arX

iv:2

006.

1154

9v1

[co

nd-m

at.s

tr-e

l] 2

0 Ju

n 20

20

An introduction to Kitaev model-I

Saptarshi Mandal1, 2, ∗ and Arun M Jayannavar1,2, †

1Institute of Physics, P.O.: Sainik School, Bhubaneswar 751005, Odisha, India.2Homi Bhabha National Institute, Mumbai - 400 094, Maharashtra, India

(Dated: June 23, 2020)

This pedagogical article is aimed to the beginning graduate students interested in broad fieldof frustrated magnetism. We introduce and present some of the exact results obtained in Kitaevmodel. The Kitaev model embodies an unusual two spin interactions yet exactly solvable model intwo dimension. This exact solvability renders it to realize many emergent many body phenomenasuch as Z2 gauge field, spin liquid states, spin fractionalization, topological order exactly. First wepresent the exact solution of Kitaev model using Majorana fermionisation and elaborate in detailthe Z2 gauge structure. Following this we discuss exact calculation of magnetization, spin-spincorrelation function establishing its spin-liquid character. Spin fractionalization and de-confinementof Majorana fermion is explained in detail. Existence of long range multi-spin correlation functionand topological degeneracy are discussed to elucidate the entangled and topological nature of anyeigenstate. Some elementary questionnaires are provided in appropriate places for assimilation ofthe technical details.

PACS numbers: 75.10.Jm

I. INTRODUCTION

With the advent of time, discovaries of various newstates of matter1,2 continues to surprise and challengethe knowledge and understanding of mankind. For longtime we were familiar with solid, liquid and gaseousstates of matter. However, gradually we discoveredthat there are subtle differences in the microscopicorganizations within various solid or liquid or gaseousstates and this needs further classifications and char-acterizations. For example, a solid can be a crystal oramorphous, can be metal, insulator or semiconductor.In this article we are going to understand quantumspin liquid states3,4 which for quite some time puzzledthe present condensed matter community for its uniqueproperty and possible application for future technology.We would take some effort to understand the each termof the phrase "quantum spin liquid". In ordinary liquidstates like water we are concerned about the positionsand orientations of each water molecule with respectto other neighbouring water molecule. In fact it is therelative position and orientation of each water moleculewith respect to all other water molecule that defines agiven state like solid, liquid or gas. In this context somepreliminary idea like symmetry, long range order etc areimportant. Take the example of water in liquid statewhere the position of a water molecule is in constantchange with respect to any other water molecule thougha very short range correlation exist. Thus we say thatthere is no long range order in liquid states. On theother hand water in ice states has a long range order.Secondly if we look at microscopically around a givenwater molecule, the neighbouring environment or thepositions of surrounding water molecules looks identical.However this is completely different in solid materialwith crystal structure. We say that for liquid state, therotational symmetry is present and for solid with a given

crystal structure, the continuous rotational symmetryis broken and becomes discrete. This idea of longrange order and symmetry helps us to classify variousstates of matter5. Fundamental degrees that is relevantin this article are spins of molecule, atoms or evenelectrons. It may happen that the molecules or atomthat constitute a particular materials may constitutea perfect crystals or ordered solid phases of matter.In that case what we are interested is the collectivestates of individual spins of the atoms/molecules. Inother case it may happen that the spins of the valenceelectron that are roaming freely inside the materials.The collective states of such large collection of spins canshows diverse states having long range order or absenceof it. Ferromagnetic state or paramagnetic states areexample of that respectively. However there is a possi-bility of states of matter governed by the fundamentalproperty of quantum mechanics which makes it possibleto have states beyond simple ordered collinear statesof paramagnetic states. The superposition principleenables this miracles due to which the system can bein a given time, probabilistically, in many states. Forexample |Ψi〉 is a many-body spin state for the systemwhere spins at a given site ‘p’ is in a state |ip〉. It ispossible to have another many-body spin configurationgiven by state |Ψj〉 where spins at a given site ‘p’ isin a state |jp〉. |Ψi〉 is said to be different then |Ψj〉where |ip〉 6= |jp〉 for at least a given site ‘p’. One caneasily construct different non-equivalent many-body spinconfiguration. The fundamental postulate of quantummechanics says that it is possible for the system to bein state |Ψ〉 =

∑

i ai|Ψi〉 where∑

i |ai|2 = 16. Thepossibility that such construction is possible is at theheart of quantum spin-liquid states. If an experimentis done on such system to investigate the many-bodyspin configurations of the collective spins, one findsdifferent outcome at different times but with certain

