Sie sind auf Seite 1von 113

Home Search Collections Journals About Contact us My IOPscience

Group theory and topology in solid state physics

This content has been downloaded from IOPscience. Please scroll down to see the full text.

1970 Rep. Prog. Phys. 33 533

(http://iopscience.iop.org/0034-4885/33/2/303)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 130.240.43.43
This content was downloaded on 26/09/2013 at 11:09

Please note that terms and conditions apply.


Group theory and topology in solid state physics

J. KILLINGRECK
Department of Physics, University of Hull

Contents
Page
1. Introduction . 534
2. The role of abstract group theory in solid state physics . 536
2.1. Isomorphism and related concepts . 536
2.2. Group characters and product representations . 541
2.3. Group products and invariant subgroups . 546
2.4. Point groups, site groups and factor groups for a crystal lattice 551
2.5. Commuting operators; the group algebra . 552
2.6. Double groups and magnetic groups . 556
2.7. Symmetry groups of nonrigid molecules . 563
3. Matrix methods and group theory . 563
3.1. The secular matrix and group theory . 563
3.2. Equivalent operators. . 569
3.3. The second quantization formalism . 575
3.4. Symmetry properties for many-electron systems . 581
3.5. Class sum operator approach to finite group theory . 586
3.6. Crystal field theory . 588
3.7. Electronic band theory . 599
4. Topological methods in solid state theory . 607
4.1. Planar graphs; Euler’s formula . 607
4.2. Topological wave functions . 609
4.3. Connected graph expansions . 613
4.4. Fixed points of mappings; compound operations . 61 5
4.5. Model calculations for one-, two- and three-dimensional
systems . 617
4.6. Critical points for functions with lattice periodicity . 624
4.7. The Fermi surface . 630
References . 635

Abstract. Firstly, various formal concepts of finite group theory are discussed
with particular reference to their use in recent work on solid state theory. The
pictorial viewpoint is stressed wherever possible throughout the discussion.
Groups of operators are treated, including the anti-unitary operators involved in
the magnetic groups.
Next, the use of group theory in conjunction with matrix methods is discussed,
and this leads to the treatment of equivalent operators. An account is given of
several methods used in the many-electron problem, including a treatment of
second quantization from a group theoretical viewpoint. The class sum operator
approach is outlined as a means of linking the group theoretical approach and the
traditional Dirac approach to quantum mechanics. Some topics from crystal field
theory and electronic band theory are treated as illustrations of the general
principles, and some recent work in these two fields is reviewed.
The basic terminology of graph theory is then given, and seyeral applications
to solid state theory are treated. Some topics which involve the fixed points and
indices of mappings are discussed, as are the similarities and differences between
the theory of one-, two- and three-dimensional lattices. An account is given of the
effect of periodic boundary conditions in reciprocal space on the dispersion curves

Rep. Prog. Phys., 1970, 33, 533-644


534 J . Killingbeck
for lattice vibrations, with particular reference to the theory of critical points in the
density of states function. Finally, a survey is given of experimental methods
which employ strong magnetic fields to invertigate the topology of the Fermi
surface in metals.

1. Introduction
Since the publication of the article on group theory by Johnston (1960) much
new theoretical and experimental work has been performed in solid state physics,
and there has been a corresponding change in the emphasis placed on various group
theoretical concepts within solid state theory. T h e present article is intended to set
out some of the basic concepts which have come to the forefront of applied group
theory in recent years, and also to give a brief survey of the theoretical and experi-
mental work of the last decade which has led to their emergence. Johnston’s article
sets out very clearly the basic methods of the theory of the unitary representations
of groups, and the reader may regard it as a source from which he can refresh his
memory on the basic points taken for granted in the present article. The recent
work on magnetic groups means that unitary matrix representations are now not
the only ones worthy of attention for finite groups, and we adopt throughout this
paper the terminology (following Melvin) ‘ rep ’ for ‘unitary irreducible representa-
tion’. T h e symmetry properties of functions are of importance in solid state
physics, but their continuity and smoothness are also of interest. The latter
properties are of the type studied in topology, and one part of the present article sets
out some topological concepts which have been used in the consideration of solid
state problems.
Any author attempting to cover such a vast range of material must make some
selection of topics. Probably the best approach is to adopt some definite personal
‘viewpoint’ of the subject, and then to develop this viewpoint, with the review
material being brought in to illustrate appropriate points, rather like beads along
the thread of the argument. In this case the subject matter contains too many
interesting side-tracks for such an approach to be carried out with complete single-
mindedness. However, the general content of the article can be fairly well grasped
by noting that it is based on the viewpoint which we now briefly summarize. If we
accept that quantum mechanics is to be the model governing the dynamics of a solid
state system, then the study of the dynamics naturally leads to the concept of the
symmetry group of the Hamiltonian operator of the system. This group is a group
of operators, but in the solid state case many of the operators correspond to physical
rotations, reflections or translations. Thus a pictorial approach is often useful, and
some of the group relationships have a clear physical interpretation. In practice, the
application of quantum mechanics involves approximation methods, particularly
perturbation theory, and this theory is most commonly used in a form which
involves the matrix of the Hamiltonian operator in some basis set of state functions.
T h e group of the Hamiltonian then determines many of the details of the matrix
structure, and the matrix partitioning technique leads to effective Hamiltonian
operators which have definite symmetry properties, and which act within particular
state vector subspaces. Besides having symmetry properties, the wave function is
required to be continuous in value and in slope in real space, and may also be sup-
posed to be continuous when regarded as a function of some perturbation para-
meter, or of a reciprocal space vector. Periodicity conditions in real or reciprocal
space may also be applied, and thus topological considerations arise naturally, often
Group theory and topology in solid state physics 535

in conjunction with a group theoretical approach. After this general outline, we now
give a more detailed account of the contents of the three parts of the article.
Subsection 2.1 illustrates how the concept of isomorphism means that one
abstract group will serve to describe many different groups of physical operations,
and also comments on the way in which a group may be defined in terms of its
generators. T h e pictorial interpretation of the class concept is illustrated for a
group of rotations, and the pictorial viewpoint is also used in $2.2 in the derivation
of a selection rule for an integral, starting from the character orthogonality relations,
Anti-isomorphism is briefly mentioned in a discussion of the relationship between
active and passive viewpoints. Section 2.3 deals with the notions of direct and semi-
direct products of groups, pointing out the reasons for the widespread use of the
latter in the literature of crystallography. Further discussion of the relevant groups
for the description of crystal lattices is given in $ 2.4, with comments on the difference
between the Bethe (site group) splittings and the Davydov (factor group) splittings
of ionic and molecular energy levels, respectively. Sections 2.5 and 2.6 stress the
operator nature of the elements of the group of the Hamiltonian, and point out that
the operators are naturally regarded as members of an algebra. T h e fact that the
time-reversal operator is a member of the symmetry group for some systems leads
to a discussion of magnetic groups, and the pictorial approach can also be used here
by representing such groups in terms of the black and white groups of Shubnikov.
T h e Hamiltonian operator, and the operators of its symmetry group, are usually
regarded as operating within some subspace of state functions; $2.7 includes a
discussion of double groups, which describe the mathematical symmetry of the
quantum dynamical problem when the electron basis states include spin factors.
Section 2 concludes with a brief mention of the finite groups of high order which
have been used to describe non-rigid molecules.
Subsection 3.1 deals with the way in which the symmetry group affects the
results of approximate calculations which employ perturbation theory and matrix
theory. T h e Lowdin matrix partitioning technique is introduced as a useful way of
linking matrix theory and perturbation theory, and the effective operator approach
is described, with the spin Hamiltonian method ($3.1) and the pseudo-angular-
momentum concept ($3.2) being used as examples. I n $3.1 the Wigner-Eckart
theorem is pointed out as providing the means by which the symmetry group affects
the structure of the secular matrix of the Hamiltonian, and $3.2 deals with the
equivalent operator method, particularly as applied in crystal field theory. I n $ 3.3
the transformation properties of creation and destruction operators in the second
quantization formalism are discussed, and the coupling of pairs of such operators
leads to the construction of the Watanabe seniority operator. T h e Holstein-
Primakoff representation and its relationship to spin-wave theory are briefly treated.
Continuing with the topic of many-electron problems, $3.4 deals with the use of
symmetry-adapted one-electron orbitals in the Hartree-Fock method, and discusses
several modern variants of that method. Methods which use group projection
operator techniques are particularly stressed, including alternant molecular orbital
theory and the band theory of ferromagnetic metals. Subsection 3.5 returns to the
topic of pure group theory, setting out the basic details of the class sum operator
approach to finite group theory; this approach will be particularly appreciated by
physicists who have a working knowledge of the Dirac approach to quantum mech-
anics. T h e last two subsections of $ 3 deal with two subjects for the study of which
group theory is a well-established technique, namely crystal field theory and
536 J. Killingbeck

electronic band theory. T h e discussion of crystal field theory ($3.6) is based on


the symmetry-adapted expansion of the field, and a survey is given of some recent
work involving the orbit-lattice coupling and the Jahn-Teller effect. T h e discussion
of band theory ($3.7) outlines the principles of several approaches to the subject,
using the central idea that a lattice-periodic function has a Fourier series expansion
involving only k-vectors from the reciprocal lattice. T h e effect of screw axis
symmetry is discussed in terms of the secular matrix approach to the determination
of the one-electron energy bands.
The first three subsections of $ 4 deal mainly with graphs. Subsection 4.1 deals
with the basic terminology of graph theory, and uses Euler’s theorem in a discussion
of the possible rotational symmetries of a lattice. Subsection 4.2 describes the free-
electron network model of quantum chemistry, and mentions some other types of
topological wave function. T h e concept of the incidence matrix of a graph is used
in both $$4.1 and 4.2. I t also plays an auxiliary role in $4.3, where a theorem is
given relating to the graphical representation of the terms in a perturbation series,
showing the importance of connected graphs in such an expansion. Subsection 4.4
treats the combination of pairs of rotations, translations, etc., making use of the
concept of the fixed points of a mapping. Subsection 4.5 reviews the similarities
and differences between one-, two- and three-dimensional lattice systems, showing
that there are some quantitative results which are peculiar to one dimension,
while other effects are qualitatively impossible in one dimension. T h e matrix
theorem of Frobenius is invoked to show that long-range order in the Ising
model requires at least two dimensions. Subsection 4.6 outlines the theory of
critical points within the Rrillouin zone for functions which are lattice-periodic in
k-space, and points out that continuity considerations and group theoretical argu-
ments are used in conjunction in that theory. Subsection 4.7 outlines the semi-
classical approach to the motion of electrons in a magnetic field, and shows how the
Onsager relation is applied to interpret the de Haas-van Alphen effect in terms of the
extrema1 areas of the Fermi surface. Magnetoresistance experiments are also men-
tioned, and the anomalous results for ferromagnetic nickel are described. Finally an
account is given of the Kohn effect, in which the phonon frequencies of a metal are
influenced by the shape and extent of the electronic Fermi surface.

2. The r81e of abstract group theory in solid state physics


2.1, Isomorphism and related concepts
Much work has been devoted by mathematicians to the problems of elucidating
the structural details of various types of group. Some of this work has been
motivated by problems arising in physical theory. For example, the theory of
elementary particles has led to the discovery (or rediscovery) of various properties
of the unitary groups and their extensions, while the classification of possible types
of magnetic order in solids has instigated detailed work on the magnetic space
groups. In both the cases mentioned above, it is true that the basic mathematical
theory was to some extent already known before the physical problem arose; never-
theless, the physical problem served to focus attention on specific parts of the theory.
I n the first part of this article, we shall point out some of the useful connections
between the pictorial (or model) viewpoint often taken by experimental physicists
and the more abstract language of pure group theory, and shall review some of the
Group theory and topology in solid state physics 537

recent work in group theory which has been motivated by the problems of solid
state physics.
T h e key concept which links abstract group theory to the symmetry groups of
physics is that of isomorphism between groups. For example, the only possible
abstract group of order 2 consists of an identity element E and an element A such
that A 2 = E , but this group is isomorphic to several different groups of symmetry
operations, e.g. ( E ,2)) ( E ,cr) or ( E ,I ) . Here the symbols 2, U and I denote a rotation
of x , a reflection in some plane, and the space inversion, respectively. By investigat-
ing the detailed multiplication tables of the 32 crystal point groups, it is found that
they fall into families of isomorphic groups. This means that a set of character tables
need be given only for the independent abstract groups involved, provided that the
isomorphisms between the various crystal point groups are displayed. Thus,
Watanabe (1966) gives only 11 distinct character tables, but these serve to describe
21 point groups. Further, by using the concept of the direct product of two groups,
the character tables for the remaining 11 crystal point groups may be derived from
the 11 basic tables given by Watanabe. I t would be possible to use the direct
product concept to reduce even further the number of tables required, since the
groups C,, D, and D,, which are each given a table, are themselves expressible as
products of smaller groups which also correspond to one of the tables. Such a
further saving would be slightly offset by the need to specify the detailed iso-
morphisms involved. Dimmock and Wheeler (1962) give a set of character tables
for the point groups which display the various isomorphisms; they also give a
diagram which shows how the various point groups form subgroups (and invariant
subgroups) of other point groups. A concept which arises in connection with iso-
morphism is that of a one-to-one correspondence between two groups. This concept
sometimes causes confusion, since it is not always realized that a one-to-one corre-
spondence need not be unique. Thus the symmetry group D, consists of the
identity operation plus rotations 2,, 2,) 2, of x about the three Cartesian axes. When
establishing a correspondence between this group and the abstract group of four
elements known traditionally as the Vierergruppe, ( E ,A , B , C), the mapping may
involve the pairing A --f 2,, B -+ 2,, C + 2, or, for example, A -+2,) B -+ 2,) C + 2,.
Both of these correspondences are one-to-one, but they differ. These two alter-
natives could of course be transformed into one another by the correspondence
2, ts 2,, within D,, and this is exactly the correspondence known as an inner homo-
morphism or automorphism for the group D,. T h e discussion so far has been
intended to show that, even when the groups dealt with are interpreted in a direct
physical sense (as rotations or reflections), a knowledge of the abstract concept of
isomorphism can help towards economy in the presentation of information. T h e
group character table was mentioned above, and this table displays another concept
which permits great economy in the presentation of information for non-Abelian
groups. This is the concept of a class function, i.e. some quantity which is the same
for all group elements in a class. T h e character of a class is the best known example
of a class function. We leave this for the moment, however, and pursue the concept
of isomorphism a little further.

2.1.1. Permutation groups. One of the simple basic theorems of group theory is
Cayley’s theorem, which asserts that every finite group is isomorphic to some group
of permutations. I n the case of physical symmetry groups, this can be clearly
displayed by numbering the vertices of some model which has the appropriate
538 J . Killingbeck

symmetry, in such a way that the group operations simply permute the numbered
vertices. T h e theory of permutation groups was one of the earliest parts of group
theory to be developed (Dehn 1960), and it is sometimes useful to treat a group in
its permutation form. T h e most commonly encountered example of this is pro-
vided by the octahedral point group oh, which, for suitably aligned axes, may be
regarded as the group of 48 permutations of the form

%: :p :J 01,

a#P#y.
P,y = x , y ,
When 0, is described in this way, it is very easy to compute the effect of the
(1)

symmetry operations of the group on polynomial functions of type x‘yh xc, and this
fact is useful in crystal field theory (Ballhausen 1962, Watanabe 1966). One way of
checking that the group (2.1) is of order 48 is as follows; if the 5 is dropped from
the three lower symbols, the symmetric group S, of permutations on three symbols
is obtained. This has 3 ! = 6 elements. On including the i: sign, each element of S,
corresponds to 2, = 8 elements of the new group of permutations, and the group
(2.1) accordingly has order 48. I t is not possible to set 48 = n ! for any integer n ,
and so 0, is not isomorphic to any of the full symmetric groups S,. However,
0, is clearly a subgroup of S,, which has 6! = 15 x 48 elements. This example
illustrates Lagrange’s theorem, which asserts that the order of a finite group must be
an integral multiple of the order of any of its subgroups. When the lowering of
symmetry caused by some small distortion of a symmetric body is considered
(e.g. trigonal or tetragonal distortion of a cube), the new symmetry group will be a
subgroup of the starting group, and Lagrange’s theorem is helpful when enumerat-
ing the symmetry elements in the new lower symmetry. We have used the example
of 0, here; Coxeter (1963) points out that the group 0 of pure rotational sym-
metries of a cube is isomorphic to S,, while the group D,, is isomorphic to S,.
Further, the group T, of the symmetry operations of a regular tetrahedron is iso-
morphic to the alternating group A,, while the rotational symmetry group of the
regular icosahedron is isomorphic to A,. (The alternating group A,, is the subgroup
of even permutations of the symmetric group S,, and thus has order & ( n ! ) . )T h e
group D,, is the symmetry group (in three dimensions) of the equilateral triangle;
the isomorphism with S, may be seen pictorially by numbering the three vertices of
the triangle. T h e operations of D311do not permute the Cartesian axes, of course,
but permute three axes with angles of $T between them. If the x axis is taken in the
direction perpendicular to the plane of the triangle, then, in terms of Cartesian
coordinates, the operations of D,, permute the five linearly dependent functions
- 8x i-4J3y, x, i: x ; this description of the group is more cumbersome than the
permutation description of oh, although the use of linearly dependent unit
vectors is common in the crystallographic description of hexagonal lattices (Buerger
1956).
2.1.2. Equivalence relations and classes. I t is well known that the groups studied in
solid state physics represent only a small fraction of the possible abstract group
structures. For example, in a regular crystal lattice with translational symmetry,
the rotational symmetries about any point in the unit cell are described by one of the
crystal point groups. These groups cannot contain, for example, a fivefold or a
sevenfold rotational axis, and so cannot be isomorphic with groups such as C, or
D5.h,or the group of rotations of the regular icosahedron. There are of course
Group theory and topology in solid state physics 539

physical objects, perhaps even isolated molecules, with these point group sym-
metries. T h e proof of the absence of such symmetry elements as fivefold rotational
axes for a lattice can be established by a directly pictorial trigonometric method
(Buerger 1956), by a group theoretical trace method (Heine 1960), or by an approach
based on Euler’s theorem for planar graphs (Ore 1963). For the crystal point groups
the concept of a class of elements can be given a useful pictorial interpretation, as
follows. If two rotation axes (or reflection planes) of the same type within the group
are related by any element of the group, then they belong to the same class of the
group. For example, in the group D,, the three dihedral axes belong to one class,
since the operations of the threefold principal axis permute the dihedral axes. I t is
important to note that if the rotations about two of the dihedral axes, A and B , are

(7

Figure 1. Two of the dihedral axes which belong to the same class of the rotation group D,.
The z-axis is perpendicular to the page, and the rotations 2, = A , 2b = B, 3, = C are as
referred to in equation (2). Rotation C takes axis a into axis b.

pictorially related by the rotation C, through angle +, then the algebraic equation
which describes this is (see figure 1)
B = CAC-1 (2)
which is the usual defining equation for the class relationship. I n general, two
elements A and B of a group belong to the same class if there is some element C of
the group for which (2.2) holds. T h e relationship of belonging to the same class of
a group is an equivalence relation between the elements of the group (Killingbeck
and Cole 1970). Equivalence relations occur in the general theory of sets, and it is a
basic result that the equivalence classes defined by such a relationship are disjoint
and exhaust the group. Thus each element of a group belongs to exactly one class
(and, of course, for an Abelian group, each class contains exactly one element).
Another equivalence relation which may be defined between group elements is that
of belonging to the same left coset with respect to some fixed subgroup of the group.
This shows that the left cosets of a subgroup are disjoint and exhaust the group, a
result used in the proof of Lagrange’s theorem (Ledermann 1969). I n the study of
540 r. Killingbeck
the basic principles of group representation theory, it is often instructive to dis-
tinguish between those concepts of the theory which belong to general set theory,
those which belong to abstract group theory and those which belong to matrix
theory. For example, the concept of an equivalence relation belongs to general set
theory and can usefully be applied both to groups and to matrices. Schur’s lemma,
on the other hand, together with the concept of the reducibility of a family of
matrices, is most accurately regarded as a matrix theorem, since it is applicable even
when the matrices do not form a group representation (Murnaghan 1963).
2.1.3. The rank of a group; group generators. As an example of a simple concept of
abstract group theory which has been little used in solid state physics until recently,
we may quote the concept of the rank of a group. This is simply the minimum
number of generators of a group, i.e. the smallest number of group elements such
that the whole group may be derived by forming products of powers of these
elements. T h e generators of a group are not unique, but the rank is (just as the
basis of a finite dimensional linear space is not unique while the dimensionality is
unique). Gabriel (1964) has suggested a new method of reducing group representa-
tions, and it is an essential part of his method that the matrix representatives of only
the group generators need be known to carry out the reduction. Thus, only a
number of representation matrices equal to the rank of the group need be known.
That Oh, of order 48, has rank 2, was noted by Wachtman and Peiser (1966), who
tabulated the ranks of several point groups in their study of the symmetry con-
ditions applying to jump rates of point defects between equivalent crystal lattice
sites. Zachariasen (1967) gave three generators for 0,1,whereas Wachtman and
Peiser (1966) noted that the two generators 4, and m (reflection in the (110) plane)
will suffice. T h e product m4, m is equal to 4,, and 4,, 4, are generators for the
group 0 (Gabriel 1964). T h e group 0 contains a twofold rotation 2, about an axis
perpendicular to the (1 10) plane, and the product 2, m is the inversion I . Thus the
whole of 0,1may be generated by (4,, m). T h e group 0, = 0 x ( E , I ) thus has the
same rank as 0. When the direct product A x 5 of two Abelian groups is formed,
however, the rank of A x 5 equals the sum of the ranks of A and 5 (e.g. D, = C, x Ch
has rank 2 = 1 + 1). This is most easily seen by using the additive convention for
the combination of group elements (Killingbeck and Cole 1970), noting that the
number of generators is then equivalent to the dimensionality of a linear vector
space. T h e group 0 has six equivalent axes involving rotations of the type 2, intro-
duced above. That it can have no more purely twofold axes follows from the
relation s2 = s4+ 3s, + 3 (Heine 1960, Zachariasen 1967) which relates the numbers
s, of rotation axes of order n for proper noncyclic point groups. Olbrychski (1963 a)
pointed out that the complete structure of a group G ( A ) may be concisely specified
by giving the generators A,, .. . , A,, of the group, together with a minimum set of
relationships which may be written in the form

(2.3)
where E is the identity element of the group. A clear account of the generator-
relationship method of defining a group is given by Grossman and Magnus (1964),
who employ a graphical approach (originally due to Cayley) in order to render the
treatment more readily visualized. The relationships (2.3) may be used to build up
the whole of the group multiplication table. A set of m matrices which are irreduc-
ible and which obey the relations (2.3) will generate, by matrix multiplication, an
Group theory and topology in solid state physics 541

irreducible representation of the elements Aj. If a second group G(B) has m


generators B,, ..., B,, which obey the same relations (2.3) as those obeyed by the
elements A j , plus some extra relationships of the same type, then the groups G ( A )
and G(B)are homomorphic. Olbrychski (1963 a, b) has used the above properties
of group generators to construct the irreducible representations of several space
groups.

2.2. Gsoup characters and product representations


2.2.1. Symmetrized Kronecker products of reps. T h e group character table is
probably the group theoretical tool most widely employed by solid state physicists.
One of the useful theorems employed in connection with the table is the theorem
which asserts that the characters of a product representation are the products of the
characters arising from each representation in the product separately. If Fl and rz
are reps of a group, then this theorem allows rlx rzto be decomposed in terms of the
reps of the group. When dealing with electronic orbitals which are labelled by the
Fj, however, the antisymmetry principle introduces some extra problems. Consider
two electrons, which can occupy any one of the orbitals #, of some degenerate rep I?.
T h e two-electron wave function, involving sums of product functions of type
#,(1) 4 ~ ~ ( will
2 ) , be a spin singlet or triplet. T h e singlet states must have a permuta-
tion symmetric orbital part, while the triplet states must have a permutation anti-
symmetric orbital part. T h e space of the product functions can be split up into two
subspaces, one containing symmetric functions of form Gn7(1)#,(2) + #,(2) #%(l),
and the other containing antisymmetric functions of form #,(1) #,(2) -&%(2) #,(l).
Since the permutation 1 .tj2 commutes with every element of the spacial symmetry
group G associated with the I' labels, the symmetric and antisymmetric subspaces
are invariant under the group operations. The representations of the group G
which are provided by the symmetric and antisymmetric subspaces are termed the
symmetrized square [PIz and the antisymmetrized square respectively, of the rep
r. The character of the symmetrized square is equal to

and can thus be found if the characters x of F are known. Since the sum of the
symmetrized and antisymmetrized characters must equal [x(R)I2, which is the
character of I ' x I?, it follows that {x}~(R)is obtained by changing the sign of the
second term in (2.4). When products of three one-electron orbitals are taken, these
can (after symmetrization) be classified as even or odd or as belonging to the two-
dimensional rep of s,, the permutation group on three symbols. T h e representations
of the group which arise from the subspaces corresponding to the two one-dimen-
sional reps of S, give the symmetrized and antisymmetrized cubes, [I?], and
of the rep F. T h e functions associated with {F}, are appropriate orbital components
for those states with total spin S = p. Since there are no totally antisymmetric
functions which can be formed by combining triple products of one-electron spin
functions, it follows that the orbital functions associated with [PI3cannot be given an
analogous interpretation. In fact, they correspond to no allowed three-electron
state. T h e three-electron states with S = 4 involve orbital functions from the sub-
space associated with the two-dimensional rep of s,, in accord with the principle
that the orbital and spin parts of the total function must belong to adjoint Young
tableaux of S, (Heine 1960, Judd 1963).
5 42 J . Killingbeck

The representations [TI3 and (rJ3 are of interest in connection with the theory of
magnetic phase transitions in crystal lattices (Dimmock 1963, Cracknell and
Joshua 1968). Arguments involving the lattice free energy suggest that at a magnetic
second-order phase transition the symmetry group G of the lattice (usually a
magnetic group, of the type discussed later) changes to a subgroup G of the original
group. Furthermore, the new symmetry can belong to one rep 'I of the original
group, provided that the rep r obeys certain conditions, which we do not consider in
detail here. However, we note that the criteria involve [r]Rand These two
representations have been decomposed in terms of the reps Fi for all the point
groups by Cracknell and Joshua (1968). While their work was motivated by the
phase transition theory, their results can be used in the classification and enumera-
tion of two-electron eigenfunctions. Symmetrized and antisymmetrized products
of reps are also used to give the selection rules which determine those electronic
states which undergo a Jahn-Teller coupling to vibrational modes of various
symmetry types (Ham 1967). This is an example of a general fact; group theoretical
results worked out in connection with one physical problem can often be inter-
preted so as to give useful results in other physical problems. The formal reason for
this is usually the one stressed in the first section of this article ; the symmetry groups
of various physical problems are isomorphic to each other by virtue of being iso-
morphic to the same abstract group. T h e general problem of the coupling of N
equivalent electrons belonging to one rep of a finite group has been studied group
theoretically by Goscinski and o h r n (1968). Some of the examples given at the end
of their paper are wrong, however, since they have omitted to make the Young
tableaux of spin and orbital functions adjoint to each other.
2.2.2. A selection rule f o r inte<grals. One basic result needed in the construction of
the group character table is the character orthogonality theorem for unitary reps of
a finite group,
R
cxp*(R)X"R) = hs,,. (2.5)

Here the superscripts label the reps, R is a typical group element, and h is the order
of the group. I n addition to (2.5), there is also an orthogonality relation involving
summation over reps

where M, N are class labels, and n,, nN are the number of elements in the classes
N and M respectively. Equations (2.5) and (2.6) are sometimes called the character
orthogonality relations of jifirst and second kinds, respectively. Their similarity
suggests a duality between the properties of the reps and classes of a finite group,
and this duality has been investigated in detail by Gamba (1968). I n this para-
graph we wish to show how the relations (2.5) lead to an instructive physical picture
of the group theoretical selection rules for matrix elements. If the identity rep is
used as one of the reps in (2.5), the following result is obtained
cx"w = 0
R
(2.7)
where ( p ) is any rep other than the identity rep. Consider the integral
Group theory and topology in solid state physics 543

where +Wbelongs to some rep of a group of spatial symmetry operators. We take


( p ) to be one-dimensional at the moment, and take the group operations R to be the
rotations or reflections of some point group (see figure 2). If in the integral
( 3 . 5 7 ) is visualized as a scalar function with an assigned value at each point of space,
then it will follow as a ‘physical invariance principle’ that I(fi)is unchanged in value
under any change of co-ordinates which involves a nonvanishing Jacobian but no
change in the length unit. T h e set of such coordinate transformations is itself an
infinite group, of which the operations of G , regarded as coordinate transformations,
will form a finite subgroup. T h e use of information concerning the effect of the
operations of G on the integral can reveal useful facts about the value of I ( P ) ;any

Figure 2. A typical star of volume elements for the group C, as referred to in the discussion
of the integral 1‘”. R is here the operation 4,.

further properties of I ( p )must be classified as accidental if they do not specifically


arise from the symmetries of G. Consider a small volume element d r which is
shaped like the traditional ‘ curved parallelopiped ’ used in integration over spherical
polar coordinates. If the operation R is applied to this element, a congruent new
volume element results which we call R d7. Using all the elements of the group in
this way gives a set of h volume elements, h being the order of the group. T h e
integrand in the element R d r is equal to x @ ) ( Rtimes ) the integrand in dT, since the
integrand belongs to the rep ( p ) . Equation (7) then shows that the total con-
tribution of the h volume elements to I @ )is zero. The set of h volume elements may
be termed the star of the starting volume element, and this terminology is in line
with that used in space group theory (Falicov 1966) except that we are dealing with
real space here, and not with k-space or an abstract representation space (Altmann
1963). If we regard the whole of space as partitioned into stars, it follows that the
integral I ( / )is zero if ( p ) is not the identity rep. There will be some special points,
lying on symmetry axes, for which the star of dr contains less than h distinct volume
elements. However, these constitute a set of measure zero and will not affect the
5 44 J . Killingbeck

value of I ( 1 )if yW is finite everywhere. We have used a one-dimensional rep in our


argument here. For a degenerate rep the full representation orthogonality relations
would be needed instead of the character orthogonality relations, but the principle is
unchanged. T h e space may be broken up into a set of regions which give zero
contribution to 1'1)for symmetry reasons.
2.2.3. Actiue and passive uiewpoints. T h e argument given above is probably more
intuitively meaningful to a physicist who thinks pictorially than is the usual approach.
T h e latter approach invokes the principle of invariance, and shows that 10')is zero
by comparing its values in different frames, whereas the argument given above
sticks to one frame throughout. T h e relationship between the two approaches is
essentially that between active and passize views of symmetry operations, e.g. a
rotation R of the space corresponds to a rotation A-l of the axes. The argument
given above is not affected even if we are careless about the convention; Cx(R) and
Cx(R-l)are identical, since a group contains the inverse of every one of its elements.
This group property, together with the property x("*(R) = X ( ~ ) ( Rof- ~a) finite
unitary group, means that correct results are often obtained even when we are
careless about the active-passive distinction. This is particularly so if the calcula-
tion involves a sum over group elements. While perhaps convenient, this fact makes
the exceptional cases appear even more puzzling to a worker who has mixed the
conventions somewhere in his calculation. The essential property needed is that
the viewpoint adopted shall provide a set of matrices M(A)which properly represent
the group, i.e. for which

M(A) M(B) = M(AB). (2.9)


If the active-passive distinction is not treated carefully, it is possible to obtain
matrices for which
M(A)M(B) = M(BA) (2.9~)
and these do not form a representation if the group is non-Abelian (Killingbeck and
Cole 1970). T h e importance of a careful choice of convention for the way in which
operators act on coordinates and wave functions is stressed by Altmann and Bradley
(1965 a), who criticize the convention chosen by Slater (1965 a) in his paper on the
electronic wave functions in crystal lattices.
Bouten (1969) discusses the case of the rotation group R,, and points out that
some of the standard texts are in error in their convention for rotation operators.
In particular, the prescription for the rotated field R# at point r
( r ) = $(R-l r ) (2.10)
which describes the active rotation of a scalar field, leads to a group of matrices
M(R) which is isomorphic to the group of spatial rotations, whereas the passive
viewpoint leads to matrices N ( R ) which obey equation (2.10). Routen terms the
two matrix groups anti-isomorphic; it is clear on considering the physical meaning
of the active and passive viewpoints that each matrix M(R) is the inverse of the
corresponding N(R).
Formulae such as (2.5) and (2.6) are commonly used, and are explicitly relevant
to unitary matrix representations, which can always be constructed for a finite group.
I n fact, this point is usually taken care of automatically in solid state physics, since
the basis sets used to set up operator matrices are usually orthonormal; the rotation,
Group theory and topology in solid state physics 545

translation and reflection operators are unitary operators, and automatically give
unitary matrices when used with an orthonormal basis (within an invariant sub-
space, of course). T h e anti-unitary time-reversal operator is an exception and is
discussed later.

2.2.4. Some theorems concerning nodes. T h e result

I$(/) dT = 0 ( p )# identity rep (2.11)

shows that, when the orbitals of an electron are classified according to reps of some
group, they give zero when integrated over all space (unless they belong to the
identity rep). This result applies to the three-dimensional rotation group as well as
to finite point groups, and shows, e.g., that p, d, f, ... type atomic orbitals give zero
l2
when integrated over all space. When the integral of [ $ ( / ) is formed, however, it
always contains some identity rep component, and this component gives the nonzero
normalization integral. If we have a Hamiltonian H with symmetry group G , and
are given a real nodeless eigenfunction $ of H , it follows that the integral of t,h over
all space is nonzero. According to (2.11) this means that $ has a component $(l)
belonging to the identity rep of G. Further we have, if h is the eigenvalue associated
with $,

since H does not change the rep label of a function. T h e second and fourth members
of (2.12) imply that $(I) is an eigenfunction of H ; in fact, this must be so, since
otherwise $ could not be an eigenfunction of H , as was assumed. T h e theorem just
sketched here has been used by Kleiner and Kaplan (1969) to investigate the
symmetry of the ground state of certain Hamiltonians. Mattis (1965) has discussed
several theorems relating to the nodeless properties of ground states of two-electron
systems. One of these theorems involves replacing $ by [ $1; since the integral of
1 $ 1 is nonzero, I $1 necessarily contains a component belonging to the identity rep
A , of the group of H . When $ is the two-electron ground state function for, say,
the helium atom, and H is the helium atom Hamiltonian, the expectation value
( I $ 11 HI1 II, I ) equals <t,b I HI $). According to the variational principle, however, the
first expectation value should exceed the second. The conclusion is that $ is
nodeless, so that $ and 141 are identical. Further, $ is an orbital S state, belonging
to the identity rep of the three-dimensional rotation group, in the case of the helium
atom. This can be seen by applying to $ the theorem derived above in connection
with equations (2.11) and (2.12). However, as pointed out by Kleiner and Kaplan
(1969), it is not necessarily true that the ground state of a system with more than
two electrons has A , symmetry. This is because the many-fermion ground state is
then not the mathematical ground state of H , owing to the effect of the Pauli
principle, which prevents the lowest mathematical eigenfunctions of H from being
occupied. Hegstrom and Lipscomb (1968) have shown that this effect means that
a closed shell molecular wave function need not necessarily be diamagnetic when
more than one electron pair is present.
Riess (1970) has given a detailed discussion of the nodal structure of the wave-
function for many particle systems.
546 J. Killingbeck

2.3. Group products and inoariant subgroups


2.3.1. Direct product groups. T h e study of the group structure or the matrix
representation theory of a large group is considerably simplified if the group can
be regarded as some kind of product of smaller groups. The properties of the large
group can then be deduced from the properties of the smaller groups from which it
is compounded. T h e best known example of a group product is the direct product of
two groups. If the first group G has elements g j and the second group H has
elements hk then the direct product group G x H may be formed provided that each
g j commutes with every hk. G x H then consists of the symbolic pairs ( g j ,h,) (the
so-called Cartesian product of the groups, in set theoretic terminology) with the
multiplication rule
(2.13)
I n the case of an abstract group, the rule (2.13) (or the stipulation that the g j and hk
commute) is a rather artificial-seeming construction from the physicist’s point of
view. However, when the groups G and H are interpreted as groups of symmetry
operations for a physical system, it becomes directly meaningful to ask whether the
operations of G and those of H commute, The symbolic pair ( g j ,h k ) is then inter-
preted as the operator product gjhk, and a direct product group formed from the
operators in this way will be isomorphic to an abstract direct product group. In
modern abstract algebra this type of product is sometimes called an inner direct
product group, to distinguish it from an outer product group, which may be formed
according to the prescription (2.13) even when no assumption is made about the
commutativity or non-commutativity of the elements of G and H (Fraleigh 1967).
A simple but important example is provided by a holohedric crystal point group,
which is the direct product of a group of pure rotations and the group C, consisting
of the identity operator and the space inversion operator I ( I commutes with any
rotation). For example, 0, is the direct product 0 x Ci. I t is possible to suppress
some of the operations of a hololedric group and still be left with a group, I not being
present as a distinct element in the new group. T h e new group is termed hemihedric,
and one useful property of such a group is that it must always have exactly half of
its elements in the product form IR, R being a pure rotation. Since a hemihedric
group does not contain I , it cannot be regarded as a direct product of Ci with a
group of pure rotations. (That it can be so regarded is implied by a comment of
Knox and Gold (1964).)

2.3.2. Semi-direct products, inoariant subgroups. It is worth while here to point out
a difference between the terminology of standard group theory, as used by non-
crystallographers, and the terminology sometimes used by writers on crystallo-
graphy. T h e latter often use the term “direct product’’ (Zachariasen 1967) or
“group product” (Buerger 1956) to denote a quantity which is known otherwise as a
semi-direct product, and which does not in general coincide with the direct product as
defined above. T h e semi-direct product can arise as follows. Suppose, for example,
that we have a group 6 = C, consisting of the operations of an n-fold rotational
axis. If we introduce the group Ci with axis perpendicular to the first axis, then the
new operation, which we denote by 2,, does not necessarily commute with the
operations of C,. However, it is still the case that the relation
2, G2,-I = G (2.14)
Group theory and topology in solid state physics 547

will hold. This may be rewritten as


2,G = G2,. (2.15)
I t is important to note that (2.15) does not necessarily imply that 2, commutes with
all the individual elements of G, although the latter property does imply (2.15). If
the elements of C, and the operations (E,2,) of the twofold axis group Ci are
multiplied together to form the set of all possible products
uj b, ( U j € c,, b, € c;) (2.16)
then a group results. This group is the dihedral group D, for our example (G = C,),
and is not necessarily Abelian, whereas the direct product of two Abelian groups
would be Abelian. T h e new group is called the semi-direct product of C,, and Cg.
T h e semi-direct product G A G ‘ can be defined for any two groups G and G‘,
provided that (2.14) holds when 2, is replaced by each element of G: in turn.
Altmann (1963) has shown how to express the larger point groups as direct or semi-
direct products of smaller point groups, thereby simplifying the theory of the larger
groups. His work is essentially a ‘feedback’ into point group theory of a method
which has long been used in the theory of space groups, and which gave rise to the
crystallographic terminology mentioned above. T h e space group of a crystal lattice
includes both translational and rotational symmetry operations ; the subgroup T of
purely translational symmetries obeys the same relationship as that expressed by
equation (2.14), namely
RTR-l= T (2.17)
where R is any element, rotational or translational, of the space group. T h e result
is very clear when R is a translation, i.e. RET, since T is Abelian. When R is a
rotation, the physical picture already explained in connection with the class concept
shows that RtR-l is a translational symmetry operation obtained from the transla-
tion t by applying rotation R to t ( t being treated as a vector). T is called an invariant
subgroup of the space group, and it is clear that this concept is closely related to that
of a semi-direct product. I n fact, from its definition, G is always an invariant
subgroup of the group G A G ’ . T h e method described by Altmann (1963) for
obtaining the reps of a point group G A G’ from those of G and G’ is essentially the
same as that commonly used for the construction of the reps of a space group by
starting from Bloch functions with given k vectors (e.g. Koster 1937, Falicov1966).
A space group cannot be regarded as a direct product, which accounts for the
precedence given to the concept of the semi-direct product in crystallography. I n
fact, only some of the space groups are semi-direct products of a lattice translation
group T with a crystal point group, and such space groups are termed symmorphic.
Seventy-three of the traditional 230 space groups are symmorphic ; they include, for
example, the space groups of the simple cubic (sc) (oh’)) face-centred cubic (fcc)
(OhT))and body-centred cubic (bcc) (Ob9)lattices, the reps of which were originally
studied by Rouckaert et al. (1936)) and the space group of the zinc blende (Td2)
lattice, the reps of which were studied by Parmenter (1955). The remaining 157
groups, the non-symmorphic space groups, contain symmetry elements such as
screw rotation axes or glide reflection planes. T h e non-symmorphic groups include
the space group Oh7 of the diamond lattice, and the space group DGh4of the hexa-
gonal close packed lattice, both of which were discussed by Herring (1942), as well
as the space group c 2 h 5 of the molecular crystal naphthalene (Davydov 1962), and
the space group D4h14of the rutile structure, the reps of which were most recently
548 J . Killingbeck

discussed by Gay et al. (1968). It is possible to find apparent screw axis operations
even for a symmorphic group (Falicov 1966) ; the important distinction between
symmorphic and non-symmorphic groups is that in the former it is always possible
to choose an origin in the lattice such that no symmetry operations involve frac-
tional lattice translations, whereas no such origin can be found in a non-symmorphic
lattice. T h e non-symmorphic diamond lattice is obtained by making identical the
two types of atom in the unit cell of the symmorphic zinc blende lattice; the
Kronecker products of the reps for both T,* and 0,17 space groups have been
given by Birman (1962).

