Lectures 12 and 13 – the Haldane model

Tight binding
In 1988, D. Haldane demonstrated that a quantized Hall effect emerges in a magnetic honeycomb
lattice at zero magnetic field. Starting from a tight binding model for graphene, he included second
neighbor interactions that manifest as hoping terms that form loops each enclosing a single
magnetic ion. The impact of the magnetic ion is accounted for via an Aharonov-Bohm phase that
is picked up once a loop is completed around at magnetic ion (path I in Fig. 1(a)) compared to
when it’s completed around the first Brillouin zone (path II in Fig. 1(b)). Paths A and B
respectively described nearest neighbor and second neighbor hopping.

FIG 1. (a) Path 1: nearest neighbor hopping loop not enclosing any magnetic moments. (b) Path
II: second neighbor hopping loop that encloses a magnetic moment. A net total Aharonov Bohm
phase equal to 3φ is accumulated when going around path II compared to path I.
The tight binding Hamiltonian describing the Haldane model comprises three parts:

𝐻0 = 𝑡1 ∑(𝐴† 𝐵 + 𝐵 † 𝐴)𝑒 𝑖𝑘⃗.𝑟

𝑟

𝐻0 is the unperturbed graphene Hamiltonian

𝐻1 = 𝑀(𝐴† 𝐴 − 𝐵 † 𝐵)
𝐻1 is an inversion symmetry breaking term that usually results from the presence of two different
atoms occupying lattice sites A and B.
𝐻2 is developed in appendix I, and yields:
 ⃗ . 𝑏⃗) − 2𝑡2 sin 𝜑 ∑ sin⁡(𝑘
𝐻2 = [2𝑡2 cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗)] 𝐴† 𝐴

⃗ . 𝑏⃗ ) + 2𝑡2 sin 𝜑 ∑ sin⁡(𝑘

+ [2𝑡2 cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗)] 𝐵 † 𝐵

The Aharonov-Bohm phase φ appears in H2 as well as the second neighbor hopping vectors 𝑏⃗. The
summation is over those vectors.
We can now write the matrix Hamiltonian resulting from this model as:

⃗ . 𝑏⃗) − sin 𝜑 ∑ sin⁡(𝑘

𝑀 + 2𝑡2 (cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗)) 𝑓(𝑘)
[ ]
𝑓(𝑘) ⃗ . 𝑏⃗) + sin 𝜑 ∑ sin⁡(𝑘
−𝑀 + 2𝑡2 (cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗))

𝑓(𝑘) is the usual graphene off-diagonal elements.

We linearize this Hamiltonian in the vicinity of the K and K’ valleys of graphene and we obtain
the following simpler massive Dirac form:
𝑚+ 𝑣 2 ℏ𝑣𝑘 −
𝐻𝐾 = [ ]
ℏ𝑣𝑘 + −𝑚+
𝑚− −ℏ𝑣𝑘 −
𝐻𝐾′ = [ ]
−ℏ𝑣𝑘 + −𝑚− 𝑣 2
3
With 𝑚± = 𝑀 ∓ 3√3𝑡2 sin 𝜑, and 𝑣 = 2 𝑎𝑡1 . a is the lattice constant.

Berry curvature for M=0

In the presence of inversion symmetry M=0 and 𝑚± = ∓3√3𝑡2 sin 𝜑 = ±𝑚𝐾 . We can compute
the Berry curvature for each valley starting with the eigenvector that we can obtain by solving the
above Hamiltonian. (𝑚𝐾′ = −𝑚𝐾 )
𝑚𝐾 𝑣 2 ℏ𝑣𝑘 −
𝐻𝐾 = [ ]
ℏ𝑣𝑘 + −𝑚𝐾 𝑣 2
𝑚𝐾′ −ℏ𝑣𝑘 +
𝐻𝐾′ = [ ]
−ℏ𝑣𝑘 − −𝑚𝐾′
For K and K’ we have the following eigenvector:
1 −ℏ𝑣𝑘 −
𝜓𝐾 = [ ]
2 𝑚𝐾 𝑣 2 − 𝐸(𝑘)
√(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 − 𝐸(𝑘))

