Quantum Matter / Matéria Quântica

Solutions to Problem set 3 — A model of strained graphene (TB and Dirac de-
scriptions)

Figure: Left: representation of the honeycomb lattice of a graphene monolayer with the three nearest-neighbor
hoppings labeled as tα (α = 1, 2, 3). Right: the corresponding reciprocal space. The shaded hexagon identifies
the first Brillouin zone and some high-symmetry points are labeled.

Problem 1 (simple model for the electronic structure of strained graphene)

a) Write down the expressions for the primitive vectors a1 and a2 shown in the figure above, as
well as the coordinates of the points labeled K and M1 at the boundary of the first Brillouin
zone (BZ). Express all of these in terms of the carbon-carbon distance, a.

Solution

With the lattice orientation given, they are

√ ! √ !
√
 
1 3 4π 2π 3 1
a1,2 = 3a ± , , K= √ ,0 , M1 = , .
2 2 3 3a 3a 2 2

b) Consider the orthogonal nearest-neighbor tight-binding (TB) description of the electrons in

the pz orbital of graphene, where the three nearest-neighbor hoppings are marked in the figure
above as t1 , t2 , t3 . In the notation of the lectures, the 2-band Hamiltonian is expressed as
h i
ψ̂k† h(k) ψ̂k , ψ̂k† ≡ c†kA c†kB ,
X
Ĥ =
k∈BZ

where c†kα creates an electron of wavevector k on sublattice α ∈ {A, B}, and h(k) is the 2 × 2
tight-binding matrix with the form
" #
0 f (k)
h(k) = . (1)
f (k)∗ 0

1
 Taking the local orbital energy as the zero of the energy scale, but assuming that the nearest-
neighbor hoppings tj (j = 1, 2, 3) may be different, express the coherence function f (k) in
terms of tj and of the primitive vectors a1,2 .

Solution

If n represent a Bravais-lattice vector connecting the unit cell R of an atom in sublattice A with the unit
cell of its nearest-neighbor atom, the coherence function is given by
X
f (k) = tR,R+n eik.n .
n

From the labeling of the 3 closest neighbors used in the figure we identify n1 = −a2 , n2 = 0, n3 = −a1 .
Therefore,
f (k) = t1 e−ik.a2 + t2 + t3 e−ik.a1 .

c) Considering the conventional graphene case where all the hoppings are equal, tα = −t:

i) Write the corresponding energy dispersion Eλ (k), where λ is the band index, consistent
with the above lattice parameterization.

ii) Plot Eλ (k) along the path ΓKM2 K ′ Γ in k-space; use t for energy units and mark each
of the high symmetry points on the horizontal axis.

Solution

(i) The energy band dispersion is given by the eigenvalue spectrum of the TB matrix h(k), which is
immediate to see:
E± (k) = ±|f (k)| = ± t1 e−ik.a2 + t2 + t3 e−ik.a1 ,

where the two bands (λ = ±) are symmetric relative to zero energy. For graphene at equilibrium, we put
tα = −t and obtain

E± (k) = ±t 1 + e−ik.a2 + e−ik.a1

h √ √   i
= ±t 3 + 2 cos( 3kx a) + 4 cos 23 kx a cos 23 ky a .

(ii) The bandstructure along this path is the following,

2
 where the red (blue) line represents the valence (conduction) band.

d) Suppose now that the three nearest-neighbor hoppings can be different and may take arbitrary
values t1 6= t2 6= t3 . (This will happen if graphene is deformed under planar tension along
some arbitrary direction.)

i) What do we obtain for Eλ (k) when t2 = −t and t1 = t3 = 0? Give a brief physical

argument for why this result is trivially expected.

ii) Compute Eλ (k) when t2 = 0 and t1 = t3 = −t. Comment on the nature of this energy
dispersion, what it says about the motion of the electrons, and why it is also an expected
outcome given the structure of the hoppings.

iii) Determine the value of Eλ (k) at k = M for arbitrary t1,2,3 .

