Effect of impurities on ground state flux configuration

in Kitaev model
VSRP 2010 Project Report

Dushyant Barot
Aerospace Engineering Department,IIT Madras
Advisor: Vikram Tripathi
DTP,TIFR

July 5, 2010

Abstract

In this report an exactly soluble model of 12 -spins on a 2-d lattice called Kitaev
model is discussed. Using a theorem by Lieb[(4)] we have discussed whether
presence of impurities viz.absence of a link or site, has a finite flux in ground
state.Our work corrects the result stated in [(1)] that breaking an odd number of
links creates a finite flux by showing that it is neccessary to remove a site to do
so.For a row of rings with all common links removed, ground state energy is found
in the presence and absence of flux.

1 The Model

It is a model of spins on a hexagonal lattice. Since each cell in the lattice has an
even number of sites the lattice is bi-partite and hence we split it into two sub-lattices with
two types of sites : ‘a’ and ‘b’and hence three types of links as shown in fig.1a.
Kitaev model considers nearest neighbour interactions with the restriction that only α-
component of spin interacts along α-link.

1
 The hamiltonian then is as follows:

X X X
H = −Jx σjx σkx − Jy σjy σky − Jz σjz σkz
x−links y−links z−links

x y x y x y x y x y x y

z z z z z z z
y x y x y x y x y x y x

z z z z z z
x y x y x y x y x y x y

z z z z z z z
y x y x y x y x y x y x

Figure 1: Honeycomb Lattice

Let us define an operator for a plaquette p:

Wp = σ1x σ2y σ3z σ4x σ5y σ6z = K12 K23 K34 K45 K56 K61 ,

α α
where , Kjk = σj jk σk jk , αjk = x, y, z depending on the type of link between j and k
and σj ’s are the pauli sigma matrices.
(This convention for αjk will be followed throughout the report).
All Wp ’s commute with the Hamiltonian and with each other. Hence, Wp ’s and H can be
diagonalised simultaneously so that eigenstates of Wp ’s are also eigenstates of H. Hence,the
Hilbertspace can be split into sectors which are eigenspaces of Wp . It can be easily shown
that Wp2 = 1,hence eigenvalues of Wp ’s are wp = ± 1. Every sector corresponds to a
particular configuration of wp0 s for each plaquette. Hence, there are 2m different sectors.
wp = −1 can be called as a magnetic flux through the plaquette and wp = 1 can be called
zero flux through the plaquettes[Sec.6]. Thus,wp = −1 will alternately be called finite flux
through plaquette p, similarly for wp = 1.
Let m be number of plaquettes and n be number of vertices then n ≈ 2m
holds(Assuming a large lattice). Thus,there are 2n possible states but there are only 2m
different sectors ,so there is a degeneracy of 2m in each sector, hence splitting the Hilbert

2
space into sectors does not fully solve the problem .But ground-state spectrum depends
only on wp ’s as different sectors corresponding to a configuration of wp ’s can be converted
to one another by a gauge-transformation[(3)]. This fact will be used to diagonalise the
Hamlitonian in momentum space.

1.1 Fermionic hamiltonian

1.1.1 Spin representation using Majorana operators

†
ak +ak
Majorana operators can be defined as: c2k−1 = ak + a†k , c2k = i
, here ak and a†k are
creation and annihilation operators. These operators are equal to their conjugate transpose
hence called Majorana operators. Spins can be represented using four majorana operators:
∼x ∼y ∼z
bx , by , bz , c as : σ = ibx c,σ = iby c,σ = ibz c. Spins must satisfy the algebra satisfied
by pauli sigma matrices. This leads to the condition that the operator: D = ibx by bz c
must act as an identity operator in the eigenspace of spins[Sec. 6]. D commutes with the
Hamiltonian. Dj changes the sign of the eigenvalues for all umn where mor n is equal to j
. But the spectrum is left unchanged. Thus it acts like a gauge transformation.[Sec. 6]

1.1.2 Free-fermion Hamiltonian

The new Hamiltonian in terms of Majorana operators looks like:

∼
H = 4i Âjk cj ck , which corresponds to free fermions ,
P
j,k
where ,Âjk = 2Jαjk ûjk if j and k are connected, else Âjk = 0.
α α
Here, ûjk = ibj jk bk jk ,
Some properties of ûjk :

• ûjk =−ûkj ,

• û2jk = 1, hence its corresponding eigenvalues are : ujk = ±1.

