Commun. 13, 771
(1973).
[5] G. K. Wertheim, L. F. Mattheiss, M. Campagna, and T. P. Pearsall, Phys. Rev.

Lett. 32, 997 (1974).
[6] T. Wolfram and Ş. Ellialtıoğlu, Phys. Rev. B 19, 43 (1979).
[7] F. L. Battye, H. Höchst, and A. Goldmann, Solid State Commun. 19, 269
(1976).
[8] S. P. Kowalczyk, F. R. McFeely, L. Ley, V. T. Gritsyna, and D. A. Shirley, Solid
State Commun. 23, 161 (1977).
[9] D. D. Sarma, N. Shanthi, S. R. Barman, N. Hamada, H. Sawada, and K. Ter-
akura, Phys. Rev. Lett. 75, 1126 (1995).
[10] V. E. Henrich, G. Dresselhaus, and H. J. Zeiger, Bull. Am. Phys. (Soc. II ), 22
(1977), Solid State Research Report, Lincoln Laboratory, MIT (1976), p.39.
[11] R. A. Powell and W. E. Spicer, Phys. Rev. B 13, 2601 (1976).
[12] G. F. Derbenwick, Ph.D. Thesis (Stanford University, 1970) unpublished.
[13] T. Wolfram and L. Sutcu, Phys. Rev. B 31, 7680 (1985).
[14] G. Hollinger, F. J. Himpsel, B. Reihl, P. Petrosa, and J. P. Doumerc, Solid State
Commun. 44, 1221 (1982).
[15] D. W. Davis and D. A. Shirley, J. Electron Spectrosc. 3, 137 (1974).
[16] D. A. Shirley, J. Electron Spectrosc. 5, 135 (1974).
[17] P. H. Citrin and D. R. Hamann, Chem. Phys. Lett. 22, 301 (1973); Phys. Rev.
B 10, 4948 (1974).
[18] J. W. Gadzuk, J. Vac. Sci. Technol. 12, 289 (1975).
[19] S. Doniach and M. S̆unjić, J. Phys. C 3, 285 (1970).
[20] M. S̆unjić and D. S̆okćević, J. Electron Spectrosc. 5, 963 (1974).
[21] J-N. Chazalviel, M. Campagna, G. K. Wertheim, and H. R. Shanks, Phys. Rev.
B 16, 697 (1977).
[22] M. Campagna, G. K. Wertheim, H. R. Shanks, F. Zumsteg, and E. Banks, Phys.
Rev. Lett. 34, 738 (1975).
1. Derive (8.25) and (8.26) from (8.23).
2. Make a graph of ρdπ (E), given in (8.25), using the parameters (pdπ) = 1 eV, Et =
– 5 eV, and E⊥ = – 8.2 eV. A table of K(x) is given in Appendix B, and ρ(E) is given
in (8.30). Discuss the results in terms of covalency of the band states.
3. The angular frequency, ωp , for plasma oscillations of a metal is given by the relation
ωp2 ≡ 4πne e2 /me , where ne is the electron density. Show that in Gaussian (CGS) units
1/2
ωp = 5.65 × 104 ne rad/s when ne is the number of electrons per cm3 . Calculate the
plasmon energy for NaWO3 in eV.
4. The constant matrix element approximation for ε2 (ω) discussed in Chapter 7 leads to
a description involving the JDOS. The constant matrix element approximation for the
XPS energy distribution curve leads to a description involving the DOS. Explain the
major factors that cause this difference.
5. Explain the following terms used in the description of photoemission:

(a) hole relaxation;
(b) shake-up peaks;
(c) shake-off satellites.
9
Surface states on d-band perovskites
When a crystalline solid is terminated by a surface new types of energy bands

can form that are localized at or near the surface. A geometrically perfect surface
may have “intrinsic” surface states with energies lying within the band-gap region,
above or below the bulk energy bands. These surface bands have wavefunctions
that decrease exponentially with increasing distance into the crystal. The further
the energy of the surface state is from the bulk band-edge energy, the more rapidly
its wavefunction decreases with distance from the surface. Localized surface states
associated with defects such as oxygen vacancies can also occur. Such surface bands
and states can play an important role in chemisorption and catalysis in transition
metal oxides.
In this chapter we will review the theoretical concepts that underlie the forma-
tion of surface bands and defect states based on our empirical LCAO model. The
material is essential to the understanding of more fundamental and accurate calcula-
tion methods. A comprehensive review of the experiments on transition metal-oxide
surfaces is available [1], but only those relevant to doped insulating perovskites and
metallic Nax WO3 will be discussed here.
9.1 Perturbations at a surface
Figure 9.1 illustrates the two types of (001) surfaces for the perovskite structure.
For the type I surface the B and oxygen ions are on the surface. For the type II
surface the lattice starts with a layer of A and oxygen ions.
A number of different perturbations occur when a surface is formed even if the
lattice is terminated in a geometrically perfect way. In many cases the atomic layers
near the surface will “relax” by changing their interlayer and interatomic distances.
For example, the distance between the first and second layers of SrTiO3 differ by
about 5% from the interior layer spacing [1–3]. The LCAO interaction parameters
199
200 Surface states on d-band perovskites
Type I (001) surface Type II (001) surface

B ion
O ion
Figure 9.1. Type I and type II (001) surfaces of the cubic perovskite ABO3 structure.
The small circles represent B ions, and the large circles are the oxygen ions. The A ions
are not shown.
such as (pdπ) and (pdσ) depend exponentially on distance, therefore an expansion

or contraction of 5% can lead to substantial changes in the parameters.
A surface can “reconstruct” by forming atomic patterns unlike those of the
interior or by forming steps and terraces or a puckered surface. Patches of type
I and type II surfaces can be expected on fractured surfaces. These perturbations
alter not only the LCAO parameters, but also the electrostatic potentials. Occupied
surface states occurring in the band gap can create a surface dipole layer, cause band
bending, and may in some cases “pin” the Fermi energy.
The surface ions have missing neighbors (unsaturated bonds) that alter the
potentials experienced by the ions at and near the surface. For example, at a per-
fectly terminated type I (001) surface of ABO3 , cubic perovskite, the Madelung
potential at a surface B-ion site is a few eV less repulsive than at an interior site
[4]. On the other hand, the Madelung potential for a surface oxygen ion is essen-
tially unchanged from its bulk value. For a type II surface the change in the oxygen
potential is large and that for the B ion much smaller (see Appendix D for details).
The point group symmetries of the surface ions are also different from that
of the interior ions, causing additional splitting of the electronic levels. The point
group of a B ion on a (001) surface is C4v rather than Oh . As a result the t2g
group, threefold degenerate for Oh symmetry, is split into a doubly degenerate “e”
level and a singlet, b2 level. The point group symmetry of a surface oxygen is also
different and splitting of the doubly degenerate oxygen pi level (p⊥ ) occurs. Figure
9.2 shows schematically how the ion splittings are changed at the surface.
In general, the effects of the surface perturbations are largely confined to the
first one or two layers of the solid, but some surface effects may extend for large
distances. For example, for ferroelectrics such as BaTiO3 the surface creates a large,
long-range depolarization field.
9.1 Perturbations at a surface 201
eg (2)
6 a1
d (5) ∆(d)
b1
6 ∆VM (d) 6
? ?
t2g (3) 6 b2
e (2)
′
VM (d) VM (d)
? ?
(Bulk) (Surface)
(a)
x
p⊥ (2)
p (3) ∆VM (p) 6 6
6∆(p) ? z
6 ?
pk (1)
′
VM (p) VM (p)
? ?
(Bulk) (Surface)
(b)
Figure 9.2. Electrostatic splitting of ion levels at a (001) surface. The left-hand side
shows the bulk splitting and the right-hand side shows the splittings of ions on the surface
for a type I surface: (a) d-orbital splittings and (b) p-orbital splittings. ∆VM is the shift
in the Madelung potential.
In addition to “natural” changes that occur when a surface is formed, the

surface properties are also strongly influenced by the method of preparation and
surface treatment. Surfaces are formed by cleaving, fracturing, or by chemical or
epitaxial growth. In the performance of experiments a surface is often subjected to
various surface treatments such as polishing, annealing, or bombardment with ions.
Annealing in the presence of a reducing atmosphere (hydrogen for example) or an
oxidizing agent results in n- or p-type doping. Bombardment of a surface with ions
(argon ions, for example) is commonly used to clean the surface of impurities, but
also results in surface oxygen vacancies, leading to surface defect states.
9.2 Surface energy band concepts
(a) The (001) surface
In this section we shall be concerned with an ideal, semi-infinite perovskite termi-

nated by a (001) surface. As mentioned above there are two possible configurations
designated as type I and type II. The type I surface has the B ion and O ions
exposed on the first atomic layer forming a BO2 surface layer. The second atomic
layer has A cations and oxygen ions present and forms an AO layer. Therefore,
the semi-infinite, type I, perovskite lattice consists of atomic layers parallel to the
surface in the sequence BO2 –AO–BO2 –AO– · · ·. The type II (001) surface begins
with the AO layer and forms the sequence AO–BO2 –AO–BO2 – · · ·.
x δO
(pdπ)′′
(pdπ)′ n=1
z
d12 (Ti)
(pdπ)
(pdπ) n=2
d23 (Ti)
n=3
Ti ion Oxygen ion
Figure 9.3. Schematic of a type I (001) surface showing the displacement of the surface
oxygen ions normal to the surface. The interlayer spacings are designated by d12 (Ti) and
d23 (Ti). Two atomic planes make up a unit-cell layer. They are indicated on the right-hand
side of the diagram. Also shown are the perturbed LCAO parameters (pdπ)0 and (pdπ)00 .
A schematic type I (001) surface is shown in Fig. 9.3. For SrTiO3 it is found
that the surface oxygen ions are slightly displaced perpendicular to the surface and
the first two layers deviate from the bulk interlayer spacing by a few percent.
Beginning at the surface, each pair of successive atomic layers forms a unit-
cell layer with the composition ABO3 . If we number the unit-cell layers starting
from n = 1 at the surface to n = ∞, the position of any atom, R ~ j,m + ~τj,m , can be
specified by
~ j,m = ρ
R ~j,m + ~zj (n) (9.1)
where ρ x
~j,m = (xj,m + τj,m y
) ~ex + (yj,m + τj,m ~ j,m on the
) ~ey is the projection of R
z
xy-plane, and ~zj (n) = [zn + τj (n)] ~ez . Here zn is the distance of a B ion from the
surface, which for the infinite lattice is 2(n − 1)a, where 2a is the lattice spacing.
For the semi-infinite lattice the interlayer spacing may not be uniform. The notation
allows for the lattice spacing perpendicular to the surface to depend upon n, but
assumes that the x − y spacing is the same as in the bulk.
To begin our study of the surface energy bands we shall use the same nearest-
neighbor model employed for the discussion of the bulk bands. Since the electronic
and surface properties are principally determined by the π ∗ bands and the band-
gap surface states we shall limit our discussion to the pi bands. A more complete
discussion of the various types of surface bands that are possible can be found in
references [4] and [5].
We need to specify the types of surface perturbations to be considered. It might
be supposed that long-range Coulomb potentials such as the Madelung potentials
could be altered over many atomic layers near the surface. However, as mentioned
earlier, calculations [4] show that the Madelung potentials approach their bulk
values after the first atomic layer.
The changes in the d-orbital site potentials and the electron–electron repulsion
energy are the largest energies involved in the surface problem. For n-doped insu-
lators the density of d electrons at the surface is much larger than for the interior
ions when states form in the band-gap region. Therefore special attention must be
paid to the Coulomb repulsion effects.
The next largest energies are the changes in LCAO two-center integrals such
as (pdπ). We shall consider the perturbations in the first unit-cell layer, but assume
that all other layers are described by the same parameters as for the infinite lattice.
For a geometrically perfect, (001) surface the pi and sigma bands do not mix
and may be considered separately just as in the case of the bulk energy bands.
When there are small displacements of the surface oxygen ions the pi and sigma
orbitals are coupled by two-center integrals that are first order in the displacement,
but the energy is affected only in second order. In this chapter we shall ignore the
small mixing between the pi and sigma orbitals.
There are three different pi-type surface bands: those involving dyz orbitals,
those involving dxz orbitals and those involving dxy orbitals. We shall refer to these
bands as the pi(yz), pi(xz), and pi(xy) surface bands. The parameters for the type
I, pi(yz) surface bands are:

£ ¤
Et (n) = Et + ∆Et + U Ns δn,1 (9.2)
E⊥ (n) = E⊥ + ∆E⊥ δn,1 (9.3)
¡ ¢

 1 + ∆00 δn,1 (pdπ) between surface d and p orbitals
¡
 ¢
1 + ∆0 δn,1 (pdπ) between surface d and the
(pdπ)n = (9.4)

 first subsurface p orbitals


(pdπ) otherwise
∆Et and ∆E⊥ are the changes in diagonal matrix elements for the B and O sites,
respectively, and ∆0 and ∆00 are the fractional changes in the p–d interactions. In
(9.2) U is the Coulomb repulsion among the electrons occupying the same surface
d orbital and Ns is the number of electrons occupying the surface state per spin.
The surface energy bands and the potential, U Ns , must be calculated self-
consistently. The total number of electrons occupying surface states is Ns =
Ns (xy) + Ns (yz) + Ns (xz), where Ns (αβ) is the number of electrons occupying
the pi(αβ) surface band per spin state. The self-consistent solutions for the surface
bands require that all three surface bands be considered simultaneously.
(b) Pi(yz) surface energy bands
Surface states involving the dyz , py (~r − a~ez ), and pz (~r − a~ey ) orbitals are symmetry
equivalent to those states involving dxz , px (~r − a~ez ), and pz (~r − a~ex ), and therefore,
we need only consider the pi(yz) states.
The LCAO equations that determine the eigenvalues and eigenvectors for the
semi-infinite lattice are:
¡ ¢ ¡ ¢ ¡ ¢
ωt0 − ω cyz (1) + 2iSy 1 + ∆00 cz (1) + 1 + ∆0 cy (1) = 0, (9.5)
¡ 0 ¢ ¡ ¢
ω⊥ − ω cz (1) − 2iSy 1 + ∆00 cyz (1) = 0, (9.6)
¡ ¢ ¡ ¢
ω⊥ − ω cy (1) + 1 + ∆0 cyz (1) − cyz (2) = 0, (9.7)
and for n > 1

¡ ¢
ωt − ω cyz (n) + 2iSy cz (n) + cy (n) − cy (n − 1) = 0 (9.8)
¡ ¢
ω⊥ − ω cz (n) − 2iSy cyz (n) = 0, (9.9)
¡ ¢
ω⊥ − ω cy (n) + cyz (n) − cyz (n + 1) = 0, (9.10)
with Sy = sin ky a. The terms cyz (n), cz (n), and cy (n) are the amplitudes of the
dyz (n), pz (n), and py (n) orbitals, respectively. In (9.5)–(9.10) we have introduced
a number of dimensionless quantities,
ω = E/(pdπ), (9.11a)
ωt = Et /(pdπ), (9.11b)
ω⊥ = E⊥ /(pdπ), (9.11c)
∆ωt = ∆Et /(pdπ), (9.11d)
∆ω⊥ = ∆E⊥ /(pdπ), (9.11e)
u = U/(pdπ), (9.11f)
ωt0 = ωt + ∆ωt + uNs , (9.11g)
0
ω⊥ = ω⊥ + ∆ω⊥ , (9.11h)
ωg = Eg /(pdπ). (9.11i)
The amplitudes cy and cz can be expressed in terms of cyz using (9.6), (9.7),
(9.9) and (9.10). Substitution of these results into (9.5) and (9.8) yields the reduced
secular equation,
£ ¤ ¡ ¢
−2 cos θ + ∆p(ω, ky ) cyz (1) + 1 + ∆0 cyz (2) = 0, (9.12)
−2 cos θ cyz (n) + cyz (n + 1) + cyz (n − 1) = 0, (9.13)
where
¡ ¢¡ ¢
−2 cos θ = ωt − ω ω⊥ − ω − 4Sy2 − 2 (9.14)
and
½ ¢¡ ¢2 ¾
¡ ¢¡ ¡ ¢ ¢
0 2 2 (ω⊥ − ω 1 + ∆00
∆p(ω, ky ) = ∆ωt + uNs ω⊥ − ω − 1 + ∆ − 4Sy ¡ 0 ¢ − 1 + 2,
ω⊥ − ω
(9.15a)
£
∆0 = (pdπ)0 − (pdπ)]/(pdπ), (9.15b)
£
∆00 = (pdπ)00 − (pdπ)]/(pdπ). (9.15c)
Equations (9.12) and (9.13) are simple second-order difference equations and
the general solutions are:
1 ³ inθ ´
cyz (n) = √ e + Λe−inθ , (9.16)
N
½ ¾
1 − ∆p(ω, ky )eiθ − ∆0 e2iθ
Λ=− (9.17)
1 − ∆p(ω, ky )e−iθ − ∆0 e−2iθ
where N is a normalization constant.

(c) Classification of the states
The solutions of (9.16) can be classified as volume states or surface states depending
upon the behavior of the wavefunction amplitudes at large n. The volume states
have wavefunctions whose amplitudes extend unattenuated throughout the entire
semi-infinite lattice while the wavefunctions for the surface states have amplitudes
that decrease exponentially with increasing distance into the solid.
Volume states
The conditions for a volume state are that its wavefunction remains bounded and
non-vanishing as n → ∞. These two conditions can be met only if the factor, θ, in
equation (9.16) is a real number. For real θ the denominator in (9.17) is the negative
complex conjugate of the numerator and hence Λ is a complex number with unit
modulus. Therefore we may write Λ = −e−iδ , where δ is a real number and
1
cyz (n) = √ (einθ − e−i(nθ+δ) ). (9.18)
N
Since θ is a real number it follows that | cos θ| ≤ 1. Using this result with (9.14)
yields the inequality:
|(εt − ω)(ε⊥ − ω) − 4Sy2 − 2| ≤ 2. (9.19)
This equation may be solved to determine the possible values of the dimensionless
energy, ω, for which volume states can exist. One finds two regions of (ω, ky )-space
which satisfy (9.19):
³ π´
ωπ∗ (ky , 0) ≤ ωvol (ky ) ≤ ωπ∗ ky , , (9.20)
³ 2a
´
π
ωπ (ky , 0) ≤ ωvol (ky ) ≤ ωπ ky , , (9.21)
2a
where ωvol (ky ) is a volume state energy and ωπ∗ (ky , kz ) and ωπ (ky , kz ) are the bulk
(infinite lattice) dimensionless energy band dispersion relations,
rh i2
¡ ¢ 1 1
ω π∗ (ky , kz ) = (ωt + ω⊥ ) ± (ωt − ω⊥ ) + 4(Sy2 + Sz2 ) . (9.22)
π 2 2
Thus the volume state energies are confined to the same ω − ky regions as the
bulk (infinite-lattice) energies. These regions are between the bottom and top of
the π ∗ (yz) and between the bottom and top of the π(yz) bands as shown in Fig.
9.4. We shall refer to these regions as the “bulk continuum of states” or just the
“bulk continuum”. The volume states form the same continuum of energies as the
infinite-lattice states. For every pair (ω, ky ) for which there is a solution of the
infinite lattice, there is also a volume state solution. The energies of the volume
states are the same as those of the infinite lattice, but the wavefunctions are quite
eiθ = −e−β Region IIa

cyz (n) = (−1)n e−nβ
Surface states
Volume states
π ∗
iθ
|e | = 1
ωt
Unitless energy ω
Surface states
0 < eiθ < 1 Region I
eiθ = e−β
cyz (n) = e−nβ
ω⊥
|eiθ | = 1
π
Volume states
Surface states
eiθ = −e−β
cyz (n) = (−1)n e−nβ Region IIb
π
0 ky a 2
Figure 9.4. Surface energy band and volume energy band regions of (ω, ky ) space for the
pi(yz) states.
different. For example, the square of the d-orbital amplitude is not uniform on the
various layers since
2£ ¤
|cyz (n)|2 = 1 − cos(2nθ + δ) . (9.23)
N
In fact, for a given volume state the square of the d-orbital amplitude will have
maxima or minima on the nth layer whenever
jπ
n= − δ, where j = 0, ±1, ±2, ±3, . . . . (9.24)
2θ
When j is even the d-orbital probability on the nth layer vanishes, while for j
odd the d-orbital amplitude is twice the average value. Since the positions of these
maxima and minima vary with the particular volume state (i.e., vary with θ) the
average d-orbital probability inside the lattice quickly approaches the infinite-lattice
average.
Surface states
The wavefunctions for the surface states have the property that cyz (n) → 0 as
n → ∞, but is non-vanishing for at least one value of n. It is obvious from (9.16)
that these conditions are met if
Im θ > 0,
Λ = 0. (9.25)
The requirement that the imaginary part of θ > 0 means that einθ → 0 with increas-
ing n as e−n(Imθ) . The second requirement, Λ = 0, imposes an eigenvalue condition,
namely the surface state condition that
1 − ∆p(ω, ky ) eiθ − ∆0 e2iθ = 0. (9.26)
The surface energy bands are specified by pairs, (ω, ky ), that satisfy (9.26).
These pairs define the surface state dispersion curves, ωs (ky ). The surface states
are highly localized since |cyz (n + m)/cyz (n)|2 = e−2m(Imθ) .
According to (9.25), we may write
θ = α + iβ,
where α and β are real numbers and β > 0. This gives
cos θ = cos α cosh β − i sin α sinh β. (9.27)
However, for real energy, ω, (9.14) requires that cos θ be real. This is compatible
with (9.27) only if
α = `π, (` = 0, ±1, ±2, . . .). (9.28)
There are two distinct cases: ` is 0 or an even integer, and ` is an odd integer. We
shall use ` = 0 and ` = 1. Other choices lead to equivalent results. We have
eiθ = e−β `=0 (9.29)
iθ −β
e = −e ` = 1. (9.30)
In either case eiθ is real and 0 < |eiθ | < 1.
Consider the case for which 0 < eiθ < 1. We have cyz (n) ∝ e−nβ and decreases
uniformly with increasing distance into the semi-infinite lattice.
Since cos θ = cosh β, and cosh β > 1 for β > 0, we have the inequality
(ωt − ωs )(ω⊥ − ωs ) − 4Sy2 − 2 < −2. (9.31)
If (9.31) is solved for ωs we find that these surface state energies must be in the
band-gap region between the π ∗ and π volume bands (region I in Fig. 9.4)
ωπ (ky , 0) ≤ ωs (ky ) ≤ ωπ∗ (ky , 0). (9.32)
For the case that −1 < eiθ < 0, the d-orbital amplitude, cyz (n), alternates in
sign from one unit layer to the next,
cyz (n) ∝ (−1)n e−nβ . (9.33)
The quantity cos θ = − cosh β < −1 so that the surface states of this type have
ωs (ky ) that satisfies the equation
(ωt − ωs )(ω⊥ − ωs ) − 4Sy2 − 2 > 2. (9.34)
The allowed regions for ωs (ky ) are

³ π´
ωs (ky ) ≥ ωπ∗ ky , (9.35)
2a
and
³ π´
ωs (ky ) ≤ ωπ ky , . (9.36)
2a
These amplitude-oscillating surface states occur only above the π ∗ band or below
the π band. The regions are designated as regions IIa and IIb in Fig. 9.4.
(d) Pi(yz) density of surface states
The density of surface states (DOSS) can be found from the eigenvalue equation,
(9.26), which may be written as
∆p − 2 cos θ 1 + ∆0
+ + 2 cos θ = 0, (9.37)
1 + ∆0 ∆p − 2 cos θ
where −2 cos θ and ∆p are defined by (9.14) and (9.15), respectively. Equation
(9.37) may be solved for the variable Sy2 (which appears in both ∆p and −2 cos θ)
to obtain the quadratic equation
A(−4Sy2 )2 + B(−4Sy2 ) + C = 0, (9.38)
with
A = ηt + η 2 , (9.39a)
B = 2γη + ξηt + γt, (9.39b)
2 2
C = γ + t + γξt, (9.39c)
2
γ = (∆ωt + uNs )(ω⊥ − ω) − t + 2 + (1 − t)ξ, (9.39d)
(ω⊥ − ω)r2
η= 0 − ω) − t, (9.39e)
(ω⊥
ξ = (ωt − ω)(ω⊥ − ω) − 2, (9.39f)
0
t = (1 + ∆ ), (9.39g)
00
r = (1 + ∆ ). (9.39h)
From (9.38) we obtain
½ sµ ¶2 ¾
1 B B
Ω(ω, ky ) ≡ − − ± − AC = Sy2 . (9.40)
4A 2 2
The DOSS is then
µ ¶−1
2a dΩ 1 1
ρs (Ω) = = p , (9.41)
π dky π Ω(1 − Ω)
and the DOSS, ρs (ω), as a function of ω, is given by
½ ½ sµ ¶2 ¾¾
dΩ d 1 B B
ρs (ω) = ρs (Ω) = ρs (Ω) − − ± − AC . (9.42)
dω dω 4A 2 2
The surface band is one-dimensional, depending only on ky , and therefore the

DOSS has square-root singularities at Ω = 0 and Ω = 1.1
The number of electrons occupying the surface band can be written as
Z ΩF
1 1
Ns (yz) = ρ(Ω) dΩ = − arcsin(1 − 2ΩF ), (9.43)
0 2 π
where ΩF corresponds to the Fermi energy
ΩF = Ω(ωF ), (9.44)
with
ωF = EF /(pdπ). (9.45)
Consider the case of the “perfect” surface defined to be a surface for which
all of the perturbation parameters, ∆ωt , ∆ω⊥ , ∆0 , and ∆00 are zero. In this case
∆p(ω, ky ) = 1 and the eigenvalue condition, (9.26), gives eiθ = 1 which violates the
surface state requirement that einθ → 0 as n → ∞. Therefore we can conclude that
there are no surface states on a “perfect” type I (001) surface. That does not mean
that the states are the same as those of the infinite lattice. It means that all of the
1
Equation (9.41) is valid so long as the quantity D = (B/2)2 − AC ≥ 0. If D < 0 it indicates that the
surface energy band is truncated by intersecting the volume continuum. The truncation occurs at
the value of ky for which D = 0. Ω = 0 and Ω = 1 correspond to the bottom and top of the surface
state band, respectively. In the case of a truncated surface band, one of the singularities occurs at the
truncation energy.
states belong to the volume continuum. The energies are the same as those of an
infinite lattice, but the wavefunctions are modified by the presence of the surface.
Of course, a “perfect” surface as we have defined it here is not the same as an
“ideal” surface which is defined to be a surface that is geometrically perfect. The
ideal surface will have non-zero perturbations even though it is atomically perfect.
Furthermore, the energies of the surface states will depend upon the position of
the Fermi energy and the number of electrons in the surface states. The Coulomb
repulsion between electrons of opposite spin occupying the same surface d orbital
will shift the energy calculated for the unoccupied surface state.
As a simple tutorial example consider the solution of (9.26) when the only
non-zero parameter is κ = (∆ωt + uNs ) and examine the solutions as κ → 0. From
(9.26) we obtain
1
eiθ = , (9.46)
κ(ω⊥ − ω) + 1
2 cos θ = 2 + κ2 (ω⊥ − ω)2 + higher−order terms. (9.47)
Using (9.14) yields the surface state eigenvalue equation,
(ωt − ω)(ω⊥ − ω) − 4Sy2 + κ2 (ω⊥ − ω)2 ∼
= 0. (9.48)
For κ = 0, (9.48) describes the bottom of the π ∗ band as a function of ky .