Solution

(i) In that case, E± (k) = ±t. Physically, the situation with t1 = t3 = 0 is equivalent to removing the
hopping paths defined by the hopping amplitudes t1 and t3 . This means that electrons in an atom labeled
A can only hop to the atom B directly above, and vice-versa, but they can never travel to their left or
right neighbors. As a result, all electrons are localized in groups of A–B dimers within their respective unit
cells. The absence of hopping to neighboring unit cells results in energy levels independent of k, because
the k dependence arises only when there are hopping pathways spanning the entire crystal.

Within each of those A–B dimers connected by the hopping parameter t2 , both carbon atoms contribute
with one local pz orbital, each with energy Epz . This is physically equivalent to a 2-atom carbon molecule
where the nonzero hopping amplitude for the electron leads to a bonding and anti-bonding splitting of the
energies that become Epz ∓ t2 , respectively. If Epz is used as the zero of energy, we get simply ∓t2 .

In conclusion, having the energy spectrum reduced to E± (k) = ±t in this case reflects the fact that,
physically, the electrons have become confined to such dimer molecules. Since they cannot hop throughout

3
 the crystal, their energy spectrum got reduced to the corresponding bonding and anti-bonding levels within
each dimer. Each of these levels has a degeneracy equal to the number of unit cells in the crystal.

(ii) From the general expression above, we obtain

√ 
t →0
E± (k) = ± t1 e−ik.a2 + t2 + t3 e−ik.a1 −−2−−→ ±2t cos 23 kx a .

Physically, the situation now reflects the motion of the electrons along the horizontal 1D “chains” defined
by the bonds labeled t1 and t3 . In a single-orbital 1D chain with atoms separated by a distance a′ and
electronic hopping amplitude −t, the energy dispersion is simply −2t cos(ka′ ). The effective 1D chains
√
of this modified graphene scenario indeed have an effective lattice parameter a′ = 3a/2, which is the
distance between nearest neighbors connected by t1,3 when projected along the x̂ direction (the direction
of effective electronic motion in this restricted case). The result above for Eλ (k) reflects precisely that,
from the point of view of the electrons, it’s as if they move along 1D chains. The partition of the system
into 2 families of similar 1D chains explains the two different signs ± of the corresponding band dispersions.
Each of these effective 1D bands has a degeneracy equal to the number of corresponding 1D chains in the
crystal when t2 = 0.

(iii) There are three non-equivalent M points at the edge of the 1st Brillouin zone:
√ ! √ !
2π 3 1 2π 2π 3 1
M1 = , , M2 = (0, 1) , M3 = − , .
3a 2 2 3a 3a 2 2

The energy at each of these is exactly given by

E± (M1 ) = ±|t1 + t2 − t3 |, E± (M2 ) = ±|t1 − t2 + t3 |, E± (M3 ) = ±| − t1 + t2 + t3 |.

We can see that, when all three hopping parameters t1,2,3 take different generic values, the energies will
be different at each of the 3 points M1,2,3 . This is one way to see that these M points are not equivalent
to each other (equivalent points in reciprocal space always have the same energy).

Note — Since the text of the question didn’t specify one particular M point to consider, it was enough to
provide only one of these results (any one) to have the response considered answered in full.

e) We now take t1 = t3 = −t and t2 = −t − δt, where δt can be thought as a perturbation to the

vertical hopping t2 relative to the case of ideal graphene. (It will happen, for example, if you
compress graphene along ŷ.)

i) In the same graph, show the curves Eλ (k) along the path ΓKM2 K ′ Γ for the following
three cases: δt = 0, δt = −t/2, δt = +t/2. From these plots, how does δt affect the
Dirac points?

ii) There is a critical value (δtcrit ) at which the Dirac points merge and, if δt is changed
further, they disappear leaving a gap in the spectrum. Assuming that −1 ≤ δt/t < ∞,
analytically find the exact value of δtcrit .

