C2: Band Structure
C2: Band Structure
C2: Band Structure
December 2005
1 Introduction
When you buy a diamond for your girl/boy friend, what you pay for is a band gap of ∼6 eV.
Band structure is one of the most important concepts in solid state physics. Actually,
any excitation in a crystalline solid can be characterized by a Bloch vector k, but usu-
ally bandstructure refers to the structure of the electronic energy levels. The electronic
structure determines the chemical bonds which keep the atoms together. A material is
a conductor or insulator depending on how the Fermi energy is located relative to the
band gaps. The electronic structure indirectly also determines the vibrational levels i.e.
the phonons. Optical and magnetic properties depend crucially on the underlying band
structure, but there properties also depend more directly on how the electrons interact
with one another.
In its most naive form, band structure theory assumes that each particle moves in an
average potential V (r) from all the nuclei and electrons in the system, so-called mean-
field theory. In a crystalline material this potential is lattice periodic, and the electron
orbitals can be chosen to have Bloch symmetry involving a Bloch label k and a band
index n, φ(r) = φkn (r). However, the band structure concept remains valid also when the
electrons interact with one another, and the energy bands are experimentally measurable
with experiments such as angle-resolved photoemission. In the real band structure, the
electron states have a finite lifetime away from the Fermi energy, and the band energies
ǫkn may differ from those obtained from simple mean-field approximations.
In order to determine ground-state properties, the mean-field approach can be made
formally exact by adding terms to the effective potential which account for the interac-
tion and Pauli principle, so-called density-functional theory (DFT). In this form, band-
structure theory has been very successful for predicting ground-state properties such as
cohesive energies, crystallographic structure and chemical bond lengths and lattice param-
eters with e2 and ~ as the only input parameters. In the adiabatic approximation, the
electrons are considered to remain in the instantaneous (electronic) ground state when the
lattice vibrates. Consequently, also adiabatic phonon frequencies can be obtained from
DFT.
1
In order to predict correct excitation energies beyond mean-field theory, the poten-
tial must include terms which account for the interaction with all other electrons in a
dynamical way. Consequently, excitation energies are more difficult to predict than are
the ground-state properties mensioned above. Approximations from so-called many-body
perturbation theory have in recent years been remarkedly successful for predicting correct
excitation energies and band gaps in sp-bonded materials like Si, GaAs, etc. There is a
large number of band-structure codes available which solve the band-structure problem
within a density-functional framework relevant to ground-state properties. Also codes
for excited-state properties are becoming available. Many of these codes are results of
open-source efforts in the scientific community and are available free of charge.
In this excercise, we will illustrate the bandstructure concept by doing simulations of
model systems in one and two dimensions. We use two Matlab scripts named band.m
and lab.m. In the first program we look at how band structure emerge in one dimension,
and how it depends on the strength of the potential and the separation between the ions.
Moreover, we will discuss the reduced band scheme and the meaning of the reciprocal
lattice vectors. In the second program a two-dimensional model is considered, and we will
look at the importance of the potential strength and how bands may overlap in energy.
3 Construction of eigenstates
Even a simple one-dimensional Schrödinger equation cannot usually be solved in closed
form by some formulae, so one needs almost always to use some approximate method
(grid methods, basis sets, etc.) For solving electronic eigenstates in solids and molecules,
methods using basis sets and the variational theorem are dominating. Thus, the eigenstates
are found by minimizing
hψ|H|ψi
E=
hψ|ψi
where the trial wave functions ψ are chosen as linear combinations of some given basis
functions φ1 , φ2 , . . . , φn , X
ψ(r) = cν φν (r).
ν
This gives P ∗
c cl Hkl
E = Pkl k∗ (3.1)
kl ck cl Skl
where Z Z
Hkl = φ∗k (r)Hφl (r)d3 r, Skl = φ∗k (r)φl (r)d3 r, (3.2)
2
Question 1: Derive Eqs. (3.1, 3.2).
is a function only of n − m. With only one orbital per atom, there is no variational degree
of freedom left in the Bloch sum in Eq. (3.3), and the energy is just given by
hψk |H|ψk i Hk
ǫk = = (3.4)
hψk |ψk i Sk
where
1 X −iak(n−m) X
Hk = e Hn−m = e−iakn Hn−0 , (3.5)
N nm n
1 X −iak(n−m) X
Sk = e Sn−m = e−iakn Sn−0 . (3.6)
N nm n
3
Question 2: derive Eqs(3.4 - 3.6).
In order to increase the variational freedom and thus the accuracy, we may use several
local orbitals per atom, φ1 (x), φ2 (x), ..., φr (x), and form the Bloch sums
N −1 r
1 XX
ψk (x) = √ ckν eikna φν (x − na).
