Chapter 1

Periodic systems and the Bloch

Theorem
1.1 Introduction
We are interested in solving for the eigenvalues and eigenfunctions of the
Hamiltonian of a crystal. This is a one-electron Hamiltonian which has the
periodicity of the lattice.
H =
p
2
2m
+V (r). (1.1)
If R is a translation vector of the lattice, then V (r) = V (r + R). To
solve for such a problem, one must rst consider the boundary conditions
that the eigenfunctions must satisfy. In a crystal, we require them to be
periodic of period N
i
R
i
; (i = 1, 2, 3) for large N
i
. Typically, N
i
is of the
order of the size of the crystal: N
1
N
2
N
3
10
23
! A good basis to describe
such functions are plane waves exp (ik.r). Applying the boundary conditions
to the plane waves implies that the momenta k must be discrete points:
k
i
= 2n
i
/N
i
R
i
; i = 1, 2, 3 where n
i
are integers. Practically, for large N
i
,
they form a continuous set of points in the reciprocal space. In the following,
we will rst introduce and prove Blochs theorem which is based on the
translational invariance of the system; then make use of it to simplify the
dicult task of nding the eigenstates and eigenvalues of the innite crystal
Hamiltonian.
1
2
1.2 Bloch Theorem
Let T
R
be the translation operator of vector R. T
R
commutes with the
Hamiltonian. Indeed, the kinetic energy is translationally invariant, and the
potential energy is periodic:
[T
R
, V ]f(r) = T
R
V (r)f(r)V (r)T
R
f(r) = V (r+R)f(r+R)V (r)f(r+R) = 0
(1.2)
On the other hand, [T
R
, T
R
] = 0. Thus, the Hamiltonian and all the transla-
tion operators of the crystal commute with each other. They possess, there-
fore, a common set of eigenstates.
Let us search then the eigenstates of the translation operators T
R
. For a
general function satisfying the boundary conditions of the problem, one can
write, after expanding the eigenstate on a plane wave basis, the eigenvalue
equation as follows:
T
R
f(r) = T
R

q
C
q
e
iq.r
=

q
C
q
e
iq.r
e
iq.R
= t
R

q
C
q
e
iq.r
(1.3)
where we denoted the eigenvalue by t
R
. In order for the above equality to be
true, e
iq.R
must be a constant: q.R = 2n + constant q = k +G where
k is an arbitrary vector and G is a reciprocal lattice vector: G.R = 2n.
The eigenvalue is therefore t
R
= e
ik.R
and the eigenvector could be any plane
wave of momentum k +G. The eigenvalue t
R
being degenerate with respect
to k +G, a general eigenvector associated with this eigenvalue can be written
as:
f
k
(r) =

G
C
k+G
e
i(k+G).r
= e
ik.r

G
C
k+G
e
iG.r
= e
ik.r
u
k
(r) (1.4)
The arbitrary vector k labels thus dierent eigenvalues and eigenstates of
translation operators. Furthermore, the above function is also eigenstate of
all possible translation operators of the lattice. Note that the innite sum
over all possible momenta q is now reduced to a discrete (still innite) sum
over the reciprocal lattice vectors, and that is a great simplication in the
problem. We can also notice that all states corresponding to k and any
k +G are equal, i.e. the function f
k
is periodic in the reciprocal space
f
k
= f
k+G
, G. Due to the previously-mentioned commutation relation,
f
k
is also eigenstate of H. To obtain its coecients C
k+G
, we just need to
insert this wavefunction into the Schroedinger equation. The Hamiltonian
3
matrix must therefore be diagonalized in the space of all plane waves of
momentum k +G for any vector k chosen in the rst Brillouin zone. Indeed
the periodicity of f
k
in the reciprocal space implies that it is sucient to
chose the arbitrary vector k in the rst Brillouin zone.
Let us nally state Blochs theorem: The eigenstates f
k
of a peri-
odic Hamiltonian can be written as a product of a periodic function with
a plane wave of momentum k restricted to be in the rst Brillouin zone
f
k
(r) = u
k
(r)e
ik.r
(with u
k
periodic in k and in r) ; furthermore f
k
(r +R) =
e
ik.R
f
k
(r).
Solution of the eigenvalue problem in general
The Schroedinger equation or the eigenvalue problem is written in the
following form: H = E. Dierent Boundary Conditions (BC) apply for
of a crystal and a cluster. The basis set must therefore satisfy the same
boundary conditions. This means that for an innite crystal, the basis must
satisfy periodic BCs, whereas for a molecule or a cluster, basis functions must
go to zero at innity. Let
i
(r)
i=1,...,N
be a nite basis set (hence incomplete
description), and let us expand the eigenfunction on this basis: (r) =