1 ℏ𝑣𝑘 +
𝜓𝐾′ = [ ]
2 𝑚𝐾′ − 𝐸(𝑘)
√(ℏ𝑘𝑣)2 + (𝑚𝐾′ 𝑣 2 − 𝐸(𝑘))

𝐸(𝑘) = ±√(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2

All we need to do now determine the Berry curvature of the massive Dirac fermion
The Berry connection can be written as:
⃗⃗⃗⃗ ⃗ |𝜓𝐾 ⟩
𝐴𝑘 = 𝑖⟨𝜓𝐾 |∇
𝜕
We will work using cylindrical coordinates. It can be shown that ⟨𝜓𝐾 | 𝜕𝑘 |𝜓𝐾 ⟩ = 0, so that we
have:
𝜕
⃗⃗⃗⃗
𝐴𝑘 = 𝑖 ⟨𝜓𝐾 | |𝜓 ⟩ 𝜙̂ = 𝐴𝜙 𝜙̂
𝑘𝜕𝜙 𝐾
[−ℏ𝑣𝑘 + 𝑚𝐾 𝑣 2 − 𝐸(𝑘)]𝜕 −ℏ𝑣𝑘 −
𝐴𝜙 = 𝑖 [ 2 ]
(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 − 𝐸(𝑘)) 𝑘𝜕𝜙 𝑚𝐾 𝑣 − 𝐸(𝑘)
2

[−ℏ𝑣𝑘𝑒 𝑖𝜙 𝑚𝐾 𝑣 2 − 𝐸(𝑘)] 𝜕 −ℏ𝑣𝑘𝑒 −𝑖𝜙

𝐴𝜙 = 𝑖 2 [ 2 ]
(ℏ𝑘𝑣)2 + (𝑚 𝑣 2 − 𝐸(𝑘)) 𝑘𝜕𝜙 𝑚𝐾 𝑣 − 𝐸(𝑘)
𝐾

ℏ2 𝑣 2 𝑘
𝐴𝜙 = 2
(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 − 𝐸(𝑘))

The Berry curvature of Berry field can found by taking the curl of ⃗⃗⃗⃗
𝐴𝑘 :
1 𝜕(𝑘𝐴𝜙 )
𝐹= 𝑧̂
𝑘 𝜕𝑘
1 ℏ2 𝑚𝐾 𝑣 4
𝐹𝑧 = −
2 [(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]3/2

Hall conductivity
The Hall conductivity for the K-valley can now be computed using the TKNN index (see appendix
II for explanation of integration limits):

𝐾
𝑒2 0
𝜎𝑥𝑦 = ∫ 𝐹 𝑑𝑘 2
ℎ2𝜋 𝑎𝑙𝑙⁡𝑜𝑐𝑐𝑢𝑝𝑖𝑒𝑑⁡𝑠𝑡𝑎𝑡𝑒𝑠 𝑧

𝐾
𝑒2 0 1 ℏ2 𝑚𝐾 𝑣 4 𝑘𝑑𝑘𝑑𝜙
𝜎𝑥𝑦 = ∫ −
ℎ2𝜋 −∞ 2 [(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]3/2
0
𝐾
𝑒 2 2𝜋𝑚𝐾 𝑣 2 1
𝜎𝑥𝑦 = [ ]
ℎ2𝜋 2 [(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]1/2 −∞

𝐾
𝑒 2 𝑚𝐾 𝑒2
𝜎𝑥𝑦 = = 𝑠𝑖𝑔𝑛(𝑚𝐾 )
2ℎ |𝑚𝐾 | 2ℎ
For the K’valley, the Berry connection acquires a ‘minus’ sign due to k+ appearing in the
eigenvector:
ℏ2 𝑣 2 𝑘
𝐴𝜙 = − 2
(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 − 𝐸(𝑘))

The resulting Berry curvature is then:

1 ℏ2 𝑚𝐾 𝑣 4
𝐹𝑧 =
2 [(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]3/2

And the Hall conductivity is given by:

′ 𝑒 2 𝑚𝐾 ′ 𝑒2
𝐾
𝜎𝑥𝑦 =− = − 𝑠𝑖𝑔𝑛(𝑚𝐾′ )
2ℎ |𝑚𝐾′ | 2ℎ
The total Hall conductivity is therefore:
𝑒2
𝜎𝑥𝑦 = (𝑠𝑖𝑔𝑛(𝑚𝐾 ) − 𝑠𝑖𝑔𝑛(𝑚𝐾′ ))
2ℎ
For 𝑚𝐾′ = −𝑚𝐾 it is evident that we get
𝑒2
𝜎𝑥𝑦 =
ℎ
Phase diagram
Generally speaking we write:
𝑒2
𝜎𝑥𝑦 = (𝑠𝑖𝑔𝑛(𝑚+ ) − 𝑠𝑖𝑔𝑛(𝑚− ))
2ℎ
Thus the TKNN index for the quantized anomalous Hall state derived by Haldane is given by:
1
𝑛= (𝑠𝑖𝑔𝑛(𝑚+ ) − 𝑠𝑖𝑔𝑛(𝑚− ))
2
With,

𝑚± = 𝑀 ∓ 3√3𝑡2 sin 𝜑

As long as 𝑀 < |3√3𝑡2 sin 𝜑 |, 𝜎𝑥𝑦 is finite, since the sign of 𝑚± will be entirely determined by
that of sin 𝜑. For > |3√3𝑡2 sin 𝜑 | ,⁡𝜎𝑥𝑦 is zero. We therefore have the phase diagram shown in
Fig. 2 for 𝑀/𝑡2 versus 𝜑.
The reasons why this problem turns out to be complex in the case of the honeycomb lattice is
mainly related to the valley degeneracy that requires having masses of opposite sign for the K and
K’ valleys. We will see that the problem becomes simpler when we have a single Dirac valley at
the center of the Brillouin zone
Fig 2. Topological phase diagram of the Haldane model. n denotes the TKNN index appearing in
the Hall conductivity. When n is non-zero a QH effect exists at zero magnetic field. It is referred
to as the quantized anomalous Hall effect.
Observation of the quantum anomalous Hall effect
The quantum anomalous Hall effect (QAHE) was first reported in 2013 by C.Z.Chang et al. (from
the Chinese Academy of Science and Tsinghua University) and later on reproduced by several
other groups. It was successfully realized in the magnetic topological insulator Cr-Bi2-xSbxTe3.
Experimentally, it manifests as a remanent anomalous Hall resistivity that is quantized at ℎ/𝑒 2.
You can refer to the following publication to read about the experimental realization (C.Z. Chang
et al. Science 340 167 (2013)).
Appendix I: Deriving H2
From Fig. 1, we will now derive H2 given the following lattice vectors (I’m setting the lattice
constant a=1 in this section).
For A lattice sites we have:
𝑟1 : (1,0)⁡⁡⁡⁡⁡⁡⁡⁡⁡𝐵1 : −𝑟⃗⃗⃗1 : (−1,0)
𝐴1 : ⃗⃗⃗

1 √3 1 √3
𝐴2 : ⃗⃗⃗
𝑟2 : (− , )⁡⁡𝐵2 : −𝑟⃗⃗⃗2 : ( , − )
2 2 2 2

1 √3 1 √3
𝐴3 : ⃗⃗⃗
𝑟3 : (− , − )⁡⁡⁡⁡⁡𝐵3 : ⃗⃗⃗
𝑟3 : ( , )
2 2 2 2
We will not determine all possible second neighbor jumps with their associated Aharonov-Bohm
phase:

Starting from A1𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗2 −𝑟⃗⃗⃗⃗1 ) 𝑒 𝑖𝜑 𝐴† 𝐴 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗3 −𝑟⃗⃗⃗⃗1) 𝑒 −𝑖𝜑 𝐴† 𝐴: see Fig. 3(a)