Therefore, for small κ the surface band must lie near the π ∗ band edge, ωπ∗ (ky , 0).
As a result, we may replace ω in the last term of (9.48) by ωπ∗ (ky , 0) so that to
second order in κ the surface band is given by
rh i2
1 1 ¡ ¢2
ωsπ∗ ≈ (ωt + ω⊥ ) + (ωt − ω⊥ ) + 4Sy2 − κ2 ω⊥ − ωπ∗ (ky , 0) , (9.49)
2 2
¡ ¢2
κ2 ω⊥ − ωπ∗ (ky , 0)
≈ ωπ∗ (ky , 0) − . (9.50)
2 ωπ∗ (ky , 0) − (ωt + ω⊥ )
The denominator in (9.50) is positive because the π ∗ band edge necessarily lies
above the mid-gap energy, 12 (ωt + ω⊥ ). Therefore it follows that there is an entire
surface band below the π ∗ band edge, ωπ∗ (ky , 0), that lies in the band-gap region.
This result is valid only for values of κ that are negative.2
For a doped, n-type perovskite, the Fermi level will be near the bottom of
the π ∗ band edge at Γ, within a few meV of Et (i.e., ωF very close to ωt ) and
the surface band will be partially occupied as indicated in Fig. 9.5(a). The pa-
rameter κ = ∆ωt + uNs is a balance between the additional negative electrostatic
potential and the positive electron–electron repulsion and must be calculated self-
consistently. However, we can see that for negative values of κ there will always
2
For positive κ the surface band is located above the top of the volume continuum. That is, above the
top of the π ∗ band (Region IIa) and/or below the bottom of the π band (Region IIb).
–3 5 2
(a) (b) (c)
DOSS(yz)
Ns (yz)
–4 π∗ 3
Volume-state 1
ω
continuum
2
Surface-state
bands
–5 EF = Et 1
Occupied
surface states
0 0
0 ky a π/2 0 Ω 1 0 Ω 1
Figure 9.5. (a) Schematic showing the occupied surface states with the Fermi energy
EF = Et , the bottom of the bulk conduction band. (b) DOSS and (c) Ns versus Ω for
pi(yz) or pi(xz) surface states.
be a surface band below the bulk band in the band-gap region. If we keep ∆ωt
fixed and increase the value of u, the value of κ becomes less negative. That will
move the surface band closer to the bulk band-edge and reduce Ns . The effect of
reducing Ns is to make κ more negative and to counteract the effect of increasing u.
Therefore, the surface band will find a self-consistent solution between the energy
of the unoccupied surface state (Ns = 0) and the lower edge of the π ∗ continuum.
With the mean-field representation of the Coulomb repulsion we are using here, U
would have to be infinite to force the surface band completely out of the band gap.
Because of the canceling effects between uNs and ∆ωt the surface bands tend to
be near to the edge of the π ∗ continuum even when ∆ωt is a few eV negative.
It should be noted that for an n-doped insulator, the average number of elec-
trons in a surface d orbital is much larger than for the interior d orbitals. For
example, for a doping level of 1018 cm−3 , the average occupation (beyond that due
to covalent bonding) of an interior d(t2g ) orbital is 6.4×10−5 electrons. On the other
hand, a surface d(t2g ) orbital’s occupation is of the order of unity when a surface
band lies within the band gap. Therefore, for the insulators, electron–electron cor-
relation effects are more important for the surface energy bands and surface defect
states than for the volume states.
(e) Pi(xy) surface energy bands
The surface band involving the dxy orbitals is easily derived since in the approxima-
tion of nearest-neighbor interactions, the unit-cell layers are uncoupled. Therefore
9.3 Self-consistent solutions for the band-gap surface states: SrTiO3 213
we can immediately express the energy bands for the surface unit-cell layer as
µ 0 ¶ sµ 0 ¶2
±
0
Et + U Ns + E⊥ Et + U Ns −E⊥ 0 £ ¤2³ ´
Es (kx , ky ) = ± +4 (pdπ)00 Sx2 +Sy2 ,
2 2
(9.51)
where the primed and double primed symbols are the perturbed surface parameters.
The form of the surface energy band dispersion is identical to the dispersion of the
bulk π and π ∗ bands and therefore the DOSS, ρs (E), can be obtained by making
the following substitutions:
ρπ (E) → ρs (E), (9.52a)

Et → Et0 + U Ns , (9.52b)
00
(pdπ) → (pdπ) , (9.52c)
0
E⊥ → E⊥ , (9.52d)
(here A → B means replace A by B),
into the expression for the pi DOS given by equation (6.28). This yields,
¯ ¯ µr ³ ε(E) ´2 ¶ · ³ ε(E) ´2 ¸
1 ¯E − 12 (Et0 + U Ns + E⊥ 0 ¯
)
ρs (E) = 2 K 1 − Θ 1 − ,
π [(pdπ)00 ]2 2 2
(9.53)
where
£ ¤2
E − 12 (Et0 + U Ns + E⊥ 0
) − Eg02
ε(E) = £ ¤2 − 2. (9.54)
2 (pdπ)00
Et0 = Et + ∆Et is the perturbed site Madelung potential at the d ion and Eg0 =
Et0 + U Ns − E⊥
0
is the perturbed energy gap for the surface unit layer.
The DOSS for the pi(xy) surface band has jump discontinuities at its band
edges and a logarithmic singularity at its band center just as in the case of the
bulk states. The energies at which these discontinuities occur depend upon the
perturbation parameters. The jump in the DOSS per spin state at E = Et0 is,
1 Eg0
ρs (Et ) = £ ¤ . (9.55)
2π (pdπ)00 2
9.3 Self-consistent solutions for the band-gap

surface states: SrTiO3
The (001) surface of SrTiO3 is typical of the surfaces of the insulating perovskites.
Low-energy electron diffraction (LEED) and reflection high-energy electron diffrac-
tion (RHEED) experiments have been used to investigate the geometry of the sur-
face [1, 2]. It is found that the surface oxygens move slightly upward, creating
a puckered surface. LEED [1] experiments indicated 2% expansion of the distance
d12 (Ti), but a contraction by 2% of d23 (Ti) (see the definitions in Fig. 9.3). The sur-
face oxygen displacement, δO , is 4% of the Ti–O spacing. For the type II surface the
first layer Sr–O distance, d12 (Sr), is contracted by 10%±2% and d23 (Sr) expanded
by 4%±2%. The surface oxygen displacement was 8.2%±2%. Later RHEED [2]
experiments confirmed the puckering due to displacement of the surface oxygens,
but found expansion of both d12 (Ti) (3.6%) and d23 (Ti) (5.1%). The displacement
of the ions near the surface creates a static dipole moment whose polarization is
estimated [1] to be about 0.17 cm−2 .
In this section we consider only the largest surface perturbations, ∆ωt , the
change in the electrostatic potential at the surface d-orbital site and uNs , the
average additional Coulomb repulsion at the B-ion site due to the occupation of the
surface states. The effects of other surface perturbations are discussed in references
[4] and [5]
As mentioned previously, because Ns is the total number of electrons per spin
state in all of the surface bands we must calculate the electronic occupations of the
pi(xy), pi(yz), and pi(xz) surface bands simultaneously to achieve self-consistent
solutions. To begin with we assume that the surface oxygen site potential and p–
0
d interactions are unperturbed. That is, E⊥ = E⊥ , and (pdπ)0 = (pdπ)00 = (pdπ).
The only perturbation is then the change in the diagonal energy at the surface
B-ion site, ∆ωt + uNs . The actual value of U for various perovskites is not known.
For the Ti ion the difference between the ionization potentials for Ti+4 and Ti+3 ,
∆Ip = 15.75 eV. This value, appropriate for atomic states, should be an upper bound
on the possible value of U for this material. A reasonable estimate is that U is one-
half to one-third of ∆Ip . For SrTiO3 , the band gap is 3.2 eV and (pdπ) is between
0.84 and 1.3 eV based on LCAO fits to different energy band calculations [4, 6].
The change in the Madelung potential, ∆ωt , for the ideal surface is about −2 eV
at a surface Ti ion. Because the actual (001) surface is puckered, the precise value
is uncertain, but most likely it is negative (less repulsive than at interior ions). We
will explore the self-consistent solutions as a function of ∆ωt and u and for the
examples here assume (pdπ) =1 eV.
To find self-consistent solutions we need to calculate Ns as a function of κ. For
the pi(yz) or pi(xz) surface bands we have the eigenvalue equation
1
(ωt − ω)(ω⊥ − ω) − 4Sy2 − 2 + ∆p + = 0, (9.56)
∆p
with
∆p(ω) = κ(ω⊥ − ω) + 1. (9.57)

For doping concentrations less than or equal to about 1018 cm−3 we may use
EF /(pdπ) = ωF = ωt , the bottom of the bulk π ∗ band. This means the occupied
surface states lie in the range ωt0 ≤ ω ≤ ωt as illustrated schematically in Fig. 9.5.
The contribution to Ns (per spin state) from pi(yz) or pi(xz) surface bands is
given by
Z Ω(ωt )
1 1 ¡ ¢
Ns (αβ) = ρ(Ω) dΩ = − arcsin 1 − 2Ω(ωt ) , (9.58a)
0 2 π
and from (9.56) and (9.57) we have
(κωg )2
Ω(ωt ) = . (9.58b)
4(1 − κωg )
For the pi(xy) band the DOSS (per spin state) is

1 0
ρxy (ω) = (ω − ωm )K(x), (9.59a)
π2
³ ε(ω) ´2
x2 = 1 − , (9.59b)
2
1h 0 2
³ ω 0 ´2 i
g
ε(ω) = (ω − ωm ) − − 2, (9.59c)
2 2
0
where K(x) is the complete elliptic integral of the first kind, ωm = 12 (ωt + κ + ω⊥ ),
0
and ωg = ωg + κ. The number of electrons occupying the surface band is
Z ωt
Ns (xy) = ρxy (ω) dω. (9.60)
ωt +κ
Ns (xy) may be approximated by

1 1 1
Ns (xy) ≈ (κ2 − ωg0 κ) + (ω 0 )2 κ2 + ω 0 κ3 . (9.61)
4π 16π g 48π g
Equation (9.61) errs less than 2% for values of κ2 < 0.2. The total number of elec-
trons occupying all three surface states is then,
Z ωt h1 1 ¡ ¢i
Ns = Ns (xy) + 2Ns (yz) = ρxy (ω) dω + 2 − arcsin 1 − 2Ω(ωt ) . (9.62)
ωt +κ 2 π
The procedure for obtaining self-consistent solutions is to first choose a value

for κ, calculate Ns using (9.62), then for a given value ∆ωt use the expression
κ − ∆ωt
u= , (9.63)
Ns
to obtain the self-consistent value of u.
Figure 9.6 shows√the self-consistent solution of Ns versus κ. The kink in the
curve at κ = −2(1 + 2)/ωg ( ≈ −1.5 for SrTiO3 ) corresponds to the point at which
12
Total Ns (per spin)
u
1
4
0 0
–2 –1 0 –2 –1 0
(a) κ (b) κ
Figure 9.6. Self-consistent parameters for the surface energy bands. (a) Ns versus κ,
(b) u versus κ (for these plots, ωg = 3.2 and ∆ωt = −2).
the pi(xz) and pi(yz) bands lie entirely below the conduction-band edge so that
they are completely occupied. Beyond this point the occupation of the pi(yz) and
pi(xz) surface bands do not change. The dispersion of the pi(xy) band is larger than
that of the pi(yz) surface band because it is two-dimensional. Therefore, it takes a
larger negative value of κ to drop it entirely below the conduction-band edge. The
parameters u, κ, and ∆ωt are in units of (pdπ). For SrTiO3 with (pdπ) =1 eV, the
results may be read in units of electronvolts. Figures 9.7(a) and (b) show the pi(xy)
and pi(yz) surface energy bands, respectively, for ∆Et = −2 eV with u = 0 and
u = 6 (U = 0 and 6 eV for SrTiO3 ). As can be seen, the Coulomb repulsion forces the
surface bands toward the edge of the continuum of states. For U = 6 eV, the pi(yz)
and pi(xz) surface bands lie within 0.053 eV of the edge of the conduction band.
The pi(xy) band lies lower in energy than the pi(yz) and pi(xz) bands but is still
within 0.055 eV of the conduction-band edge for U = 6 eV. The energy displacement
of the pi(xy) surface band at Γ below the bulk conduction-band edge is in general
equal to κ(pdπ). Some results are summarized in Table 9.1. They show that Ns per
surface unit cell is in the range of 1014 cm−2 for a wide variety of the parameters.
The entries in the table can be used for any value of (pdπ), but the energy gap is
–3 –3
Bulk Bulk
–4 6u = 6 –4
continuum continuum
6u = 6
–5 –5
u=0
ω
ω
u=0
–6 –6
–7 –7
(a) (b)
–8 –8
Γ X M Γ X
Figure 9.7. Self-consistent surface bands. (a) Pi(xy) surface band for u = 0 and u = 6
and (b) pi(yz) surface band for u = 0 and u = 6.
fixed at 3.2 eV and the electron concentration is calculated for a lattice spacing of
3.905 Å: values appropriate for SrTiO3 .
Table 9.1. Self-consistent surface state parameters.
Surface concent.
U ∆ωt κ Ns (total) Ns (xy) 2Ns (yz) (1014 cm−2 )
15 – 2.0 – 0.0609 0.1293 0.0157 0.1136 0.8481
15 – 1.0 – 0.0294 0.0648 0.0075 0.0572 0.4249
15 – 0.5 – 0.0144 0.0324 0.0037 0.0287 0.2125
6 – 2.0 – 0.1589 0.3070 0.0415 0.2654 2.0132
6 – 1.0 – 0.0738 0.1545 0.0191 0.1355 1.0132
6 – 0.5 – 0.0354 0.0775 0.0091 0.0684 0.5082
3 – 2.0 – 0.3236 0.5589 0.0861 0.4729 3.6652
3 – 1.0 – 0.1458 0.2849 0.0381 0.2468 1.8683
3 – 0.5 – 0.0684 0.1441 0.0177 0.1264 0.9450
The results obtained here do not explicitly include the effects of surface charge.
A high density of occupied surface states in the band gap can cause “band bend-
ing” if the bulk density of electrons is insufficient to screen these surface charges.
However, the term U Ns has the same effect and therefore no explicit surface charge
term needs to be added to the model.
(a) Other types of surface bands
In the previous discussion it has been assumed that ∆Et is negative. This prejudice
is based on model calculations of the electrostatic (Madelung) potential for a B ion
on a ideal type I (001) surface for which ∆Et is several eV negative. However, for
non-ideal surfaces there is the possibility that ∆Et and κ could be positive; that is,
more repulsive than at an interior site. (See Appendix D for a table of Madelung
potentials.) In this case, the surface theory produces truncated surface bands that
lie just above the top of the π ∗ conduction band. Surface bands are also produced
just above the π valence band, but they do not lie in the fundamental band-gap
range (i.e., Γ-band-gap region between Et and E⊥ ). Such states would be difficult
to observe optically or by photoemission. The surface bands near the top of the
conduction band will be unoccupied and close to the jump in the bulk density of
states at the top of the band. The surface bands split off from the valence bands
would be very near to the bulk non-bonding band energies. Thus their contribution
in optical or photoemission experiments could also be obscured by the high bulk
density of occupied valence-band states. Figure 9.8(a) shows the surface bands for
both positive and negative values of κ with U = 0. More details on these types of
surface bands can be found in [4].
On type II surfaces calculations of the electrostatic potentials indicate that
∆Et is nearly unchanged, but that ∆E⊥ is several eV positive (less attractive). For
this perturbation nearly flat surface bands appear in the fundamental band-gap
region above the π valence bands whose wavefunctions are composed primarily of p
orbitals [4]. These bands are illustrated in Fig. 9.8(b). For the p-like surface bands
the electron–electron repulsion is much smaller than that for the d orbitals and
because the bands are flat a large DOSS results. In n-doped insulators and metallic
perovskites such band states would be occupied.
For metallic perovskites such as Nax WO3 (x > 0.3) the theory of the surface
bands described above applies with some modifications. First, the Fermi level is no
longer pinned at the bottom of the conduction band and hence the surface bands
will have much higher electron concentrations. Second, the concentration of bulk
electrons is sufficient to screen the surface charge associated with the occupied
surface states. Third, although the Coulomb repulsion energy is still large, the dif-
ference between the bulk and surface state density of electrons is much smaller.
The Coulomb repulsion, U Ns is still operative for both the bulk and surface states,
but it is roughly the same for both types of bands. For our empirical LCAO model
this means that the correlation effects can be assumed to be incorporated into the
parameters of the model and therefore the effects of self-consistency are less im-
portant when employing the theory. As a first approximation, we can calculate the
surface band energies in the same manner as the bulk energy bands are calculated,
as though the Coulomb repulsion parameter, U , were zero. Therefore, we expect
κ = +1
π∗ π∗
ωt ωt
Unitless energy ω
Unitless energy ω
–2
∆ω⊥ = 2.0
–4
∆ω⊥ = 1.0
ω⊥ ω⊥ ∆ω⊥ = 0.5
+1
π π
–2
–4
π π
(a) 0 ky a 2 (b) 0 ky a 2
Figure 9.8. Type I (001) surface bands with U = 0. (a) Pi(yz) surface band for κ =
+1, −2, and −4. For negative values of κ surface bands lie in the fundamental gap and
below the valence band. For positive values of κ, surface bands lie above the top of the
conduction band and above the valence band. (b) Band-gap surface bands for the type II
(001) surface for different values of the perturbed p-orbital site potential.
that the calculated surface bands for the metallic perovskites will lie deeper in the
band gap than for the insulating perovskites. For example, referring to Figs 9.7(a)
and (b), the positions of the surface bands for u = 0 would apply to a metallic
perovskite while the surface bands pushed up to the edge of the conduction band
would apply for n-type doped insulating perovskites. One would expect the metallic
behavior to dominate for electron concentrations greater than or about 1021 cm−3
and insulator behavior for electron concentrations less than or about 1018 cm−3 .
(b) Experimental results: SrTiO3 , TiO2 , and Nax WO3
The range of concentrations of electrons in the surface bands displayed in Table 9.1
for n-doped perovskites should be detectable in photoemission experiments.
Angle-resolved photoemission studies have been carried out for several
Nax WO3 samples. Surface bands in the band gap were not found for insulating
WO3 but a surface band was reported [6] for metallic Na0.85 WO3 . In this latter
case, the surface band has a band width of about 0.9 eV and dispersion similar to
the surface band shown in Fig. 9.7(a) with U = 0. However, UPS and XPS pho-
toemission experiments [7–9] performed on SrTiO3 and on the closely related oxide
TiO2 as well as WO3 , have not detected the presence of any “intrinsic” surface
states in the band-gap region. The Fermi energy appears to be pinned at or near
the conduction-band edge. Whether this is pinning by the bulk conduction band or
a nearby surface band is not known, but in either case the concentration of electrons
must be below the detection threshold of about 1013 cm−2 for the photoemission
experiments. (For n-type materials with concentrations of the order of 1018 elec-
trons per cm−3 , the surface concentration in the absence of band-gap surface states
is only about 4×1010 cm−2 .)
Inverse photoelectron spectroscopy has also been used to search for unoccupied
surface states; however, these experiments probably did not have the energy reso-
lution required to separate bulk states from the unoccupied surface band near the
edge of the volume continuum. For example, the rise in the photoemission at the
conduction-band edge seen in the inverse photoelectron experiments is spread out
over an energy interval of 1.6 eV and the quoted energy resolution [7] was 0.7 eV.
Why “intrinsic” band-gap surface states are not observed for SrTiO3 , TiO2 ,
or WO3 but are observed for metallic Nax WO3 is unclear. As mentioned above,
surface charge may play a role for the doped insulators. For WO3 it has been
suggested that band bending due to surface charge depletes the surface bands of
electrons [6]. However, unless the charge is due to sources other than electrons
in the surface states, band-bending is already effectively included in the Coulomb
repulsion parameter, U .
Many conjectures can be put forth for why surface states are not seen in the
band gap of the n-doped insulators. First, there is the question as to whether
energy band theory can be applied to these materials since the correlation energy
among the surface electrons is large compared to the bulk. The mean-free path in
photoemission is less than 10 Å and therefore photoemission samples principally
the first few layers. Nevertheless, the bulk electronic structure predicted by band
theory is clearly evident. Thus it is difficult to argue that the band theory applies to
the bulk but not the surface. One may wonder if the LCAO model is too simplistic
to correctly describe the surface electronic structure. This is certainly a possibility,
particularly since the model employed here has only nearest-neighbor interactions.
However, the addition of more distant interactions will not change the qualitative
features of the surface bands. Furthermore LCAO models correctly describe most
features of the bulk electronic structure observed in photoemission and optical
experiments on insulating perovskites as well as the electronic dispersion observed
in high-temperature superconducting metal oxides [10, 11].
There is the possibility that the mean-field approximation used for the
electron–electron repulsion is inadequate to treat surface states. A dynamic the-
ory may be more aggressive and force the surface bands to within a few meV of
the bottom of the conduction band instead of a few hundredths of an eV. That
would reduce the electron concentration in the surface band to nearly that of the
bulk, a concentration that is below the threshold for detection in photoemission
experiments.
Fracturing the crystal in high-vacuum conditions would be expected to produce
both type I, the surface type II surfaces. Surface bands split off from the valence
bands on the type II surface are also predicted to lie in the fundamental band-gap
region but are also not seen experimentally for the n-doped insulators. Depending
upon the size of the various patches of type I and type II surfaces it is conceivable
that the long-range Madelung potentials of the two different surfaces cancel one
another approximately, leaving the surface with an average potential of the bulk.
In such a case surface bands would not occur.
In summary, the band-gap surface bands expected on the basis of the LCAO
model have been observed in metallic Nax WO3 but not for SrTiO3 , TiO2 , or WO3 .
The reason why surface bands are not detected for the doped insulators is not
certain. Stronger electron correlation at the surface or the approximate cancelation
of changes in the site potentials are possibilities.
9.4 Surface–oxygen defect states
Bombarding the surface of a perovskite with Ar+ (argon ions) results in the removal
of oxygen from the surface layer and, at high doses, from subsurface layers as well.
According to an ionic model each oxygen removed donates two electrons to the
material.
UPS and XPS measurements on SrTiO3 (also TiO2 and Ti2 O3 ) exhibit emis-
sion from oxygen defect states in the band-gap region, centered roughly 1.0–1.3 eV
below the conduction-band edge. The intensity of the emission increases with in-
creasing Ar+ dose until saturation occurs (sometimes accompanied by reconstruc-
tion or formation of a different phase on the surface). Since the emission from these
states is reduced rapidly by exposing the surface to oxygen, the states are believed
to result from oxygen vacancies. The wavefunctions for these defect states involve
the d orbitals on the B ions that are adjacent to one or more of the oxygen vacancies.
When an oxygen ion is removed from the surface of a perovskite several pertur-
bations occur: (1) electrons are released into the bulk conduction bands; (2) the two
B ions adjacent to the vacancy will have unsaturated bonds and lowered symmetry;
(3) the removal of the O2− ion makes the electrostatic potential at the two B-ion
sites much more attractive (negative); (4) but the occupation of the band-gap de-
fect states leads to a repulsive electron–electron potential between electrons in the
same orbital; (5) unless the vacancies form an ordered array, translation symmetry
is destroyed and the defect states are not characterized by a wavevector.
Theories of the surface defect states are beset with difficulties because the
actual surface geometry, surface defects, and surface potentials are not known. In
addition, the electron–electron correlation energy for surface d ions is much larger
than for an interior ions when defect states lie in the band gap.
Theoretical analysis of vacancy-induced states center around two different ap-
proaches. The first is an atomic-like picture. It is supposed that the electrons do-
nated by the removal of an oxygen ion are retained by the d orbitals of the B ions
adjacent to the vacancy site. This approach suggests the formation of different B-ion
oxidation states for the surface ions. For the oxides of titanium, SrTiO3 , TiO2 , and
Ti2 O3 for example, the model supposes that Ti3+ (or even Ti2+ ) ions are formed on
the surface with energies in the band-gap region. The bulk states are assumed to be
band states, but the defect states are assumed to have atomic-like wavefunctions.
This picture can be conceptually useful, but treating electrons occupying surface
states on a different footing from these occupying bulk states is difficult to justify
theoretically.
The second approach is band theory. In this model spatially localized vacancy-
induced states are derived from delocalized energy band wavefunctions. The oxygen-
donated electrons are not necessarily retained by the surface ions alone, but may
spread throughout the bulk conduction bands. The number of extra electrons re-
siding in surface defect states is determined by the energy of the defect state/band
relative to the Fermi energy. However, if the vacancy-induced state/band lies com-
pletely within the band gap it will be occupied in n-type, doped insulators, so that
the state may begin to resemble a Ti3+ ion state. In this case the localized, atomic
view and the band theory picture are conceptually similar.
With either the atomic approach or the band theory approach, the electron–
electron repulsion energy is large for the surface ions. The difference in ionization
energies of different oxidation states is large. For example, the difference in the
ionization energies of Ti4+ and Ti3+ is nearly 15 eV and for Ti4+ and Ti2+ it
is nearly 30 eV. Therefore, if the vacancy-induced state is to appear in the band
gap, there must be a correspondingly large decrease in the Madelung/electrostatic
potential for the surface sites. While it is easy to see that such a reduction is
likely to result when repulsive O2− ions are removed, calculating these long-range
electrostatic potentials requires detailed knowledge of the arrangement of the ions
and their charges. A large surface charge can result if all of the oxygen-donated
electrons reside on the surface ions. This charge can lead to band bending if it is
not canceled by other local charges. For SrTiO3 the excess charge may be partially
neutralized by surface and subsurface Sr2+ vacancies. For TiO2 it is suggested that
the Ti4+ ions are converted to Ti3+ ions by forming a surface layer of Ti2 O3 .
There is currently no reliable theory for accurately treating surface defect states
such as oxygen vacancy states on the surface of actual crystals. However, the quali-
tative features of oxygen vacancy states can be understood by studying some simple
LCAO energy band models. In this section we look at two such models that relate
to the oxygen vacancy states.
(a) A line of oxygen vacancies on a type I (001) surface
The planes containing dyz , py , and pz orbitals that are perpendicular to the sur-
face of a cubic perovskite are uncoupled in the nearest-neighbor approximation.
Therefore the surfaces states discussed in Section 9.2(d) are actually “edge” states
or “line” states, and that is why they behave as one-dimensional systems. Figures
9.9(a) and (b) show schematically the geometry of a yz-layer before and after re-
moving pz surface orbitals along a line in the y-direction. The resulting surface line
consists of only dyz orbitals on B ions. Since the layers are uncoupled we can assume
all the other yz-planes have their full compliment of surface pz orbitals. Therefore
the model is a single line of alternating B ions and surface oxygen vacancies on an
otherwise normal (001), type I surface. For the layer with the line of vacancies the
surface parameter, (pdπ)00 , is equal to zero. That is, ∆00 = −1 in (9.15b). Clearly,
the same model applies to the symmetry-equivalent xz-planes with a line of pz
vacancies along the x-direction.
y
dyz pz
z
py
(a) (b)
Figure 9.9. Schematic of a yz-layer (a) without vacancies and (b) with oxygen vacancies
extending along the y-direction.
Using the surface state condition, Λ = 0, of (9.17) and the definitions of the
surface perturbations of (9.15) with ∆0 = 0, ∆00 = −1, E⊥

0
= E⊥ gives:
1 − ∆p(ω, ky )eiθ = 0 (eigenvalue condition), (9.64)
where
−2 cos θ = (ωt − ω)(ω⊥ − ω) − 4Sy2 − 2 (9.65)

∆p(ω, ky ) = κ(ω⊥ − ω) + 1 + 4Sy2 , (9.66)
κ = ∆ωt + uNs .
For this model it is expected that the parameter, ∆ωt is much more negative than
in the case of the ideal surface because an entire row of repulsive, O2− ions has been
removed. Consequently, we would expect that the “line” energy band will lie deeper
in the gap than the pi(yz) surface states previously discussed (Table 9.1). Exactly
where the band lies depends upon the balance between the Coulomb repulsion and
increased negativity of ∆ωt .
The DOSS for the line energy band is given by (9.41).
³ 2a ´³ dΩ ´−1 1 1
ρvac (Ω) = = p . (9.67)
π dky π Ω(1 − Ω)
µ ¶
1 1
Ω(ω, ky ) = − η + , (9.68)
4 η+ξ
η = κ(ω⊥ − ω) + 1,
κ = (∆Et + U Nv )/(pdπ), (9.69)
ξ = (ωt − ω)(ω⊥ − ω) − 2. (9.70)
In (9.69) Nv is the number of electrons occupying the vacancy-induced surface

band. The square root singularities occur at Ω = 0 and Ω = 1, corresponding to the
bottom and top of the line band. The eigenvalue condition, ∆p + 1/∆p − 2 cos θ =
0, can be written as a cubic equation in the variable F = (ω⊥ − ω):
κF 3 + F 2 [κ(κ + ωg ) + (1 − Z)] + F [(κ + ωg )(1 − Z) − κ] + Z = 0, (9.71)
with Z = −4Sy2 . Solutions may be obtained using the standard formulae for the
roots of a cubic equation. The three vacancy-line bands are shown in Fig. 9.10. The
two vacancy bands, labeled V 1 and V 2 will be completely occupied with electrons,
but do not enter the fundamental band-gap region between Et and E⊥ .
The line band at –6 eV, labeled C1, has wavefunctions composed of nearly
pure d orbitals and is nearly dispersionless. This oxygen-vacancy-produced band
lies about 1 eV below the conduction band for κ = −1.3 eV and (pdπ) = 1 eV. Be-
cause the band lies totally within the band gap it would be completely occupied
by electrons in an n-type, doped insulator such as SrTiO3 . That is, Nv = 2 includ-
ing both spin states. Here we have a situation of an extremely narrow d-electron
Bulk continuum
–4
Et
C1
Energy (eV) –6
–8
E⊥ V1
Bulk continuum
–10
–12 V2
π
0 ky a 2
Figure 9.10. Surface bands due to a line of vacancies. The C1 band, derived from the
d orbitals centered on the surface B ions, lies in the band-gap region. V 1 and V 2 are
p-orbital bands.
band with large correlation energy. The surface B ions neighboring the vacancies
are forming atomic-like states. Therefore, the first electron occupying the state en-
counters a Coulomb repulsion energy roughly equal to the Ti+4 → Ti+3 ionization
energy of about 15 eV. Placing a second electron in the state costs 30 eV relative to
occupying a bulk state. Therefore, the Coulomb repulsion parameter is U for the
first electron and 2U for the second electron. As a result the C1 band for double
electron occupation lies well above the conduction band. Consequently, the vacancy
line-band will have only a single electron per d orbital and the Ti ions are, roughly
speaking, Ti3+ ions while the bulk Ti ions are Ti4+ ions.
(b) Isolated vacancy states
The previous section dealing with a line of vacancies applies when the vacancy
concentration is high. At the other extreme, that of low concentration, the vacancies
(–1,–1) (–1,0) (–1,1)

- y
=
x Y
(0,–1) (0,0) (0,1)
oxygen
vacancy
(1,–1) (1,0) (1,1)
Figure 9.11. Schematic of an isolated vacancy on the xy-plane. For the pi(xy) band the
vacancy is represented by the absence of a px orbital in the unit cell at the origin. The unit
cells are indicated by the x–y coordinates in parentheses and their locations are specified
by two-dimensional vectors ρ ~j,m . The values of j and m are shown in the figure.
are non-interacting and may be considered as isolated. The model for an isolated
oxygen vacancy on a (001) type I surface is illustrated in Fig. 9.11. The surface
unit-cell layer, parallel to the xy-plane, is uncoupled from the other unit-cell layers
in the nearest-neighbor approximation. The orbitals for the pi(xy) states are shown
schematically. In this case the oxygen vacancy takes the form of a missing px orbital.
The B ions are located on the xy-plane by the set of two-dimensional vectors,
ρ
~j,m = 2a(j~ex + m~ey ), where m and j are integers. The px orbitals of the O ions
are located at ρ~j,m + a~ey and the py orbitals at ρ
~j,m + a~ex . If we assume the missing
px orbital is in the unit cell at the origin, then, the equations for cx , cy , and cxy ,
the amplitudes of the px , py , and dxy orbitals respectively, are
(ωt + κ − ω) cxy (~
ρ0,0 ) − cx (~
ρ0,−1 + a~ey ) + cy (~
ρ0,0 + a~ex )
−cy (~
ρ−1,0 + a~ex ) = 0, (9.72a)
(ωt + κ − ω) cxy (~
ρ0,1 ) + cx (~
ρ0,1 + a~ey ) + cy (~
ρ0,1 + a~ex )
−cy (~
ρ−1,1 + a~ex ) = 0, (9.72b)
(ω⊥ − ω) cy (~
ρ0,0 + a~ex ) + cxy (~
ρ0,0 ) − cxy (~
ρ1,0 ) = 0, (9.72c)
(ω⊥ − ω) cx (~
ρ0,1 + a~ey ) + cxy (~
ρ0,1 ) − cxy (~
ρ0,2 ) = 0, (9.72d)
(ω⊥ − ω) cy (~
ρ0,1 + a~ex ) + cxy (~
ρ0,1 ) − cxy (~
ρ1,1 ) = 0, (9.72e)
κ = (∆Et + U Nv )/(pdπ). (9.72f)
For j 6= 0, m 6= 0 or 1:
(ωt − ω) cxy (~
ρj,m ) + cx (~
ρj,m + a~ey ) − cx (~
ρj,m−1 + a~ey )
+cy (~
ρj,m + a~ex ) − cy (~
ρj−1,m + a~ex ) = 0, (9.73a)
(ω⊥ − ω) cx (~
ρj,m + a~ey ) + cxy (~
ρj,m ) − cxy (~
ρj,m+1 ) = 0, (9.73b)
(ω⊥ − ω) cy (~
ρj,m + a~ex ) + cxy (~
ρj,m ) − cxy (~
ρj+1,m ) = 0. (9.73c)
In (9.72f) ∆Et is the change in the Madelung potential and Nv is the number
of d electrons occupying the vacancy states. By direct substitution we can elim-
inate the p-orbital amplitudes and obtain equations involving only the d-orbital
amplitudes,
£ ¤
(ωt + κ − ω)(ω⊥ − ω) − 3 cxy (~
ρ0,0 ) + cxy (~
ρ0,−1 ) + cxy (~
ρ1,0 )
+cxy (~
ρ−1,0 ) = 0, (9.74a)
£ ¤
(ωt + κ − ω)(ω⊥ − ω) − 3 cxy (~
ρ0,1 ) + cxy (~
ρ0,2 ) + cxy (~
ρ1,1 )
+cxy (~
ρ−1,1 ) = 0. (9.74b)
For j 6= 0, m 6= 0 or 1:
£ ¤
(ωt − ω)(ω⊥ − ω) − 4 cxy (~
ρj,m ) + cxy (~
ρj+1,m ) + cxy (~
ρj−1,m )
+cxy (~
ρj,m+1 ) + cxy (~
ρj,m−1 ) = 0. (9.74c)
The equations represented by (9.74a)-(9.74c) can be written in matrix form:

~ xy = 0,
[Hd (ω) + ∆Hd ] C (9.75)
where Hd (ω) is the effective Hamiltonian describing the interactions between the
d orbitals for the unperturbed surface unit-cell layer, ∆Hd describes the perturba-
tions due to the vacancy, and C~ xy is a vector whose components are the d-orbital
amplitudes, cxy (~
ρj,m ).
Referring back to (9.74a) and (9.74b) we see that ∆Hd is a null matrix except
for a 2×2 block centered on the diagonal.
Ã !
∆H(0, 0) ∆H(0, 1)
∆Hd = , (9.76)
∆H(1, 0) ∆H(1, 1)
where
∆H(0, 0) = ∆H(1, 1) ≡ ∆h = κ(ω⊥ − ω) + 1, (9.77)