2.3.3. Factor groups. In general the notation ( R I t ) is used for a space group
operation, the rotation R about the origin being followed by the translation t .
If the rotation (RIO)is a symmetry operation about the origin, so will be
( E I t,) (RI 0) (El tn)-l,since it is the same rotation about a translationally equivalent
site. Since the lattice translations form an invariant subgroup, the triple product
can be expressed as the product ( R IO) (Eltin) for some lattice translation th. By
similar arguments the whole of the space group G may be expressed as the semi-
direct .product T AG/T. T h e zinc blende lattice provides an important example.
I t consists of two fcc lattices with a spacing equal to one quarter of the cube diagonal.
If an atom A at (O,O, 0) is taken as origin, than the rotation 2,(B) about the atom B
at (t,2,a) may be written as 2,(B) = (2,(A) I t,,,) where t,,, is a lattice translation for
the fcc lattice. In a similar way all the elements of G may be referred to origin A.
T h e resulting factor group is isomorphic to T, and the site symmetry at both sites
A and B is also T,. If the A and B type atoms are identical, a fourfold screw axis
symmetry appears, and we have the diamond lattice. If a point half-way between A
and B is used as a new origin, the diamond lattice has inversion symmetry about this
origin (Mariot 1962, Kittel 1963). The factor group for the diamond lattice is
isomorphic to 0, (Mariot 1962). Nonsymmorphic groups may be regarded as semi-
direct products of the appropriate translation group with a group which is not a
crystal point group, although it is isomorphic with a crystal point group. This
group with the abstract structure of one of the crystal point groups is the factor-
group G / T of the space group G with respect to the invariant subgroup T of
translations. Zachariasen (1967), in his account of the space groups, expresses them
as semi-direct products of translation groups and factor groups, and further decom-
poses the factor groups into products of smaller groups. Zachariasen also gives the
expressions for the direct and semi-direct products of the various point groups.
(As another matter of terminology, it should be noted that Zachariasen uses the
terms “isomorphism” for our “homomorphism” and “simple isomorphism ” for
our “isomorphism”.) If the group T of lattice translations has order fV (owing to
the imposition of cyclic boundary conditions) then there is an N-to-one homo-
morphism between the group G and the group G/T. Identical rotations about sites
related by a lattice translation belong to the same coset and are thus absorbed into
one element of G / T ; this element can thus be conveniently labelled by the rotation
concerned, possibly together with some associated translation. T h e invariant sub-
group T of G is the kernel of the homomorphism G ++GIT, i.e. the set of elements
of G which are mapped into the unit element of G/T. This view of an invariant
subgroup as the kernel of a homomorphism is stressed by Birkhoff and MacLane
(1953), and is useful in the treatment of magnetic space groups (Bradley and Davies
1968). T h e usual approach to the concept of factor group involves the formation of
Group theory and topology in solid state physics 549

the cosets of an invariant subgroup. From a pedagogical point of view, it is instruc-


tive to proceed from the direct product to the semi-direct product, and then to the
factor group concept, as we have here. An invariant subgroup H of a group G is one
for which a relationship analogous to (17) is obeyed,

gHg-l = H (all gE G) (2.18)


g being an arbitrary element of G. T h e left (or right) cosets of G formed with
respect to H then form a group denoted by G / H and called the factor group of G
with respect to H. I t is a simplification to note that the multiplication table of G / H
need not be obtained by fully multiplying out the complete cosets, but is obtainable
by multiplying together only arbitrary representative elements from each coset
(Fraleigh 1967), once the elements belonging to each coset have been tabulated.
Equation (2.18) is similar to the defining equation for a class, and in general an
invariant subgroup contains a whole number of classes. One fact which is often
useful in the study of crystal point groups is the following; any subgroup of index 2
(i.e. a subgroup with half as many elements as the group from which it arises) is an
invariant subgroup, For example, the group C, of the threefold axis in D, is an
invariant subgroup of D,, and this is represented pictorially by the fact that the
rotations associated with the dihedral axis simply permute the operations of the
principal axis. This permutation property of an invariant subgroup follows from
(2.15). T h e factor group of D, with respect to C, is a group isomorphic to C,.
(Note the abstract nature of this result; when forming C,A Ci = D,, the group CL
was taken as a specific physical group.) Since the concept of the semi-direct
product appeared explicitly in the solid state literature only comparatively recently,
this section has stressed the semi-direct product concept rather than the better
known factor group concept. It is worth while to point out that the two concepts
cannot be used interchangeably in every discussion. For example, if the semi-
direct product G A G’ can be defined for two groups G and G’ (i.e. if the generalized
form of the criteria of equation (2.14) are satisfied) then it follows that G is an
invariant subgroup of G A G’, and also that the factor group G A G’/G is isomorphic
to G’. However, as pointed out by Olbrychski (1963 a), the statement that a group
F is such that FIG is isomorphic to G’ does not uniquely determine F. G A G’ would
obey this requirement, but so in general would several different and non-isomorphic
groups. A group F I+-hichobeys the above criteria is said to be an extension of the
group G by the group G’. As an example, consider the group C, with elements
(E,4,,2,,4,3). The subgroup (E,2,) has index 2 and is necessarily invariant. T h e
two cosets (E,2,) and (4,,4,,) are the elements of the factor group concerned, and
this group is isomorphic to C, (note that the coset multiplication procedure involves
the rule that multiply appearing elements in the product should be counted only
once). T h e group D,, of order 4, is not isomorphic to C,, and may be written as
( E ,2,, 2,, 2,). T h e invariant subgroup ( E ,2,) then yields the factor group consisting
of the cosets (E,&) and (2,,2,), this factor group also being isomorphic to C,.
Thus both C, and D, may be called extensions of C, by C,, and are non-isomorphic.
Note that the set {4,, 4,,} is not a group.
2.3.4. Solwability and decomposability. When a group is a direct product A x B the
reps of the group follow at once by taking Kronecker products of the reps of A and
B. When a group is a semi-direct product A A B (the invariant subgroup is men-
tioned first by convention), the reps of the group follow by considering the effect
550 J. Killingbeck

of the operators of B on the basis vectors of the various reps of A. Note here that a
definite physical picture of the group elements as operators is used, although the
resulting reps are equally valid for the abstract group isomorphic to A A B. When
A A B is a space group and A is T, the appropriate lattice translation group, the
method described here is exactly that which involves applying the operations of B
to a Bloch function basis. T h e concepts of the star and the little group of a k-vector
arise in this method, which is described in many texts (e.g. Koster 1957, Falicov
1966). As mentioned above, Altmann (1963) extended these concepts in order to
develop an analogous method of treating the reps of those point groups which may
be written as semi-direct products of smaller point groups. Zak (1960) has pointed
out that the method may be extended further within space group theory itself. H e
showed that all the 230 space groups have invariant subgroups of index 2 or 3, and
that the reps of a space group may be systematically derived from those of such an
invariant subgroup. I n particular, the reps of the non-symmorphic space groups,
which are in general harder to obtain than those of the symmorphic groups, can be
obtained by step-wise repetition of this decomposition process until the last
invariant subgroup reached is a known symmorphic space group.
The basic theorems involved in the process of building up reps of a group from
those of related smaller groups are given in a classic paper by Seitz (1936). I n
abstract group theory, a group is called simple if it has no invariant subgroups, and
semi-simple if it has no Abelian invariant subgroups. The space groups clearly do
not satisfy either criterion; in fact, as Seitz pointed out, the space groups have the
property that it is possible to form a sequence of groups of decreasing order
G , G , , ..., T (2.19)
in which G is the space group, T is the translation subgroup and the factor group
Gj,'Gj+l is an Abelian group of prime order. This property arises from the fact
that the crystal point groups are solvable, i.e. they show the same property as above,
except that the last member of the sequence is the identity operator and the Gj are
point groups, I n fact, the smallest nonsolvable group is the alternating group A,,
of order 60 (Fraleigh 1967). As mentioned previously, A, is isomorphic to the
rotational symmetry group of the icosahedron, which is not a crystal point group,
since it includes fivefold axes. T h e concept of solvability is related to that of
semi-direct product; an analogous concept involving the direct product is that of
decomposability. A group is decomposable if it is isomorphic to the direct product
of two of its proper subgroups. I n the case of an Abelian group, the decomposition
can be continued until the group is expressed in the form of a direct product of
cyclic groups,
G = C, x CA,,,'x . . . (2.20)
where iW,Ll is an integer multiple of Mj. T h e integers M j are called the torsion
coeBcients of G . ,4s a simple example, the Abelian group D, is a direct product
C, x Ck; to visualize this physically, the axis of C; must be set perpendicular to
that of C,, so that the operations of the two groups do indeed commute. If the
Abelian group G is infinite, it will be necessary to add a number of factors of type
C, at the end of the product, C, representing the cyclic group of infinite order.
T h e number of these factors is called the Betti number of G . For example, the
translation group of a three-dimensional infinite lattice has a Betti number of 3.
T h e finite group of translations of a lattice with cyclic boundary condition has Betti
number zero, as does any finite Abelian group.
Group theory and topology in solid state physics 551

2.4. Point groups, site groups and factor groups for a crystal lattice
T h e symmetry group of a regular crystal lattice is a space group and includes
translations, rotations about various points in each unit cell and, for non-sym-
morphic space groups, screw axis and glide plane operations. When an electronic
or vibrational eigenvalue problem is solved for the regular lattice, the translational
invariant subgroup symmetries are taken care of first, and produce k-vector labels
for the resulting eigenvectors. The result is an eigenvalue problem for one unit cell,
with an effective Hamiltonian or dynamical matrix which depends on k . Various
symmetry groups have been defined in the recent literature in connection with the
latter part of the problem, and it is useful to note the similarities and differences
between these groups. T h e factor group of a space group with respect to the
translation subgroup has already been mentioned. T h e factor group is isomorphic
to some point group, but need not be identical with it. For example, Mariot (1962)
gives the factor group of the diamond lattice; this factor group includes screw axis
operations, but is isomorphic to the point group Oil. Thepointgyoup associated with
a space group is defined to be the set of rotational operations which appear in the
space group operations, i.e. essentially the space group elements with any transla-
tional parts omitted. I t is this group to which the factor group is isomorphic.
Kopelmann (1967) defined a unit cellgroup as the group consisting of the elements of
the space group modulo primitive translations, and pointed out that this group is
isomorphic to, but not always identical to, the point group and the factor group.
Each particular site in the unit cell has a site symmetry, consisting of rotations and
reflections at that point, which leave invariant the rest of the lattice. T h e site
symmetry at each site in the unit cell is a subgroup of the factor group, and the
factor group contains interchange operations which exchange equivalent sites in the
unit cell. T h e theory of these interchange operations has been developed in detail by
Kopelmann (1967) in connection with the theory of excitons in molecular crystals.
T h e site symmetry group is the group involved in the Bethe (1929) theory of the
splitting of the degenerate energy levels of an ion in a crystal. According to the
Bethe theory, a singly degenerate level of the ion cannot split further in the site
group symmetry. However, if the unit cell contains more than one of these identical
ions, an apparent splitting can occur. For example, a weak interaction between two
ions in the unit cell will produce two two-ion levels from the identical singly
degenerate levels at each ion. T h e small splitting between these two levels is called
a Davydov splitting or factor-group splitting and is treated in the book by Davydov
(1962), where molecular crystals are the topic of discussion. Schaak (1963) applied
the Davydov theory to the vibrational spectrum of CaS0,.2H20. When the total
wave function has k = 0 in the space group labelling, it follows that the behaviour
in every unit cell is identical, and at k = 0 the states which show a Davydov splitting
can be labelled according to the factor group symmetry. For those optical transi-
tions in which the crystal ground state before absorption has k = 0, and the wave-
length of the light is long compared with the lattice spacing, then the crystal state
after absorption has k e 0, and the selection rules for absorption may be fairly well
described in terms of the factor group theory. Winston (1951) investigated the way
in which the factor group reps can be found once the site symmetry reps are given
for the sites in the unit cell, and applied his results to give a partial explanation of
additional lines in the Eu3'(EuF3) spectrum which could not be explained by the
single ion site group (Bethe) splitting theory. T h e presence of four Eu3+ ions per
unit cell leads to the extra factor group (Davydov) splitting. Winston's work
552 J . Killingbeck

treated only the special case in which the site symmetry reps involved are of the
identity (ill)type; this is appropriate for molecular crystals, in which the individual
molecules usually have an A, ground state, but is usually not appropriate for para-
magnetic ions in crystals. Schulz (1961) derived Winston’s results within a more
general treatment based on group algebra techniques, and later (Schulz 1962)
extended the theory to deal with cases in which the site states are not of A, type.
His second paper uses character analysis to show how the qualitative energy level
spectrum for a pair of neighbouring rare-earth ions may be derived from a know-
ledge of the Bethe energy level spectrum of the individual ions. Richman et al.
(1963) gave an account of the factor group theory as applied to lattice vibrations,
and showed that the factor group (k = 0 ) classification gave a fairly good account of
the many vibronic lines observed in the polarised spectra of the rare-earth ion
Pr3f in several rare-earth trihalide lattices.
2.4.1. Cyclic boundary conditions. T h e use of cyclic boundary conditions for a
crystal lattice leads to a finite group of lattice symmetry operations. Cyclic boundary
conditions are nowadays used without comment in most of the literature, although
their validity and significance were originally a matter of some controversy. From a
physical point of view it is to be expected that lattice waves with very long wave-
length will be most affected by crystal size or shape; however, it is exactly for this
doubtful region k N 0 that the standard theory is used to interpret optical absorption
by lattice vibrations! This important point has been investigated by Englman and
Ruppin (1968). The effect of finite crystal size on the long-wavelength modes is
most marked for ionic crystals, in which long-range Coulomb forces are present,
and the general result is that the absorption spectrum can contain extra lines which
are not predicted by the usual theory. Englman and Ruppin (1968) refer to experi-
mental work which appears to show a shape dependence of the optical absorption
spectrum of a crystal lattice.

2.5. Commuting operators; the group algebra


While the symmetry groups of solid state physics are often directly visualized
as the symmetries of a physical model (for example, a regular lattice with spherical
ions at each site), the basic group studied in quantum mechanics is the symmetry
group of the Hamiltonian operator H . Solid state physics is a subject in which
group theoretical arguments can be readily visualized, but there are still some
exceptions. While it is generally true to say that any symmetry operation derived
from a pictorial model of a system will correspond to a member of the symmetry
group of H , there are some group members, such as space inversion or time
reversal, which cannot be simply visualized in terms of a model employing rigid
objects. The formal procedure most meaningful to those physicists who are used
to the Dirac version of quantum mechanics (Dirac 1958) is to define the group of H ,
G ( H ) as the set of operators which commute with H. Those operators which
correspond to rotations, translations or reflections are linear operators, while the
time-reversal operator is anti-linear. Since in general the symmetry operations of
solid state physics possess inverses, and since the identity operation can be included
as a trivial symmetry operation, it follows that the group of H is a group in the formal
sense. T h e closure property follows on noting that the following operator identity
holds,
[ H , A B ]= [ H , A ] B + A [ H , B ] (2.21)
Group theory and topology in solid state physics 553

so that AB belongs to G(H)if A and B do so. The usual theory implicitly contains
the restriction that only products of symmetry operators are to be considered.
However, if A and B commute with H then so does A + z B , where z is a complex
number. Thus the set of all operators which commute with H is strictly speaking
an algebra. Since the addition of linear operators is a natural concept in quantum
mechanics, the theory of groups of operators involves the group algebra as well as
the group itself. Lowdin (1967) has reviewed the theory of group algebras and its
connection with quantum mechanics; one obvious fact is that the rep projection
operators of finite group theory are particular members of the group algebra which
have the special property of idempotency. T h e paper by Schulz (1961), already
mentioned in the preceding section of this article, sets out the basic properties of the
group algebra, and uses the group algebra formalism to discuss factor group
splittings.

Figure 3. A simple planar graph with two connected components. If the points 1 , 2 , 3 are
associated with d(t,) orbitals and the points 4, 5 with d(e) orbitals, this graph illustrates
the arguments of the text concerning energy level degeneracies.

2.5.1. A basic theorem. T h e basic result which links the degeneracy of an energy
level with group theory is the following; any operator of G(H)acting on an eigen-
function of H will produce a function which is also an eigenfunction of H associated
with the same eigenvalues. This is simply shown. If
HlA) = AlA) (2.22)
then
HA I A ) = AH1 A ) = AA I A>. (2.23)
Suppose we are given an eigenfunction of H. By acting on this function with all
elements of G(H)in turn we generate a family of degenerate functions. These
functions define a linear subspace such that any function in that subspace is an
eigenfunction of H with the originally given eigenvalue. This may be illustrated
pictorially by means of a graph of the type studied in planar graph theory (Berge
1962, Ore 1963). The functions are represented by the vertices of the graph (see
figure 3), and an arc of the graph goes from a to b if b occurs as a component in the
resultant function obtained when a is acted on by any of the group functions. A
graph is called connected if any vertex may be reached by some path from any other
vertex. I n this case, the connected components of the graph correspond to degenerate
554 J . Killingbeck

energy levels. For example, if the five traditional d wave functions


xz-yz, 32'2-r2, xy, y x , x x (2.24)
are represented by the five vertices of a graph, then all vertices are directly con-
nected to all others for the case in which the full rotation group is the group
employed. However, if the group operators are those of the cubic group 0,1,the
graph has two disjoint components, one containing two vertices (x2- y 2 , 3z2- r2)
and the other containing three ( x y , y x , x x ) (figure 2). This pictorially represents the
way in which the fivefold d level splits up into e and t, components in a cubic
crystal field. I n this d-electron example we are actually using an extension of the
result (2.23); this extension states that the expectation oalues (alHI a ) and
(blHI b ) are equal if I b ) is produced from I a ) by some operator of G(H).T h e
planar graph picture can only be used with approximate eigenfunctions, such as the
d orbitals, if the states (corresponding to the vertices) are properly symmetry
adapted to belong to the reps of G(H).Thus this pictorial approach, while very
useful for general discussion of principles, must be used carefully for particular
systems. For example, if the d functions Y p are used instead of the five 0,
symmetry-adapted functions in the example, the connectivity of the resulting graph
will be different. A similar comment can be made concerning the pictorial repre-
sentations of the atomic orbitals, showing disposition relative to the ligands, which
often appear in works on crystal field theory (e.g. McClure 1959). While such a
picture shows, for example, that ( x y [ HI x y ) = ( x x [HI xz) for an 0, crystal field,
it does not make clear the vital point that ( x y IHI x x ) is zero. I t is important to note
that the group theoretical approach only proves that eigenfunctions belong to the
same energy if they can be related by the group operators. Thus, it may be the case
that the time-reversal operator should be included in the group of H , and that it will
link functions which are not linked by the spatial symmetry operators belonging to
G(H).This situation gives the well-known Kramers degeneracy, and the arguments
given above are not affected by the anti-unitary nature of the time-reversal operator.
On the other hand, if two families of functions give exactly the same rep, it does
not follow that the two families are degenerate. For example, if a state with L = 2
is modified by the application of a crystal field of C,, symmetry, the character
decomposition gives D ( 2 -+ ) Fl + 2r3(,4, + ZE,). T h e two r3 levels are not neces-
sarily degenerate, since their corresponding functions do not mix under the group
operations. I n this respect some of the degeneracies given by Stevenson (1965) in
his energy level diagrams are incorrect.
2.5.2. The model approach; effectizie symmetries. Given the abstract group iso-
morphic to G(H), then the dimensionality of the reps of G(H)is fixed and, apart
from accidental degeneracy, or extra degeneracy due to the anti-unitary time
reversal operator, the possible degeneracies of the energy levels of H are fixed. An
alternative approach, which is often useful, is to explore the reps of the group by
deliberately constructing a model Hamiltonian with the given symmetry group, and
then solving the eigenvalue problem of H explicitly within some invariant subspace.
T h e formal problem remains of ensuring that the model H does not possess hidden
symmetries or accidental degeneracies, e.g. the use of the hydrogen-atom
Hamiltonian as a model for the group R, would give degeneracies which are too
great, while the use of a non-Coulombic central potential in H would give the
correct 21+ 1-fold degeneracies appropriate to the reps D(I)of R,. As pointed out
Group theory and topology in solid state physics 555

by Gabriel (1964), the set of all operators A ( G ) which commute with a given group
G constitutes an algebra, the commutator algebra of G . This algebra includes all
possible Hamiltonians with symmetry group G , and degeneracies common to all
elements of the algebra must be regarded as essential, i.e. due to the symmetry of G .
Gabriel (1964) discusses these ideas from the viewpoint of formal group theory; in
practice any degeneracy is termed ‘ accidental ’ if it is not due to any known element
of G ( H ) . As a simple example of this model approach, we consider the case (dis-
cussed above) of the splitting of an L = 2 state of an ion in a c,, crystal field. T h e
tables of Prather (1961) show that the crystal field may be written in the form below
for C,, symmetry:
V = C f?(r) YR(O,+) n 2 m m = 0, F 3, F 6, etc. (2.25)
nm

Within a manifold of states with L = 2, V may be represented by means of the


equivalent operators given by Stevens (1952) and Orbach (1961). Thus, for example,
a simple crystal field with C,, symmetry and having only fi andfi nonzero would be
represented within the L = 2 manifold by the operator
V ( L )= A(3L,2-2)+B{Lz(L-3+ L-3)+(L+3+L-3)Lz). (2.26)
Solution of the 5 x 5 secular matrix in the basis of states 12, M ) gives a doublet with
energy # A+ #(9A2+ 64B2)”2, a doublet with energy $A- $(9A2+ 64B2)’/2, and a
singlet with energy - 6A. T h e calculation thus shows that the two doublet E, levels
need not be degenerate by virtue of C,, symmetry. As a by-product of this calcula-
tion we obtain states of A, and E, symmetry, namely the eigencolumns of the 5 x 5
matrix.
As pointed out previously, it is not possible for the site symmetry at a sub-
stitutional or interstitial site in a regular lattice to be of icosahedral type (unless, of
course, a very special and unlikely type of local distortion occurs around the site
considered). However, it is really the symmetry of H which counts, and H may
show an effective symmetry which is different from the geometrical symmetry.
For example, Judd (1957) showed that the crystal field acting on a rare-earth ion
in the rare-earth double nitrates is almost of icosahedral symmetry; this manifests
itself by the occurrence of high near-degeneracies in the energy level scheme. T h e
point is that the particular crystal field parameters which cause the crystal field
splittings happen in this case to have relative magnitude very close to those which
would be found for the field due to a regular icosahedron with equal point charges
at its vertices. T h e actual geometrical symmetry around the rare-earth site is
probably roughly of C,, symmetry (Judd 1957). As another example, a C,, type
geometrical array of ligands gives a splitting which is describable by using a crystal
field operator with the apparent symmetry D311. (c3h is a subgroup of D3h.) Murao
et al. (1967) showed, however, that it is possible to distinguish whether the crystal
field is of C,, or D,, type if the variation of optical absorption intensity with
applied magnetic field direction is considered. Crystal field theory is discussed
further in $ 3 of this article.

2.5.3. Anti-commutators and time reversal. T h e usual approach to groups of


operators makes extensive use of operator commutators, but it is interesting to note
that anti-commutators can also be introduced into the theory to a small extent,
For example, the x component of angular momentum anti-commutes with the
556 J . Killingbeck

reflection in the x x plane,

(2.27)

This result is easily obtained by using the method already explained for the class
relationship, i.e. by making the substitution y -+ -y in the 1, operator once to obtain
the correct result for the triple product of operators. T h e result (2.27) may be used
in an argument analogous to that used in angular momentum theory, and shows that
ul,acts on an eigenfunction of 1, to reverse the eigenvalue. Since 1, is invariant under
x rotations, and since u y can be written as I x 2,, where the inversion I commutes
with 1,) it follows that
2, YT = N Y i m , (2.28)
I n (2.28)) y is any axis perpendicular to the 0 = 0 axis of the spherical harmonic,
and cy is of unit modulus. Since the operation of complex conjugation also changes
the sign of m, it follows that the two operations can be made identical ( f o r the
particular operand YT)by an appropriate choice of a. For the motion of an electron
in a potential field, it is well known that the transformation #+#* of the wave
function may yield degeneracies additional to those given by the rotational or other
operators of G ( H ) . When spin is included, the anti-linear complex conjugation
operator is generalized to the full time-reversal operator. T h e discussion above
shows that for the orbital wave function the complex conjugation operator can be
made equivalent to the operator 2,. While the former operator is meaningless for a
spin function, the latter is not, and leads to a generalization of the conjugation
operator, i.e. the time reversal operator, in which the operator 2, acts on the spin
factors of the wave function (Watanabe 1966). Fano and Racah (1959) discuss the
relationship between the complex conjugation operation and the operator 2, at some
length in their treatment of contragredient tensorial sets. I n one-electron band
theory, for those cases in which the effective one-electron Hamiltonian possesses
both space group invariance and time reversal invariance, the effect of the time-
reversal invariance is to ensure that Bloch eigenfunctions are always associated in
pairs, with wave vectors & k , for a given energy.

2.6. Double groups and magnetic groups


2.6.1. The Bethe double group. T h e last few years have involved a vast proliferation
of the number and type of symmetry groups used to describe solids and molecules,
and it is important to distinguish between the various levels of physical theory at
which the use of such groups becomes imperative. Probably the longest-known
example of an extension of the ordinary space-symmetry groups is provided by the
double groups associated with the point groups of crystal field theory. T h e use of
the double group is necessary when the basis functions belong to half-integral reps
of the group of rotations for the problem. This arises when spin is included, with an
odd number of electrons, and is required even when the Hamiltonian H includes
only orbital operators, and also even when the time-reversal operation is not con-
sidered as a possible symmetry operator of H.I n this case, then, the use of the double
group formalism is not necessitated by a change in the symmetry group of the
physical problem, but by a change in the properties of the basis functions on which
the symmetry operators act. This point is perhaps best displayed by the original
procedure of Bethe (1929), who introduced a fictitious group element R to represent
Group theory and topology in solid state physics 557

a rotation of 277 about any axis. R2 he then regarded as the identity E, and further
postulated that R commutes with any rotation. When the space inversion I appears
as a factor in a group element, the property 1 2 = E (not R ) is retained in treating that
element (see e.g. Elliott 1954). Bethe's procedure amounts to describing the usual
rotations of real space in terms of angles defined modulo 47~;in short, the physical
symmetry group of H is unchanged, but its elements are described in a way which is
appropriate for double-valued basis functions. We stress this point, since almost all
works on the subject concentrate on the technical problems of character decom-
position, introducing the double group as though its only purpose were to give
single-valued quantities in the theory. Such an approach omits the important
demonstration that the groups thus introduced are still symmetry groups of H ;
without this link the physical value of the extension to the double group is not
established. T h e double group description is still applicable for basis states with
integer angular momentum, and then leads to the same results as the single group
description, since such basis states give the same character for the group elements
y and R y . T h e double group G+ of a spatial symmetry group G has twice as many
elements as the group G, but usually does not have twice as many classes. Classes
of G which include twofold rotations can coalesce in pairs to give one class of G+.
T h e group D,+ illustrates this; from the fact that R commutes with any rotation,
and the relation R2 = E , it might appear that R behaves like the usual space inver-
sion, so that D,+ is isomorphic to DZh. This is not so; the use of angles modulo
477 applies for all rotations in G , and this further property actually makes D,-
isomorphic to the non-Abelian group D,. T h e elements E, R , each form a class of
one element in D2+,while the pairs (2,, R2,), (2u,R2,), (2*,R2,) each form classes of
two elements. T h e latter statement is proved as follows:
(R2,) 2,(R2,)-1 = (R2,) 2,2,-1 R-l = R22,2,2,3 = R22,2,2,2 = R2,. (2.29)
T h e group D,+ thus has five classes, with reps of dimension 1,1,1,1,2. T h e two-
dimensional rep is the only one appropriate for half-integer D(J)basis states; in the
simple one-electron case this result is often directly obvious in terms of a twofold
spin degeneracy of each orbital state, but it persists for any number of electrons, and
also for the case in which spin-orbit coupling is included in the Hamiltonian. T h e
group D,+ was used by Koningstein (1968) in a discussion of the Raman effect of
the Yb3f ion (configuration 4f13) in ytterbium gallium garnet. In a recent note
Sivardikre (1969) has pointed out that the theory of the reps of double point groups
can be simplified by noting that the double groups are solvable.
2.6.2. The Opechowski double group. For all the groups G- treated by Bethe (1929)
there is a coalescence of the classes of G which involve twofold axes, i.e. the classes
R and R C , coincide to give one class of G+, as illustrated above for D,+. Ballhausen
(1962) erroneously asserts that this result is always valid. Opechowski (1940)
showed that this is not so for all point groups, but that on passing from G to GT
either all such classes coalesce in pairs, or none of them do. Opechowski further
formalized Bethe's work by noting that G- is essentially the group formed by the
D(':g) matrices M(R) for the rotations R of G with +M(R) and -M(R) being
considered as distinct elements of G-. Since the groups G are subgroups of the
full rotation group R,, this viewpoint is directly using the well-known two-to-one
homomorphism between SU, and R3. Opechowski's approach can be extended to
give double groups for space groups; in the double space group G+ the translational
558 J . Killingbeck

operations are left unchanged but the elements 5 D(l’z)(R)are introduced for the
rotational parts R of the space group operations. This is because the spin functions
are only two-valued with respect to rotational operations. I n a treatment of the
effects of spin-orbit coupling in band theory, Elliott (1954) has discussed the double
groups of the three cubic lattices, the diamond lattice and the hexagonal close-
packed lattice.
2.6.3. Comments on double groups. The Bethe approach to the double group seems
more physically intuitive than the Opechowski approach, and double group
character tables are universally displayed in a form which contains the fictitious
element R. However, the Bethe approach is mathematically unsound. T h e full
mathematical details on this point require a lengthy exposition (Killingbeck, to be
submitted for publication), but the gist of the argument is as follows. For the case
of D2+,treated above, it follows from the Bethe prescription that the triple products
(2,2,)2, = R2, and 2,(2,2,) = 2,2, = 2,
are not equal. This lack of associativity means that the Bethe D,+ is not a group!
However, the Opechowski D,+ is a group (matrix multiplication is always associa-
tive). I t turns out that by applying the standard formulae of finite group theory to
the Bethe D,+ as if it were a group, the same rep dimensionalities are obtained as
would result for the Opechowski D,+ group. (Triple products of the special type
used in equation (29) are still associative.) T o show that the Opechowski group is
also the symmetry group for the physical problem we must remember that in a
basis of spin-orbitals the Hamiltonian operator H should be regarded as a 2 x 2
matrix in spin space as well as a continuous matrix in coordinate space. T h e eigen-
functions thus became columns of two elements, each element being a function of
the orbital coordinates. T h e set of operators which commute with H also acquires
a 2 x 2 matrix form in terms of the spin indices, and it is in this way that the
Opechowski group of 2 x 2D(’/2)matrices turns out to be the appropriate symmetry
group when the electron spin is included in the problem. T h e eigenfunctions
(essentially eigencolumns) must accordingly be classified according to the reps of
the Opechowski group. Thus, while the discussion of the Bethe double group given
in the previous section appears deceptively simple, it is not completely rigorous
mathematically. However, it is still true to say that the double group formalism
does not arise because of a change in the physical symmetry, but rather because of a
change in the nature of the basis states when spin is included. T h e double group
then appears as the mathematical symmetry group of the secular matrix problem of
H in this spin-orbital basis; this viewpoint is in line with that adopted in the general
discussion of $ 2.5 of this article, if the operators A , B and H of equation (2.21) are
interpreted as partially continuous and partially discrete matrices.
In $2.5 emphasis has been placed on the fact that the symmetry of the
Hamiltonian operator H is the most important symmetry in a quantum mechanical
problem. The use of an approximate or effective Hamiltonian may thus impose a
particular symmetry group on a problem, and this idea, already discussed in $2.5,
may be further illustrated by the use of the double group in crystal field theory.
For a d electron in a strong 0, field there are two possible levels, the triply degenerate
T, and the doubly degenerate E. Allowing for spin degeneracy there are thus six
states of type I T,, m ) 14, S,), where m labels the T, state and can be chosen, for
example, according to the convention of Griffith (1962). Since the crystal field acts
Group theory and topology in solid state physics 559

on the orbital function only, and there is no spin-orbit coupling, the symmetry
group of H is the direct product of the group 0 for the orbits and the group R, for
the spins. T h e six functions arising from T, may be classified according to their
behaviour under 0+,which is essentially the subgroup of 0 x R, for which the spin
and orbital rotations are restricted to be the same rotations of 0. This classification
according to Of is mathematically possible, since the space of the six functions is
invariant under the operations concerned, and becomes physically useful when Of
actually becomes the symmetry group of H . This happens when spin-orbit coupling
is introduced, and the sixfold level splits to give one fourfold and one twofold level.
This result can be easily obtained by remembering that the T, states have a pseudo-
angular momentum of 1 (see § 3 of this article). T h e spin-orbit coupling thus yields
j = 4 a n d j = Q. This example illustrates clearly the relevance of the double group
formalism; the maximum dimension of the reps of 0 is 3, while O+ has a rep of
dimension 4, to which the fourfold level belongs.
2.6.4. Magnetic groups; coreps. Much of the recent literature dealing with group
theoretical methods in solid state physics has been devoted to the magnetic point
groups and space groups. A review of this field has already appeared (Cracknell
1969), and so we restrict ourselves to making a few basic points concerning magnetic
groups. T h e key fact about these groups is that they contain anti-unitary operators,
consisting of products of the anti-unitary time-reversal operator 8 with unitary
spatial symmetry operators, as well as containing unitary symmetry operators. T h e
unitary operators clearly form an invariant subgroup H of index 2 in the group,
and the group may be written as the sum of two cosets,
G = H+XH. (2.30)
If B belongs to G , we may set X = 8 in (2.30); otherwise X = BY for some unitary
element Y of G . With,H taken to be one of the usual point groups or space groups,
the possible magnetic groups G are isomorphic to point groups and space groups of
the usual type. This is similar to the way in which hemihedric point groups turn out
to be isomorphic to proper point groups; indeed if the space-inversion operator I
replaces 8 in the discussion above, the discussion remains valid (for point groups) as
far as the abstract group structure is concerned. However, 8 is anti-unitary and I
is unitary, and this fact produces differences between the theories of the matrix
representations in the two cases (Dimmock and Wheeler 1962, Bradley and Davies
1968). T h e matrix representations of the magnetic group are in fact better termed
corepresentations (Wigner 1959), since they cannot in general be arranged to give
the usual group multiplication properties, but rather obey the multiplication scheme
below :
D(a) D(b) = D(ab) D(a) D(m) = D(am) (2.31)
D(m) D*(a) = D(ma) D(m) D*(n) = D(mn). (2.32)
I n these equations, a and b are typical elements from the coset H , and m and n are
typical elements from the coset X H. An irreducible corepresentation (corep) may
be defined (by analogy with a rep) to be a corepresentation provided by a subspace
which is irreducible with respect to the operators in G . T h e usual procedure is to
take subspaces which gives reps of H, and then to see how they combine to give
subspaces irreducible under G . This procedure is carried out formally in various
papers (Dimmock and Wheeler 1962, Bradley and Davies 1968, Cracknell 1968).
5 60 J . Killingbeck

I n a more directly physical context, namely that of band theory, Herring (1937 a)
investigated the way in which the addition of 8 to the symmetry group of the
Hamiltonian would affect the degeneracies found by studying the spatial sym-
metries alone. Herring's work thus applied to groups for which H in (2.30) is a
unitary symmetry group and X = 8, i.e. direct product groups. Cracknell (1968)
showed that for these groups (the so-called grey groups) Herring's work is exactly
equivalent to the formal corep theory. This is to be expected, since both approaches
are essentially searching for the irreducible subspaces associated with G , starting
from those associated with H.

2.6.5. Groups of operators. It is worth noting that the usual character invariance
property cannot in general be applied to the corep matrices of G , since their traces
are not necessarily invariant under unitary transformations. Furthermore, the
discussion given in $ 2.5 of this article must be modified if the symmetry group of H
contains anti-linear operators such as 6'. Suppose that 8 commutes with some
operator A , and we have an eigenfunction 4 of A with complex eigenvalue x. Then,
A04 = 6'A+= 8x4 = Z* 94 (2.33)
so that 04 has eigenvalues z*. If we set A = H , the eigenvalues x all become real,
and the statements about energy level degeneracy made in $2.5 remain valid.
However, if A is a rotation operator, with a complex eigenvalue, belonging to a
group H, extra degeneracy can result, For example, the functions Y:(6',+)and
Yy1(8, 4) belong to different reps of C,, and have eigenvalues exp ( -t i71.14) for the
operator 4,. T h e operator 8 links the two functions, and they thus give a single
corep of the group C, + BC,. T h e basic step taken by the papers referred to above is
to give criteria for deciding when such extra degeneracy occurs (including those
cases when 6' alone does not appear in G ) , these criteria being expressed in terms of
the characters of the reps of H. These criteria are given for both single-valued and
double-valued reps of H, since a double-valued rep of H (i.e. a rep of H+) leads to
a double-valued corep of the magnetic group. Such a double-valued corep is
associated with an irreducible basis set of product functions arising from an odd
number of electronic spins. The same remark may be made at this point as was
made concerning double groups, namely that much of the literature concentrates
on technical problems without making clear their links with physical theory. T h e
group of the Hamiltonian is a group of operators, and the group elements thus can
have extra properties (linearity, unitarity, anti-unitarity, etc.) which are additional
to their abstract group multiplication properties. In the quantum mechanical
approach, the operators are usually represented by matrices with respect to some
orthonormal basis in an invariant subspace, and these matrices will then reflect the
unitary, anti-unitary or other properties of the operators. For example, the group
+
C,+ BC, is isomorphic in an abstract sense to the group C, IC, = C4h, if we set
82 = E (which is appropriate for orbital operands). However, C,, has only one-
dimensional reps if we apply unitary rep theory, whereas we have already seen that
C,+ BC, has an irreducible subspace of dimension two, associated with some
corep. Thus the formal theory of coreps is needed in the physical theory when the
Hamiltonian operator H has 8 as one of its symmetry operators, or 8 in combination
with some unitary operator. Thus, while the abstract isomorphism between groups
is useful, as outlined in $2.1, the rider must be added that for a group of operators G
the representation theory can involve different degeneracies if some of the elements
Gvoup theory and topology in solid state physics 561

of G are said to be anti-unitary. I n other words, isomorphism of group structure


need not imply isomorphism of the rep structure for groups of operators. Those
symmetry operations of physical theory which may be directly visualized lead to
unitary operators. For groups of these operators the isomorphism of groups does
lead to isomorphism of reps. Dimmock and Wheeler (1962) pointed out that
Indenbom (1959) had obtained incorrect results for the magnetic groups by taking
the operator 8 to be linear instead of antilinear.

2.6.6. Shubnikov groups. T h e magnetic space groups with X f 8 are isomorphic to


the black and white groups of Shubnikov and Belov (1964) and are often displayed
in this useful pictorial form. T h e basic theory of space groups was worked out
before experimental x-ray work made it possible to verify that it was the appro-
priate theory for the description of perfect crystalline media; similarly, Shubnikov
and others introduced the concept of antisymmetry into space group theory before
it was shown to be appropriate for the description of ordered magnetic lattices.
T h e anti-symmetry operation P may be thought of as the interchange operation
black ts white if the lattice points are thought of as having two possible colours,
black and white. Many workers have investigated the way in which the inclusion of
an anti-symmetry operation modifies (or, rather, enlarges) the traditional theory
based on Bravais lattices (Mackay 1957, Holser 1961, Shubnikov and Belov 1964).
For example, if alternate points on a sc lattice are coloured black and white, the
spatial symmetry group of the resulting structure is that of the fcc lattice. However,
some operations of form PR are also symmetry operations, where R belongs to the
original symmetry group of the simple cubic lattice. If the original cubic lattice
parameter is taken to be unity, and the space group of the fcc lattice is called H , the
resulting black and white group is
G = H+T,,,PH (2.34)
where JT,,,,, is a translation along (100) by the distance 1. T h e group G is iso-
morphic to the group of the original sc lattice, but leads to different reps (in fact
coreps) if P is interpreted as being 8, the anti-unitary time-reversal operator. It is
this interpretation which is necessary in the quantum mechanics of magnetic
crystals; 8 is then the operator which reverses the sign of the magnetic moment at
each site, and this reversal operation replaces the black t-)white interchange
operation of Shubnikov’s theory. Ordered magnetic arrangements are often shown
by placing arrows on the atomic sites to indicate the magnetic moment direction
(Atoji 1965, Bacon 1965, Tahir-Kheli et al. 1966, Bertaut 1969), and it is possible
to determine the magnetic order by neutron scattering experiments (see Cracknell
1969). This pictorial view is useful in developing an intuitive grasp of the formal
theory. For example, if a lattice is invariant under some operation P X , where X is
a spatial operation, then it follows that under operation X alone the lattice must
simply suffer a reversal of all the site arrows while being otherwise invariant
(figure 4). Thus, if we take those one-dimensional reps of an ordinary point or space
group G’ in which half the elements are represented by + 1 and half by - 1, we
may expect to obtain a possible magnetic group structure (regarding - 1 as equiva-
lent to the reversing of an arrow). In fact the formal theory (Indenbom 1959,
Bradley and Davies 1968) shows that all the abstract magnetic point and space
groups of black and white type can be obtained in this way from the ordinary point
and space groups. (The black and white groups have X f 8 in equation (2.30).)
562 J. Killingbeck

T h e group H of equation (2.30) is then that group of elements which are associated
with + 1 in the relevant rep and is an invariant subgroup of G ’ , since it is the kernel
of the homomorphism between G’ and the pair of numbers I 1. G of equation (2.30)
is then isomorphic to the generating group G ’ , but, of course, yields coreps rather
than reps. From this point of view Indenbom (1959) was essentially finding the
unitary matrix reps of the magnetic groups; these do exist, but it is the coreps which
are relevant for physical theory, as discussed in detail above. Starting with one of
the 32 point groups as H in equation (2.30), it is clearly possible to obtain 32
magnetic groups (the grey groups) by setting X = 8. It is also possible to obtain
58 magnetic groups (the black and white groups) with Xf 0 in (2.30). For the space
groups there are 230 grey groups and 1191 black and white groups, which are given
by Opechowski and Guccione (1965).

(U) (b) (C 1

Figure 4. (c) possesses the symmetry operation P4,, P being the black t+ white interchange
operation. ( a ) also possesses this symmetry if P is interpreted as arrow reversal, with the
arrow fixed radially outward during the rotation 4,. I t is in this way that rotations are
interpreted in the theory o f magnetic symmetry, e.g. figure (b), in which the arrow
‘floats’ during the rotation 4,, does not have the symmetry operation P4,.

2.6.7. Spin-space groups. Section 2.5 has emphasized that it is the symmetry group
of the Hamiltonian which is central to any quantum mechanical problem, and this
fact seems to be somewhat obscured in the literature on magnetic groups. If a
polarized neutron beam is diffracted by a magnetically ordered crystal, the Born
scattering theory of this process will involve an interaction term determined by the
arrangement of magnetic moments, and thus involving the magnetic space group.
However, the array of moments itself, while describable by a magnetic group, is
actually the ground state arising from some spin Hamiltonian, which is often taken
to be the isotropic Heisenberg interaction
(2.35)
T h e symmetry group G of this H is much larger than that of the magnetic space
group of its ground state, and is one of the spin space groups discussed by Brinkman
and Elliott (1966). G includes the full group R, of simultaneous rotation of all the
spins, as well as the spatial lattice symmetry group. A spin wave represents a small
disturbance of the ground state, and thus in the theory of spin waves the relevant
symmetry group depends on the spin symmetry in the ordered ground state. If
spin-orbit coupling is present the possible spin and spatial symmetry operations
will be tied together, and the spin-wave problem will have a magnetic space group
as its symmetry group (Brinkman and Elliott 1966). More general spin Hamiltonian
terms than the isotropic one of equation (2.35) are compatible with the existence of
an ordered magnetic ground state (i.e. are invariant with respect to magnetic space
group operations); one particular term increasingly studied in recent years is the
Group theory and topology in solid state physics 563

anisotropic exchange operator of form C D i j .[Si x S,] (Stevens 1953 a, Moriya


1960, Erdos 1966, Tachiki 1968, Bertaut 1969). Bradley and Davies (1968) state
that no cubic point group can be ferromagnetic, meaning that an axial vector (which
would represent a permanent magnetic moment) cannot be invariant under all the
operations of such a group. They assert that this implies that no undistorted cubic
lattice can be ferromagnetic; the present author interprets the result,simply to imply
that if a cubic crystal becomes ferromagnetic, then the resulting ordered lattice
cannot be described by a cubic group. Ferromagnetic crystals with the NaCl
structure are known experimentally e.g. EuO, EuS (Matthias et ~1.1961).