4
 iii) In the same graph, show the curves Eλ (k) for δt = δtcrit and for δt = 2δtcrit .

Solution

(i) You should obtain a plot like the following,

where each color represents the cases δt = 0 (dashed black), δt = −t/2 (blue), δt = +t/2 (red). Although
Dirac points remain along the ΓKM2 K ′ Γ line, we see that their position changes with δt: if δt < 0 the
distance between Dirac points increases; on the other hand, they approach each other if δt > 0 (red curve).

(ii) From the previous question, we expect the merging to occur at some value δt > 0. Given the
symmetric approach of the two Diract points that we see in the plot of the previous response, the merging
should occur at k = M2 . We have also seen above that

E± (M2 ) = ±|t1 − t2 + t3 | −→ ±|t − δt|.

Clearly, when the merging occurs we have E± (M2 ) = 0, from where we conclude that the critical value is

δtcrit = t,

or, equivalently, the two Dirac points merge at k = M2 when t2 = 2t1 = 2t3 .

(iii) The plot shown below represents Eλ (k) in the cases δt = 0 (dashed black), at the critical value
δt = t (red), and δt = 2t (blue). We can see that, if δt increases further above the critical value for the
merge, an energy gap appears in the spectrum, which grows with the magnitude of δt.

5
Comment — To confirm this interpretation that for positive and increasing δt the Dirac points approach
along the path ΓKM2 K ′ Γ, merge at δt = t and we have a gap at k = M2 if δt > t, we may plot the energy
dispersion over the entire Brillouin zone. This is illustrated in the panels below that show density plots of
the conduction band energy, E+ (k)/t, for different values of δt/t. It is clear that, by perturbing (only) the
hopping parameter t2 , the Dirac points are displaced horizontally and eventually merge at k = M2 .

6
f) Dirac approximation — To study the behavior near K/K ′ , we set k = ξ K + q (ξ = ±1, valley
index) and approximate the TB Hamiltonian to lowest order in q. In ideal graphene case (all
hoppings the same), the coherence function in the TB matrix Eq. (1) approximates to

3ta
f (ξ K + q) ≃ ~vF ξqx − iqy , vF = , |q| ≪ K.
2~

Considering again the scenario of question (e) above, where t1 = t3 = −t and t2 = −t − δt:

i) Obtain the corresponding lowest-order expansion in q for f (k).

(ξ)
ii) In this low-energy approximation, obtain the energy dispersion Eλ (q) for valley ξ in the
presence of the perturbation δt. Describe the shape and location of the Fermi surface
and relate this analytical result with what you observed in question (e).

iii) Write the effective Hamiltonian Ĥ(ξ) (r) in the real space, continuum approximation
for valley ξ. Show that, despite the hopping perturbation δt, you still get a Dirac
Hamiltonian. In particular, show that we can write

Ĥ(+) = vF σ · (p̂ − A(+) ), Ĥ(−) = −vF σ∗ · (p̂ − A(−) ),

where A(ξ) is a 2D vector that depends on δt and valley index. Obtain the expression
for A(ξ) in terms of δt and the other parameters of the system.

Remark — Note that this last result tells us that the net effect of perturbing the hoppings is to
replace p → p − A(ξ) . In other words, a hopping perturbation δt acts similarly to adding a magnetic
vector potential, A(ξ) , to each valley (remember how a magnetic field enters in the classical Hamil-
tonian of a particle, for example). There is of course no magnetic field here; but, since the dynamical
implication is the same as if there was one, we call it a (strain-induced) pseudomagnetic field.

7
Solution

(i) The starting point is

f (k) = t1 e−ik.a2 + t2 + t3 e−ik.a1 ,

which becomes, after replacing the tα :

 
f (k) = −t e−ik.a2 + 1 + e−ik.a1 − δt.