∆H(1, 0) = ∆H(0, 1) = −1. (9.78)
Equation (9.75) can be rewritten as

~ xy = 0,
Hd [I + Gε ∆Hd ] C (9.79)
where I is the unit matrix and Gε = Gε (ω) = [Hd ]−1 is the d-orbital lattice Green’s
function. The matrix elements of Gε (ω) are derived and their behavior discussed in
Appendix C.
The eigenvalues are given by the zeros of the determinant of (9.79),
det{Hd [I + Gε ∆Hd ]} = {det Hd }{det[I + Gε ∆Hd ]} = 0. (9.80)
Since the zeros of {det Hd } occur at the unperturbed energies it follows that the
perturbed energies are given by the zeros of the 2×2 determinant, det[I + Gε ∆Hd ].
Thus the energies of the vacancy states are given by
Ã !
1 + ∆h G(0) − G(1) ∆h G(1) − G(0)
det = 0, (9.81)
∆h G(1)∗ − G(0) 1 + ∆h G(0) − G(1)∗
where G(0) ≡ Gε (ρ, ρ) and G(1) ≡ Gε (~ρ, ρ
~ + a~ex ) = Gε (~
ρ, ρ
~ + a~ey ). For energies in
the band gap, G(0) and G(1) are real functions and G(1) = G(1)∗ . Equation (9.81)
yields two solutions for the vacancy states corresponding to symmetric and anti-
symmetric combinations of the d orbitals adjacent to the vacancy.
£ ¤
G(0) + G(1) (∆h − 1) + 1 = 0 (9.82)
and
£ ¤
G(0) − G(1) (∆h + 1) + 1 = 0 . (9.83)
In Appendix C it is shown that in the band gap

1 ³2´
G(0) = K (9.84)
πε ε
and
1h1 i
G(1) = − ε G(0) , (9.85)
2 2
where
1
ε= (ωt − ω)(ω⊥ − ω) − 2. (9.86)
2
So that (9.82) and (9.83) take the form
h ³ ε´ 1i
G(0) 1 − + (∆h − 1) + 1
2 4
h 1 ³ 2 ´³ ε ´ 1 i£ ¤
= K 1− + κ(ω⊥ − ω) + 1 = 0, (9.87)
πε ε 2 4
h ³ ε´ 1i
G(0) 1 + − (∆h + 1) + 1
2 4
h 1 ³ 2 ´³ ε ´ 1 i£ ¤
= K 1+ − κ(ω⊥ − ω) + 2 + 1 = 0, (9.88)
πε ε 2 4
where K is the complete elliptic integral of the first kind.
Figure 9.12 shows the solutions of (9.87) as a function of the perturbation,
Conduction band
–5.0
6
–6.0
Band gap at Γ
ω
–7.0
–8.0 ?
Valence band
0.0 –5.0 –10.0 –15.0

κ
Figure 9.12. Energy of the oxygen defect state as a function of the perturbation param-
eter, κ. The gray areas at the top and bottom indicate the bulk continuum of states.
κ = ∆ωt + uNd . The solutions of (9.87) and (9.88) are virtually identical.3 For zero
or positive values of κ there are no vacancy states in the band-gap region. For
negative κ the vacancy states move downward into the band gap with increasing
negative values of κ. The energy moves down rapidly at first then approaches the top
of the valence band asymptotically for very large negative values of κ. For an n-type
insulator such as SrTiO3 with (pdπ) = 1 eV, the abscissa of Fig. 9.12 corresponds
to energy in eV. As mentioned earlier, vacancy states appear about 1 eV below
the conduction band. In Fig. 9.12 a vacancy state 1 eV below the conduction band
corresponds to κ of about −2.05 eV. As in the case of the vacancy line band, Nv
will be equal to 1 per unit cell since the doubly occupied state will lie above the
conduction-band edge and the Fermi level. For SrTiO3 each of the pair of Ti ions
adjacent to the vacancy will correspond approximately to a Ti3+ ion while the
interior ions are approximately Ti4+ .
The vacancy states provide coordinatively unsaturated bonds that are chem-
ically active. These states provide d orbitals that can act as a source or sink for
electrons to catalyze surface chemical reactions. A more complete discussion of the
surface reactive properties can be found in [5].
3
If the second-neighbor oxygen–oxygen interactions are included the two solutions will be slightly
different, corresponding to symmetric and antisymmetric combinations.
References
[1] V. E. Henrich and P. A. Cox, The surface science of metal oxides (Cambridge,
Cambridge University Press, 1996) p. 1.
[2] N. Bickel, G. Schmidt, K. Heinz, and K. Müller, Vacuum 41, 46 (1990).
[3] T. Hikita, T. Hanada, M. Kudo, and M. Kawai, Surf. Sci. 287/288, 377 (1993).
[5] T. Wolfram and S. Ellialtioglu, Concepts of surface states and chemisorption
on d -band perovskites. In Theory of chemisorption, ed. J. R. Smith, Topics in
current physics, Vol. 19, Ch. 6 (Heidelberg, Springer-Verlag, 1980) pp. 149-181.
[6] H. Höchst, R. D. Bringans, and H. R. Shanks, Phys. Rev. B 26, 1702 (1982).
[7] B. Reihl, J. G. Bednorz, K. A. Müller, Y. Jugnet, G. Landgren, and J. F. Morar,
Phys. Rev. B 30, 803 (1984).
[8] V. E. Henrich, G. Dresselhaus, and H. J. Zeiger, Phys Rev. B 17, 4908 (1978).
[9] R. A. Powell and W. E. Spicer, Phys. Rev. B 13, 2601 (1976).
[10] D. H. Lu, D. L. Feng, N. P. Armitage, K. M. Shen, A. Damascelli, C. Kim, F.
Ronning, Z.-X. Shen, D. A. Bonn, R. Liang, W. N. Hardy, A. I. Rykov, and S.
Tajima, Phys. Rev. Lett. 86, 4370 (2001).
[11] M. C. Schabel, C.-H Park, A. Matsuura, Z.-H. Shen, D. A. Bonn, R. Liang, and
W. N. Hardy, Phys. Rev. B 57, 6090 (1998).
1. For the pi(xy) surface bands with the only non-zero perturbation, ∆00 = −1, (a) find
the eigenvalue equation for the two surface bands. (b) Give analytic expressions for
the eiqenvalues. (c) Show that the two surface bands are not truncated, that is, those
surface bands that exist for all values of ky .
2. Make graphs of the pi(xy) bulk band edge (ky = 0) and the two surface bands of
Problem 1 for ωt = – 5 and ωt = – 8.
3. Using the eigenvalue expressions of Problem 1, find an analytical expression for the
DOSS, ρs (ω), for the surface bands and show that ρs (ω) has square-root singularities
at the top and bottom of both surface bands.
4. Make a graph of ρs (ω) in Problem 3 for ωt = – 5 and ω⊥ = – 8.
5. If the Fermi energy corresponds to ωF = – 4.9, find the number of electrons per spin
occupying the surface bands in Problem 4. (Hint: use ρs (Ω) rather than ρs (ω) for the
calculation.) If U = 6 eV, what is the surface Coulomb repulsion potential for electrons
occupying the surface band?
10
Distorted perovskites
The majority of perovskites are not cubic, but many of the non-cubic structures
can be derived from the cubic (aristotype) structure by small changes in the ion
positions. Several types of distortions occur among the perovskites. The most im-
portant types are those involving (a) ion displacements in which, for example, the
B ion or A ion (or both) moves off its site of symmetry; (b) rotations or tilting of
the BO6 octahedra; and (c) both tilting and displacements.
Departure from the cubic perovskite structure will occur whenever distortions
lead to a lower total energy. The lowering of the total energy in most cases is small,
typically of the order of a tenth of an eV/cell and dependent upon the tempera-
ture. The additional stabilization energy of one structure over another depends in
a subtle manner upon the competition between a number of electronic factors in-
cluding changes in the Coulomb interactions (Madelung potentials), changes in the
degree of covalent bonding, and the number of electrons occupying the antibonding
conduction bands. In many cases changes in the A–O covalent bonding is thought
to play a key role in determining the distorted structure.
Clearly it is not possible to predict structures based on the simple LCAO model
we have been studying. However, given a particular structure that is close to the
cubic structure one can examine the changes in the electronic states with the goal
of understanding why the distorted structure is more stable. In addition, there are
important new electronic features that result from small changes in the structure
that can be explored with the simple LCAO model.
10.1 Displacive distortions: cubic-to-tetragonal

phase transition
A number of perovskites that have the ideal cubic structure undergo a cubic-to-
tetragonal phase transition as the temperature is lowered. BaTiO3 and SrTiO3 , for
example, are cubic at temperatures above their transition (or Curie) temperatures
231
232 Distorted perovskites
TC = 408 and 378 K, respectively, but are tetragonal below TC .1 For BaTiO3 the
tetragonal structure is achieved by what is called a “displacive transition”. The B
ions and A ions are displaced from their cubic positions in one direction and the
oxygen ions in the other direction as shown in Fig. 10.1. In this tetragonal phase
BaTiO3 is ferroelectric. The situation for SrTiO3 is different. Its transition involves
the rotation of alternate octahedra in opposite directions as well as displacements
that result in a tetragonal state that is antiferroelectric.
Figure 10.1. Tetragonal displacements: the displacements of the ions for the cubic-to-
tetragonal phase transition. The A and B ions move upward along the z-direction while
the oxygens move down. The cubic symmetry changes to tetragonal as a result of the
displacements of the ions.
The spontaneous displacements (or octahedral rotations) are usually associated

with the existence of a “soft” optical phonon. That is, a particular lattice vibra-
tion belonging to an optical phonon branch whose frequency tends to zero at TC .
As the phonon frequency decreases the vibrational motions increase and become
anharmonic. Conceptually, one can imagine the vibrational motions of the ions in
the soft mode becoming “frozen in” as T approaches TC .
The magnitudes of the displacements in a displacive transition can be different
for the different ions. For BaTiO3 the magnitude of Ti-ion displacement is larger
than that of the A ion and O ions [1]. Also, the Oz displacement is larger than that
of the Ox and Oy ions so that the oxygen octahedron becomes slightly elongated
in the z-direction. The displacements reduce the symmetry to tetragonal, but the
departure from cubic symmetry is small.
1
Additional structural transitions can occur at even lower temperatures. For example, BaTiO3 is tetrag-
onal in the range 285 < T < 400 K, but undergoes a tetragonal-to-orthorhombic transition at 285 K.
10.1 Displacive distortions: cubic-to-tetragonal phase transition 233
For a displacive transition the B ion of the ABO3 structure is no longer at a site
of inversion symmetry. As a result the electronic charge distribution is asymmetric
and hence there is an electric dipole associated with each unit cell. In the ferroelec-
tric state applying an external electric field can orient all the dipoles. The field of
the dipoles produces long-range effects which, in fact, depend upon the macroscopic
shape of the sample (or domains that form within the sample). The polarizability of
the dipoles leads to a low-frequency dielectric function that is extremely large and
temperature sensitive, particularly as the temperature approaches TC . For BaTiO3
the dielectric “constant” ranges from 1200 to 1600 at a frequency of 1 kHz. By con-
trast, in Chapter 6, we found ε2 (ω) ≈ 6 at the band-gap energy where ω ≈ 5 × 1012
kHz. Understanding the low-frequency behavior of the dielectric properties of the
perovskites requires consideration of the lattice dynamics and polarizability of the
dipoles and is not described by conventional band theory.
Many of the ferroelectric perovskites are also piezoelectric. If electrodes are at-
tached to opposite faces of such crystals and a voltage applied, the electric field will
induce dimensional changes. Conversely, application of an axial compressive force
can produce a voltage on the electrodes. Because of these properties piezoelec-
tric perovskites are used in electronic devices such as transducers and electrooptic
modulators, and for the fine control of scanning tunneling microscope (STM) tip
motion.
In Fig. 10.1 it can be seen that the B ion moves up along the z-axis toward one
of the neighboring Oz ions and away from the other. As a result the (pdσ) and (pdπ)
interactions are increased for one of the B–O bonds and decreased for the other.
Also, the line joining the Ox (or Oy ) ions to the B ion is no longer perpendicular to
the z-axis. Consequently, the wavefunctions at Γ, which are pure d orbital or pure
p orbital in the cubic case, will have a small p–d mixing in the tetragonal phase.
Another related effect is the splitting of the energy band degeneracies at Γ. The
tetragonal displacement splits the t2g states into a doubly degenerate group, dxz
and dyz , and a non-degenerate dxy state. Similarly, the doubly degenerate σ ∗ states
are split into non-degenerate dx2 and dz2 states. This splitting, however, is not the
same as the Jahn–Teller splitting that occurs in molecules.
For perovskites that undergo cubic-to-tetragonal phase transitions the displace-
ments of the ions are small, usually less than a few percent of the lattice constant.
For example, the calculated displacements for the Ba, Ti, Ox , Oy , and Oz ions
in BaTiO3 are 0.012, 0.039, 0.014, 0.014 and 0.025 Å, respectively [1]. The exper-
imentally observed c/2a ratio is 1.0086, indicating that the departure from cubic
symmetry is very small [2]. The displacements are larger for PbTiO3 where Ti-
ion displacement is calculated to be 0.1006 Å and the experimental c/2a ratio is
1.0649. The stabilizing energy per unit cell for the tetragonal phase is calculated to
be about – 0.4 meV for BaTiO3 and – 40 meV for PbTiO3 [1].
To explore the effects of the displacements on the energy levels we can specify
the ion locations for the tetragonal structure as:
B ions at : 2a(nx , ny , nz ),
h³ 1´ ³ 1´ ³ 1 í
A ions at : 2a nx + , ny + , nz + − δA ,
2 2 2
h³ 1´ i
Ox ions at : 2a nx + , ny , (nz − δOx ) , (10.1)
2
h ³ 1´ i
Oy ions at : 2a nx , ny + , (nz − δOy ) ,
2
h ³ 1 í
Oz ions at : 2a nx , ny , nz + − δOz ,
2
where nx , ny , and nz are positive or negative integers and
δA = (dB − dA )/a,
δOx = δOy = (dB − dOx )/a, (10.2)
δOz = (dB − dOz )/a.
Here, dB , dA , dOx , and dOz are the displacements of the B, A, Ox , and Oz ions,
respectively, and a is the B–O distance in the cubic phase. Equation (10.1) can be
understood in the following way. Each type of ion can be assigned to a simple cubic
sublattice, but not all of the atoms are on the lattice points. We choose a B ion as
the origin, and then displace each of the sublattices relative to the B sublattice.
The Hamiltonian, H, for the cubic perovskite is given in Table 4.1. The changes
in the matrix elements at Γ to first order in the displacements are:
∆H(1, 2) = 2∆σ ≡ d
√
∆H(1, 11) = ∆H(1, 14) = δOx [(pdσ) − 2 3(pdπ)] ≡ e
√
∆H(3, 11) = −∆H(3, 14) = −δOx [ 3(pdσ) − 2(pdπ)] ≡ f
√
∆H(9, 4) = ∆H(5, 12) = −2δOx [ 3(pdσ) − (pdπ)] ≡ g (10.3)
∆H(9, 7) = ∆H(8, 12) = −2δOx (pdπ) ≡ h
∆H(9, 10) = ∆H(12, 13) = 2∆π ≡ u
∆H(I, J) = ∆H(J, I).
In equation (10.3) ∆σ ∝ δOz is the increase (decrease) in (pdσ) along the positive
(negative) z-axis. Similarly, ∆π ∝ δOx is the increase (decrease) in (pdπ) along the
positive (negative) z-axis. We have neglected the tetragonal perturbation on the
oxygen–oxygen interaction in (10.3).
By rearranging rows and columns (H + ∆H) can be reduced to block-diagonal
form. There is a 5×5 block (rows/columns 1, 2, 3, 11, 14), a 1×1 (row/column 6)
and two symmetry-equivalent 4×4 blocks (rows/columns 9, 4, 7, 10 and 12, 5, 8,
13).
0
We label the energies as in Table 5.1. The “primed” energies of the form, EΓn ,
indicate energies for the tetragonal phase and those without a superscript “prime”
are for the cubic phase.
The block-diagonalized secular equation, (H + ∆H − EΓ0 ) takes the forms
shown below.
(a) 1×1; row/column 6. This solution is the π ∗ (xy) band edge.
0
EΓ3 = EΓ3 = Et (unshifted eigenvalue) . (10.4a)
∗
(b) 4×4; rows/columns 9, 4, 7, 10. Solutions yield the shifted π (xz) band
edge and shifted valence-band energies.
Orbital 9 4 7 10
9 Et − EΓ0 g h u
4 g Ek − EΓ0 2c 2c (10.4b).
7 h 2c E⊥ − EΓ0 p
10 u 2c p E⊥ − EΓ0
(c) 4×4, rows/columns 12, 5, 8, 13. Solutions yield the shifted π ∗ (yz) band
edge and shifted valence-band energies. These solutions are symmetry-equivalent
to the solutions of (10.4b) (10.4c)
(d) 5×5, rows/columns 1, 2, 3, 11, 14. Solutions yield the shifted σ ∗ band
edges and shifted valence-band energies:
Orbital 1 2 3 11 14
1 Ee − EΓ0 d 0 e e
2 d Ek − EΓ0 0 2c 2c
(10.4d)
3 0 0 Ee − EΓ0 f −f
11 e 2c f E⊥ − EΓ0 p
14 e 2c −f p E⊥ − EΓ0
The matrix elements d, e, f, g, h, and u are defined in (10.3). The elements,

p ≡ 4(ppπ) and c ≡ (ppσ) + (ppπ) are the oxygen–oxygen interactions for the cubic
phase.
0
The solutions of det(H + ∆H − EΓn ) = 0 involve first-order changes in the
eigenvectors and second-order changes in their eigenvalues from their cubic values.
As an example we consider the case of BaTiO3 for which δOx= 0.027, δOz=
0.033. The LCAO parameters for BaTiO3 are (in eV):
Et = –6.70 E⊥ = –10.00 (pdπ) = 1.00 (ppπ) = 0.05 ∆π = 0.038

Ee = –5.40 Ek = –10.50 (pdσ) = –2.00 (ppσ) = –0.15 ∆σ = –0.076.
To estimate the change in (pdπ) and (pdσ) integrals (∆π and ∆σ ) we made use of
hydrogenic orbitals to calculate the change in the overlap due to the displacements
and then scaled LCAO parameters accordingly.
(a) Pi-like tetragonal states
The solutions of the 4×4 secular equation are given in Table 10.1. The splittings
caused by the tetragonal distortion are shown schematically in Fig. 10.2. An im-
portant result in Table 10.1 is that to second order in the displacements there is
no effect on the EΓ3 (the band edge for π ∗ (xy) at Γ). This result can be seen by
inspection since the angular integral of dxy with any pair of neighboring p orbitals
except px (~r − a~ey ) and py (~r − a~ex ) vanishes by symmetry independent of the dis-
placement along the z-direction. Furthermore, the change in the interaction with
the px (~r − a~ey ) and py (~r − a~ex ) orbitals is second order in the displacements and
hence higher order in the perturbed energy.
Table 10.1. Pi-like states at Γ. (Here µ = x, y, z and ν = y, z.)
Band at Γ Orbitals Orbitals Energy ∆EΓn

(Table 5.1 involved involved EΓn (eV) shift
notation) Cubic case Tetragonal case Cubic case (meV)
0
EΓ3 (π ∗ ) dxy dxy –6.70 0.00
0
EΓ4 (π ∗ ) dxz dxz , px (~r − a~eµ ) –6.70 17.63
0
EΓ9 (π 0 ) px (~r − a~eν ) dxz , px (~r − a~eµ ) –10.20 –2.33
0
EΓ10 (π + σ) px (~r − a~eµ ) dxz , px (~r − a~eµ ) –9.70 –1.40
0
EΓ11 (π + σ) px (~r − a~eµ ) dxz , px (~r − a~eµ ) –10.60 –13.90
The perturbed wavefunction for the π ∗ (xz) state, EΓ4 0

, is an admixture of
(mostly) dxz with a small amount of px (~r − ~eµ ) orbitals (µ = x, y, and z). Symmetry
equivalence tells us the same energy shift occurs for the dyz state and mixing occurs
between the dyz and the oxygen py (~r − ~eµ ) orbitals (µ = x, y, and z).
Table 10.1 shows that the triple degeneracy of the π ∗ band is split into
0
an unshifted singlet EΓ3 (π ∗ (xy)) and a doublet EΓ4
0
(π ∗ (xz)) and the symmetry-
0 ∗ ∗ ∗
equivalent EΓ5 (π (yz)). The π (xz) and π (yz) band edges are shifted upward in
energy as shown schematically in Fig. 10.2. The shift increases quadratically with
the relative displacements of the oxygen ions.
Γ4t (xz), Γ5t (yz)
Γ3 (xy), Γ4 (xz), Γ5 (yz) Γ3t (xy)
Γ10 , Γ13
Γ10t , Γ13t
Γ9 , Γ12
Γ9t , Γ12t
Γ11 , Γ14
Γ11t , Γ14t
Figure 10.2. Tetragonal splitting of the pi-like states at Γ: the labeling of the levels is
according to the convention of Table 5.1. The subscript ‘t’ refers to the tetragonal phase.
The threefold degeneracy of the π ∗ band edge is split into a doubly degenerate and a non-
degenerate level. The twofold degenerate σ ∗ band edge is split into two non-degenerate
levels.
For insulating BaTiO3 (with empty π ∗ bands) one can see that the energy of
the system is lowered because all of the filled valence states are lowered in energy. To
calculate the actual stabilization energy we would need the perturbed energies over
the entire Brillouin zone. The results at Γ are only suggestive and likely overesti-
mate the effect since the perturbations vary as cos(kα a) and are therefore maximal
at Γ. Nevertheless, it is interesting to look at the contributions from the states
at Γ. The stabilization energy due to these occupied valence states amounts to
−35.26 meV/cell (2×(– 13.90 – 1.40 – 2.33) meV/cell) and results from the increased
Ti–O bonding. From Table 10.1 it is also obvious that the stabilizing energy will
decrease if electrons are added to the π ∗ bands.
The change in the degeneracy of the π ∗ band edge at the phase transition can
produce important effects. For example, a lightly n-doped material may have its
Fermi energy within a few meV of the bottom of the π ∗ band. A concentration
of 1020 electrons per cubic centimeter (6.4×10−3 electrons per unit cell) produces
a Fermi energy about 5.6 meV above the bottom of the π ∗ band edge at Γ (see
Subsection 6.4(a)). In the cubic phase the electrons reside equally at the bottom of
the three degenerate π ∗ bands. During the cubic-to-tetragonal transition the Fermi
energy, EF , will change abruptly because the degeneracy of the band edge changes
from 3 (above TC ) to 1 (below TC ). Below TC all of the electrons must now reside in
the π ∗ (xy) band. To achieve the same number of occupied states the Fermi energy
0
must change as shown schematically in Fig. 10.3. In addition, if EF < Et + ∆EΓ4 ,
0 ∗ 0 ∗
the Fermi level will not intersect the EΓ4 , (π (xz)) or the EΓ5 , (π (yz)) bands so
the Fermi surface loses two of its arms. The three-dimensional “jack” described in
Chapter 6 and shown in Fig. 10.3(a) degenerates into the surface of a single circular
rod oriented along the√kz -axis as shown in Fig. 10.3(b). The radius of the rod must
increase by a factor of 3 to accommodate the same number of electrons. The above
comments would be strictly true at low temperature. However, the thermal energy,
kB T , is about 35 meV at 400 K and 25 meV at 285 K so the thermal energy is larger
than the splitting throughout the temperature range of the tetragonal phase. Thus,
Figure 10.3. Splitting and the Fermi surface. (a) Above the Curie temperature, TC . The
three π ∗ bands are degenerate and the Fermi surface is the “jack” described in Chapter
6. (b) Below the Curie temperature. The π ∗ (xz) and π ∗ (yz) energies are raised while the
π ∗ (xy) is unshifted. Abrupt changes occur in the Fermi energy and the Fermi surface at
TC that lead to abrupt changes in the electron transport properties.
0 0
one would expect thermally excited electrons in the EΓ4 , EΓ5 bands and holes in
0
the EΓ3 band.
Nevertheless the abrupt change in the Fermi energy and the sudden appearance
of electric dipoles will produce abrupt changes in the electron transport properties.2
For example, one might expect to see the sudden appearance of anisotropy in the
conductivity or in the Hall effect. In fact, such effects are seen experimentally in
samples of BaTiO3 . Measurements on single-crystal BaTiO3 [3] show a step-like
increase of the resistivity by a factor of 2 at TC as the crystal passes from the cubic
to the tetragonal phase. In addition, measurements of the Hall-effect mobility, µ,
indicate µa /µc & 10 for electrons [4, 5] in the tetragonal phase.
(b) Sigma-like tetragonal states
The secular equation resulting from the 5×5 matrix equation can be condensed into
the following form:
h in o
ee E
E e − − 2f 2 E ee (E
ek E
e + − 8c2 ) − d2 E
e + − 2e2 E
ek + 8cde = 0, (10.5)
⊥ ⊥ ⊥
e ± = E⊥ ± 4(ppπ) − E 0 ,
where E⊥ Γ
ee = Ee − E 0 ,
E Γ
ek = Ek − E 0 .
E Γ
The first factor in brackets yields

rh i2
0 1 − 1 −
EΓ2,6 = (Ee + E⊥ )± (Ee − E⊥ ) + 2f 2 . (10.6)
2 2
The second factor is a cubic equation in EΓ0 . The solutions are given in Table 10.2.
0 0
The energies, EΓ1 and EΓ2 correlate with the eg d-orbital states, dz2 and dx2 ,
which are degenerate in the cubic case with energy Ee (equal to –5.40 eV in this
example). In the tetragonal phase they are split with the dz2 raised in energy
by 13.64 meV and the dx2 by 9.05 meV. All of the valence states are lowered in
energy. The contribution to the stabilization energy from these valence states at
Γ is –22.69 meV, an amount that is about two-thirds of the contribution of the
pi-like states. The total stabilization energy from all the valence states at Γ is
–57.95 meV/cell.
The energy shifts due to the tetragonal distortion are a result of increased B–
O covalent bonding (and antibonding) They are not analogous to the Jahn–Teller
effect that occurs in molecules. Here both the t2g and eg antibonding states are
2
The energy band description for electron transport in BaTiO3 is appropriate as recent experiments
have shown [3]. Older results have been interpreted in terms of transport by small polarons [6].
Table 10.2. Sigma-like states at Γ. (Here µ = x, y, z, and ν = x, y.)
Band at Γ Orbitals Orbitals Energy ∆EΓn

(Table 5.1 involved involved EΓn (eV) shift
notation) Cubic case Tetragonal case Cubic case (meV)
0
EΓ1 (σ ∗ ) dz2 dz2 , pz (~r − a~eµ ) –5.40 +13.64
0
EΓ2 (σ ∗ ) dx2 dx2 , pz (~r − a~eν ) –5.40 +9.05
0
EΓ6 (π 0 ) pz (~r − a~eν ) dx2 , pz (~r − a~eν ) –10.20 –9.05
0
EΓ7 (π + σ) pz (~r − a~eµ ) dz2 , pz (~r − a~eµ ) –9.70 –4.90
0
EΓ8 (π + σ) pz (~r − a~eµ ) dz2 , pz (~r − a~eµ ) –10.60 –8.74
pushed up in energy whereas for Jahn–Teller splitting generally one of the d-orbital
levels is lowered and other is raised in energy.
While it is not expected that the LCAO model can yield accurate results for the
stabilization energy, it is expected that the results reflect the dominant energy-band
mechanisms that act to stabilize the distorted structure.
10.2 Octahedral tilting
(a) Classification of tilting systems and space groups
Many of the cubic perovskites distort by tilting their oxygen octahedra. Phase tran-
sitions involving octahedral tilting alter the symmetry and can result in tetragonal,
orthorhombic, rhombohedral, or monoclinic structures. The occurrence of a large
number of different perovskites is often attributed to the fact that the octahedral
structure can tilt to accommodate metal cations of widely varying sizes.
A system for classifying the tilt structures and determining their space-group
symmetries has been developed by Glazer [7] and extended by others [8]. To under-
stand this system we need to look at the aristotype (cubic) structure as an array
of oxygen octahedra as depicted in Fig. 10.4(a). The oxygen ions are located at the
corners of the octahedra. The B ions are located at the center of the octahedra and
the A ions nestle in the spaces (interstices) between the octahedra. The tilting of
an octahedron in a particular layer can be defined by specifying the angle of tilt
(rotation) about each of the three pseudocubic axes. That is, the cubic axes before
tilting has occurred. Since each oxygen (corner) is shared by two octahedra, it is
clear that the tilting of one octahedron will require tilting of all the octahedra in a
given plane. For example, rotating one of the octahedra about the z-axis will cause
its neighbors to rotate oppositely, leading to an entire layer in which alternate oc-
Figure 10.4. Octahedral tilting: ABO3 cubic perovskite showing the oxygen octahedra.
The oxygen ions are at the corners of the octahedra. The B ions are at the center of the
octahedra and the A ions reside in the interstices between the octahedra. (a) Untilted,
cubic case. (b) View looking down the ~c-axis of the a0 a0 c+ tilt system. The sense of
rotation about the ~c-axis is the same in adjacent layers. (c) View looking down the ~c-axis
of the a0 a0 c− tilt system. The sense of rotation about the ~c-axis is opposite in adjacent
layers.
tahedra are rotated in opposite senses. The tilts in the adjacent layers must also be
specified. The octahedra in the next layer may be rotated in the same sense or in
the opposite sense.
In Glazer’s system the octahedra themselves are assumed to remain regular
and tilts of a particular octahedron are described by specifying three symbols, ~a,
~b, ~c which relate to the magnitude of the rotation about the ~a, ~b, and ~c (x, y, and
z) axes in a given layer. A repeated symbol indicates that the tilts are equal. Thus
“aac” would indicate equal rotations about the ~a and ~b axes and a different rotation
about the ~c-axis. To specify the sense of rotation in the adjacent layer each of the
symbols is assigned a superscript, “+”, “–”, or “0”. If there is no rotation about a