2

probability. However all the |Ψi〉 that appears in theexpressions of |Ψ〉 are not arbitrarily chosen. They aregoverned by the microscopic Hamiltonian or mutualinteractions among the spins. There are specific relationsamong all the |Ψi〉 that appears in the expressions of|Ψ〉. Such relations can be of diverse nature and verydifferent from model to model or system to system.This particular aspect is known as entanglement. In thepresent study, in view of the Kitaev model, we wouldlike to explain some concepts and advanced idea of themodern condensed matter physics and give examplesof other system in comparison with Kitaev model.We begin by introducing Kitaev model in section II.Here we describe in detail the exact solution of Kitaevmodel, its Z2 gauge structure and ground state wavefunction explicitely. Various exercizes are also givenfor beggining graduate students to assimilate technicaldetails. Doing this would help appreciating the Kitaevmodel further. In section III, we evaluate various orderparameter such as magnetization, two-spin correlationfunction to establish the spin-liquid character of Kitaevmodel. Multispin correlation function and toplologicaldegeneracy is also explained in a simple way to show thelong range entanglement hidden in any eigenstate. Wepresent a brief recapitulation and discuss some recentdevelopment for further reading in section IV.

II. KITAEV MODEL

At zero temperature, a quantum mechanical systemminimizes its energy by occupying its ground states.Hamiltonian is the operator form of energy whose eigen-values and eigenstates determines all the property of thesystem in principle6. Lets consider a particular Hamilto-nian that is of our interest in this article, so called KitaevHamiltonian7. The system is defined on a honeycomblattice and at each site of this lattice there is a spin-1/2particle. The spin 1/2 particles interact with each otherin a specific way. The origin of such interaction is beyondthe scope of this pedagogical article and interested read-ers are requested to have a look at the recent progresses8.We will only discuss few elementary reasons and conse-quences of the origin of interactions which involves spin-magnetic moments9. Lets consider the situation wherewe bring in two hydrogen atoms close to each other froman initial position where they were at large separation.Hydrogen atom is made of one proton and one electron.The electron at ground state occupy the 1s state withorbital angular momentum L = 0 state. This happensbecause 1s state is spherically symmetric. However theintrinsic spin angular momentum of the electron couldbe quantized in +z or −z direction, this we representsas | ↑〉 and | ↓〉 respectively. When the two Hydrogenatoms are far apart the relative spin states of the twoelectrons can be shown to be arbitrary. However whenthe two hydrogen atoms come nearby then the electronic

orbital of two electron overlaps which means that electro-static potential energy increases. To minimise this elec-trostatic energy and also to maintain the two electronicwave function to be antisymmetric according to Pauliprinciple, one can show that the electronic spins forms asinglet, which can be shown to be an eigenstate of the

operator ~S1 · ~S2 where ~S1/2 represents the spin angularmoments of the electrons. In a nutshell, the origin ofmagnetic interaction is due to the interplay of minimiza-tion of electrostatic energy due to overlapping orbitals inconjunction with the antisymmetrization of wave func-tion according to Pauli principle. Extending this idea tocomplex condensed matter system composed of varietiesof atoms/molecules, different complex spin Hamiltoniancan emerge. The most well known of them is so called

Heisenberg Hamiltonian H =∑

<ij> Jij~Si · ~Sj . Here

< ij > denotes a pair of sites which could be nearestneighbour or next nearest neighbour. The sign of Jijcould be positive or negative depending on the relativeorbital properties of the participating atoms/molecules.With this prelude we describe the Hamiltonian that iscelebrated as Kitaev model. To understand that we firstnote that for a hexagonal lattice, there are three differ-ent kind of bond that can be grouped according to theiralignation as showin in Fig. 1. There are vertical bondsand lets call it z-bond. There are bonds with positiveslope and lets call it x-bond and the bonds which arehaving negative slope are termed as y-bonds. Now it isclear that a given site is connected to three other sites bythese three different type of bonds. The spin-spin inter-action between two neighbouring spins depends on thekind of bonds they are connected with. For example, agiven spin located at site ‘i’ interact with another spinlocated at site i + δx with Sx

i Sxi+δx

, where as the samespin interact with its neighbouring site located at sitei+ δy and i+ δz with Sy

i Syi+δy

and Szi S

zi+δz

respectively.

In the above Sαi , α = x, y, z denotes the spin compo-

nent of the spin angular moment. With this in mind theHamiltonian for the system can be written as follows,

H =∑

<ij>x

JxSxi S

xj +

∑

<ij>y

JySyi S

yj +

∑

<ij>z

JzSzi S

zj(1)

In the above equation the coupling parameterJx, Jy, Jz are taken positive. However the analysis we willfollow and the details of the property that we will discussremains unchanged for any combinations of signs of thecoupling strength. This is another remarkable character-istics of Kitaev model.

In this pedagogical article we try to give a few exactresults and exciting physics that this Hamiltonian con-tains. To get insight what does this innocently lookingHamiltonian as given in Eq. 1 contains, we take simplelimits and readers are asked to find the answers them-selves before proceeding further.