2.7. Symmetry groups of nonrigid molecules


If a system can tunnel quantum mechanically between various spatial con-
figurations, with a characteristic time which is short compared with the time scale
of some experiment performed on the system, then the experiment will ‘see’ an
average symmetry which arises from all the possible configurations. For example, in
a magnetic resonance experiment in which the ligands of the magnetic ion are
molecules which are capable of rotation, the average symmetry as revealed by the
resonance lines may vary with temperature, since at low temperatures the molecules
may freeze into particular arrangements (Bates and Stevens 1969). This idea was
originally developed by Longuet-Higgins (1962) for the case of isolated molecules,
and he defined groups of feasible transformations for nonrigid molecules. A
feasible transformation is one which can occur rapidly (this definition is clearly
somewhat imprecise in that it depends on the time scale of the particular experi-
ment). Longuet-Higgins defined groups of feasible operations for various nonrigid
molecules; these groups have many more elements than the usual point groups,
e.g. the boron trimethyl molecule has an associated group of order 324. Stone
(1964) described how to calculate character tables for the groups of certain mole-
cules, and Mortimer (1969) has used some of the ideas of the theory in a discussion
of the nmr spectra of nonrigid molecules.

3. Matrix methods and group theory


3.1. The secular matrix and group theory
3.1.1. Degeneracies of approximate eigenfunctions. Most texts dealing with group
theory in physics proceed by proving theorems about the symmetry properties of
the exact eigenstates of a Hamiltonian H with symmetry group G(H).I n practice,
of course, exact eigenstates are rarely encountered, and it is much more common to
deal with approximate eigenfunctions which result from the solution of the secular
matrix of H in some finite set of basis states. In this case, as already pointed out in
$2.5, the group operators link together functions with equal expectation values of H.
In most problems the finite basis set employed to set up the matrix of H constitutes
an invariant subspace for the operators of G(H), and under these circumstances the
matrix calculation yields approximate eigenfunctions which have the same
degeneracies and rep labels as those appropriate to the exact eigenfunctions.
Suppose that the invariant subspace of orthonormal functions In), n = 1 , 2 , ...,N ,
is used to set up the matrix of H. Solving the secular matrix of H in this subspace
is equivalent to solving exactly the eigenvalue problem for the operator
H’= PHP
5 64 J: Killingbeck

in the whole of Hilbert space. T h e projection operator P belongs to the identity


rep of G ( H ) ,Le. it commutes with all the operators of G ( H ) . Thus H ' has the same
symmetry group as H , and the eigenfunctions of H ' , which are the approximate
eigenfunctions of H , show the typical degeneracy and rep labelling of the group
G ( H ) . Any function not in the subspace of the I n ) is formally an eigenfunction of
H ' with eigenvalue zero, and as the size of the subspace increases the operator P
approaches the identity operator, i.e. H ' tends to H in the limit in which the whole
of Hilbert space is used as the basis set. The Sternheimer Hamiltonian, H,, associated
with an approximate eigenfunction may also be discussed in group theoretical
terms. If 4 is an approximate eigenfunction with eigenvalue E of H = T + V , where
T and V are the kinetic and potential energy operators, then 4 is an exact eigen-
function with energy E of the Sternheimer Hamiltonian

Equation (3.2) leads to an effective local potential even when 4 arises from a cal-
culation involving nonlocal potentials (Goddard 1968); in general H ' of (3.1) will
include nonlocal parts. If 4 belongs to a one-dimensional rep of some group of
rotations or translations, it follows that H, is invariant with respect to that group,
since T is invariant for any such group. H, will then have other eigenfunctions,
some of them associated with degenerate reps of the group. These other eigen-
functions may be reasonable approximations to the eigenfunctions of H itself
(Goddard 1968), since H, = H if the function 4 is an exact eigenfunction of H , for
the case in which H has a local potential term. We would also expect that a set of
degenerate approximate eigenfunctions will each yield the same H,, and that this
H, will belong to the identity rep of the group for which the starting functions form
a rep basis. A proof for the case of spherical symmetry, and for a one-electron
problem, is as follows. T h e approximate functions are taken to be in the form

Since E and V in (3.1) are independent of m, and V is assumed spherically sym-


metric, only the term involving T need be examined. We have, since y? is a solid
harmonic of degree n,

the last step involving the Euler identity


r . grady: = ny;.
In this case, H, is spherically symmetric and independent of m.
3.1.2. The generalized Wigner-Eckart theorem. As the discussion above has indi-
cated, the characteristic degeneracies and rep types of G ( H ) will appear in the
secular matrix calculation for H . This is true whether or not explicit account of
symmetry is taken in setting up the matrix. If no account is taken of symmetry, the
resulting matrix may appear quite complicated and of high dimension ; however, the
relative sizes of the various matrix elements will be determined by symmetry
(usually via the Wigner-Eckart theorem (Koster 1958, Judd 1963) and the explicit
diagonalization of the matrix will yield the characteristic degeneracies which could
have been predicted by means of a group character analysis on the basis functions.
Group theory and topology in solid state physics 5 65

T h e resulting eigenvectors of the matrix are symmetry adapted functions (SAF)


belonging to reps of G(H).From the ‘model’ viewpoint discussed in $2.5, the
SAFS could be constructed by diagonalizing the matrix of some H with the correct
symmetry group. I t is much more common, however, to rearrange the basis set into
symmetry-adapted form by means of projection operators, so that the matrix of H in
a SAF basis is reduced as far as possible into smaller submatrices. A few comments
may be made here about projection operators. Firstly, since functions belonging to
different reps of a group are orthogonal, projection operators provide a means of
orthogonalizing sets of overlapping wave functions; probably the best known
example is the projected pair 4(1)*+(2), for the case in which 4 represents an
atomic 1s orbital in the molecular orbital theory of H,, and the group involved is the
symmetric group S,. These projected functions are orthogonal on mathematical
grounds, irrespective of what may be said about the symmetry of the Hamiltonian
for the molecule. Secondly, although this does not seem to have been pointed out
in the literature, the usual proof that the projection operator

projects out a function belonging to the pth row of the representation rei) does not
involve the assertion that r ( i )is irreducible. Equation (3.4) can thus be used to
produce families of functions with identical transformation properties even when
more than one rep is involved. For the cases in which is a rep the projection
process is particularly important in connection with the Wigner-Eckart theorem.
For a simply reducible group this theorem asserts that matrix elements are pro-
portional to coupling coefficients as follows,
( r ” m ” ) T , ( m ) I r ’ m ’ )= ~sc A v ( r r d I ‘ mrl ”
~ ” ) { r ~ ~ ” ~ ~(3.5) r~~r’
I n (3.5) the labels m‘, m”, denote specific members of the rep families I?’, r”,
respectively, the tensor operators Tr(m)being defined to transform according to the
rep r of the group. T h e coefficients C( I ) are those which describe the Kronecker
product decomposition
I I”’m” N , (I? x I?’)) =
m,m’
I
C,( r m r ’ WZ’I r”m”)I r m ) I”m’). (3.6)

The index ill allows for the possibility that a given rep may occur more than once
in the Kronecker product. This complication does not arise for the full rotation
group R,, but does for some of its finite subgroups, and the result (3.5) for these
cases is derived by several authors (e.g. Heine 1960, Judd 1963). Judd (1959) gives
an example of the use of the general equation in connection with spin Hamiltonian
theory. The coefficients C( I ) are only well defined if the rep matrices for each rep
are taken as given in some standard form. For the group R, the standard choice is
that involving the matrices D ( J )set up in a basis IJ, M ) for which the usual shift
operator relations
J & M ) = ( J ( J + l ) - M ( M + I ) y q J , M + 1) (3.7)
hold (Fano and Racah 1959). I n any other case the basis functions for each rep of
the group must be chosen in a standard form, and operators such as (3.4) can then
be used to arrange any basis set for a rep so that it has the standard transformation
matrices (Griffith 1962, Mariot 1962). If the operator A is invariant under the
25
566 J . Killingbeck

group G , and the projection operators P ( p )for the reps of G are normalized to be
idempotent, then the following result holds, P being any PQ-'),

= (+IAP21+) = (+lAPI+>. (3.8)


This means that the matrix element of A between the SAFS derived from and + +
can be evaluated without explicitly forming P @ )+, For the permutation group S,,
the result (3 .S) leads to the well-known simplification in the evaluation of matrix
elements between determinantal states of an n-electron system (Condon and
Shortley 1935).
3.1.3. iwatrix partitioning; efJective operators. The usual sum over states ' formulae
(

of quantum perturbation theory may be regarded as mathematical expressions for


the eigenvalues and eigencolumns of the infinite secular matrix of the perturbed
Hamiltonian H in the basis afforded by the complete set of eigenstates of the
unperturbed Hamiltonian H,. If only a finite number of these basis states are used
in setting up the matrix of H , then the first, second, etc., eigenvalues of the matrix
will be upper bounds to the first, second, etc. exact eigenvalues of H , but these
approximate eigenvalues may be evaluated by summing only over the finite basis
set in the usual perturbation formulae. T h e link between perturbation theory and
the matrix approach is brought out elegantly by Lowdin's matrix partitioning
technique (Lowdin 1962 a) which we present here in a slightly specialized form.
Consider a degenerate level of H,; under the influence of the perturbation H - H ,
this level may be split. We label the subspace of eigenstates (eigencolumns) of Ho
for the level by a, and the rest of Hilbert space by b. The eigenvalue problem for H
may be written in the partitioned matrix form

(3.9)
Solving for cb and then inserting the result back in (3.9) gives
guaC, = {HUa+ Hub(E1- Hhb)-'Hha}C , = EC,. (3.10)
If the a subspace contains n functions, the effective matrix operator Ha, acts within
a only, and the second term in the curly bracket is an effective operator which
represents the perturbing effects from all other states outside the a subspace.
Equation (3.9) has the feature, like the Brillouin-Wigner perturbation theory, of
having the unknown E on both sides of the eigenvalue equation. I n exact numerical
work some iterative procedure is therefore necessary, but for a weak perturbation
some appropriate first-order estimate of E can be used. Effective operators of the
type (3.10) or very similar to it have been used in recent years to describe configura-
tion interaction in atoms (Wybourne 1968), relativistic corrections to the theory
of atomic energy levels (Armstrong 1966), and second-order crystal field effects
for rare-earth ions (Killingbeck 1969 a). Hubbard et aZ. (1966) constructed an
effective crystal field operator for systems such as R/ln2+F,- in which one of the
effective operator terms represents the mixing in of charge transfer configurations
such as R;In+FF,-, and the operator acts within the manifold of n/In*T states. T h e
most widely known example of an effective operator is the spin Hamiltonian of
magnetic resonance theory, and we now use an example of this to illustrate several
comments on the use of effective operators. T h e Co2+ ion has an unclosed d7 shell,
Group theory and topology in solid state physics 5 67

with a 4F ground term. I n a crystal field of Oh or T, symmetry this term splits to


give 4A2,4T2,4T, levels, with energies in that order, for the cases in which 4A is the
ground state. I n first order the perturbing Hamiltonian
H‘ = AL.S+p.H.(L+ZS) (3.11)
leads to ‘spin-only’ magnetic effects, since only the term H . 2 S has a nonzero
expectation value for the 4A, state. By using the equivalent operator AL. S for the
spin-orbit coupling we are limiting the discussion to basis states within the 4F
manifold, which is sufficient for the purposes of the present illustration. T h e
angular momentum operator L is of T, type in Td symmetry, and the Kronecker
product expression T, x A, = T, shows that H’ will have a matrix element between
A, and T, but not between A, and T,. T h e perturbing effect of the T, level on the
A, level is then described by the following effective operator within the 4A, manifold
(3.12)
where the unperturbed energy separation may be used as an approximation in the
energy denominator. T h e zero-field splitting term arising from (3.12) is zero,
since the fact that the projection operator 2 I 4T,j)(4T,jI and the operator AL. S are
invariant under the T, group of simultaneous orbital and spin transformations
means that the zero-field effective Hamiltonian also has this property. T h e 4A,
states belong to a single fourfold rep of this group. T o represent this effect an
effective spin operator proportional to S ( S + 1) would suffice. This very simple
case illustrates clearly a fact about the spin Hamiltonian formalism which is not
alwaj-s sufficiently stressed, namely that the formalism is designed to give the
correct energies but not the correct eigenstates. The general equations (3.9, 3.10)
show that the effective operator acts on the column c, alone. If the spin states
I $, S,?are used in our CO,+ example, the effective operator hS(S+ 1) will give the
correct energy shift if h is chosen appropriately, but the eigenstates according to
(3.11) should be of the form
AL.SIA,,S,).
S ~ ) - ( E ~ - ~ ~ ~ ) - ’ ~ I TSz‘)(T2,j’,SB‘/
1242, 2,j’, (3.13)
This mixing of different F eigenstates is not described by the spin Hamiltonian
formalism ; fortunately, the spin resonance selection rules are not affected by this,
since the spin Hamiltonian eigenfunctions do show the correct transformation
properties of the functions (3.13). If a tetragonal distortion is suffered by the
tetrahedral environment of the CO,- ion, giving a Dzd crystal field, then the 4T,
and *T, levels split (T, -+A, + E, T,+ R, + E) while the A, level remains single
(-A2-\B1), in the absence of spin-orbit coupling. If the spin-orbit coupling is
weak, so that the first-order splitting of the T, level is almost exclusively due to the
tetragonal field, then the effective Hamiltonian for the 4B, manifold becomes
ptH. 2 s - { ( E E -EA)-‘ 2 H’ I E,j) (E,j 1H’)- {(EB - E-L)-lH’ I A) (A Iff’}. (3.14)
j

If the two energy denominators were equal this operator would belong to the A,
rep of the T, group, as already discussed. However, the tetragonal splitting of the
T2 level makes the denominators unequal. T h e effective operator then has only
n2d symmetry, and produces a zero-field splitting of the 4B, manifold. This
shows up qualitatively in the group character analysis, since the 4B1states give two
distinct reps of the D,, double group. T h e result may be summarized as follows:
568 J , Killingbeck

the character analysis indicates that the 4A, level should split on passing to DZd+
symmetry; in first order no splitting of 4A, occurs, but a splitting of T, does occur;
in second order the splitting of T, leads to a small splitting of 4A,. This example
illustrates a basic result ; provided that the basis functions form invariant subspaces
for the group of H, and that G ( H ) is a subgroup of G(H,), then the level degeneracies
at each order of perturbation theory must be in accord with the group character
analysis. I t may be, however, that the second or a higher order of perturbation
theory is needed to split some levels which need not be degenerate on symmetry
grounds. Resides our example here, another problem which involves considerations
of this type is the problem of the zero-field splitting of S state ions, where high
orders of perturbation theory are needed to yield the full splitting effect allowed by
the results of group character analysis (Sharma et al. 1966, Wybourne 1966). I t is
worthy of note that the splitting effect for our Co2- example is only obtained when
thefimt-order energy gaps are used in the energy denominators, in accord with the
matrix result of equation (3.10). Lowdin (1968) has discussed some problems
which arise in the interpretation of low-order perturbation theory when the total
Hamiltonian has a definite symmetry group. Explicit evaluation of (3.14) shows
that the zero-field splitting operator is quadratic in S ; this, together with the Elzd
symmetry and the operator equivalent method of Stevens (1952) shows that the
Hamiltonian must take the form
+
H = A S ( S + 1) D{S,'- $ S ( S +1)) +gllH, S, +g,(H, S,+ H, S,) (3.15)
when the Zeeman terms are included. These latter terms bring home forcibly the
fact that the spin Hamiltonian does not exactly describe the eigenstates; the devia-
tion of theg factors from the spin-only value is due to unquenched orbital magnetism
caused by the T, mixing into A,. Since in T, symmetry the 4A, level is unsplit,
it follows that we must have
D = Kh2{(EE- EA)-'- ( E , - E&'}. (3.16)
Evaluation of the relevant matrix element gives the result K = 4 (Kamimura 1962),
and the g factors are given by
2 -gll 8h(E, - E-k)-' 2 -gL = 8 X ( E E - E*&)-', (3.17)
Thus on this model the sign of D and also the sign of g,,-g, should be related to the
sign of the tetragonal splitting of the 4T level. However, extra effects, such as the
tetragonal field mixing of 4T1(E)and 4T,(E) or the inclusion of higher LS manifolds,
could spoil this relationship by changing the relative effects of the last two terms in
(3.14). Work on CO,+ (Van Stapele et al. 1966, Jesson et al. 1968) and Cr3+
(MacFarlane 1955, Sugano and Peter 1961) suggests that such extra effects are of
importance for Cr3+(d3)in ruby. For Co2- in Cs3CoC1, detailed crystal field cal-
culations show that the corrections should be small, but there is some uncertainty
about the order of the tetragonal components of 4T as given by optical absorption
experiments (Van Stapele et al. 1966, Jesson et al. 1968). The inclusion of the extra
effects does not change the form of the zero-field splitting operator in (3.15), which
is fixed by the DZdsymmetry and the fact that the ground manifold has a spin of $.
Within these restrictions it is still possible to add extra magnetic terms, however,
and these take the form A, H, S: and fL(H,SX3+ Hy Sv3)(Henning et al. 1966,
Van Stapele et al. 1966). The procedure described above uses perturbation theory
and is originally due to Pryce (1950). Koster and Statz (1959) have described how
Group theory and topology in solid state physics 5 69

the most general form of the spin Hamiltonian can be obtained from group theory.
Their procedure yields the form of the spin operators which would result by taking
together all orders of perturbation in the Pryce approach, and can in addition yield
some extra terms which would not arise in the Pryce approach, although Jarrett
(1959) points out that they could arise in a generalization of Pryce’s approach which
includes covalency effects as well as simple crystal field theory. Koster and Statz
use an S = g ground manifold as an example, and point out that an operator of form
H. + b S ( 3 +) c S ( ~ ) ] (3.18)
would be of correct symmetry to represent the Zeeman effect in 0, symmetry,
where S ( L )denotes the combination of products of spin operators which belongs to
DCL) of R, and to T, of 0,. T h e usual crystal field perturbation approach to
Hamiltonian theory would not give the terms involving parameters b and c. Judd
(1959) also gave an example of the use of group theory in determining the terms in a
spin Hamiltonian. Rae (1969) has recently discussed the way in which the site
symmetry of a paramagnetic ion can be investigated by studying the angular varia-
tion of the esr spectrum.

3.2. Equivalent operators


I t is useful to make a distinction between effective operators, which simulate
within a given manifold the effects of perturbing mixtures from other manifolds,
and equivalent operators, which have exactly the same behaviour within a given
manifold or between two manifolds as some given operator. For example, the
operator H’ of (3.14) is an effective operator, while the spin Hamiltonian (3.15) is
an equivalent operator (or operator equivalent, in the terminology of Stevens
(1952)) for the operator H‘ in the S = 8 manifold. Thus the emphasis is on per-
turbation theory when choosing effective operators, and on group theory when
choosing equivalent operators, although the two aspects may combine in any
particular problem. One important operator which can be included in the second
category defined above is the coordinate dipole operator r , the matrix elements of
which between different manifolds are used to calculate optical absorption intensi-
ties. I n such a calculation this operator is actually an equivalent operator for the
velocity dipole operator p (Schiff 1955), with a proportionality constant which
depends on the energy difference of the states involved in the transition. When
approximate eigenstates are used in the calculation, it is debatable whether the co-
ordinate dipole or velocity dipole formalism gives the better results, since the
numerical value to be used for the energy difference becomes uncertain.
3.2.1. The operators TE. I n crystal field theory the representation of the crystal
field operator within a given L (or J ) manifold by means of an equivalent operator
which is a polynomial in the components of the operator L (or J ) is well known
(Stevens 1952, Orbach 1961, Casimir 1963, Smith and Thorley 1966, Watanabe
1966). T h e operator T p ( L ) which is equivalent to the coordinate operator Y?(r)
is not obtained by setting x - + J x , y - + J u , x + J , directly in the function YE(r),since
allowance is needed for the fact that the component operators of J do not commute.
However, the operators representing the normalized spherical harmonics are related
by the equation (Racah 1942, Judd 1963, Brink and Satchler 1968)
[L,, TE(L)]= (n(n+ 1) - m(m k 1))1’$T:*l(L). (3.19)
570 J . Killingbeck

This equation enables the operators T;(L) to be generated by starting from the
operator for Y;(r), which is always proportional to L+*. T h e procedure is the
operator analogue of the well-known procedure involving L , for the generation of
atomic angular momentum eigenstates (Condon and Shortley 1935, Judd 1963).
T h e operators TE(L)and Y E ( r )have identical rotational transformation properties,
and thus have proportional matrix elements within a given L manifold, by virtue
of the Wigner-Eckart theorem, equation (3.5). This means that the matrix elements
of T;(L) are actually proportional to vector coupling coefficients, and it follows that
for the rare-earth ions the operators T T ( J ) can be used to evaluate off-diagonal
matrix elements of the crystal field between manifolds of type [ L S J ) and I L ' S J ) ,
e.g. the 3P, and 3F, multiplets of the Pr3+(4f2)ion. I n many applications of crystal
field theory the operator T ; ( L ) is used to represent part of the crystal field, but the
symmetry is high enough to suppress any of the other Tp(L) operators. It is thus
easy to forget that T:(L) represents Y:(r)by virtue of the fact that it is a member
of the D(" type family, which includes these unused operators. Thus, if T i ( L )has
the true D(2)family classification as well as the m = 0 property, then it must have
zero matrix elements in manifolds of angular momentum 4. If the operator has the
form ALB2 - L(L + 1) this criterion immediately gives A = 3. Similarly T t ( L )must
have zero matrix elements within manifolds L = & , 1 , $ ; in general the operators
T i ( L )take the form
T i ( L )= p ( k )L (n-2k) (3.20)
k

where F ( k )is a polynomial of degree k in L ( L + 1) and only positive powers of L,


are included in the sum. For example, T t ( L ) has six coefficients when written in
this form; the conditions that (LM1 T t ( L )I L 1 V ) be zero for all states I L M ) with
L = 0, & , 1 , # yields six linear equations which determine all the coefficients. This
approach provides a means of constructing the operators, or of quickly checking
them, e.g. the equivalent operator for a cubic field given by Ballhausen (1962) is
misprinted, having ( L + 1) instead of L(L+ 1) in one of its terms. T h e checking is
sometimes made simpler by using the characteristic operator identity for the mani-
fold I L M ) , namely
l!f=--L
n
M=-L
(L,-M)=O (3.21)

to lower the order of the terms LZn. For S = 4,(3.21) becomes Sa2= $, and any
polynomial in S, can be reduced to a linear term plus a constant. Van Wageningen
(1964) has used (3.21) to give explicit polynomial expressions for finite rotation
operators. Corio (1968) has noted that any linear transformation in the manifold
ILiV) must be expressible as a sum of operators TF ( n < L ) . Thus the tensor
operators form a basis for the linear vector space of such transformations. Further-
more, the basis is orthogonal if the inner product of operators is defined by the
'trace metric'
(AIB) = Tr(A+B). (3.22)
Starting from the basis (J='J,"} ( Y = 0, 1, ...,2 L ; s = 0 , 1, ..., 2 L - r ) it is thus
possible to use the Schmidt orthogonalization method of linear space theory to
obtain the T;. Corio used a different procedure to project out the T g , and listed
them for n = 1,2, .. ., 8. T h e concept of the trace metric has also been used in the
so-called 'direct method' of nmr theory, which aims at obtaining the resonance
frequencies and transition amplitudes directly from a matrix, rather than obtaining
Group theory and topology in solid state physics 571

them as an after-product of a standard energy matrix diagonalization (Banwell and


Primas 1962, Anderson 1969).
T h e basis of the equivalent operator approach is the replacing of one operator
family (e.g. YE(r))by another operator family (e.g. T ? ( L ) ) with proportional
matrix elements within a given manifold. If the constant of proportionality happens
to be zero or infinity, false conclusions may result. For example, a vector operator
will in general have matrix elements between states which differ in their angular
momentum ( J ) values by 0 or f 1. Thus the operator L, has a nonzero matrix
element between the states I 3P2,1,= 1> and I 3P1,J, = 1) of the f 2 configuration,
whereas the operator J, does not. T h e use of J, would thus give the false impression
that all vector operators are diagonal in J. Brink and Satchler (1968) do give such
a result, but it is only valid for states of type Ij., J ) which involve the j-j coupling
of equivalent particles.

3.2.2. Pseudo-angular momentum. As an example of a case in which the states (as


opposed to the operators) have some effective property we may quote the fact that
in cubic symmetry T, and T, type states have a pseudo-angular momentum of 1.
For the group 0 the angular momentum operators L belong to rep T,. Use of the
Wigner-Eckart theorem shows that, for example,
(T,xlLIT,y) = cofistantx(x1LIy) (3.23)
so that the matrices of L within any T, manifold can be made proportional to those
for a P manifold if the T, states are appropriately chosen (Griffith 1962, Watanabe
1966). For example, in the CO,- system discussed previously, (3.23) will hold;
crystal field mixing of higher terms into the ground one by means of the cubic field
will merely change the value of the constant of proportionality. T h e CO,+ case
illustrates that the pseudo-angular momentum is simply a convenient way of des-
cribing the effects of L within the TI manifold, since the true angular momentum
for the F term is 3. T h e representations T, and T, are connected by the relationship
T, = A, x T,; an interesting way to see this is to note that for the cubic group the
reciprocal functions x-I,y-l, z-l will belong to the T, rep with x,y, x. On forming
the products of these three functions with the familiar A, function xyz,the three T,
functions y z , xz,xy are obtained. From another viewpoint the relationship
T, = A, x T, describes the change from a polar vector to an axial vector, and A,
has characters + 1 or - 1 only for each class. Consider now the matrix element
(Tzxl FI T , Y ) = (?/XI FI x 4 (3.24)
where the T, functions have been labelled T,x = y z , etc., as suggested by the
argument above involving the reciprocal coordinates. F in (3.24) is a set of three
orbital functions of T, rep type. On forming the integral which gives the value of
the matrix element, the squared A, function will belong to the A, rep, and thus the
resulting values will be proportional to those for a T, manifold. This still holds if F
is replaced by L , and thus a T, manifold has a pseudo-angular momentum of 1.
For the case of one p electron and one d electron, respectively, we have
a
(xlx7--y71y)
a = (xlx) = 1
oy ox
(3.25)
0 0
(yzIx--y-lxz) = -(yzIyz) = -1
ay ax
572 J . Killingbeck

showing that if the pseudo-angular momentum approach is applied to the d(t2)


manifold the true magnetic moment is minus the pseudomagnetic moment. For a
d(t,) electron the matrix of the spin-orbit coupling within the t, manifold is pro-
portional to that for a p electron. T h e resulting two levels have a p s e u d o j value of
$ and 3. Remembering (3.25), the statej, = 8 has magnetic moment
(3.26)
(not 1 + 1 = Z), showing that the spin and orbital moments exactly cancel for the
states with j = #. For a real p electron, of course, the moments would add, and j
would be a true angular momentum. I t is very common to use the operator 6. s in
crystal field theory as the operator for the spin-orbit coupling of one d electron.
Strictly speaking the operator should be (Schiff 1955)

(3.27)

where V is the potential field. For the free ion there is some debate as to exactly
which effective potential should be used for V when interelectronic forces are
present, but it is clear that the crystal field should be added to the free-ion V . T h e
spin-orbit coupling operator thus becomes t , , s in octahedral symmetry (Stevens
1953 b). If the equivalent operator 6 is used within the d(t,) manifold, the result is
that the free ion operator 9,6. s is replaced by the operator 92. s within d(t,) where
(Schmidtke 1963, Killingbeck 1969 b)
6
27 = 9 ? o - - . l e Q l ( ~ 2 ) . (3.28)
R5
for six octahedrally arranged ligands of charge Q at distance R. This gives a
reduction in the value of 9 from the free-ion value go.Such reductions are observed
experimentally, but are traditionally interpreted in terms of a covalency effect.
One approach is to note that plotting out the wave functions de and dt, shows that
for the t, orbitals the overlap with the six ligands is larger than that for the e orbitals.
A rough way to introduce covalency is thus to slightly expand the t, orbitals from
their free-ion size, while leaving the e orbitals unchanged. T h e result is to reduce
the value of 27(tz) even when the isotropic operator is used to represent the spin-
orbit coupling. While covalency effects are undoubtedly needed to obtain the
qualitative occurrence of phenomena such as transferred hyperfine interaction, it is
possible that at least part of the reduction in 27 may be explained by a simple crystal
field approach using (3.28); further numerical work is needed to decide this. T h e
procedure of expanding the t, orbitals was originally used by Koide and Pryce
(1958) for the S state ion Mn2T(d5),and that case provides an interesting example of
an accidental degeneracy. In the crystal field approach the states 4Ag,4Egarising
from the free-ion *G term have the same energy in a cubic field. T h e same result
holds for the strong field approach, in which the states arise from the tZ3e2con-
figuration (Stevenson 1965). If, however, the t, orbitals are supposed to be slightly
radially expanded relative to the e orbitals then the (*A,, 4Eg)degeneracy is removed,
explicitly demonstrating that the original degeneracy was not imposed by the
octahedral symmetry. For a very small expansion the theory of Koide and Pryce
predicts the *A, level to be below the 4E, level; this seems to be in disagreement
with experiment for some salts containing Mn2- (Tsujikawa and Kanda 1963,
Ferguson et al. 1966). For the Co2+ ion in 0, symmetry (as in MgO) the 41;(T,)
Group theory and topology in solid state physics 5 73

orbital triplet is the lowest level, but is split by spin-orbit coupling to give levels
with pseudo-j values of 4,#,Q. T h e splittings between these spin-orbit levels should
theoretically obey a Land6 interval rule, and t h e j = 4 doublet is lowest. T h e doublet
can be split by a magnetic field, and the energy levels for the,j = 4 manifold may be
described by an operator which is equivalent to a 2 x 2 matrix. Since the unit 2 x 2
matrix and the three Pauli matrices form a complete set for the representation of any
2' x 2 matrix, it follows that the 'spin' Hamiltonian is a linear combination of the
spin operators S,, S,, S, for an effective spin S = 4. This is true no matter what the
further mixing produced by weak noncubic parts of the crystal field. I n this case
the effective spin differs from the true spin (#) and, in fact, equals the pseudo-j
value.
3.2.3. The Dirac identity; exchange forces. The operator equivalence which has
probably had the most far-reaching consequences in solid state theory is the Dirac
identity (Dirac 1958)
- p1 2(orb) E p12(spin) '2(1 + 4S1. S2). (3.291
Equation (3.29) describes the effect of a transposition 1-2 of the labelling of the
spin variables in a many-electron product function, and shows that the spin operator
on the right-hand side produces the same effect. T h e equivalence may be verified
by operating on two-electron product states with total spin 0 and 1. T h e -P12(Orb)
on the left-hand side of (3.29) may only be included for operands which are anti-
symmetric for the transposition 1 ++ 2, and this means that orbital permutations can
be represented by an equivalent spin operator. Dirac (1958) uses two kinds of
permutations operators in his discussion, label permutations and place permutations.
Thus the label permutation P12when acting on the function
I a>, I I I
b>3 C>Z 0 4 (3.30)
(we take four particles as an example) would interchange particle labels 1 and 2,
whereas the place permutation P12'would interchange the labels of the first and
second orbitals (a standard ordering of the orbitals being assumed). Dirac points out
that P12'is only meaningfully defined for product states such as (3.30); more
recently Ohnuki and Kamefuchi (1969) have stressed that place permutations are
not well defined even then if the particles concerned obey statistics other than the
Fermi or Bose type. Their paper is one of several which have investigated the
theoretical properties of systems of identical particles for which the state vectors
belong to reps of S, other than the identity (boson) or alternating (fermion) rep.
For a many-electron system we may use either label or place permutations with
product states of type (3.30). If a fixed set of orbital functions is taken, for example
the four functions of (3.30), and one particle is assigned to each orbital with up or
down spin then in the most general case (all orbitals different) there are 16 x 24
states of type (3.30) when label permutations are allowed for. If the Coulombic
electron-electron interaction V is 'turned on', matrix elements of V will arise
between some of the 24 states. However, only antisymmetric linear combinations of
states of type (3.30) correspond to allowed physical states, and these form a subspace
in the linear space spanned by the 24 functions. I n this subspace the full equa-
tion (3.29), including P l 2 ( O r h ) , is valid, and the result is that V may be replaced by
the effective spin-dependent operator (Dirac 1958)
(3.31)
5 74 J . Killingbeck

(apart from a constant term), where is the matrix element of V between states
in which orbitals j and k have been interchanged. (It is characteristic of anti-
symmetrization procedures that sums over particles may be replaced by sums over
orbitals.) Dirac’s proof holds for a very special model, and we may ask if anything
can be said about the way in which the eigenvalues of (3.31) depend on the total
coupled spin S of the system. Koster (1955) showed that if a , b, c, d are distinct and
orthonormal it is further possible to prove that the lowest energy state is one of
maximum multiplicity. His proof rests on the fact that Tk for the Coulomb inter-
action is positive when orbitalsj and k are orthogonal. We give a simple example
here to illustrate the idea behind his proof. The states I L , S , L,, Sa) of the f 2 con-
figuration may be expressed in terms of determinantal states ; for example,
L-;
I 3H,5,1> = {3,2}
(3.32)

O n introducing the Coulomb interaction the energy difference E(l1) - E(3H) is


found to be the exchange integral (321 V123). Since this is positive (Racah 1942)
it follows that E(l1) > E(3H). T h e usual Slater theory (Condon and Shortley 1935)
gives for the energy difference the quantity SOF, + 60F4+ 14F,, which is also
necessarily positive. T h e argument given here uses the fact that both 3H and lI
contain states arising from the (3,2) orbital pair, and so it cannot be extended to give
an energy ordering for all the terms of f2. T h e ordering of some of the terms
depends on the ratios of the Fkparameters. This fact is most notable for the con-
figuration g2; Judd (1967 a) pointed out that the use of hydrogenic 5g ratios for the
Fk gives E(3H)< E(3K), showing that Hund’s rules (in so far as they refer to the L
value of the ground term) are not universally valid even for configurations of
equivalent electrons. Suppose that the fixed set of orbitals can be regarded as made
up of two fixed subsets, associated with two different atoms. For example, two
Mn2+ions in their 6S ground states have the fixed occupied I, orbitals (2,1,0, - 1, - 2)
in the weak field approach. Ferguson et al. (1966) have shown that even in the strong
field approach several of the experimentally important levels of Mn2+arise from the
same t,3e2 strong field configuration. T h e terms in (3.31) for which j and k refer to
different atoms will now yield an interaction term which corresponds to an effective
magnetic interaction between the atoms. T h e mathematical arguments involved
here are the same as those used for the spin-orbit coupling within the atom, and the
analogy may be set out schematically as shown below.
l,s+L,S: z g i l i . S i= h L . S
(3.33)
sl,s, + S,, S,: Jkl sk .s1= JS, . S,.
I n a manifold in which the total spins S , and S , of the individual atoms are given,
the interaction term takes the form JS, . S , for some constant J. Strictly speaking,
the assumption has been made that the orbitals on different sites are orthogonal,
which is not true if the interatomic effects are worth bothering about! Lowdin
(1962 c) has pointed out that if orthogonalized atomic orbitals are used in an
attempt to simplify the formalism, then J must always be positive (together with the
exchange integrals Jkl),and the resulting interaction is always of ferromagnetic type
T o obtain an antiferromagnetic contribution to the interaction energy it is then
Group theory and topology in solid state physics 575

necessary to allow mixing in of higher configurations, such as states in which an


electron has transferred between the atoms. T h e second-order energy corrections
then yield an antiferromagnetic contribution to J, and they are dominant for the
majority of insulators, I n the perturbation series for the energy there may arise
terms referring to the simultaneous exchange between the atoms of a pair of
electrons with parallel spin. Allan and Betts (1967) pointed out that such terms may
be formally described by that generalization of (3.29) which applies for spins
S = 1, namely
PI*= (S1.S2)2+(S1.S2)-1. (3.34)
T h e result is a small biquadratic contribution which should be added to the bilinear
spin Hamiltonian (3.33), although such terms are not found to be of important
magnitude experimentally. T h e interaction operator (3.33) has been widely used
to describe the magnetic interaction between paramagnetic ions in magnetically
dilute solids, although its use is not always rigorously justifiable. Stevens (1953 a)
has investigated the theoretical form of the magnetic interaction between ions when
each ion has one electron, which may be in either of the two states of a Kramers
pair, and spin-orbit coupling is present. T h e resulting operator includes aniso-
tropic terms as well as an isotropic term of form (3.33). Such anisotropic terms have
been investigated by later workers (e.g. Moriya 1960, Bertaut 1969, Levy 1969).
Levy and Copland (1969) have treated the case in which the orbital magnetism of
each ion is not quenched, and have pointed out that the usual procedure (Mattis
1965) of replacing Si and Sj in (3.33) by (gJ,- l)Ji and (gJ,- 1)4,gj being the
LandCg factor, does not fully describe all the contributions to the isotropic exchange
interaction. T h e operator (3.33) is the basic model Hamiltonian for the theory of
ferromagnetism and antiferromagnetism. l l u c h work has been done on the
theoretical consequences arising from this model operator, but it is still a subject
of research to show that for a metal the operator is derivable from first principles
within the context of band theory. Several authors discuss the problems (Ruijgrok
1962, Anderson 1964, Mattis 1965, Beeby 1967). I t should be noted that the
exchange integral J of the original Heisenberg theory of ferromagnetism is not the
same as the Jof the approach outlined above, since the Heisenberg integral involves
the full Hamiltonian and not just the Coulomb interaction. Anderson (1964)
pointed out that the Heisenberg integral may have either sign, although it does not
accurately describe the magnetic coupling at all interatomic distances. Stuart and
Marshall (1962) evaluated the Heisenberg integral for iron and found it to be much
smaller than the phenomenological J value required by experiment. Lowdin
(1962 c) has discussed the way in which the Heisenberg integral differs from an
exchange integral which is defined to give the spin energies correctly. T h e ways in
which exchange interactions between atoms and molecules can be systematically
described by perturbation theory have been reviewed by Herring (1966) and
Jansen (1967).

3.3, The second quantization formalism


3.3.1. The occupation number representation. I n recent years several books have
treated the general second quantization formalism (e.g. Dirac 1958, Abrikosov et al.
1963, March et al. 1967, Mattuck 1967, Ziman 1969) while others have particularly
dealt with solid state applications (Kittel 1963, Anderson 1964, Mattis 1965) to the
theory of electrons, phonons and magnons. Our present interest is with the group
5 76 J . Killingbeck

theoretical and matrix theoretical aspects of the formalism, and only a few key
principles will be discussed.
T h e way in which the name ‘second quantization’ arose is made clear by a
study of Dirac’s paper (Dirac 1927) on the theory of radiation. However, the name
is not properly descriptive of the modern approach to the subject ; what is essentially
involved is the use of an occupation number representation for a quantum mechanical
many-body system. T h e most obvious group involved in the theory of an AT-
electron system is S ,, the symmetric group. T h e usual wavefunction approach
involves choosing the physically realizable wave functions to belong to the one-
dimensional antisymmetric or symmetric rep of s,. I n the case in which the
Hamiltonian has symmetries other than S,, the eigenfunctions of H ill be linear
combinations of determinantal product states; nevertheless we may use a single
determinant as operand when setting up the occupation number representation. If
a list of the one-electron basis states is made, then a determinantal state may be
specified by giving the occupation numbers 0 or 1 of each state. T h e determinantal
states may thus be symbolically represented by ket symbols of form I 101.. .) which
display the occupation numbers. It is now useful to define equivalent operators
which, when acting between the symbolic kets, give the same matrix elements as
would be obtained by using the traditional operators between determinantal wave
functions. T h e well-known result is as follows; for a two-body operator,
(3.35)

where the operators C ~ , Care the creation and annihilation operators for the one-
electron states. C; = (ci)t, and the ci have the properties
ciIn, ...nd...) = ( - l ) x n i l n l . . . n i - l . . . ) (3.36)
where is the sum of the occupation numbers (in the operand) for states occurring
before state i in some fixed conventional listing of the orbitals. T h e operators
obey the anticommutation relations,
(3.37)
Kobe (1965) has presented the second quantization formalism concisely in terms of
the theory of a graded Hilbert space.
As shown in the standard texts, the use of equations (3.35) with the symbolic
kets is simply expressing the S, antisymmetry requirement in terms of the operators
rather than the state vectors. For example, in the simple case of a two-electron
system, the Coulomb interaction energy would take the form, for the state {a, b},
< I r, Glmn Cl,+ Clk+ c, I
c, I ) = (ab V I ab) - (ab I V I ba) (3.38)
since c, cb = - cb c,, etc. The exchange term thus arises exactly as in the traditional
method which employs determinantal functions. I n this sense there is nothing new
in the occupation number language, but there is the procedural advantage that the
mathematical problems are now concentrated on the algebra of the operators.
This means that the calculations become independent of N , the number of electrons,
although this may be specified in the last numerical stage of a calculation. T h e
occupation number approach is thus suited to the description of general many-body
effects which occur in many-particle systems regardless of the exact value of N
(assumed large). Heisenberg (1931) used the occupation number approach in his
paper investigating the concept of holes in a many-electron system. He showed that
Group theory and topology in solid state physics 577

the Schrodinger equation for an atom with n electrons missing from a closed shell
is equivalent to a Schrodinger equation for n positive holes, except for a slight
change in the central potential. He also showed that a band with n electrons
missing responds to external fields as if it contained n holes. Both these results are
only correct if configuration interaction (or band-band mixing) is ignored, and
Heisenberg pointed out the way in which this is achieved in the occupation number
formalism. We give here an extended version of his comment. Suppose that the
single-particle states are those belonging to some central potential, so that we have
some specific ground configuration, e.g. 3dn plus closed shells, in the absence of the
Coulomb interelectronic repulsion. T h e Coulomb repulsion when introduced leads
to the effective Hamiltonian
I:(kZl Vlmn)cltcktc,c, (3.39)
where the sum is over all one-electron orbital indices, including those referring to
higher configurations. If now the sum in (3.39) is restricted to indices referring to
the ground configuration, the resulting theory is equivalent to that which involves
a traditional matrix diagonalization of the Hamiltonian within the manifold of
antisymmetric states arising from the ground configuration. A simple way to see
this is to recall the discussion of $3.1, noting that the formal infinite sum in (3.39)
can be retained if we use a modified potential which has no matrix elements between
the ground configuration and higher configurations. If the states from two bands
are included in the sum (3.39) and the Coulomb interaction is present, then the
resulting theory is capable of describing electron-hole pair exciton phenomena
(Kubler 1967). As another example, R'lcCumber (1964) studied Hamiltonian
operators of the form
Eocot co + E1(clt c1 + c,+ c2} + I:{cn a,+ + c,+ a,}. (3.40)
n,q

Here the fermion operators co, (c1, c z ) , refer to the states of a nondegenerate and a
degenerate level, respectively, for an electronic system, while the a,, a,+ are lattice
phonon (boson) operators, which obey commutation relations with one another
rather than anticommutation relations. Expression (3.40) describes the effect of
the lattice coupling to the two-levels, assuming that phonon coupling between 0
and 1 is neglected. T h e last term in (3.40) will give rise to a Jahn-Teller effect in
the degenerate 1 level. Optical transitions between 0 and 1 levels will show a
phonon structure due to the coupling, and McCumber shows that the absorption
spectrum has a line shape given by the Fourier transform of correlation functions
of type
(3.41)
where a phonon ensemble average is implied by the symbol {n,), and where N is the
occupation number of the state j . T h e use of such correlation functions or of
related Green functions is very common in the occupation number formalism (e.g.
Kittel 1963). Kubler (1967) has used such techniques to discuss excitons, and
Richmond and Sewell (1968) have used them in a discussion of the Mott insulating
state.