given axis the superscript 0 is used. A “+” superscript indicates rotation in the same
sense in the adjacent layer and a “–” superscript indicates rotation in the opposite
sense. With this system the cubic structure is represented by “a0 a0 a0 ”, indicating
no rotation relative to ~a, ~b, or ~c in the layer and none in the adjacent layer. Consider
the triad, a+ b− c− . The first symbol, a+ , indicates a rotation about the ~a-axis in
the first layer and the same rotation in the second layer. The second symbol, b− ,
indicates a different amount of rotation about the ~b-axis and the opposite rotation
in the adjacent layer. The third symbol, c− , indicates a third amount of rotation
about the ~c-axis and the opposite amount in the adjacent layer. The tilt systems
a0 a0 c+ and a0 a0 c− are shown in Figs 10.4(b) and (c), respectively.
Glazer’s system has proven very useful for characterizing the space-group sym-
metry of distorted perovskites in a compact and convenient way. One should note,
however, that it is assumed that the repeat of the interlayer pattern is no more
than two layers of octahedra and the octahedra are assumed to remain (nearly)
undistorted, or “regular”. Furthermore, displacements of the ions are not described
by Glazer’s notation.
Others [8] have extended the system to include displacements of the B cation
by adding a subscript to each of Glazer’s three symbols. According to this notation,
a00 a00 c0+ is an untilted tetragonal system (because all the superscripts are zero) with
displacement of the B ion along the positive ~c-axis (+ subscript). This notation
would apply, for example, to the displacive phase transition we discussed for BaTiO3
in Section 10.1. SrTiO3 in its antiferroelectric tetragonal phase would correspond
to a00 a00 c−+ . This notation describes a crystal with tetragonal distortions consisting
of opposite tilting about the ~c-axis in adjacent layers and a B ion displaced along
the positive ~c-axis.
Glazer’s system leads to 10 distinct tilt systems and 23 space groups. The
extended notation, including displacements, yields 61 distinct space groups. Crys-
tals representing 15 of Glazer’s 23 tilt systems have been found experimentally.
The most common system for the perovskites appears to be a− b+ a− (55%) while
cubic crystals with a0 a0 a0 account for about 10% of the identified structures [9].
Of course, the actual structure depends upon the temperature. It is not unusual
for a perovskite to go through several different octahedral-tilting transitions with
increasing temperature, ending up in the cubic phase at the highest temperature.
NaNbO3 , for example, undergoes seven transitions with increasing temperature, fi-
nally taking on the cubic structure above 923 K. A few examples of tilting systems
are shown in Table 10.3.
Table 10.3. Examples of tilt systems. (TC =110 K for SrTiO3 .)
a0 a0 a0 a0 a0 c− a− b+ a− a− a− a−
SrTiO3 (T > TC ) SrTiO3 (T < TC ) CaTiO3 LiNbO3
ReO3 NaTaO3 SrZrO3 NdAlO3
NaWO3 BaTiO3 YCoO3 LaCoO3
(b) Causes of octahedral tilting
Perovskites will have a tilted structure whenever tilting lowers the total energy of
the system. Understanding exactly how the energy is reduced is not a simple matter.
The physical and chemical forces (energies) that come into play are large in number
and often competitive in action. A delicate balance between these various energies
determines the structure with the lowest energy. This balance of energies changes
with temperature, indicating that we are dealing with stabilization energies of a
few tens of meV/cell at most.
The physics and chemistry of octahedral tilting in perovskites is a topic receiv-
ing a great deal of attention. Woodward [10] has given an overview of the factors
that influence the tilt systems. Approaches to understanding the phenomenon are
varied and include: (1) empirical correlations focused on ion sizes, coordination
spheres, and various geometrical considerations; (2) empirical model calculations of
total lattice energy; and (3) energy band calculations using various approximations.
Each of these different approaches has its strengths and weaknesses and each offers
a different perspective on the problem.
(c) Geometric considerations
From a geometrical point of view, the largest effect of octahedral tilting is the
change in the immediate environment of the A ion. Tilting can substantially al-
ter the A–O distances and the coordination spheres of the A ion, while the B-ion
coordination sphere remains roughly octahedral. Figure 10.5 shows the oxygen co-
ordination spheres for the cubic, a0 a0 a0 and the a0 b− b− tilt systems. For the ideal
cubic system the A ion is surrounded by 12 equidistant oxygen ions. For the a0 b− b−
tilt system there are seven nearest and next nearest-neighbor oxygens (Fig. 10.5(b)).
The nearest five enclose the A ion in a pyramidal cage and the remaining two lie
further out. The nearest-neighbor A–O distances are less than for the cubic case,
suggesting increased orbital interaction between the A and O ions. Such interactions
tend to depress the energy of the non-bonding oxygen valence states and therefore
lower the total energy of the system if the antibonding states are unoccupied.
Figure 10.5. A-ion coordination spheres. (a) A cubic perovskite showing the 12 oxygen
ions that form the coordinations sphere of the A ion. (b) The first and second nearest-
neighbor oxygen ions for the a0 b− b− tilt structure. The first coordination sphere consists
of five oxygens that enclose A in a tetrahedral cage. The A–O distance for these nearest
neighbors is less than in the cubic case. Two oxygen ions at a greater distance constitute
the second nearest oxygens.
Some theories focus on the size of the A ion as the key factor in tilting. The
geometrical “fit” of the A ion in the structure is measured by the Goldschmidt
tolerance factor, “t”, which is defined in terms of the ionic radii RA , RB , and RO
of the constituent ions:
(RA + RO )
t= √ . (10.7)
2(RB + RO )
The ideal fit occurs for t = 1 and significant departures from unity suggest the
crystal will tilt to accommodate the mismatch. Most of the known perovskites have
0.78 < t < 1.05 [8]. Most cubic crystals are in the range 0.986 < t < 1.049, but many
tilted structures also lie in this range. The tolerance factor is suggestive of when
tilting might be expected but gives no information about what tilt system should
occur. Thomas [8] has proposed an empirical, geometric approach that yields a
predictive relationship between the tilt angles and the polyhedral volumes:
VA
= 6 cos2 θm cos θz − 1, (10.8)
VB
where VA is the volume of the A–O12 coordination polyhedron and VB is the volume
of the BO6 polyhedron, θm is the average of the tilt angles in the xy-plane and
θz is the rotation angle about the z-axis. According to equation (10.8) VA /VB is
maximized for untilted structures where VA /VB = 5, and decreases with increasing
tilt angles. For a number of perovskite structures good agreement is found between
the predicted volume ratio and measured tilt angles. For example, for NaNbO3
(293 < T < 773 K) the measured tilt angles are θm = 3.617◦ and θz = 4.209◦ and
(10.8) yields VA /VB = 4.964. The ratio calculated from the actual structure is 4.956.
Thomas defines the degree of tilt by the parameter Φ = 1 − cos2 θm cos θz , which is
zero for untilted systems. The experimental data points for most of the perovskites
examined lie on the straight-line plot of VA /VB versus Φ. Those data which do not
fall on the line (e.g., NaTaO3 ) correspond to structures believed to have significantly
distorted octahedra.
(d) Lattice energy calculations
Models that use empirically derived atomic potentials to represent Coulombic and
short-range forces have been employed to examine the stability of various tilting
systems [11, 12]. Woodward [10] used this approach to obtain results for idealized
YAlO3 subject to various normalizing constraints in order to compare the repulsive,
attractive, and total lattice energy of different tilt systems. The repulsive energy
is minimized by the cubic structure, but the attractive potential is maximized for
the a+ b− b− system. Table 10.4, second column, shows the stabilization energies
relative to the cubic structure for several of the tilt systems [10].
Table 10.4. Total energies of tilt systems relative to the cubic phase [10].
Tilt Empirical lattice Extended Hückel Extended Hückel

system energy YAlO3 (eV) YAlO3 (eV) AlO3 (eV)
a+ b− b− – 0.72 – 1.76 +0.08
− − −
a a a – 0.61 – 1.60 +0.07
0 − −
a b b – 0.55 – 1.62 +0.05
a0 a0 c− – 0.20 – 1.49 +0.19
a0 a0 c+ – 0.19 – 1.44 +0.19
0 0 0
a a a 0.00 0.00 0.00
The total lattice energy difference between different tilt systems ranges from
0.19 to 0.72 eV. The Al–O–Al angles were found to vary from 145.4◦ (for a0 b− b−
and a0 a0 c− ) to 180◦ for the cubic case. Woodward suggested that oversized A
cations (t > 1) are best accommodated by the cubic structure because that structure
minimizes the repulsive energy. When the A ion is small (t < 0.975) the tilt rotation
angle becomes large and the orthorhombic a+ b− b− tilt system is favored according
to Woodward [10].
(e) Electronic structure considerations
For many of the tilted structures the BO6 octahedron remains nearly undistorted.
The energy band structure will differ from the cubic structure for a number of
reasons: (1) the interaction of the A-ion orbitals with the oxygen orbitals may be
substantially increased and (2) the changes of the ionic site potentials shift the
diagonal energies, (3) the angles of the O–B–O and O–O interactions change, and
(4) splitting of degeneracies occur because of lower symmetry. All of these effects
change the energy band structure from that of the cubic crystal. Most of the changes
are minor in terms of the overall band structure but are of vital importance for
properties such as ferroelectricity or magnetic order, metal–insulator and structural
phase transitions. The decreased A–O distance can lead to admixing the A orbitals
into the wavefunctions for the p–d valence and conduction bands and the p–p non-
bonding bands. If the A-orbital state is at an energy higher than the d-orbital
diagonal energy, as is usually the case, the effect of A–O interactions is to lower
the energy of the valence bands and raise the energy of the A-orbital antibonding
bands. In extreme cases where the A orbitals are strongly interacting even in the
untilted phase, the basic structure of the energy bands is substantially altered. For
example, for BaBiO3 the 6s Bi orbitals hybridize with the oxygen sigma orbitals to
form an antibonding band that is partially occupied. Similarly, for PbTiO3 the Pb
6s and 6p orbitals are involved in the primary electronic structure. In these cases,
tilting further enhances the roles of the A-ion orbitals.
Calculations of the total energy using the extended Hückel band structure
model [10] have also been carried out for YAlO3 . The stabilization energy relative
to the cubic phase is shown in the third column of Table 10.4. The results are
quite different from those of the empirical potential calculations. In order to isolate
the contribution of the Y ion, similar calculations were carried out for AlO3− 3 . The
results are shown in the fourth column of Table 10.4. Omitting the Y (A ion)
resulted in much smaller differences in the energies of the various tilt systems.
More importantly, the results indicate cubic structure is the most stable system.
Comparing columns three and four of Table 10.4 indicates that the majority of
the tilt stabilization energy for YAlO3 is derived from the presence of the Y ion.
These results might suggest that tilting is mainly due to the A ion, but that is
certainly not true. On the contrary, the perovskite, WO3 , has no A-site ion yet
undergoes four phase transitions between 0 and 950 K. Furthermore, displacement
of the W ion from its center of symmetry appears to occur in all of the phases [13–
15]. Therefore, electronic effects alone are sufficient to cause tilting. For example,
theoretical energy band calculations using density functional theory [14] for WO3
as a function of electron doping produce all four of the tilting phases observed for
NaWO3 despite the fact than the A ion (Na) is omitted from the calculation. In
this case adding electrons to the conduction bands drives the phase transitions.
As the number of electrons is theoretically increased from zero to one, the charge
goes into the W 5d-like antibonding bands. This reduces the stabilization energy of
the filled valence bands and allows tilting structures to compete. Doping above 21
electron per unit cell leads to displacement of the W ion along the z-axis. The 5dxz
and 5dyz bands are then preferentially filled in a manner similar to that discussed
in Section 10.1 for tetragonal BaTiO3 .
In summary, octahedral tilting occurs in most perovskites. Tilting provides
the perovskite structure a way to accommodate a wide range of cation sizes
and to lower its total energy. Typical stabilization energies are small, usually in
the range 0.01–0.1 eV per unit cell. Despite the small energies involved, tilting
phase transitions can lead to dramatic changes in the physical properties of the
perovskite (e.g., ferroelectricity, magnetic order, superconductivity). Theoretical
prediction of which tilting system will occur at a given temperature is a formidable
task because there are a large number of different, competing mechanisms at
work. Calculations of energy differences must be accurate to better than 0.01 eV
to identify the lowest-energy tilt system with certainty.
References
[1] T. Nishimatsu, T. Hashimoto, H. Mizuseki, Y. Kawazoe, A. Sasaki, and Y.
Ikeda, http://xxx.lanl.gov/PS cache/cond-mat/pdf/0403/0403603.pdf
[2] Y. Kuroiwa, S. Aoyagi, A. Sawada, J. Harada, E. Nishibori, M. Takata, and
M. Sakata, Phys. Rev. Lett. 87, 217601 (2001).
[3] T. Kolodiazhnyi, A. Petric, M. Niewczas, C. Bridges, A. Safa-Sefat, and J. E.
Greedan, Phys. Rev. B 68, 085205 (2003).
[4] C. N. Berglund and W. S. Bear, Phys. Rev. 157, 358 (1967).
[5] P. Bernasconi, I. Biaggio, M. Zgonik, and P. Günter, Phys. Rev. Lett. 78, 106
(1997).
[6] E. Iguchi, N. Kubota, T. Nakamori, N. Yamamoto, and K. J. Lee, Phys. Rev.
B 43, 8646 (1991).
[7] A. M. Glazer, Acta Cryst. B 28, 3384 (1972).
[8] N. W. Thomas, Acta Cryst. B 52, 16, (1996); H. T. Stokes, E. H. Kisi, D. M.
Hatch, and C. J. Howard, Acta Cryst. B 58, 934 (2002).
[9] M. W. Lufaso and P. M. Woodward, Acta. Cryst. B 57, 725 (2001).
[10] P. M. Woodward, Acta Cryst. B 53, 44 (1997).
[11] T. S. Bush, J. D. Gale, R. A. Catlow, and P. D. Battle, J. Mater. Chem. 4, 831
(1994).
[12] T. S. Bush, C. R. A. Catlow, A. V. Chadwick, M. Cole, R. M. Geatches, G. N.
Greaves, and S. M. Tomlinson, J. Mater. Chem. 2, 309 (1992).
[13] P. M. Woodward, A. W. Sleight, and T. Vogt, J. Phys. Chem. Solids 56, 1305
(1995).
[14] A. D. Walkingshaw, N. A. Spaldin, and E. Artacho, Phys. Rev. B 70, 165110
(2004).
[15] F. Cora, M. G. Stachiotti, C. R. A. Catlow, and C. O. Rodriguez, J. Phys.
Chem. B 101, 3945 (1997); M. G. Stachiotti, F. Cora, C. R. A. Catlow, and
C. O. Rodriguez, Phys. Rev. B 55, 7508 (1997).
1. Calculate the magnitude of the spontaneous polarization (dipole moment/unit volume)

in C/m2 for BaTiO3 using the displacements given in this chapter (see Fig. 10.1)
and assuming charges of +2e, +4e, and −2e for the Ba, Ti and O ions, respectively
(a = 3.992 Å, c = 4.036 Å).
2. The perturbation matrix element ∆H(1,14) is the change in the LCAO interaction
between the dz2 (~r) orbital and the two orbitals pz (~r + a~ey ) and pz (~r − a~ey ) due to the
√
relative oxygen-ion displacement. Show that ∆H(1, 14) = δOx [(pdσ) − 2 3(pdπ)] to
first order in δOx .
3. Show that the unitary transformation, U , block-diagonalizes the 5×5 matrix of (10.4d)
into a 2×2 and a 3×3 block, where
√ 
2 0 0 0 0
√
 0 2 0 0 0
 √ 
1  
U = √  0 0 2 0 0 .
2 0 0 0 1 1

0 0 0 −1 1
Using this transformation derive the eigenvalue equations and find the eigenvectors (in
terms of the amplitudes of the orbitals) for the 2×2 block.
4. Describe the tilting and ion displacements of a crystal with the a+ 0 +

0 b0 c− structure.
5. Consider the interactions between an s orbital on the A ions and the d orbitals on a
B ion. The B ion at a(0, 0, 0) has eight nearest-neighbor A-ion s orbitals located at
(±a, ±a, ±a). Find the Slater–Koster parameters for the s–d interactions for the five
types of d orbitals. What effect would you expect this to have on the (a) π and π ∗
bands at Γ and (b) the σ and σ ∗ bands at Γ?
6. Suppose the B ion in Problem 5 is displaced by a small amount, δa, along the positive
z-axis. Find the new Slater–Koster s–d parameters to first order in δ. What effect would
you expect this to have on the (a) π and π ∗ bands at Γ and (b) the σ and σ ∗ bands at
Γ?
11
High-temperature superconductors
11.1 Background
In 1986 Bednorz and Müller [1] made the surprising discovery that the insulating,
ceramic compound, La2 CuO4 , was superconducting at low temperature when suit-
ably doped with divalent ions. In fact, all of the members of the class of copper
oxides, La2−x Mx CuO4 (where M is Ba2+ , Sr2+ , or Ca2+ ions), were found to be
superconducting for x in the range, 0.1 . x . 0.25. At optimal doping of x ≈ 0.15,
the superconducting transition temperature, Tc , of La2−x Srx CuO4 was about 38 K.
Since 1986 there has been an enormous scientific effort focused on copper oxides
with similar structures. Over 18 000 research papers were published in four years
following Bednorz and Müller’s report. As work progressed around the world, new
compounds were discovered with higher Tc ’s. In 1987 doped samples of YBa2 Cu3 O7
(“YBCO”) were found to be superconducting at 92 K, thus becoming the first super-
conductors with Tc higher than the boiling point of liquid nitrogen (77 K). Recent
studies [2] on mercuric cuprates report Tc in excess of 165 K.
A very brief list of some of the most studied high-temperature superconduc-
tors (HTSC) and their transition temperatures is given in Table 11.1. In column
two “alias” refers to the frequently used name of the undoped, “parent” composi-
tion. For example, Tl1223 refers to: one Tl, two Ba, two Ca, and three Cu atoms.
The tetragonal structure for the La2 CuO4 , is shown in Fig. 11.1(a). The copper
ions are located at the centers of the octahedra and the oxygen ions occupy the
corners of the octahedra. The La ions fill the spaces between the octahedra. The
figure also shows the details of one of the octahedra. It is similar to the perovskite
metal–oxygen octahedron, but elongated in the z-direction. The Cu–Oip (in-plane
oxygen) distance is about 1.9 Å, while the Cu–Oz distance is significantly greater
at around 2.4 Å. As a result the Oz ions interact only weakly with the central
copper ion compared with the Oip ions. The strongly interacting Cu-Oip ’s of the
octahedral arrays form planar sheets. Each sheet consists of a square copper-oxygen
lattice whose unit cell contains one Cu and two oxygen ions. The Cu–O2 sheets are
249
250 High-temperature superconductors
Table 11.1. Some high-Tc superconductors and their Tc values. The formula for
the undoped compound and its alias in the scientific literature are given along with
the number of Cu–O2 layers, n, in close proximity.
Compound formula Alias n Tc (K)

La2 CuO4 La214 1 38
TlBa2 CuO5 Tl1201 1 50
Tl2 Ba2 CuO6 Tl2201 1 80
Bi2 Sr2 CaCu2 O8 BiSCO 2 85
YBa2 Cu3 O7 YBCO 2 92
TlBa2 Ca2 Cu3 O9 Tl1223 3 110
TlBa2 Ca3 Cu4 O11 Tl1234 4 122
HgBa2 Ca2 Cu3 O8 Hg1223 3 166
separated by a large distance, about 6.6 Å. Figure 11.1(b) shows the structure of
La2 CuO4 projected onto the ac-plane.. The Cu–O2 superconducting layers are in
the ab-plane separated from each other by Oz and La ions. Other high-Tc cuprates
have two or more Cu–O2 sheets in close proximity. For example, Y123 (YBa2 Cu3 O7 ,
Tc = 92 K) has two copper-oxygen layers about 3.2 Å apart. This pair of adjacent
layers is separated (8.2 Å) from the next pair by three Y–Ba–O isolation layers.
Generally, it is found that Tc increases with, n, the number of adjacent Cu–oxygen
layers. The schematic phase diagram, T versus acceptor (hole) concentration, for
La2−x Srx CuO2 shown in Fig. 11.2 is typical for the cuprates. The undoped parent
compound from which the HTSC material is derived by acceptor doping1 with Ba,
Sr, or Ca is antiferromagnetic with the nearest-neighbor Cu spins aligned in op-
posite directions. The Neel temperature, TN , for undoped La2 CuO4 is 340 K, but
decreases sharply with hole-doping. The magnetic phase exists up to about x = 0.04.
There is a small region between the antiferromagnetic phase and the superconduct-
ing phase in which La2 CuO4 is a non-magnetic insulator. In the antiferromagnetic
and insulating phases the carrier dynamics are described by a Hubbard model or
the so-called “t–J model” (discussed later). The mobile holes tend to hop between
sites on the same spin sublattice and have large effective masses due to the antifer-
romagnetic correlations. Increased doping begins to interfere with the long-range
antiferromagnetic order resulting in some cases in the formation of a spin glass.
Further doping produces a poor metallic material that becomes superconducting
at low temperatures. In the under-doped region the material is no longer antifer-
romagnetic but the spin correlations persist. The coexistence of antiferromagnetic
1
Not all HTSC materials are hole-doped. The superconducting compound Nd2−x Cex CuO4 is electron-
doped. Similar to the hole-doped cuprates this compound is antiferromagnetic, but structurally it
lacks the apical oxygen and superconductivity exists only over a narrow range of doping.
11.1 Background 251
Figure 11.1. (a) Structure of La2 CuO4 . The Cu ions are located at the center and the
oxygen ions occupy the corners of the tetragonally distorted octahedra. The Oz (apical)
oxygens are 2.4 Å above or below the central Cu, but the in-plane oxygens are at a distance
of only 1.9 Å. The La and apical oxygens act as isolation layers separating the Cu-in-plane-
oxygen layers. (b) Projection of the structure onto the (001) plane showing the layers
end-on.
correlations and superconductivity appears to be common in all the HTSCs and

both phenomena may arise from the Coulomb repulsion effects at the Cu site.
As the doping increases, Tc increases in the region, 0.1 . x < 0.15. At the “op-
timal doping” concentration, x ' 0.15, Tc is maximized. Beyond that point further
doping results in a decrease in Tc (“over-doped” region). At even higher doping con-
centrations (x > 0.25 or so) the material is no longer superconducting. The “normal”
metal phase is characterized by extreme anisotropy in the transport properties. The
conductivity parallel to the Cu–O2 planes is orders of magnitude greater than that
perpendicular to the layers. While the resistivity of conventional metals typically
varies as A + BT 5 at low temperatures due to phonon scattering, HTSC materials
display a nearly linear decrease with temperature of the form C + DT (or in some
cases C/T + DT ) down to the temperature at which superconductivity sets in. The
in-plane conductivity is unaffected by the addition of some impurities, but modified
by others. Dopants that increase resistivity and reduce Tc are those which substi-
tute for Cu and form localized spins in the Cu–O2 layers. Also shown in Fig. 11.2
is a structural phase boundary (dashed line) which separates the tetragonal phase
from the orthorhombic phase. The occurrence of such structural transitions in the
HTSC materials is typical.
Although the scientific literature dealing with the electronic properties of
La2−x Srx CuO4

300
Temperature (K)
200 Antiferromagnetic “Normal” metal
1 Tetragonal
)
Orthorhombic
100
Optimally
Under- doped Over-
doped doped
Superconducting
0
0.0 0.1 0.2 0.3
x (hole concentration)
Figure 11.2. Schematic phase diagram for La2−x Srx CuO4 . The dashed line running diag-
onally across the sketch separates the tetragonal from the orthorhombic structures. Three
regions in the superconducting phase are identified as “under-doped”, “optimally doped”,
and “over-doped”. There is a small insulating region between the antiferromagnetic and
superconducting phases at small doping.
HTSCs is large and rapidly growing, there is as yet no single, complete, and widely
accepted quantitative theory. Theoretical understanding of how the electronic struc-
ture evolves from an antiferromagnetic insulator to a superconductor and then to
a “normal” metal as a function of the doping concentration is still incomplete. It
is generally agreed that electron pairs are responsible for the superconductivity,
but the mechanism of pairing is still uncertain. BCS (Bardeen–Cooper–Schrieffer)
theory, which has been so successful for “conventional” (lower-temperature) super-
conductors, does not appear to adequately describe the behavior of HTSC materi-
als. It is also generally agreed that the superconducting energy gap is anisotropic
in ~k-space and usually has d-wave symmetry. Theoretical explanations of the elec-
tronic properties of the “normal” state are still controversial. A variety of models
for transport in the normal state have been proposed including the total absence
of coherent states and collective modes in which charge and spin are separated and
propagate at different velocities [3].
The development of a quantitative theory for the HTSCs has been difficult
because not only are the systems in the strong electron-correlation regime, but
also there are a large number of competing phenomena with small but comparable
stabilization energies to be considered. These phenomena include magnetic order,
superconductivity, charge density waves, spin density waves, formation of charge
and spin stripe domains, as well as structural phase transitions. In addition, the
number of different HTSC compounds discovered is growing rapidly. Perhaps the
most remarkable thing is not the differences displayed by this highly diverse collec-
tion of compounds, but rather the striking similarities.
A review of the field of HTSC in general, and the various current candidate
theories in particular, is beyond the scope of this introductory text. Instead, in this
chapter we shall concentrate on discussions of some of the qualitative electronic
models that can be related to experimental results, particularly results obtained
from angle-resolved photoemission. For this purpose we will relate our results
mainly to the La2−x Srx CuO4 (LSCO) system because it has a relatively simple
crystal structure, has single, isolated Cu–oxygen layers, and can be reliably doped
up to a hole concentration of x = 0.35.
11.2 Band theory and quasiparticles
The application of conventional energy band theory to the electronic structure

of the Cu–oxygen layer leads to the prediction that the parent cuprate materials
should be metals with half-filled, d-like conduction bands. This is not the case. Band
theory fails because the electron correlation energy associated with two electrons
occupying the same copper site is large.
A simple ionic model of La2 CuO4 , for example, would have La3+ , Cu2+ , and
O ions. Thus, the copper ions would have the electronic configuration [Ar] 3d9
2−
while the La3+ and O2− ions have closed-shell or subshell configurations. In the
solid the d orbitals are split into the t2g and eg states that can accommodate six
and four electrons, respectively (including spin states). The tetragonal symmetry
further splits the levels. The t2g states split into a singlet, dxy , and a doublet dxz
and dyz . The eg states split into dz2 and dx2 −y2 levels. According to energy band
theory these states are broadened and covalently admixed with oxygen p orbitals
to form one-electron pi and sigma bands. The valence-band states below the p–d
gap are filled and the nine outer electrons of the copper ion are distributed among
the antibonding σ ∗ and π ∗ bands. Six of the outer nine d electrons will occupy
the π ∗ bands, two will occupy the lowest σ ∗ band, and the remaining electron will
reside in the upper σ ∗ band. Since each σ ∗ band can hold two electrons per unit
cell, the highest occupied band is half-filled and therefore the material should be
metallic. However, a transition metal oxide with a half-filled d band usually forms
a Mott–Hubbard antiferromagnet insulator as a result of the strong correlation
effects. In the antiferromagnetic regime the Hubbard model for strong correlations
[4, 5] leads to splitting (of what would have been the upper conduction band) into
two bands separated by a gap of several electronvolts.2 The lower Hubbard band is
2
Depending on the relative magnitudes of the band gap, repulsive correlation energy, and band width,
the filled, non-bonding bands may hybridize with the upper Hubbard band. This contributes additional
structure to the filled bands [6]. Nevertheless, the upper Hubbard band remains empty and is still
separated from the filled states by a sizable energy gap.
fully occupied and the upper band is empty so the material is an insulator and the
states are localized rather than extended as in band theory. In this situation the
p–d interaction causes virtual “hopping” of electrons between the metal and oxygen
ions rather than forming delocalized energy band states. Acceptor dopants such as
Sr2+ that substitute for the trivalent ions introduce holes in the lower Hubbard
band (as for example in LSCO = La2−x Srx CuO4 ). The holes reduce the effect of
the correlation energy. As x increases, a point is reached where delocalized electron
states are competitive with the localized Hubbard states. Further acceptor doping
results in superconductivity at low temperatures. This, however, does not mean
that conventional band theory applies. Strong correlations still operate and must be
taken into account in any description of the electronic states. Experimental results,
however, suggest that a quasiparticle-like description of the low-energy electronic
excitations is appropriate in the superconducting phase.
The details of the low-energy excitations and the Fermi surface (FS) of the
HTSCs have been investigated extensively using angle-resolved photoemission spec-
troscopy (ARPES) [7]. These experiments show that in the doped materials a FS or
at least portions of a FS and quasiparticle peaks can be identified that are related
to states derived from the d- and p-orbital bands. The quasiparticle states are sim-
ilar to the one-electron states of energy band theory, but their energies are much
smaller than what would be expected from band theory. These quasiparticle states
include electron correlation effects that spread the spectral weight of a state over
a range of energies and ~k-vectors and introduce lifetime effects. Such quasiparticles
are described by Landau’s Fermi liquid theory, which, for example, provides a jus-
tification for mean-field theories of electronic structure and a starting point for the
BCS theory of superconductivity [8].
The character of Fermi-liquid quasiparticles can be explained heuristically by
extending the density of states (DOS) concept. In one-electron energy band theory
the states have sharp energies in ~k-space. The DOS for a single band state is
( )
1 1 ¡ ¢
ρ(ω, ~k) = − Im = δ ω − E(~k) . (11.1)
π ~
ω − E(k) + i0+
The effects of electron–electron interactions lead to a complex, proper self-energy

correction, E(~k) → E(~k) + Σ(~k, ω) with
Σ(~k, ω) = Σ0 (~k, ω) + iΣ00 (~k, ω) . (11.2)
The DOS takes the form