N
i=1
C
i

i
(r). We have already seen that minimizing the expectation value
of the Hamiltonian with the constraint of orthonormality (the variational
formulation) is equivalent to inserting the expansion of the wavefunction into
the Schroedinger equation and projecting it onto each of the basis functions.
Both lead to the following set of linear equations:
N

j=1
[H
ij
ES
ij
] C
j
= 0; i = 1, ..., N (1.5)
where H
ij
=
_

i
(r) H
j
(r) dr and S
ij
=
_

i
(r)
j
(r) dr are the Hamilto-
nian an overlap matrix elements respectively.
Solution of the Schroedinger equation in a periodic
box
[
h
2
2m
+V (r)](r) = E(r)
The crystal potential V (r) is a periodic function: V (r +R) = V (r)
4
We will therefore make use of the Fourier decomposition to describe it as:
V (r) =

V
G
e
iGr
where

V
G
=
_
cell
V (r) e
iGr
dr/; Gs are reciprocal lattice vectors dened
by G R = 2n, Rs are translation vectors of the crystal lattice and is
the unit cell volume. If {R
1
, R
2
, R
3
} is the basis of the primitive cell, then
the reciprocal lattice vectors are dened by:
G
i
R
j
= 2
ij
, or more explicitly: G
1
= 2(R
2
R
3
)/ with = R
1

(R
2
R
3
) and so on ... We see that the Gs appear naturally because of the
periodicity of the potential.
The BCs satised by the wavefunctions come from Blochs theorem:
For every eigenfunction , there exists a vector k such that :
k
(r + R) =
e
ikR

k
(r); in other words
k
(r) = e
ikr
u
k
(r) where u
k
is a periodic function.
We see therefore that is not periodic! To solve the eigenvalue problem, one
may take any basis set satisfying the above theorem.
Plane wave method:
One simple and easy choice would be the plane waves:
k
(r) = e
ikr
/

which are eigenstates when V (r) = 0. The eigenvalue equation in this basis
may be written as:

q
[H
kq
E S
kq
] C
q
= 0; k
But S
kq
=
k,q
and H
kq
=
k,q
k
2
/2m +

V
kq
. Therefore, k q = G
must be a reciprocal lattice vector. So the overlap S is the identity and the
Hamiltonian H couples any arbitrary vector k only to some k + G. The
problem of having to diagonalize an innite matrix H has been highly
simplied to the diagonalization of many smaller matrices (one for each k)
of nite dimension (dened by the number of Gs included in the Fourier
expansion of G). The nal solutions for each k are therefore (this is another
statement of Blochs theorem):

k
(r) =

G
C
k+G
e
ik+Gr
/

for a Hamiltonian H
k+G, k+G
= H
GG
(k) =
GG
(k+G)
2
/2m+

V
GG
. Note
that
k
has the additional property of being periodic in the reciprocal space:

k
(r) =
k+G
(r).
5
The number of obtained eigenvalues is equal to the number of chosen
basis functions. The eigenvalues and eigenstates for each k will also be
labeled by an integer which is called the band index ( = 1 corresponds to
the rst band, = 2 represents the second band and so on...). Hence, the
eigenfunctions and eigenvalues will be denoted respectively by
k
and E
k
.
The form of the Schroedingers equation in the plane-wave basis is:
_
(k +G)
2
/2m+