The term in (· · · ) is what we get for ideal graphene and the correction is simply a constant. Hence, there’s
actually no extra work compared to the case of ideal graphene when we put k = ξ K + q and expand to
linear order in qa:
  3ta
f (ξK + q) = −δt − t e−ik.a2 + 1 + e−ik.a1


≃ ξqx − iqy − δt.
k=ξK+q 2

To lowest order in qa we then have

3ta

f (ξK + q) ≃ ξqx − iqy − δt.
2

(ii) In the previous question we obtained a correction to f (k) that is purely real. It is therefore helpful
to express such result as

3ta


f (ξK + q) ≃ ξqx − iqy − δ = ~vF ξ q̃x − iq̃y ,
2
where we defined
ξδt 3ta
q̃x ≡ qx − , q̃y ≡ qy ~vF ≡ .
~vF 2

From here, the energy dispersion follows immediately:

r
q
2 2
ξδt 2
Eλ (ξK + q) = ±|f (ξK + q)| ≃ ±~vF q̃x + q̃y = ±~vF qx − + qy2 .
~vF

At some given Fermi energy, EF , the Fermi surface is defined by the condition EF = Eλ (k). In the current
approximation, this condition becomes
r
ξδt 2
EF = ±~vF qx − + qy2 ,
~vF
p
which defines two circles (one for each ξ = ±) of radius qF ≡ EF /(~vF ) in reciprocal space. The center
of these circles marks the position of the two Dirac points, which are thus located at the wavevector
 
(ξ) ξδt
kD = ξK + x̂.
~vF

Note that the energy dispersion remains linear (conical) relative to these shifted Dirac points and the
corresponding Fermi velocity, vF , is unchanged relative to the ideal graphene case. Consequently, the

8
leading effect of the hopping perturbation δt is simply to shift each Dirac point by the vector

(ξ) ξδt
δkD ≡ x̂
~vF

along the horizontal direction.

This immediately shows that, when δt > 0, the two Dirac points approach each other relative to their
separation in ideal graphene. Conversely, their distance increases if δt < 0. This is precisely what we have
seen in the explicit plots of question (e) above: when compared to the case δt = 0, the Dirac points are
nearer to each other for δt = t/2, but further apart for δt = −t/2.

Comment 1 — We should keep in mind that the analytical result above has been derived in a first-order
(ξ)
expansion of the Hamiltonian near the Brillouin zone points ξK. Therefore, the result for δkD is valid
(ξ)
provided that |δkD a| remains small or, equivalently, while δt/t remains small. For example, we may use
(ξ)
the expression above for δkD to obtain a rough order-of-magnitude estimate of the δt needed to merge
the Dirac points; but it fails to provide the exact value δt, which is why in the previous question we resorted
to the full tight-binding dispersion relation to obtain δtcrit = t.
(ξ)
Comment 2 — The fact that the Dirac point displacement vector δkD obtained above is parallel to x̂ is
due to the specific hopping perturbation chosen, which only affects t2 . But the result is easily generalized
to an arbitrary perturbation of the type tα → −t − δtα for all α ∈ {1, 2, 3}: we need only consider the
expansion of the coherence function
X X X
f (ξK + q) = tα ei(ξK+q).nα = −t ei(ξK+q).nα − δtα ei(ξK+q).nα ,
α α α

for small qa. The first term is just what we get for ideal graphene (δtα = 0). In the second term, if we
neglect small products like qa×δtα , it reduces to α δtα eiξK.nα , which is some complex-valued constant
P
(ξ) (ξ)
that we can always express in the form Ax − iAy . In the case of a general perturbation with all δtα 6= 0,
(ξ) (ξ)
both components Ax and Ay will be non-zero, which leads also to a more general 2D displacement
(ξ)
vector δkD for the Dirac points. For example, if we have t1 = t2 = − t and t3 = − t − δt, the Dirac points
move along the oblique direction KM1 (M1 is defined in Figure 1) and eventually merge at k = M1
when t3 = 2t1 = 2t2 . (In this case, the pseudomagnetic vector potential A(ξ) that appears in the following
question has components along both x̂ and ŷ.)