( )
1 1
ρ(~k, ω) = − Im ¡ ¢
π ω − E(~k) + Σ0 (~k, ω) + iΣ00 (~k, ω)
1 Σ00 (~k, ω)
= £ ¡ ¢¤ £ ¤ . (11.3)
π ω − E(~k) + Σ0 (~k, ω) 2 + Σ00 (~k, ω) 2
The result in (11.3) is the one-particle spectral function, usually denoted by A(~k, ω).
It includes the effects of electron–electron interactions, which broaden the energy
and introduce lifetime effects. The spectral weight described by A(~k, ω) is dis-
tributed according to the ~k and ω dependences of Σ0 (~k, ω) and Σ00 (~k, ω). The Fermi
liquid quasiparticle state is not an eigenstate of the system. Instead it is coherent
superposition of eigenstates with a narrow spread of energies and momenta. As a
result, the quasiparticle has a finite lifetime that is related to Σ00 (~k, ω)−1 . Never-
theless, there remains a FS similar to that for non-interacting electrons, but in this
case the sharp edges of the Fermi distribution function are smeared out, even at
T = 0 K. The “line shape” described by (11.3) consists of a Lorentzian peak and
a broad, relatively smooth background. In fact, A(~k, ω) can be separated into two
parts [9, 10]:
A(~k, ω) = A(~k, ω)coherent + A(~k, ω)incoherent , (11.4)
representing the “coherent” and “incoherent” parts of the spectral function. The
coherent part is given by
Γ(~k)/π
A(~k, ω)coherent = Z(~k) ¡ ¢2 (11.5)
ω − ²(~k) + Γ(~k)2
µ ¶−1 ¯
~ ∂Σ0 (~k, ω) ¯
¯
Z(k) = 1 − ¯ , (11.6)
∂ω ω=²(~
k)
¯ ¯¯¯
Γ(~k) = Z(~k)¯Σ00 (~k, ω)¯¯ , (11.7)
ω=²(~
k)
£ ¤¯¯
²(~k) = Z(~k) E(~k) + Σ0 (~k, ω) ¯ . (11.8)
ω=²(~
k)
For non-interacting electrons the occupation number of a state, n(E(~k)), has a

discontinuity of 1 as E(~k) passes through the Fermi energy. For quasiparticles the
discontinuity is reduced due to electron–electron interactions. Z(~k) is a measure of
the discontinuity in n(~k) across the FS surface and of the validity of the quasipar-
ticle representation. Small Z(~k) indicates that the quasiparticles undergo strong
interactions and have short lifetimes, while Z(~k) = 1 indicates a sharp Fermi dis-
tribution, sharp energies, and infinite lifetimes. Equation (11.8) shows that the
“quasiparticle” energy, ²(~k), is smaller than the bare energy, E(~k), when Z(~k) < 1.
Fermi liquid theory is rigorously applicable only when the effects of correlation
are weak. This does not seem to be the case for the HTSCs at low temperatures. The
electron–electron repulsion on the Cu ions, Ud , is comparable to or greater than the
p–d interaction parameters. On the other hand, ARPES experiments clearly show a
well-defined FS and electronic excitations with the characteristics of quasiparticles

exist for the HTSC materials in the superconducting phase.
11.3 Effective Hamiltonians for low-energy excitations
The shortcoming of band theory is that the total wavefunction contains states that
have finite probabilities of two electrons of opposite spin simultaneously occupying
the same Cu orbital, but the one-electron Hamiltonian either ignores this situation
or includes a repulsive energy through the introduction of a mean-field potential.
The LDA or local density approximation method, often used to calculate the elec-
tronic structures, assumes that the exchange and correlation can be represented by
a potential that is a function of the average local density of charge. The potential
is obtained self-consistently by iterative calculations. Unfortunately, the LDA does
not capture the full effect of strong electron correlations.
A different approach is to exclude the double occupation states from the sys-
tem’s wavefunction (e.g., by using a reduced Hilbert space for the possible solu-
tions). The problem may be treated by a variety of methods including “renormal-
ization” schemes and so-called “slave-boson” formulations. The slave-boson scheme
uses two separate particle operators to describe the spin (spinon) and charge (holon)
degrees of freedom of the electrons [11–14]. In many cases the Hilbert space used
excludes the doubly occupied d orbitals. Both renormalization and slave-boson ap-
proaches lead to the same type of effective Hamiltonian for the case when the re-
pulsion parameter Ud is large (usually taken to be infinitely large). The “extended
Hubbard model” Hamiltonian, HeH , obtained in this manner has the form:
X 1 X
HeH = hij a†iσ ajσ + Uij a†iσ aiσ a†jσ0 ajσ0 , (11.9)
ijσ
2 0
ijσσ
where hij are the “renormalized”, metal–oxygen interactions for i 6= j and the
“renormalized” diagonal site energies for i = j. Uij are the electron–electron
Coulomb repulsion parameters of which Uii = Ud or Up . In (11.9), the subscript, σ,
indicates the spin state. Equation (11.9) looks relatively simple, but the solutions
are extremely complex even with a myriad of approximations. Exact solutions have
been found for one-dimensional systems but not otherwise. Because of the com-
plexity of finding meaningful approximate solutions, usually only a single band is
considered. Approximate effective Hamiltonians can be obtained from (11.9) by
retaining nearest- (and sometime next nearest) neighbor interactions. An approx-
imate effective Hamiltonian called the “Rice model” is also often employed. This
Hamiltonian is of the form
X X· 1
¸
HR = P tij a†iσ ajσ P + J ~si · ~sj − niσ nj,−σ . (11.10)
ijσ ijσ
4
The operator, P , projects out the zero and single occupancy states and J is the
effective exchange energy. The parameters, tij are the renormalized tight-binding
(LCAO-like) parameters and ~si is the electron spin operator. For a half-filled band
the second term of the Hamiltonian is dominant and the low-lying quasiparticles are
antiferromagnetic spin excitations. For small doping, mobile holes become possible.
In this regime the holes acquire a sizable effective mass due principally to the
“resistance” of the ordered spin system through which they move. At higher hole
concentration the first term of (11.10) becomes competitive with the second term,
resulting in the loss of long-range spin order.
Effective Hamiltonians (essentially equivalent to the Rice model) called the
“t–J” model (nearest neighbors) and the “tt0 –J” model (nearest- and next-nearest-
neighbor interactions) are also commonly used. The Rice or “t–J” model recasts the
calculation of the electronic/spin structure and low-energy excited states into a form
that is similar to conventional tight-binding (LCAO) band theory. However, the ef-
fective parameters such as Eg , (pdσ) and (pdπ), are typically 5–10 times smaller
than those of conventional band theory [6, 15, 16]. On the other hand, it has been
argued [17, 18] that the O–O parameters, (ppσ) and (ppπ), which govern the hop-
ping between oxygen orbitals are much less affected than the p–d parameters. In our
previous discussions of the band structure of the perovskites these O–O interactions
were of minor importance since they were roughly an order of magnitude smaller
than the (pdσ) and (pdπ) parameters. For the cuprates, after renormalization, it
is likely that the effective (ppσ) (oxygen–oxygen, second-neighbor parameter) is
comparable to (pdσ) and therefore the O–O interactions must be considered in
calculating the quasiparticle band states.
In general, it is found that the energy scale for the effective Hamiltonian is of
the order of 0.1 eV when the band is approximately half-filled and the scale for the
superconducting gap is of the order of a few times kB Tc , a few hundredths of an eV.
These scales are much smaller than typical band structure parameters that tend to
be of the order of 0.5 eV to several electronvolts.
11.4 Angle-resolved photoemission
Angle-resolved photoemission spectroscopy (ARPES) is one of the most important

tools for investigating the electronic properties of the HTSC materials. The process
of photoemission and its relation to the energy band states was described in Chap-
ter 8. In that discussion we concentrated on the emitted electron intensity, I(E),
versus the energy of the initial state. In practice a beam of monochromatic ultra-
violet photons of energy, ~ω (usually less than 100 eV), is incident on the surface
of the sample. Photoelectrons are emitted from the surface in all directions with a
distribution of kinetic energies. For a fixed kinetic energy the photoelectron inten-
sity is measured as a function of the polar angle, θ, relative to the surface normal of
the sample. Using this information the initial state energy and wavevector parallel
to the surface can in principle be inferred if the momentum perpendicular to the
surface is assumed to be conserved,
Ekin = ~ω − Φ − E(~k), (11.11)

p
~~kk = 2mEkin sin θ. (11.12)
If Ekin is fixed and θ varied, a peak in the photoelectron intensity is expected when
~kk corresponds to an initially occupied state on the energy-band dispersion curve
defined by (11.11). Conversely, if ~kk is fixed and Ekin varied, a peak in the intensity is
expected when the energy corresponds to an initially occupied state on the energy-
band dispersion curve. For a typical solid with three-dimensional energy bands,
such peaks are broadened because the component of momentum perpendicular to
the surface, k⊥ is not conserved. Therefore the photoemitted electrons arriving at
the detector are coming from a range of states with different k⊥ values. On the other
hand, for the HTSCs the CuO2 -layer energy bands are two-dimensional and E(~kk )
is the same for all values of k⊥ . As a result, the two-dimensional dispersion curve
E(~k) = E(~kk ) for states below the Fermi energy can be mapped with reasonable
accuracy.
In the BCS theory of superconductivity the formation of Cooper pairs opens
a small gap, 2∆, in the quasiparticle density of states near the Fermi energy. The
DOS is given by
E − EF
ρBCS (E) = ρ(EF ) p (11.13)
(E − EF )2 − ∆2
where ρ(EF ) is the DOS at EF in the normal state. The DOS has a square-root sin-
gularity at E = EF ± ∆ that is clearly observed in electron tunneling experiments.
HTSC material may not be described by BCS theory, but in the superconducting
phase they possess a similar energy gap that can be measured by ARPES experi-
ments.
Figure 11.3 illustrates the type of data obtained from ARPES experiments.
Figure 11.3(a) shows energy distribution curves for a typical HTSC for T > Tc and
T < Tc . The wavevector ~kk is fixed at a point on the FS and the energy scanned.
The midpoint of the leading edge of the intensity curve (indicated by horizontal
tick marks) is taken as the position of the quasiparticle energy. For T > Tc , the
midpoint of the intensity edge corresponds to the state E = EF . For T < Tc , the
midpoint moves away from EF to a lower energy leaving an energy gap, ∆, in
which no quasiparticle states exist. By choosing different wavevectors on the FS
the dependence of ∆ on ~kk can be determined. In many cases ∆(~kk ) possesses
(a) - ∆ (b)
Intensity
Intensity
T < TC
T > TC
Energy EF Energy EF
Figure 11.3. Typical ARPES data for a typical HTSC. (a) Photoemission intensity ver-
sus energy for T > Tc and T < Tc , showing the formation of a superconducting energy
gap, ∆, near the Fermi surface. (b) Energy distribution curves. Each curve is data taken
along a line in the Brillouin zone that cuts across the Fermi surface. The locus of the peaks
defines the energy dispersion, E(~kk ).
approximate d-wave symmetry given by the expression, ∆(~kk ) = constant·| cos kx a −

cos ky a|.
Figure 11.3(b) is an illustration of a series of ARPES energy distribution curves.
Each curve corresponds to a particular ~kk along a line segment in the Brillouin zone.
The intensity, I(E), shows two features. On the right-hand side of the figure it can
be seen that I(E) drops to zero as the energy scans across the Fermi energy. In
the middle of the figure are a series of quasiparticle peaks, one for each energy
distribution curve. Therefore the locus of these peaks is the energy dispersion of
the quasiparticles, E(~kk ).
It should be mentioned that the vector ~kk , determined by Ekin and θ in (11.11)
and (11.12), often lies in the second Brillouin zone, but the results can be projected
back to the first Brillouin zone by symmetry considerations.
It is evident that tracking the quasiparticle dispersion up to the Fermi energy
along different line segments in the Brillouin zone provides a means of determining
the shape of the FS. Thus, ARPES experiments are extremely important because
they provide direct measurements of both the quasiparticle dispersion and the FS.
Finally, we mention that photoemission is a many-body process. The pro-
cess may be equated to the operation of the initial-state destruction operator
~a~k initial (0) acting on the N -electron state followed by the final-state creation oper-
ator, a~† (t), acting on the (N − 1)-electron state. Ignoring final-state effects, the
k final
shape of the quasiparticle peak is directly related to the spectral intensity function,
A(~k, ω) discussed above. Approximate results for real and imaginary parts of the
self-energy function have been derived from ARPES energy distribution curves [16].
11.5 Energy bands of the Cu–O2 layers
As mentioned above, an empirical LCAO model with effective parameters that

approximately include correlation effects can be employed to investigate the nature
of the Fermi surface and density of states when the high-Tc material is sufficiently
doped so that the FS and quasiparticle peaks are identifiable experimentally in
ARPES data.
In this section we use our LCAO model to describe the energy bands of the
Cu–O2 layers. The results can not be applied to the parent (undoped) cuprates
because of the strong correlation effects discussed above. They can, however, be
used to describe the cuprates when they are suitably doped. That is, when the
materials are in the superconducting phase.
We begin by considering the square, planar array formed by the Cu and in-plane
oxygen ions. The LCAO Hamiltonian can be derived directly from the Hamiltonian
for the perovskites given in Chapter 4 by setting to zero the matrix elements asso-
ciated with the Oz ions. That is, by simply deleting the rows and columns labeled
2, 10, and 13 in Table 4.1. This eliminates the orbitals pα (~r − a~ez ) (α = x, y, and
z) and reduces the matrix to that appropriate for the square, copper–oxygen layer
whose unit cell is CuO2 . The result is the 11×11 matrix in Table 11.2. We have
introduced the notation Ez and Ex for the diagonal energies of the dz2 and dx2 −y2
orbitals in place of the cubic parameter, Ee . Due to Jahn–Teller and tetragonal
ligand-field effects, Ez is expected to be more negative than Ex and therefore the
uppermost, half-filled band is derived from dx2 −y2 . For convenience we show in Ta-
ble 11.2 the orbitals involved as well as retaining the labeling used in Table 4.1.
(We remind the reader that we use “a” as the metal–oxygen distance, a distance
that is half of the lattice spacing, a0 . Discussions in the scientific literature use “a”
as the lattice spacing. As a result the high symmetry points such as X and M are
designated by wavevector components apparently larger by a factor of 2. Also it
should be mentioned that most papers seem to neglect the factor of a0 altogether.
Thus one sees the wavevector for X written as k = (π, 0) instead of ka0 = (π, 0). In
our notation the X point is ka = (π/2, 0).)
To begin with we consider the solutions neglecting the oxygen–oxygen interac-
tions. Then, as can be seen from Table 11.2, the Hamiltonian reduces to a block-
diagonal form. There is (a) a 4×4 sigma block involving the σ-type p orbitals,
Table 11.2. The 11×11 matrix (H − E~k I), with vanishing determinant, for Cu–O2 bands.
dz 2 dx2 −y2 pxx pyy dxy pxy pyx dxz pzx dyz pzy
1 3 4 5 6 7 8 9 11 12 14
1 Ez−E~k 0 −i(pdσ)Sx −i(pdσ)Sy 0 0 0 0 0 0 0

√ √
3 Ex−E~k 3i(pdσ)Sx − 3i(pdσ)Sy 0 0 0 0 0 0 0
4 Ek−E~k −2bSx Sy 0 2cCx Cy 0 0 0 0 0
5 Ek−E~k 0 0 2cCx Cy 0 0 0 0
6 Et−E~k 2i(pdπ)Sy 2i(pdπ)Sx 0 0 0 0
7 0 0 0 0
261
E⊥−E~k −2bSx Sy
8 E⊥−E~k 0 0 0 0
∗
9 Hij = Hji Et−E~k 2i(pdπ)Sx 0 0
11 E⊥−E~k 0 4(ppπ)Cx Cy
12 Et−E~k 2i(pdπ)Sy
14 E⊥−E~k
Parameters used are b ≡ (ppσ) − (ppπ), c ≡ (ppσ) + (ppπ), Sα ≡ sin kα a, and Cα ≡ cos kα a.
pαβ ≡ pα (~r − a~eβ ), Ez = dz2 site energy, Ex = dx2 −y2 site energy, Ex − Ez = tetragonal splitting of eg levels.
Table 11.3. Energy bands for the Cu–O2 plane in the nearest-neighbor approxima-
tion. AB, B, and NB stand for antibonding, bonding, and non-bonding, respectively.
Band # Band Type Energy band E~k EΓ

Pi-type bands q
2
6 π ∗ (xy) AB Emπ + ( 1 Egπ ) + 4(pdπ)2 (Sx2 + Sy2 ) Et
q 2
2
7 π(xy) B Emπ − ( 12 Egπ ) + 4(pdπ)2 (Sx2 + Sy2 ) E⊥
8 ππ0 (xy) NB E⊥ E⊥
q
2
9 π ∗ (x) AB Emπ + ( 12 Egπ ) + 4(pdπ)2 Sx2 Et
q
11 π(x) B Emπ − ( 12 Egπ )2 + 4(pdπ)2 Sx2 E⊥
q
2
12 π ∗ (y) AB Emπ + ( 12 Egπ ) + 4(pdπ)2 Sy2 Et
q
2
14 π(y) B Emπ − ( 12 Egπ ) + 4(pdπ)2 Sy2 E⊥
Sigma-type bands
1 σ∗ AB E0 + E1 cos(ϕ/3) Ex
3 σz02 NB E0 + E1 cos(ϕ/3 − 2π/3) Ez
4 σ B E0 + E1 cos(ϕ/3 + 2π/3) Ek
5 σ0 NB Ek Ek
p = −(Ex + Ek + Ez ), q = Ez Ek + Ex Ez + Ex Ek − 4(pdσ)2 Ω, s = (p2 − 3q)/9, E0 = −p/3.
√ √
r = −Ex Ez Ek + (3Ez + Ex )(pdσ)2 Ω, t = (9pq − 2q 3 − 27r)/54, cos(ϕ) = t/ s3 , E1 = 2 s.
and the dz2 and dx2 −y2 orbitals, (b) a 3×3 block involving the oxygen π orbitals,
px (~r − a~ey ), py (~r − a~ex ), and dxy , (c) a 2×2 block involving the oxygen π orbital,
pz (~r − a~ex ), and dxz , (d) a 2×2 block involving the oxygen π orbital, pz (~r − a~ey ),
and dyz . The eigenvalues of the block-diagonalized matrix are easily obtained.
(a) Filled valence bands: the pi bands
The pi-band energies are given in Table 11.3. There are several interesting features.
The π ∗ (xy) and π(xy) bands have the same type of energy dependence as the
perovskite pi bands and hence the DOS is the same as that given by equation
(6.28). Therefore, we expect two dimensional, logarithmic singularities in the π ∗ (xy)
and π(xy) density of states. The π ∗ (α) and π(α) (α = x or y) however, are one-
dimensional (depend on only one component of the wavevector) and therefore their
DOSs will have square-root van Hove singularities at the top and bottom of the
band. The DOS is given by

Z π/2
1 dx 1 Θ(ε)Θ(ε − 1)
ρ(ε) = = p (11.14)
π2 −π/2 ε − Sx2 + i0+ π ε(ε − 1)
(E − Emπ )2 − ( 12 Egπ )2
ε(E) = , (11.15)
4(pdπ)2
dε
ρ(E) = ρ(ε) , (11.16)
dE
where Emπ = 12 (Et + E⊥ ) and Egπ = (Et − E⊥ ).
The ideal logarithmic, square-root and flat band singularities in the DOS will
be broadened in actual data, but should be observable in photoemission from the
filled valence bands below the Fermi level [19].
(b) Sigma bands for the Cu–O2 layer
The secular equation for the four sigma bands yields one flat band with E = Ek
and three bands whose energies are given by the solution of the cubic equation,
£ ¤
(Ez − E) (Ex − E)(Ek − E) − 3(pdσ)2 Ω − (Ex − E)(pdσ)2 Ω = 0, (11.17)
Ω = sin2 (kx a) + sin2 (ky a). (11.18)

Analytical expressions for the three roots are given in Table 11.3.
The band or bands that intersect the Fermi energy determine the topology
of the FS in the cuprates; that is, the highest partially occupied bands. In the
case of the Cu–O2 layer, this is usually the antibonding σ ∗ band. We shall refer to
the four sigma bands in Table 11.3 above as the four-band model. This model is
generally not the model used to discuss the electronic properties of the cuprates.
Instead, it has been customary to ignore the role of the dz2 orbitals. A three-
band model is employed that considers only the dx2 −y2 , px (~r − a~ex ) and py (~r −
a~ey ) orbitals and the interactions among these three orbitals. The model is defined
by the matrix elements shown in Table 11.4, which includes the second-neighbor
interactions between the two oxygen orbitals. It should be noted, however, that the
model neglects other comparable oxygen–oxygen interactions such as the interaction
between px (~r − a~ey ) and py (~r − a~ex ). The justification for neglecting these other
oxygen–oxygen interactions is that the orbitals px (~r − a~ey ) and py (~r − a~ex ) have
no direct interaction with the dx2 −y2 orbital. As a result, they have only a very
minor effect on the σ ∗ band (see Problem 11.7). In Table 11.4 (and Table 11.3) we
use the simplified notation pxx = px (~r − a~ex ) and pyy = py (~r − a~ey ).
Again, ignoring the oxygen–oxygen interaction (i.e., setting b = 0) the resulting
Table 11.4. Matrix for the three-band model, with vanishing determinant.
d(x2 ) pxx pyy

√ √
d(x2 ) Ex − E(~k) 3i(pdσ)Sx − 3i(pdσ)Sy
√
pxx − 3i(pdσ)Sx Ek − E(~k) −2bSx Sy
√
pyy 3i(pdσ)Sy −2bSx Sy Ek − E(~k)
solutions include one non-bonding band with E = Ek , together with an antibonding

band and a bonding band:
Eσ0 (~k) = Ek , (11.19)

r³ ´2
1 ¡ ¢
Eσ∗ (~k) = Emσ + Egσ + 3(pdσ)2 Sx2 + Sy2 , (11.20)
2
r³ ´2
1 ¡ ¢
Eσ (~k) = Emσ − Egσ + 3(pdσ)2 Sx2 + Sy2 . (11.21)
2
The parameters in (11.20) and (11.21) are defined as Emσ = 12 (Ex + Ek ) and Egσ =
(Ex − Ek ). In Fig. 11.4 the energy bands for the four-band model are compared with
those of the three-band model using the same parameters. For the figure we plot
the dimensionless energy, ² ≡ (E − Ex )/|(pdσ)|. As can be seen the bonding and
antibonding bands are similar for both the three- and four-band models with the
four-band model having somewhat larger dispersion. The non-bonding band labeled
σ 0 corresponds to the state with E(~k) = Ek (that is, at ² = (Ek − Ex )/|(pdσ)| =
−2.3333). It is the same for both models. However, the nearly flat band derived
from dz2 and labeled σz02 does not occur in the three-band model. This band has
weak dispersion, indicating weak covalent bonding to the oxygen ions, but it is not
entirely dispersionless. The Fermi energy for the undoped material, ²F , is at the
center of the σ ∗ band, the energy at the X-point,
E(~kF ) − Ex ~kF = (π/2a, 0) .
²F = ,
|(pdσ)|
For the cuprates all of the bands are filled except for the σ ∗ band, which for the
parent material is half-filled. Since the electronic and superconducting properties
are determined by the states near the Fermi level the only band that is significant
is the σ ∗ antibonding band. Therefore it can be argued that the three-band model
is sufficient. This argument will be valid provided the σz02 band is well below the
σ ∗ band. If not, hybridization of the two bands will change the composition of the
wavefunctions. We shall return to this point later in this chapter.
Much of the theoretical work in the field is based on the three-band model
Four-band model
2 Three-band model
ǫF σ∗
1
ǫ = (E − Ex )/|(pdσ)|
0 σz02 Ex
Ez
–1
–2 σ0
Ek
–3
σ
–4
Γ X M Γ
Figure 11.4. Comparison of the energy bands of the three-band and four-band mod-
els. The dotted curves are for the three-band model. The solid curves are for the four-
band model. The flat, non-bonding, σ 0 band occurs in both models. The nearly flat band
σz02 , derived from the dz2 orbitals occurs only in the four-band model. The parameters
used for the calculations of the curves are Egσ /|(pdσ)| = 2.3333, Ex /|(pdσ)| = −1, and
Ez /|(pdσ)| = −1.6667.
because it is simpler to work with than the four-band model. The energy bands
of the three-band model given by (11.20) and (11.21) are of the same analytical
form as those encountered for the pseudo-two-dimensional pi bands of the cubic
perovskites. Therefore we can use (4.41) and (6.28)
√ to immediately obtain the DOS
for the σ ∗ band by the replacement (pdπ) → 23 (pdσ). This gives
¯ ¯ µr ³ ε(E) ´2 ¶ ³ ε(E) ´2
1 ¯ E − 12 Emσ ¯
ρσ∗ (E) = 2 ¯ ¯K 1 − , for ≤ 1, (11.22)
π ¯ 3 (pdσ)2 ¯ 2 2
4
[E − Emσ ]2 − [Egσ /2]2
ε(E) = 3 2
− 2, (11.23)
2 (pdσ)
where K is the complete elliptic integral of the first kind. It follows that the DOS
for the σ ∗ band of the three-band model has the same van Hove singularities as
the DOS for the perovskite π ∗ band including the jump discontinuities at the band
edges and the logarithmic singularity at the center of the band. There is, however,
a critical difference. For the perovskites the three, symmetry-equivalent π ∗ bands

can accommodate six electrons so the possibility of a half-filled conduction band
is not realized for any known perovskite oxide. For example, to half-fill the π ∗ -
conduction bands of WO3 would require doping of 3 electrons per unit cell. On the
other hand, for the cuprates, the σ ∗ band is a single band that can accommodate
only two electrons and hence the parent compounds invariably have a half-filled Cu–
O2 conduction band. Therefore, if energy band theory were applicable to the parent
materials it would predict that the Fermi energy is always precisely at the logarithmic
singularity in the DOS. This common feature is undoubtedly an important factor
in the electronic properties of the high-Tc superconductive materials.
(c) Effects of the oxygen–oxygen interactions
Now we consider the effect of the second nearest neighbor, oxygen–oxygen interac-
tions on the three-band model. Using the full matrix of Table 11.4 gives the secular
equation
£ ¤
(Ex − E) (Ek − E)2 − 4b2 Sx2 Sy2 − 3(pdσ)2 (Ek − E)(Sx2 + Sy2 ) + 12b(pdσ)2 Sx2 Sy2 = 0.
(11.24)
Figure 11.5 shows the results for the three-band model for different values of b.
Since the oxygen–oxygen interaction parameters b and b2 in (11.24) are multiplied
by Sx2 Sy2 the dispersion along the Γ → X (along the kx or ky directions) in the
Brillouin zone is unchanged from the case with b = 0. In the interior of the Brillouin
zone the dispersion is larger for b > 0 than for b = 0, while for b < 0 the dispersion
is reduced. The parameter b = (ppσ) − (ppπ) enters (11.24) linearly through the
last term and hence the results are sensitive to the sign of b. Negative b leads to a
fairly flat dispersion as can be seen in the figure.
DOS plots are shown in Fig. 11.6. Several features are evident. First, the peak
position always occurs at the same energy regardless of the value of b. That is
because the logarithmic singularity in the DOS occurs at a saddle point in the
two-dimensional energy dispersion. The saddle points are the X-points where the
energy is independent of the value of b. The proof of this is not difficult. In two
dimensions a saddle point occurs at E0 when the energy dispersion has the form
E(~k) → E0 + E1 α2 − E2 β 2 , where α and β are small, orthogonal components of the
wavevector and E1 and E2 are positive constants. Writing a power-series expansion
of (11.24) near the X point with kx a = π/2 + α, and ky a = β, we obtain
£ ¤
(Ek − E)(Ex − E) − 3(pdσ)2 (Ek − E) + 3α2 (pdσ)2 (Ek − E)
©£ ¤ ª
−β 2 3(Ek − E) − 12b (pdσ)2 + 4b2 (Ex − E) ∼ = 0. (11.25)
b
= +1
ǫ = (E − Ex )/|(pdσ)| |(pdσ)|
2
0
1
–1
0
Γ X M Γ
Figure 11.5. Energy bands for the Cu–oxygen layer (three-band model) showing
the effect of the oxygen–oxygen interaction parameter, b. The parameters used are
Egσ /|(pdσ)| = 2.3333, Ex /|(pdσ)| = −1, and Ez /|(pdσ)| = −1.6667.
Next, write E = E0 + ∆E, where E0 is the energy at X and satisfies the equation
(Ex − E0 )(Ek − E0 ) − 3(pdσ)2 = 0 at X. (11.26)
Using (11.25) and (11.26) we find that ∆E = E1 α2 − E2 β 2 , where E1 and E2 are

positive coefficients. Therefore, if (pdσ) 6= 0, X is a saddle point. Furthermore, the
saddle-point energy at X is unaffected by the value of b since Sx Sy = 0 at the X
points. We can conclude that (i) there is a logarithmic singularity at X and (ii)
it occurs at E0 independent of the value of b. This does not mean that the Fermi
energy, ²F , is the same for different values of b. On the contrary, ²F depends on
the magnitude and sign of b. It is clear from Fig. 11.6 that for a fixed number of
electrons in the band, ²F moves to higher energy for b > 0 and to lower energy for
b < 0. The effect of b on the shape of the FS is illustrated in Fig. 11.7. The straight
line is the edge of the FS for b = 0 and ²F = ²0 = ²(π/2a, 0) corresponding to the
van Hove singularity and exactly one electron. The curves for b/|(pdσ)| = 1 and –1
are calculated for the same Fermi energy, ²F = ²0 . For b/|(pdσ)| = 1 the curve is
bowed inward (concave) compared to the b = 0 curve. For b/|(pdσ)| = –1 the curve
is bowed outward (convex).
The Fermi level will also shift with the number of holes. Clearly, as the hole
concentration increases ²F must decrease, therefore qualitatively hole-doping coun-
teracts the effect of b > 0 interactions. Arguments have been advanced that the
optimal hole-doping occurs when the effect of the holes just offsets the effect of
the oxygen–oxygen interactions leaving the Fermi energy at the logarithmic sin-
gularity. The positions of the Fermi energies are indicated in Fig. 11.6 for a hole
concentration of 15%, an amount that is usually near optimal hole-doping. The
results of Figs 11.6 and 11.7 assume that the LCAO parameters are unchanged by
the doping. That is unlikely to be the case. In Chapter 6, Subsection 6.4(b) we
3
DOS (spin states/ǫ)
ǫF = 0.87
2
b=0
ǫF = 0.78 ǫF = 0.94
w
w b = +|(pdσ)|
1 b = −|(pdσ)| =
0
0.0 0.5 1.0 1.5 2.5
ǫ
Figure 11.6. Density of states for different values of the oxygen–oxygen interaction pa-
rameter, b = 0, 1, and – 1 (in units of |(pdσ)|). The logarithmic singularity occurs at the
same value of ² for all values of b. For b > 0 the band is expanded. For b < 0 the band is
contracted. The positions of the Fermi energy, ²F , are shown for a hole concentration of
0.15, corresponding to optimal doping.
discussed the x dependence of the LCAO parameters for NaWO3 and found sub-
stantial changes, particularly in the energy gap and the p–d interaction parameter
[20]. Band calculations [21] for the cuprates indicate that the parameters change,
but that ²F remains near the logarithmic singularity as the hole-doping varies. The-
oretical studies indicate that Egσ /b and (pdσ)/b decrease by a factor of about 6
and 3, respectively, as the hole concentration varies from 0.3 to 0.05 in LSCO [17].
This dramatic reduction presumably results because the correlation effects increase
as the band approaches a half-filled condition.
(d) Extended singularity in the DOS
Returning to the energy bands of Fig. 11.5, for the case of b < 0 it can be seen
that the dispersion curve is flattened. In fact, it turns out that there is a value of b
for which the energy band is mathematically flat along the entire line X → M with
energy fixed at E0 . Consider (11.24) with kx a = π/2 and E = E0 . For this choice
of parameters the secular equation reduces to
©£ ¤ ª
12b − 3(Ek − E0 ) (pdσ)2 − 4b2 (Ex − E0 ) Sy2 = 0 . (11.27)
π
2
b = −(pdσ)
b=0
ky a
b = +(pdσ)
0 π
0 2
kx a
Figure 11.7. Fermi surface for different values of, b, the oxygen–oxygen interaction pa-
rameter. The upper right-hand quadrant of the Brillouin zone shows the effect of b on the
shape of the Fermi surface with the energy fixed at its value (²F = 0.9217) at the X-point.
For b < 0 the surface bows outward, while for b > 0 it bows inward. The parameters used
are Egσ /|(pdσ)| = 2.3333, Ex /|(pdσ)| = – 1, and Ez /|(pdσ)| = – 1.6667.
If there is a value of b that satisfies the equation then E = E0 independent of the

value of ky . That is, the band is flat along X to M. The solutions to the quadratic
equation for b are doubly degenerate with the root, bcrit , given by
3 2
2 (pdσ)
bcrit = . (11.28)
(Ex − E0 )
This critical value of b is necessarily negative because E0 , the energy at X, is
always greater then Ex , the energy at Γ. For the parameters used for Fig. 11.5 this
corresponds to bcrit = – 1.6275 in units of |(pdσ)|. For this special value of b, the
energy is independent of ky along X → M. Furthermore, according to (11.25), near
the X → M line, the energy varies as E0 − 3α2 (pdσ)2 /|(Ek − Ex )| for small values of
α. That means the energy behaves as a one-dimensional system along the line from
X to M. Therefore the DOS singularity changes from logarithmic to a square-root
singularity as E approaches E0 and has the form
ρ(E) ∝ |(E − E0 )|−1/2 , as E → E0 . (11.29)
This type of singularity is called an “extended singularity” [25, 26] because the
2
0.08 h
1 ǫ0
ǫF
0 0.77 e
ǫ = (E − Ex )/|(pdσ)|
–1
M
–2
Γ X
–3
–4
–5
–6 b = bcrit
ǫF = 0.7
–7
Γ X M Γ
Figure 11.8. Energy bands showing the flat antibonding band that occurs for the critical
value, b = – 1.6275 (in units of |(pdσ)|). The flat portion of the upper band along X to M
leads to an extended singularity of the inverse square-root type. The other parameters
used are Egσ /|(pdσ)| = 2.3333, Ex /|(pdσ)| = –1, and Ez /|(pdσ)| = – 1.6667. Inset: Fermi
surface for ²F = 0.7, below the top of the upper band, ²0 = 0.9217.
singularity is extended from a point at X to a line passing through X. Extended

singularities have been reported for a number of HTSCs including Bi2212, Bi2201,
Y123, and Y124. However, it seems that these features are not consistent with the
extended singularity of the three-band model. The problem is discussed in the fol-
lowing sections. The energy bands for b = bcrit are shown in Fig. 11.8. As is evident
the upper band is flat along X → M. Another new feature is that the previously flat
non-bonding band develops dispersion along X → M and touches the upper band at
M. An interesting result is that for EF = E0 the FS for b = bcrit is the entire first
Brillouin zone and includes exactly two electrons while for b = 0 the same energy
corresponds to exactly one electron. Because of the dispersion of the second lowest
band, the FS for fewer than two electrons consists of two areas; one centered at Γ
due to the highest band and a second centered at M due to the lower band. The
FS is illustrated in the inset in Fig. 11.8 using an extended zone scheme to show
the symmetry.
M X M
n = 1.95
1.75
1.50
1.25
1.00 Γ X
0.05
0.25
0.50
0.75
M X M
Figure 11.9. Equi-energy contours (Fermi surfaces) when the oxygen–oxygen interaction,
b, is zero. The rectangle in the center occurs when the σ ∗ band is half-filled (one electron).
Contours external to the rectangle correspond to increasing electron concentrations for
which the Fermi surface consists of four empty pockets centered on the M points in the
Brillouin zone. This is called a “hole-like” Fermi surface. Contours internal to the rectangle
correspond to decreasing electron concentrations for which the Fermi surface consists of a
single filled pocket centered on Γ called an “electron-like” Fermi surface.
(e) Comparison of the three-band results with Fermi surface

experimental data
The shape or topology of the FS of the HTSC materials can be determined exper-
imentally by careful ARPES measurements. It is important to know if empirical
tight-binding (LCAO) models are capable of reproducing the FS topologies in or-
der to establish whether the models are qualitatively useful in understanding the
electronic properties of the cuprates. Before discussing the experimental results it is
worthwhile recalling some of the important features of the FS. First, in the absence
of oxygen–oxygen interactions the density of states, ρ(ε), for the Cu–O2 layer is a
universal function just as it is for the perovskites. This means that the FS topol-
ogy is independent of the empirical parameters. The shape of the FS is determined
entirely by the number of electrons occupying the band. Figure 11.9 shows the uni-
versal FSs for b = 0 for different numbers of electrons occupying the σ ∗ band. The
figure is an extended zone scheme with the constant-energy curves continued into
the adjacent Brillouin zones. The square curve is the FS for exactly one electron in
the σ ∗ band. Moving inward toward the center starting from the side of the square
corresponds to decreasing the number of electrons. Moving outward from the square
Table 11.5. LCAO parameters for La2−x Srx CuO4 .
x Egσ /b (pdσ)/b p/d ratio at X p/d ratio at M