3.3.2. Transformation properties of the ck. McCumber (1964) discusses the sym-
metry properties of the operators appearing in (3.40) for the case in which the
electronic levels belong to an ion at a lattice site with some point group symmetry.
578 J . Killingbeck

T h e general principle is that the terms in H should remain invariant under all the
group operations; in this case we must ask how to use the usual group theoretical
arguments for operators such as c, ct. By using the spectral representation of the
standard quantum theory we may write a one-electron operator in the form
v = Clk><kl Y I W I (3.42)
Comparison with (3.35) suggests that the operator t k should be assigned the trans-
formation properties of the wave function +k* while ck+ transforms as +k, and this
result is confirmed by checking the other formulae of the theory. Armed with this
basic principle, it is possible to form coupled products of creation and annihilation
operators by analogy with coupled products of wave functions. For example, Judd
(1963) considers tensor operators with the reduced matrix elements
(nl /I d k ) 11 n'Z') = (2k + 1)1'2 s n , s,, (3-43)
where the nl are one-particle quantum numbers for motion in a central field. T h e
relevant group is R,, and the operators are essentially a coupled product family,
= C C(1lmm' I k m ) cTLlntt
vm(k) cnlmr (3.44)
where the operators create or destroy one-electron states with quantum numbers
I)
nlm and the C( are vector coupling coefficients. Judd showed the way in which
the tensor operators so defined can be grouped to form Lie algebras when their
commutation relations are calculated, and how the resulting operator families may
thus be regarded as the generators of various continuous groups which are useful in
classifying the wave functions of many-electron atoms. These developments were
originally due to Racah (1942). Judd (1967 b) later presented some of the theory in
an explicit second quantized form, and this second reference contains an account of
several important applications of the second quantization formalism. Wybourne
(1968) used the more general tensor operators Vck)(l, Z') = (c,+ x c?),~(~) in a treat-
ment of configuration interaction in atoms. His idea was to construct an effective
operator (in the spirit of $3.2) which represents within one configuration the effects
due to mixing with other configurations. T h e particular variant which he employed
was in the treatment of the sum

(3.45)

which expresses mixing effects to the second order. One common approximation
is to use closure in (3.45) to yield a matrix element (01 V210). Some average
excitation energy must then be guessed to use as a denominator, and this procedure
is known as the Unsold approximation. Wybourne instead suggested splitting up V
into a sum of operators of form ( V ( k lx) V ( k z ) ) (so
k )that each operator links IO) only
with one excited configuration. If we denote the portion of Y which links IO) and
In) by V,, we have
I;(Eo - En)-1 (0 I &'on2 I 0).
n
(3.46);

(3.46) may give a more accurate estimate of (3.45) than that obtained by the Unsold
approximation. We point out here that, while second-order contributions to the
energy from higher configurations are additive, third-order ones need not be so in
general, since linked chains of form (01 VI a ) (nl VI m) (ml VI 0) can occur. This is
relevant to the discussion preceding equation (7) of Wybourne's paper.
Group theory and topology in solid state physics 579

If the spin is included in the specification of the one-electron states, then the
operator c t for such a state I lsm,m,) must be regarded as a double tensor operator
(Judd 1963, Armstrong 1968). By the use of such operators much work has been
done on deriving relationships between matrix elements within different con-
figurations; in the traditional theory in which the wave functions were emphasized
such relationships were derived by the use of determinantal wave functions and
coefficients of fractional parentage. Since our concern is with group theoretical
aspects we give only one example, the pair operator of Watanabe (1966). Consider
the operator
A= c CVnl,,; C-,,,,-m,t(
ml;m8
- l)ml+ms++ (3.47)

As far as the orbital magnetic quantum numbers m are concerned, (3.47) is a second
quantized scalar product, on formally taking:,c to transform as qml(c,,t xmz). N

As far as spin is concerned, (3.47) is also a scalar product if c + l , , t ~ a c-l,;-P.


, The
operator A thus appears to have the following properties; it adds two electrons to
the system; it leaves L and S unchanged. I n the terminology of the traditional
I
theory, A produces when acting on a term 11" L S ) a term ln+2L S ) , and thus gives
rise to a chain of functions with the same seniority. As a simple example, the
operator A for a d shell is
A = [c2$ c-,$ - C1$ CMIJ.t + co+t CJ - c-1l.t c,$ (3.48)+c 4 t c,J
as is found by using (3.47) and the anticommutation relations (2.37). T h e con-
figuration d3 contains two 2D terms, and the one with seniority 1 can be generated
from the ,D state of dl.
- +- -
Alioooo) = lfoooi)-l
ti-
iioio)-l iioio)+lioloo) (3.49)
i
where the occupation numbers refer to the five d orbitals and 1 denotes a doubly
occupied orbital. I t is possible to use the usual vector coupling coefficients to con-
struct other operators which link states differing by two electrons. For example, the
simple operator c,+t c2+t has L = 4,L, = 4, S, = 0 by the vector coupling rules, and
should generate a state of d3 with L, = 5 and L = 5 or 6 when acting on the operand
zk
01000) belonging to dl. I n fact we have the correct result,
+ *+
cz+ c2+~01000)= I 11000). (3.50)
I n some cases the operator cannot be constructed; for example, an attempt to
construct the coupled operator with S = 0, L = 3, L, = 3 from the d shell operators
gives an operator which is identically zero by virtue of the anticommutation rela-
tions (3.37). T h e pair operators can be used to produce states of configuration such
as d7 or fll, belonging to shells which are more than half filled. However, the usual
procedure is to use a configuration of holes in such a case; the relationships between
configurations of n particles and n holes has also been discussed by means of the
second quantization procedure (Watanabe 1966, Armstrong 1968).
3.3.3. A boson representation f o r angular momentum. An attempt to construct an
operator with S = 1, S, = 0 from two spin i operators in the spirit of the last para-
graph would lead to cl.t c+t + c+t eft, which is zero by the anticommutation rules.
The situation can be saved by using annihilation operators, and the three operators
3s.+
= 7icj+t cj+ sj-= hC*J ci+ SSj= @(C,f+ c*+- cj+t Cj+) (3.51)
5 80 J . Killingbeck

do have the required properties S = 1. I n fact these three operators obey the
traditional angular momentum commutation rules, and also the operators for i # j
commute if we suppose the fermion operators for i#j to anticommute. Equa-
tions (3.51) thus provide a fermion representation of the angular momentum
operators for a set of independent particles, and are used in the theory of magnetism
(Mattis 1965). A representation of the angular momentum operators in terms of
boson (harmonic oscillator) operators is also used in the theory of magnetism.
This is the Holstein-PrimakofS representation used in the theory of ferromagnetic
spin waves at low temperatures (Mattis 1965). T h e idea of the representation is as
follows. If each spin has 2 S + 1 states, we may formally think of the fully ferro-
magnetic state of a lattice as the standard state, and regard the quantity S - (Si,)
as the spin deviation at a site i. T h e spin deviation number can be used to label the
eigenstates instead of S,, and with this viewpoint the angular momentum commuta-
tion relations become
(n, I Si,I n, + 1) = - i(ni I Si, I n, + 1) = ${(n,+ 1) (2s- ni)}’l2
(3.52)
(n,ISi,In,) = S-n, (O<ni<2S).
In (3.52) the kets are specified by giving the numbers n, for each site, but the
operators S, act only on those for the ith site. If we had a system of oscillators, with
canonical coordinates and momenta pi,P, and commutation relations
[pi,41 = is,,,
then we could think of the numbers n, as oscillator occupation numbers. The
usual harmonic oscillator rules then give
I I
(n, pi S’/g ni + 1) = I
- i(ni I Qi S’/Qni + 1) = ${(ai + 1)2S}”%
(3.53)
(nil S - $(GZ+ Qiz- 1)I ni) = S-n,.
These are not quite the same as (3.52); however, for n,<S, i.e. for tiny deviations
from the ferromagnetic ordered state, they suggest that a boson quasiparticle picture
might be useful. I n fact, on using the Heisenberg Hamiltonian, and replacing the
spin operators by oscillator variables as suggested by equations (3.52) and (3.53),
the resulting Hamiltonian may be linearized to resemble that for lattice vibrations.
T h e only new feature is the presence of an effective momentum coupling term Pi
between neighbouring sites. This does not affect the translational invariance
properties of the equations, and the Hamiltonian may be diagonalized to give spin
waves with definite k-vectors. T h e resulting theory is valid in the low-temperature
limit Enk<NS where N is the number of sites and n k the number of magnons in
the mode k. At higher temperatures, terms must be included to represent the
difference between (3.52) and (3.53)) and they lead to a so-called ‘kinematic inter-
action’ between the spins, in addition to the ‘dynamic interaction’ caused by small
extra terms in the original Hamiltonian for the spin system (Dyson 1956, Kenan
1967). There are various other formal devices for the representation of angular
momentum operators in terms of fermion or boson operators (Mattis 1965, Kenan
1967). Schwinger used such a representation to give an elegant approach to the
theory of angular momentum; his work is briefly described by Mattis (1965) and is
published in full in the book edited by Biedenharn and van Dam (1965). T h e book
of articles by Moshinsky et al. (1968) contains a detailed account of the group
Group theory and topology in solid state physics 581

theoretical aspects of the second quantization approach to many-body theory, but


the applications given are mainly to the theory of nuclei.

3.4. Symmetry properties f o r many-electron systems


3.4.1. Generalized Unsold’s theorem. If the one-electron orbitals of a system are
classified according to the reps of some symmetry group, then a determinantal
function in which each state of a rep is filled by two electrons (with opposite spin)
gives a charge density with A, (identity rep) symmetry for that group. This result
is a generalized form of Unsold’s theorem, which asserts that a closed atomic shell
is spherically symmetric. Furthermore, such a determinantal function has total
spin S = 0, as is most easily verified by operating on it with the three total spin
operators S,,St. For example, the four traditional tetrahedral sp3 hybrids of the
carbon atom (Pauling 1960), when each doubly (or singly) occupied, give a
spherically symmetric charge distribution about the carbon nucleus, exactly as if the
hybrids had not been formed. I t is still the case, however, that the total energy,
including exchange and Coulomb repulsion within the carbon atom as well as
bonding effects via the hybrids, leads to a tetrahedral minimum energy configuration
for CH, and analogous systems. As Simpson (1962) has pointed out, if correlation
effects are included, the most probable arrangement for the four sp3 electrons would
involve a tetrahedral array, although the orientation of the tetrahedron would be
arbitrary in the absence of ligands. This concept of correlation as giving a ‘shape’
to a free atom is also briefly discussed by Pauling (1960); Luder (1967), following
Linnett (1964) has pressed the idea to the limit by using such atomic shapes in a
geometrical discussion of the bonding between atoms. His approach presumably
implicitly involves some dubious assumptions about the relative magnitudes of the
intra-atomic correlation energy and the interatomic bonding energy. T h e generalized
Unsold theorem has some relevance to the simple pictorial view of the static Jahn-
Teller effect. As pointed out by Clinton and Rice (1959) it is instructive to consider
the Jahn-Teller effect in terms of the forces exerted on the ligands by the charge
distribution of the central ion. From the viewpoint of the physicists who prefers
pictorial arguments, this procedure gives a link with the pictorial approach to
crystal field splittings (commented on in $2.5). Thus, if an ion has the strong field
ground state t,3 or t,3 e2 in cubic symmetry, the resulting charge distribution has full
cubic symmetry provided that each level of t, and e is occupied. T h e resulting
forces on the ligands thus contain no symmetry-violating components to give a
Jahn-Teller effect. Furthermore, if the ion undergoes a transition to a higher state
belonging to the same strong field configuration, no change takes place in the basic
charge distribution symmetry. This means that no marked vibrational effects are
to be expected; it is the case that such transitions give comparatively narrow lines
in absorption experiments (e.g. Ferguson et al. 1966).
3.4.2. Selfconsistency and symmetry. T h e A, symmetry of the projection operator
associated with a subspace which spans a full rep of a group has been noted in
$3.1. I n this subsection we comment on an extended form of this result which is
applicable to the Hartree-Fock theory. T h e projection operator can be written as
follows, when acting on some operand $,

(3.54)
582 J . Killingbeck

where p(r, r’) is the Dirac density matrix (Seitz 1940, Ter Haar 1961), defined by
the equations above. T h e Hartree-Fock equations €or a determinantal trial function
can be written as follows (Seitz 1940);

I n (3.54) the Dirac density matrix is defined by summing over the occupied orbitals
in the determinant. If these exactly fill one rep (or a whole number of reps) for a
group, the Dirac density matrix and also the effective one-electron operator in
(3.55) have A, symmetry. I n this case it is thus possible to obtain a self-consistent
solution of the Hartree-Fock equations in which the orbitals are symmetry adapted,
usually for the symmetry group of the one-electron potential V . This self-con-
sistency result has been discussed for particular groups by Seitz (1940) and Lowdin
(1962 b). For example, in an atom with a closed shell configuration, the Hartree-
Fock equations may be solved self-consistently by putting each electron in a
traditional central field orbital with angular factor Yp(6’,4). For the Hartree
equations, with no exchange terms, this is not so without some kind of spherical
averaging process, since the potential acting on each orbital is the closed shell
potential minus that of one orbital. In the case in which V in equation (3.55) has a
lattice translational symmetry, a self-consistent solution results on simply filling
up Bloch functions of definite k with opposite spins. This gives a value for the direct
and exchange energies which is directly calculable for the ‘jellium model’ in which
a metal is represented by a free electron gas with a uniform positive background
(Kittel 1963, Ziman 1965). This model gives unsatisfactory results for the band
width and density of states at the Fermi surface (Kittel 1963), but one of the
‘jellium’ results has been widely used in atomic calculations (Herman and Skillman
1963) and in the band theory of magnetism (Slater 1968). This is the result that the
exchange energy for the free electron gas, while due to a nonlocal operator, can be
simulated by using a local energy operator proportional to p1/3,p being the local
charge density. For the free-electron gas p is constant, of course, but the p1i3
approximation has been widely used as a simple method of estimating exchange
energies in many-electron systems in which p does not vary too rapidly. T h e
approximation has been used in calculations which are outside the closed-shell
single determinant Hartree-Fock formalism ; for example, in the band theory of
ferromagnetism the ground state is required to have more spins up than down, and
so cannot be the usual paramagnetic state with doubly occupied Bloch orbitals. I t
is possible, however, to perform self-consistent calculations which lead to such a
magnetized state, and this is done by using ( p f ) ” 3 and ( p $ ) ” 3 respectively, in the
4
exchange potential terms for the .f‘ and Bloch orbitals.
3.4.3. Modified Hartree-Fock theories. We now discuss some developments of the
Hartree-Fock theory, starting from equation (3.55). Firstly, while the para-
magnetic (doubly occupied Bloch orbital) state gives a self-consistent solution of
(3.55) when V has lattice periodicity, it is conceivable that there are other self-
consistent solutions with lower energy (Adams 1962 a). For example, Bloch (1929)
showed that if all the spins of a free-electron system were up, then a self-consistent
solution of (3.55) would result, with a lower energy than the paramagnetic state at
high densities. Overhauser (1962) showed that for the free-electron gas there is
another self-consistent solution of the equations (3.55) with lower energy than the
Group theory and topology in solid state physics 583

paramagnetic state, and he termed this a spin density wave (SDW) state. I n this
state the charge density is uniform, but the spin density shows a wavelike variation
with a wave vector roughly equal to twice the Fermi wave vector. Correlation
effects and solid state lattice periodicity effects would modify the results in a manner
which is not completely certain, but experiment suggests that the antiferromagnetic
ground state of chromium may involve such an SDW state (Overhauser 1962). (For
later work on the SDW state see for example Penn and Cohen 1967, Berggren and
Johansson 1968.) Suppose now that we use a single determinantal function in which
each orbital in a rep is filled once, but the spins are in some arbitrary arrangement.
T h e resulting equations (3.55) may be satisfied self-consistently by making each
spatial orbital belong to a rep, but the spatial function for up and down spins should
differ, since the up and down spins have different exchange terms acting on them.
(In Overhauser’s work the one-electron functions were not taken to be separable into
spatial and spin factors, but this separability is imposed in all other work described
here.) T h e approximation in which the up and down spins are allocated the same
spatial orbital is called the restricted Hartree-Fock method, while, following Watson
and Freeman (1960), the approach in which this requirement is not imposed is
termed the ‘spin-polarized’ method. When the reps filled by electrons are all one-
dimensional, the restricted and spin-polarized methods are the only two which
need to be discussed; however, for an open shell in spherical symmetry there is also
possible an unrestricted Hartree-Fock method, in which the radial one-electron
functions are allowed to vary with I,. I n fact the restricted method for an open shell
specifically requires a spherical averaging of the potential, and so gives perfect self-
consistency only for the spherically averaged potential, and not for the individual
one-electron effective potentials. For the atomic case it is also common to assume
that the one-electron radial functions are the same for all terms of a configuration
(Freeman and Watson, 1962), although a less restrictive calculation by Synek and
Corsiglia (1968) for Nd3+ indicates that this is not strictly correct. This particular
assumption of fixed radial functions within a configuration is central to the widely
used Slater-Racah theory which involves the electrostatic Fk integrals, and within
that theory would be explained in terms of configuration interaction effects. T h e
Freeman and Watson (1962) paper did include one calculation for the Ce3+ ion in
which the dependence of the radial functions on J was calculated after inclusion of
spin-orbit coupling effects. The problem of including spin-orbit coupling in self-
consistent field calculations has been discussed more recently by Condon and
Odabasi (1966). T h e unrestricted method leads to certain qualitative effects which
seem to be appropriate for the explanation of some experimental results, although a
detailed numerical check has not been carried out for all such cases. For example, to
explain the epr results on 0,- (Zeller and Kanzig 1967) and S- (Vannotti and
Morton 1968) ions in alkali halide lattices, it was found necessary to use different
values of (r-3) for the spin and spatial parts of the orbitals when evaluating the
hyperfine interaction. This effect is possible within the unrestricted method but not
within the restricted method, T h e former method also makes it possible for nega-
tive spin densities to occur in certain regions of space, and this is necessary to explain
some magnetic resonance experiments on organic molecules containing 7~ electrons
(see e.g. Streitwieser 1961, -Amos and Snyder 1964).

3.4.4. Projection operator methods. I n the case of either the spin-polarized method
or the unrestricted method, the determinantal trial state used will generally not be
5 84 J . Killingbeck
an eigenfunction of total spin, and perhaps w7ill not belong to a particular single rep
of the spatial symmetry group concerned. This symmetry problem leads to a choice;
either the single determinant is used, leading to equations ( 3 . 5 9 with a later use of
projection operators to project out functions of the desired symmetry, or the pro-
jection is made first, the variational procedure then being used for the projected
function. T h e second process, the so-called extended Hartree-Fock scheme,
yields more complicated equations than ( 3 . 5 5 ) , and is really only feasible for small
systems such as the Li atom (Goddard 1968). Most numerical work employs the
first alternative, so that the results obtained will not be the very best possible. For
example, it is possible to form a determinantal function using Hartree orbitals; the
resulting function will usually have a lower energy than the original Hartree pro-
duct function, but not quite the same energy as would result from solving the full
Hartree-Fock equations. I n the case of the atom the LS eigenfunctions are linear
combinations of determinantal states, and a slight difference will be obtained if one
determinant is used in the variational principle instead of the full L S state. Con-
sider the case in which a trial function can be written in the form
(3.56)
where the functions [ A ) belong to the reps of the group G(H)of the Hamiltonian.
Then we have
( 3 . 5 6a)
from which it follows that at least one of the components I A) of [ 4 ) will give a
lower expectation value for H than that produced by the trial function itself. I n
some of the unrestricted Hartree-Fock calculations the projecting out of the correct
symmetry components IA) has not been carried out before evaluation of the final
total energy. Amos and Snyder (1964) pointed this out, and described an approxi-
mation in which only the major contaminant contribution I A) is removed from the
trial determinant. T o explain this comment we must look at the projection operators
introduced by Lowdin (1964); these operators take the form below, S2being the
total spin operator for a many-electron system,

(3.57)

T h e operator Os,, when acting on a determinantal function, eliminates one at a time


all the components with total spin different from S’ (Lowdin 1964). If a trial
determinantal function contains many total spin components, it is very tedious to
apply the operator (3.57) in full. Amos and Snyder (1964) pointed out that for a
determinant with a given S, value, it is usually the case that the components
S = S, and S = S,+ 1 are dominant; in this case a good approximation to the
S = S, component is obtained by using only that factor of Os, which refers to
k = Sa+1. A discussion of the amplitudes of the various spin components in a
determinant was given by Sasaki and Ohno (1963). If L2 is used instead of S2in
( 3 . 5 7 ) , states of given total orbital angular momentum can be projected from a
determinant. If the determinantal states are written in an occupation number
notation similar to that of § 3 . 3 , they may conveniently be expressed in binary code.
Rotenberg (1963) has used this to construct a computer program which projects
out atomic LS functions from determinantal starting functions by acting on them
with operators of type (3.57). I n recent years many authors have dealt with
Group theory and topology in solid state physics 585

projection methods for obtaining appropriately symmetrized many-electron func-


tions from simple starting functions (Matsen 1964, Zeh 1965, Goddard 1967, Gallup
1968, Musher and Silbey 1968, Sullivan 1968, Smith and Harris 1969). One method
which involves projection methods is the alternant molecular orbital (AMO)method,
which has been applied to alternant hydrocarbon molecules (Lindner and Lunnell
1967, Pauncz 1967) and in some cases to solid state problems (Dermit 1962, Calais
1965, Larson and Thorson 1966). T h e method is designed to allow for correlation
effects in those systems (such as a bcc lattice) which consists of two types of equiva-
lent sites A and B, each site of one type having neighbours only of the other type.
In a Hartree-Fock function there is some correlation between parallel spins, so
that most of the remaining correlation error is due to electronic repulsion effects
between electrons of opposed spin. The idea of the AMO method is to use a trial
wave function in which the up and down spins are predominantly on separate sub-
lattices, the extent of this separation being adjustable by means of a variational
parameter. T h e resulting optimum state has a reduced energy because of the built-
in correlation, although, of course, after the correct projection the total spin density
on A and B sites may be zero. T h e point is that while an electron is equally likely
to be up or down at an A site, its presence there increases the probability that any
electron with opposite spin will be on a B site rather than that A site. I t is possible
to build this into the ALTO function while at the same time preserving near each site
the predominantly parallel alignment of spins which is described by Hund’s rules
for isolated atoms (Dermit 1962).
T h e comment made above about projections is also relevant to the Overhauser
SDW state. While the SDW state satisfies the Hartree-Fock equations it is still not
an eigenfunction of total spin, so that, for example, a state with S = 0 could be pro-
jected from it; for such a state there would be no fluctuating spin density. From a
more physical viewpoint we could say that the position of the origin (i.e. the phase
of the wave) does not affect the energy, so that there ought to be an equal likelihood
of each possible phase, giving a state with no spin density fluctuations. If the
projected ground state has Sf0, however, some fluctuations will remain. I n the
solid state it is conceivable that the effect of the lattice locks in an SDW state with a
specific phase as the ground state. T h e SDW and AMO approaches clearly have
various similarities, but differ in some respects. Firstly, they were designed for
systems with high- and low-electron densities respectively. Secondly, the wave
vector of the SDW is mainly determined by the Fermi momentum of the electron gas,
which does not have a universal ratio to the reciprocal lattice spacing. Thirdly,
the Overhauser picture, being based on the Hartree-Fock equations, may be expected
to be modified by correlation effects, whereas the AMO attempts to make allowance
for these.
Most of the discussion above has concentrated on the symmetry conditions
associated with the Hartree-Fock equations and their extensions, but a final
comment is made here on the matrix theoretical interpretation of the Hartree-Fock
theory. T h e orthonormal one-electron orbitals which are occupied in the Hartree-
Fock ground state are such that the total Hamiltonian H has no matrix elements
between the ground state determinant and those determinants obtained by replacing
one orbital by an orbital which is orthogonal to it (Brillouin’s theorem). If H con-
tains only one- and two-particle operators, this means that only doubly excited
states link with the ground state Hartree-Fock function. I n this sense the Hartree-
Fock theory yields a particular complete orthonormal basis in which the matrix
5 86 J . K i l lingb eck

of H is simplified. (The orthonormal product states outside the occupied set may
be chosen in any convenient manner.) If the Hartree-Fock determinant were an
exact eigenfunction of H , of course, even these matrix elements to doubly excited
states would vanish, and in some cases (Anderson 1964) it can be argued that they
are small. For an open shell atom the Hartree-Fock equations are usually solved
by some kind of spherical averaging of the potential. This means that Brillouin's
theorem should not be exactly valid for the resulting functions, although it should
still be a useful guide. For example, Wong (1963) calculated the energy shift of the
levels of the Pr3+(4f2)configuration due to mixing with 4f6p to be only about 10% of
the shift due to mixing with 5d2. If the orthonormal occupied orbitals in the
Hartree-Fock determinant are replaced by orthonormal linear combinations of
themselves, the determinantal wave function is unchanged, but the orbitals in general
will no longer satisfy Brillouin's theorem and will not obey the Hartree-Fock
equations. This is discussed by Goddard (1967); Adams (1962 b) developed an
approach in which the one-electron basis consists of linear combinations of the
standard orbitals which satisfy the Hartree-Fock equations.
If an attempt is made to improve the Hartree-Fock ground state by first-order
perturbation theory, Brillouin's theorem leads to the result that doubly excited
determinantal states should be mixed into the Hartree-Fock determinant. Those
terms in which only the orbitals of electrons 1 and 2 are different from their ground
state forms can be grouped together to give a pair function u(l,2), and the first-
order correction to the wave function takes the form

(3.58)

where d is the antisymmetrizer and the 4k are the ground state orbitals. T h e pair
correlation functions u ( i , j ) have been discussed by Sinanoglu (1961), and several
authors have discussed the application of perturbation theory to Hartree-Fock
wave functions (Sinanoglu 1961, Lowdin 1962 d, Sharma 1968, Yin and Silverston
1968). Steiner (1968) has discussed the symmetry properties of the u ( i , j ) and of the
correlation functions which occur in higher-order terms in the perturbation series.
A very good account of correlation effects for atoms, molecules and the electron gas
is given by the articles in the recent volume edited by Lefebvre and "loser (1969);
several of the articles employ the second quantization formalism. Condon (1968)
has reviewed the recent approaches to the correlation problem in atoms.

3. S . Class sum operator approach to finite group theory


It was pointed out in $2.5 that the familiar concept of the addition of linear
operators leads to a group algebra when the operators form a group. T h e most
familiar members of this group algebra are the traditional projection operators of
finite group theory. T h e two commonly used types of operator are of the form

(3.59)
Group theory and topology in solid state physics 5 87

When acting on a function + h ( j ) belonging to the hth row of the j t h rep these opera-
tors give (Falicov 1966)
pP U (i)+ h ( j ) = 823, . 8uh +/ L (i) (3.61)
P ( i )$ h ( j ) = Sij # h ( j ) . (3.62)
If a subspace is invariant under the operators R of the group, then it will contain
functions +*(j) for all h if it contains one of them. I n that case the Pfiu(i)operators
with fixed v and variable p will be able to project out a full basis for the rep J?(i),
If the operand picked on happens to belong exactly to the vth row of the
PFu(i)will generate from it the full basis and are thus analogous to the shift operators
L, of the rotation group R,. Pci) acting-on such an operand would simply reproduce
it, so that some other operand in the invariant subspace would have to be chosen.
T h e character operators are easier to apply, but the functions which they
produce have to be sorted out into orthonormal linearly independent families by
an auxiliary calculation following the projection process. T h e class sum operator for
a class of nk elements in a finite group is defined by

( jin class k). (3.63)

Dirac (1958) used these operators in his study of the symmetric group S ,. Lowdin
(1967) has given an account of several important operators in the group algebra, and
Hall (1967) has given an excellent account of the class sum operators. Hall’s book
is particularly recommended to those physicists who are accustomed to the Dirac
(1958) viewpoint, which stresses the algebra of operators. Heine (1960) briefly
discussed the relationship between the Dirac viewpoint and that of group repre-
sentation theory. Killingbeck (1970) has set out the basic principles which link the
two approaches, and has applied them in a treatment of the group D3h, which is
the appropriate group for the description of the crystal field at rare-earth impurity
sites in various solids. The class sum operators (3.63) commute with one another,
even for a non-Abelian group, and they may be regarded as a family of commuting
operators of the type employed in the Dirac approach to quantum mechanics. T h e
product of any two class sum operators is expressible as a sum of class sum operators,
and this makes it possible to find the eigenvalues of the operators. With each rep of
the group is associated a set of eigenvalues for the class sum operators. Within
an irreducible subspace it is further possible to diagonalize individual group opera-
tors, and thus to construct a modified group character table which displays the
possible eigenvalues for each class and each rep. Killingbeck showed how the
resulting information leads to an easy method of analysing the Kronecker product
of reps. At this point we shall concentrate on another aspect, namely the relation-
ship between the finite group projection operators (3.60) and the Lowdin projection
operators (3.57) for the continuous group R,. T h e former operators project out a
function of the type required, while the latter knock out one at a time the unwanted
components. There is no essential difference, however; if the class sum operators
Vk have eigenvalues X k ( f i ) for the yth rep of a finite group then the analogue of
Lowdin’s operator is
n(Vk -
V+P
(3.64)

if all the A,@) differ for the reps p occurring in the operand. Since any polynomial
in the Vk can be reduced to a linear combination of the Vk,the operator (3.64) in
588 g. Killingbeck
general reduces to one of the traditional projection operators .of finite group theory.
However, if the operand is known to contain only specific rep types, the operator
(3.64) can be specially adapted to the operand. For example, if the operand con-
tains only two rep types, for which some Vk has different eigenvalues h ( l ) A(,),
, then
the single operator Vk- h(l)will suffice to separate the components. I t is also possible
to use the group operators themselves to form tensor operators belonging to some
(but not all) of the reps of the finite group (Gamba 1968, Killingbeck 1970). There
are still several interesting problems connected with the class sum operator approach.
Apart from such specific research problems, however, the present author feels that
the approach could be valuable in the postgraduate teaching of group theory to those
students who are accustomed to the Dirac approach to quantum mechanics.

3.6. Crystal jield theory


3.6.1. Recent experimental work. Many of the examples used in the discussion of the
preceding sections have been drawn from crystal field theory. T h e intention of this
subsection is to emphasize further some of the recent theoretical and experimental
developments in crystal field theory itself, with the group theoretical approach as a
guide. 1Zs a few examples of recent experimental advances we may quote the
following : the measurement of the free spectra of highly ionized rare-earth atoms,
so that comparison with crystals spectra becomes possible (Dieke and Crosswhite
1963, Sugar 1965, Dupont 1967; for a study of 4fn-4F-l5d spectra in crystals see
Loh 1968) ; the direct infrared measurement of crystal-field level-energy separations
of about 100 cm-I, which otherwise are obtained indirectly from optical spectra
(Slack et al. 1967, Bloor et al. 1968, Wheeler et al. 1968); the measurement of
properties of 5fn and 5dn type ions in crystals, which requires extension of the
theory of 4fn and 3dn configurztions to the case of strong spin-orbit coupling
(Dsrain and Wheeler 1966, Rahman and Runciman 1966); experiments in which the
ionization state of a paramagnetic ion is changed by bombardment of the solid
(Merz and Pershan 1967; Sabisky and Anderson 1967); investigation of the epr
of metastable excited states of paramagnetic ions in crystals (Sabisky and Anderson
1966, 1967) which leads to accurate g values for such levels; lLlossbauer effect
measurements on ions in crystals (Eck et al. 1967, Siegwarth 1967, Jesson et al.
1968), some of which give information about the crystal field symmetry and magni-
tude. A review of experimental work up to about 1965 is given by Wybourne (1965).
3.6.2. Classification of free-ion levels. All versions of crystal field theory, whether for
d or f electron systems, start from a consideration of the energy levels of the free ion.
T h e theory of the classification of the energy levels of the free ion has itself been the
subject of much work; some of this has been mentioned in earlier sections, but the
topic is vast enough to merit a complete article on its own. T h e major group
theoretical principle involved is that the many-electron ionic wave functions for the
terms of an f n configuration can be labelled by rep labels for the groups R, and G,
as well as by the usual R, rep labels L and S. This was pointed out by Racah (1942),
and the use of these groups in the evaluation of matrix elements of the Coulomb
interaction, spin-orbit coupling and other operators has been extensively developed
by later workers. T h e R,, G, quantum numbers may not be good quantum numbers
in the presence of the interelectronic Coulomb repulsion forces, but this is not
important when compared with the advantages gained by using the systematic
classification. Other groups could be used to classify the f n states, but, as pointed
Group theory and topology in solid state physics 5 89

out by Wybourne (1965), the (R,,G,,R,) scheme is more convenient for the
evaluation of matrix elements. As a simple example, by classifying the tensor
operators for the Coulomb interaction in terms of their R, transformation pro-
perties, it may be shown that the energy separations of the f" L S terms of maximum
multiplicity are numerical multiples of the combination
E 3 = Q(5FZ + 6F4- 9 1Fe)
of the standard Slater parameters. T h e Racah parameterization in terms of
EO, E l , E 2 ,E 3 is now often used instead of the traditional F,, F,, F4,F, parameteriza-
tion for f n configurations. Elliott et al. (1957) have given the group theoretical
classification of the states of the f n configurations, together with their energies, as a
multiple of F,, on the assumption that the ratios F41F,and F,/F, may be set equal to
their values for a hydrogenic 4f orbital. These ratios as determined semi-
empirically do not deviate very far from the hydrogenic ratios, although the one-
electron radial wave functions themselves seem to be noticeably non-hydrogenic.
T h e group theoretical interpretation of fractional parentage coefficients and of
Wigner 6j and 9j symbols has been investigated at a formal level in recent years
(Holman 1969, Moshinsky and Devi 1969); from the viewpoint of applications,
however, the most important development is the appearance of numerical tables of
these coefficients (Nielson and Koster 1964) and symbols (Rotenberg et al. 1959),
since these facilitate the evaluation of the matrix elements of tensor operators
between f" type wave functions. T h e theoretical techniques mentioned above have
mainly been developed for f" configurations ; they are applicable to dn configurations
also, but the latter can often be treated by simpler methods. T h e spectra off" ions
in crystals are usually much sharper than those of dn ions, so it is worth while to
consider for f" ions the small effects associated with spin-other-orbit coupling
(N'alli 1968), orbit-orbit coupling (Malli 1967), and spin-spin coupling (Malli and
Saxena 1969). Configuration interaction effects of the Coulomb interaction have
also been considered (Rajnak and Wybourne 1963), and any such configuration
mixing can be represented by using an effective operator in the sense of 5 3.1. This
effective operator viewpoint has been applied in detail by Wybourne (1968). I t
sometimes happens that the total effective Hamiltonian in a manifold contains
operators of similar form from first and second order, and a semi-empirical approach
cannot separate these. For example, Araki (1948) showed that within an LS term
spin-spin coupling would modify the Land6 interval rule to become
E ( J )- E ( J - 1) = aJ+ bJ3. (3.65)
Equation (3.65) does fit some of the experimental results, but Judd (1956) pointed
out that a similar equation results from the second-order spin-orbit coupling
between some LSJ multiplets. Judd further showed that the numerical values of
6 for the spin-spin first-order effect were much smaller than the required values.
As another example, Trees (1964) showed that several observed spectra could be
better fitted if a term aL(L + 1) were included in the Hamiltonian in addition to the
Fk parameters and the spin-orbit parameter 9. Such a term can arise from orbit-
orbit coupling in first order, but can also arise as a result of configuration interaction
with some excited configurations (Wybourne 1965).
3.6.3. The crystaljield expansion. T h e original work of Bethe (1929) supposed the
crystal field at a site to be a static potential field with some definite point group
5 90 J . Killingbeck

symmetry. This assumption alone leads to the well-known results for the rep labels
and degeneracies of the levels arising when the crystal field splits a state of given
angular momentum. If spherical polar coordinates are used to describe the potential
field, then it can be expanded in the form

V ( r ) = X:fT(Y) y n YZ(0,$1. (3.66)

Prather (1961) gives tables which show those m, n terms in the expansion which are
nonzero when the point group symmetry of V ( r ) is given. When the resulting Y;
terms are used as operators in a basis of states of definite angular momenta, they may
be treated by means of the Wigner-Eckart theorem ($3.1) or may be written in
terms of equivalent operators (Stevens 1952). T h e usual tables of such operators
may also be used in spin Hamiltonian theory ($3.1) if L is replaced by S , and thus
Prather’s tables may be used in conjunction with the spin Hamiltonian theory of
Koster and Statz (1959). T h e particular symmetry-adapted combinations of the
function y n Y z which belong to the A, rep of the various point groups have been
given by several authors (Von der Lage and Bethe 1947, Bell 1954, Altmann and
Cracknell 1965) and it is in principle possible to expand any scalar field of a given
point group symmetry in terms of the appropriate symmetry-adapted harmonics.
T h e practical use of a finite-series expansion of type (3.66) was studied by Houston
(1948), and his method has been applied by several authors (Horton and Schiff
1959, Miasek 1966, Bennett 1967). For the case of crystal field theory only a few
terms of the series (3.66) are needed in any case; for example, within an f n con-
figuration those terms of (3.66) with n>6 would yield a zero matrix element, as
would the odd parity terms with an odd n value. T h e final result of setting up the
secular matrix of V ( r ) in an L S or LSJ manifold is thus a set of energy levels with
energies which depend on the expectation values (ynfZ(y)) taken over the radial
one-electron functions of the f shell. I n the special case in which the field V ( T )is
produced by a set of point charges, the functionsf;(Y) become independent of Y ,
and the contribution of a specific point charge varies as R-(n+l),where R is the dis-
tance of the point charge from the origin; it is usually considered adequate to use the
interior ( Y < R ) solution only. T h e standard expressions for the terms of the series
(3.66) in the point charge model have been given by several authors (Low 1960,
Hutchings 1964) and are widely used in the literature to estimate crystal field para-
meters. As noted by Bradbury and Newman (1967), the ratios of the factors
attached to the Yp(0,C$)will be the same for fixed n regardless of the force law which
describes the potential field of the ligands, provided that the ligand effects are
additive, and that the potential due to each ligand is axially symmetric about the line
joining the ligand to the origin. We shall use group theory here to set their result in
a broader context. We assume that the scalar field V ( r ) belongs to the A, rep of a
point group G , and that it may be expanded as follows:

V ( r )= c F;(r) YE(0, +) (3.47)


where we take the Y;(O,C$)to be integral normalized to unity. We then have

F ~ ( Y=) P(r) YE8(0,+) d0 d+. (3.68)

Suppose now that the rep D(%)of R, is such that the identity rep 4 , of G occurs k
Group theory and topology in solid state physics 591

times (k = 0 , 1 or 2 for the cases of interest). Yz(B,+)may be written as


k
+
YF(0,+) = E ol(j)m+(j)(A1) other reps. (3.69)

This gives the result

FF(r) = 2 n ( j ) mJ+(j)*(Al)V ( r )d0 d+ (3.70)


j=1

since V ( r )belongs to the A, rep. If k = 0, F z ( r ) is zero; this criterion gives the set
of nonzero terms in the crystal field as tabulated by Prather (1961). If k = 1, the
ratio of the various F z ( r ) for fixed n is independent of the detailed form of V ( r )and
so may be obtained by means of a Coulombic point charge model, even if such a
model does not give the correct magnitudes of the F ~ ( Y )For . k = 2, the ratio may
not in general be independent of the detailed form of V ( r ) ,but will be so for the
special cases pointed out by Bradbury and Newman (1967), as mentioned above.
As an example we may consider 0, symmetry and the terms with n = 4 in the
expansion (3.67). The general function of fourth order may be written in the
alternative forms
+ + +
f 4 ( r ){A(x4+ y 4 z4) B(x2y2 x2z2+ y 22))
2A-B
= f4(Y) { - T
+y4
(x4+ z4 - 33Y4 3A+B
)+-+4),. (3.71)

For a Coulombic point charge model the potential obeys the Laplace equation, f ( r )
is a constant, and B = -3A. This equality also holds for the case discussed by
Bradbury and Newman, but in the more general casef4(r) may not be a constant and
the r4 term may not vanish. T h e point is that when V ( r )is not harmonic the terms
of type and the terms of nth order in x , y , z are no longer necessarily the same.
This is shown by the equation which describes how the squared angular momentum
operator2Z2acts on a homogeneous function F, of order n in x , y , z (Killingbeck and
Cole 1970):
2Z2FF,(x,y,z)= (n(n+ 1)-r2C2)F,(x,y,z). (3.72)
Group theory shows that a cubic field cannot contain a D(2)component. However,
a general non-Coulombic cubic field can contain a term r2, which is of D(O) type,
not D(2)type. Terms such as rn are isotropic and give no crystal field splittings.
O n lowering the symmetry from 0, to D, (tetragonal) the general D(4)term in V ( r )
may be written as a linear combination of a term with FE ratios arising from a model
point charge calculation and the cubic term f4(r) (x4+y4+ z4- $r4). This decom-
position would be particularly adapted for use with a cubic symmetry-adapted
basis. I n general, terms other than that with the point charge F z ( r ) ratio can arise
when D(n)contains the A, rep of G more than once. Such terms would presumably
arise if an attempt were made to describe the nonlocal overlap and exchange effects,
as well as the effects of the continuous ligand charge distribution, by means of a
local potential field V ( r ) . Such a V ( r )could be meaningful in the sense of pseudo-
potential theory, and it may be possible to allocate it in a semi-empirical manner for
complexes involving regular series of ions and ligands. T h e precedent of the Born-
IVIadelung theory for ionic crystals suggests that it may be possible to assign a
specific pseudopotential to each ligand and then evaluate the coefficients in (3.67).
This does not seem to have been attempted, although Phillips (1959) suggested a
5 92 J . Killingbeck

method of calculating crystal fields which is based on the ideas of pseudopotentia1


theory, and a pseudopotential approach to colour centres has been given by Bartram
et al. (1968). (Since writing the first draft of this manuscript, the author has received
a reprint of a paper by Newman and Stedman (1969) which does carry out a
parameterization of rare-earth crystal fields similar to that suggested above.)
Various workers have concluded that the ratios of the FE(r) for fixed n are quite
well given by the Coulomb point charge model, and it is also the case that the
correct reversal of crystal field energy levels is observed to occur when a dn type ion
is studied in both octahedral and tetrahedral surroundings. Even in such a case the
crystal field parametws should involve the quantities ( f E ( r ) rn) over the radial
wavefunction instead of the (Y").

