Quantum Theory of Matter 1

Tutorial 1
Bloch Theorem, Tight-Binding Model and Van-Hove Singularities

The Technion, Faculty of Physics

Spring Term 2024

1 Introduction
In this tutorial, we will recap a few of the concepts we’ve met in the under-grad. level solid state course,
which will be on some topics in our course the foundations.

1.1 Bloch’s Theorem

One of the essential concepts we’ve met in solid state physics is the constraint on the electronic wave-
function due to the presence of a periodic potential, V (r), as naturally appears in periodic structures.
The Schödinger equation reads (ℏ = 1)
1 2
− ∇ ψ(r) + V (r)ψ(r) = Eψ(r) ,
2m
where V (r) = V (r + R) is a periodic potential, and R = n1 a1 + n2 a2 + n3 a3 is any of the lattice vectors.
Note that here ni are integers and ai are the primitive vectors of the lattice.
This statement of periodicity can also be written using translation operators

T̂R = exp(−ip̂ · R)
† −1
Where p̂ is the momentum operator, the generator of translations. T̂R is unitary, as T̂R = T̂R and its
action on a wavefunction ψ(r) is

T̂R ψ(r) = ψ(r − R) .

Since discrete translations are a symmetry of the system, we have that the translation operators commute
with the Hamiltonian. They also commute with different translation operators. That is, ∀R, R′ ∈
{Lattice} we have
h i h i
H, T̂R = 0 T̂R , T̂R′ = 0 ,

hence we can choose the energy eigenstates to be simultaneous eigenstates of the translations. Be ψ(r)
an eigenfunction of both H and T̂r with eigenvalues ε, c(R) correspondingly. Due to the additivity of
the translation group

T̂R T̂R′ = T̂R+R′

c(R + R′ ) = c(R) · c(R′ ) .

Since T̂R is unitary, |c(R)| = 1, so we can write c(R) = exp(iφ(R)). We can consider R = ai to have
c(ai ) = exp(i2πxi ), thus

c(R) = c(n1 a1 + n2 a2 + n3 a3 )
= c(n1 a1 )c(n2 a2 )c(n3 a3 )
= cn1 (a1 )cn2 (a2 )cn3 (a3 )
= exp(2πin1 x1 ) exp(2πin2 x2 ) exp(2πin3 x3 )

1
 = exp(iR · k) ,

where ki = xi bi , for bi a reciprocal lattice vector (obeys ai · bj = 2πδi,j ). It means that k is a good
quantum number (associated with T̂R ), called ”crystal momentum” or ”quasimomentum”. This is
not the real momentum of the electrons, as it comes from the discrete translation operator. A
priori, each k sector can be degenerate, namely to have more than one eigenenergy per crystal momentum
value, distinguished by the index n (called ”the band index”).
Thus, on the one hand,

T̂R ψk,n (r) = eik·R ψk,n (r) ,

and on the other hand, T̂R ψk,n (r) = ψk,n (r + R), which is exactly the Bloch’s theorem,

ψk,n (r + R) = eik·R ψk,n (r) Bloch’s theorem .

To be noticed that with periodic boundary conditions,

!
ψk,n (r + N1 a1 ) = eik1 L1 ψk,n (r) = ψk,n (r) ,

therefore k1 L1 = 2πn1 for n1 ∈ Z and L1 = N1 a1 ,

 
2πn1 2πn2 2πn3
k= , , ,
L1 L2 L3

so ki is defined modulo 2πni /Li .

Another useful representation of Bloch’s theorem is the decomposition of the wavefunction to a periodic
function multiplied by a plane wave. Let us define the following function

uk,n (r) = e−ik·r ψk,n (r)

and see what happens when translating it with straight lattice vector R,

T̂R uk,n (r) = uk,n (r − R)

= e−ik·(r−R) ψk,n (r − R)
Bloch’s theorem = e−ik·(r−R) e−ik·R ψk,n (r)
= uk,n (r) .

