Streda’s Formula, Integer Quantum Numbers, and the Diophantine Equation∗

A. Streda’s formula

Suppose that we start with a system where the Fermi energy is in a gap. We have seen that as long as F
is in a gap, the Hall conductivity must be given by
νe2
σ xy = = −σyx , where ν ∈ Z (1)
2π~
Now if we vary the magnetic field continuously, keeping ρV and other quantities fixed (so that the flux
through the unit cell changes), the density of electrons ρe will also vary. Then, according to Streda’s
theorem, which we prove below, we must have
∂ρe ν ∂ρe
= , which is equivalent to =ν (2)
∂B φo ∂ρB
Proof : Consider turning on a weak electric field in the y direction, via a vector potential given by

A = A0 + δAy (t), where δAy (t) = Re[A0 ei(kx−ωt) ]

so that E 0 (k, ω) = iωA0 (k, ω) and j x (k, ω) = σ xy (k, ω)E 0 (k, ω) (3)

As long as the Fermi-energy is in a gap, we can replace σ xy (k, ω) by σ xy (0, 0) ≡ σ xy for a slowly varying
electric field in space and time. In this limit, we can use the continuity equation for charge to write
∂ρ h i ∂ ∂B
= −∇ · j = −∂ x j x = −σ xy Re (ik)(iω)A0 ei(kx−ωt) = σ xy (∇ × A)z = σ xy (4)
∂t ∂t ∂t
Noting that the charge density ρ = −e ρe , we have proved Streda’s theorem.
It should be emphasized that Streda’s formula applies only when the Fermi-level is in a true energy gap.
It does not apply when the Fermi level is a region of localized states, because the derivative of the electron
density with respect to the magnetic field in the limit of small but finite frequency does not then coincide
with the derivative of the density with respect to the magnetic field in an equilibrium state.

B. Application to a periodic potential.

Consider, now a system of electrons in a periodic potential with a Bravais lattice with primitive vectors
and a and b. Then, the density of cells per unit area is ρV = |a × b|−1 . Defining the flux density ρB = B/φo ,
the ratio of the flux passing through a cell to the quantum of flux is given by φ = ρB /ρV .

∗
These notes are based on lectures given by Prof. B. I. Halperin at Harvard University during Fall, 2013. Transcribed by Shubhayu
Chatterjee, and edited by BIH.
 2

First, consider the case when φ = p/q is a rational number, p and q being mutually prime integers. Then
we can construct a supercell with area q|a × b| which contains an integer number (p) of flux quantum, and
apply Bloch’s theorem with this supercell to get the band structure. Let the Fermi energy F be in a gap.
Then band theory dictates that the number of electrons per supercell must be an integer nF (ignoring spin).
Then the number density of electrons ρe is given by

n F ρV
ρe = (5)
q
Now, let us vary the flux continuously. Starting with an initial rational value of φ0 = p0 /q0 , let us change
it so that φ = p/q. Note that the change in flux can be thought of instead as expanding or contracting the
lattice, which will change the period of the periodic potential. However, we can do a series expansion of the
changed potential in powers of the small change in lattice parameters, and therefore we effectively have a
small perturbation in the potential which cannot affect the quantized Hall conductance or close the gap. So,
we expect that we can make the following expansion of density of electrons:

ρe
= α + β(φ − φ0 ) + γ(φ − φ0 )2 + higher order terms (6)
ρV
Let η = α − βφ0 . Recalling that ρe /ρV = nF /q, we have

nF = ηq + βp + γ(φ − φ0 )2 q + ... (7)

Initially we have φ = φ0 , so we know that nF = ηq0 + βp0 ≡ n0F is an integer. Now, consider the sequence
of integers pn = np0 + 1, qn = nq0 + 1, so that limn→∞ pn /qn = φ0 . For n  1, we have φ − φ0 ∼ 1
qn ∼ n1 .
Hence,
!
1
nF = ηqn + βpn + γ(φ − φ0 ) qn + ... = η + β + n(po + qo ) + γ O
2
. (8)
n
Therefore, for any value of γ, the term proportional to γ will be very small compared to unity for large
enough n,. The first two terms may be written as nn0F + η + β. This will be an integer if and only if η + β is
an integer. Since p0 and q0 are mutually prime, knowing that ηp0 + βq0 is an integer, and η + β is an integer,
we can conclude that η and β are both integers. Since the LHS is an integer, looking at the large n limit we
conclude that γ = 0 and η + β is also an integer. By a similar argument, all higher order terms will also be
zero, and we have

nF = ηq + βp (9)

By Streda’s formula, we must have

∂ρe ∂ ρe ∂ nF ∂
! !
ν= = = = (η + βφ) = β, (10)
∂ρB ∂φ ρV ∂φ q ∂φ
 3

So

nF = ηq + νp. (11)

Now let us divide both sides of this equation by q/ρV . The result is

ρe = ηρV + νρB (12)

Equation (11) was derived, and only is meaningful, under the assumption that φ is rational. Equation
(12), however, is true even if φ is irrational, with η and ν each taking on a fixed integer value, over an
entire range of φ. In fact the integers cannot change as long as the Fermi energy remains in an energy
gap. (We already know that ν cannot change, while the three densities in the equation change only change
continuously, so the integer η cannot jump to a different value. Thus the electron density ρe is locked to
an integer multiple of the flux density plus an integer multiple of the density of unit cells, over the entire
interval where the Fermi energy remains in the gap.