N n=0 ν=1
In this case, there are n free parameters ck,1 , ck,2 , . . . , ck,n to optimize for each k. The
variational calculation leads in this case to a n × n matrix eigenvalue problem to solve
for each k. This is how the matlab script in the exercise obtains the one-dimensional
band states. The potential is modeled by an attractive Gaussian potential well, and the
basis functions of the form xν exp(−αx2 ). By increasing the size of the basis, we obtain
gradually a more and more accurate solution to the lattice Schrödinger equation.
where G are reciprocal lattice vectors. As basis functions we take plane waves,
1
φk+K = √ ei(k+K)·r (3.7)
Vc
and the the variational Ansatz takes of the form
1 X
ψ(r) = √ ck+K ei(k+K)·r ,
Vc K
where k is the Bloch vector and K reciprocal lattice vectors. In the plane wave basis the
Hamiltonian matrix becomes
~2 (k + K)2
Hk+K,k+K′ = + VK−K′
2m
and the overlap matrix is unity. The secular problem becomes
µ 2
~ (k + K)2
¶ X
− ǫ ck+K + ck+K−G VG = 0
2m
G
4
εkn
Figure 1: Band structure in 1D with (solid line) and without (dashed line) a weak lattice potential
Generally, the second-order contribution is small, but near a zone boundary where |k +
K + G| = |k + K| the basis functions φk+K+G and φk+K become degenerate and non-
degenerate perturbation theory breaks down. To describe the band structure near the
degeneracy point we make a simple variational calculation with the two nearly degenerate
states above as basis functions. This gives a Hamiltonian matrix
µ 0 ¶
ǫk+K V−G
VG ǫ0k+K+G
with energy eigenvalues
à !1/2
ǫ0 + ǫ0k+K+G (ǫ0k+K − ǫ0k+K+G )2
ǫ = k+K ± + |VG |2
2 4
5
We see that a band gap of 2|VG | opens up, and this occurs in general for each G, i.e. at
every zone boundary.
Question 3: How many electron states (including spin) are there in each band if the
lattice consists of N sites?
Question 4: Consider a hypothetical 1D solid with two electrons per atom. Is the material
metallic or is it an insulator?
4 Simulations
4.1 Simulations in 1D
In this part you will use the program band.m. The use of the program is quite easy. Try
for yourself to figure out how it works and then answer the following question.
Question 5: What is the blue line? The red lines? And the yellow line?
Question 6: What happens when you enlarge the lattice constant? And when it is
diminished?
Look at the form of the single ion potential. It is chosen to be Gaussian.
Question 8: What happens with the band structure? Does it look like something you
know?
Reset the program again, and then lower the potential to its minimum value. The potential
is almost flat and we see a band scheme as for a free electron. Now increase the potential
strength and try to describe what happens to the bands.
Question 9: What is the difference between the lowest band and the higher ones? Hint:
Look at the energy of the bands and compare it to the potential. Try also to plot more
bands.
Question 10: Reset the program once more, and then increase the lattice constant. What
do you see? Plot more bands and see what happens.
Finally we look at Band structure for different lattice constants for a fixed potential.
Change the lattice constant and describe what you see. After having done that for a few
different lattice constants, press All lattice constants. Do the same for different potentials.
This will summarize what you have already done.
6
4.2 Simulations in 2D
These simulations intend to illustrate among other things then different bands may overlap
in energy as soon as the dimension is higher then one. The simulations are making use
of plane waves as basis functions for solving the Schrödinger equation. We consider a
square lattice in 2D with lattice parameter a. In Fig. 2 we show the 2D Brillouin zone
and the lowest free-electron bands. The free-electron bands are obtained by folding in the
free-electron parabola in the first Brillouin. The lowest ones are
~2 k 2
ǫ00 (k) = ,
2m
~2 (kx ± G1 )2 ~2 ky2 ~2 kx2 ~2 (ky ± G1 )2
ǫ01,2 (k) = + , ǫ03,4 (k) = + ,
2m 2m 2m 2m
~2 (kx ± G1 )2 ~2 (ky ± G1 )2
ǫ05,6,7,8 (k) = + ,
2m 2m
where G1 = 2π/a.
Γ X Y Γ
Figure 2: Left: Brillouin zone of a square lattice with special points Γ = (0, 0), X = (π/a)(1, 0)
and Y = (π/a)(1, 1). Right: Free-electron bands along the path Γ → X → Y → Γ.
To compute the 2D bandstructure when a lattice potential is added, we use the Matlab
script lab.m. The script has to be called with two arguments
lab(U , µ)
Question 11: Try to call it with (U, µ) = (0, 0). What do you see?
Now increase U and see what happens (try to rotate the figure). Increase the Fermi level
afterwards.
7
Question 12: What do you see on the figure and what does it represent?
Question 13: From the output, sketch how a weak lattice potential modifies the free-
electron bandstructure in Fig. 2
Question 14: Can a divalent ‘solid’ in 2D be metallic? What about divalent solids’ in
1D?
For a fixed potential vary the Fermi level.
Question 16: In the metallic case, sketch the shape of the Fermi surface of the lowest
band when it start to touch the zone boundaries. Is the effective mass positive (electron-
like) or negative (hole-like)?
Question 17: When the second band starts to fill: Sketch the shape of the Fermi surface
of the second band. Is the effective mass electron-like (> 0) or hole-like?
5 Final remarks
As stated in the beginning a diamond has a band gap of ∼ 6 eV. Do you understand why
this makes it expensive? Hint: Try to find the energy of a photon in the visible range.
Another question: Are you aware of why a solid with a band gap cannot conduct
current? See Ibach and Lüth.
6 References
Ibach, H. and Lüth, H.,Solid State Physics, Springer 2003.
Ashcroft, N.W. and Mermin, N.D., Solid State Physics, Saunders College Publishing 1976.