• ûjk commutes with the Hamiltonian and all ûjk ’s commute with each other.

Because of the last property above the Hilbert space can be split into sectors each corre-
sponding to a particular configuration of ujk 0 s over the lattice. Restriction to a particular

3
sector gives the following Hamiltonian:

∼ iX
H = Ajk cj ck , (1)
4 j,k

1.2 Ground-state Spectrum and Phase-Diagram

1.2.1 Spectrum

It has been shown in (4) that zero flux ie wp = 1 in all plaquettes corresponds to
ground-state energy. Thus we can restrict the hamiltonian to this sector for calculating
ground-state energy and choose all u0jk s = 1.
The free-fermion Hamiltonian restricted to a sector can be written as:

∼ iX
Hu = Ajk cj ck , (2)
4
j,k

For a hamiltonian of this form it is shown in [(4)] that ground state spectrum is determined
by Aij ’s and is independent of c0j s.
In this sector there is translational symmetry in the lattice thus spectrum is determined via
a fourier transform. This is done by choosing the following basis for a unit cell as shown
in fig. 2a:
Hamiltonian can alternately be written as:

i X
H = Asλtµ csλ ctµ
4 s,t,λ,µ

where s, t are labels for the cells and λ, µ are labels for the type of sites in each cell.
rt → ~
Performing fourier transform and using the condition:~ rt + ~
rs leaves H unchanged
(This is a statement of translational symmetry), we get:

iX ∼ X ∼ ∼
⇒H = Aq,−q ( exp(i~
q .~
rt )) c q,λ c −q,µ
4 q,q 0 ,λ,µ t

Choosing a particular cell say,s = 0 we get a block-diagonal matrix with 2X2 blocks
corresponding to a particular q in pairs of ±q along the off-diagonal.(Unless there are zero

4
 n2
n1 fig. 2c

unit cell

Plane of symmetry
for the case of a site removed

fig. 2a

fig. 2b
fig. 2d
Hexagonal
Lattice
Plane of symmetry for
the case of link removal

energy modes). Thus, we have:

 
∼ 0 if (~
q)
iAq =  ,
∗
−if (~
q) 0
q
∼
f (~
q ) = 2(Jx exp(i~
q .~
n1 ) + Jy exp(i~ n2 ) + Jz ), Energy,q is given by: det(Aq )
q .~

1.2.2 Phase-Diagram

Broadly the spectrum splits into two phases:

• Gapped-phase- In this phase (q) 6= 0 hence all excitations from ground-state to

higher energy levels require finite energy. Spin correlations decay exponentially with
distance thus spatially separated quasiparticles do not interact directly[(1)].

• Gapless-phase: In this phase zero energy modes are present ie ε(q) = 0 for some
q = q∗ .

5
2 Effects of site-dilution on Ground-state Spectrum in the pres-
ence and absence of flux

Following theorem proved in [(4)] has been used for our analysis:

Theorem. Given a bi-partite lattice with a plane of symmetry the ground-state configuration
for loops containing the plane of symmetry: 0 flux for loops with 4n+2 sites and π flux for
loops with 4n sites.

2.1 Site dilution

Honeycomb lattice can be obtained from a square lattice by removing alternate links as
shown in fig. 2b. From the theorem proved above,hexagonal lattice with no sites removed has
6=2(mod4) sites hence, 0 flux is the lowest energy configuration.For the case of a site removed
from the lattice as shown in fig.2c, we get one plane of symmetry along which we have circuit
with 0(mod4) sites(fig. 2c),with lowest energy configuration having π flux through it. which
corresponds to wp = −1. Similarly, one can remove more sites maintaining reflection
symmetry and show that when odd number of links are removed then the circuit with the
vacancy(ies) is stable in a finite-flux state as stated in [(1)]

2.2 Link removal with no site removed

In this case we get a circuit containing the impurity with 2(mod4) sites(fig. 2d), hence
lower energy configuration is vortex-free[(4)]. Hence, it is not sufficient to remove an odd
number of links to create finite flux in the ground state configuration,as mentioned in [(1)]
,an odd number of sites must be removed to create finite flux in ground state.