V
0
E
k+G
_
C
k+G
+

=G

V
GG
C
k+G
= 0 (1.6)
or in terms of dierent bands (corresponding to dierent G vectors)
1
.
_
(k +G)
2
/2m+

V
0
E
k,
_
C

k+G
+

=G

V
GG
C

k+G
= 0 (1.7)
In a plane wave calculation, the Fourier expansion has to be truncated at
some G
max
corresponding to a cuto energy of E
cuto
= h
2
G
2
max
/2m. This
basically denes the smallest wavelength one uses to describe the potential
and the charge density. It only depends on the nature of the elements one is
treating and not on the unit-cell size. However, the number of plane waves
(Gs) used in a calculation relate the two through: N
PW
G
3
max
E
3/2
cuto
.
The cuto energy should be chosen such that the total energy of the system
has converged as a function of the latter. To achieve such convergence, rather
high cuto energies are required. In practice, however, usually dierences in
total energies are of interest. In this case, smaller cutos will be sucient
since systematic errors due to lack of convergence will cancel each other out.
Periodicity, broken symmetry and gap openings:
The eect ofperiodicity on the band structure can be investigated by solving
exactly solvable models such as the Kroenig-Penney model in one dimension
2
Here, however, we will use the perturbation approach, where analytical
formulas are available for the eigenstates and eigenvalues. Assuming the
potential is weak, one can expand the eigenvalues and eigenstates in powers of
1
In the absence of a periodic potential, there is a one to one correspondance between a
band labeled by and a reciprocal lattice vector G, so that E
k,
= E
k+G
and C

k+G
= 1
and C

k+G
= 0 for all other G

. In the presence of a weak periodic potential, a perturbative

argument shows that these coecients slightly vary from 1 and 0
2
the potential is either that of a superlattice i.e. constant with dierent values in 2
chosen intervals, the sum of which is the period of the lattice; or it can be a comb
function (sum of Dirac delta functions).
6
the potential.To rst order in the eigenstates and second order in eigenvalues
we have:

k
= e
i(k+G).r
+

=G
e
i(k+G

).r

V
GG

E
k+G
E
k+G

(1.8)
E
k
=
(k +G)
2
2m
+

V
0
+

=G
|

V
GG
|
2
E
k+G
E
k+G

(1.9)
this expansion is valid whenever

V
GG
|E
k+G
E
k+G
|. We can there-
fore note that near the zone boundaries, where there is degeneragy, the de-
nominator can become zero, and the expansion breaks down. In this case,
one will keep in the expansion only the terms that have small denominator,
the contribution of others being small in the

V 0 limit, as they are in

V
2
. The divergence in the denominators happens for the Bragg condition
E
k+G
= E
k+G
implying either k = 0 and |G| = |G

|, or, k = G
1
/2 and
G = 0; G

= G
1
where G
1
is the smallest vector of the reciprocal lattice.
Here, for simplicity, we will restrict ourselves to the one-dimensional case
where we only have two-fold degeneracies and diagonalizations can be done
analytically. The Hamiltonian matrix in the case where there is only two-fold
degeneragy (or near degeneracy) is:
_
E
k+G

V
GG

GG
E
k+G

_
=>

=
E
k+G
+E
k+G

2

_

2
+|

V
GG
|
2
with = (E
k+G
E
k+G
)/2. At the exact Bragg condition where =
0, we have splitting of the two eigenvalues by the perturbation:

=
E
k+G
|

V
GG
|. So, the two degenerate bands become separated by a
gap given by 2|

V
GG
|. As we previously said, this can happen at the zone
center or boundary where the group velocity
k
E is also equal to zero.
The eigenstates are then either the sum or the dierence of the two plane
waves