Starting from A2: 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗1 −𝑟⃗⃗⃗⃗2 ) 𝑒 −𝑖𝜑 𝐴† 𝐴 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗3 −𝑟⃗⃗⃗⃗2 ) 𝑒 𝑖𝜑 𝐴† 𝐴⁡ see Fig. 3(b)
 ⃗ ⃗
Starting from A3: 𝑒 𝑖𝑘(𝑟⃗⃗⃗⃗1 −𝑟⃗⃗⃗⃗3 ) 𝑒 𝑖𝜑 𝐴† 𝐴 + 𝑒 𝑖𝑘(𝑟⃗⃗⃗⃗2 −𝑟⃗⃗⃗⃗3 ) 𝑒 −𝑖𝜑 𝐴† 𝐴 see Fig. 3(c)

Starting from B1: 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗2 +𝑟⃗⃗⃗⃗1 ) 𝑒 𝑖𝜑 𝐵 † 𝐵 + 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗3 +𝑟⃗⃗⃗⃗1 ) 𝑒 −𝑖𝜑 𝐵† 𝐵 see Fig. 3(d)

Starting from B2: 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗1 +𝑟⃗⃗⃗⃗2) 𝑒 −𝑖𝜑 𝐵 † 𝐵 + 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗3 +𝑟⃗⃗⃗⃗2 ) 𝑒 𝑖𝜑 𝐵† 𝐵⁡ see Fig. 3(e)

Starting from B3: 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗1 +𝑟⃗⃗⃗⃗3) 𝑒 𝑖𝜑 𝐵† 𝐵 + 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗2+𝑟⃗⃗⃗⃗3 ) 𝑒 −𝑖𝜑 𝐵† 𝐵 see Fig. 3(f)
In this notation, in the full phase that is picked out around a look is 3φ.

FIG 3. (a-c) Second neighbor hopping loops for A sublattice. Clockwise loop (CW) is shaded in
yellow and counterclockwise loop (CCW) is shaded in blue. Two purple arrows show the hop of
interest in each case for each A sublattice position (A1,A2 and A3). (e-f) Same but for B sublattice.
Putting together the terms derived above yields:

𝐻2𝐴 = 𝑡2 [⁡𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗2 −𝑟⃗⃗⃗⃗1 ) 𝑒 𝑖𝜑 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗3 −𝑟⃗⃗⃗⃗1 ) 𝑒 −𝑖𝜑 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗1 −𝑟⃗⃗⃗⃗2) 𝑒 −𝑖𝜑 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗3 −𝑟⃗⃗⃗⃗2 ) 𝑒 𝑖𝜑 + 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗1 −𝑟⃗⃗⃗⃗3) 𝑒 𝑖𝜑
+ 𝑒 𝑖𝑘⃗(𝑟⃗⃗⃗⃗2 −𝑟⃗⃗⃗⃗3 ) 𝑒 −𝑖𝜑 ] 𝐴† 𝐴

⃗ (𝑟⃗⃗⃗2 − ⃗⃗⃗
𝐻2𝐴 = 2𝑡2 𝐴† 𝐴(𝑐𝑜𝑠[𝑘 ⃗ (𝑟⃗⃗⃗3 − ⃗⃗⃗
𝑟1 ) + 𝜑] + 𝑐𝑜𝑠[𝑘 ⃗ (𝑟⃗⃗⃗1 − ⃗⃗⃗
𝑟2 ) + 𝜑] + 𝑐𝑜𝑠[𝑘 𝑟3 ) + 𝜑])