3.6.4. Euen parity terms; energy level splittings. T h e even parity terms in the
expansion (3.67) will have matrix elements within a given dn or f n configuration, and
give rise to crystal field splitting of the free ion terms or multiplets. I n a basis of
I
states J M ) or I L M ) the operator Yg(8, +) will satisfy the usual angular momentum
selection rules. Furthermore, the condition that the V ( r ) be real (essentially the
time reversal symmetry condition) leads to relationships between the pair FE(r) and
F-F(r'), and in this way the levels resulting from the matrix (lnJlylI VI l n J ' M ' )
are ensured to have the correct theoretical degeneracies. If the crystal field has a
principal axis of order AT,only terms with m = 0, i-AT, i- 21V, etc. will be nonzero
in (3.66), and thus the resulting operator will only link states I J M ) with '%1 values
which differ by these m values. Thus the resulting eigenstates can be given a
crystal field quantum number (Hellwege 1949), equal to M ( m o d IC'). Further
symmetries in the group may suppress some of the terms allowed by the symmetry
of the axis. For example, the group C3h includes a threefold rotation as well as a
sixfold rotation inversion. T h e condition that a term YE should be invariant under
both these operations (i.e. under the latter operation) shows that m must be a
multiple of 3 and that n - m must be even. Similar results hold for D3h, but with
one difference. T h e presence of the dihedral axes in D,, makes it possible to chose
a dihedral axis as x-axis in such a way that all the terms in (3.66) may be written as
cosines (for n even) or sines (for n odd) of the azimuthal angle 9. Such a simplifica-
tion is not in general possible for C3h; however, if only the terms Y!j,Y i , Y:, Y:6
are employed in calculating crystal field splittings the Yt6 terms can be taken to
give a single cos6+ type term for both C,, and D3h, This amounts to a definite
choice of x-axis, and Murao et al. (1967) have investigated the physical effects which
can distinguish between C3h and D,, symmetries if quantities other than the crystal
field splittings are measured. YbC1,. 6H,O provides an example of a similar effect
(Eisenstein 1961). T h e site symmetry of the Y b 3 ion ~ is C,, and the fitting of the
crystal field levels could proceed without any specific choice of x-axis. However,
the Zeeman effect shows a typical x-axis splitting when the magnetic field is per-
pendicular to the geometrical twofold axis. This disagreement between the geo-
metrical and physical principal axes led Eisenstein to suggest that YbC1,. 6H,O is
ferroelectric. In the case of D,, symmetry the crystal field quantum numbers p
are equal to M (mod6), but are usually chosen to be 0, 2 1, + 2 , 3 . Since the ,u
classification refers only to the properties of the principal axis it will not in general
lead to as complete a simplification of the crystal field secular matrix as the use of the
full rep classification for D3,. This has been pointed out by Wybourne (1965).
However, Hellwege (1949) actually introduced a secondary crystal quantum
Group theory and topology in solid state physics 593

number v, essentially the eigenvalue rt 1 of the operation 2,. v is well defined for
the one-dimensional reps of Dah, and the ( p , v ) classification is equivalent to the
D,, rep classification. For example, Wybourne (1965) points out that p = 0 states
belong to both A, and A, reps. T h e A, and A, reps have different v labels, and so
are distinguished in the ( p , v) notation which is widely used by Hellwege's group to
label their experimental crystal field levels. The effect of time reversal symmetry is
to make pairs of levels c p stick together in such a classification.

3.6.5. Oddparity terms; absorption intensities. T h e odd parity terms ( n odd) in the
symmetry-adapted expansion (3.66) do not cause crystal field splittings within an
f n or dn configuration of definite parity, but mix it with higher configurations of
opposite parity. T h e electric dipole moment operator will then have a nonzero
matrix element between states of the nominal f n or dn configuration. T h e crystal
field spectra thus contain Laporte-forbidden transitions which cannot be seen in the
free-ion gaseous phase. As an example, consider a dn ion in a tetrahedral field.
Such a field contains a YB2part, together with higher-order ( n > 6) odd parity parts.
If, on a point charge approach, we think of the Y z 2part as dominating the odd parity
field, then this will link the dn states to states of type dn-lx, where x is some odd
parity type, most importantly p. T h e matrix element of r between the nominal
states I dTL I
L) and dn L') will then include terms of type

(3.73)

SO that the resulting electric dipole transitions will obey, besides the necessary T,
rep selection rules, further R, selection rules e.g. AL < 4,A S = 0. For a rare-earth
ion in T, symmetry the effects of spin-orbit coupling will lift the A S = 0 rule and
the other rule will become AJ< 4. I n general the terms in (3.73) can be represented
by an effective operator within the dn or f n configuration, and it is clear that the
resulting operator must be a combination of just the same even parity operators
which give the crystal field splittings. T h e most general dipole transition selection
rule due to the mechanism described here will thus be A J < 6 for f n type ions and
AL < 4 for dn type ions, although spin-orbit coupling effects may modify the latter
by mixing the different L states slightly. If the appropriate closure approximations
are made in the sum over L" and x in (3.73), then the same effective operator will
describe the transition amplitudes between all terms of the configuration. T h e
operator will consist of a linear combination of TZ type tensor operators, arising
from the Kronecker product r x J&, and the coefficients of each T g can be deter-
mined semi-empirically, in much the same way as the standard crystal field splitting
parameters are treated. The theoretical treatment of intensities according to this
approach has been investigated in detail by several authors (Judd 1962, Ofelt 1962,
Judd 1966, Wybourne 1968) and has been compared with experiment (Axe 1963,
Carnal1 et al. 1965). If the static field is of holohedric point symmetry, then odd
parity ligand vibrations are needed to give odd components in V ( r ) . The observed
spectrum then consists only of phonon-assisted lines. Vibrations may also occur for
the hemihedric case, of course, and Judd (1962) considered them in his paper,
showing that they still lead to the same kind of effective operator to describe the
intensities. Koide and Pryce (1958) also invoked lattice vibration effects to explain
optical transitions between levels of the S state ion Mn2+; Liehr and Ballhausen
(1957) had previously treated the intensities for T i and Cu by considering odd parity
5 94 J , Killingbeck

vibrations. Hagston (1967) pointed out that when the mixing of higher states into
both the states I A ) and I B ) involved in a transition is taken into account then the
dipole matrix element has two terms

(3.74)

When a mean energy denominator AE is used in a closure approximation, this sum


is approximated by

However, Hagston noted that, although the operators r and V,,, commute, the
inequality
I c
( A r I m> ( 4VI B ) # ( A I V Z I m> (ml r l B ) (3.76)
will hold if the Im) do not form a complete set. I n particular, he showed that the
approach of Koide and Pryce (1958), which replaces r V + Vr by 2rV when the I m )
form a finite set of p orbitals, need not give the same transition selection rules and
amplitudes as would result from (3.74). Kiel (1966) has applied the closure pro-
cedure to obtain effective operators within the ground configuration when the odd
parity field is an externally applied field. Kid's work treated only the direct effect
of the field on the paramagnetic ion wave functions; Bates (1968) gave a theoretical
treatment which also allowed for the indirect effect due to the field-induced dis-
placement of the lattice ions, which produces an extra contribution to the apparent
field at the paramagnetic ion.
3.6.6. Cozalency effects and shielding ejfects. T h e theory of the transition proba-
bilities involves coupling of the ground dn or f'L configuration to configurations in
which one electron is in an orbital of larger radial extent than that of the d or f
orbitals. Such an orbital should have a strong overlap with the ligand orbitals, and so
we are led to some kind of molecular orbital theory; although this problem is a
genuine solid state one, it is often skirted for lanthanide ions by keeping to the model
?n which the ion has its own free ion excited configurations intact (i.e. the crystal field
sources are taken as being effectively at infinity). T h e molecular orbital approach
would regard some of the excited states in the sum (73) as charge transfer states, and
this more realistic approach has been taken for the intensity theory of transition ions
by Englman (1961) and Balt (1968). Covalency effects may also be taken into account
in the theory of the crystal field splittings within the ground configuration of the ion.
This LCAO-MO approach has been widely applied to first transition series ions, and
has been represented in semi-empirical treatments by the use of the orbital reduc-
tion factors which multiply the orbital angular momentum matrix elements com-
puted as if for the pure central ion function (Stevens 19.53 b, Low 19.58), or which
regard the ionic wave functions as expanded by some scaling factor on passing from the
free ion to the crystal (Marshall and Stuart 1961, Jorgensenas 1962, Shrivtava 1969).
Covalency need not be treated only in the molecular orbital model, but also in
the configuration interaction approach. Hubbard et al. (1966) have adopted this
approach and point out that it is better capable of allowing for correlation effects
than the usual molecular orbital method. They also showed that the 4s orbitals
play a role in the covalency of Mn2+. Bartram et al. (1968) used a configuration
interaction approach to the theory of g shifts for the F centre in alkali halides, I n
Group theory and topology in solid state physics 595

recent years the molecular orbital approach has also been applied to lanthanide ions
(Jorgensen et al. 1963, Axe and Burns 1966, Thornley 1966, Watson and Freeman
1967), the usual procedure being to take a strong field basis set consisting of
symmetry-adapted linear combinations of the central ion f orbitals and the ligand
orbitals. Phillips (1959) suggested that an estimate of the effect of the ligands can
be obtained by specifically using a modified central ion orbital obtained by Schmidt
orthogonalizing the free central ion orbital to the ligand orbitals. This ortho-
gonalization process is widely used in the calculation of transferred hyperfine inter-
actions (Schmid 1966, Spaeth 1966); in the ligand field theory of the rare earths
it has been used, with various modifications and extensions, by Ellis and Newman
(1967) and Ellis (1968). Intermediate between simple point-charge crystal field
calculations and full molecular orbital calculations are those approaches which stick
to a potential field of appropriate symmetry but allow it to be modified in some way.
For example, if the electrostatic potential due to the extended electronic charge
cloud is calculated in detail, the resulting potential field will differ from a simple
point charge one. This kind of calculation was undertaken by Kleiner (1952) and
Freeman and Watson (1962), and has been carried out for rare-earth ions in crystals
by Raychaudhuri and Ray (1967). Further, if the resulting field is regarded as
arising from sources which are 'external' to the ion, then the filled shells 5s26p6of a
rare-earth ion will shield the internal f shell from the field. The shielding effect has
a dominant part which is linear in the field, and the various Y z terms are thus
reduced by factors which depend on n, but not m.
We may see this as follows. If we consider mixing in by the crystal field of only
those configurations in which the 5s26pGshells have been disturbed, and if we label
the state of the 5s25pGshell (or its rearrangement) by the quantum numbers J M ,
then the perturbed functions take the form
I n J M ) (nJJ4I Y-Yl00) (3.77)
100>-C
n En - Eo
where the unchanged f shell quantum numbers have been suppressed, and n labels
the excited configurations. T h e resulting charge density is to first order

(3.78)

Thus the Y y term in the crystal field produces yj*iw terms in the distorted charge
distribution and hence in the potential due to that distribution. T h e real crystal
field contains equal amplitudes of YJ*-Ifhowever, and the result is an equal reduc-
tion in both the YJ*-'l field components acting on the 4f electrons. Since
(nJMI YJfl 00) is independent of M and -Eoo may also be taken to be inde-
pendent of M with high accuracy, it follows that the reduction factor for the field
components Y y is independent of 144, so that the ratios Y,"/Yp' are not affected by
the screening. Watson and Freeman (1964) have pointed out that the distortion
of the outer closed shells can also in principle give rise to nonlinear effects; to
represent these the apparent crystal field for the 4f shell should be augmented by
small terms not corresponding to the original external crystal field symmetry. I n
the literature this nonlinear effect is taken to be negligible in comparison with the
linear shielding effect. Sternheimer (1966) also calculated the exchange contribu-
tion arising from the distortion of the outer closed shells, and found it to be very
small compared with the direct electrostatic potential of the distorted shells. T h e
596 J . Killingbeck

screening effect has been estimated by Burns (1962) and by Lenander and Wong
(1963), the latter authors obtaining much greater screening effects (70% for Y i ,
as compared with lo?; from Burns’ results). A more recent calculation by Stern-
heimer et al. (1968) seems to confirm the higher value. Newman and Curtis (1969)
have argued that the interaction between 4f electrons should also be reduced by a
closed-shell screening effect, and have suggested that the neglect of such many-body
effects leads to too large a variation of the radial wave functions from term to term
in the Hartree-Fock Nd3+ calculation of Synek and Corsiglia (1968). Newman and
Curtis point out that the variation of (rn.>values from term to term predicted by the
Hartree-Fock results implies a much greater term dependence of the crystal field
parameters than is found from a fitting of observed crystal field levels for Nd3++.
Eisenstein (1963) suggested that for the P r 3 ~ion in PrCl, the variation of the wave
function from term to term is responsible for the apparent term dependence
of crystal field parameters. We comment here that experimental workers on
luminescence often use the concept of the configurational coordinate diagram
(e.g. Curie 1963), in which the ion-ligand distance is regarded as depending on the
ionic LS term. From this point of view a variation of crystal field with the state
of the ion will necessarily occur. T h e present author is not aware of any detailed
study of this specific problem; Eisenstein makes the comment that the experiment-
ally observed sharpness of the lines in PrCl, implies that the error is not in the term
dependence of the crystal field but in that of the central ion wave functions. T h e
same comment may be made about screening calculations as was made about
intensity calculations, nameIy that they involve free ion orbitals which cannot exist
as such in the solid state. Thus the 5s2p6shell orbitals will directly overlap the
ligand orbitals and lead to molecular orbitals. Bishton et al. (1967) have taken this
more realistic solid state model and have concluded that the apparent crystal field
parameters for the 4f shell are also reduced in this model; they also included
ligand-ligand overlaps, which are almost universally neglected in the literature.
Bishton et al. found the 5s2p6 overlap effect to be largest for the Y : parameter,
while Ellis and Newman (1967) showed that the direct overlap between the 4f
orbitals and the ligand orbitals causes a contribution to the crystal field parameters
which is greatest for the sixth-order parameters.

3.6.7. Orbit-lattice coupling; Jahn-Teller eflect. 3Cluch recent effort in ligand field
theory has been devoted to the theoretical and experimental study of the effects of
ligand displacements or vibrations. T h e vibronic structure associated with elec-
tronic transitions has long been known (Hellwege and Hellwege 1952) and has been
studied in more detail in recent years (e.g. Yamamoto et al. 1967). T h e effect of
symmetry lowering distortions of the ligands around a paramagnetic ion has been
investigated theoretically in terms of simple crystal field theory (Fick 1962, Stevens
1967) and of covalency theory (Bates 1964, Englman 1966, Zdansky 1967). T h e
distortions may be produced experimentally by the application of pressure or
directed stress (Zahner and Drickamer 1961, Burns and Axe 1966, Hughes and
Runciman 1967) or by the application of an external electric field which will produce
distortions in an ionic crystal (Kiel 1966, Williams 1967, Bates 1968). In some
cases it is possible to study the change in the vibronic structure which occurs when
different isotopes are used for one of the types of atom in the lattice (Hughes 1966).
I n addition to these static effects, the dynamic Jahn-Teller effect has been intensively
explored (Ham 1967, Hochli 1967, Coffman 1968, Bates and Dixon 1969). If it is
Group theory and topology in solid state physics 597

remembered that the wave functions used in ordinary crystal field theory should
really include vibrational factors, then the matrix elements of the theory should
include vibrational overlap integrals as factors; Ham (1965) has pointed out that
the dynamical Jahn-Teller effect can then lead to a large reduction in the effective
matrix elements of orbital and spin operators between the ionic states. We shall
mention in detail only a few works, which have in common the fact that they use
equivalent operators or effective operators to represent the effects of the lattice on
the ionic energy levels. Several authors have discussed the inclusion of lattice
displacement effects in the spin Hamiltonian formalism (Mattuck and Strandberg
1960, Ray et al. 1966, Bottger 1967), and Ham (1967) has discussed in group
theoretical terms the way in which the epr g factors and zero field splitting para-
meters vary with the ligand displacements. T o deduce the numerical coefficients
of the operators involved would require a perturbation treatment (like the Pryce
one outlined in 8 3.1) including the spin-orbit coupling, since the ligand displace-
ments themselves affect the orbital wave functions by changing the crystal field.
If the crystal field is described in a point charge model, explicit expressions for the
extra field due to the ligand movements may be obtained (Fick 1962, Stevens 1967)
giving a unified theory of the crystal field splittings and of the orbit-lattice coupling.
Covalency effects may modify the details, but such a unified theory should give a
reasonable semiquantitative guide. For example, O’Brien (1965) showed that the
point charge model coupling of an E, type vibration to the jD(T2,) electronic state
of Fe2+ in cubic symmetry is considerably greater than that to the 5D(E,) state,
although the point charge ratio was still too small to agree with experiment. Orbach
(1961) used rough order-of-magnitude arguments based on traditional crystal field
theory to estimate the orbit-lattice coupling parameters in his treatment of spin-
lattice relaxation in rare-earth salts. Manenkov and Orbach (1966) have edited a
book of collected papers on spin-lattice relaxation.
T h e basic principle involved in the evaluation of the crystal field due to the
ligand displacements is the invariance of the total energy terms under simultaneous
group operations applied to electronic and lattice coordinates (Ballhausen (1962)
erroneously states that the crystal field factor alone belongs to the A, rep; if this
were so there would be no Jahn-Teller splitting effect). If the displacement-
induced part of the crystal field Hamiltonian is written as
AV = gz(y)yp(e, 4) ~
m,n,A,v
pK,,,A~L
) (3.79)

where Q 2 v ( A ) is the amplitude of the vth component of the (A) rep type normal mode,
then the nonzero terms in (3.79) will be such that the Y p and Qv(h)belong to the
same rep of the site symmetry group. For example, the even parity vibrational
modes of six ligands situated octahedrally about a paramagnetic ion are of types A,,
E and T,. T h e form of the even parity part of AV is thus
...I+
AV = a~Q(Al)+b{(3x2-r2)Q(E8)+J3(x2-y2)Q(EE))+c(yxQ(T2~)+ ...
(3.80)

where the sum is over all reps A, and over even-order harmonics, n = 0,2,4. ..
T h e K are the (p-independent) coupling constants, and the P & ) ( h ) ( n ) ( r ) are
symmetry-adapted functions with the correct (A, p) type transformation properties.
For an E type manifold the E type mode is the only effective one, whereas for T,
26
598 J . Killingbeck

or T, states both E and T, modes are effective. On using the generalized Wigner-
Eckart theorem the matrices of the E and T, type operators within the T, manifold
are determined except for constant factors; if the T, type states are classified in
accord with L = 1 effective angular momentum viewpoint ($3.2) then the orbit-
lattice coupling takes the equivalent operator form (with A, included)
A S = AL(L + 1) Q(A,) + B{L,2- +(Lz2+ L,'))) Q(E0) + J 3 B 3 ( L z 2- LU2)Q(EE)
, ) ~ ( T. , ~ ) (3.81)
+ C { ( ~ , ~ , + ~ , ~ , ) Q Q ( T 2 ~ ) + ( ~ , L , + L , L...}+... +
Such equivalent operator forms for the orbit-lattice coupling terms are given by
Van Vleck (1960), Fischer (1967), Ham (1967) and Stevens (1969). -4s pointed out
by Stevens, there could in principle be a term involving products of type
Q(T,x) ( L z )but time-reversal forbids this. In the Jahn-Teller theory only the even
parity normal modes are considered; nlenne et al. (1968) have given the odd parity
terms in the orbit lattice interaction for the case of eightfold cubic coordination.
When both b and c are nonzero in (3.80), the calculation of the eigenstates of the
Hamiltonian is difficult. T h e inequality b $ c is often applicable (Sturge 1967),
so that the term involving c can be treated by a perturbation approach after the
problem has been solved using the b term. Stevens (1969) has recently used a
transformation approach in which the Hamiltonian is transformed to yield an
effective Hamiltonian in which the E type terms have been replaced by an energy
shift term. If c is set equal to zero, and if terms in Q2 and P 2 are added to (3.80)
in order to represent ligand potential and kinetic energy, then three distorted
equilibrium ligand configurations are found. These correspond to x, y and x tetra-
gonal distortions of the ligand octahedron, with a wave function equal to a product
of an electronic wave function and a ligand harmonic oscillator function centred on
the tetragonal equilibrium position. If the energy barriers between the three
equilibrium positions are large, then the situation is conveniently regarded in terms
of tunnelling effects. Bersuker (1963) suggested that the small coupling between
the three degenerate tetragonal states of minimum energy would produce a further
splitting into a doublet and a singlet. Ham (1965) commented that the three
electronic-vibronic states concerned transform as T, and should not split under any
perturbation of cubic symmetry. Bersuker and Vekhter (1966) then showed that the
tetragonal minima must always have a small trigonal component as well, and sug-
gested that the presence of this component does lead to a splitting. They suggested
that such a splitting was needed to explain the experimental results of Sturge
(1965). T h e small effect treated by Bersuker and Vekhter only arises when the
original treatment of Opik and Pryce (1957) is extended to take into account the
variation of the electronic wave function with ligand position; the Opik-Pryce
approach predicts the possibility of pure tetragonal type minima. Bersuker and
Vekhter do not discuss the transformation properties of their three pseudo-tetra-
gonal states, but state that a cubic perturbation gives nonzero matrix elements
between them. If their three states still belong to T, of Oh, then Ham's objection
still applies. It may be that the following comment is relevant to the dispute;
if an A, type operator acts between non-orthogonal functions belonging to a
single rep of a group, then nonzero cross matrix elements result; the correct
secular determinant should include off-diagonal overlap terms also, and their
effect is to restore exactly the full energy level degeneracy appropriate to the rep
involved.
Group theory and topology in solid state physics 599

Much of the work on the ion-lattice coupling effects uses the single complex
model in which only the effect of the nearest neighbour ligands on the ion is con-
sidered. I t is true that any movement of these ligands can be represented in terms
of the standard normal modes of the complex, with appropriate point group rep
types, but the rest of the lattice will ' drive ' these modes at lattice frequencies which
are not the natural frequencies of the complex. It is one of the basic problems of
spin-lattice relaxation theory, for example, to allow for the spread of the lattice
frequencies while retaining the useful aspects of the site symmetry group (Orbach
1961, Stevens 1967). I n the case of the Jahn-Teller effect it is apparent that any
distortion around one ion will in some way be communicated to other ions by
means of the lattice. As a simple example, if two paramagnetic ions had a common
ligand between them, an outward distortion for one ion would be an inward dis-
tortion for the other one, and the cancellation of the two energy changes would
presumably suppress such a distortion. Sarfatt and Stoneham (1967) used a
continuum model together with a nonrelativistic form of the Goldstone theorem to
suggest the existence of wavelike coupled Jahn-Teller distortions which could
propogate through a lattice at a speed considerably less than that of sound. More
recently Stevens (1969) has used a model in which the lattice is represented by a
continuum, and has shown that a Jahn-Teller distortion around a site will produce
displacements varying as cos at other sites, 0 being the angle between R and
the axis of the distortion.
T h e operator 4 V of (3.79) can be used in an effective operator

(3.82)

which given an effective matrix element between crystal field states which are
otherwise degenerate. Huang (1967 b) and Menne (1968) calculated the contribu-
tion of this orbit-lattice interaction to the zero field splitting of the ground state of
the rare-earth S state ion Gd3+(%7/,) in CaF,. Their numerical results were not in
agreement, but both concluded that the orbit-lattice effect is much too small to give
the observed splitting. Inoue (1963) used a similar approach to calculate phonon
induced g shifts, Menne et aZ. (1968) calculated the temperature dependence of the
hyperfine coupling for the S state ion Euz+ in CaF, by considering the lattice-
induced mixing of configuration 4f7 with configurations 4f6ns. They used a
nearest-neighbour point charge model, which gave the correct temperature depen-
dence but too small a magnitude. Shrivastava (1969) has studied the lattice
vibrational contribution to the transferred hyperfine coupling of Mn2+ in L i F ; his
calculations had of necessity to include the modulation of the LCAO molecular
orbitals by the ligand vibrations, whereas most of the previous calculations men-
tioned in this paragraph were carried out in terms of point charged crystal field
theory.

3.7. Electronic band theory


T h e approaches to one-electron band theory may be roughly divided into three
categories. T h e first type consists of those methods which still adopt an essentially
atomic viewpoint, namely the cellular method and the tight-binding method. T h e
second type consists of methods which stress the delocalization of the electrons in
the choice of basis set, namely the nearly free electron (nfe) approach, the ortho-
gonalized plane wave method and the pseudopotential method. Thirdly, there are
600 J . Killingbeck

methods which show a mixture of the two aspects mentioned above; these latter
methods include the augmented plane wave method and the Korringa-Kohn-
Rostoker method. I n the present section a brief survey is given of the symmetry
considerations used in the various approaches, and mention is made of some of the
ideas developed in recent work on band theory.
3.7.1. The NFE approach; screw axis symmetry effects. Probably the most convenient
starting point for our discussion is the fact that the crystal potential V , which we
take to have the space group symmetry, will only have nonzero matrix elements
between two plane wave states exp (ik. r ) and exp (ik' .r ) if the vector k - k' is a
vector K of the reciprocal lattice for the crystal considered. T h e matrix element is
of course the Fourier transform component V ( K )of V . This criterion involves only
the translational symmetry, and would thus yield the same type of results for Al,
NaC1, diamond, and zincblende lattices, since all have an underlying fcc trans-
lational symmetry. Extra rotational symmetries may make several of the V ( K )
equal to each other, and screw axis symmetries (as in diamond) may cause some
V ( K )to be zero even though K belongs to the reciprocal lattice. A wave function
with wave vector k may thus be written as
(3.83)
K

n being a band index. At high symmetry points in the Brillouin zone, of course,
the ratios of some of the A K are fixed by symmetry considerations, but (3.83)
represents the general form of the one-electron eigenfunctions. T h e most direct
approach is to take (3.83) and insert it in the Schrodinger equation

" V 2 f V )#=E#.
(-m (3.84)

T h e resulting problem is essentially that of solving the secular matrix of V in the


basis of states exp {i(k + K ) , r}. T h e nfe approach assumes that this is possible by
using perturbation theory, i.e. V is assumed to be weak. Two or more plane waves
may have the same kinetic energy and belong to wave vectors which differ by
reciprocal lattice vectors; in that case there will be a strong splitting of their
degeneracy by the periodic potential V. It is well known that this splitting can
occur for states on the boundary of the first Brillouin zone (Seitz 1940, Anderson
1964), so that a band gap usually arises in the nfe approach at the Brillouin zone
boundaries in k-space. If the unit cell of the lattice has extra symmetry which
makes some of the V ( K )vanish, then a gap may not appear at some of the Brillouin
zone boundaries, and it is possible to define Jones zones which are enclosed by those
Brillouin zone boundaries for which a first-order energy gap does occur, I n the
nfe approach the energy discontinuities at the zone boundaries arise in a per-
turbation calculation, and are essentially dictated by the spherical symmetry of the
unperturbed free-electron wave function. This point is discussed at some length
by Seitz (1940).
I n the diamond lattice the V(2,0,0) and V(2,2,2) Fourier components of the
crystal potential vanish if the atoms are taken to be identical and spherically
symmetric. Thus the waves ( & 1,0,0) have no direct matrix element of V between
them and are degenerate in first order ; they correspond to the X, rep as given by
Cardona (1963) and other workers. T h e fact that there must be this twofold
degeneracy of X, type may be proved by group theory, using the presence of the
Group theory and topology in solid state physics 60 1

screw axis symmetry (for a simple example of this type of argument see Heine
(1960)). At points in k-space on the plane bisecting (2,0,0) in the reciprocal lattice
the twofold degeneracy still exists in first order (since V(2,0,0) is zero), but is not
required by group theory. T h e coefficients V(1, 1, 1) and V(1, - 1, - 1) are not
zero, and in second order they may be combined into an effective operator V’(2,0,0)
in the manner described in 5 3.1. V’(2,0,0)then can produce a small splitting of the
double degeneracy, except at the point XI.
T h e effective operator contribution from the V (111) and equivalent Fourier
coefficients is
V(1-1-l)V(lll) V(1-ll)V(ll-1)
(+100[v,(-100) =
E(loO)-z(iTi~--
+ E(100)-E(01-1)-
-~ (3.85)

at X,. T h e relationships V(l1- 1) = iV(111), etc. hold between the Fourier


coefficients, so that (3.85) vanishes. For a pair of states (1, a , p), (- 1, a, p), how-
ever, with CY and /3 small, the energy denominators in the various terms of (3.85) will
be modified, since not all the intermediate states involved will have the same energy.
T h e various terms will then not cancel exactly, and so a small second-order splitting
will appear, tending to zero at X,, where a = /3 = 0. (Mariot’s (1962) discussion of
this is somewhat misleading, in that he appears to suggest that the states (loo),
( - loo), etc. need not be considered at all.) Similar arguments may be used when
discussing the structure factors for x-ray scattering by a diamond type lattice. T h e
structure factors Fzooand Fzzzvanish if the atoms are identical and spherically
symmetric. If, however, a small amount of bonding charge is supposed to be
situated between neighbouring carbon atoms, these factors become non-zero.
Kleinman and Phillips (1962) give expressions for the structure factors in this case;
if a small term E is included to represent a spherical bonding charge half-way
between the two atoms we have
Fhkl= 1+ exp {i&v(h+ K + 1 ) ) + E exp {i&r(h+ k + 1)). (3.86)
Equation (3.86) shows that Fzzz and Fzooare nonzero if E is nonzero, and V(200) is
also nonzero in that case. This appears to imply a first-order splitting at X , if we
consider the electronic bands. However, V(ll1) also changes, and the total effect
of V, and higher-order terms must exactly annul the apparent first-order splitting
at X , (although the present author has not seen an explicit proof of this in the
literature). Colella and Merlini (1966) have performed careful experiments which
yield a nonzero Fzzzvalue for x-ray diffraction in Ge and Si. The site symmetry at
each atom in the diamond lattice is the tetrahedral T, symmetry, which does not
contain the inversion operation. If an origin is taken at a point in space half-way
between two nearest-neighbour carbon atoms, then the site symmetry about this
new origin (sometimes called the ‘standard origin’) does include the inversion.
Mariot (1962) gives the factor group symmetry operations for the diamond lattice
referred to the standard origin. With respect to this origin, the potential V for a
3-5 or 2-6 type zincblende lattice may be split up into a small even parity part and
a larger odd parity part, which must be added to the lattice potential of the appro-
priate 4-4 element. T h e 4-4 type band wave functions will have definite parity with
respect to the standard origin, and the antisymmetric perturbing potential links
only those of them which have opposite parity. For example, the X , degeneracy of
states ( F 1,0,0) is broken, since they may be rewritten as a cosine function and a
sine function of opposite parity, From the formal group theoretical point of view,
602 J . Killingbeck

the absence of the fourfold screw axis operation in the zincblende lattice means that
the twofold X, degeneracy is no longer required by symmetry. T h e perturbation
approach described here has been applied in actual calculations by several workers
(Callaway 1957, Bassani and Celli 1961, Cardona 1963, Cohen and Bergstresser
1966).
3.7.2. Orthogonalized plane waces ; pseudopotentials. T h e expansion (3.83) is always
formally possible in the sense that the plane waves form a complete set. However,
standard perturbation theory or matrix methods may not converge rapidly to give
the AK. In fact, from first principles it would appear that the crystal potential V is
too strong for perturbation theory to be useful. However, there is an extra con-
sideration to be allowed for, namely the fact that the band states +k should be
orthogonal to the occupied core states, which may be reasonably described by a
tight-binding picture. This suggests the use of a basis of orthogonalized plane
waves (opws) of form
+k = IQ-Clc)(clk) (3.87)
where [ k ) denotes the plane wave of wave vector k and I c) is a core Bloch function
I
with the same k-vector. The sum over c) may be extended to all core band func-
tions, since those with wave vector different from k give zero terms. Further, by the
unitary invariance property mentioned in $3.4, the core band Bloch functions may
be replaced by the core band Wannier functions in the sum; the latter functions
will only be atomic functions if the tight-binding approximation is valid for the
core bands. By the arguments of $3.4, +k will have the same space group trans-
formation properties as [ k ) , and all the symmetry arguments of nfe theory apply
with equal force when the opws are used as a basis set instead of plane waves. T h e
opw basis set gives better convergence in most cases; the use of the true crystal
potential V in an opw basis can be regarded as equivalent (as far as the energy
eigenvalues are concerned) to the use of a weak pseudopotential V,, in a plane wave
basis, and this means that the general philosophy of the nfe approach is more widely
applicable than had been previously supposed. The formal details of pseudo-
potential theory have been discussed by many workers (Cohen and Heine 1961,
Austin et al. 1962, Harrison 1966), and it is clear that there are some formal diffi-
culties associated with the energy dependence and nonlocal nature of the pseudo-
potential. However, the theory has been used semi-empirically with reasonable
success (Cohen and Bergstresser 1966, Picard and Hulin 1967, Rehwald 1967), and
seems to provide a rationale for an approach in which the V , are regarded as dis-
posable parameters in much the same way as the crystal field parameters V ; of
$ 3.6. Luehrmann (1968) has given tables of the symmetry-adapted combinations
of plane waves (or opws) which are needed at the various high symmetry points of
the Brillouin zone of symmorphic space groups. His article gives a good account of
the important group theoretical theorems which are employed in band theory
calculations.
3.7.3. The k . p method. As pointed out above, a function +k which is a sum of
opws can also be expressed in the form (3.83), since the core band states themselves
can be written (not very economically!) as a sum of plane waves. An alternative
approach is to factor out the exp (ik. r ) factor and set
$k = e x p ( i k . r ) C A,exp(iK.r) = &(r)exp(ik.r) (3.88)
Group theory and topology in solid state physics 603

where uk(Y) will be periodic in the direct lattice. Equation (3.88) is just the tradi-
tional form for a Bloch function. T h e insertion of (3.88) into (3.84) gives an
effective Schrodinger equation for uk(r);

(3.89)

This leads to the k . p approach; in this approach the states uo(r)are used as basis
set and the k . p operator is regarded as a perturbation. Equation (3.89) is exactly
equivalent to (3.84), and so would be applicable at all points in the Brillouin zone
if the complete set of u,,(r) were used. Consider a set of degenerate functions of
definite parity at k = 0, for example a set of three functions transforming as T,
about a lattice site in a cubic lattice. (When studying such transformation pro-
perties only one unit cell need be taken, since u(r)k has lattice periodicity.) T h e
matrix of k . p in this basis is identically zero. For the case in which k is along the
x-axis we have an effective operator V1 given by

(3.90)

which gives the energy up to second order. Clearly the T, x and T,y states remain
degenerate, while the T , x state splits off from them. T h e presence of the k,p,
term in (3.89) indeed directly suggests a classification of the uk in terms of the reps
of D4h,which is the group of the k-vector (0, 0, k,). T l x and T , y belong to one rep of
D4h, while T,x belongs to another. Falicov (1966) has stressed the usefulness of
the k . p formalism in making explicit the various compatibility relationships
(Bouckaert et al. 1936) which describe the way in which level degeneracies along
lines of symmetry in k-space match up with those at points of high symmetry. For
each of the bands an effective mass m* can be defined by setting the energy resulting
from V1 as being equal to Vi2m* k,2. I n practice a semi-empirical k . p approach,
linked with opw theory, is often employed. T h e states uo(r)may be taken as opws,
the E,, being the opw energies. This still leaves the task of finding the matrix
elements o f p between opws. T h e task is often avoided by regarding the independent
matrix elements of p (i.e. those which cannot be related to one another by sym-
metry arguments) as being parameters which should be adjusted to give agreement
with observed absorption frequencies, energy minima in k-space, m* values from
cyclotron resonance, etc. (Cardona 1963, Pollack and Cardona 1966, Morgan and
Galloway 1967). Whether m* as derived from energy band theory can be set exactly
equal to m Q as derived from dynamical experiments involving externally applied
forces is a complicated problem in its own right (Wannier 1960, Weinreich 1965)
but this first-principles question is ignored in the semi-empirical k .p theory. When
spin-orbit coupling is included an extra term should be added to the Hamiltonian,
and this leads to an additional term
72 , k . ( S x g r a d V )
__ (3.91)
2m c
in the k . p effective Hamiltonian. (We keep here to the approach in which spin
effects are represented by effective operators in the nonrelativistic theory; most
workers have adopted this approach in practice, although Johnston (1960) has
formally discussed the use of the Dirac relativistic equation in solid state problems.)
When the term (3.91) is included, a choice is possible; the spin-orbit operator can
604 J . Killingbeck

be taken along with the k . p term as a perturbation on states uo(r) obtained from a
calculation not including spin-orbit coupling; or the uo(r) can be taken as already
including spin-orbit coupling, so that they are classified according to reps of some
double group. T h e latter approach was used for Ge by Morgan and Galloway
(1967), who treated states belonging to the r7rep of the D4h double group and set
up the k . p matrix for k-vectors along the (1,0,0) direction in k-space. T o obtain
I?, type states it is necessary to combine Bloch functions at k = 0 which are of type
r,-, I?,+, rS-,rs+. Broerman (1968 a) gave double group symmetry-adapted
functions along other directions in k-space for the diamond and zincblende lattices.
He also pointed out (1968 b) that in the limit of zero spin-orbit coupling the
momentum matrix elements of I?,+ and rs+type valence band states with other
band states have definite symmetry-determined ratios. These ratios are little
changed by the spin-orbit effects, and Broerman showed that their values as found
semi-empirically by Galloway and Morgan (who allowed them to be freely variable
parameters) were somewhat smaller than the correct theoretical values. This
illustrates a point of general validity; when the total symmetry group of the
Hamiltonian H is determined by an operator V of low symmetry and with small
physical effects, it is not realistic to ignore those approximate relationships (between
reduced matrix elements and other quantities) which can be derived by using the
larger symmetry group of H-V. More directly speaking, continuity considerations
are important as well as symmetry considerations; a group symmetry is formally
either present or absent, but the energies, matrix elements, etc., of a numerical
calculation usually vary smoothly through the parameter values which represent a
formal change in symmetry. T h e inclusion of spin-orbit coupling effects in the
k . p theory for GaAs was treated by Cardona et al. (1965) without a double group
formalism; they directly applied the spin-orbit coupling terms as a perturbation
to eigenstates obtained from a standard k . p calculation. Luehrmann (1968) dis-
cusses the advantages and disadvantages of directly using double group symmetry-
adapted basis sets in calculations, while Litvin and Zak (1968) discuss the method of
construction of the double space group symmetry adapted functions in the course
of their study of Clebsch-Gordan coefficients for space groups.
3.7.4. The tight-binding approach; Wannier functions. Several authors have dis-
cussed the incorporation of spin-orbit coupling effects into tight-binding theory or
closely related schemes (Doggett 1966, Lenglart 1967, Kunz 1967), and in the tight-
binding theory the major parameter involved is simply the single ion central field
spin-orbit parameter Y, T h e tight-binding method forms Bloch functions by
applying a projection operator to an atomic orbital,
$ka = N X exp ( - ik. R,) $a(Rn) (3.92)
RTl

where N is a normalizing factor and a labels the atomic orbital concerned. Rn is a


lattice translation. When the unit cell contains several atoms, Bloch functions of a
given k may be formed from orbitals at each site, and can then be used as a basis
for the matrix of the crystal potential. For the special high-symmetry points in the
Brillouin zone, the must be chosen to belong to a rep of the group of the k-vector,
and this leads to definite selection rules for the mixing of orbitals arising from
different atoms in the cell. This fact has been discussed, usually in connection with
the cellular method, by several authors (Bell 1954, Altmann and Bradley 1965 b,
Onodera 1968). Altmann and Bradley pointed out that Bell’s results are in error
Group theory and topology in solid state physics 605

for the hcp lattice, since proper account is not taken of the phase factors associated
with the orbitals at the two atoms in the unit cell. A good account of the tight-
binding approach (including spin-orbit coupling) is given by Lenglart (1968)) who
applies it to the d bands of transition metals. T h e transition metals provide an
interesting problem, in that their inner d bands are mainly of tight-binding type,
while the outer s and p bands are of the type amenable to opw theory. A certain
amount of hybridization between these bands occurs, and the explanation of the
magnetic properties of transition metals in terms of band theory is one of the
research problems of current interest in the theory of magnetism (see e.g. the
volume edited by Marshall (1967)). Mueller (1967) has applied a combined tight-
binding-pseudopotential approach to the transition metals ; Rehwald (1967) used
both pseudopotential and tight-binding viewpoints for CdIn,S,, and Beall Fowler
(1963) treated the electronic bands of solid krypton by means of a combined tight-
binding opw approach. If equation (3.92) is inverted, with the $k being taken as the
band functions (not necessarily of tight-binding type) from one band, the resulting
functions
c
d(r - R,) = exp {ik (U - R,)} #/e
k
(3.93)

will form an orthonormal set, each function being localized around an atomic site.
These functions are the Wannier functions for the band concerned (Wannier 1937).
T h e localization (but not the orthonormality) of the functions can be adjusted by
including factors exp (ih(k)} in the terms of the sum (3.93). This property has been
discussed by several workers (e.g. Kohn 1959 b, Weinreich 1965)) and plays an
important r61e in effective mass theory (Wannier 1960, Weinreich 1965). Des
Cloiseaux (1963) has discussed the general problem of the construction of equiva-
lent localized orbitals in a lattice of given space group symmetry, and points out
that Kohn’s (1959 b) results, which were worked out for a one-dimensional lattice,
cannot be directly generalized to three dimensions. When there is more than one
atom per unit cell, the use of the correct band Bloch functions in (3.93) will lead to
localized functions which are associated with more than one site, and which may
roughly resemble the bonding or antibonding molecular orbitals of quantum
chemistry. It is possible to incorporate directly the known bonding properties of
isolated complexes of atoms into the band theory by deliberately using bonding or
antibonding several-centre LCAO functions as a basis set and then projecting out
from them Bloch functions by means of (3.92) (see e.g. Bates and Stevens 1969,
Doggett 1966). This bond-orbital approach provides a useful intuitive link with
molecular theory; if the atomic orbitals are taken separately, of course, a similar
bonding or antibonding effect will emerge naturally on diagonalization of the matrix
of the crystal potential.
3.7.5. Mufin tin models; Green functions. T h e insertion of (3.83) into (3.84) leads
to the result

(3.94)

On using the orthonormality of the plane waves I k + K ) the value of AK may be


formally obtained from (3.94) and then re-inserted in (3.83). T h e result is

i’
$(r) = G(r,r’) V ( r ’ )$(r’) dr‘ (3.95)
606 J . Killingbeck

where the Green function G(r, r’) is given by


1 exp {i(k + K ) . ( r - r’))
G(r, r’) = - (3.96)
k+
T K E - ( 7 ~ ~ / 2(m ) K)’ *
T h e Korringa-Kohn-Rostoker (KKR) approach to band theory concentrates on the
solution of the integral equation (3.95) by means of a variational procedure, and is
used in conjunction with a muffin tin potential, which is radially symmetric about
each site for Y < R and constant in the rest of the lattice. T h e basis functions used
in the variational process are the solutions of the radial Schrodinger equation for
Y < R ; because of the assumed spherical symmetry of the muffin tin there will be a
degeneracy for functions with the same 1 and differing m. T h e use of the non-
overlapping muffin tin potentials is an approximation also employed in the aug-
mented plane wave (apw) method (Slater 1964); it greatly simplifies the calculations
to use this model. If the potential were a sum of oaerlapping muffin tin potentials
from each site, then the resulting potential would not be spherically symmetric
near each site. For example, Jones et al. (1955) used x-ray results to obtain Do)and
D ( Q components of the charge density about the sites in bcc iron by taking the
charge density to be a sum of identical spherically symmetric distributions at each
site. T h e details of the variational calculation associated with the K K R method are
given by Ham and Segall (1961); an important r6le is played by the structure
constants, which are lattice sums of the general form
(3.97)

Here Y T ( R ) is a spherical harmonic involving the polar angles of the vector R.