C. Solutions of the Diophantine Equation

Equation (11) is linear diophantine equation, which relates the integers nF , ν, and η, if p, q are known..
Suppose we can solve for the band structure for a given flux φ = p/q, and compute nF given the Fermi
energy. Given this information, we can ask if we can find η and ν. However, from the above equation we
cannot determine them uniquely, as a linear diophantine equation has an infinite number of solutions. If
(ν1 , η1 ) are solutions, then so are (ν1 + qN, η1 − pN) for any N ∈ Z. We can therefore see that if two bands
touch and then separate, the Chern number ν can change by a multiple of q and η will change by a multiple
of p. We shall see later, in more detail, how this can occur.
Let us now consider one example of the diophantine equation.. Let p = 3 and q = 7 so that φ = 3/7. If
we restrict ourselves to solutions with |ν| ≤ q/2, we have the following solutions for nF :

nF ν η

0 0 0
1 -2 1
2 3 -1
3 1 0
4 -1 0
5 -3 1
6 2 -1
7 0 1
 4

Of course, many other solutions are possible, which would give large values of |ν|. However, these happen
to be the correct solutions for the Hofstadter model at flux φ = 3/7.

D. Current induced by a moving potential.

Let us now consider what happens if we have a quantized Hall state in a periodic poten-
tial and we slowly displace the potential in space. For example, consider a potential V(r) =
V cos(2π(x − X)/a) + cos(2π(y − Y)/b) , and change the periodic potential by varying X(t) and Y(t) slowly,
 

We may also apply a small electric field. We claim that the current response is given by

dR
j = eνρB vD + eηρV (13)
dt

where vD = (E × B)/B2 is the drift velocity of a classical charged particle in perpendicular electric and
magnetic fields, The proof of this equation is a homework problem.
We may also interpret vD as the drift velocity of the flux lines, because, from Maxwell’s equations,

dρB /dt = −∇ · (ρB vD ) (14)

Therefore, we see that the electron charge seems to break into two pieces, one piece tied to the lattice and
the other to the flux.

E. Generalization to Several Incommensurate Potentials

Formula (12) can be generalized to the case of a quasiperiodic potential, which may be described as
a combination of several periodic potentials with different periodicities. For example, suppose that the
potential is given by
! ! !
2πx 2πx 2πy
U(r) = λ1 cos − θ1 + λ2 cos − θ2 + λ3 cos − θ3 . (15)
a1 a2 b

We define the densities of unit cells by ρV1 = 1/ba1 and ρV2 = 1/ba2 Then if the Fermi energy lies inside
an energy gap, we can show that

ρe = ν ρB + η1 ρV1 + η2 ρV2 (16)

where ν, η1 , η2 are all integers. Furthermore, if any or all of the parameters B, a1 , a2 , b are varied continu-
ously, the three integers cannot change, as long as the Fermi level remains in a gap. We say that the integers
are quantum numbers characterizing the energy gap. We note that this relation holds, and is meaningful,
even in the absence of a magnetic field, in which case we may simply set ν = 0.
 5

When the three densities ρB , ρV1 , ρV2 happen to be commensurate with each other, then one may define
an enlarged unit cell, which contains an integer number of flux quanta as well as an integer number of unit
cells of each potential. Then the integer quantum numbers may be expressed in terms of Chern numbers
derived for this band structure in the enlarged unit cell For example, the quantum number η1 is the sum over
occupied bands of the Chern numbers of the bands that one obtains in a cycle where when one varies θ1 and
k x , the x-component of the Bloch momentum. Specifically, we consider a in cycle where θ1 is varied from
0 to 2π, then k x is allowed to increase by one reciprocal lattice vector, then θ1 is decreased and then k x is
decreased. The proof of this will not be given here, but a similar theorem, for an incommensurate potential
in one dimension, appears as a problem on Problem Set 3.
It is also possible to define a Chern number for a cycle in which, say, θ1 and θ3 are varied around the
unit cell. The quantum number η1 can be related to this Chern number provided that φ , 0.
The above discussion can easily be extended to a quasiperiodic potential with an arbitrary number of
incommensurate periodicities.