= e
i(k+G).r
e
i(k+G

).r
which are reduced to a sin and cos function.
Near the zone boundary, however, where there is near degenearcy, the term
is small but nonzero. Qualitatively, the level repulsion still persists and
the gap is even larger than at the zone boundary. It can be veried that the
eigenstates become the product of the sin or cos functions times an enve-
lope function of large wavelength. This wavelength being the inverse of the
distance of the actual wavenumber from the zone boundary = k G
1
/2.
Note that for the higher bands, the dierence of the G vectors becomes
larger, and since usually

V
G
is a decreasing function of G, the gaps become
7
increasingly small for higher energy bands. The dierence between a metal
and a semiconductor is in the position of the Fermi level. If the latter lies
within a band, one is dealing with a metal, and if it is in the gap (fully lled
bands), one is dealing with a semiconductor.
Tight-binding or LCAO method:
In cases where some electronic states are very localized, it would take a large
number of plane waves to describe them. Examples are core electrons in
all-electron calculations, and d and f orbitals in transition metals and heavy
atoms. In this case, one might prefer to expand the wavefunctions on local-
ized, atomic-like basis functions. This method is called Tight-Binding (TB)
or LCAO (standing for Linear Combination of Atomic Orbitals). Sometimes,
TB also refers to semi-empirical methods in which the hopping matrix ele-
ments between neighboring atoms are determined from a t to ab-initio or
experimental data. Such a choice of basis also exists in mixed-basis methods,
where the basis consists of both plane waves, to describe extended elecron
states, and atomic-like orbitals to describe localized electron states. In a
periodic system, the proper form of the basis functions satisfying Blochs
theorem should be:

k
(r) =

R
b

(r R) e
ikR
/

N
where the function b

(r R ) is an atomic-like orbital localized on the

atomic position labeled within the unit cell dened by R, and is the
orbital index (s, p
x
, p
y
, p
z
...).
Note that this expression is manifestly G-periodic (i.e. periodic in the
reciprocal space). This change of basis from (R, , ) representation to the
Bloch basis: (k, , ) is just a unitary transformation which makes use
of the Bloch theorem (

k
(r + R) =

k
(r) e
ikR
) to make the matrix of
Hamiltonian, as we will show below, block-diagonal, meaning that:
<

k
|H|

k
>
k,k

+G
which is a great simplication of the diagonalization problem. The size of
each bloch for a sampling k-point in the FBZ, is equal to the total number
of orbitals present in the unit cell: (

1=size of the Hamiltonian matrix in

the Bloch basis).
8
<

k
|H|

k
> =

RR

1
N
e
ikR
e
ik

< b

(r R)|H|b

(r R

) >
=

RR

1
N
e
ikR
e
ik

f
,

(RR

)
=

R
1
N
e
i(kk

)R

R
e
ik

f
,

(R

)
=
k,k

+G
[

e
ik

f
,

(R

)] (1.10)
showing the decoupling of dierent k vectors within the rst Brillouin zone.
Use of translational symmetry was made going from the rst to the second
line.
Note that the sum over R