T h e group G(k) of the vector k is the set of elements of the space group which leave
the wave vector of a Bloch function $k unchanged modulo a reciprocal lattice vector.
For any space group the lattice translations belong to G(k) for any k, and for a
symmorphic space group the lattice translations form an invariant subgroup of
G(k). I n a nonsymmorphic group this is not usually true, but the translations t
for which exp (ik, t ) = 1 do still form an invariant subgroup of G(k) (Herring 1942,
Altmann and Bradley 1965 a). For a symmorphic group the factor group G(k)/T
is a point group, and the operations of this group on the Bloch function $k is
usually visualized as a rotation of the vector k about the origin of k-space. As Re11
(1954) pointed out, at any point in the direct lattice one may define a ‘site’ sym-
metry, which can vary from the identity operator to the full point group of the
(symmorphic) space group. This implies that the group to be considered when
evaluating (3.97) should be the set of elements which is common to the site sym-
metry group of the origin and to the factor group G(k)/T. About a lattice site of a
Bravais lattice this group is the usual site symmetry for the case k = 0, and the
sum (3.97) for k = 0 becomes one of the type encountered in crystal field theory
( 5 3.6). The sum (3.97) is then zero for k = 0 unless Y,“ is a member of the A, rep
of the point group. For k # 0 and a lattice site origin, (3.97) is zero unless Y m is a
member of the A, rep of the point gioup of k (this group is a subgroup of the site
symmetry, i.e. the full point group). Ham and Segall (1961) tabulated the structure
constants of the K K R theory for fcc and bcc lattices, and Segall (1961) applied the
K K R method to the fcc metal Al. T h e opw and K K R methods have been briefly
discussed above, starting from the plane wave expansion (3.83). This suggests that
the opw and K K R results can probably be obtained by using some appropriate
Group theory and topology in solid state physics 607

effective potential function in a plane wave basis. For the opw theory this is the
pseudopotential already referred to; Lloyd (1965) has shown how to construct
effective potentials appropriate to the opw, K K R and apw methods. Ziman (1966)
has developed the K K R viewpoint into a more general theory which is applicable
for nonregular arrays of muffin tin potentials, such as would arise in a model of a'
liquid metal.

4. Topological methods in solid state theory


T h e general level of awareness of group theoretical concepts is fairly high
amongst solid state physicists, and the use of formal group theoretical techniques
is common in the literature, although for specific problems an ad hoc approach is
sometimes more easy than the application of the full general group theoretical
formalism. Section 3 of this article has essentially been devoted to pointing out
that the most fruitful approach is to use a judicious mixture of group theory and

/
/-- --.

I
Figure 5 . The planar graph for which the adjacency matrix is given in the text.

matrix theory. I n the case of topology things are somewhat different; while many
of the intuitive arguments used in solid state theory are of a topological nature, they
represent such simple examples of the general theorems of topology that they are of
little interest to the professional topologist. T h e use of formal topological theorems
and concepts in solid state physics is very much less developed than the use of
formal group theory. This is not surprising, since the rewards gleaned from
topology have not so far been as great as those obtained from group theory; never-
theless, the future will probably see an increasing use of topological results and
terminology in solid state physics, and this part of the article treats some basic
topological concepts which have occurred in the literature.

4.1. Planar graphs; Euler's formula


Figure 5 shows an example of a graph, which consists of a set of n vertices
connected by edges. T h e adjacency matrix of the graph is an n x n matrix with
elements Mjkequal to the number of edges which join vertices j and k. For example,
608 J. Killingbeck

the adjacency matrix for the graph of figure 4 using the solid edges is

T h e jk element of Mp gives the number of paths between j and k which involve p


steps. I n a directed graph the edges have arrows assigned to them, so that 1+2 and
2 + 1 are counted as different, and the elements MI, and Mzl can differ. An
undirected graph, however, has a symmetric adjacency matrix. If the graph
vertices are numbered differently, the adjacency matrix will be changed in appear-
ance, and it is a basic problem of graph theory to discover criteria which will
indicate that two different n x n adjacency matrices are actually describing the same
graph (Heap 1967). I t is accordingly useful to set up some standard way of labelling
vertices. For example, the vertex with highest degree (i.e. number of incident edges)
could be labelled 1, and so on in order of decreasing degree. A further criterion
would then be needed to order vertices of the same degree; Nagle (1966 a) has
discussed these problems and has given a unique labelling prescription for the
vertices. T h e graph of figure 4 is planar, i.e. it can be drawn on a plane without the
need for any pair of edges to intersect; Kuratowski found the necessary and sufficient
conditions for a graph to be planar (see Berge 1962, Ore 1963).
Consider a planar graph, such as that of figure 4, in which the edges form a set
of adjoining polygons. If we concentrated attention on the polygonal areas them-
selves, we would have a map (of the geographical type). We denote the number of
vertices, edges and faces of the graph by U , e and f , respectively. Suppose that
vertices 2 and 4 in figure 4 are joined directly by a new edge, or that the midpoints
of edges (1,2) and (1,3) are so joined. Then the changes in z', e and f are, for the
first process, (0,1,1) and for the second process (2,3,1). I n both cases, and for
more complicated edge insertion processes, the number z' +f - e remains unchanged
and can be seen to have the value 2. T h e Euler formula
v+f-e = 2 (44
for this case may be extended to polygonal maps on other surfaces. For example,
v + f - - e equals 0 for a polygonal map on the surface of a torus (doughnut)
(Patterson 1956), while it still equals 2 for a polygonal map on the surface of a
sphere. In topological terms this arises because the surface of a sphere can be
homeomorphically mapped on to the two-dimensional plane, while the torus
cannot. This can be seen in terms of connectivity; on a sphere any closed curve
divides the surface into two parts; on a torus there are some closed curves with
this property and some without it. T h e two types of closed curve belong to different
homotopy classes, i.e. a curve of one type cannot be continuously deformed into a
curve of the other type without leaving the surface of the torus. T h e Euler
characteristic z' +f - e is only one of several quantities of interest in connection with
planar graphs; another is the cyclomatic number, e - v + 1. If a vertex of degree 2
is suppressed, by simply deleting it and joining its two incident edge; up to form a
single longer edge, then the cyclomatic number is unchanged. If a graph can be
amended in this way to look like another graph (vertex numbering being disregarded)
then the two graphs are termed homeomorphic. By repeated suppression of vertices
Group theory and topology in solid state physics 609

of order 2, we must finally obtain a graph with no such vertex, and such a graph is
termed homeomorphically irreducible. Heap (1967) gives a clear account of the
construction of homeomorphically irreducible star graphs (a star is a graph which
is connected, in the sense that there is some path between any two vertices, and in
which it is not possible to break the graph up into disconnected pieces by removing
any single vertex and its incident edges). T h e Euler formula (4.2) applies to solid
polyhedra as well as to planar polygonal graphs, and has been used in that context by
Frank and Kasper (1958) in a discussion of the packing of atoms in complex alloy
structures. Essam and Fisher (1970) have recently listed many basic definitions of
graph theory.
4.1.1. Rotational symmetry axes f o r a lattice. If we regard a three-dimensional
lattice as obtained by the stacking of two-dimensional lattices (e.g. Buerger 19-56),
then we may proceed by first studying the possible symmetries of a two-dimensional
mosaic (Ore 1963), which is essentially a translationally periodic infinite planar
polygonal graph. We suppose that the faces are identical polygons, with a edges at
each vertex, each face having /3 boundary edges. Denoting the number of vertices,
edges and faces by v, e,f , as before we have
e = + v a e = +f/3 (4.3)
where the factors 4 allow for ‘double counting’. Euler’s formula may be divided
by e and combined with (4.3) to yield
2 2
v f = 1 + -2 = -+-
-+-
e e e cy. /3’ (4.4)
For the infinite lattice we are taking the ratios vie and f i e to have definite numerical
values, actually limiting ratios in the limit e+co (Ore 1963). T h e term 2/e then
vanishes in (4.4) in this limit, so that the only possible integer pairs ( a ,/3) which can
obey (4.4) are (3,6), (4,4) and (6,3). Thus the unit cell shapes must have three, four,
or six edges. Ore’s (1963) diagrams seem to imply that the shapes must be regular
but this is not required by the mathematical proof. For example, a lattice of
oblique parallelograms has a = 4 just as well as a lattice of squares. Thus, twofold
rotation axes can be seen to be possible, whereas five- or sevenfold rotation axes are
not possible. Turner (1968) has given a treatment of planar lattices using graph
theoretical methods, and for the square lattice has studied the ways in which the
sides of the squares may be assigned a direction (denoted by an arrow) with the
lattice still retaining some rotational symmetry. I n the latter case the lattice becomes
a directed graph. Kastelyn (1967) has surveyed several counting problems for
random walks on planar lattices and has pointed out their relevance to problems of
crystal physics.

4.2. Topological waue functions


4.2.1. Methods using the incidence matrix. I n simple organic molecules such as
benzene and anthracene, which consist of one or more linked planar rings of
identical atoms, the T electron approximation is commonly used. This approxi-
mation takes a basis set consisting of identical carbon p~ orbitals, perpendicular
to the plane of the molecule, one at each site. T h e other two p orbitals and the s
orbitals are of o type, or of even parity for reflection in the molecular plane, and so
have zero overlap with the prr orbitals considered; they do not mix with them either,
610 J . Killingbeck

if the one-electron Hamiltonian is assumed to have reflection symmetry in the plane


of the molecule. If the atoms are taken to have equal separations, the Huckel
approximation (Streitwieser 1961, Simpson 1962, Murre11 1963) proceeds by
setting the matrix elements of the (unspecified) one-electron Hamiltonian equal to
some constant B for nearest neighbours, and the overlap matrix element equal to S
for nearest neighbours. Other off-diagonal matrix elements are neglected. Clearly
the resulting secular determinant is of form
A1 +BM-E(l +SM)I = 0 (4.5)
on including the diagonal terms A and 1. Here M is the adjacency matrix for the
molecular structure, with bonds (edges) between neighbouring atoms (vertices).
Many applications of the Huckel theory are discussed by Streitwieser (1961), and
Simpson (1962) discusses it in terms of a more fundamental Hartree-Fock approach.
T h e use of the adjacency matrix in quantum chemistry is exemplified by several of
the papers of the Chicago group, now collected in book form (Platt et al. 1964).
Schmidtke (1966, 1968 a, b) has treated the LCAO molecular orbital theory using
graph theory, and has suggested an alternative approach to the theory of sym-
metrical complexes such as NIX,, MX,. His idea is to take the symmetrical array
of X atoms on its own, and to work out its energy levels using only nearest-
neighbour interactions. T h e secular matrix is then of the type encountered in the
Huckel theory of hydrocarbons. T h e orbitals of the central ion M are mixed in by
a perturbation approach at the last stage. Schmidtke’s approach is equivalent to
assuming large ligand-ligand (X-X) overlap and small M-X overlap. I t is thus
making assumptions contradictory to those of the usual covalency theory for
complexes, but Schmidtke notes that the ordering of energy levels obtained is the
same as that for crystal field theory or the usual covalency theory. Group theory is
often useful in treating the adjacency matrix; as a simple example we may consider
six identical s orbitals at the six X sites in octahedral MX,. T h e ordering of energy
levels is satisfactorily given (Schmidtke 1968 a) by diagonalizing the adjacency
matrix alone, neglecting overlap. On numbering the X sites from 1 to 6 so that the
pairs (1,2), (3,4), (5,6) are directly opposite to each other, we obtain a symmetric
matrix with the elements M j j all zero, with M12= MS4= M5, = 0, and with the
other independent elements all equal to 1. This is essentially the same matrix that
would arise in band theory for the interaction of the plane wave state (1,0,0) (in
the reciprocal lattice of a cubic lattice) with the five other degenerate plane waves.
T h e six plane waves are known to give the reps ill,E, T for the group 0, of the
wave vector (1,0,0). T h e appropriate projection operators could thus be applied
to our X, orbitals in order to diagonalize the matrix directly. -4 useful check is
obtained by calling the three roots x, y and z and computing a few low powers of M,
setting for the diagonal sum
x(MP) = xp + 23.’”+ 3 ~ ? 3 (4.6)
T h e roots are found to be 4(A1), -2(E), O(T,) for our example; when p orbitals are
used at the X sites, symmetry arguments may be used to find the way in which
p lobes at one X site will interact with lobes at another site. I n this case more
undetermined constants will appear in the secular matrix (Schmidtke 1968 b) and,
as for traditional Huckel theory, they must be estimated by a rough argument or
semi-empirically .
Group theory and topology in solid state physics 61 1

4.2.2. Free-electron network theory. T h e book of papers by the Chicago group


(Platt et al. 1964) contains not only work which employs the adjacency matrix, but
also many examples of the so-called free-electron network theory. While the Huckel
approach to a hydrocarbon ring molecule would involve a basis set of atomic orbitals,
the network theory treats the electrons as though they were free to travel along the
bonds, but have to obey certain boundary conditions (microscopic Kirchoff laws) if
several bonds have a junction at an atom. Along each bond the electronic motion is
treated as one-dimensional. For example (Platt et al. 1964), the ring of six carbon
atoms in benzene is simply treated as a ring of constant potential, and the cyclic
boundary conditions directly give one-electron orbital energies equal to n27i2/2mL2,
L being the perimeter length and n an integer. T h e free-electron network theory
gives reasonable qualitative and semi-quantitative results for simple hydrocarbon
molecules. This is an example of a general self-consistency principle; if the wave
function $ is known to be concentrated in some region for a particular type of
eigenstate, only small error is introduced by constraining # to be zero outside that
region during the calculation of $. (This idea has also been used in the calculation
of local modes of vibration in lattices (Dean 1968).) The wavefunction constraint
$ = 0 is replaced by the condition V = CO in the potential. Psychologically, the
latter appears highly unphysical, but is equivalent to the (not very harmful) wave-
function constraint. I n the case of p~ networks the Huckel theory and the network
theory can only be made logically consistent if the network is a cylinder rather than
a line, since a pn orbital is zero on the line joining adjacent nuclei. This point seems
to have been widely overlooked in the literature; fortunately, it is not numerically
important if the bond ‘width’ is small compared with its length. The relationships
between LCAO theory and free-electron network theory have been discussed by
several authors (Murre11 1963, Platt et al. 1964). Hoerni (1961) applied the network
approach to the diamond lattice, with the electron paths being along the traditional
tetrahedral bond directions between neighbouring atoms. The results were only
in very rough agreement with those of band theory calculations of other types. T h e
free-electron network approach is clearly more akin to a covalent bond type picture
of the carbon lattice than to what the solid state physicist would think of as a free-
electron theory (§3 of this article), and it is necessary to watch out for a confusion of
terminology in the literature. A benzene ring is very similar to a one-dimensional
lattice with cyclic boundary conditions; if an atomic potential were incorporated
at each site in the network approach, the resulting mixing of the unperturbed
running waves would give the nfe approach of solid state band theory. This is not
so for the three-dimensional case, since the nfe wave functions of band theory are
non-zero at all parts of the unit cell, not just on link lines between the atoms.
4.2.3. Cyclic boundary conditions. The adjacency matrix for a one-dimensional
lattice provides a simple example of the effect of cyclic boundary conditions in solid
state theory. Consider N identical atoms on a linear lattice and set up the i V x N
adjacency matrix with elements Mik = 2ik,i+l, MAv1 = 0. T h e eigenvalues of this
matrix will give a qualitative idea of the band states in a tight-binding theory which
includes only nearest-neighbour interactions. The eigenvalues of such a matrix
were evaluated by Rayleigh, and the solution is also given by Webster (1955); the
exact eigenvalues are

( n = integer). (4.7)
612 J . Killingbeck

If periodic boundary conditions are imposed we make MAyl equal to 1 instead of 0,


and the usual Bloch function transformation gives for the eigenvalues

(n = integer).

(4.7) and (4.8) show that the eigenvalues resulting from the use of cyclic boundary
conditions differ only by order N-I from the exact eigenvalues. Chan (1969) has
treated the eigenvalue problem for secular matrices with all elements equal to zero
except the elements Hji,Hj,i+hand Hj,j-h. For the case in which all the H j j are equal
and all the nondiagonal elements Hj,j*h are equal, the Bloch transformation may be
used to obtain the eigenvalues as the dimension of the matrix tends to infinity;
Chan apparently did not note this simple special case. If the matrix considered
above in connection with (4.7) and (4.8) is modified by making one of the diagonal
elements nonzero, then the resulting eigenvalue problem gives a qualitative picture
of the effect on the energy bands of a point impurity at one site, and Chan’s con-
tinued fraction method would give the eigenvalues.
T h e adjacency matrix approach may be used in a simple argument to show that
the tight-binding model of band theory gives a band density of states g ( E ) which is
symmetric about the band centre. This result holds if only nearest-neighbour
interactions are considered and if overlap between atoms is neglected. T h e explicit
E ( k ) functions for the three cubic Bravais lattices under these assumptions are given
by several authors (e.g. Jelitto 1969)) and do show the above-mentioned feature.
T h e tight-binding secular matrix, set up within a basis of identical orbitals from
each site, is essentially the adjacency matrix M for the lattice. T h e trace of the nth
power of the matrix, x(Mn), thus gives the nth moment of the band about its centre.
A diagonal element of Mn equals the number of connected paths of n steps which
start from a given site and end on that site. For the three cubic Bravais lattices
there are no such paths if n is odd. Thus, the odd moments vanish, and the density
of states is symmetric about the band centre. This argument will clearly fail for
lattices in which two nearest neighbours can have a common nearest neighbour.
4.2.4. Wintner’s theorem on nodes. T h e topological wave functions of the Huckel
or network types have the property that the number of nodes in the function
usually increases monotonically with the energy, This property may be rigorously
derived for functions $(x) of one variable which obey a differential equation of
Sturm-Liouville type (Killingbeck and Cole 1970)) and Rosenstock and McGill
(1962) have shown that it is also true for the vibrational modes of a disordered linear
lattice. I n the case of a one-electron three-dimensional eigenfunction #(r) the
following theorem holds (Wintner 1948); in any region bounded by nodal surfaces of
#(r) within which $(r) does not change sign, any wave function of higher energy
must change sign at least once. As an example, the d orbitals xyf(y) and (x2- y 2 ) f ( y )
are essentially identical except for a rotation, so that they have interlocking nodal
planes. Their ordering can be reversed by changing the sign of the octahedral
crystal field. They show that it is not true to assert that the number of nodes must
always increase with the energy, but they do obey Wintner’s theorem. Herzfeld
(1949) has discussed nodal surfaces in simple molecules. T h e topological wave
functions treated in free-electron network theory refer to an electron moving on a line
of constant potential; in an approximate treatment of the F centre, Hunter (1968)
has treated an electron moving on a spherical surface of constant potential. T h e
Group theory and topology in solid state plzysics 613

radius of the sphere was taken equal to the distance from the centre of the F centre
vacancy to the first cation layer. T h e relationship between Hunter’s approach and
Schmidtke’s strong ligand overlap topological matrix approach is thus similar to
that between Huckel theory and free-electron network theory.

4.3. Connected graph expansions


T h e graphs talked about so far in this part have mainly been able to be inter-
preted directly as physical networks in a lattice or molecule. Graphs or diagrams of
a more abstract kind occur in the pictorial representation of perturbation series or
other algebraic summations. Such diagrammatic techniques are not restricted to
solid state theory, but some of them, when used in connection with the statistical
mechanics of a regular lattice system, lead back to diagrams in which the vertices
can once again be thought of as lattice sites (see e.g. Domb and Wood 1965). Many
authors have dealt with graphical methods in statistical mechanics (e.g. Hurst 1964,
1966, Nagle 1966 b, Sykes et al. 1966, Domb and Heap 1967, Nagle 1968, Nagle and
Temperley 1968, Woodbury 1969) and in quantum mechanical perturbation theory
(Brueckner 1967, Mattuck 1967, March et al. 1967, Salzman 1968). Diagram-
matic techniques have recently been developed in angular momentum theory (see
e.g. Brink and Satchler 1968) and in the LCAO-MO modification of crystal field theory
(Bishton and Newman 1968).
T h e particular way of classifying and interpreting the diagrams varies from one
application to another, but the general principle is that each diagram represents a
particular term in some expansion. Topologically equivalent diagrams are often
grouped together, and in most applications the connected diagrams can be isolated
in such a way that they alone give the required sum. Kubo (1962), in a discussion
of a generalized cumulant expansion method, indicated on general statistical
grounds why the linked diagrams play an important r6le in various theories. A
clear discussion of the way in which the use of linked Feynman diagrams leads to
the Goldstone (1957) linked cluster theorem of quantum mechanical perturbation
theory is given by March et al. (1967). Goldstone’s (1957) approach used time-
dependent perturbation theory to obtain a time-independent result. Kittel (1963)
also gives an example of such a procedure. Goldstone did point out that the linked
cluster theorem could be obtained by time-independent methods. Salzman (1968)
has devised a diagrammatical representation of the Rayleigh-Schrodinger time-
independent perturbation theory, and has pointed out the relationship of his work
to that of Goldstone, Wannier (1966) treats the Mayer cluster expansion for the
partition function of an imperfect gas; Domb and Wood (1965) apply the theorem
to the Heisenberg spin-& model of ferromagnetism, showing that their cluster
expansion approach involves connected graphs of n nearest-neighbour lattice bonds
at a time (n = 1,2,3, .. .). T h e basic principle behind the various linked-diagram
methods may be described as follows. Consider the symbolic formal sum

I n this sum, the terms involving x are those of the exponential series. Each bracket
contains a formal sum over all graphs with a fixed number of vertices and all
21
614 J: Killingbeck
possible numbers of edges. (Multiple edges are not allowed in the present dis-
cussion, nor are loops with both ends at one vertex. Such edges are considered in
some applications, and lead to adjacency matrices in which integers greater than
one can occur in off-diagonal positions, while the diagonal elements can be nonzero
integers.) T h e topological degeneracy factor attached to each diagram describes
the number of graphs of identical connectivity which would arise if the vertices were
numbered; one useful way to compute it is to take a fixed natural ordering 1,2, ...,n
as labelling the columns of the adjacency matrix, and find the number of distinct
adjacency matrices which can be constructed by using the given number of edges,
For example, with three vertices and one edge, the degeneracy factor is 3, corre-
sponding to (12) (3), (13) (2) and (23) (1). Now consider the symbolic expression

(4.10)

in which the terms are constructed as for (4.9), but using only the connected graphs
of each order. Expansion of the exponential yields the formal result

1+sum of all linked cluster terms + i x z


(I1
where formal products of graphs have been indicated by drawing the graphs side
by side. Comparison of (4.11) and (4.9) shows that the two series agree if we set
I and 1 formally equal. T h e term +x3 will occur in the third-power term of
the expansion (4.10). That the expressions (4.9) and (4.10) agree to all orders may
be proved by induction (Wannier 1966) if it is assumed that the formal product of
any number of graphs is independent of the order in which they are written. This
result holds true when the graphs are taken to represent numbers, i.e. each graph is
replaced in the expressions (4.9) and (4.10) by a weight function. T h e value of the
weight function must depend on the topology (connectivity) of the graph and not
on the labelling of the vertices, and its value for an unconnected graph must equal
the product of its values for the connected components of the graph. T h e choice
and interpretation of the weight function vary from one application to another, as
may the choice of the values of the dummy variable x. For example, March et al.
(1967), in deriving a linked cluster expansion for the expectation value of the
quantum mechanical development operator in the interaction picture, represent the
relevant multiple integrals by means of diagrams. T h e topological classification of
their diagrams is such that the contribution of a diagram consisting of a pair of
linked clusters (a, b) is given by

(4.12)

where n,,nb are the number of vertices in the connected components a, b, and
(a), (b) are the individual cluster contributions. To relate this case to that des-
cribed above would accordingly involve incorporating a factor n! in the weight
function associated with a graph of n vertices. Such factors usually occur in the
interaction picture of quantum mechanics because of the process of time-ordering
of the perturbing operators V ( t )in the various terms of the series expansion of the
Group theory and topology in solid state physics 615

development operator. T h e quantum mechanical calculation has x = 1; if, how-


ever, (4.9) represents the grand partition function for a system of interacting
classical particles, then x becomes the temperature-dependent fugacity and the
sums in brackets in (4.10) represent free-energy functions for cluster subsystems
with definite particle numbers.

tI O’ /

I
/
//
I /
\
\
I /
\ I /

Figure 6. A compound rotation-translation. Point a is taken to b by the rotation about 0, and


translation t takes b back to a.

4.4. Fixed points of mappings ; compound operations


Much of topology is concerned with mappings of one space into another. I n
the case of a mapping of a given space into itself, an important problem is to estab-
lish the existence of ajixedpoint of the mapping, i.e. a point which goes over into
itself. For example, Brouwer’s fixed point theorem asserts (as a particular case) that
every continuous mapping of the closed sphere I r l 6 1 into itself has a fixed point.
Schauder’s theorem (again a special case) extends this to the interior and surface
of any convex body. A convex body in three dimensions is one which, for every
pair of points which it contains, also contains all points on the straight line joining
them. T h e first Brillouin zones or Wigner-Seitz cells of crystal lattices are convex.
Many mathematical theorems have been derived for convex figures (Lyusternik
1963), but the present author does not know of any work which has so far used
these theorems in solid state physics. I n the case of the rotations, reflections and
translations of solid state physics the situation is intuitively clear. A translation has
no fixed point (unless we agree to use only lattice translations and formally identify
equivalent points in different unit cells). A rotation has a line of fixed points along
the rotation axis, while a reflection has a plane of fixed points. Suppose now that we
follow an operation which has some fixed points by an operation which has none.
Has the resulting compound operation any fixed points ? If we consider the fixed
points of the first operation, then they clearly are not fixed under the compound
operation. However, there may be some points which are moved by the first
operation and are then moved back again by the second operation. Figure 6 gives
616 J . Killingbeck

+
an important example. T h e rotation through angle about the axis 0 takes the
point a over into the point b. If now we wish to restore b to a by a further operation,
there are several possibilities, We could have another rotation axis at some position
0' involving an appropriate rotation angle, or we could have a translation through
distance t. In particular, the rotation about 0 followed by the translation t has a
as a fixed point. A quick check reveals that the points around a are not fixed under
the compound operation, and the physical conclusion to be reached is that the com-
pound operation is identical with a rotation about a. T h e angle of rotation is +,
as may be obtained by noting that 0 suffers only the translation t . In a crystal
lattice the vectors Oa, Ob and t would be required to be lattice translation vectors in
order to leave the lattice invariant under the various operations, and this restricts
+
the possible values for the rotation angle +. T h e possible values for are then just
those corresponding to 1-, 2-, 3-, 4-or 6-fold axes. T h e preceding example makes it
clear that the concept of a 'screw axis' operation with the translation perpendicular
to the rotation axis is redundant, since such an (apparently compound) operation
would actually turn out to be a pure rotation. T h e only screw axis operations
needed to describe a lattice have identical screw and translation directions. T h e
axes 0 and 0', and the axis at a, discussed above, are all parallel to one another.
If two rotation axes intersect at a point, then the effect of the separate rotations can
be represented by drawing arcs on the surface of a sphere centred on the intersection
point. This is essentially a modification for the spherical surface of the kind of con-
struction for the plane as shown in figure 6. T h e resulting Euler construction,
described, e.g. by Buerger (1956), yields a fixed point on the sphere. This fixed
point is not changed by the successive performance of the two rotations (in a
definite order) and may be joined to the centre of the sphere to give the rotation axis
of the single rotation to which the combined rotations are equivalent.
4.4.1. Phase plane methods for differential equations. I n various solid state problems
the need arises to discuss the rate of change of the populations of various energy
levels when an externally applied radiation field and an internal relaxation
mechanism are present. For example, the Statz-de Mars equations for the photon
density X and the population inversion N = N 2 - N l in a three-level laser may be
written in the form
2 =-aX+bXN r\i = e - d N - 2 b X N (4.13)
where a, b, c, d are constants characteristic of the laser system. T h e equations
(4.13) are special examples of a general type of nonlinear equation system
2 = F(x,Y) g = G(x,y). (4.14)
Equations such as (4.14) are conveniently studied in terms of the x y phase plane,
and the possible forms of the trajectory x ( t ) ,y ( t ) in the phase plane may be studied
by topological arguments. If the equations (4.14) possess a periodic solution, then
this must be represented by a closed curve in the phase plane, with the representa-
tive point for the system continuously moving round the curve. Green's theorem
gives, for an integration around the closed trajectory,

(4.15)
Group theory and topolofy in solid state physics 617

the double integral being taken over the area within the closed curve. This yields
Bendixson’s theorem, which states that there cannot be a periodic solution curve for
the equations (4.14) in any region of the xy-plane throughout which the integrand
of the integral (4.15) has fixed sign (i.e. has no nodes). Thus any periodic solution
trajectory must cross the curve i?F/ax+i?G/ay = 0 in the phase plane. I n general,
the rate of change of any function A ( x , y ) is given by

(4.16)

A point at which F ( x , y ) = G ( x , y ) = 0 is termed an equilibrium point (or singular


point). Suppose now that we can construct the function A ( x , y ) so that it is zero at
some equilibrium point, but increases along any direction away from the equili-
brium point. If B ( x , y ) is negative definite throughout some region Y including the
equilibrium point it follows that, for any initial values (xo,yo)in the region Y, the
solution function x(t), y ( t ) will tend asymptotically to the equilibrium point. T h e
proof of this asymptotic stability thus can be achieved by constructing an appro-
priate Liapunov function F ( x ,y ) . Hofelich-Abate and Hofelich (1968 a) applied this
approach to the Statz-de Mars equations, and also (1968 b) drew out some phase
space trajectories for the equations. Kleinman (1964) discussed approximate
solutions of the Statz-de Mars equations. T h e use of phase plane and topological
arguments in the treatment of nonlinear differential and integral equations is
treated in several recent books (e.g. Davis 1962, Krasnoselskii 1963, Saaty and
Bram 1964). Consider now a region in the phase plane throughout which the ratio
FIG as obtained from (4.14) is well defined and single-valued. This ratio gives the
slope dyidx of the (unique) phase plane trajectory through each point. Suppose
that a circle is drawn in the region under consideration, and that the angle 6 of the
trajectory line with respect to, say, the x-axis is noted for the points on the circle.
On traversing the circle once, 6 must vary continuously in such a way that it
changes by an amount 2N77 for one circuit. T h e integer A T is called the index of the
curve. If the circle is gradually distorted, with its circumference still in the region
in which the trajectories are uniquely defined, then the value of N will change
continuously; since N must be an integer, it must remain fixed. Topological argu-
ments often proceed in this manner, by constructing functions which can take on
only discrete values, but which must also be single-valued and continuous ; Arnold
(1962) gives several examples. (In this case, of course, N is strictly speaking a
functional of the closed curve.) It is possible for the interior of the curve to contain
a singular point (at which F = G = 0 ) without invalidating the above argument,
provided that the circumference of the curve does not pass through the singular
point while the curve is being distorted. Not all singular points are stable equili-
brium points ; the classification of singular points into nodal points, vortex centres,
spiral points and saddle points is given by Davis (1962). Each type of singular point
has a certain associated value of N , which is additive if several singular points are
within the closed curve. By evaluating N for a closed curve it is thus possible to say
something about the nature of the singular points within it.

4.5. kfodel calculations foy one-, two- and three-dimensional systems


Many calculations have been made on one-dimensional lattices, for which
mathematical difficulties are kept to a minimum, while it may be hoped that the
results obtained will give useful qualitative insight into the particular phenomenon
618 J . Killingbeck

under investigation. For example, many texts give examples of one-dimensional


band theory or of lattice vibration theory for a linear chain, and these examples
show clearly the existence of energy bands or of wavelike normal modes in a periodic
structure. However, some of the details of band theory do change according to the
number of dimensions involved, while there are some phenomena which cannot
occur for systems with less than two dimensions.
4.5.1. Band theory in one dimension. Consider the one-dimensional band problem
for an electron in a linear lattice. It is clear that the real nature of the potential
function (i.e. time-reversal symmetry) means that the eigenfunctions go in pairs
$ck with equal energy. This result also holds in three dimensions, provided that
proper account of the time reversal of the spins is included when spin-orbit coupling
is present, and k replaces k. Suppose we make the reasonable assumption that the
energy E(K) varies continuously with k. If there is a maximum or minimum in E(k)
at any point 12 = k,, other than the origin or the Brillouin zone boundary, then for
small values of h we shall have
E(k, + h ) = E(k, - h).
By time reversal symmetry we also have
E(K,+h) = E(&-h) = E ( - k , - h ) = E(-k,+h). (4.17)
This requires that there be four different solutions of the Schrodinger equation
with the same E-value. For fixed E , however, the Schrodinger equation is a second-
order equation with only two independent solutions. Thus, there cannot be a
maximum or minimum of the type postulated above for the E ( k ) curve, and so the
energy varies monotonically with k. Wannier (1960) derives this result by a more
involved argument based on the Floquet theory, and comments that the same result
holds for one-dimensional lattice vibrations. This is not generally so if forces other
than nearest-neighbour forces are allowed, since the squared angular frequency of
the vibrations is
Wk2 = f,( 1 - cos nka) ka I 6 7T (4.18)
n

where f, is a force constant for nth-nearest-neighbour interactions and a is the


lattice spacing. I t is possible to chose the f, so as to produce non-monotonic
variation of uk2with k, although such a choice may not correspond to a good
physical model. No matter what the choice of the f,,,w 2 as given by (4.18) has zero
I
slope at the Brillouin zone boundary, kal = T . If we assume the E(K) curve for
the band theory example to be smooth as well as continuous, and to be periodic in
the reciprocal (linear) lattice, then it follows that for a non-degenerate band the
E(k) curve has zero slope at the zone boundary. (If two bands touch at the zone
boundary only their average slope need be zero.) T h e assumption of smoothness
(continuity of slope) is important in the argument stated above; if a free-electron
parabolic E ( k ) curve is drawn out in the reduced zone scheme it does not have zero
slope at the zone boundary until the periodic potential is 'turned on'. T h e nearly
free-electron theory leads to the zero slope at the zone boundary; for example, on
setting n / a = K , the secular determinant for the mixing of the nearly degenerate
waves, K+h, - K + h , takes the form

I ( K +V "
h)z- E V
( K - h)' - E
(4.19)
Group theory and topology in solid state physics 619

and direct calculation shows that the two roots vary quadratically with h as h
increases from zero, with a band gap 21 VI at h = 0. I n three-dimensional band
theory the relevant quantity is the component of grad,(E) perpendicular to the
Brillouin zone boundary. This component is not always zero; for example, it is not
necessarily zero at a general point on the plane hexagonal faces of the Brillouin zone
of the fcc lattice (Bouckaert et al. 1936). However, it is zero if there exists a sym-
metry plane perpendicular to that reciprocal lattice vector which generates the
section of zone surface under consideration, and this condition is fulfilled in the
majority of cases, except for triclinic lattices. From a topological point of view these
results can be obtained by packing the Brillouin zones together to fill all of reciprocal
space, and then demanding that the E(k) function shall vary smoothly and periodic-
ally in the reciprocal lattice as well as having the correct rotational symmetry. This
is much more difficult to visualize than the corresponding one-dimensional case.
Kittel (1963) gives some examples of the way in which the isotropic free-electron
E(K) parabola fits into the reduced zone of the three cubic lattices. Slater (1965 b)
gives similar illustrations for some other lattices, and points out that for many
metals such plots give a reasonable qualitative picture of the band structure as
obtained from detailed calculations. For the case of lattice vibrations, equation
(4.18) shows that long waves ( K N 0) have a wavelength-independent speed. A long
wave in this context is one for which X 9 a. I n three-dimensional lattices, each of the
three acoustical modes has a velocity which is independent of I kl for long wave-
lengths, but which does depend on k (i.e. is non-isotropic). If the k-vector of a
vibrational mode lies along a high-symmetry direction in a three-dimensional
lattice (e.g. the 100 direction in a cubic lattice) then the lattice planes may move
rigidly for that mode, so that an effective one-dimensional theory will describe that
mode. Cochran (1963) gives some examples, including the case of the fcc diamond
lattice, for which he points out that two of the modes are degenerate at the zone
boundary along the 100 direction; this is due to the same kind of screw axis sym-
metry effect which was discussed for electronic band theory in $ 3 of this article.
I n the vibrational case, the symmetry leads to relationships between the force
constants, and these relationships then show up by causing degeneracies in the
eigenvalues of the dynamical matrix (see e.g. Maradudin and Vosko 1968).
T h e notions of continuity and smoothness are invoked in an intuitive manner
in many of the works cited in this article. One of the basic requirements of quantum
mechanics is that the wave function t,L shall be continuous in value and in slope.
T h e second requirement is violated by those model calculations which employ delta
function potentials. T h e Koster-Slater (1954) model of electronic impurity levels
employs a localized delta function perturbing potential which mixes the Bloch
functions of the lattice. Such models are often useful as an indication of the
qualitative effects which can occur in a given system, even though they violate the
requirement of continuity of slope. T h e delta function also has the special property
that its matrix elements are identical between any two distinct members of a set of
plane wave functions. This often simplifies the mathematics, but does incidentally
mean that it cannot be completely correct to neglect interband matrix elements when
the impurity site potential is taken to be a delta function, since all the band Bloch
functions together are needed to give a complete orthonormal set. I t is possible for
considerations of smoothness and of dimensionality to both be involved in pro-
ducing unrealistic results for a model calculation. For example, in a one-dimensional
binary alloy lattice, consisting of equally spaced delta potentials of strengths a and p,
620 J . Killingbeck
which represent the two types of atom, the following result may be demonstrated
(Luttinger 1951); energy levels which are forbidden in a pure a: lattice and in a
pure ,# lattice are also forbidden in an alloy lattice with any proportions of a: and ,#
type sites. This theorem does not in general hold for lattices which violate the
structural requirements described above. This example and many others are
quoted in the book edited by Lieb and Mattis (1966), which gives a good survey
of the use of one-dimensional models in several branches of theoretical physics.

4.5.2. Non-monotonicity in three dimensions. T h e E ( k ) curve is monotonic for the


one-dimensional electronic band theory. For 0 < k < rria there is thus a one-to-one
relationship between E and k . For a two-dimensional square lattice E(k) would be
specified at each point in a square Brillouin zone. A basic result of topology, con-
cerning dimensionality, is the following; it is not possible to map the points of a line
(one-dimensional space) onto the points of a square (two-dimensional space) in
such a way that the mapping is both one-to-one while at the same time the mapping
and its inverse are both continuous. Since there is a one-to-one relationship
between E and k for the one-dimensional band, it follows that the one-dimensional
E values cannot be assigned to the k-vectors in the first quadrant of the square zone
without there being some degeneracy or discontinuity. In practice, of course,
E(k) surfaces in two or three dimensions do not arise from such a topological
mapping process, but are found by starting from a periodic potential with the
appropriate number of dimensions. Suppose that a potential V ( x )gives rise to the
monotonic E ( k ) curve. One simple (but not very realistic) three-dimensional
potential with cubic symmetry would be
V ( r ) = V ( x )+ V ( y )+ V ( x ) (4.20)
with the three (identical) component functions having a periodic distance a. Such
simple additive potentials have been considered as models by Seitz (1940) and
Mattis (1965). I t may be noted that the usual so-called one-dimensional band theory
is actually an additive potential theory with the y and x terms suppressed. T h e
free-electron network theory (§4.2.2) is nearer to a true one-dimensional theory.
If the potential is of form (4.20)) with E ( k ) monotonically increasing along k,, k,
or k, axes individually, then the energy will increase monotonically along any line
through the origin and directed into the first octant. (Symmetry arguments give
E ( k ) in the other octants.) This arises because the radial gradient along direction
I, m , n is equal to
2E 2E i3E
grad,E = l-+m----+n, (4.21)
ak, 2ku dk,

with all terms being positive. T h e same is not true for the tangential component,
so that E cannot vary monotonically on any circle round the origin in k-space; this
can be seen by noting that cyclic conditions must hold on the circle. As an example
of a function in k-space which has cubic symmetry we may quote the sum of cubic
harmonics
E ( k ) = A I kI2+ B(kZ4+kU4+k,4- kI4) ‘(4.22)
For this case the energies are additive (and could thus arise from an additive
I 14,
potential) except for the term - 8 k With A and B positive, E increases mono-
tonically along each coordinate axis, but need not do so for all radial directions; in
Group theory and topology in solid state physics 62 1

fact it will vary non-monotonically along directions for which l 4 + m4 + n4 < +. This
monotonicity along each axis plus cubic symmetry does not imply monotonicity
in all directions. I n fact, the one-dimensional theorem cannot even be extended to
apply for k vectors of type (k,O,O), since C2 replaces E2/ax2 in the kinetic energy
operator.