0.05 0.33 0.33 0.53 0.93
0.10 0.50 0.50 0.66 1.11
0.15 1.00 0.84 0.64 1.18
0.22 1.50 1.00 0.75 1.78
0.30 2.33 1.00 0.75 2.45
toward the edges of the diagram corresponds to increasing the number of electrons
in the σ ∗ band. Thus if the number of electrons in the band is increased beyond
one, the FS is bounded externally by four areas each centered at an M point. This
is often referred to as a “hole-like” Fermi surface. On the other hand, starting from
one electron, the addition of holes leads to a FS, that is a single area centered on
Γ. This is called an electron-like FS.
When the oxygen–oxygen interaction, b, is added to the model, the universality
of the FS is lost. The FS shape becomes dependent upon both the size and the sign
of b. For the three-band model, the spectrum of FS topologies are determined by
only two parameters. We shall take the two parameters to be the ratio, Egσ /b, and
the ratio, |(pdσ)|/b. For b > 0 the hole-doped FS is concave, bowing inward toward
Γ. For b < 0 it is convex, bowing out toward M.
Some experimental results for the FS of La2−x Srx CuO4 (LSCO) are compared
with theoretical results in Fig. 11.10. The solid lines and shaded areas are theo-
retical FSs and the tick marks indicate the experimental data and its uncertainty.
It is apparent that the FSs bow inward and therefore the parameter b > 0. The
empirical parameters used to fit the data are shown in Table 11.5. Similar results
have been reported by Mrkonjic and Barisic [17, 18] for both La2−x Srx CuO4 and
YBa2 Cu3 O6.95 . Within the accuracy of the experimental data the LCAO param-
eters are not precisely determined and variations of 10%–15% in the parameters
can lead to fits to the data that are equally “good”. The first column of Table 11.5
lists the hole concentration per CuO2 unit cell. The second and third columns give
the ratio of the energy gap and p–d interaction parameter to the oxygen–oxygen
interaction parameter, b. These ratios show a reduction in (pdσ) and the energy gap
as the doping decreases which is consistent with the idea that correlation increases
as a half-filled band is approached.
The last two columns of Table 11.5 show the ratio of the p-orbital to d-
orbital
q¯ composition of the wavefunction at X and M. The ratios are defined as
¯ ¯ ¯
¯a2p + a2p ¯/¯a2 ¯, where the wavefunction is Ψ = ap pxx + ap pyy + ad dx2 .
xx yy d xx yy
The values obtained indicate bonding similar to that of the d-band perovskites
with a trend toward less covalency as x decreases.
x = 0.30
b = +3(pdσ)
ǫF = 0.863
X M
I
I I I
I
I
III
III
I
I
I
I
Γ X
II
II
I
I
I
I
II
III
I
I II
I I
II I x = 0.15
b = +3(pdσ)
I
I
ǫF = 1.32
I I I
I I I
X M
I
II I
Γ X
I x = 0.05
b = +3(pdσ)
ǫF = 2.14
X M
I
I
Γ X
Figure 11.10. Fermi surfaces (extended into the adjacent Brillouin zones) versus the hole
concentration, x, for La2−x Srx CuO4 . The shaded areas are the filled states. The shapes are
calculated from the three-band theory with oxygen–oxygen interaction, b, included. The
parameters employed in the theoretical calculations are summarized in Table 11.5. The
“I-shaped” tick marks are ARPES data from [22, 23]. (a) x = 0.30 (over-doped region)
showing an isolated “electron-like” Fermi surface. (b) x = 0.15 (optimal doping) show-
ing the Fermi surface evolved into a large “hole-like” surface. (c) x = 0.05 (under-doped)
showing a “hole-like” Fermi surface.
In Fig. 11.10, the agreement between experimental and theoretical FSs is quite
encouraging, especially considering that there are only two free parameters. How-
ever, there are other aspects of the experimental data that do not agree with the
three-band model. One important disagreement is that experiment shows a definite
flattening of the dispersion curve near X for low hole concentrations. This implies
an extended singularity. The problem is, as discussed above, that b is required to be
negative in order to achieve the flat-band condition near X. On the other hand, b
must be positive to match the inward curvature of the experimental FS curves. This
contradiction can not be resolved within the framework of the three-band model.
It should also be mentioned that obtaining good theoretical agreement with
the measured FSs does not guarantee that the theoretical energy bands are also
in agreement. Choosing parameters to fit the FS only fixes certain ratios and does
not fix the absolute energy scale. For example, in Table 11.5 for x = 0.30 two sets
of LCAO parameters with the same ratios are {b = 1, Egσ = 2.33, (pdσ) = 1} and
{b = 3, Egσ = 7, (pdσ) = 3}. The two sets of parameters give σ ∗ band widths that
are very different, but yield the same FS. Therefore, the parameters should be
determined from the experimental energy band dispersion curves when they are
available. Unfortunately, complete dispersion curves can not be obtained experi-
mentally because the data is cut off at the Fermi energy. In addition, very near
the FS the dispersion curve is distorted by the formation of the energy gap in the
superconducting state.
Figure 11.11 shows a sketch of the energy band dispersion curves obtained
by ARPES experiments [22, 23] on La2−x Srx CuO4 for values of the hole-doping
ranging from x = 0.05 to 0.30. As should be expected the Fermi energy moves down
the dispersion curve as doping is increased. The FSs and dispersion curves for
x ≥ 0.15 (metallic and superconducting phases) can be fitted by the simple three-
band model we have been discussing above. However, for x ≤ 0.10 the dispersion
is flat near the X point. For the three-band model this flat behavior can only be
achieved with the oxygen–oxygen interaction negative (b < 0). However, it is clear
that the Fermi energy is well above the energy of the X-point energy for x = 0.10
and 0.05 and this feature requires b > 0. In fact, b/|(pdσ)| must be large and positive
to position the Fermi level as far above the X-point energy as shown for x = 0.05
in Fig. 11.10(c). This failure suggests that other many-body mechanisms [27] such
as antiferromagnetic spin correlations that are not represented by the three-band
model are operative in the insulating phase of LSCO.
(f ) Possible role of the dz2 non-bonding band
As mentioned earlier in this chapter the validity of the three-band model rests on
the assumption that the non-bonding dz2 band is sufficiently far removed from the
0.0
–0.1
x = 0.30
–0.2
0.0
–0.1
x = 0.22
Relative Energy E − EF (eV)
–0.2
0.0
–0.1
x = 0.15
–0.2
0.0
–0.1
x = 0.10
–0.2
0.0
–0.1
x = 0.05
–0.2
Γ X M
Figure 11.11. A sketch of ARPES data for the energy band dispersion as a function of
the hole concentration, x. As x decreases EF moves up the quasiparticle dispersion curve. A
flat band (extended singularity) occurs for x ≤ 0.15. The energy band parameters change
as x changes (Table 11.5). The experimental results are from [22, 23].
σ ∗ band to be neglected. If Ez lies within the energy range of the σ ∗ band the
two bands will hybridize and repel one another. Figure 11.12 illustrates the results
of the nearest-neighbor four-band model for Ez > Ex , at an energy that intersects
the σ ∗ band. The wavefunctions for the two bands are mixed. Starting from Γ on
the left, the d-orbital composition of the lower band is mostly x2 − y 2 in character
while the upper band is mostly dz2 . Proceeding left to right, toward X, the d-
2.5
2.0
ǫ = (E − Ex )/|(pdσ)|
1.67
1.5
1.33
1.0 1.00
0.5
0.0
Γ X M Γ
Figure 11.12. Hybridization of the dz2 and dx2 −y2 bands in the four-band model. The
non-bonding band (lower band) hybridizes with the antibonding band (upper band). From
Γ to X the upper band d-orbital composition is predominantly dx2 −y2 and the lower
band mostly dz2 . Along X to M the d-orbital character reverses, then reverses again
along M to Γ. The Fermi energy for hole-doped material will lie below the energy at
X and therefore the FS states are predominantly of dx2 −y2 character. The values of the
dimensionless parameters used for the calculation of the three examples are Ex /|(pdσ)| = –
1, (Ez − Ex )/|(pdσ)| = 1.0000, 1.3333, and 1.6667, Egσ /|(pdσ)| = 2.3333.
orbital composition of the lower band evolves into mostly dz2 and vice versa for the
upper band. Near M the wavefunctions again reverse their d-orbital characters. In
the absence of the oxygen–oxygen interaction this hybridization does not alter the
shape of the FS. The lower band will be fully occupied and the DOS as well as the
FS of the upper band are still universal curves, independent of the parameters. The
addition of the oxygen–oxygen interactions will have the same effect as described in
the previous section. Because there is always a sizable gap between the upper and
lower curves, ²F always intersects the upper band and never intersects the lower
band.
Although the shape of the FS is not significantly altered by the hybridization
with the dz2 non-bonding band, the matrix elements that govern the ARPES in-
tensity as well as resolution effects can be significantly modified. However, if ²F
lies reasonably near the energy of the logarithmic singularity (energy at X) the
portion of the hybridized σ ∗ band that determines the FS will be predominantly of
x2 − y 2 character. Therefore it appears that the qualitative FS results based on the
11.6 Chains in YBa2 Cu3 O6.95 277
three-band model are not invalidated by hybridization with the dz2 , non-bonding
band.
Theoretical justification for why Ez may lie above Ex has been given by Perry
et al. [24] based on including certain “static correlations” omitted in local-density-
approximation band calculations. These authors propose a model in which the z 2
band possesses modest dispersion in the kz -direction and intersects the x2 − y 2
band at the Fermi energy. In their model the interaction is weak and the resulting
hybridization gap is very narrow. This model does not seem to be compatible with
the wide gap that occurs naturally in the four-band model when Ez > Ex .
11.6 Chains in YBa2 Cu3 O6.95
The high-temperature superconductor, YBa2 Cu3 O6.95 , is slightly orthorhombic

with (b/2a) = 1.025 (2a = lattice spacing). It has a pair of closely spaced, Cu–O2
planes parallel to the a–b plane separated by Y ions. In addition to these super-
conducting planes it also has copper–oxygen chains oriented along the b-axis. A
schematic of the structure is shown in Fig. 11.13(a). Figure 11.13(b) and (c) illus-
trate the Brillouin zone and one of the chains, respectively. It is evident that the
unit cell has one copper and three oxygen ions. Using the LCAO model including
(b) T(0ππ) R(πππ)

c Z(00π)
U(π0π)
b
U(0π0) S(ππ0)
a Γ(000) X(π00)
p+
z
z
py dx2
y
p−
z
(c) unit cell

(a)
Figure 11.13. YBa2 Cu3 O6.95 . (a) Crystal structure. (b) Unit vectors and the Brillouin
zone. (c) Cu–O3 chain showing the orbitals.
6
–1.5
–1
5
ω
4
b=1
3
Y Γ
Figure 11.14. Chain dispersion curves for different values of the oxygen–oxygen inter-
action, b. Other parameters used are ωx = 3 and ωk = 0.
oxygen–oxygen interactions give the following matrix equation for the chain bands:
  
(ωx − ω) −β 1 −1 Cx 2
 −β ∗ C 
 (ωk − ω) bβ −bβ   py 
  =0 (11.30)
1 bβ ∗ (ωk − ω) 0 C
 p+ z

−1 −bβ ∗ 0 (ωk − ω) Cp −
z
where ωx = Edx2 −y2 /σ, ωk = Ek /σ, β = 1 − exp(2iky a), b = 12 [(ppσ) − (ppπ)]/σ,

√
and σ = 23 (pdσ). In (11.30) Cx2 , Cpy , Cp+ z
, and Cp−
z
are the amplitudes of the
dx2 −y2 , py , p+
z and p−
z orbitals for the chain wavefunction. One of the eigenvalues
is ω = ωk corresponding to a non-bonding band for the combination √12 (p+ −
z + pz ).
The secular equation determining the three remaining eigenvalues is
(ωx − ω)(ωk − ω)2 − 2(ωk − ω) − {(ωk − ω) − 4b + 2b2 (ωx − ω)}4Sy2 − 32bSy4 = 0.

(11.31)
The roots of (11.31) can be obtained by use of standard formulae for the roots of a
cubic equation. The root of principal interest is the partially occupied antibonding
chain energy band. Its dependence on the oxygen–oxygen interaction parameter,
b, is illustrated in Fig. 11.14. Positive values of b depress the dispersion relative
to b = 0. Negative b results in a curve with an initial flat portion and then a very
sharp transition to a rapid rise that eventually results in an enhancement of the
11.7 Summary 279
0.0
△
△
△
– 0.2 △
△
E − EF (eV)
△
△
△△
△
△
– 0.4 △△
△△
△
△
△△△△△△△
– 0.6
Y Γ
Figure 11.15. Comparison of experimental and theoretical dispersion curves. The tri-
angles represent the data [28].The solid curve is for b = −1.5 and the dashed curve is for
b = −1.3. Other parameters used are EF = 2.175 σ (eV), ωx = 0.5, and ωk = 0.
dispersion compared to b = 0. The antibonding energy band of the Cu–O3 chain

is one-dimensional and the dispersion along Γ → Y (and Γ → S) in the Brillouin
zone have been observed experimentally [28, 29]. The bottom of the chain band lies
about 0.5 eV below the Fermi level and appears to be rather flat as it approaches Γ,
suggesting that the oxygen–oxygen interaction, b, is negative. Figure 11.15 gives a
comparison of the LCAO chain band with the experimentally observed band. The
theoretical curve for b = −1.5 yields an excellent fit to the experimental data.
11.7 Summary
Since the discovery of HTCS in LSCO by Bednorz and Müller in 1986 a large
number of similar Cu–oxygen layered compounds have been discovered. With time
the record superconducting transition temperature has increased from 30 K to over
165 K for the mercuric cuprates. Essentially all of these compounds are antiferro-
magnetic in the undoped condition and become superconductors when the dop-
ing concentration is greater than about 0.10 holes per unit cell. Optimal dop-
ing at around 0.15 maximizes Tc and above this concentration Tc decreases. For
hole-doping in excess of about 0.25 the material is no longer a superconductor.
The superconducting energy gap is anisotropic and usually has d-wave symmetry.
Tunneling and other experiments indicate that electron pairs are responsible for
the superconductivity. The mechanism(s) for electron pair formation is uncertain
but likely results from electron–electron correlations instead of (or in addition to)
phonon-mediated pairing. The HTSC materials are unstable to the formation of
charge density waves, spin density waves, and charge and spin segregation into
stripes [20].
Conventional energy band theory does not describe the HTSCs because of
strong electron correlations. The antiferromagnetic and insulating phases are de-
scribed by the Mott–Hubbard theory and renormalized effective Hamiltonians that
include correlation effects approximately. At temperatures above Tc the HTSCs are
poor metals with highly anisotropic transport properties and a nearly linear de-
pendence of the resistivity on temperature. Conductivity parallel to the Cu–oxygen
layers is several orders of magnitude greater than that perpendicular to the lay-
ers. Despite strong electron–electron correlations, Fermi-liquid-like quasiparticles
are seen in ARPES experiments that are reasonably well defined in the super-
conducting phase. In addition, Fermi surfaces have been mapped out by ARPES
experiments and fitted to tight-binding models based on the three-band energy
band model with renormalized parameters. In the antiferromagnetic and insulating
regions the dispersion curves for the electron excitations display a flat region in the
neighborhood of the X-point in the Brillouin zone that can not be explained by the
three-band model. The utility of the quasiparticle description is questionable in the
“normal” metal phase and it has been suggested that in that phase there are no
coherent excitations and that collective modes, holons, and spinons, involving the
separation of the spin and charge are operative.
The field of HTSC is moving at a rapid pace both experimentally and theoret-
ically. New materials are being discovered and explored. Experiments are becoming
more precise and able to distinguish between different proposed theories. As a re-
sult one can expect that a better understanding of the physics and chemistry of the
HTSC materials will emerge in the near future.
References
[1] J. G. Bednorz and A. Z. Müller, Z. Physik B 64, 189 (1986).
[2] D. A. Pavlov, Doctoral Thesis, Academic Archive On-Line. Oai:DiVa.se:su-246
(2004); M. Monteverde, M. Núñez-Regueiro, C. Acha, K. A. Lokshin, D. A.
Pavlov, S. N. Putilin, and E. V. Antipov, Physica C 408–410, 23 (2004).
[3] P. W. Anderson, The theory of superconductivity in the high-Tc cuprates,
Princeton Series in Physics, (Princeton, NJ, Princeton University Press, 1999).
[4] P. W. Anderson, Science 235, 1196 (1987).
[5] J. Hubbard, Proc. Roy. Soc. London A 277, 237 (1964).
[6] F. C. Zhang and T. M. Rice, Phys. Rev. B 37, 3759 (1988).
[7] A. Damascelli, Z. Hussain, and Z.-X. Shen, Rev. Mod. Phys. 75, 473 (2003),
(A. Damascelli, Z.-X. Shen, and Z. Hussain, arXiv:cond-mat/0208504 v1 27
Aug 2002).
[8] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957).
[9] D. Pines and P. Nozières, The theory of quantum liquids, Vol. 1 (New York,
Benjamin, 1966).
[10] P. Nozières, The theory of interacting Fermi systems (New York, Benjamin,
1964).
[11] G. Kotliar and A. E. Ruckenstein, Phys. Rev. Lett. 57, 1362 (1986).
[12] H. Hasegawa, J. Phys. Soc. Jpn. 66, 1391 (1997).
[13] J. E. Hirsch, arXiv:cond-mat/0111294 v1 16 Nov 2001.
[14] R. Fresard and M. Lamboley, arXiv:cond-mat/0109543 v1 28 Sep 2001.
[15] E. A. Dagotto, A Nazarenko, and M. Boninsegni, Phys. Rev. Lett. 73, 728
(1994).
[16] V. J. Emery, Phys. Rev. Lett. 58, 2794 (1987).
[17] I. Mrkonjic and S. Barisic, arXiv:cond-mat/0103057 v2 5 Mar 2001.
[18] I. Mrkonjic and S. Barisic, arXiv:cond-mat/0206032 v1 4 Jun 2002.
[19] K. M. Shen, F. Ronning, D. H. Lu, W. E. Lee, N. J. C. Ingle, W. Meevasana,
F. Baumberger, A. Damascelli, N. P. Armitage, L. L. Miller, Y. Kohsaka, M.
Azuma, M. Takano, Hi. Takagi, and Z.-X. Shen, Phys Rev. Lett. 93, 267002
(2004), (arXiv:cond-mat/0407002 v2 6 Jul 2004).
[20] B. G. Levi, Phys. Today 57, 24 (December 2004).
[21] D. M. Newns, P. C. Pattnaik, and C. C. Tsuei, Phys. Rev. B 43, 3075 (1991).
[22] A. Ino, C. Kim, M. Nakamura, T. Yoshida, T. Mizokawa, A. Fujimori, Z.-
X. Shen, T. Kakeshita, H. Eisaki, and S. Uchida, Phys. Rev. B 65, 094504
(2002); A. Ino, C. Kim, T. Mizokawa, Z. X. Shen, A. Fujimori, M. Takaba,
K. Tamasaku, H. Eisaki, and S. Uchida, arXiv:cond-mat/9809311 v3 27 Mar
2000.
[23] T. Yoshida, X. J. Zhou, M. Nakamura, S. A. Kellar, P. V. Bogdanov, E. D. Lu,
A. Lanzara, Z. Hussain, A. Ino, T. Mizokawa, A. Fujimori, H. Eisaki, C. Kim,
Z.-X. Shen, T. Kakeshita, and S. Uchida, Phys. Rev. B 63, 220501 (2001).
[24] J. K. Perry, J. Phys. A 104, 2438 (2000), (arXiv:cond-mat/9903088 v2 2 Sep
1999).
[25] A. A. Abrikosov, J. C. Campuzano, and K. Gofron, Physica C 214, 73 (1993).
[26] K. Gofron, J. C. Campuzano, A. A. Abrikosov, M. Lindroos, A. Bansil, H. Ding,
D. Koelling, and B. Dabrowski, Phys. Rev. Lett. 73, 3302 (1994).
[27] N. Bulut, D. J. Scalapino, and S. R. White, Phys. Rev. B 50, 7215 (1994).
[28] D. H. Lu, D. L. Feng, N. P. Armitage, K. M. Shen, A. Damascelli, C. Kim, F.
Ronning, Z.-X. Shen, D. A. Bonn, R. Liang, W. N. Hardy, A. I. Rykov, and S.
Tajima, Phys Rev. Lett. 86, 4370 (2001).
[29] M. C. Schabel, C.-H. Park, A. Matsuura, Z.-X. Shen, D. A. Bonn, R. Liang,
and W. N. Hardy, Phys. Rev. B 57, 6090 (1998).
1. The electron creation and destruction operators, a†iσ and aiσ , anticommute:
aiσ a†kσ0 + a†kσ0 aiσ = δik δσσ 0 ,

a†iσ a†kσ0 + a†kσ0 a†iσ = aiσ akσ0 + akσ0 aiσ = 0,
the latter equations require that aiσ aiσ = −aiσ aiσ , and a+ + + +
iσ aiσ = −aiσ aiσ and therefore
+ +
aiσ aiσ = aiσ aiσ = 0. If [A, B] ≡ AB − BA, show that
· X ¸ X
aiσ , Hjk a†jσ0 akσ0 = Hik akσ .
j,k,σ 0 k
2. The time derivative of an operator A is given by i~ dA/dt = [A, Hop ], where Hop is the
Hamiltonian in operator form. Find the time derivative of aiσ for the Hamiltonian
X X
Hop = Ej a†jσ0 ajσ0 + Hjk a†jσ0 akσ0 .
j,σ 0 j,k6=j,σ 0
3. The operator form of the three-band model for Cu–O2 can be expressed as
X X
Hthree−band = Eα,i a+
α,i aα,i + Hα,i;β,j a+
α,i aβ,j
α,i α,i;β,j(nn)
X
+ Hα,i;β,j a+
α,i aβ,j
α,i;β,j(nnn)
where α, i denotes an α-type (dx2 , pxx , and pyy ) orbital centered at R ~ i . The notation
α, i; β, j(nn) indicates a sum over (β, j)th orbitals that are nearest neighbors of the
(α, i)th orbital. Similarly, the notation α, i; β, j(nnn) indicates a sum over (β, j)th
orbitals that are the next-nearest neighbors of the (α, i)th orbital. The Hamiltonian
components, Hα,i;β,j are the LCAO two-center interactions between the α-type
orbital centered at R ~ i and the β-type orbital centered at R~ j . (a) Find the equations
for the time derivatives of the three aα,i (α = dx2 , pxx , and pyy ). (b) Assume
aα,i (t) = cα exp(−iωt + ~k · R
~ i ) and show that the equations are equivalent to the
matrix equation of Table 11.4.
4. Consider the three-band model of Table 11.4 and suppose (pdσ) = 0.

(a) Find expressions for the three energy bands and make a graph of them for ~k along
the lines Γ → X → M → Γ for b = 1, Ex = −3, and Ek = −7.
(b) Show that the p density of states for the two dispersive energy bands, is given by
ρ(ξ) = (4/π 2 ) K( 1 − ξ 2 ), where ξ = (E −Ek )/(2b).
5. Make a graph of the curves of constant energy (Fermi surfaces) in kx –ky space for the
two dispersive energy bands in Problem 4 for ξ = 0, 0.01, 0.25, 0.5, 0.75, 0.99, and 1.
6. For the three-band model, assume b = – 3, Ex = – 3, Ek = – 7, and (pdσ) = 0.707 eV.

(a) Make a graph of the three energy bands for ~k along the lines Γ → X → M → Γ.
(b) Discuss the nature of the wavefunction of the upper band along Γ → X → M.
7. Using the five unit-cell basis orbitals, dx2 −y2 (~r), px (~r − a~ex ), py (~r − a~ey ), px (~r − a~ey ),
and py (~r − a~ex ) construct the 5×5 matrix eigenvalue equation. Find the eigenvalues at
Γ, X, and M in the Brillouin zone. Show that the energies for the σ ∗ band are identical
to those of the three-band model at Γ, X, and M.
Appendix A
Physical constants and the complete elliptic integral
of the first kind
A.1 Selected physical constants
Symbol Constant Value Unit

c Speed of light in vacuum 2.9979×108 m/s
me Electron rest mass 9.1094×10−31 kg
mp Proton rest mass 1.6726×10−27 kg
mn Neutron mass 1.6749×10−27 kg
mp /me Proton mass/electron mass 1836.1527
e Electron charge 1.6022×10−19 C
eV electronvolt 1.6022×10−19 J
e/me Electron charge/mass ratio –1.7588×1011 C/kg
h Plank constant 6.6261×10−34 Js
~ Reduced Plank constant 1.0546×10−34 Js
6.5821×10−16 eV s
µe Electron magnetic moment –9.2848×10−24 J/T
µB Bohr magneton 9.2740×10−24 J/T
5.7884×10−5 eV/T
a0 Bohr radius 5.2918×10−11 m
Å Angström 1×10−10 m
NA Avogadro constant 6.0221×1023 mol−1
kB Boltzmann constant 1.3807×10−23 J/K
8.6173×10−5 eV/K
Ry Rydberg 2.1799×10−18 J
13.6057 eV
Cal Mean Calorie 4.1868 J
285
286 Physical constants and the complete elliptic integral of the first kind
A.2 The complete elliptic integral of the first kind
Z π/2
dφ
K(x) ≡ p (A.1)
0 1 − x2 sin2 φ
p
K 0 (x) = K(x0 ) , x0 ≡ 1 − x2 (A.2)
K(−x) = K(x) (A.3)

π
K(0) = (A.4)
2
³4´
K(x) → ln →∞ as x → 1 (A.5)
x0
Z 1
1 1 1
K(x) dx = 2 G , G= 2
− 2 + 2 − · · · ≈ 0.915 9656 (A.6)
0 1 3 5
Z 1
π2
K 0 (x) dx = (A.7)
0 4
1 ³1´
K = K(x) + iK 0 (x) (A.8)
x x
µr ¶
1 x2
K(ix) = √ K . (A.9)
x2 + 1 x2 + 1
A.2 The complete elliptic integral of the first kind 287
Table A.1. Numerical table of K(x).
x2 K(x) K(x0 ) x02 x2 K(x) K(x0 ) x02

0.00 1.570 796 ∞ 1.00 0.26 1.691 208 2.138 970 0.74
0.01 1.574 746 3.695 637 0.99 0.27 1.696 749 2.122 132 0.73
0.02 1.578 740 3.354 141 0.98 0.28 1.702 374 2.105 948 0.72
0.03 1.582 780 3.155 875 0.97 0.29 1.708 087 2.090 373 0.71
0.04 1.586 868 3.016 112 0.96 0.30 1.713 889 2.075 363 0.70
0.05 1.591 003 2.908 337 0.95 0.31 1.719 785 2.060 882 0.69
0.06 1.595 188 2.820 752 0.94 0.32 1.725 776 2.046 894 0.68
0.07 1.599 423 2.747 073 0.93 0.33 1.731 865 2.033 369 0.67
0.08 1.603 710 2.683 551 0.92 0.34 1.738 055 2.020 279 0.66
0.09 1.608 049 2.627 773 0.91 0.35 1.744 351 2.007 598 0.65
0.10 1.612 441 2.578 092 0.90 0.36 1.750 754 1.995 303 0.64
0.11 1.616 889 2.533 335 0.89 0.37 1.757 269 1.983 371 0.63
0.12 1.621 393 2.492 635 0.88 0.38 1.763 898 1.971 783 0.62
0.13 1.625 955 2.455 338 0.87 0.39 1.770 647 1.960 521 0.61
0.14 1.630 576 2.420 933 0.86 0.40 1.777 519 1.949 568 0.60
0.15 1.635 257 2.389 016 0.85 0.41 1.784 519 1.938 908 0.59
0.16 1.640 000 2.359 264 0.84 0.42 1.791 650 1.928 526 0.58
0.17 1.644 806 2.331 409 0.83 0.43 1.798 918 1.918 410 0.57
0.18 1.649 678 2.305 232 0.82 0.44 1.806 328 1.908 547 0.56
0.19 1.654 617 2.280 549 0.81 0.45 1.813 884 1.898 925 0.55
0.20 1.659 624 2.257 205 0.80 0.46 1.821 593 1.889 533 0.54
0.21 1.664 701 2.235 068 0.79 0.47 1.829 460 1.880 361 0.53
0.22 1.669 850 2.214 022 0.78 0.48 1.837 491 1.871 400 0.52
0.23 1.675 073 2.193 971 0.77 0.49 1.845 694 1.862 641 0.51
0.24 1.680 373 2.174 827 0.76 0.50 1.854 075 1.854 075 0.50
0.25 1.685 750 2.156 516 0.75 0.50 1.854 075 1.854 075 0.50
x02 K(x0 ) K(x) x2 x02 K(x0 ) K(x) x2
Appendix B
The delta function
As we have seen in Chapter 6 the δ function (sometimes called the Dirac delta
function) is a useful mathematical tool. In this Appendix we derive formulae for
the representation of the delta functions employed in Chapter 6.
The δ function is defined by its properties:
δ(x − x0 ) = 0 for x 6= x0 , (B.1)

Z
δ(x − x0 ) f (x) dx = f (x0 ) , (B.2)
where f (x) and its derivative are continuous, single-valued