We found that uk,n (r) = uk,n (r + R) is indeed periodic,

ψk,n (r) = eik·r uk,n (r) .

Remarks
• Under Bloch’s theorem conditions, the hamiltonian splits into blocks of different k values, reads
 
Hk
H=
 Hk ′  ,

..
.

where each block diagonlized with the n quantum numbers which can be, in principle, infinite
(unless introducing an cut-off due to an approximation, as in the tight-binding model).

• Any k in k-space can be confined to the (first) Brillouin zone (BZ), as k defined up to modulo in
the primitive reciprocal lattice vectors, thus the transformation k′ = k + G for a reciprocal lattice
vector G does not change the Bloch wavefunction.
The BZ might be finite, with N1 × N2 × N3 values of k.

2
1.2 Tight Binding Approximation
Suppose we’re given the atomic orbitals of the lattice sites, ϕn,R (r) (centered at the atom in position
R). We can construct Bloch wavefunctions from these atomic orbitals
1 X
ψn,k (r) = p ϕn,R (r)eik·R
N (1 + αn (k)) R

where N is the number of atoms and the overlap between atomic orbitals at different positions is

αn (R − R′ ) ≡ ⟨ϕn,R |ϕn,R′ ⟩
Z
= dr ⟨ϕn,R |r⟩ ⟨r|ϕn,R′ ⟩
Z
= dr ϕ∗n (r − R)ϕn (r − R′ )
Z
= dr ϕ∗n (r)ϕn (r − R′ + R) .

The prefactor normalizes the wavefunction (you are welcome to convince yourselves). Note that it
depends only on the displacement between the two orbitals. Moreover, we have
X
αn (k) = eik·R αn (R) .
R̸=0

Recall that the Bloch wavefunction ψn,k (r) above are degenerate (for the same n) within the atomic
Hamiltonian. The next step is to find the splitting to first order in perturbation theory by diagonalizing
the Hamiltonian within the degenerate subspace. We can write the lattice potential as
X
VR (r) = Vat (r − R′ )
R′
= Vat (r − R) + ∆VR (r)

where we’ve split the part of V arising from the R’s atom at and from the rest of the lattice (the
perturbation)
X
∆VR (r) = Vat (r − R′ ) .
R′ ̸=R

We can now calculate the energy

εn (k) = ⟨ψn,k |H|ψn,k ⟩

Z
= drdr ′ ⟨ψn,k |r⟩ ⟨r|H|r ′ ⟩ ⟨r ′ |ψn,k ⟩

∇2
 
1
Z X ′
= dr eik·(R−R ) ϕ∗n (r − R′ ) − + Vat (r − R) + ∆R V (r) ϕn (r − R)
N (1 + αn (k)) ′
2m
R,R
1 X ′
= ε(0)
n eik·(R−R ) ϕ∗n (r − R′ )ϕn (r − R)
N (1 + αn (k)) ′
R,R
1 X ′
+ eik·(R−R ) ϕ∗n (r − R′ )∆VR (r)ϕn (r − R)
N (1 + αn (k)) ′
R,R

(0)
where we’ve used the fact that ϕn are atomic orbitals, so Hat ϕn = εn ϕn . Overall we found that
1 X
ε0n αn (R) − γn (R) eik·R


εn (k) =
1 + αn (k)
R
Z
αn (R) ≡ drϕ∗n (r − R)ϕn (r)
Z
γn (R) ≡ − drϕ∗n (r − R)∆V0 (r)ϕn (r) .