4.5.3. Some examples, mainly from lattice oibration theory. As examples of the use
of one-dimensional models to suggest more general three-dimensional effects the
following may be quoted, T h e original work of Overhauser (1960) on the spin-
deviation wave state of an electron gas was carried out for a one-dimensional system,
and the extension of his results to the three-dimensional case was fraught with
difficulties and controversy. Neumark (1964) has used a one-dimensional model to
‘explain’ the observed fact that the interband momentum matrix elements at a given
point in k-space do not vary much for band calculations on diamond and zincblende
type semiconductors. Neumark’s model is a generalization of the Kronig-Penney
model (Pippard 1965)) and employs wells of differing depth to simulate sites of
different polarity. T h e dispersion curve for a vibrating diatomic linear chain
(e.g. Brillouin 1953) shows that, when the two masses in the unit cell differ con-
siderably, the optical band is much narrower than the acoustical one. Further, the
optical branch has zero slope at k = 0, so that w varies very slowly with k near to
the origin. These results carry over qualitatively to three dimensions, and Huang
(1967 b) has invoked them to suggest that the high density of states in the optical
branches can give an appreciable contribution to electron spin-lattice relaxation
effects, even though the acoustical modes are more highly populated. Rosenstock
(1 960) investigated the infrared absorption of a one-dimensional ionic lattice,
including long-range Coulomb forces in order to assess the role of the boundary
conditions. H e found that the absorption line shape is modified when the long-
range forces are included, but pointed out that his one-dimensional calculation was
not capable of describing subsidiary absorption lines which can occur when boundary
conditions are imposed for a three-dimensional lattice (see 0 2.4). Pirenne and
Renson (1962) have re-investigated the thermodynamic properties of the mon-
atomic linear chain and have shown that the results of Blackman (1935) were
partially in error. They point out that in any theory which treats isotopic impurities
it is essential to know the exact properties of the unperturbed lattice, and such exact
results are usually only obtainable for one-dimensional lattices. T h e effects of
impurities on the vibrations of linear lattices have been studied by several workers
(e.g. Bjork 1957, Rosenstock and McGill 1962). T h e theory for the three-
dimensional case is reviewed by Maradudin et al. (1963) and Elliott (1966). There
is a difference between the one-dimensional and three-dimensional cases. If the
impurity mass M is lighter than the other masses m in a monatomic lattice, then a
localized high-frequency mode may be expected to exist around the impurity, with
a frequency dependent on the mass ratio E = Mim. If E < 1, such a mode does
appear in the one-dimensional case, but it generally does not appear in the three-
dimensional case until E is less than some critical value less than unity. This result
is related to the different behaviour of the density of states in one and three
dimensions, as is made clear by the Green function approach. T h e one-dimen-
sional case may be treated as follows. For an isotopic impurity at the site n = 0 we
have
- mw2A , + ( m- M ) w2A , ,a, = K(A,+, + A,-, - ZA,). (4.23)
622 J . Killingbeck

I n (4.23) nearest-neighbour forces have been taken, and the normal mode assump-
tion x, cc exp(iwt) has already been made in the original equations of motion.
From (4.23) we obtain the result
(wk2- w2)-’( 1- e) w2 A, = 2 A, exp (ikna) (4.24)
n

where the perfect lattice frequencies w 2 = 2K( 1 - cos ka) m-l have been introduced.
T h e sum on the right-hand side of (4.24) may be interpreted as the partial amplitude
of the unperturbed normal mode with wave vector k in the local mode motion,
and the summation over n then corresponds to passing from a crystal coordinate
representation to a crystal momentum representation. A summation over the
allowed k-values in the Brillouin zone leaves only the coefficient A, on the right-
hand side of the equation. A, may then be cancelled to give the equation

E
i
(1 - 6) w2 ( w ’-
~ w2)>-’g(w’)dw‘ (4.25)

where g(w) is the density of states function in the variable w . If we try to obtain a
solution w which is greater than any of the wk, it is clear that a solution will only
exist in the limit E-+ 1 if the integral exhibits some kind of singularity for w’+w,,
where w, is the maximum frequency of the unperturbed band. For the one-
dimensional case g(w) varies as (wm-w)-’’2, and so a solution is obtainable no
matter how close E is to 1. I n three dimensions, however, g(w) usually varies as the
higher power (wm - w)*’z, and the resulting integral does not diverge strongly enough
t o ensure a solution w > w, for E very close to 1. For the three-dimensional case, of
course, the equation (4.25) must be appropriately modified to include the three
types of mode (usually termed ‘longitudinal’ and ‘transverse’, although they need
not always be exactly longitudinal or transverse with respect to their associated k-
vector). Elliott (1966) points out that, in the special case of an isotropic Debye
model for which the three modes are assumed degenerate, the local mode solution
with w > U , does exist right up to E = 1. T h e discussion above referred to lattice
vibrations ; the electronic energy bands for amorphous one-dimensional systems
have been investigated (Phariseau 1960, Taylor 1966); an amorphous system is
represented by a set of potential wells with a nearest-neighbour spacing which is not
constant, but has some probability distribution of values between two prescribed
limits. T h e calculations suggest the persistence of reasonably well-defined energy
bands.
4.5.4. Effectively one- and two-dimensional systems. If a full theory of some three-
dimensional lattice effect is needed, then any simple one-dimensional model must
be improved to apply to three dimensions. There is an interesting halfway house,
however, namely the case of two dimensions. Some of the simple one-dimensional
mathematical models can be extended to two dimensions without too much diffi-
culty, and it may be experimentally possible to make surface layers of a material so
that it behaves as if effectively two dimensional (e.g. Barnes and Steel 1966).
Alternatively, within a three-dimensional lattice there may exist equivalent layers
of atoms which are so widely separated that they behave almost as isolated two-
dimensional systems. Examples of the latter type of behaviour are given by De Vries
Group theory and topology in solid state physics 623

et al. (1968), D e Jongh et al. (1969) and Kamimura and Nakao (1968). T h e work of
Overhauser on spin density waves has been mentioned several times in the present
article; Wolff (1966) has pointed out that Overhauser's results, whatever their
status for a three-dimensional electron gas, are still true in one dimension, and has
suggested that they could be checked by using large magnetic fields in order to
render some favourable solids effectively one-dimensional. Besides the use of
explicitly one- or two-dimensional models to qualitatively represent three-dimen-
sional phenomena, there are some three-dimensional theories which make specific
topological assumptions about the lattice under consideration. For example, the
two-sublattice theory of antiferromagnetism is designed for systems in which the
atom on a site of one magnetic (A) sublattice is surrounded only by neighbours from
the other (B) sublattice. This criterion is satisfied for example by the bcc lattice,
which consists of two interlocking simple cubic lattices. T h e zincblende lattice
consists of two interlocking fcc lattices. Lattices of such a topological type are also
needed for the application of the alternant molecular orbital method to solid state
band theory (Dermit 1962, Calais 1965) since the correlation effects are treated by
allowing the up and down spins to be on separate sublattices. T h e final projection
(t
process (see 0 3) of course gives equal weight to A, J, B) and ($A, t B) arrange-
ments, but the AMO approach reduces the correlation energy.

4.5.5. The Ising model; the theorem of Frobenius. No discussion of the relationship
between lattice calculations in various numbers of dimensions could omit some
reference to the Ising model. Newel1 and Montroll(l953) reviewed the basic known
results for the model, and Mattis (1965) has given a discussion of it, together with
references to other reviews. Wannier (1960) gives an account of the arguments
which establish that a one-dimensional Ising lattice cannot be ferromagnetic at any
temperature, while a square Ising net can become spontaneously magnetized below
some nonzero critical temperature. These results can be phrased in terms of the
concept of long-range order. If we build up a linear Ising chain by adding one
particle at a time to a starting particle, then, if we take only nearest-neighbour
forces, we may use the canonical ensemble to evaluate the probability a that the
1' 1'
(n+ 1)th particle is in, say, the state if the nth particle is in the state. T h e 2 x 2
matrix which describes the four possibilities will take the form

= (;: s) ( a + b = 1). (4.26)

If we take the nth power of M, we obtain a matrix which tends in the limit n+co
towards a matrix with all four elements equal to 4. T h e state of the first particle may
be specified by giving the probabilities p , and qo that it is up or down, respectively.
O n writing p , and qo as the elements of a two-column, x,, the matrix product
MaxO= x, gives the resulting probability distribution for the nth particle. p , and
qn actually tend to 8 as n+m, so that there is no long-range order in this case, i.e.
the state of widely distant particles is not correlated. For a two-dimensional lattice
of finite width, we may repeat these arguments by adding one row at a time to
increase the length of the lattice. T h e same result holds, as may be proved by
using the theorem of Frobenius which asserts that a finite matrix with positive
definite elements has a non-degenerate largest eigenvalue. T h e limiting distribution
for n --f CO then corresponds to that single eigencolumn of the transition probability
matrix which is associated with the largest eigenvalue. T h e situation is essentially
624 J . Killingbeck

that of a Markov chain process with a finite number of possible states (Killingbeck
and Cole 1970). I n order to make long-range order possible, then it is necessary to
introduce a non-finite matrix, and this may be accomplished by making the lattice
infinite in both directions, so that the theorem of Frobenius is not applicable. Hunt
and Newman (1969) have recently used the theorem of Frobenius, together with a
continuity argument, to establish the symmetry type of the greatest eigenvalue
of a matrix which arises in their Ising model approach to PrC1,.
4.5.6. The percolation problem. If a ‘particle’ of some kind is situated on a site in a
regular lattice, and if it can travel along links (bonds) to the nearest-neighbouring
sites, then it may be thought of as hopping (or percolating) along various possible
routes within the lattice. If each bond has a probability p of being ‘open’, and if
there are Z nearest neighbours around each site, then there is a probability (1 -p)”
that the particle will be trapped at the initial site (or at any site which it arrives at
along the route). Each site is taken to have a like environment. (While the papers
on the subject say nothing about this point, the present author surmises that one
of the ‘rules of the game’ must be that the open bonds at a site are each equally
likely to be traversed.) On starting from a given site, a particle will travel for some
average distance, dependent on p , before it is trapped. However, an interesting
result of the theory is that if p exceeds a critical value p,, then there is a nonzero
probability that the particle will visit an infinite number of distinct sites (we take the
lattice to be unbounded). One way to estimate p , is to obtain a power series for the
mean size P ( p ) of the cluster of sites which can be reached from an arbitrary
starting site, and then takes p , to be the value of p for which the series diverges.
Sykes and Essam (1964), use this approach, and also quote some values for p ,
obtained by Monte Carlo methods. Ziman (1969) noted that their results for the
three-dimensional sc, bcc, fcc and diamond lattices could be fairly well summarized
by the rule Zp, = $. H e used this result in a discussion of the localization and
delocalization of the electronic states in a disordered lattice; the p of the percolation
problem is then taken to be dependent on the height of the saddle point in the one-
electron potential which occurs between nearest-neighbour sites.

4.6. Critical points for functions with lattice periodicity


4.6.1. Critical points and the density of states function. I n the previous subsection
some comments have been made on the different behaviour of the density-of-states
function g ( E ) in one and three dimensions, and also on the criteria which determine
whether the normal component of grad, E is zero on the Brillouin zone boundary.
Both these topics are part of the theory of critical points for lattice-periodic func-
tions. This theory deals with continuous functions which have the translational and
rotational periodicity of a lattice, and thus applies to the function E(k) of band theory
or the function w2(k) of lattice vibration theory. Note that the analogy is between
E and w 2 , not w ; for an acoustic mode near the origin, g ( w ) = 2wg(w2)tends to zero
more rapidly than g(w2). We speak of E ( k ) for convenience, but shall quote
examples which involve w 2 ( k ) . T h e density of states g ( E ) may be expressed in the
form
(4.27)

where V is the volume of the lattice and the integral is taken over a constant energy
Group theory and topology in solid state physics 625

surface associated with the energy E. T h e function g ( E ) will be a smooth function


of E unless there are on the surface some points at which ‘U is zero; these are termed
critical points. An analytical critical point is one about which E may be expanded
in a Taylor series (at least up to some nonzero distance). On referring the quadratic
terms to principal axes, the function E ( k ) becomes
E ( k o + A k ) = E ( k o )+ a(Ak,)2+p(Ak,)2+y(4k,)2 (4.28)
x , y , z being the local principal axes. T h e critical point can be assigned an index,
equal to the number of the coefficients which are positive, following van Hove
(1953), or negative, following Phillips (1956). We consider a maximum, with
van Hove index 0. T h e surface of constant energy for an energy slightly less than
E(k,) will be an ellipsoid. There may be other surfaces associated with this energy
at other parts of k-space, so that the total surface of constant energy actually con-
sists of disconnected parts. We are only interested here in the contribution to g ( E )of
the ellipsoidal surface; the other contributions may be combined additively with it,
T h e x,y and z length scales may be changed to make the new coefficients a‘, p‘, y‘
all equal. This simply scales up or down the constant density of states in k-space,
and the contribution to g ( E ) of the resulting spherical shell, computed with origin
at KO, is proportional to
4 4 kI2 4rlk21 47Iizl2 Eo-E
2n ~-
-- --=-
laE/akl - ~ o - a ’ I k 1 2 ) / i 3 k -
I 2a’lkl a ( 01 1
”2


(4.29)

This is the result quoted previously in connection with the theory of localized
modes of vibration. At E = E, the contribution from the ellipsoid thus gives a
discontinuity in the slope, dg/dE being negative infinite at E = E,. T h e contribu-
tion to g ( E ) due to the ellipsoid is clearly zero for E > E,. When E, is a minimum,
the same kind of arguments can be used, except that E-E,, replaces E o - E in
(4.29), and the slope dg/dE is negative infinite as E approaches E, from above.
These slope discontinuities are superimposed on the background contribution from
those other portions of the constant energy surface which are not connected to the
ellipsoid. For the acoustic modes of a lattice, with E , = W , ~= 0, the origin in
k-space is a non-analytical critical point (see below) contributing a density of
states given by g ( u 2 ) K w , which gives the general result that the lattice specific
heat varies as T 3 at low temperatures. For a saddle point, with index 1 or 2, the
constant energy surfaces become hyperboloids near k , and so need not form an
isolated surface; the contribution from the portion of the surface near to k , may
still be evaluated, and leads to an infinite slope discontinuity in g ( E ) ,with the slope
being positive infinite from above and from below for types 1 and 2, respectively.
If one of a,p, y is zero in (4.28), the constant energy surfaces become cylinders
instead of spheres, and the result is a discontinuity in g ( E ) at E = E,,. Stronger
singularities could be brought about by considering lines of critical points instead
of isolated critical points, but computed E ( k ) curves for band theory and lattice
vibration theory do not usually possess such critical points.
T h e discussion given above has involved the expansion (4.28); there could in
principle be non-analytic critical points; for example, E ( k ) might have a sharp peak
maximum at k,; at such a point ‘U would not be defined, whereas for an analytic
critical point it is well defined and zero. Further, E, - E would be non-analytic
around k , ; it might for example vary as a I k - k , I, where 01 depends on the direction
of k - k , , . With a isotropic the calculation leading to (4.29) may be modified to
626 J . Killingbeck
show that g(E) varies as (E,- E)2 below E,. For this case, then, both g(E) and its
slope are continuous at E = E,, but d2g/dE2 is discontinuous at E = E,. This
result holds also when 01 is non-isotropic; in general, if Q: can change sign we are
dealing with a non-analytic saddle point, and g(E) and dg/dE are still continuous
at E,. If a is zero, and higher powers of I k - k, I are needed to describe the variation
of E near k,, then stronger singularities can result; for example, a term /3 1 k - k, l2
gives a discontinuity of dg/dE at E,, as can be seen by noting the result (4.29) for
an analytic critical point. For a non-analytic critical point, the index may be
defined as the number pair (a,b) where a is the number of solid angle regions
starting at k, within which (for small k - k,) E - E, is of positive sign, and b is the
number of negative regions. For isolated analytic critical points we have only the
possibilities (l,O), (2, l), (1,2) and (0, l), but Phillips (1956) pointed out that there

~ ~ ~~

k
Figure 7. An illustrative example of crossing bands in one dimension. If the bands are labelled
1 and 2 as shown, there will be an apparent singularity in the density of states for each
band separately at the crossover energy, but not in the total density of states.

can be non-analytic critical points with other indices. These can arise when sym-
metry reasons make several bands degenerate, and are termed JEuted points by
Phillips. Maradudin et aZ. (1963) list several types of critical point and the char-
acteristic associated behaviour of g(E) near E = E,. T h e need for the study of non-
analytic critical points is due to the fact that the eigenvalues E(k) or cu2(k) are the
eigenvalues of a matrix, and so need not be analytic functions of k. Our discussion
so far has spoken of the function E(k) and has essentially referred to a single band.
T h e total density of states is a sum over all bands, and so the possibility arises that at
some k, values the effects of singularities in different bands may cancel. For
example, if two bands are accidentally degenerate at k,, and if the bands are
numbered in order of increasing energy, then both bands might apparently have a
non-analytic singular point at k, although the total g(E) curve will have no singular-
ity. Phillips (1956) gave the name singular criticalpoints to such critical points arising
from accidental degeneracy. Herring (1937 b) discussed accidental degeneracies of
bands, and concluded that they could occur only at points or along curves, but not
over areas or throughout volumes in k-space. This result is discussed by Weinreich
(1965), and it ensures that the labelling of the bands in order of increasing energy
gives continuous E(k) surfaces. Figure 7 shows a hypothetical case of the effect
Group theory and topology in solid state physics 627

discussed above. If E(k) has, for example, the symmetry of a square in k-space,
then it must have no gradient perpendicular to (100). This may be seen pictorially
in terms of vectors, or simply by noting that ( x , y ) do not belong to the A, rep of C,,
whereas z does. If two bands accidentally cross on the (100) line then we have a case
in which the bands 1 and 2 separately have non-analytic critical points.
4.6.2. The numbers of criticalpoints of various types. T h e effect of singular points on
the density function g ( E ) has been discussed above under the assumption that E ( k )
has zero gradient at certain points. If we further add the requirement that E(k)
must have reciprocal lattice periodicity in k-space, it is possible to show that this
forces E(k) to have a certain minimum number of critical points of each type in each
kY

I
w w
I w w
C B A D

I
c------ -k,

XI

unit cell of k-space. Topological arguments can be used (Phillips 1956, Phillips and
Rosenstock 1958) to relate the numbers of critical points of each type within the
unit cell. Group theory can be used to deduce the existence of certain critical points
in k-space (Phillips 19-56), and the topological relationships then yield more critical
points. T h e smallest collection of critical points which is consistent with both the
group theoretical and topological requirement is called the minimal set of critical
points (Phillips and Rosenstock 1958). We give a brief discussion of a simple
example in order to illustrate some of the ideas of the theory. Consider the square
planar lattice in k-space (figure 8). T h e function E ( k ) is supposed to have the same
value at points D and B because of periodicity. However, if E ( k ) also has reflection
symmetry across the y-axis, it follows that E ( A ) = E(B) = E ( C ) = E(D). Thus, if
E(k) is continuous it will have a maximum or minimum as we pass through k , = T
on the line AD. If E(k) is smooth this means that the normal component of grad, E
is zero on the unit cell boundaries. However, the symmetry arguments do not force
the tangential component of grad,E to be zero, except at the special points
(k, = i T , k, = f T ) , ( k , = 0, k , = i T ) and (k, = f T , k , = 0). T h e point (0,O)
must also be an extremum if E ( k ) is analytic there. If we had known that there is,
628 y. Killingbeck
say, a minimum at X,, the translational periodicity would then produce one at X,.
On proceeding along X, X, (i.e. k, = T ) we would by continuity find some point at
which E is a maximum (provided that it is not constant along k , = T ) . This would
give aEl?k, = 0 at that point, on assuming smoothness as well as continuity; it would
not, however, ensure that this gave us a critical point unless we knew that 2E/ak,
were zero along X,X, as well as at the end points X, and X,. This point seems to
be overlooked in the discussion of Weinreich (1965) ; Wannier (1960) points out that
in the general case the path X,X, can be constructed to pass from XI to X, via
points at which aEjak, is zero, but may then be a curved path. If we invoke the
reflection symmetry in the y-axis then we can derive the result %E/%x= 0, and thus
conclude that there will be a saddle point somewhere along the straight line X,X,.
If we also add the x-axis reflection, we can see that the saddle point is exactly halfway
between X, and X,, as concluded above. There may be other critical points along
X,X,, but these arguments provide a minimum number. As a particular example
of an analytic function with the symmetry properties of the square lattice, consider
the function cosxcosy. This function has critical points at the points already
found by the general arguments above, and also has critical points at the four
points (k, = I. +T,k, = i 8 ~ ) T . h e latter four points are actually critical points at
which two perpendicular surfaces of constant energy intersect and should be
classified as non-analytic. T h e critical points at the origin and at the zone boundary
stay fixed for the more general function
E = 2 A, cos nk, cos nk, (4.30)
n

whereas the non-analytic critical point moves. The saddle points on the zone
boundaries give rise to logarithmic singularities in g ( E ) . If an isotropic function
(e.g. E = I k 1') is used in a density of states calculation in which only states from
within the Brillouin zone may be used, then the critical points on the zone boundary
cause singularities in g ( E ) or its slope as the surface of constant energy passes
through them. Wannier (1960) illustrates this for the fcc lattice. Figure 9 shows
the effect for the square planar lattice. T h e details of the singularities i n g ( E ) will
not necessarily be the same as those obtained by using a periodic E ( k ) , since it is
really a geometrical space-filling property which is being explored, while the true
periodic E ( k ) would have zero gradient at the relevant critical points, although the
integral in (4.27) is also over only that portion of the constant energy surface which
lies within the Brillouin zone. There seems to have been no detailed study of this
point in the literature, but presumably a symmetry argument can be concocted to
prove that the use of the space-filling picture does give correctly the zone boundary
critical points. Our arguments so far suggest the following analytic critical points
in two dimensions; one maximum, one minimum, and one saddle point of each type.
This is actually the minimum number possible. If a function is periodic in x with
period a, it must have at least one maximum and one minimum in one linear unit
cell, and the kind of argument above for the line X, X, can be invoked to show that
the addition of a periodicity in another direction will lead to a minimum total of four
critical points in the two-dimensional unit cell; there will be at least one maximum,
one minimum and one saddle point of each type. There are also relationships
between the numbers no of maxima, n, of saddle points and n, of minima (Phillips
1956)
z o > 1 n 2 > l n, = n,+n, (4.31)
Group theory and topology in solid state physics 629

and these again follow by applying the intuitive argument used for XI X,, remember-
ing that in the general case a curved path will be required. Phillips (1956) gives the
analogues of (4.31) for three dimensions; by adding a third dimension the intuitive
path tracing approach indicates that within the three-dimensional cell E ( k ) must
have at least one maximum, one minimum and three saddle points each of index 1
and 2. I n fact, this building-up process leads to the result (e.g. Weinreich 1965)
that in N dimensions there must be at least C; critical points of index n in each unit
cell, where

(4.32)

is the usual binomial coefficient. T h e Cl?;were shown by van Hove (1953) to be


equal to the Betti numbers of topology; they are then interpreted as the maximum

E
Figure 9. Density-of-states function for a free-electron sphere (circle) in a square planar
lattice. T h e peak occurs when the circle touches the side of the square, while the
vertically descending tail region corresponds to the filling of the four corners of the
square.

number of types of closed n-dimensional surfaces (in the closed N-dimensional


manifold) which cannot be continuously deformed into one another while staying
in the manifold. For example, on a torus, C: = 1 (the point); Ci = 2 (a loop on the
surface of the torus, and one around it); Cg = 1 (a patch of area on the surface of
the torus). Nusimovici (1966) derived the C; from this more abstract viewpoint.
T h e non-analytic critical points which we obtained for the function cos x cosy are
not included in the formulae such as (4.28) and (4.29) for analytic critical points.
Phillips (1956) has discussed how non-analytic critical points may be incorporated
in the equations of the theory. It should also be pointed out that the usual dis-
cussions assume the critical points to be isolated; Jelitto (1969) has given three-
dimensional examples for which this is not the case, and the result is a logarithmic
infinity in g(E) instead of an infinite discontinuity in its slope. One of his examples,
for the bcc lattice, is also discussed by Maradudin et al. (1963). Phillips (1959)
630 J. Killingbeck
showed how information about the E ( k ) curves at principal symmetry points of the
Brillouin zone could be used to interpolate E ( k ) throughout the Brillouin zone and
thus obtain g(E) for diamond type lattices. Both van Hove (1953) and Phillips
(1956) point out that Houston's method (0 3) can produce spurious singularities in
g(E). Houston's method involves interpolation by means of a cubic harmonic
type expansion; if the E ( k ) function along a few high-symmetry directions is used
as initial data, the resulting three-dimensional interpolated function will have
characteristic one-dimensional properties imposed upon it in regions off the high-
symmetry axes, T h e resulting g ( E ) may then have spurious singularities which are
actually appropriate to one-dimensional systems.

4.7. The Fermi surface


4.7.1. The semi-classical approach. T h e use of strong magnetic fields to investigate
the topology of the Fermi surface of metals and semiconductors is very widespread.
These methods have in common a particular geometrical viewpoint, namely the
idea that the Fermi surface in k-space may be regarded as a passive geometrical
object, the dimensions and shape of which may be gauged by studying various
high-field phenomena, without the high field causing a disturbance of the surface
itself. This geometrical viewpoint has been well described by Pippard (1965), and
essentially rests on the use of semi-classical methods, which experience shows to
work well in this case. T h e starting point of the semi-classical approach is the result

k = Ge ( w X H ) (4.33)

which expresses the rate of change of the k-vector of a free electron when it is
acted upon by the Lorentz force due to the magnetic field H . If we replace 'U by
grad, E and take this to give k for a band electron (regarded as a wave packet) in a
solid, then (4.33) implies that the k-vector of an electron at the Fermi surface will
travel around the Fermi surface on a plane perpendicular to H . For a free electron
the time for one orbit of the Fermi surface would be T = 2nmc/(eH). I n the case of
a non-isotropic Fermi surface, however, this result must be modified to the form
(Kittel 1963)
(4.34)

Here S(E) is the area which the €-constant energy contour gives in the plane per-
pendicular to M. T h e real-space motion of the electron is a spiral, and if the motion
of the electron perpendicular to H is quantized in the semi-classical manner, then
the energy eigenvalues include a term nRw,, which leads to the Landau levels of the
electron in the magnetic field. Of course, if H is directed at a given direction
through the Fermi surface, each slice perpendicular to H will have its own wc value.
It is basic to the interpretation of many experimental results to assume that the
dominant effects arise from those regions at which S ( E )is extremal. At such regions
many slices have the same we value, and detailed calculations for simple models
(Kittel 1963, Pippard 1965, Ziman 1965) show that the extremal cross sections do
dominate the results (at least for the oscillatory phenomena which we are dis-
cussing here). I n order to make the theoretical ideas sketched above applicable to
experimental work, it is necessary to perform the experiments on single pure
Group theory and topology in solid state physics 63 1

crystals, so that H can be aligned with respect to well-defined crystal axes, and to
work at low temperatures, so that the electrons can complete many orbits at the
cyclotron frequency wc before being scattered. H must also be large in order to
satisfy the latter criterion, Microwave resonance at the cyclotron frequency in
metals (Azbel-Kaner resonance) can be observed by letting H be parallel to the
metal surface. T h e microwaves (of frequency w ) only penetrate down to the skin
depth, but those electrons which move on spiral paths so that they re-enter the
microwave field region in phase with the oscillations will gain energy. T h e relevant
electrons are those for which w is a multiple of w c ; if the we values corresponding
to extremal Fermi surface sections dominate, only one or two w c values will be
effective. As H is varied, the microwave absorption should then vary periodically
in the variable l / H , since the electron spirals may re-enter the oscillating field region
after any integer number of microwave oscillations. Each extremal wc value will
give its own period in 1/H.

4.7.2. A fundamental theoretical problem. Another aspect of working at low tem-


peratures is that, on the independent-particle model, it keeps the Fermi surface
sharp, whereas it would become blurred at high temperatures. Many quantities of
simple theory (e.g. the electronic specific heat, the Pauli spin paramagnetism)
depend upon the density of states at the Fermi surface, and several workers (Tachiki
and Teramoto 1966, Sat0 et al. 1967) have discussed the stability of superlattice
structures in terms of the stabilizing effect produced when a Brillouin zone boundary
touches the Fermi surface for the conduction electrons. It is thus an important
question of first principle to decide whether the concept of a sharp Fermi surface
is still valid in the presence of electron-electron interaction. This has been dis-
cussed from a theoretical point of view (Migdal 1957, Luttinger 1960, Luttinger and
Ward 1960) and it appears that there is some theoretical justification for retaining the
Fermi surface concept. It is, of course, certainly the case that the rigid Fermi
surface picture fits very well to the majority of experimental results.

4.7.3. The Onsager relation; the de Haas-van Alphen efleect. If the field H is varied,
the Landau levels will one by one pass through the Fermi energy, which is assumed
to be negligibly dependent on H . T h e rate at which the levels pass through the
surface will once again depend on which slice is considered, and each slice will give
a contribution to the thermodynamic properties of the system which is periodic in
1/H. T h e extremal area regions dominate, and the resulting thermodynamic
properties show oscillatory components in 1/H, with periods depending on the
extremal areas. This means that information about the Fermi surface can be
gained by measuring various properties as functions of 1/H at high fields. For
example, the heat capacity of a small specimen may be measured as a function of
1/H (McCombe and Seidel 1967, Sullivan and Seidel 1967), or the absorption
coefficient of ultrasonic waves may be measured as a function of 1/H (Fletcher et al.
1969). T h e basic relationship between the period A ( l / H ) in the reciprocal field
and the corresponding extremal area S of the Fermi surface which gives rise to the
periodicity is the Onsager relation

A-
(k) E.
=- (4.35)

T h e formula (4.35) in principle allows direct determination of the extremal cross


632 J . Killingbeck

sections S for varying field directions, whereas the formula (4.34) gives dS/de in
terms of the cyclotron frequency we. Equation (4.34) may be used to give an
effective mass mY. No mass appears in (4.35), but m* does appear in the expression
for the amplitude of the various de Haas-van Alphen effect harmonics as a function
of temperature; (4.35) gives the fundamental frequency, but harmonics are also
present at very low temperatures. Further, the effect of scattering can be repre-
sented by adding an effective temperature, the Dingle temperature, to the true
temperature when computing the relative magnitudes of the various harmonics
from the zero-scattering theory (Kittel 1963, Shepherd and Gordon 1968). T h e
de Haas-can Alphen eflect is the periodic variation of the magnetic susceptibility in
the variable 1/Hfor high fields; this fluctuating component is superimposed on the
steady Pauli paramagnetism. T h e de Haas-van Alphen effect experiments give a
set of extremal areas (one or more) for each setting of the field. T h e reconstruction

Figure IO. Two simple cylindrically symmetric shapes, showing the necks (broken curves) of
extremal area perpendicular to their symmetry axes.

of the Fermi surface from these data is a difficult task, and is usually attempted in
conjunction with a rough model of the Fermi surface which has been obtained from
band calculations. T h e starting model serves to help in getting several major
features correct, and may then be refined as the rest of the detailed experimental
results are compared with it. T h e general mathematical problem of reconstructing
the Fermi surface from the values of extremal areas as a function of field orientation
was treated by Lifshitz and Pogorelov (1954); their procedure applies for inversion
symmetric closed convex surfaces (convexity is defined in 9 4.4) which involves a
restriction to surfaces for which the radius vector r from any internal origin to the
surface is single-valued. Mueller (1966) has suggested a more practical procedure,
although it is still useful only for surfaces obeying the above criteria. Mueller
pointed out that for cubic lattices the extremal areas along high-symmetry direc-
tions can be used to obtain several of the coefficients ,8 in the cubic harmonic
expansion
c
A ( 4 4) = RG(44) (4.36)

of the area perpendicular to the e,+ direction. T h e squared radius vector length
[ r(8, +)I2 from the inversion centre to the Fermi surface can be similarly expanded
with coefficients a?, and Mueller derived the relationship
%T = 7rPL(0)p2. (4.37)
P&) in (4.37) is a Legendre polynomial. Thus, the first few terms of the expansion
for r2 can be obtained, enabling the Fermi surface to be constructed. Mueller and
Priestley (1966) applied this method to experimental results for palladium, and it
has also been used by later workers. Figure 10 shows two hypothetical surfaces,
both having non-central extremal areas as well as central extremal areas. One of the
surfaces is not convex, however, so that the methods outlined above cannot be
Group theory and topology in solid state physics 633

applied. If the region near the neck of this noh-convex surface is approximated by a
cylinder, then Onsager’s equation (4.35) shows that the period A(l/H) of the
d e Haas-van Alphen oscillations would vary as (cos as the field direction
swings about the axis 8 = 0. This kind of near-(cos O)-l dependence is widely
illustrated in the literature, and is used to detect the existence of necks in the Fermi
surface. A(l/H) will vary a little less rapidly than (cos or a little more rapidly,
depending on whether a neck or a dumb-bell shape is involved. At 0 = 3.. for the
dumb-bell there will be only one extremal area, showing that one extremal area
must vanish as 8 varies from 0 to &T. I n general, some of the extremal areas can only
be observed over a definite range of settings of the field.

4.7.4. Field dependence of Fermi surface in nickel. For metals such as iron or nickel,
which are ferromagnetic, the study of the de Haas-van Alphen effect involves
special problems, T h e most obvious one is that the internal field is not equal to the
applied external field, and so must be calculated, making allowances for shape-
dependent demagnetizing fields (e.g. Stark and Tsui 1948). A more novel effect is
that relating to the band structure in ferromagnetic metals. T h e notion of different
bands for different spins was mentioned in $ 3 of this article; in a ferromagnetic
metal the combined effect of a ferromagnetic band splitting and spin-orbit coupling
can give rise to a small change in some sections of the Fermi surface as the external
field direction is varied (Hodges et al. 1967, Gold 1948). This is a case in which the
rigid Fermi surface model has to be modified. For nickel the Fermi surface includes
hole pockets at the X point of the Brillouin zone (reciprocal lattice vector 0, 0 , l ) .
I n a tight-binding approach two appropriate d-band degenerate orbital basis
functions at X would be the Bloch orbitals formed from xx and yx atomic orbitals.
Including spin, there would be a fourfold degeneracy. However, in ferromagnetic
nickel there will be a splitting AE between the 4
and pairs, ( x x f , y x f ) and
(.xi, yx.1). Spin-orbit coupling matrix elements link the states tt, $4 and ti.
If, following Hodges et al. (1967), we assert that the spin labels f, J, must refer to the
direction of the strong external field, which is aligning the magnetization, then the
matrix elements of the spin-orbit coupling operator, which is taken in the isotropic
form 1 . s, will include factors cos 8 or sin 8. Here 6 is the angle between H and the
axis (0, 0 , l ) . For example, the matrix element ( x x f I 1 . s I y z f ) will be equal to
(xx I I, [ y x ) (f 0 I s, 1 f 6) with I,, s, referred to (0, 0 , l ) . This matrix element thus has
a cos 0 dependence. As the.H direction varies, so does the spin-orbit splitting of the
t band (and similarly of the $ band). There will also be some second-order effects
due to mixing of the f and $ bands across the gap AE. This field dependence of the
levels at X is thought to be responsible for the behaviour of some of the hole pockets
i n the Fermi surface of nickel (Gold 1948); the directional dependence of these
pockets does not reasonably fit into the rigid Fermi surface model.

4.7.5. The magnetoresistance of nickel. We have outlined above the basic principles
of the de Haas-van Alphen effect, and of those experimental methods which employ
the Onsager equation (4.35). There are other experiments which give information
about the topology of the Fermi surface; the measurement of the high-field magneto-
resistance is one of these. (Low-field magnetoresistance experiments do give useful
information (e.g. Jeavons and Saunders 1969) but are usually difficult to interpret.)
High-field magnetoresistance measurements are interpreted to give information
about open orbits on the Fermi surface. Polyvalent metals do not in general have
634 J , Killingbeck
simple closed convex Fermi surfaces which are entirely within the first Brillouin
zone. A useful way to see this is to draw the constant energy surfaces for a free
electron, folding them back so as always to lie within the reduced zone of some
reciprocal lattice. Very complicated shapes can be produced in this way (e.g.
Ziman 1963, Pippard 1965), and the effect of a weak lattice potential (nfe approach)
is to give energy gaps at the zone boundary which distort or nip off some parts of the
free-electron shape. Various portions of the Fermi surface will then belong to
different bands, and in a high magnetic field the electron orbits in k-space will stick
to the separate band Fermi surfaces, except for the cases where band gaps are so

Figure 11, An orbit A-B-C-D in the repeated zone scheme, for the case of a square lattice in
which the Fermi surface touches the zone boundary. T h e interior of the orbit is the
unoccupied shaded region, and the orbit is termed a ‘hole orbit’ (electron orbits have
occupied interior regions). The wave function is continuous around the orbit, whereas
an apparent series of jumps (e.g. B +- B’) would appear in a reduced zone representation.
T h e formulae of the theory are applicable to hole orbits if S(E)is interpreted as the
shaded area.

small that magnetic breakthrough can occur (Harrison and Webb 1960, Pippard
1965). If the electron path around the Fermi surface is viewed in the reduced zone
scheme, then jumps may be involved, whereas the extended zone scheme makes it
clear that the path is essentially continuous in k-space (figure 11). When the
magnetic field H is aligned in certain directions it may be the case that open orbits
(i.e. orbits which are not closed and periodic in the extended zone picture) exist
perpendicular to the field. It is characteristic of such orbits (Harrison and Webb
1960, Gold 1968) that they produce a transverse magnetoresistance which does not
saturate in high fields. T h e simple theory asserts that in such cases an H2-depen-
dence of the magnetoresistance should be found at high fields ; the experimentally
determined exponent is often a little less than 2. As the field direction is rotated,
there may be directions for which the high-field magnetoresistance does saturate,
and this implies that no open orbits exist perpendicular to H . For a dosed convex
Fermi surface the magnetoresistance will saturate for all field directions, since there
Group theory and topology in solid state physics 635

are no open orbits. There are circumstances under which the H2-dependence can
persist in the absence of open orbits (Gold 1968), namely when the metal is com-
pensated, containing equal numbers of electrons and holes in the various sheets of
the Fermi surface. Nickel, with atomic number 28, and a cubic structure, was
thought to be compensated, so that the H2-dependence of the transverse magneto-
resistance would hold for all field directions. However, Reed and Fawcett (1965)
found that for some ranges of field direction saturation did occur. Their suggestion
was that there is in ferromagnetic nickel an extra band splitting between and 1' 4
1'
spin bands, so that the and J bands have different energies and should be con-
sidered separately. This is exactly the effect predicted by the band theory approach
to ferromagnetism, and, as discussed in the preceding paragraph, the same idea has
been used to explain the anomalous de Haas-van Alphen results for the X hole
pockets in nickel. These results for nickel are of more recent origin than the work
surveyed in the Fermi surface volume edited by Harrison and Webb (1960), but
that volume gives a good survey of basic theoretical and experimental techniques.
4.7.6. The Kohn anomaly. I n the preceding section of the present article the
behaviour of the w2(k) function for phonons was discussed. Kohn (1959 a, b)
pointed out that in a metal the force-constant model may not describe the w 2 ( k )
function, because the electron gas will give a wave vector-dependent screening of the
interaction between the vibrating ions. T h e phonon frequency w may be set
effectively equal to zero with negligible error when this screening is computed. T h e
dielectric constant function E(k, 0) for a free-electron gas has a logarithmic singu-
I
larity at k I = I 2kF I (Ziman 1965), and this would in principle give rise to an abrupt
I
variation (an infinity in slope) in w ( k ) at k I = I 2kF1. For the case of a nonspherical
Fermi surface in a metal the effect will occur for k = 2k, where 2kF is the appro-
priate vector diameter of the Fermi surface (and may have more than one value),
and can also occur at k = 2k,+ K , where K belongs to the reciprocal lattice. This
new possibility is a typical solid state Umklapp effect. Woll and Kohn (1962)
concluded that the effect is also reduced in magnitude when the one-electron wave
functions are more realistically taken to be Bloch functions rather than pure plane
waves, but did interpret experimental neutron scattering experiments on lead as
showing the effect. T h e jump in w ( k ) is positive when an electron portion of the
Fermi surface is spanned by k , and negative when a hole portion is spanned, so that
in principle the Kohn anomaly should provide information not readily obtained
from the de Haas-van Alphen effect. So far the effect has not been clearly identified
for many metals; Sharp (1969) has treated Kohn anomalies for niobium and refers
to earlier results for aluminium and for niobium-molybdenum alloys,

References
ABRIKOSOV,A. A., GORKOV, L. P., and DZYALOSHINSKI, I. E., 1963, Methods of Quantum Field
Theory in Statistical Physics (New Jersey : Prentice-Hall).
ADAMS,W. H., 1962a, Phys. Rev., 127, 1650-8.
ADAMS,W. H., 1962 b , J . chem. Phys., 37, 2009-18.
ALLAN,G. A. T., and BETTS,D. D., 1967, Proc. Phys. Soc., 91, 341-52.
ALTMANN, S.L., 1963, Rev. mod. Phys., 35, 641-5.
ALTMANN, S.L., and BRADLEY, C. J., 1965 a, Rev. mod. Phys., 37, 33-45.
ALTMANN, S. L., and BRADLEY, C. J., 1965 b, Proc. Phys. Soc., 86, 915-31.
ALTMANN, S. L., and CRACKNELL, A. P., 1965, Rev. mod. Phys., 37, 19-32.
AMOS,T., and SNYDER, L. C., 1964,J. chem. Phys., 41, 1773-83.
636 J. Killingbeck
ANDERSON, J. M., 1969, J . magn. Res., 1, 89-97.
ANDERSON, P. W., 1964, Concepts in Solids (New York: Benjamin).
ARAKI,G., 1948, Progr. theor. Phys., 3, 152-9.
ARMSTRONG, L., 1966,J. math. Phys., 7, 1891-9.
ARMSTRONG, L., 1968, Phys. Rev., 166, 63-7.
ARNOLD, B. H., 1962, Intuitive Concepts in Elementary Topology (New Jersey: Prentice-Hall).
ATOJI,M . , 1965, A m . J . Phys., 33, 212-19.
AUSTIN,B. J., HEIKE,V., and SHAM,L. J., 1962, Phys. Rev., 127, 276-82.
AXE,J. D., 1963, J. chem. Phys., 39, 1154-60.
AXE,J. D., and BURKS,G., 1966, Phys. Rev., 152, 331-40.
BACON,G. E., 1965, Sci. Prog., Oxf,, 56, 59-81.
BALLHAUSEN, C. J., 1962, Introduction to Ligand Field Theory (New York: McGraw-Hill).
BALT,S., 1968, .Mol. Phys., 14, 233-9.
BANWELL, C. N., and PRIMAS, H., 1962, Molec. Phys., 6, 225-56.
BARNES, &I. W., and STEEL,W. A., 1966, J , chem. Phys., 45, 461-465.
BARTRAM, R.H., STONEHAM, A. M., and GASH,P. x., 1968, Phys. Rev., 176, 1014-24.
BASSANI,F., and CELLI,V., 1961,J. phys. Chem., 20, 64-75.
BATES,A., and STEVENS, K. W. H., 1969,J. Phys. C: Solid St. Phys., 2, 1573-85.
BATES,C. A., 1964, Proc. Phys. Soc., 83, 465-72.
BATES,C. A., 1968,J. Phys. C: Proc. Phys. Soc., 1, 877-88.
BATES,C. .4.,and DIXON,J. M., 1969, J . Phys. C: Solid St. Phys., 2, 2209-24.
BATES,C. A., and STEVENS, K. W. H., 1961, Proc. Phys. Soc., 68, 1321-39.
BEALLFOWLER, W., 1963, Phys. Rev., 132, 1591-9.
BEEBY,J. L., 1967, Proc. Phys. Soc., 90, 765-77.
BELL,D. G., 1954, Rev. mod. Phys., 26, 31 1-20.
BENNETT, A. J., 1967, Phys. Rev., 153, 482-7.
BERGE,C., 1962, The Theory of Graphs and its Applications, translated by A. Doig (London:
Methuen).
BERGGREN, K. F., and JOHANSSON, B., 1968, Physica, 40, 277-89.
BERSUKER, I. B., 1963,J. ex@.theor. Phys., 16, 933-8.
BERSUKER, I. B., and VEKHTER, B. G., 1966, Phys. Stat. Solidi, 16, 63-8.
BERTAUT, E. F., 1969,J. Phys. Chem. Solids, 30, 763-73.
BETHE,H. A., 1929, Ann. Phys., Lpz., 3, 133-208.
BIEDENHARN, L. C., and VAN DAM,H., 1965, Quantum Theory of Angular iWomentum (New
York: Academic Press).
BIRKHOFF, G . A., and MACLANE, S., 1953, A Survey of Modern Algebra (New York: Mac-
Millan).
BIRMAN, J. L., 1962, Phys. Rev., 127, 1093-106.
BISHTON,S.S., ELLIS,M. M., NEWMAN, D. J., and SMITH,J., 1967,J. chem. Phys., 47, 4133-6.
BISHTON,S. S., and NEWMAN, D. J., 1968, J . Phys. Chem. Solids, 29, 1245-53.
BJORK,R. L., 1957, Phys. Rev., 105, 456-9.
BLACKMAN, M . , 1935, Proc. R. Soc. A, 148, 365-83.
BLOCH,F., 1929,Z. Phys., 57, 545-55.
BLOOR,D., ELLIS,E., MARTIN,D. H., and WADHAM, A. J., 1968,J. appl. Phys., 39, 971-2.
BOTTGER, H., 1967, Phys. Stat. Solidi, 24, 65-75.
BOUCKAERT, L. P., SMOLUCHOWSKI, R., and WIGKER,E. P., 1936, Phys. Rev., 50, 58-67.
BOUTEN,M.,1969, Physica, 42, 572-80.
BRADBURY, ILI. I., and I\;EWMAN, D. J., 1967, Chem. Phys. Lett., 1, 44-5.
BRADLEY, C. J., and DAVIES, B. L., 1968, Rev. mod. Phys., 40, 359-79.
BRILLOUIN, L., 1953, Wave Propagation in Periodic Structures (New York: Dover).
BRINK,D. NI,, and SATCHLER, G. R., 1968, Angular Momentum (Oxford: Clarendon Press).
BRINKMAN, TV. F., and ELLIOTT,R.J., 1966, Proc. R. Soc. A, 294, 343-58.
BROERMAN, J. G., 1968 a,J. Phys. Chem. Solids, 29, 1147-66.
BROERMAN, J. G., 1968 b, Phys. Stat. Solidi, 25, 757-62.
BRUECKNER, K. A., 1967, in Mathematical Methods in Solid State and Superfluid Theory
(Edinburgh: Oliver and Boyd).
BUERGER, M. J., 1956, Elementary Crystallography (New York: Wiley).
BURNS,G.. 1962, Phys. Rev., 128, 2121-30.
BURNS,G., and AXE,J. D., 1966, J . chem. Phys., 45, 4362-3.
Group theory and topology in solid state physics 637

CALAIS, J. L., 1965, A r k . Phys., 28, 479-98.