R functions and the inte-
gral is over any range containing x0 . The result, δ(x − x0 ) dx = 1, follows from
(B.2) for f (x) = 1. Another property, δ(x − x0 ) → ∞ as x → x0 , is implied by (B.1)
and (B.2). Clearly if (B.1) holds, the δ function must be arbitrarily large at x0 if
(B.2) is valid.
There are numerous analytical representations of the delta function. We shall
use a frequently employed representation wherein δ(x − x0 ) is the limit of a partic-
ular function:
1 1
δ(x − x0 ) = − Im , in the limit as λ → 0 (B.3)
π x − x0 + iλ
where “Im” indicates the imaginary part of the quantity and λ is a small positive
number. In using this representation there is an implied order of doing things.
The limiting process λ → 0 (λ > 0) is to be performed last. This means one must
calculate the imaginary part first, then take the limit as λ → 0. This limiting process
is often indicated by using the symbol 0+ as we did in Chapter 6.
The imaginary part of (B.3) is
λ 1
δ(x − x0 ) = lim . (B.4)
λ→0 π (x − x0 )2 + λ2
288
The delta function 289
It is easy to show that the delta function defined by (B.4) satisfies the equations
(B.1) and (B.2). For x 6= x0 , in the limit as λ → 0 the right-hand side of (B.4) tends
to zero as the first power of λ and hence δ(x − x0 ) = 0. However, when x = x0 , then
the right-hand side tends to infinity as 1/λ. To show that (B.2) holds we use (B.4)
and write
Z ³λ´Z f (x) dx
δ(x − x0 ) f (x) dx = lim . (B.5)
λ→0 π (x − x0 )2 + λ2
The integration range in (B.5) may be taken as a small interval around x0 , say
from x0 − a to x0 + a. For our purposes we may assume that f (x) possesses a
convergent power-series expansion near x0 ,
X∞
1 (n)
f (x) = f (x0 )(x − x0 )n , (B.6)
n=0
n!
where the constant, f (n) (x0 ), is the nth derivative of f (x) with respect to x eval-
uated at x0 . Inserting (B.6) into (B.5) and changing the integration variable to
z = (x − x0 )/λ, we obtain
∞ Z +a/λ (n)
1 X λn f (x0 ) z n dz
lim . (B.7)
λ→0 π n=0 n! −a/λ z2 + 1
The nth term of the sum vanishes as λn and therefore only the n = 0 term will
survive in the limit as λ → 0. Thus, (B.7) becomes
Z +∞ ¯+∞
1 dz 1 ¯
f (x0 ) 2
= f (x 0 ) arctan(z)¯ = f (x0 ) . (B.8)
π −∞ z +1 π −∞
This Rresult shows that the delta function Rdefined by (B.3) satisfies condition (B.2)
that δ(x − x0 ) f (x) dx = f (x0 ) and that δ(x − x0 ) dx = 1 (by choosing f (x)=1).
As mentioned in Chapter ¯ 6 it is often convenient to work with a function, f (x),
rather than x. If (df (x)/dx)¯x ≡ f (1) (x0 ) 6= 0, we can define δ[f (x) − f (x0 )] in the
0
same way as δ(x − x0 ),
n1 1 o
δ[f (x) − f (x0 )] = lim − Im
λ→0 π f (x) − f (x0 ) + iλ
( )
1 λ
= lim £ P∞ 1 ¤2 . (B.9)
λ→0 π f (n) (x0 )(x − x0 )n + λ2
n=1 n!
For x 6= x0 , δ[f (x) − f (x0 )] → 0 in the limit λ → 0. As x → x0 we have two limiting

processes, but as mentioned earlier, the limit λ → 0 is to be performed last. There-
fore, for x sufficiently close to x0 only the linear term of the power-series expansion
290 The delta function
of f (x) need be retained. Thus,

( )
1 λ
δ[f (x) − f (x0 )] = lim £ ¤2
λ→0 π f (1) (x0 )(x − x0 ) + λ2
n 1 1 o
= lim − Im (1) . (B.10)
λ→0 π f (x0 )(x − x0 ) + iλ
Equation (B.10) can be rewritten as
1 1 1
δ[f (x) − f (x0 )] = − lim Im (B.11)
(1)
π f (x0 ) λ→0 (x − x0 ) + iλ0
where λ0 ≡ λ/f (1) (x0 ). We note that when f (1) (x0 ) < 0, λ0 will approach zero from
the negative side as λ → 0 from the positive side. As a result the imaginary part
will also change sign. This change in sign is canceled by the change in sign of the
multiplying factor, 1/[πf (1) (x0 )]. Thus, the final result can be written for either
case by using the absolute value of f (1) (x0 ):
1 1 1
δ[f (x) − f (x0 )] = − lim Im
π |f (1) (x0 )| λ→0 (x − x0 ) + i|λ0 |
1 1 1
=− Im (B.12)
(1)
π |f (x0 )| (x − x0 ) + i0+
where i0+ is shorthand for the limiting process. The relationship between δ func-
tions is then,
1
δ[f (x) − f (x0 )] = δ(x − x0 ) . (B.13)
|f (1) (x0 )|
Appendix C
Lattice Green’s function
We first consider the lattice Green’s function for the d orbitals alone. In Section
C.3 we calculate the total Green’s function including the p-orbital functions. The
lattice Green’s function, G, for the pi(xy) states of the cubic perovskite is defined
here as the inverse of the matrix, Hd , that describes the interactions between the d
orbitals for a unit-cell layer parallel to the xy-plane. Within the nearest-neighbor
approximation different unit-cell layers are uncoupled and may therefore be treated
as two-dimensional systems. The B ions are located on the xy-plane by the set of
two-dimensional vectors, ρ ~j,m = 2a(j~ex + m~ey ), where j and m are integers. The
px orbitals of the O ions are located at ρ~j,m +a~ey and the py orbitals at ρ
~j,m + a~ex .
For the pi(xy) states the equations for cx , cy , and cxy , the amplitudes of the px ,
py , and dxy orbitals, respectively, are
¡ ¢ ¡ ¢ ¡ ¢
(ωt − ω) cxy ρ~j,m + cx ρ ~j,m + a~ey − cx ρ ~j,m−1 + a~ey
¡ ¢ ¡ ¢
+cy ρ ~j,m + a~ex − cy ρ ~j−1,m + a~ex = 0, (C.1)
¡ ¢ ¡ ¢ ¡ ¢
(ω⊥ − ω) cx ρ ~j,m + a~ey + cxy ρ ~j,m − cxy ρ ~j,m+1 = 0, (C.2)
¡ ¢ ¡ ¢ ¡ ¢
(ω⊥ − ω) cy ρ ~j,m + a~ex + cxy ρ ~j,m − cxy ρ ~j+1,m = 0. (C.3)
Using (C.2) and (C.3) to eliminate the p-orbital amplitudes from (C.1) we
obtain an equation involving only the d-orbital amplitudes.
£ ¤ ¡ ¢ ¡ ¢ ¡ ¢
(ωt − ω)(ω⊥ − ω) − 4 cxy ρ~j,m + cxy ρ~j+1,m + cxy ρ ~j−1,m
¡ ¢ ¡ ¢
+cxy ρ ~j,m+1 + cxy ρ ~j,m−1 = 0. (C.4)
The equations represented by (C.4) can be written in matrix form:
~ xy = 0,
Hd (ω) C (C.5)
where Hd (ω) is the effective Hamiltonian describing the interactions between the
d orbitals and C~ xy is a vector whose components are the d-orbital amplitudes,
291
292 Lattice Green’s function
cxy (~
ρj,m ). The matrix elements of Hd (ω) are given by
Hd [~
ρj,m ; ρ
~j 0 ,m0 ] = (ωt − ω)(ω⊥ − ω)δj,j 0 δm,m0 + δj+1,j 0 δm,m0
+δj−1,j 0 δm,m0 + δj,j 0 δm+1,m0 + δj,j 0 δm−1,m0 . (C.6)
Using a more compact notation (C.6) can be expressed as
Hd (~ ~0 ) = (ωt − ω)(ω⊥ − ω)δρ~,~ρ 0
ρ, ρ
+δρ~,(~ρ 0 +a~ex ) + δρ~,(~ρ 0 −a~ex ) + δρ~,(~ρ 0 +a~ey ) + δρ~,(~ρ 0 −a~ey ) (C.7)
In this notation eigenvectors, C~ k , are characterized by a two-dimensional wavevec-

~ ~ k
tor, k. The components of C are
1 ~
ckxy (~
ρ) = √ e−ik·~ρ . (C.8)
N
Our aim here is to construct the matrix elements of the function [Hd ]−1 . To do
this we note that Hd (~k, ~k 0 ) is diagonal in ~k-space and hence its inverse also diagonal.
Then, transforming Hd−1 (~k, ~k 0 ) back to lattice space we obtain the desired matrix.
Using (C.8) we have,
1 X −i~k·~ρ ~0 0
Hd (~k, ~k 0 ) = e Hd (~ ~0 ) e−ik ·~ρ
ρ, ρ
N 0
ρ
~,~
ρ
= {(ωt − ω)(ω⊥ − ω) − 4 + 2C2x + 2C2y }δ~k,~k0 . (C.9)
Therefore for the matrix elements of the inverse:
Hd−1 (~k, ~k 0 ) = {(ωt − ω)(ω⊥ − ω) − 4 + 2C2x + 2C2y }−1 δ~k,~k0 , (C.10)
where C2α = cos(2kα a) and a is the B–O distance. Since we know the matrix ele-
ments of Hd in ~k-space, we can easily obtain the desired lattice space function, G.
We transform the matrix from the ~k-space representation of (C.10) to lattice-space
representation using the eigenvectors.
~ ~ 0
0 1 XX eik·~ρj,m e−ik ·~ρj0 ,m0 δ~k,~k0

Gε [~ ~ ]= √
ρ, ρ (C.11)
N ~ ~ 0 (ωt − ω)(ω⊥ − ω) − 4 + 2C2x + 2C2y
k k
~
1 X
0
eik·(~ρ−~ρ )
= , (C.12)
N 2ε + 2C2x + 2C2y
~
k
where ε has its usual definition 2ε = (ω − ωm )2 − (ωg /2)2 − 4. Because of trans-

lational symmetry, Gε (~ ~ 0 ) depends only on the difference ρ
ρ, ρ ~−ρ ~ 0 . Since Hd is a
0 0 ∗
hermitian matrix so is Gε and therefore Gε (~ ρ, ρ
~ ) = Gε (~ ρ ,ρ
~) . In the limit of large
N , the sum can be converted to an integral,
Z ~ 0
(2a)2 eik·(~ρ−~ρ )
Gε (~ ~ 0) =
ρ, ρ d~k, (C.13)
(2π)2 2ε + i0+ + 2C2x + 2C2y
where the integration is over the two-dimensional Brillouin zone. We now make use
of an integral representation of the Bessel function
Z π
i−p
Jp (t) = dx eit cos(x) eipx (C.14)
2π −π
and
Z ∞
1 +
= −i dt ei(λ+i0 )t
, (C.15)
λ 0
to arrive at the result

Z ∞
Gε (~ ~ 0j 0 ,m0 )
ρj,m , ρ =i (q+r−1)
ei(εt) Jq (t) Jr (t) dt (C.16)
0
Z ∞
= i(q+r−1) cos(εt) Jq (t) Jr (t) dt
0
Z ∞
+ i(q+r) sin(εt) Jq (t) Jr (t) dt (C.17)
0
where q = j − j 0 and r = m − m0 .
C.1 Function Gε (0)
For ρ ~ 0 in (C.13) we obtain the diagonal d-orbital, lattice Green’s function,

~=ρ
Z π Z π
1 dx dy
Gε (0) = (C.18)
(2π)2 −π −π 2ε + i0+ + 2 cos(x) + 2 cos(y)
Z πZ π
2 dx dy
= . (C.19)
(2π)2 0 0 2ε + i0+ + 2 cos(x) + 2 cos(y)
Inside the pi(xy) band, |ε| ≤ 2, the imaginary part of Gε (0) is related to the
DOS function. In fact, from (6.21) we see that,
1
− Im Gε (0) = 2ρ(ε) = ρ(2ε) . (C.20)
π
Making use of (C.16) and (C.17) we have

Z
i ∞
Gε (0) = − dt eiεt J0 (t)2
2 0
Z Z
i ∞ 1 ∞
=− dt cos(εt) J0 (t)2 + dt sin(εt) J0 (t)2 . (C.21)
2 0 2 0
The result may also be written in terms of the complete elliptic integrals:

1  (k1 ) K(k1 ) |ε| > 2,
Gε (0) = (C.22)
2π  sign(ε) K(k) − iK(k 0 ) |ε| < 2,
√
with k = ε/2, k 0 = 1 − k 2 , and k1 = 1/k. The imaginary part of Gε (0) vanishes
for |ε| > 2. Inside the band, the relation between the imaginary part of Gε (0) and
the DOS is
1 1
− Im Gε (0) = ρ(ε) = ρ(2ε). (C.23)
π 2
0.50
0.25
Re Gε (0)
0.00
–0.25
–0.50
–6 –4 –2 0 2 4 6
ε
Figure C.1. Real part of Green’s function Gε (0).
Figure C.1 shows a graph of the real part of Gε (0). The function possesses
logarithmic singularities at the band edges (ε = ±2) and a jump discontinuity of
magnitude 1/2 at ε = 0. For large values of ε the function decays as 1/(2ε). It is
antisymmetric about ε = 0.
C.2 Function Gε (1)
We define
Gε (1) = Gε (~
ρ, ρ
~ + 2a~eα ), (α = x or y), (C.24)
and obtain the result:

Z π Z π
1 eix dx dy
Gε (1) = 2 +
, (C.25)
2(2π) −π −π ε + i0 + cos(x) + cos(y)
Z Z
= cos(εt) J0 (t) J1 (t) dt + i sin(εt) J0 (t) J1 (t) dt. (C.26)
The imaginary part of Gε (1) vanishes for ε outside the pi energy band, that is for
|ε| > 2. From (C.25) it follows that
1
Gε (1) + Gε (−1) = − ε Gε (0), or
2
1h1 i
Re{Gε (1)} = Re{Gε (−1)} = − ε Gε (0) . (C.27)
2 2
Using (C.22) we can write

1  π/2 − K(k1 ) |ε| > 2,
Re{Gε (1)} = (C.28)
2π  π/2 − |k| K(k) |ε| < 2.
0.50
0.25
Re Gε (1)
0.00
–0.25
–0.50
–6 –4 –2 0 2 4 6
ε
Figure C.2. Real part of Green’s function Gε (1).
Figure C.2 shows a graph of real part of Gε (1). The function possesses logarithmic
singularities at the band edges and a cusp at ε = 0. For large ε it tends to zero as
−1/(4ε2 ). Unlike Re Gε (0), Re Gε (1) is symmetric about ε = 0.
C.3 Lattice Green’s function for the pi bands
In the previous sections we obtained the Green’s function for the d orbitals. In this
section we calculate the Green’s function for the pi bands including the Green’s
functions for the p orbitals. From these results we calculate the “partial” DOS
functions associated with the square of the amplitudes of each of the orbitals in-
volved in the pi bands. The partial DOS functions provide a way of quantifying the
degree of covalent mixing for the pi bands and are employed in obtaining the XPS
photoelectron cross-sections in Chapter 8.
(a) The pi-band lattice Green’s function
According to Chapter 4, equation (4.43), the matrix equation that determines the
pi(αβ) bands, with αβ = xy, xz, or yz, is
£ ¤
ĥ(~k) − E C(~k) = 0, (C.29)
£ ¤
where the 3×3 matrix ĥ(~k) − E is given by
 
Et − E 2iSβ (pdπ) 2iSα (pdπ)
 −2iSβ (pdπ) E⊥ − E 0  (C.30)
−2iSα (pdπ) 0 E⊥ − E
with Sα = sin(kα a). The components of the vector,£ C(~k), ¤are the amplitudes of
the dαβ , pα , and pβ orbitals. The total Hamiltonian H − E is a 3N × 3N matrix
given by
£ ¤ £ ¤
H − E ~k,~k0 ;r,s = ĥ(~k) − E r,s δ~k,~k0 . (C.31)
We define the lattice Green’s function by the equation,

£ ¤ £ ¤−1
H − E G(E) = I or G(E) = H − E , (C.32)
where I is a 3N × 3N unit matrix. In ~k-space the matrix elements of the inverse

are given by
£ ¤−1 £ ¤−1
H − E ~k,~k0 ;r,s = ĥ(~k) − E r,s δ~k,~k0 . (C.33)
The matrix elements of the inverse,

£ ¤−1 1
ĥ(~k) − E r,s = Ms,r , (C.34)
D
where
h i
D = det(~k, E) = (E⊥ − E) (E⊥ − E)(Et − E) − 4(pdπ)2 (Sα2 + Sβ2 ) (C.35)
is the determinant and Ms,r is the (s, r) element of the matrix of the minors, given
by
 (E − E)2 −2i(E − E)S (pdπ) −2i(E − E)S (pdπ)

⊥ ⊥ β ⊥ α
 2i(E⊥ − E)Sβ (pdπ) [(E⊥ − E)(Et − E) − 2

4Sα (pdπ)2 ] 2
4Sα Sβ (pdπ) .
2i(E⊥ − E)Sα (pdπ) 4Sα Sβ (pdπ)2 2
[(E⊥ − E)(Et − E) − 4Sβ (pdπ)2 ]
(C.36)
The lattice-space Green’s function is obtained by transforming G(E)~k,~k0 ;r,s to

lattice-space by means of the unitary transformation
X ~ ~ ~0 ~
G(E; R ~ n )r,s = 1
~ m, R ei(k·Rm −k ·Rn ) G(E)~k,~k0 ;r,s . (C.37)
N
~
k,~
k0
The d-orbital Green’s function discussed earlier in this Appendix is related to the
(1, 1) elements of the full lattice Green’s function in (C.37) as follows:
Gε (~
ρ, ρ ~ m, R
~ 0 ) = (pdπ)2 G(E; R ~ n )11 ~m − R
(R ~n = ρ ~ 0 ).
~−ρ (C.38)
(b) Relation to the density of sates
An important result is the relationship between the DOS and the trace of the
3N × 3N Green’s function matrix. Consider
1 n o 1 n1 X £ ~ ¤−1 o
− Im Tr G(E + i0+ ) = − Im Tr ĥ(k) − E + i0+ (C.39)
π π N
~
k
where Tr indicates the trace. Since the trace of a matrix is invariant

£ under unitary
¤−1
transformation it follows that we may evaluate the trace of ĥ(~k) − E + i0+ in
its diagonalized form. This gives
£ ¤−1 X £ ¤−1
Tr ĥ(~k) − E + i0+ = Ej (~k) − E + i0+ (C.40)
j
where Ej (~k) are the pi-band energies for the π 0 , π ∗ and π,
E1 (~k) = Eπ0 = E⊥ , (C.41)

q
E2 (~k) = Eπ∗ (~k) = Em + (Eg /2)2 + 4(pdπ)2 (Sx2 + Sy2 ), (C.42)
q
E3 (~k) = Eπ (~k) = Em − (Eg /2)2 + 4(pdπ)2 (Sx2 + Sy2 ). (C.43)
With these results, equation (C.39) takes the form

Z
1 n³ 2a ´3 X o X
Im d~k [Ej (~k) − E + i0+ ]−1 = ρj (E), (C.44)
π 2π j j
© ª
where the integration is over the first Brillouin zone. Thus, − π1 Im Tr G(E + i0+ )
is the sum of the density of states functions for each of the three bands and therefore
it is the total density of states for the pi(αβ) bands.
(c) Partial density of states
The “partial” DOS functions associated with each type of orbital involved in the pi
bands can be obtained from the lattice Green’s function. From (C.36) and (C.37)
we have
Z
X£ ¤−1 ³ 2a ´3 (E⊥ − E)2
G(E; R ~ m )1,1 ≡ 1
~ m, R ĥ(~k) − E 11 = d~k , (C.45)
N
~
2π D(~k, E)
k
³ 2a ´3Z [(E⊥ − E)(Et − E) − 4(pdπ)2 Sα2 ]
G(E; R ~ m )2,2 =
~ m, R d~k , (C.46)
2π D(~k, E)
³ ´3Z [(E⊥ − E)(Et − E) − 4(pdπ)2 Sβ2 ]
G(E; R ~ m )3,3 = 2a
~ m, R d~k . (C.47)
2π D(~k, E)
The partial DOS functions are given by
1 n o
ραβ (E) = − Im G(E + i0+ )R~ m ,R~ m ;1,1 , (C.48)
π
1 n o
ρα (E) = − Im G(E + i0+ )R~ m ,R~ m ;2,2 , (C.49)
π
1 n o
ρβ (E) = − Im G(E + i0+ )R~ m ,R~ m ;3,3 , (C.50)
π
where ραβ (E) is the part of the total DOS that the d orbitals participate in. Simi-
larly ρα (E) and ρβ (E) are the parts of the total DOS that the pα orbitals and pβ
orbitals participate in. Since ραβ (E) + ρα (E) + ρβ (E) = − π1 Im{Tr [G(E + i0+ )]}
it follows that ρπ (E) = ραβ (E) + ρα (E) + ρβ (E), where ρπ (E) is the total pi-band
DOS for all three bands together. From (C.35) and (C.36),
½
1 1
ραβ (E) + ρα (E) + ρβ (E) = − Im
π (E⊥ − E + i0+ )
³ 2a ´3 Z π/2 Z π/2 ¾
dkx dky [(E⊥ − E) + (Et − E)]
+ .
2π −π/2 −π/2 [(E⊥ − E + i0+ )(Et − E + i0+ ) − 4(pdπ)2 (Sα2 + Sβ2 )]
(C.51)
The first term on the right-hand side of (C.51) is δ(E⊥ − E), the DOS for the flat,
non-bonding, π 0 band. The second term on the right can be evaluated from (6.28)
as
|E − Em |
ρπ (E) = ρπ (ε(E)). (C.52)
(pdπ)2
The function in (C.52), discussed in Chapter 6, is the DOS for the π and the π ∗
bands. It is given by
µr ³ ε(E) ´2 ¶
1
ρπ (ε(E)) = 2 K 1− , (C.53)
π 2
(E − Em )2 − (Eg /2)2
ε(E) = − 2. (C.54)
(pdπ)2
For E < Em , ρπ (E) gives the DOS of the π band and for E > Em it gives
the DOS of the π ∗ band. This proves that ρπ (E) = ραβ (E) + ρα (E) + ρβ (E) =
ρπ0 (E) + ρπ (E) + ρπ∗ (E).
(d) d-Orbital and p-orbital partial DOS functions
The d-orbital partial DOS, ραβ (E), is given by (C.45). The integral can be evaluated
immediately in terms of ρπ (E). We have
½³ ´ Z ¾
1 ~ m )1,1 = − 1 Im 2a 3 (E⊥ − E)2
− Im G(E; R ~ m, R d~k
π π 2π D(~k, E)
½³ ´ Z π Z π ¾
1 2a 3 2 2 dkx dky (E⊥ − E)
= − Im
π 2π − π2 − π2 [(E⊥ − E + i0+ )(Et − E + i0+ ) − 4(pdπ)2 (Sα2 + Sβ2 )]
|E − E⊥ |
= ρπ (ε(E)). (C.55)
2(pdπ)2
The final result of (C.55) for the d-orbital partial DOS gives the contribution of
the square of the dαβ -orbital amplitude to the states in the range E to E + dE.
For convenience we shall name this function ρπd (E). It is clear that ρπd (E) = 0 at
E = E⊥ , showing that pi bands are pure p orbital in composition at E⊥ . In contrast
to this result, at E = Et , ρπd (E) = ρπ (Et ) (the total DOS at Et ), showing that the
states are pure d orbital in composition at the edge of the conduction bands.
For the p-orbital partial DOS, ρπpx (E) and ρπpy (E), we have
1 ~ m, R
~ m )2,2 + G(E; R ~ m )3,3 } = δ(E − Et ) + |E − Et | ρπ (ε(E)).
~ m, R
− Im {G(E; R
π 2(pdπ)2
(C.56)
The first term of (C.56) is the partial DOS for the pure non-bonding p bands.
The second term is the sum of the two p-orbital partial DOS functions for the π
and π ∗ bands. Symmetry indicates that the two partial DOS functions are equal,
ρπpx = ρπpy , thus

1 1 |E − Et |
ρπ px (E) = δ(E − Et ) + ρπ (ε(E)) = ρπ py (E) . (C.57)
2 4 (pdπ)2
As may be seen from (C.57), these partial DOS functions vanish at E = Et because
the states are pure d orbital at that energy. For E = E⊥ , ρπpx (E) = 12 ρπ (E⊥ ). The
partial DOS functions are obtained in Chapter 8 directly from the wavefunction
amplitudes.
(e) Covalency ratio
For the π and π ∗ bands we define

1 |E − Et |
ρπ p (E) = ρπ (ε(E)), (C.58)
2 (pdπ)2
the sum of the p-orbital partial densities excluding the non-bonding delta function.
Figure C.3 shows ρπ d (E) and ρπ p (E) as functions of E. The parameters used for
the figure are appropriate for SrTiO3 or BaTiO3 . The total area under the two
curves for each function is 1. For ρπ d (E) the conduction-band area is 0.82 and
the valence-band area is 0.18. For ρπ p (E) the valence-band area is 0.82 and the
conduction-band area is 0.18.
1.4
ρπd (E)
1.2 ρπp (E)
1.0
0.8
Valence Conduction
band band
0.6
0.4
0.2
0.0
–10 –9 –8 –7 –6 –5 –4 –3
Energy (eV)
Figure C.3. Partial density of states. Parameters used are E⊥ = – 8.2 eV, Et = –5.0 eV,
and (pdπ) = 1.0 eV.
The d-orbital mixture into the π valence band is

ρπ d (E) (E − E⊥ )
R(E) ≡ = for E2 ≤ E ≤ E⊥ (C.59)
[ρπ d (E) + ρπ p (E)] 2(E − Em )
p
where E2 = Em − (Eg /2)2 + 8(pdπ)2 is the bottom of the valence band. Equation
(C.59) is a measure of the covalent mixing of the d orbitals into the valence band
as a function of E. If we make the substitution, E 0 = E − Em in (C.59) the result
is
(E 0 + Eg /2)
R(E 0 ) = for E2 − Em ≤ E ≤ −Eg /2. (C.60)
2E 0
This result was obtained in Chapter 4 by other means and hRi, the value of R(E)
averaged over the valence band is shown in Fig. 4.5 as a function of Eg and of (pdπ).
Equation (C.60) shows that for a given E 0 , the d-orbital mixing into the valence
band is dependent only on the energy gap, Eg , however, the total mixing for the
entire band depends on both Eg and (pdπ) because the band width depends on
both of these parameters. Symmetry demands that the probability of the p orbitals
in the π ∗ conduction band is the same as the probability of the d orbitals in the
valence band. That this requirement is met is evident in Fig. C.3.
Appendix D
Surface and bulk Madelung potentials
for the ABO3 structure
The electrostatic potentials (Madelung potentials) due to the ionic charges of the
ABO3 perovskite structure are given in this appendix for the infinite and semi-
infinite lattices. The potential for an electron at the point ~r due to the charged ions
is
X e2 q B X e2 qA X e2 qO
VM (~r) = + − , (D.1)
|~r − R~ B| ~ A|
|~r − R ~ O|
|~r − R
~B
R ~A
R ~O
R
where R~ B, R
~ A and R~ O are the vector positions of the B, A, and O ions respectively.
For the infinite lattice the sum extends over all of the lattice sites, while for the
semi-infinite lattice the sum extends over the half-space bounded by a type I or
type II (001) surface. In (D.1), qB , qA , and qO are the magnitudes of the charges
on the B, A, and O ions, respectively.
A bit of finesse is required to carry out these sums and various methods are
discussed in the scientific literature [1–4]. The potentials at the sites are summarized
in Table D.1 for a perovskite such as SrTiO3 with qB = 2qA = 2qO .
In Column 1, the notation “Site(z)” indicates the type of site and z, the dis-
tance below the surface for the type I and type II (001) surfaces. The convention
is that a positive potential is repulsive to an electron and a negative potential is
attractive. The reduction in the B-ion potential at the type I surface means that
the surface site potential is less repulsive.
The potentials in Table D.1 are in units of e2 /2a and the charge is e =
4.802 × 10−10 esu. To convert these potentials to electronvolts multiply the entry by
(14.3942/2a), where 2a is the lattice constant in angstroms. For SrTiO3 , for exam-
ple, 2a = 3.92 in angstroms and the conversion factor is 3.6720. Thus, the Madelung
potential at the Ti site for a type I (001) surface is: 11.7045 × 3.6720 = 42.9789 eV.
For the infinite crystal the value is 12.3775 × 3.6720 = 45.4502 eV. The reduction
of the repulsive potential at the surface is 2.4713 eV. These values are for the full
302
Surface and bulk Madelung potentialsfor the ABO3 structure 303
Table D.1. Bulk and surface electrostatic potentials.
Ion site(z) Infinite lattice Surface Change from

potential potential bulk value
Type I (001)
B(0) 12.3775 11.7045 – 0.6730
A(a) 5.3872 5.4159 0.0287
O(0) – 6.4559 – 6.4590 – 0.0031
O(a) – 6.4559 – 6.4845 – 0.0029
Type II (001)
B(a) 12.3775 12.4092 0.0317
A(0) 5.3872 4.9783 – 0.4089
O(0) – 6.4559 – 5.5127 – 0.9432
O(a) – 6.4559 – 6.4590 0.0031
ionic charges qSr = 2, qTi = 4, and qO = 2. The charge may be adjusted for covalent
effects, however charge neutrality must be maintained. This means that
qA + qB − 3qO = 0. (D.2)
For bulk SrTiO3 it is found that due to covalent mixing the effective charge
is approximately 85% of the ionic charge so that qA = 1.7 for the Sr ions, qB = 3.4
for the Ti ions, and qO = 1.7 for the oxygen ions. With this covalency reduction,
the change in the Madelung potential on the type I (001) surface is 2.1003 eV (less
repulsive).
Using Table D.1, the surface unit-cell site potentials can be calculated for the
type I and type II (001) surfaces. Additional tables and results may be found in
[1–4].
References
[1] E. A. Kraut, T. Wolfram, and W. Hall, Phys. Rev. B 6, 1499 (1972).