involves only a small number of terms for

which the hopping integrals f
,

(R

) are nonzero. A similar expression

holds for the overlap matrix, and one can, for each chosen k-point in the rst
Brillouin zone, solve the generalized eigenvalue problem with a matrix of size
much smaller than in the plane wave method.
As will be shown in the exercises on the linear chain, the bandwidth
wihin the TB formalism is proportional to the hopping integrals, and N-fold
degenerate atomic levels fan into a continuum of width 4t as the interatomic
distance becomes smaller and the hopping integral gets larger.
Figure 1.1: Energy levels of the system as at the atoms get closer to each
other, and the hopping integral increases. The bandwidth essentially scales
with the latter.
First Brillouin Zone and general considerations:
The LCAO Hamiltonian for k and any k + G however, has the same ma-
trix elements. This is similar to the plane wave method, where both the
Hamiltonian and its eigenstates were shown to be periodic in k. Due to this
periodicity in the reciprocal space, there is no need to treat all ks; one only
needs to treat the ks inside the First Brillouin Zone (FBZ) which is the
Wigner-Seitz cell of the reciprocal space. In practice, for numerical calcula-
tions, involving integrations over the k-points in the FBZ, one has to choose
a grid of k-points. Because of the symmetries of the cell, if any, one needs to
9
consider only the k-points in some portion of the FBZ. Furthermore, some
special k-points may be used to increase the computational eciency. For
more details about this topic, one may refer to the following papers:
M. J. Mehl, et al., Phys. Rev. B 41, 10311 (1990);
H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976).
or consult the following homepage to download k-point sets:
http://dave.nrl.navy.mil/bind/kpts/index.html
Using the power of Blochs theorem, one has been able to reduce the
problem of diagonalizing an innite Hamiltonian matrix, to diagonalizing
a nite matrix (treating the atoms of the unit cell only) but for all the k-
points in the FBZ. For metals or systems without a gap, the mesh should be
a ne one, but for semi-conductors or insulators where the gap is non-zero,
one may take a coarser k-point mesh. In any case, checks must be performed
to make sure of the convergence of the results as a function of the number of
k-points taken in the FBZ.
Band Structure and Density Of States (DOS)
In an electronic structure calculation based on the density-functional or
Hartree-Fock theory, once the self-consistency iterations have converged, one
obtains the ground state charge density and the crystal potential. Using
the eigenvalues at the k-points taken in the FBZ, one may also extract the
density of states (DOS). After this calculation, one may also obtain the band
structure curves by chosing another set of k-points taken along the symmetry
directions in the FBZ, and calculating the eigenvalues for these k-points by
using the converged and self-consistent charge density and crystal potential.
The DOS is dened (including the spin degeneracy) as
DOS(E) = 2

k
w
k
(E E
k
)
If a ne k-grid is chosen, one can replace the delta function by a broadened
Gaussian and perform the sum numerically. The broadening must be large
enough or k-grid ne enough so that the delta contribution of each grid point
does not appear as a spike in the DOS; and it should not be too large as to
wash out the ner details of the DOS curve. The weighting factor w
k
are there
10
rst to assure proper normalization
3
The best method for calculating the
crystal DOS, is the tetrahedron method. In this method, one has calculated
the eigenvalues at some k-points in the FBZ. It is possible to connect the
k-points nearest to each other to obtain small tetrahedra (in 3D) or triangles
(in 2D). One then assumes for each band a linear E(k) in this region, and
integrates over the tetrahedron volume analytically:
DOS(E)

_
dS
k
|
k
E
k
|
For the integration over the full volume of the FBZ, one also needs the energy
eigenvalues at the zone boundaries and symmetry directions. These special
points would come into the calculation with a dierent weight. Likewise, the
band energy and the ground state charge density are a weighted sum of the
eigenvalues and square of the eigenvectors respectively, over the FBZ:
E
band
=

k
w
k
E
k
; (r) =
occupied

k
w
k
|
k
(r)|
2
E
band
is the total energy for a non interacting system. If interactions are
present, even within a mean-eld theory, a double counting correction term
must be subtracted from this energy.
The general owchart of a band structure calculation is described below:
3
typically for a k-grid not following any symmetry, w
k
= 1/N
k
so that the sum is not
proportional to the number of chosen k-points, and is properly normalized to the number
of bands times number of spin states. In case there is a symmetry in the k-grid, dierent
k-points might have dierent weights.
11
Initialization: Atomic Positions and Parameters,
Unit Cell: R and G Vectors, Special k-points
?
Assume Initial Charge Density
?
Calculate Crystal Potential
?
Do loop over k-points in the FBZ
?
Calculate Matrix Elements of H(k) and S(k)
?
Diagonalize to get eival E
k
and eivec
k
?
-
Calculate New Charge Density (r) (by Mixing or Broyden)
?

P
P
P
P
P
P
P
P
P
P
P
P
P
P

, E
k
converged?
NO
-
YES
?
Store Charge and Crystal Potential
?
Do Loop over k-points for Band Structure
?
Calculate Matrix Elements of H and S using the self-consistent potential
?
Diagonalize to get eival E
k
store for Band Structure
-