CALLAWAY, J., 1957,J. Electronics, 3, 330-40.
CARDONA, M., 1963, J . Phys. Chem. Solids, 24, 1543-55.
CARDONA, M., POLLACK, F. H., and BROERMAN, J. G., 1965, Phys. Lett., 19, 276-7.
C ~ R N A LW.L , T., FIELDS, P. R., and WYBOURNE, B. G., 1965,J. chem. Phys., 42, 3797-806.
CASIMIR, H. B. G., 1963, On the Interaction between Atomic Nuclei and Electrons (San Francisco:
Freeman).
CHAN,Y. W., 1969, Brookhaven Rep., KO.BNL 10155.
CLINTON, W. L., and RICE,B., 1959,J. chem. Phys., 30, 542-6.
DESCLOIZEAUX, J., 1963, Phys. Rev., 129, 554-66.
COCHRAN, W., 1963, Rep. Prog. Phys., 26, 1-45.
COFFMAN, R. E., 1965, J. chem. Phys., 48, 609-18.
COHEN, ill. L., and BERGSTRESSER, T. K., 1966, Phys. Rev., 141, 789-96.
COHEN, M.L., and HEINE,V., 1961, Phys. Rev., 122, 1821-6.
COLELLA, R., and MERLINI, A., 1966, Phys. Stat. Solidi, 18, 157-66.
CONDON, E. U., 1968, Rea. mod. Phys., 40, 872-5.
CONDON, E. U., and ODABASI, H., 1966, in Quantum Theory of Atoms, Molecules, and the Solid
State (New York: Academic Press).
CONDON, E. U., and SHORTLEY, G. H., 1935, The Theory of Atomic Spectra (Cambridge:
Cambridge University Press).
CORIO,P. I,., 1968,J. math. Phys., 9, 1067-71.
COXETER, H. S. &I., 1963, Regular Polytopes (New York: Macmillan).
CRACKNELL, A. P., 1969, Rep. Prog. Phys., 32, 633-707.
CRACKNELL, A. P., 1968, Adv. Phys., 17, 367-420.
CRACKNELL, A. P., and JOSHUA, S. J., 1968,J. Phys. A: Proc. Phys. Soc., 1, 40-2.
CURIE,D., 1963, Luminescence in Crystals, translated by G. F. J. Garlick (London: Methuen).
DAVIS,H. T., 1962, Introduction to Nonlinear Darerential and Integral Equations (Xew York:
Dover).
DAVYDOV, A. S., 1962, Theory of Molecular Excitons, translated by NI. Kasha and M.
Oppenheimer (New York: McGraw-Hill).
DEAN,P., 1968, J . Phys. C: Proc. Phys. Soc., 1, 22-34.
DEHN,E., 1960, Algebraic Equations (New York: Dover).
DERMIT, G., 1962, Phys. Rev., 127, 1110-21.
DIEKE,G. H., and CROSSWHITE, H. M., 1963, Appl. Opt., 2, 675-86.
DIMMOCK, J. O., 1963, Phys. Rea., 130, 1337-44.
DIMMOCK, J. O., and WHEELER, R. G., 1962,J. phys. Chem., 23, 729-41.
DIRAC, P. A. M., 1927, Proc. R. Soc. A, 114, 243-65.
DIRAC, P. A. M., 1958, The Principles of Quantum Mechanics (Oxford: Oxford University Press).
DOGGETT, G., 1966, J . Phys. Chem. Solids, 27, 99-110.
DOMB,C. A., and HEAP,B. R., 1967, Proc. Phys. Soc., 90, 985-1001.
DOMB,C. A., and WOOD,D. W., 1965, Proc. Phys. Soc., 86, 1-16.
DORAIN, P. B., and WHEELER, R. G., 1966, J. Phys. Chem., 45, 1172-81.
DUPONT, A., 1967,J. Opt. Soc. Am., 57, 867-9.
DYSON, F. J., 1956, Phys. Rev., 102, 1217-30.
ECK,J. S., LEE,Y. K., WALKER, J. C., and STEVENS, R. R., 1967, Phys. Rev., 156, 246-50.
EISENSTEIN, J. C., 1961,J. chem. Phys., 35, 2097-100.
EISENSTEIN, J. C., 1963, in Loa Symposium on Paramagnetic Resonance, Vol. 1 (New York:
Academic Press).
ELLIOTT, R. J., 1954, Phys. Rev., 96, 280-7.
ELLIOTT,R. J., 1966, in Phonons in Perfect Lattices and in Lattices with Point Imperfections
(Edinburgh: Oliver and Boyd).
ELLIOTT, J. P., JUDD,B. R., and RUNCIMAN, W. A., 1957, Proc. R. Soc. A, 240, 509-23.
ELLIS,M. M., 1968,J. chem. Phys., 49, 4037-43.
ELLIS,M. M., and NEWMAN, D. J., 1967,J. chem. Phys., 47, 1986-93.
ENGLMAN, R., 1961, Molec. Phys., 4, 183-188.
ENGLMAN, R., 1966,J. chem. Phys., 45, 3862-9.
EXGLMAN, R., and RUPPIN,R., 1968,J. Phys. C : Proc. Phys. Soc., 1, 614-29.
ERDOS,P., 1966, J. Phys. Chem. Solids, 27, 1705-20.
ESSAM, J. W., and FISHER, M. E., 1970, Rev. Mod. Phys., 42, 271-88.
63 8 J . Killingbeck

FALICOV,L. M., 1966, Group Theory and Its Physical Applications (Chicago: University of
Chicago Press).
FANO,U., and RACAH,G., 1959, Irreducible Tensorial Sets (New York: Academic Press).
FERGUSON, J., GUGGENHEIM, H. J., and TANABE, Y., 1966,J. Phys. Soc.Japan, 21, 692-704.
FICK,E., 1962,Z. Phys., 169, 100-13.
FISCHER, F., 1967,Z. Phys., 204, 351-74.
FLETCHER, R., MACKINNON, L., and WALLACE, W. D., 1969, Phil. Mag., 20, 245-58.
FRALEIGH, J. B., 1967, A First Course in Abstract Algebra (Reading, Massachusetts: Addison-
Wesley).
FRANK, F. C., and KASPER,J. S., 1958, Acta Cryst., 11, 184-90.
FREEMAN, A. J., and WATSON,R. E., 1962, Phys. Rev., 127, 2058-75.
GABRIEL, J., 1964, J. math. Phys., 5, 494-504.
GALLUP,G. A., 1968, J . chem. Phys., 48, 1752-9.
GAMBA, A., 1968, J. math. Phys., 9, 186-92.
GAY,J. G., ALBERS, W. A., and ARLINGHAUS, F. J., 1968,J. Phys. Chem. Solids, 29, 1449-59.
GODDARD, W. A., 1967, Phys. Rev., 157, 81-93.
GODDARD, W. A., 1968, Phys. Rev., 174, 659-62.
GOLD,A. V., 1968,J. appl. Phys., 39, 768-774.
GOLDSTOKE, J., 1957, Proc. R. Soc. A, 239, 267-79.
GOSCIKSKI, O., and ~ H R N Y.,, 1968, Quantum Chemistry Group, Uppsala, Rep., No. 214.
GRIFFITH, J. S., 1962, The Irrreducible Tensov Method f o r Molecular Symmetrry G~oups(London :
Prentice-Hall).
GROSS~LZAN, I., and MAGNUS, W., 1964, Groups and Their Graphs (New York: Random House).
HAGSTON, W. E., 1967, Proc. Phys. Soc., 92, 1101-5.
HALL,G. G., 1967, Applied Group Theory (London: Longmans).
HAM,F. S., 1965, Phys. Rev., 138, 1727-40.
HAM,F. S., 1967, General Electric Rep., No. 67-C-315.
HAM,F. S., and SEGALL, B., 1961, Phys. Rea., 124, 1786-96.
HARRISON, W. A., 1966, Pseudopotentials in the Theory of Metals (New York: Benjamin).
HARRISON, W. A., and WEBB,M. B., (eds), 1960, The Fermi Surface (New York: Wiley).
HARTMANN, H., and SCHrlRMANK, A., 1969, 2. ivaturf., 24a, 1117-23.
HEAP,B. R., 1967, NPL report Ma. 57.
HEGSTROM, R. A., and LIPSCOMB, W. Ii,,1968, Rea. mod. Phys., 40, 354-8.
HEINE,V., 1960, Group Theory in Quantum iwechanics (Oxford: Pergamon).
HEISENBERG, W., 1931, Ann. Phys., Lpz., 10, 888-904.
HELLWEGE, A. M., and HELLWEGE, K. H., 1952, 2. Phys., 133, 174-86.
HELLWEGE, K. H., 1948, Ann. Phys., Lpz., 4, 95-160.
HENNING, J. C. M., VAN DEN BOOM,H., and DIELEMAN, J., 1966, Philips Res. Rep., 21, 16-26.
HERMAN, F., and SKILLMAN, S., 1963, Atomic Structure Calculations (New Jersey: Prentice-
Hall).
HERRING, C., 1937 a, Phys. Rev., 52, 361-5.
HERRING, C., 1937 b, Phys. Rev., 52, 365-73.
HERRING, C., 1942, J . Franklin Inst., 233, 525-43.
HERRING, C., 1966, in Magnetism, Vol. IIB, Eds G. Rad0 and H. Suhl (A-ew York: Academic
Press).
HERZFELD, K. F., 1949, Rev. mod. Phys., 21, 527-30.
HOCHLI,U. T . , 1967, Phys. Rev., 162, 262-73.
HODGES, L., STONE,D. R., and GOLD,A. V., 1967, Phys. Rea. Lett., 9, 655-8.
HOERKI, J. A., 1961,J. chem. Phys., 34, 508-13.
HOFELICH-ABATE, E., and HOFELICH, F., 1968 a, 2. Phys., 211, 142-51.
HOFELICH-ABATE, E., and HOFELICH, F., 1968 b, 2. Phys., 209, 13-32.
HOLMAN, W., 1969, Ann. Phys., N . Y., 52, 176-91.
HOLSER, W. T., 1961, Acta cryst., 14, 1236-42.
HORTON, G. K., and SCHIFF,H., 1959, Prroc. R. Soc. A, 250, 248-65.
HOUSTON, W. V., 1948, Rev. mod. Phys., 20, 161-5.
HUANG, C. Y., 1967 a, Phys. Rev., 154, 215-9.
HUANG, C. Y., 1967 b, Phys. Rev., 199, 683-6.
HUBBARD, J., RIMMER, D. E., and HOPGOOD, F. R. A., 1966, Proc. Phys. Soc., 18, 13-36.
HUGHES, A. E., 1966, Proc. Phys. Soc., 87, 535-42.
Group theory and topology in solid state physics 639

HUGHES, A. E., and RCNCIMAN, W. A., 1967, Proc. Phys. Soc., 90, 827-38.
HUNT, R. A., and NEWMAN, D. J., 1969, J . Phys. C: Solid St. Phys., 2, 75-83.
HUNTER, T. F., 1968, hfolec. Phys., 14, 171-81.
HURST, C. A., 1964,J. math. Phys., 5, 90-100.
HURST,C. A., 1966,J. math. Phys., 7, 81-7.
HUTCHINGS, M. T., 1964, in Solid State Physics, Vol. 16, Eds F. Seitz and D. Turnbull (New
York: Academic Press).
INDENBOM, V. L., 1959, Kristallograjiya, 5, 513-525.
INOUE,M., 1963, Phys. Rev. Lett., 11, 196-7.
JANSEN,L., 1967, Phys. Rev., 162, 63-8.
JARRETT, H. S., 1959,J. chem. Phys., 31, 1579-85.
JEAVONS, A. P., and SAUNDERS, G. A., 1969, Proc. R. Soc. A, 310, 415-32.
JELITTO,R. J., 1969,J. Phys. Chem. Solids, 30, 609-26.
JESSON,J. P., 1959,J. chem. Phys., 48, 161-8.
JESSON,J. P., WEILER, J. F., and TROFIMENKO, S., 1968,J. chem. Phys., 48, 2058-66.
JOHNSTON, D. F., 1960, Rep. Prog. Phys., 23, 66-153.
JONES,W., MARCH,N. H., and TUCKER, J. W., 1965, Proc. R. Soc. A, 284, 289-301.
DEJONGH,L. J., BOTTERMAN, A. C., DE BOER,F. R., and MIEDEMA, A. R., 1969,J. appl. Phys.,
40, 1363-5.
JORGENSEN, C. K., 1962, Absorption Spectra and Chemical Bonding in Complexes (Oxford:
Pergamon).
JORGENSEN, C. K., PAPPALARDO, R., and SCHMIDTKE, H. H., 1963,J. Chem. Phys., 39, 1422-30.
JUDD,B. R., 1956, Proc. Phys. Soc., 69, 157-64.
JUDD,B. R., 1957, Proc. R. Soc. A, 241, 122-31.
JUDD,B. R., 1959, Proc. Phys. Soc., 74, 330-9.
JCDD,B. R., 1962, Phys. Rev., 127, 750-61.
JUDD,B. R., 1963, Opevator Techniques in Atomic Spectvoscopy (New York: McGraw-Hill).
JUDD,B. R., 1966,J. chem. Phys., 44, 839-40.
JUDD,B. R., 1967a, Phys. Rev., 162, 28-37.
JUDD,B. R., 1967 b, Second Quantisation and Atomic Spectroscopy (Baltimore: John Hopkins
Press).
KAMIMURA, H., 1962, Phys. Rev., 128, 1077-84.
KAMIMURA, H., and NAKAO,K., 1968, J . Phys. Soc. Japan, 24, 1313-25.
KASTELYN, P. W., 1967, in Graph Theory and Theoretical Physics, Ed. F. Harary (New York:
Academic Press).
KENAN,R. P., 1967, Phys. Rev., 159, 430-8.
KIEL,A., 1966, Phys. Rev., 148, 247-56.
KILLINGBECK, J., 1969 a, J . Phys. C: Solid St. Phys., 3, 19-22.
KILLINGBECK, J., 1969 b, Phys. Stat. Solidi, 36, K67-69.
KILLINGBECK, J., 1970, J. math. Phys., in the press.
KILLINGBECK, J., and COLE,G. H . A., 1970, Mathematical Techniques and Physical Applications
(New York: Academic Press).
KITTEL,C., 1963, Quantum Theory of Solids (New York: Wiley).
KLEINER, W. H., 1952,J. chem. Phys., 20, 1784-91.
KLEINER, W. H., and KAPLAN,T. A., 1969,J. math. Phys., 10, 236-8.
KLEINMAN, D. R., 1964, Bell System Tech.J., 43, 1505-32.
KLEINMAN, D. R., and PHILLIPS, J. C., 1962, Phys. Rev., 125, 819-24.
KNOX,R. S., and GOLD,A., 1964, Symmetvy in the Solid State (New York: Benjamin).
KOBE,D., 1965, Preprint 158. Quantum Chemistry Group, Uppsala.
KOHN,W., 1959 a, Phys. Rev. Lett., 2, 393-4.
KOHN,W., 1959 b, Phys. Rev., 115, 809-21.
KOIDE,S., and PRYCE, M. H. L., 1958, Phil. Mag., 3, 607-24.
KONINGSTEIN, J. A., 1968, Phys. Rev., 174. 477-8.
KOPELMANN, R., 1967,J. chem. Phys., 47, 2631-48.
KOSTER,G. F., 1955, Phys. Rev., 98, 514-5.
KOSTER, G. F., 1957, in Solid State Physics, Vol. 5 , Eds F. Seitz and D. Turnbull (New York:
Academic Press).
KOSTER,G. F., 1958, Phys. Rev., 2, 227-31.
KOSTER, G. F., and SLATER, J. C., 1954, Phys. Rev., 96, 1208-23.
640 J . Killingbeck

KOSTER,G. F., and STATZ,H., 1959, Phys. Rev., 113, 445-54.


KRASNOSELSKII, M. A., 1963, Topological Methods in the Theory of Nonlinear Integral Equations
(Oxford : Pergamon).
KUBLER, J., 1967,Z. Phys., 201, 172-99.
KUBO,R., 1962,J. Phys. Soc.Japan, 17, 1100-20.
KUNZ,A. B., 1967, Phys. Reo., 159, 738-41.
LARSON, E. G., and THORSON, W. R., 1966, J . chem. Plzys., 45, 1539-54.
LEDERMANN, W., 1949, The Theory of Finite Groups (Edinburgh: Oliver and Boyd).
LEFEBVRE, R., and MOSER,C., 1969, Adv. chem. Phys., Vol. 14 (New York: Wiley).
LENANDER, C. J., and WONG,E. Y., 1963, J . chem. Phys., 38, 2750-2.
LENGLART, P., 1967, J . Phys. Chem. Solids, 28, 2011-25.
LENGLART, P., 1968, Ann. Phys., Paris, 3, 27-61.
LEVY,P. M., 1969,J. appl. Phys., 40, 1139-41.
LEVY,P. NI., and COPLAND, G . M., 1969, Phys. Rev., 180, 439-41.
LIEB,E., and MATTIS,D. C. (eds), 1966, Mathematical Physics in One Dimension (New York:
Academic Press).
LIEHR, A. D., and BALLHAUSEN, C. J., 1957, Phys. Rev., 106, 1161-3.
LIFSHITZ, I. ILL, and POGORELOV, A. V., 1964, Dokl. Akad. Nauk. S S S R , 96, 1143-60.
LINDNER, P., and LUNNELL, S.,1967, Preprint 189, Quantum Chemistry Group, Uppsala.
LINNETT, J. W., 1964, The Electronic Structure of Molecules (London: Methuen).
LITVIN, D. B., and ZAK,J., 1968, J . math. Phys., 9, 212-21.
LLOYD,P., 1965, Proc. Phys. Soc., 86, 825-32.
LOH,E., 1968, Phys. Rea., 175, 533-6.
LONGUET-HIGGINS, H. C., 1962, illolec. Phys., 6, 445-60.
Low, W., 1958, Phys. Rev., 109, 247-55.
Low, W., 1960, Solid State Physics, Suppl. 2 (New York: Academic Press).
LOWDIN,P. O., 1962 a,J. math. Phys., 3, 969-82.
LOWDIN,P. O., 1962 b,J. appl. Phys., Suppl., 33, 251-80.
LOWDIN,P. O., 1962 c, Rev. mod. Phys., 34, 80-7.
LOWDIN,P. O., 1962 d,J. math. Phys., 3, 1171-84.
LOWDIN, P. 0.)1964, Rev. mod. Phys., 36, 966-76.
LOWDIN, P. O., 1967, Rev. mod. Phys., 39, 259-87.
LOWDIK,P. 0.)1968, Quantum Chemistry Group, Uppsala, Rep., Bo. 219.
LUDER, W.F., 1967, The Electron-repulsion Theory of the Chemical Bond (New York: Reinhold).
LUEHRBIANN, A. W., 1968, Adv. Phys., 17, 1-77.
LUTTINGER, J., 1951, Philips Res. Rep., 6, 303-10.
LUTTINGER, J. M., 1960, Phys. Rezl., 119, 1153-63.
LUTTINGER, J. 14,)and WARD,J. C., 1960, Phys. Rea., 118, 1417-26.
LYUSTERNIK, L. A., 1963, Convex Figures and Polyhedra (New York: Dover).
MACFARLANE, R. M., 1955, J . chem. Phys., 39, 3118-26.
MACKAY, A. L., 1957, Acta cryst., 10, 543-8.
RIALLI, G., 1967, J . chem. Phys., 47, 2217-19.
MALLI,G., 1968,J. chem. Phys., 48, 1088-91.
MALLI,G., and SAXENA, K. M. S., 1969, J . chem. Phys., 50, 174-81.
MANENKOV, A. A., and ORBACH, R. (eds), 1966, Spin-Lattice Relaxation in Ionic Soleds (New
York: Harper and Row).
MARADUDIN, A. A., MONTROLL, E. W., and WEISS,G. H., 1963, Theovy of Lattice Dynamics in
the Harmonic Approximation (New York : Academic Press).
MARADUDIN, A. A., and VOSKO,S.H., 1968, Rev. Mod. Phys., 40, 1-37.
MARCH, N. H., YOUNG,W.H., and SAMPANTHAR, S., 1967, The Many-bodjj Problem in Quantum
hfechanics (Cambridge : Cambridge University Press).
M A R I O T , L., 1962, Group Theory and Solid State Physics, translated by A. Nussbaum (London:
Prentice-Hall).
MARSHALL, W. (ed.), 1967, Theory of Magnetism in Transition ,Vetals (New York: Academic
Press).
MARSHALL, W., and STUART, R., 1961, Phys. Rev., 123, 2048-58.
M.4TTHIAS, B. T., BOZORTH, R. M., and VANVLECK,J. H., 1961, Phys. Rev. Lett., 7, 160-1.
MATTIS, D. C., 1965, The Theory of Magnetism (iVew York: Harper and Row).
MATSEN, F. A., 1964, in Advances in Quantum Chemistry, Vol. I (New York: Academic Press).
Group theory and topology in solid state physics 64 1

MATTUCK, R. D., 1967, A Guide to Feynman Diagrams in the Many-body Problem (New York:
McGraw-Hill).
MATTUCK, R. D., and STRANDBERG, M. W. P., 1960, Phys. Rev., 119, 1204-17.
MCCLURE, D. S., 1959, in Solid State Physics, Vol. 8 , Eds F. Seitz and D. Turnbull (New York:
Academic Press).
MCCOMBE, B., and SEIDEL, G., 1967, Phys. Rev., 155, 633-41.
MCCUMBER, D. E., 1964,J. math. Phys., 5, 221-30.
MENNE,T. J., 1968, Phys. Rev., 170, 356-8.
RIENNE,T. J., Ames, D. P., and LAC,S., 1968, Phys. Rev., 169, 333-9.
MERZ,J. L., and PERSHAN, P. S., 1967, Phys. Rev., 162, 217-35.
MIASEK, M., 1966, J . math. Phys., 7, 139-47.
MIGDAL, A. B., 1957, Sov. Phys.-JETP, 5, 333-4.
MORGAN, D. J., and GALLOWAY, J. A., 1967, Phys. Stat. Solidi, 22, 491-7.
MORIYA, T., 1960, Phys. Rev., 120, 91-8.
MORTIMER, F. S., 1969,J. Magn. Res., 1, 1-6.
MOSHINSKY, M., 1968, Group Theory and the Many-body Problem (Kew York: Gordon and
Breach).
MOSHINSKY, M., and DEVI,V. S., 1969, J. math Phys., 10, 455-66.
MUELLER, F. M., 1966, Phys. Rev., 148, 636-7.
MUELLER, F. M., 1967, Phys. Rev., 153, 659-69.
MUELLER, F. NI., and PRIESTLEY, 1cI. G., Phys. Rev., 148, 638-43.
NIURAO, T., et al., 1967, J , chem. Phys., 47, 1572-80.
RTURNAGHAN, F. D., 1963, The Theory of Group Representations (New York: Dover).
MURRELL, J. N.,1963, The Theory of the Electronic Spectra of Organic Molecules (London:
Methuen).
MUSHER, J. I., and SILBEY,R., 1968, Phys. Rev., 174, 94-103.
NAGLE, J. F., 1966 a,J. math. Phys., 7, 1588-92.
~ - A G L E ,J. F., 1966 b, Phys. Rev., 152, 190-7.
NAGLE,J. F., 1968, J. math. Phys., 9, 1007-19.
NAGLE,J. F., and TEMPERLEY, H . K.V., 1968, J . math. Phys., 9, 1020-6.
NEUMARK, G. F., 1964, Philips Res. Rep., 19, 349-58.
NEWELL, G. F., and MONTROLL, E. W., 1953, Rev. mod. Phys., 25, 353-89.
NEWMAN, D. J., and CURTIS,M. M., 1969, Chem. Phys. Lett., 3, 158-60.
NEWMAN, D. J., and STEDMAN, G. E., 1969,J. chem. Phys., 51, 3013-23.
NIELSON, C. W., and KOSTER,G. F., 1964, Spectroscopic Coefficients for pn, d" and f" Con-
figurations (Cambridge, M I T Press).
NUSIMOVICI, M., 1966, Phys. Stat. Solidi, 14, 77-9.
O'BRIEN,M. C. N., 1965, Proc. Phys. Soc., 86, 847-56.
OFELT,G. S., 1962, J . chem. Phys., 37, 511-20.
OHKUKI, Y., and KAMEFUCHI, S., 1969, Ann. Phys., N.Y., 51, 337-58.
OLBRYCHSKI, K., 1963 a, Phys. Stat. Solidi, 3, 1868-75.
OLBRYCHSKI, K., 1963 b, Phys. Stat. Solidi, 3, 2143-2154.
ONODERA, Y., 1968,J. Phys. Soc.Japan, 25,469-80.
OPECHOWSKI, W., 1940, Physica, 7, 552-62.
OPECHOWSKI, W., and GUCCIONE, R., 1965, in Magnetism, Vol. 2 (New York: Academic
Press).
OPIK,U., and PRYCE, RiI. H. L., 1957, Proc. R. Soc. A, 238, 425-47.
ORBACH, R., 1961, Proc. R. Soc. A, 264, 458-84.
ORE,O., 1963, Graphs and their Uses (New York: Random House).
OVERHAUSER, A. W., 1960, Phys. Rev. Lett., 4, 462-5.
OVERHAUSER, A. W., 1962, Phys. Rev., 128, 1437-52.
PARMENTER, R. H., 1955, Phys. Rev., 100, 573-9.
PATTERSON, E. M., 1956, Topology (Edinburgh: Oliver and Boyd).
PAULING, L., 1960, The Nature of the Chemical Bond (Ithaca, N.Y.: Cornel1 University Press).
PAUNCZ, R., 1967, Alternant Molecular Orbital Method (Philadelphia: W. B. Saunders).
PENN,D. R., and COHEN,M. H., 1967, Phys. Rev., 155, 468-77.
PHARISEAU, P., 1960, Physica, 26, 1185-91.
PHILLIPS, J. C., 1956, Phys. Rev., 104, 1263-77.
PHILLIPS, J , C.. 1959, Phys. Rev., 113, 147-55.
642 J. Killingbeck
PHILLIPS,J. C., and ROSENSTOCK, H. B., 1958,J. Phys. Chem. Solids, 5, 288-92.
PICARD, M.,and HULIN,M., 1967, Phys. Stat. Solidi, 23, 563-70.
PIPPARD, A. B., 1965, The Dynamics of Conduction Electrons (London: Blackie).
PIRENNE, J., and RENSON,P., 1962, Physica, 26, 233-50.
PLATT,J. R., et al., 1964, Free Electron Theory of Conjugated Molecules (Chicago: Wiley).
POLLAK, F. H., and CARDONA, M., 1966,J. Phys. Chem. Solids, 27, 423-5.
PRATHER, J. L., 1961, Atomic Energy Levels in Crystals, Nut. Bur. Stand. Monogr. No. 19.
PRYCE,M. H. L., 1950, Phys. Rev., 80, 1107-8.
RACAH,G., 1942, Phys. Rev., 62, 438-62.
RAE,A. D., 1969,J. chem. Phys., 50, 2672-85.
RAHMAN, H. U,, and RUNCIMAN, W. A., 1966,J. Phys. Chem. Solids, 27, 1833-35.
RAJNAK, K., and WYBOURNE, B. G., 1963, Phys. Rev., 132, 280-90.
RAY,D. K., RAY,T., and RUDRA,P., 1966, Proc. Phys. Soc., 87, 485-99.
RAYCHACDHURI, A. K., and RAY,D. K., 1967, Proc. Phys. Soc., 90, 839-46.
REED,W. A., and FAWCETT, E., 1965, in Proc. Int. Conf. on Magnetism, Nottingham (London:
T h e Institute of Physics and T h e Physical Society).
REHWALD, W., 1967, Phys. Rev., 155, 861-8.
RICHMAN, I., SATTEN, R. A., and WONG,E. Y., 1963, J . chem. Phys., 39, 1833-46.
RICHMOND, P., and SEWELL,G . L., 1968,J. math. Phys., 9, 349-56.
RIESS,J., 1970, Ann. Phys., N . Y., 57, 301-21.
ROSENSTOCK, H. B., 1960,J. Phys. Chem. Solids, 15, 50-3.
ROSENSTOCK, H. B., and MCGILL,R. E., 1962,J. math. Phys., 3, 200-2.
ROTENBERG, M., 1963, J. chem. Phys.. 39, 512-17.
ROTENBERG, M., BIVINS,R., METROPOLIS, h-.,and WOOTEN, J. K., 1959, The 3j and 6 j symbols
(Cambridge, Massachusett: M I T Press).
RUIJGROK, T. W., 1962, Physica, 28, 877-92.
SAATY, T. L., and BRAM,J., 1964, Nonlinear Mathematics (New York: McGraw-Hill).
SABISICY,E. S., and ANDERSON, C. H., 1966, Phys. Rev., 148, 194-7.
SABISKY, E. S., and ANDERSON, C. H., 1967, Phys. Rev., 159, 234-8.
SALZMAN, W. R., 1968,J. chem. Phys., 49, 3035-40.
SARFATT, J., and STONEHAM, A. M., 1967, Proc. Phys. Soc., 91, 214-21.
SASAKI, F., and OHNO,K., 1963,J. math. Phys., 4, 1140-7.
SATO,H., TOTH, R. S., and HONJO, G., 1967, J. Phys. Chem. Solids, 28, 137-60.
SCHAAK, G., 1963, 2. Phys., 176, 67-83.
SCHIFF,L. I., 1955, Quantum Mechanics (New York: 11cGraw-Hill).
SCHMID, D., 1966, Phys. Stat. Solidi, 18, 653-66.
SCHMIDTKE, H. H . , 1963, 2. Naturf., 18, 276-80.
SCHMIDTKE, H. H., 1966,J. chem. Phys., 45, 3920-8.
SCHMIDTKE, H. H., 1968 a,J. chem. Phys., 48, 970-1.
SCHMIDTKE, H. H., 1968 b, Theor. Chim. Acta, 9, 199-209.
SCHULZ, K., 1961, Z . Phys., 163, 293-308.
SCHULZ, K., 1962,Z. Phys., 166, 299-310.
SEGALL, B., 1961, Phys. Rev., 124, 1797-806.
SEITZ,F., 1936, Ann. Math., 37, 17-28.
SEITZ,F., 1940, The Modern Theory of Solids (New York: ILIcGraw-Hill).
SHARMA, C. S., 1968,J. Phys. B: Proc. Phys. Soc., 1, 1016-22.
SHARMA, C. S., DAS,T. P., and ORBACH, R., 1966, Phys. Rev., 149, 257-69.
SHARP, R. I., 1969, J . Phys. C: Solid S t . Phys., 2, 43243.
SHEPHERD, J. P. G., and GORDON, W. L., 1968, Phys. Rev., 169, 541-52.
SHRIVASTAVA, K. N., 1969, J . Phys. C: Solid St. Phys., 2, 777-84.
SHUBNIKOV, A. V., and BELOV,N. N., 1964, Colored Symmetry (Oxford: Pergamon).
SIEGWORTH, J. D., 1967, Phys. Rev., 155, 285-96.
SIMPSON, W. T., 1962, Theories of Electrons in Molecules (London: Prentice-Hall).
SINANOGLU, 0.)1961, Phys. Rev., 122, 493-9.
SIVARDI~RE, J., 1969, C. R. Acad. Sci., Paris, 268, 1174-6.
SLACK, G. A., ROBERTS, S., and HAM,F. S., 1967, Phys. Rev., 2, 170-7.
SLATER, J. C., 1964, in Advances in Quantum Chemistry, Vol. 1, ed. P. 0. LSwdin (Kew York:
Academic Press).
SLATER, J. C., 1965 a, Rev. mod. Phys., 37, 68-83.
Group theory and topology in solid state physics 643

SLATER,J. C., 1965 b, Quantum Theory of Molecules and Solids, Vol. 2 (New York: McGraw-
Hill).
SLATER,J. C., 1968,J. uppl. Phys., 39, 761-7.
SMITH,D.,and THORLEY, J. H. M., 1966, Proc. Phys. Soc., 89, 779-81.
SMITH,V. H., and HARRIS, F. E., 1969,J. math. Phys., 10,771-8.
SPAETH,J. M., 1966,Z. Phys., 192,107-41.
STARK,R.W., and TSUI, D. C., 1968,J. appl. Phys., 39, 1056-60.
STEINER,E., 1968, Molec. Phys., 15, 459-68.
STERNHEIMER, R. M., 1966, Phys. Rev., 146,140-60.
STERNHEIMER, R. M., BLUME,M., and PEIERLE, R. F., 1968, Phys. Rev., 173,376-89.
STEVENS,K. W. H., 1952, Proc. Phys. Soc. A, 65,209-15.
STEVENS,K.W. H., 1953 a, Rev. mod. Phys., 25, 166.
STEVENS,K. W. H., 1953 b, Proc. R. Soc. A, 218,542-55.
STEVENS,K.W. H., 1967, Rep. Prog. Phys., 30, 189-226.
STEVENS,K. W. H., 1969,J. Phys. C: Solid S t . Phys., 2, 193446.
STEVENSON, R.,1965, Multiplet Structure of Atoms and Molecules (Philadelphia: Saunders).
STONE,A. J., 1964,J. chem. Phys., 41, 1568-79.
STREITWIESER, A.,1961, Molecular Orbital Theory f o r Organic Chemists (New York: Wiley).
STUART,R., and MARSHALL, W., 1960, Phys. Rev., 120,353-7.
STURGE,M. D., 1965, Phys. Rev., 140,880-91.
STURGE,M.D., 1967, in Solid State Physics, Vol. 20, Eds F. Seitz and D. Turnbull (New York:
Academic Press).
SUGANO, S., and PETER,M., 1961, Phys. Rev., 122,381-6.
SUGAR,J., 1965,J. Opt. Soc. Am., 55, 1058-60.
SULLIVAN, J. L., 1968, J . math. Phys., 9, 1369-74.
SULLIVAN, P., and SEIDEL,G., 1967, Phys. Lett., 25A,229-30.
SYKES,M.F., and ESSAM, J. W., 1964, Phys. Rev., 133A,310-15.
SYKES,M.F., ESSAM, J. W., HEAP,B. R., and HILEY,B. J., 1966,J. math. Phys., 7, 1557-72.
SYNEK,M.,and CORSIGLIA, L., 1968,J. chem. Phys., 48,3121-5.
TACHIKI, M.,1968,J. Phys. Soc.Jupan, 25, 686-90.
TACHIKI, M., and TERAMOTO, K., 1966,J. Phys. Chem. Solids, 27, 335-48.
TAHIR-KHELI, R.A., CALLEN, H. B., and JARRETT, H., 1966,J. Phys. Chem. Solids, 27,23-32.
TAYLOR, P. L., 1966, Proc. Phys. Soc., 88, 753-6.
TERHAAR,D., 1961, Rep. Prog. Phys., 24,304-62.
THORNLEY, J. H. M., 1966, Proc. Phys. Soc., 88, 325-32.
TREES,R. E., 1964,J. Opt. Soc. Am., 54,651-7.
TSUJIKAWA, I., and KANDA, E., 1963, Proc. Phys. Soc.Japan, 18, 1382-90.
TURNER, J., 1968, J. Franklin Inst., 285, 52-8.
VANHOVE,L., 1953, Phys. Rev., 89, 1189-93.
VANNOTTI, L. E., and MORTON, J. R., 1968, Phys. Rev., 174,448-53.
VANSTAPELE, R. P., BELJERS, H. G., BONGERS, P. F., and ZIJLSTRA, H., 1966,J. chem. Phys., 44,
3719-25.
VANVLECK,J. H., 1960, Physica, 26,544-52.
VANWAGENINGEN, R., 1964, Nucl. Phys., 60,250-63.
VONDER LAGE,F. C., and BETHE,H. A., 1947, Proc. Phys. Soc. A, 65,209-19.
DEVRIES,G., BREED, D. J., MAARSCHALL, E. P., and MIEDEMA, A. R.J. uppl. Phys., 39,1207-8.
WACHTMAN, J. B., and PEISER,H. S., 1966,J. Phys. Chem. Solids, 27,975-82.
WANNIER, G.H., 1937, Phys. Rev., 52, 191-7.
WANNIER, G.H., 1960, Elements of Solid State Theory (Cambridge: Cambridge University
Press).
WANNIER, G. H., 1966, Statistical Physics (New York: Wiley).
WATANABE, H., 1966, Operator Methods in Ligand Field Theory (New Jersey: Prentice-Hall).
WATSON, R. E., and FREEMAN, A. J., 1960, Phys. Rev., 120,1125-34.
WATSON, R. E., and FREEMAN, A. J., 1964, Phys. Rev., 133,1571-84.
WATSON, R.E., and FREEMAN, A. J., 1967, Phys. Rev., 156,251-8.
WEBSTER, A. G., 1955, Partial Differential Equations of Mathematical Physics (New York:
Dover).
WEINREICH, G., 1965, Solids. Elementary Theoryf o r Advanced Students (New York: Wiley).
WHEELER, R. G., REAMES, F. M., and WACHTEL, E. J., 1968,J. appl. Phys., 39, 915-22.
644 j? Killingbeck
WIGNER,E. P., 1959, Group Theory and its Application to the Quantum Mechanics of Atomic
Spectra, translated by J. J. Griffin (New York: Academic Press).
WILLIAMS, F. I. B., 1967, Proc. Phys. Soc., 91, 111-23.
WINSTON,H., 1951,J. chem. Phys., 19, 156-60.
WINTNER, A., 1948,J. chem. Phys., 16, 405-6.
WOLFF,P. A., 1966, J. Phys. Chem. Solids, 27, 685-700.
WOLL,E. J., and KOHN,W., 1962, Phys. Rev., 126, 1693-7.
WONG,E. Y., 1963,J. chem. Phys., 38, 976-8.
WOODBURY, G. W., 1969,J. chem. Phys., 50, 2247-54.
WYBOURNE, B. G., 1965, Spectroscopic Properties of Rare Earths (New York: Wiley).
WYBOURNE, B. G., 1966, Phys. Rev., 148, 317-27.
WYBOURNE, B. G., 1968,J. chem. Phys., 48, 2596-611.
YAMAMOTO, H., MAKISHIMA, S., and SHIONOYA, S., 1967, J. Phys. Soc.Japan, 23, 1321-32.
YIN, M. L., and SILVERSTONE, H. J., 1968, J. chem. Phys., 49, 3160-4.
ZACHARIASEN, W. H., 1967, Theory of X-ray diffractionin Crystals (New York: Dover).
ZAHNER, J. C., and DRICKAMER, H. G., 1961,J. chem. Phys., 35, 1483-90.
ZAK,J., 1960,J. math. Phys., 1, 165-71.
ZDANSKY, K., 1967, Phys. Rev., 159, 201-8.
ZEH,H. D., 1965,Z. Phys., 188, 361-73.
ZELLER,H. R., and K ~ V Z I GW.,
, 1967, Helv. phys. Acta, 40, 845-72.
ZIMAN,J. M., 1963, Electrons in Metals (London: Taylor and Francis).
ZIMAN,J. M., 1965, Principles of the Theory of Solids (Cambridge: Cambridge University
Press).
ZIMAN,J. M., 1966, Proc. Phys. Soc., 88, 387-405.
ZIMAN,J. M., 1969, Elements of Advanced Quantum Theory (Cambridge: Cambridge University
Press).

Das könnte Ihnen auch gefallen