2.3 A possible explanation

This can be explained because finite flux state corresponds to unpaired majorana modes
[(1)]. Cutting a link leaves even number of majorana operators,bi ’s unpaired which is
repairable. But removing odd number of sites leaves an odd number of majorana operators
unpaired which is irrepairable. Hence, this effect is seen only for the case odd number of
sites removed.

6
Note:Our analysis works only when there is a plane of symmetry. Reflection symmetry can
be broken due to site/link removal at multiple sites.

3 Spectrum for a row of rings with all common links removed

Consider 2n hexagonal rings in a row with interaction identical to Kitaev Model except
here number of links per site is 2 with odd number of links removed as shown in fig.3. The
P
hamiltonian can be written as: H = Aij ci cj
<i,j>
Such a system possesses translation symmetry in the sense that :

ci+2n = ci (3)

where N is the number of points. This allows us us to diagonalise the system in momentum
space.

1 3 5
n 2 4

Row of rings with odd number of sites removed

Fig. 3

3.1 Spectrum with no flux

In momentum space we have,

1 X ∼
cp = √ exp(ikm p) c km
2n m

We again have a bi-partite graph and following the convention as before we get:

i 1 X X ∼ ∼
H = ( (−1)q (δp,q+1 + δp+1,q )exp(i(km p + kn q))) c km c kn (4)
4 2n k ,k p,q
m n

7
 2n−1
i 1 X X ∼ ∼
⇒H = ( (−1)q δp,q+1 exp(i(km p + kn q))) c km c kn
2 2n k q=0
m ,kn

2n−1
i 1 X X ∼ ∼
⇒H = ( exp(i(km + kn + π)q)).exp(ikm ) c km c kn
2 2n k q=0
m ,kn

Using eqn. 4 we get:

mπ
exp(ikm .2n) = 1 ⇒ km = (5)
n
∼ ∼ ∼ ∼
collecting terms of the form: c km c kn and c kn c km and using the property of fermions:ci cj =
−cj ci we get,

2n−1
i 1 X X ∼ ∼
H = (exp(ikm ) − exp(ikn ))( exp(iαq)) c km c kn ,
2 2n k q=0
m ,kn

where α = km + kn + π.

i 1 X exp(i2nα) − 1 ∼ ∼
H = (exp(ikm ) − exp(ikn )))( ) c km c kn
2 2n k exp(iα) − i
m ,kn

From eqn. 6 it follows that Hamiltonian vanishes unless:

α = 0(mod2π) ⇒ km = −kn − π (6)

as the expression exp(i2nα) − 1 vanishes ∀km . Hence, using eqn. 7 we get:

i X ∼ ∼
H = (exp(ikm ) + exp(−ikm )) c km c kn
2k
m ,kn

2n−1
X ∼ ∼
H =i cos(km ) c km c kn (7)
m=0

where , km ’s are given by eqn.6 and km and kn are related by eqn.7

3.2 Spectrum with flux

Flux in this system can be obtained by flipping uij along a link say 1 → 2. But by a gauge-
transformation one can transform it to uij ’s being flipped on all alternating links starting

8
from 1. Following the analysis as above we get:

2n−1
X ∼ ∼
H = sin(km ) c km c kn (8)
m=0

km ’s are again given by eqn.6. km and kn are related as:

km + kn = 0(mod2π) (9)

3.3 Comparison of the two cases

Ground-state spectrum is given by sum of negative eigenvalues.

• Zero-flux case
Energy is given by:
n−1
2
X
Enof lux = 1 − 2 cos(km )
m=0

• Non-zero flux case

n−1
2
−1
X
Ef lux = −2 sin(km )
m=0

We find,
π π
Enof lux = −cosec( ), Ef lux = −cot( ),
2n 2n

Hence, finite-flux state is higher in energy than zero flux configuration.