3
We separate the contribution from R = 0 and use αn (0) = 1 to obtain
! !
1 0
X
ik·R
X
ik·R
εn (k) = ε 1+ αn (R)e − γn (R)e .
1 + αn (k) n
R̸=0 R
| {z }
αn (k)

Thus
γn (0) − γn (k)
εn (k) = ε0n −
1 + αn (k)
Since |γn |, |αn (k)| ≪ 1 we can to a good approximation omit αn (k) since
γ 2
1+α = γ(1 − α + α + ...) ≈ γ. So
X
εn (k) ≈ ε0n − γn (R)eik·R
R

The overlap integrals γn decay exponentially with lattice separation. We therefore keep only the on-site
(R = 0) and nearest neighbor terms (R = aj ). Furthermore, we define
γn (0) ≡ βn energy shift
γn (aj ) ≡ tjn nearest neighbor hopping term

Problem 1 - Tight-Binding on Honeycomb Lattice

Consider a honeycomb lattice as shown in the figure 1. All atoms are the same, with on-site energy
ε0 = 0.

a2 r2 a1

Figure 1: The honeycomb lattice

1. Write down the primitive vectors of the lattice, the basis vectors, and the reciprocal lattice vectors.
2. Calculate the dispersion within tight-binding approximation.
3. Draw the Brillouin zone and mark the Γ point (the center of the BZ), the M point(s) (the middle
of the BZ sides), and the K point(s) (the corners of the BZ).
Calculate the effective mass tensor (m∗ )−1
ij at both the conduction and the valence band for the
M and Γ points.1
1 The K point is special, as the effective mass there divergence (cannot be approximated to a parabolic dispersion) and

will follow significant discusses later on in the course.

4
Solution
1. The primitive vectors are given by
√ !
1 3
a1,2 =a ± ,
2 2

and we can choose the basis vectors to be

r1 = 0
a
r2 = √ ŷ .
3

The volume of a unit cell is

A = |a1 × a2 |
√ ! √ !
2 1 3 1 3
=a x̂ + ŷ × − x̂ + ŷ
2 2 2 2
√
3 2
= a ,
2
and the reciprocal lattice vectors are
2π
b1 = √ (a2 × ẑ)
3a2 /2
2π √ 
=√ 3, 1
3a
2π
b2 = √ (ẑ × a1 )
3a2 /2
2π √ 
=√ 3, −1 .
3a

2. Let us assume that all tunneling amplitudes are of magnitude t. Note that there are two atoms
in a unit cell. We denote the orthonormal Wannier states for the unit cell in location na1 + ma2
by |n, m, A⟩ , |n, m, B⟩, where A, B are the basis index inside the unit cell. The Hamiltonian is
therefore given by
X
H = −t (|n, m, A⟩ ⟨n, m, B| + |n, m, A⟩ ⟨n + 1, m, B| + |n, m, A⟩ ⟨n, m + 1, B| + h.c.) .
n,m

We diagonalize using the Fourier transform

1 X −ik·(na1 +ma2 )
|n, m, i⟩ = √ e |kx , ky , i⟩ ,
N kx ,ky

where i = A, B, which will yield

" #
t X X −i(k−k′ )·(na1 +ma2 ) 
′ −ik′ ·a1 ′ −ik′ ·a2 ′

H=− e |k, A⟩ ⟨k , B| + e ⟨k , B| + e ⟨k , B| + h.c.
N n,m ′
k,k

By using
1 X −i(k−k′ )·(na1 +ma2 )
e = δk,k′ ,
N n,m

we obtain the hamiltonian in k-space

X
|k, A⟩ ⟨k, B| + e−ik·a1 ⟨k, B| + e−ik·a2 ⟨k, B| + h.c.