⃗ (𝑏⃗) + 𝜑] = 2𝑡2 𝐴† 𝐴 ∑(cos(𝑘

𝐻2𝐴 = 2𝑡2 𝐴† 𝐴 ∑ 𝑐𝑜𝑠[𝑘 ⃗ 𝑏⃗) cos 𝜑 − sin(𝑘
⃗ 𝑏⃗) sin 𝜑)
⃗
𝑏 ⃗
𝑏

Similarly,
 ⃗ ⃗ ⃗ ⃗ ⃗
𝐻2𝐵 = 𝑡2 [⁡𝑒 𝑖𝑘(−𝑟⃗⃗⃗⃗2 +𝑟⃗⃗⃗⃗1) 𝑒 +𝑖𝜑 + 𝑒 𝑖𝑘(−𝑟⃗⃗⃗⃗3 +𝑟⃗⃗⃗⃗1 ) 𝑒 −𝑖𝜑 + 𝑒 𝑖𝑘(−𝑟⃗⃗⃗⃗1 +𝑟⃗⃗⃗⃗2 ) 𝑒 −𝑖𝜑 + 𝑒 𝑖𝑘(−𝑟⃗⃗⃗⃗3 +𝑟⃗⃗⃗⃗2 ) 𝑒 𝑖𝜑 + 𝑒 𝑖𝑘(−𝑟⃗⃗⃗⃗1 +𝑟⃗⃗⃗⃗3 ) 𝑒 𝑖𝜑
+ 𝑒 𝑖𝑘⃗(−𝑟⃗⃗⃗⃗2 +𝑟⃗⃗⃗⃗3 ) 𝑒 −𝑖𝜑 ] 𝐵 † 𝐵

⃗ (𝑟⃗⃗⃗2 − ⃗⃗⃗
𝐻2𝐵 = 2𝑡2 𝐵 † 𝐵(𝑐𝑜𝑠[𝑘 ⃗ (𝑟⃗⃗⃗3 − ⃗⃗⃗
𝑟1 ) − 𝜑] + 𝑐𝑜𝑠[𝑘 ⃗ (𝑟⃗⃗⃗1 − ⃗⃗⃗
𝑟2 ) − 𝜑] + 𝑐𝑜𝑠[𝑘 𝑟3 ) − 𝜑])

⃗ (𝑏⃗) − 𝜑] = 2𝑡2 𝐴† 𝐴 ∑(cos(𝑘

𝐻2𝐵 = 2𝑡2 𝐵 † 𝐵 ∑ 𝑐𝑜𝑠[𝑘 ⃗ 𝑏⃗) cos 𝜑 + sin(𝑘
⃗ 𝑏⃗) sin 𝜑)
⃗
𝑏 ⃗
𝑏

We obtain the Haldane term by combining the Hamiltonian obtained for the A and B sublattices.

⃗ . 𝑏⃗) − sin 𝜑 ∑ sin⁡(𝑘

𝐻2 = 2𝑡2 [cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗)] 𝐴† 𝐴

⃗ . 𝑏⃗ ) + sin 𝜑 ∑ sin⁡(𝑘
+ [cos 𝜑 ∑ cos(𝑘 ⃗ . 𝑏⃗)] 𝐵 † 𝐵

Appendix II: Integration limits of the Hall conductivity

The Hall conductivity was determined by integrating the Berry curvature over ‘all filled bands’.
However, the form used to derive the eigenvectors is a low energy limit (for small k). The integral
limits used for 𝜎𝑥𝑦 between – ∞ and 0, are therefore inconsistent with our assumptions. A second
issue is the “– ∞” that cannot make physical sense in a lattice model. The more consistent solution
to this problem, is to solve the full lattice Hamiltonian and without making the linearizing
assumption. This is a bit tedious. A simpler approach to justify our choice of “– ∞”, is to actually
use a cut-off wavevector “– κ∞” as the bottom integration limit such that:

𝐾
𝑒2 0 1 ℏ2 𝑚𝐾 𝑣 4 𝑘𝑑𝑘𝑑𝜙
𝜎𝑥𝑦 = ∫ −
ℎ2𝜋 −κ∞ 2 [(ℏ𝑘𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]3/2

𝐾
𝑒 2 𝑚𝐾 𝑚𝐾 𝑣 2
𝜎𝑥𝑦 = [ − ]
2ℎ |𝑚𝐾 | [(ℏκ∞ 𝑣)2 + (𝑚𝐾 𝑣 2 )2 ]1/2

The integral imposes ℏκ∞ 𝑣 ≫ 𝑚𝐾 𝑣 2 to satisfy summation over “all filled states”. Accordingly,
the second term will be negligibly small, and will vanish.
References and further reading
F.D.M. Haldane, Phys. Rev. Lett. 61 2015 (1988)
S.Q. Shen “Topological Insulators: Dirac equation in condensed-matter.” Springer Series in Solid
State Science. 2nd Edition 2017.
B.A. Bernevig and T. Hughes “Topological Insulators and Topological Superconductors.”
Princeton University Press. 1st Edition 2013.
C.Z. Chang et al. Science 340 167 (2013)