[3] F. Hund, Z. Physik 94, 11 (1935).
[4] Y. Sakamoto and U. Takahasi, J. Chem. Phys. 30, 337 (1952).
Index
ABO3 structure, 9, 26, 199, 232, 300 APW, see augmented plane wave method
absorption, 20, 23, 106, 137, 139, 140, 142, 156, ARPES, see angle-resolved photoemission
165, 172, 179, 181 spectroscopy
optical, 104, 157, 161 atomic layer, 201, 202
admixture, 11, 13, 63, 67, 79, 95, 97, 101, 120, 235 atomic orbitals, 27, 35–38, 41–43, 46–50, 57, 150,
161
covalent, 9, 10, 12, 15, 17, 51, 68, 73, 74, 89,
294, 299, 301 hydrogenic orbitals, 235
affinity, electron, 21, 58 Löwdin orbitals, 38, 47, 49, 55, 57, 141,
147–150, 152, 153, 185
allowed transitions, 143, 154
LCAO method, 35, 37, 40, 47, 50
AlO3 , 244
augmented plane wave method (APW), 120, 121
amplitude-oscillating surface states, 208
angle-resolved photoemission spectroscopy

Ba1−x Kx BiO3 , 19
(ARPES), 181, 253, 254, 256–259, 270,
272–275, 279 BaBiO3 , 19, 245
antibonding band bending, 23, 199, 216, 219, 221
bands, 13, 230, 245, 246, 252, 261 band edge, 160, 161, 172, 210, 215, 229, 234–237
chain energy bands, 277, 278 band-edge behavior, 157, 173
cluster states, 12, 95, 97 band-edge contribution, 171

∗
σ bands, 19, 262, 263, 269, 275 band gap, 14, 15, 20, 23, 67–69, 74, 76, 79, 103,
122, 135, 137, 199, 211, 213, 216, 227, 252
states, 11, 21, 22, 51, 93, 238, 242
at Γ, 80, 131, 208
antiferroelectric, SrTiO3 , 231, 241
at X, 83
antiferromagnetic
ionic, 80
coupling, 18
band-gap defect states, 221
correlations, 249, 279
band-gap emission, 194
phase, 249, 251, 252, 278, 279
band-gap energy, 24, 80, 156, 159, 160, 166, 179,
spin excitations, 256
232
antiparallel spins, 33
band-gap photons, 24
305
306 Index
band-gap region, 24, 191, 194, 198, 202, 208, 210, bulk band-edge, 211
211, 217, 219, 220, 221, 223, 224, 228
bulk continuum, 205, 216, 224, 228
band-gap surface states, 202, 218–220
bulk energy bands, 198, 202, 212, 215, 217,
band width, 16, 68, 74, 75, 112, 119, 194, 219, 219–221, 229
252, 299
bulk potential, 199, 201, 220, 300, 301
basis orbitals, 30, 50, 57, 76, 77, 89, 106, 184,
185, 281
C4v , 57, 199
basis states, 11, 37, 53, 56
CaMnO3 , 1, 18
BaTiO3 , 1, 2, 4, 9, 15, 20, 23, 26, 68, 69, 156,
catalysts, 2, 21–23, 198
161, 179, 230–233, 235–237, 242, 298
CaTiO3 , 1, 242
dipole moment, 213
charge density, 32, 34, 36, 37, 42, 251, 279
ferroelectricity, 245, 246
chemisorption, 22, 198, 229
Hall effect, 238
cluster model, 12, 14, 15, 103, 104
mobility, 238
cluster states, 10–12, 17, 26, 89, 93, 99–104, 105
phase transition, 241
CMEA, see constant matrix element
polarization, 17, 22, 34, 137, 138, 146, 152,
approximation
213, 247
coherent excitations, 279
Bi2 Sr2 CaCu2 O8 , 249
collective excitations, 142, 193
Bi2201, 269
collective modes, 251, 279
Bi2212, 269
conduction band, 9, 64, 71, 77, 83, 89, 102, 142,
binding energy, 17, 30–32, 192
144, 156, 194, 157, 221, 223, 224, 230, 245,
BiSCO, 249 252
Bloch’s theorem, 55 empty (unoccupied), 15, 103, 141, 143, 152,

157, 179, 188
BO6 cluster, 11, 26, 89–91, 93, 94, 100, 103
half-filled, 19, 265
bonding states, 11, 21, 97
π ∗ , 15, 65, 68, 73, 112, 122, 125–127, 129, 130,
Brillouin zone
145, 217, 297–299
extended, 126
σ ∗ , 72, 73, 118, 154
first, 13, 53, 54, 75, 76, 106–108, 115, 141, 158,
conduction band edge, 126, 211, 215, 218–220,
258, 269
228
Γ point, 76, 100, 102, 159, 281
conduction electrons, 130, 131, 134, 193
M point, 83, 270, 281
configuration interaction, 39
R point, 67, 73, 85, 87, 100, 102, 176
constant energy surfaces, curves, 126, 281
second, 258
constant matrix element approximation (CMEA),
symmetry points, 76, 80, 146, 259 183, 184
two-dimentional, 291 coordination number, 5
X point, 80,110, 126, 136, 159, 279, 281 coordination polyhedron, 243
Brillouin zone integration, 165, 296 coordination sphere, 242, 243

Index 307
core levels, 182, 193 partial (PDOS), 185–188, 190, 195, 294,
296–298
correlation energy, 16, 219, 224, 252, 253
π 0 band, 112, 113
correlations, 16, 156, 217, 220, 254, 271
π band, 108–112, 122, 127, 128, 130–132, 135,
antiferromagnetic, 249, 259, 267
261, 264
cluster states with band states, 100, 102
σ 0 band, 119, 120, 296
electron–electron, 33, 73, 211, 221, 279
σ band, 113–116, 118, 119, 264, 265, 267, 268,
exchange and, 17, 37, 255 275
static, 276 surface states (DOSS), 208, 209, 211, 212, 214,
217, 223, 229
Coulomb repulsion, 16, 33, 156, 202, 203, 210,
211, 215, 217, 219, 223, 224, 250, 255 destruction operator, 258
covalency, 2, 9, 12, 14, 26, 196, 271, 301 diagonal blocks, 63
covalent, 4, 13, 69, 211 dielectric function, 2, 139–142, 179, 181, 232
covalent bonding, 230, 238, 263 dipole moment, BaTiO3 , 213
covalent mixing, 9, 10, 12, 15, 17, 51, 68, 73, 74, dispersion, 14, 65, 73, 110, 115, 125, 136, 140,
89, 294, 299, 301 205, 257, 258, 263, 265, 267, 269, 273, 274,
276–279
critical points, 110, 111, 114, 116
surface state, 207, 212, 215, 219,
cubic perovskites, 4, 13, 52, 60, 109, 124, 135,
144, 156, 239, 264 dispersion curves, 273, 277
cubic-to-tetragonal transition, 237 displacement field, 138
cubic symmetry, 8, 57, 231, 232 displacive transition, 231, 232
Curie temperature, 237 distortions, 4, 230, 241
donor centers, 15
d bands, 52, 73, 85, 183
DOS, see density of states
d-orbital mixture, 12, 299
DOSS, see density of states, surface
decomposition coefficients, 92, 93
decomposition formula, 92 ε2 (ω), 141, 143, 146–148, 152–154, 156, 157,

159–162, 165, 166, 168–177, 179, 180, 196
defect, 156, 157, 198, 200, 211, 220–222, 228
eg states, 8, 17, 83, 95, 252, 260
defect states, 198, 200, 211, 220–222
effective cross-section, 184
degeneracy, 4, 236, 237
effective mass, 122–129, 132, 134, 256
delocalization, 13, 16, 156, 253
tensor, 125
delta function, 107, 108, 110, 112, 139, 158, 167,
170, 176, 183, 286, 287, 298 eigenvalue condition, 65, 207, 209, 223
density of states (DOS), 2, 19, 26, 67, 73, 104, eigenvalue equation, 34, 51, 55, 56, 60, 63, 69, 70,
106–114, 118–122, 136, 143, 158, 163, 165, 113, 208, 210, 213, 229, 281
183, 184, 191, 194, 253, 257, 262, 270, 281,
electrochromic, 2, 20, 21
291, 292, 295
electrolysis, 23, 24
joint (JDOS), 144–147, 160, 179, 180, 197
electromagnetic theory, 139, 140, 179, 180
308 Index
electron affinity, 21, 58 three-band model, 263, 264, 266, 267, 269, 279,
281
electron concentration, 122, 125, 129, 216, 220
energy band calculations, 2, 52, 213, 242, 245
electron screening, 17, 22
energy band diagram, 13, 14, 82, 141
electron–electron interactions, 253, 254
energy band dispersion, 205, 212, 273, 274
electron–electron repulsion, 16, 202, 210, 217,
220, 221, 254 energy band model, 52, 73, 166, 222
electron-like Fermi surface, 271, 272 energy band structure, 76, 188, 245
electronic configuration, 6, 17, 18, 252 energy band theory, 193, 219, 252, 253, 265, 279
electron correlation, 156, 252, 255, 259, 279 energy bands at Γ, 76, 78, 105
electronic structure, 9, 25, 35, 38, 40, 52, 60, 181, energy bands at X, 80, 81
219, 245, 251, 253
energy bands at M, 83, 84
electronic transitions, 104, 137, 139
energy bands at R, 85–87
electrostatic field, 7, 8
energy distribution curve, 196
electrostatic interactions, 9
energy gap, 9, 26, 66, 74, 83, 124, 125, 130, 141,
electrostatic potential, 7, 210, 213, 220 157, 215, 299
electrostatic splitting, 11, 58 superconducting, 251, 252, 257, 258, 267, 271,
273, 278
elliptic integrals, 110, 112, 145, 164, 165, 169,
214, 227, 264, 283, 284 energy loss, 139, 140, 179
emission rate, 183 exchange energy, 17, 18, 256
empirical parameters, 50, 135, 270, 271 exchange hole, 33, 39
energy band, 2, 9, 12, 26, 40, 53, 54, 60, 63, 64, exchange operator, 29
72, 101–105, 107–111, 119, 120, 123, 124,
exchange potential, 32–35
130, 131, 140, 148, 157, 160, 179–182, 190,
256 excitations, 137, 142, 148, 193, 253, 255, 256, 279
bulk, 198, 202, 217 extended singularity, 268, 269, 273, 274
chain, 277, 278 extended zone scheme, 269, 270
Cu–O2 layer, 257, 259, 261 extinction coefficient, 140
degeneracy, 232
Fermi distribution function, 122, 130, 140, 254
delocalized, 1, 156, 221, 253
Fermi energy, 122, 126–128, 130, 131, 136, 193,
four-band model, 263, 264
199, 209–211, 219, 221, 229, 237, 238, 254,
LCAO, 52, 73, 88, 100, 133, 222 257, 258, 262, 263, 265–267, 273, 275, 276
line, 223 Fermi level, 15, 193, 194, 210, 217, 228, 237, 262,
263, 266, 273
mechanisms, 239
pinned, 217, 219
pi, 136, 293
quasi-, 21
sigma, 17, 114, 174
Fermi liquid, 253, 254, 279
SrTiO3 , 89
Fermi surface (FS), 106, 122, 123, 125–129,
surface, 201–203, 206, 207, 209, 211, 212, 215
Index 309
132–134, 182, 237, 253–255, 257–259, 262, Cu–O2 layer, 248–250, 252, 257, 259, 262, 266,
266, 268–273, 275 270, 276
electron-like, 271, 272 Cu–O3 chain, 276–278
hole-like, 127, 270–272 cuprate superconductors, 19
Fermi–Thomas, 34 FS (Fermi surface), 262, 266, 267, 273, 278
ferroelectric, 2, 4, 125 hole-doping, 249, 266, 267, 273, 278
BaTiO3 , 231, 232 holons, 279
ferroelectricity, 245, 246 optimal doping, 248, 250, 251, 266, 267, 272,
278
FeSiO3 , 1
over-doped, 250, 251, 272
final state, 139–141, 144, 148, 153, 180–183, 190
quasiparticles, 252–259, 274, 279
finite lifetime, 254
under-doped, 249, 251, 272
four-band model, 263, 264, 274–276
hole-like Fermi surface, 127, 270–272
free-carrier absorption, 142, 156
hole potential, 193
frequency-dependent dielectric function, 156–179
hole relaxation, 193, 194, 197
FS, see Fermi surface
optimal, 266
FWHM (full-width at half-maximum), 112, 119,
189 HTSC, see high temperature superconductivity
Hubbard model, 249, 252, 255

g states, 100, see gerade
Hund’s rule, 17, 18
Gaussian, 112, 194, 196
hybridization gap, 276
gerade, 26, 92
Glazer’s system, 240, 241 impurity, 20, 156, 157
Goldschmidt tolerance factor, 243 index of refraction, 140, 161
ground state, 18, 21, 35, 100, 192, 193 infinite lattice, 55, 202, 205, 207, 209, 300
initial state, 39, 139–141, 143, 148, 153, 180–182,

half-filled band, 19, 126, 252, 256, 259, 263, 265, 188, 190, 194, 256, 257
267, 270, 271
insulating perovskites, 9, 15, 20, 137, 156, 198,
Hall effect, 238 212, 218, 219
Hamiltonian, 27, 29, 36, 55, 57, 90–93, 95, 119, interband transitions, 137, 142, 144, 146, 148,
138, 153, 156, 193, 233, 261, 281, 294 149, 152, 154, 156, 166, 170, 172, 179–181
effective, 226, 255, 256, 259, 289 interlayer, 198, 201, 202, 241
Hg1223, 249 interlayer spacing, 201, 202
HgBa2 Ca2 Cu3 O8 , 249 interpolation formula, 112, 133
high spin, 5, 18 intrinsic surface states, 198, 219
high-temperature superconductivity (HTCS), inverse mass tensor, 124

248–252, 255–258, 270, 279
inversion, 26, 92, 93, 100, 232
CuO2 unit cell, 259, 271
ion splittings, 199
310 Index
ionic band gap, 79, 80 lattice space function, 290
ionic charges, 6, 7, 26, 301 LaWO3 , 1
ionic model, 2, 4, 6, 10, 12–14, 25, 52, 77, 220, LCAO, see linear combinations of atomic orbitals
252
LCMTO, see linear combinations of muffin-tin
ionic radius, 5, 7, 243 orbitals
ionic site potentials, 245 LDA, see local density approximation
ionicity, 15 LEED, see low-energy electron diffraction
ionization energy, 21, 22, 35, 36, 58, 224 ligand field, 259
IR (irreducible representations), 11, 57, 90–94, splitting, 14, 17, 18, 88

101
states, 11
isolation layers, 19, 249, 250
LiNbO3 , 1, 20, 242
line band, 223, 228

Jahn–Teller effect, 18, 100, 232, 238, 239, 259
line of oxygen vacancies, 222, 224
JDOS (joint density of states), 144–147, 160, 179,
180, 196 linear combinations of atomic orbitals (LCAO),
43, 184, 213, 278
kinetic energy, 16, 41, 138, 181, 182, 192, 193, 256 energy bands, 88, 100, 133
KKR (Korringa–Kohn–Rostroker method), 121 equation, 55, 60, 203
KMoO3 , 1, 2 integrals, 36, 46, 48, 56, 57, 59, 60, 63, 72, 73,
89, 202, 281
Kramers–Kronig relation, 140
matrix elements, 40, 41, 56–59, 90, 146, 247
KTaO3 , 1, 2, 4, 26, 89, 120, 121, 156
model, 14, 27, 35, 37, 39, 40, 47, 50, 52, 73, 89,
121, 125, 129, 134, 148, 179, 198, 217, 219,
La2 CuO4 , 19, 248–250, 252
220, 222, 230, 239, 256, 259, 270, 276
La2−x Srx CuO2 , 249
parameters, 40, 63, 64, 72, 73, 88, 99, 104, 125,
La214, 249 198, 199, 201, 235, 256, 266, 267, 271, 273
LaCoO3 , 1, 2, 23, 242 -Xα, 36, 37
LaCrO3 , 1, 18 linear combinations of muffin-tin orbitals

(LCMTO), 37
LaFeO3 , 1, 18
local band function, 100, 101
LaFeO3 , 1, 18
local density approximation (LDA), 88, 89, 255
Lagrange multipliers, 30
local exchange potential, 35
LaMnO3 , 1, 2, 17
local potential, 33, 34
LaMnO3 , 1, 2, 17
local symmetry, 87
Landau, 253
localized d electrons, 3, 17, 18
LaNiO3 , 17
localized electrons, 156
LaRuO3 , 22
localized wavefunctions, 89
lattice constant, 3, 124, 232, 300
logarithmic infinity, 165, 194
lattice energy, 242, 244
Index 311
low spin, 5, 18 NdAlO3 , 242
low-energy electron diffraction (LEED), 212, 213 nearest-neighbor approximation, 202, 222, 225,
261
low-energy excitations, 253, 255
nearest-neighbor interactions, 59, 151, 211, 219
Löwdin orbitals, 38, 47, 49, 55, 57, 141, 147–150,
152, 153, 185 nearest-neighbor ions, 13, 18, 60, 149
Neel temperature, 249

Madelung potential, 6–8, 58, 199, 200, 202, 212,
non-bonding bands, 13, 15, 51, 63–65, 73, 79, 83,
213, 220, 226, 300, 301
126, 144, 145, 152, 157, 186, 188, 194, 217,
magnetic susceptibility, 129, 132, 133, 135 242, 252, 297
magnetism, 17 cluster states, 12, 93, 98, 99
magnetoresistance, 125 delta function, 112, 298
magnons, 137, 179 π 0 , 65, 112, 296
matrix-element effects, 160, 179 σ 0 , 69, 119, 263, 264, 269, 273, 275–277
metal–oxygen bonding, 12 non-local exchange potential, 32, 34
metallic bronzes, 19 normalization, 45, 71, 101, 149, 187, 204
metallic perovskites, 194, 218

Oh , 3, 54, 55, 57, 90–92, 100
MgSiO3 , 1
octahedral rotations, 231
mixed compounds, 15
octahedral tilting, 239, 242, 246
mobility, 238
octahedron, 1, 3, 9, 10, 25, 89, 231, 239, 240, 245,
molecular levels, 22
248
momentum matrix elements, 165
one-center integrals, 41
momentum operator, 147
one-electron Hamiltonian, 138, 255
monoclinic, 239
one-electron operators, 29
Mott–Hubbard, 252, 279
optical absorption, 104, 157, 161
Mott-type insulator, 129
optical constants, 179
multiplet, 17
optical properties, 20, 89, 104, 137, 138, 142–144,
148, 156, 157, 179–181
N -electron system, 28, 30, 37, 39
optical transitions, 148, 153, 156
n-doped insulators, 202, 217, 219, 220
optimal doping, 248, 250, 267, 272
n-type SrTiO3 , 15, 18, 23, 125, 142
over-doped, 250, 251, 272
Nax WO3 , 1, 6, 9, 15, 18, 19, 121, 129–132, 134,
overlap, 9, 10, 12, 13, 16, 37, 43, 45, 50, 90, 148,
136, 142, 157, 191, 192, 194, 217–220
150, 179, 235
Na0.85 WO3 , 219
overlap integral, 39, 161
NaNbO3 , 241, 243
overlap matrix, 36, 47, 49
NaTaO3 , 1, 242, 244
oxygen defect states, 220
NaWO3 , 1, 2, 4, 6, 9, 15, 120, 127, 129, 196, 242,
245, 267
312 Index
oxygen ion, 4, 7, 10, 12, 13, 17, 18, 57, 102, 199, cubic-to-tetragonal, 232
220, 221
phonon-mediated pairing, 279
oxygen site, 23, 57, 213
phonons, 125, 137, 179
oxygen vacancy, 222, 225
photochromic, 2, 20
isolated, 225
photoelectrolysis, 2, 23
oxygen–oxygen interactions, 63, 64, 73, 79, 80,
photoemission, 2, 32, 73, 181, 182, 190, 192, 193,
83, 112, 136, 228, 234, 261, 262, 265, 266,
197, 217–220, 252, 253, 256, 258, 262
270, 275, 277
photoexcitation, 20, 39
p–d mixing, 73, 232 photohole, 193
parallel spins, 32, 33 photoionization, 185, 191, 193
partial density of states (PDOS), 185–188, 190, threshold, 193

195, 294, 296–298
pi band, 13, 14, 65, 73, 106, 108–112, 202, 261,
PbCrO3 , 1, 18 264, 294, 296, 297
PbTiO3 , 20, 232, 233, 245 pi(xy), 202, 212–215, 225, 229, 289, 291
PbZrO3 , 20 pi(yz) and pi(xz), 215
PDOS, see partial density of states piezoelectric, 2, 20, 232
perovskite oxides, 1–5, 20, 265 plane waves, 183
perovskites, 6, 7, 12–23, 25, 26, 53–56, 64, 68, 72, plasma oscillations, 193, 196
76, 104, 106, 108, 113, 120, 127, 148, 161,
179, 181, 184, 186, 201, 213, 220, 229, 232, plasmon, 193
240–246, 248, 256, 259, 261, 270, 271, 300 plasmon absorption, 137, 142, 157
cubic, 3, 4, 13, 19, 52, 54, 60, 109, 122–124, plasmon creation, 194, 195
126, 134, 135, 144, 156, 199, 222, 230, 233,
239, 240, 264, 289 plasmon effect, 191, 194
insulating, 9, 15, 20, 103, 122, 137, 141–143, plasmon energy, 142, 194, 196
152, 156, 157, 198, 212, 218, 219
point group, 3, 4, 7, 54, 57, 90, 91, 199
metallic, 15, 69, 100, 126, 137, 142, 157, 194,
polarizability, 232
195, 198, 217, 218
polarization, 17, 22, 34, 137, 138, 146, 152, 213,
semiconducting, 122, 137, 156
247
distorted, 19, 230, 241
polarons, 125, 238
n-type, 124, 210, 218
polyhedron, 243
perturbation, 100, 220, 227–229
potential, 16, 203, 244, 245, 255
surface, 198, 199, 202, 210, 213, 217, 223
Madelung, 6, 7, 8, 57, 58, 199, 200, 202, 212,
tetragonal, 233, 236, 247 213, 217, 218, 220, 221, 226, 230, 300, 301
surface vacancy, 226 electrostatic, 6, 7, 8, 9, 13, 52, 199, 210, 213,

217, 220, 221, 300, 301
perturbation parameters, 209, 212
Coulomb, 32, 202, 229
perturbation theory, 119, 138, 139, 193
local, 33–35, 42
phase transitions, 230, 231, 241
Index 313
non-local, 32, 34 resolution, 181, 189, 194, 219, 275
localized, 41 resonance integral, 39
ionization, 213 RHEED, see reflection high-energy electron

diffraction
periodic, 183
Rice model, 255, 256
nuclear, 32
rigid band model (RBM), 131, 132, 134
exchange, 32–35, 41, 57
Hartree–Fock, 32, 34
saddle points, 110, 115, 136, 265, 266
vector, 138, 182
scanning tunneling microscopy (STM), 232
hole, 193
secular equation, 65, 80, 85, 204, 235, 238, 262,
site, see Madelung 265, 267, 277
surface, 221 block-diagonalized, 234
mean-field, 225 secular matrix, 64, 73, 76, 83
power-law exponent, 176 selection rules, 146, 153
projection, 92, 127, 128, 132, 202 self-consistent solutions, 34, 36, 213, 214
puckered surface, 213 self-energy, 253, 259
PZT, Pb(Zrx Ti1−x )O3 , 20 self-interaction, 32, 35
semi-infinite lattice, 202, 203, 205, 207, 300

qualitative theory, 179
semiclassical theory, 138, 180
quantum numbers, 43, 156
semiconducting perovskites, 122, 137, 156
quantum states, 139
shake-off, 193, 197
quasi-Fermi level, 21
shake-up peaks, 193, 197
sigma band, 13, 14, 19, 69, 70, 73, 106, 109, 113,
radial functions, 149
114, 175, 186, 188, 202, 252, 262
RBM, see rigid band model
singular points, 115
real-space wavefunctions, 66, 87
singularities, 104, 107, 112, 129, 209, 223, 229,
reflectance coefficient, 140 262, 293
reflection high-energy electron diffraction extended, 268, 269, 273, 274

(RHEED), 212, 213
jump discontinuity, 115, 145, 160, 161, 165,
reflectivity, 106, 161, 181 168, 194, 292
relaxation energy, 192 logarithmic, 110, 136, 146, 159, 179, 188, 212,
265–267, 275
relaxation shifts, 193
square-root, 257, 268
ReO3 , 1, 2, 4, 6, 9, 15, 69, 89, 120, 121, 142, 157,
194, 195, 242 van Hove, 106, 111, 143, 144, 159, 261, 264, 266
representation, 30, 39, 47, 57, 79, 87, 90–93, 101, site potential, 7, 300, see Madelung potential
211, 254, 286, 290, 291
Slater and Koster, 40, 42, 46
resistivity, 238, 250, 279
Slater determinant, 28, 29, 33, 35, 39
314 Index
Slater–Koster surface charge, 216, 217, 219, 221
model, 41, 56, see LCAO model surface defect states, 200, 211, 221, 222
parameters, 51, 247, see two-center integrals surface dipole layer, 199
soft optical phonon, 231 surface geometry, 221
space groups, 239, 241 surface layer, 201, 220, 222
specific heat surface oxygen displacement, 213
coefficient, 130–132 surface oxygen vacancies, 23, 200, 222
electronic, 129, 130 surface perturbations, 199, 202, 213, 223
spherical harmonics, 7, 57 surface states, 202, 203, 205, 207–212, 216, 217,
221, 223, 229
spin orbitals, 28, 29, 32, 35
intrinsic, 198, 219
spin waves, 137
surface state condition, 222
splitting, 80, 85, 200, 245, 252
surface state requirement, 209
electrostatic, 8, 11, 58, 77, 200
symmetry, 10, 20, 40, 46, 59, 72, 82, 87, 92–94,
ionic, 89, 199
115, 135, 150, 163, 167, 220, 221, 230, 235,
Jahn–Teller, 232, 239 245, 269, 290
ligand-field, 11, 14, 17, 18, 26, 88 allowed, 21, 22
tetragonal, 235–237, 260 atomic orbitals, 42–44, 185
spontaneous polarization, 247 cubic, 8, 26, 54, 57, 100, 231, 232
SrNbO3 , 1 d-wave, 251, 258, 278
SrRuO3 , 22 forbidden, 21
stabilization energy, 230, 233, 236, 238, 239, 245, inversion, 232
246
point-group, 3, 4, 7, 199
static correlations, 276
tetragonal, 4, 100, 231, 232, 239, 252
static dipole moment, 213
symmetry coordinates, 90, 93, 94, 98–101, 104,
step function, 107 105
STM, see scanning tunneling microscopy symmetry-equivalent, 54, 65, 67, 76, 124,
152–154, 156, 158, 203, 222, 234–236, 265
structural transitions, 4, 231, 250
symmetry group, 90, 199, 241
sublattice, 233, 249
symmetry index, 55, 77, 141
subsurface layers, 220
symmetry points, 53, 54, 74, 76, 80, 123, 146, 259
sudden approximation, 193
symmetry properties, 11, 38, 39, 47, 49, 50, 75,
superconducting energy gap, 182, 251, 256, 258, 102, 133, 148, 149
278
superconductivity, 3, 18–20, 25, 246, 249–251, t2g states, 8, 9, 16, 17, 57, 58, 97, 100, 105, 232
253, 257, 279
tetragonal, 19, 129, 234–238, 241, 246, 248, 250,
surface bands, 198, 202, 203, 207, 211, 213–220, 251
229
Index 315
tetragonal perturbation, 233 nearest-neighbor, 148
tetragonal symmetry, 4, 100, 231, 232, 239, 252 rate of, 179, 182
thermal energy, 237 site-diagonal, 148, 149
thermally excited electrons, 238 symmetry-equivalent, 156, 158
three-band model, 263–266, 269, 271, 273, 276, unmixed, 153, 170
279, 281
transport, 2, 106, 123, 237, 238, 250, 251, 279
three-center integrals, 41
transverse effective mass, 125
threshold energy, 166
triply degenerate t2g , 100
Ti2 O3 , 220–222
trivalent, 3, 253
tight-binding method
tungsten bronzes, 2, 4, 15, 18, 19, 142, 194
conventional, 256
two-center integrals, 39, 41, 43, 48, 57, 61, 72
empirical, 270
two-dimensional Brillouin zone, 291
renormalized, 256 see LCAO model
two-dimensional behavior, character, 2, 19, 73,
tilt angles, 243 106, 125, 159
tilting systems, 239, 241, 244, 246 two-electron operators, 29
TiO2 , 23, 24, 218–222 type I surface, 198, 200, 201, 222, 225, 300
Tl2 Ba2 CuO6 , 249 type II surface, 198, 199, 213, 220
Tl1201, 249
u states (ungerade), 26, 92, 93, 100
Tl1223, 248, 249
Tl1234, 249
van Hove singularity, 106, 111, 143, 144, 159, 261,
Tl2201, 249 264, 266
TlBa2 Ca2 Cu3 O9 , 249 volume continuum, 210, 219, see bulk continuum
TlBa2 Ca3 Cu4 O11 , 249 volume states, 205, 206, 211, see bulk energy
bands
TlBa2 CuO5 , 249
transition matrix elements, 141, 142, 144, 146, WO3 , 1, 2, 4, 6, 9, 15, 19, 21, 219, 220, 245, 265
148, 151–156, 159, 161, 172, 179, 191
transition metal ions, 9, 21, 22, 24, 74 Xα method, 36, 37

transition metal oxides, 198, 252
transition temperature, 4, 19, 248 Y123, 249, 269
transitions Y124, 269
electronic, 104, 137, 139 YAlO3 , 244, 245
equivalent, 154 YBa2 Cu3 O7 , 248, 249
inequivalent, 156 YBa2 Cu3 O6.95 , 271
interband, 137, 148, 152, 154, 159, 160 YBCO, 248, 249
mixed, 160, 172 YCoO3 , 242

Electronic and Optical Properties of D Band Perovskites

Uploaded by

Copyright:

Available Formats

Electronic and Optical Properties of D Band Perovskites

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

Electronic and Optical Properties of D Band Perovskites

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

Thomas Wolfram is a retired professor and Chairman of the Department of

ELECTRONIC AND OPTICAL

Cambridge University Press

© T. Wolfram and S. Ellialtioglu 2006

This publication is in copyright. Subject to statutory exception and to the provision of

ISBN-13 978-0-511-34856-3 eBook (EBL)

ISBN-13 978-0-521-85053-7 hardback

3 Empirical LCAO model 41

7.6 Frequency dependence of ε2 (ω)

8 Photoemission from perovskites 182

9 Surface states on d-band perovskites 199

10 Distorted perovskites 231

11 High-temperature superconductors 249

Table 1.1. Some perovskite and related oxides.

Insulating Metallic Magnetic Superconducting

1.2 The perovskite structure

1.3 Ionic model

Table 1.2. Cations commonly found in perovskite-type oxides. In parentheses is

Dodecahedral A site (Z = 12) Octahedral B site (Z = 6)

Ion Electrons Radius Ion Electrons Radius

by qA then qB = 6 − qA where the three oxygen ions contribute the factor of 6. A

1.4 Madelung and electrostatic potentials

(a) (b) (c)

1.5 Covalent mixing

cluster can be written in the form:

(a) 2t2g (b) 1t2g (c) 1t1g

1.6 Energy bands

1.7 Localized d electrons

1.8 Magnetism in the perovskites

The occurrence of magnetism in the perovskites is closely connected to the exis-

As a consequence of (1) and (3) the electronic configuration of the cation

allel spin electrons, the intra-atomic exchange involves

WO3 is an insulator with a tetragonally distorted perovskite structure. With

in these materials can be explained within the framework of “conventional”, BCS

1.10 Some applications of perovskite materials

The technological uses of perovskite and perovskite-related materials are extensive

“written” by a light beam that causes photoexcitations of these metastable valence

Another important feature of perovskite catalysts is that they can be designed

Suggested additional reading material

J. B. Goodenough, Metallic oxides. Prog. Solid State Chem. 5, 145 (1971).

T. Wolfram, E. A. Kraut, and F. J. Morin, Qualitative discussion of energy bands.

[14] J. B. Goodenough and P. M. Roccah, J. Appl. Phys. 36, 1031 (1965).

Problems for Chapter 1

1. The perovskite structure is often depicted as an array of octahedra. Each octahedron

array of octahedra and indicate the positions of the different ions.

2.1 The Hamiltonian

The total N - electron Hamiltonian, HT , for an atomic, molecular, or solid system

where Ψ(τ ) is the N -electron wavefunction with τ representing the N electron

2.2 The Slater determinant state

The N -electron wavefunction Ψ(τ ),

The Slater determinant wavefunction for an N -electron system is

terminant is referred to as configuration interaction. In many applications to solids

and two-electron operators gij ,

The expectation or average value of the energy of a particular N -electron Slater

P12 ψj (2) ψi (1) = ψj (1) ψi (2). (2.13)

2.3 Koopman’s theorem

2.4 Hartree–Fock equations

The Hartree–Fock equations are obtained by application of the variational princi-