H = −t
k

5
 Thus, for a specific value of k we can write the Hamiltonian in matrix form, with the basis
{|k, A⟩ , |k, B⟩}
 
0 f (k)
Hk = −t
f ∗ (k) 0
f (k) = 1 + e−ik·a1 + e−ik·a2 .
To find the dispersion, we calculate the eigenvalues of H. We can use this nice trick of multiplying
it by its’ Hermitian conjugate, and by that getting the eigenenrgies squared.
|f (k)|2
 
0
H2 = t2
0 |f (k)|2
r  
ε = ±t 1 + e−ik·a1 + e−ik·a2 1 + eik·a1 + eik·a2
p
= ±t 3 + 2 cos(k · a1 ) + 2 cos(k · a2 ) + 2 cos(k · (a1 − a2 ))
√ √
v ! !
u
u k x a 3k y a kx a 3k y a
= ±tt3 + 2 cos + + 2 cos − + 2 cos(kx a)
2 2 2 2
√
v !
u  
u kx a 3k y a
= ±tt3 + 4 cos cos + 2 cos(kx a) .
2 2

3. The BZ is in figure 2.

M ′′
K K′ b1

M′ M

K′ Γ K

M M′

K K′ b2
′′
M

Figure 2: The BZ of the Honeycomb lattice

Recall that the effective mass is given by

∂ 2 ε(k)
(m∗ )−1
ij (k) = .
∂ki ∂kj
We then will calculate first the dispersion in the vicinity of Γ, M and K points (up to second order
in k) to make the derivative easier.
Around Γ (kx a, ky a ≪ 1) the dispersion of the n’th band (1/2) is
v
√
u  !2 
  q a 2   
1 1 3q a 1
u
Γ x y 2
εn (q) ≃ ±tt3 + 4 1 − 1− + 2 1 − (qx a)
u  
2 2 2 2 2

6
 r
1 1
= ±3t 1 − (kx a)2 − (ky a)2
6 6
 
1 1
≃ ±3t 1 − (kx a)2 − (ky a)2
12 12
t 
= ±3t ∓ (kx a)2 + (ky a)2
2
4 
a t 1 0
(m∗n )−1
ij (Γ) = ∓
2 0 1
 
2 1 0
(m∗n )ij (Γ) = ∓ 2 .
a t 0 1

In the vicinity of Γ point, at the upper (lower) band the electrons behave hole(electron)-like due to
the negativity(positivity) of the effective mass tensor. In addition, as the coupling coefficient (t)
grows, the effective mass is reduced.
√
Similarly (but with much more sweat), around the M ′′ point we can expand k = π/a(0, 2/ 3) +
(qx , qy ) where qx a, qy a ≪ 1. We choose specifically the M ′′ point and not M or M ′ because the
boundary of the BZ is aligned with the kx axis this way (and will be easier to discuss the interesting
property of this point later). To find the rest of the M points’ effective mass tensor, one can simply
rotate the matrix.

a2 t −1 0
!
∗ −1 ′′
(mn )ij (M ) = ± 3
2 0
2
−2 0
!
∗ ′′ 1
(mn )ij (M ) = ± 2 2 .
a t 0
3
Regarding the upper band (for example), the electrons have a negative mass when moving along the
kx axis (the boundary of the BZ), namely, they move on a hole-like trajectory, while when moving
inward/outward the BZ they have a positive mass, corresponding to an electron-like trajectory.
This is a good example of how the effective mass differs from real mass, as not only it is anisotopic,
but also can hold negative values!

7
Problem 2 - Van-Hove Singularities
The tight binding dispersion of the 2D square lattice with a lattice constant a is given by:

ε(k) = −2t (cos(kx a) + cos(ky a))

1. Calculate the density of states g(ε) near the bottom of the band.
2. The dispersion has saddle points at k = (π/a, 0) and k = (0, π/a). Compute the contribution of
the saddle points to the density of states.