(iii) From the results obtained in the previous question, we recognize that the 2×2 tight-binding Hamil-
tonian matrix h(k) becomes, near the K points,
 ξδt 
h(ξ K + q) ≃ ~vF ξσx q̃x + ~vF σy q̃y = ~vF ξσx qx − + ~vF σy qy .
~vF

The replacement q → −i∇ = p̂/~ leads to the following effective low-energy Hamiltonian operators on
each of the valleys labeled by ξ = ±:
 δt 
valley at + K : Ĥ(ξ=+) = +vF σx p̂x − + vF σy p̂y
vF
 δt 
valley at − K : Ĥ(ξ=−) = −vF σx p̂x + + vF σy p̂y
vF

If we introduce the vector of Pauli matrices σ ≡ σx x̂ + σy ŷ, the effective Hamiltonian can be written as

Ĥ(+) = vF σ · (p̂ − A(+) ), Ĥ(−) = −vF σ ∗ · (p̂ − A(−) ),

9
where the 2D vector A(ξ) is given on each valley by

ξ δt 3ta
A(ξ) ≡ x̂, vF ≡ .
vF 2~

As discussed in the previous response, this vector A(ξ) defines amount and the direction along which the
Dirac points are displaced due to the hopping perturbation.

Comment 1 — It’s very interesting to see that a hopping perturbation has an effect on the electrons similar
to that of a magnetic field. To see why we can say that, consider the low-energy Dirac Hamiltonian of
unperturbed graphene:
Ĥ(+) = vF σ · p̂, Ĥ(−) = −vF σ ∗ · p̂.

If we subject the system to an external magnetic field B = ∇ × A, the Hamiltonian is updated to

Ĥ(+) = vF σ · p̂ − eA , Ĥ(−) = −vF σ ∗ · p̂ − eA ,

where we performed the minimal coupling substitution p → p − eA in order to obtain the Hamiltonian that
describes the system in the presence of the magnetic field. Hence, there is a clear formal similarity:

hopping perturbation: Ĥ(+) = vF σ · (p̂ − A(+) ).

Ĥ(+) = vF σ · p̂ − eA .


magnetic field:

The electrons, of course, only “care” about the Hamiltonian itself, not the origin of the vector potential
term in either case. A consequence of this is that, by modifying the hopping parameters in smart ways
by strain (lattice deformations), we can simulate the effect of magnetic fields without actually subjecting
the system to a magnetic field! Because of this, the vector A(ξ) that encodes the correction to the
Hamiltonian due to hopping perturbations is commonly designated by “pseudomagnetic vector potential”.
What is striking about this is that, when we put in actual numbers, we see that it’s possible to generate
the equivalent of 100’s of Tesla in this fashion relatively easily, which is obviously impossible with actual
magnetic fields (the highest continuous magnetic fields achievable in a lab are around 50 T). Effective
strain-induced pseudomagnetic fields of this type larger than 500 T have been confirmed experimentally
[See, for example, Science 329, 544 (2010); Nature Physics 6, 30 (2010)].

Comment 2 — There is, however, an important difference between a strain-induced hopping perturbation
and a real magnetic field: the latter breaks time reversal symmetry while the former does not. Time-reversal
symmetry is ensured by the fact that, as we see in the expression obtained above, the pseudomagnetic vector
potential A(ξ) has opposite sign on each valley, which means that dynamical effects on the electrons in one
valley are compensated by the opposite effect happening at the other valley (electrons in different valleys
feel exactly opposite pseudomagnetic fields). If you’re interested in reading further on this, you may begin
with Chapter 10 of Katsnelson’s book.

10