4 Discussion

Plane of symmetry for the case of site-removal passes through the vertices unlike the
case considered in the proof[(4)] where plane of symmetry passes through the links. Results
are still assumed to be true as flux should be independent of the plane of symmetry chosen.
We have corrected the result stated in [(1)] that removal of odd number of links results
in finite flux in the ground state of lower energy as we have shown that lattice with a link
removed has lower energy in a flux-free state. This effect was verified for a row of rings with

9
odd-number of links removed,indicating the effects of link removal seen could possibly be
local.

5 Acknowledgement

I would like to thank Dr. Vikram Tripathi for guiding me through my project. I thank
Dr. Deepak Dhar ,Kusum Dhochak, R. Loganayagam and my VSRP colleagues for their
illuminating discussions and constructive criticisms. I would also like to thank Dr. Nilmani
Mathur and Dr. Roop Malik for coordinating VSRP 2010 and TIFR for supporting this
project and staff of Department of Theoretical Physics for all the help during the course of
this programme. Above all, I thank my parents and family for their constant support and
encouragement.

References

1. Kitaev A Yu 2006 Ann. Phys.321 2

2. E.H.Lieb 1992 Helv. Phys. Acta 65 247

3. E.H.Lieb and M. Loss 1993 Duke Math. J. 71 337

4. E.H.Lieb 1994 Phys.Rev. Lett. 73 16

6 Appendix

6.1 Gauge-operator Dj

The four majorana operators,bx , by , bz , c span a four-dimensional space though spins lie in a
two-dimensional space. So,we apply the following condition to restrict to the physical space
of spins:
D|ψ >= |ψ >, (10)

where ,D = ibx by bz c D commutes with the Hamiltonian.This condition physically is nec-

essary for spins to obey algebra which the pauli-sigma matrices obey: Dj changes the sign of

10
the eigenvalues for all umn where either mor n is equal to j leaving the spectrum unchanged.

[σ i , σ j ] = iεijk σ k

σiσj σk = i
∼x ∼y ∼z
For example ,[σ , σ ] = 2 ∗ bx by = −2 ∗ D ∗ ibx by bz c ∗ bx by = 2D σ , hence forcing
the above condition then gives the same algebra as for pauli-sigma matrices. We have

α α
Dj ûjk = ibxj byj bzj cj ibj jk bk jk

Let αjk be x , Hence, Dj ûjk = ibjx bjy bjz cj ibjx bkx = −ûjk Dj . Dj ûjk |ψ >= −ujk |ψ >,
Hence , Dj changes the sign of all umn if mor n is equal to j,leaving spectrum unchanged
hence it acts like a gauge transformation and ujk ’s form a Z2 gauge field.
Infact it leads to the interesting result that absolute ground-state energy is independent of
the sign of the constants Jα . eg. Jx → −Jx can be compensated by ujk → −ujk for all
x-links. But this can be compensated by a gauge-transformation on all x-links,so that Ajk ’s
do not change. This requires some cj → −cj which does not affect the spectrum.

6.2 Path-operator

n
!
Y
W (j0 , ......., jn ) = −iûjs js−1 cn c0
s=1

where s is a label for a site. Such an operator is gauge-invariant and can be interpreted
as transfer of fermion from j0 to jn . For j0 =jn it reduces to the flux operator defined
previously,this corresponds to fermion going in a loop around a plaquette. Thus, eigenvalues
of flux operators can be called a magnetic flux as they are in correspondence with the phase
an charged particle gains as it goes in a loop around a region. wp = −1 can be said to
correspond to a vortex∗ in a plaquette as wavefunction gets an overall minus sign as it
goes in a loop, which is analogous to Aharonov-Bohm effect where wavefunction catches
a phase as charged particle goes in a loop around a region of magnetic field.
Superconductors behave as diamagnetic material in the presence of a magnetic field but when
field strength is large it is energetically more favourable for the material to allow magnetic

11
field to penetrate regions in the form of thin lines. These regions are termed as vortices.

12