Solution
1. Recall that the DoS is defined by
1X
g(ε) = 2 δ(ε − εk )
Ω
k

Where Ω is the volume and the factor of 2 comes from the spin. In the thermodynamical limit,
this becomes
d2 k
Z
g(ε) = 2 δ(ε − εk )
(2π)2
The DoS is useful for calculating many transport and thermodynamic properties of the electron
gas. For free electron gas, the specific heat, the susceptibility, and the resistivity are proportional
to g(εF ), e.g. cV = π 2 kB
2
T g(εF )/3 [1, ch.2]. Near the bottom of the band,

d2 k
Z
g(ε) ≃ 2 δ(ε + 4t − ta2 k 2 ) .
(2π)2
Let’s use



X δ x − x0
δ f (x) = (1)
|f ′ (x0 )|
x0 ∈{roots of f }

to obtain
1 1
Z
g(ε) = dkk p (δ(k − k+ ) + δ(k − k− )) Θ(4 + ε/t) ,
π 2ta 4 + ε/t
p
where Θ is the Heaviside step function and ak± = ± 4 + ε/t are the desired roots. Solve the
integral to get
1
g(ε) = Θ(4 + ε/t) .
2πta2

2. For ε = 0, the contour line we find for the given dispersion is

π
 −k k ∈ [0, π/a]
a



π


− + k k ∈ [π/a, 0]


C(k) = a .
π

 − − k k ∈ [0, −π/a]


 a

 π
 +k
 k ∈ [−π/a, 0]
a
In this case, we can no longer use 1 but have to take into account that we have a continuous
spectrum of zeros2 ,
1
Z   Z
dd xδ f (x) = dd−1 x .
−1
f (0) |∇f (x)|
2 Known as a ”simple layer integral”, for those of you who desire the derivation you can find it extensively online.

8
It means, that generally,

dd−1 k 1
Z
g(ε) = 2 .
surface ε (2π)d |∇k ε(k)|

For this dispersion, the divergence is contributed from the (±π/a, 0) and (0, ±π/a) points (you can
see the notes from workshop 10 in the solid state course). Thus, if indeed we find a divergence, we
can estimate

the integral only around one of these points. Near the saddle points, for example for
k = 0, πa + (qx , qy ) we have

εq ≃ −ta2 qx2 − qy2 ,

since the contour is a hyperbola, it leaves the region of the valid expansion. Thus, we introduce
an artificial
p upper cutoff in kmax a = Λ where Λ > 0 is dimensionless and must be greater than
kmin a = ε/t as can be seen in figure 3. It will not affect the divergence.

ky

ε=0
ε>0
kmax
ε<0
kmin

kx

Figure 3: The dispersion counter around ε = 0

dq
Z
g(ε) ≈ 2 2
δ(ε + ta2 (qx2 − qy2 ))
C (2π)
dq 1
Z
=2 2


C (2π) ∇q ε + ta (qx − qy )
2 2 2

1 1 1
Z
= 2 2
dq
2π ta C |(qx , −qy )|
1 1 1
Z
= 2 2
dq q
2π ta C q2 + q2 x y
kmax
1 1 1
Z
parametrization =4× dqy r
2π 2 ta2 kmin ε
2qy2 −
ta2
kmax
2 1 1
Z
= dqy r
π 2 ta2 kmin ε
2qy2 −
ta2
√ kmax
2 1 1
Z
2 2
εt ≪ a q ≃ 2 2 dq
π ta kmin q

9
 √  
2 1 kmax
= log
π 2 ta2 kmin
√ !
2 1 Λ
= 2 2 log p
π ta ε/t
√   
2 1 1 t
= 2 2 log(Λ) + log
π ta 2 |ε|

The cutoff we introduced contributes a constant, thus, compering with a divergence of the ε at 0
it can be neglected. From symmetry the contribution of the other saddle point is similar. Hence
for two saddle points
1 t
g(ε) ∼ √ log .
2
2π ta 2 |ε|

1d 2d 3d
DOS(ε)

0
εmin εmax εmin εmax εmin εmax

Figure 4: Van-Hove singularities in the DoS for a cubic lattice: in 1d there is a divergence, in 2d there
is a marginal divergence and in 3d there is a divergence in the derivative of the DoS (namely a cusp)

References
[1] N. W. Ashcroft and N. D. Mermin. Solid State Physics. Holt-Saunders, 1976.

10