• Consider the spins as classical spins and find theclassical configurations of spins that minimizes the

3

1

23

4

56

z

y

x

z

y

x

p

zxy

Quantum Spin 1/2 at each site

B

A

PSfrag replacements

~a1

~a2

FIG. 1. In the left we have shown a cartoon picture of hon-eycomb lattice. Black and green filled circles shows two sublattices that honeycomb lattice is made of. The green, blackand red bonds respectively contain x-type, y-type and z-typeinteractions respectively. ‘~a1’ and ‘~a2’ are two basis vectors.In the lower right corner an elementary hexagonal plaquette isshown with a particular enumeration of sites which has beenused to define the conserved quantity Bp in Eq. 2.

Hamiltonian in the limits a) Jx = Jy = 0, Jz 6= 0,b) Jx 6= 0, Jy 6= 0, Jz = 0 and c) Jx 6= 0, Jy 6=0, Jz 6= 0.

• For more on classical solution the reader is re-quested to have a look at the reference by10.

A. Exact Solution of Kitaev model

The Hamiltonian is an effective representation of howthe constituent particles or spins in question interact andthere are energy cost for each specific interactions. Forinteracting particles, there are varying degrees of interac-tions associated with different energy cost. The Hamil-tonian is an operator form of energy and its spectrum orpossible values of energy that it can have gives a very im-portant physical insight of the system.Thus our primaryaim will be to find the eigenvalues and eigenfunction ofEq. 1. Once we have that, we can calculate any physi-cal quantity of our interest in principle. We are familiarwith eigenvalue problem of hydrogen atom, harmonic os-cillator, particle in a box etc11. Usually this involvessolving certain differential equations. For the Hamilto-nian in Eq. 1, individual spins are represented by Paulimatrices which has dimensions two. The individual stateof a given spin-1/2 particle can be represented by the twoeigenstates of the Pauli matrix. If there are ‘N ’ such spin-1/2 objects, the total dimensions becomes 2N and for thisproblem it becomes the dimension of Hilbert space whichis 2N . The Hamiltonian, in Eq. 1 can be representedsuitably in 2N × 2N dimensional hermitian matrix. Toobtain the eigenvalue and eigen function one has to di-agonalize this matrix. However given the modern daycomputer which is commonly available, one can solve asystem of 30-40 spins. The system with such small num-

ber of spins often misses to capture the long range orderand the true characteristics of the model developed onlyin the thermodynamic limit. For one dimensional sys-tem this might be enough. This fundamental limitationis the main challenge to modern condensed matter re-searcher. One commonly used technique to deal with thespin-Hamiltonian is that, we use suitable auxiliary vari-able to express the spins12. Many times we use fermioncor bosonic field operators to express the spin-1/2 angularmomentum algebra. This process is called fermionisationor bosonisation depending on the auxiliary field operatorthat are used. Below we provide two such very commonlyused fermionisation process implemented for one dimen-sional spin problem and reader is asked to implement itthemselves to understand the usefulness of it.

Exercise-1

• Jordan –Wigner Fermionisation and its applicationto transverse field Ising Model13,14, 1d X-Y model15

• Spin-1/2 one dimensional Kitaev model16

• Apply the Jordan-Wigner Fermionisation for 1dHeisenberg model and explain the differences withthe transverse field Ising model, 1d X-Y model or1d Kitaev model. Compare the results with the onedimensional antiferromagnetic Heisenberg chain17.

The Hamiltonian in Eq. 1 is special in the sensethat usually the spin-spin interaction between the spinshave all the component involved with the expressionSx1S

x2 + Sy

1Sy2 + Sz

1Sz2 . However for the Kitaev interac-

tion only one component of spins is involved with a givenspin. For this reason Kitaev interaction is also called ananisotropic Heisenberg interaction. Another key featureof this model is the presence of large number of conservedquantities. For each plaquette we can define an opera-tor which commutes with the Hamiltonian. We call thisplaquette operator as Bp where the subscript ‘p’ standsfor the plaquette index. Plaquette operators defined ondifferent plaquettes commute among themselves. Withreference to the Fig. 1 Bp is defined as,

Bp = σy1σ

z2σ

xxσ

y4σ

z5σ

x6 . (2)

Thus for each plaquette ‘p’ we can define a Bp. It can beeasily checked that,

[Bp, H ] = 0 , [Bp, Bq] = 0, p 6= q, (3)

In the above p, q indicate different plaquette indices. Thisimplies that Bp’s are conserved quantities for this model.It is easy to verify that B2

p = 1 which implies that eigen-values of Bp are ±1. We will see later that this conservedquantity plays a significant role in the dynamics of Kitaevmodel. We now present the formal solution of this spinmodel as obtained by Kitaev himself7. He showed thatthis spin model can be solved exactly using a fermioni-sation procedure which expresses the spin 1/2 operatorsin terms of Majorana fermion operators. In the next weelaborate on this.

4

B. Fermionisation of spin 1/2 operators

Apart from the Fermionisation procedure mentioned inthe exercises in Sec. II B, here we outline another fermion-isation procedure which was used by Kitaev himself. Weagain ask the reader to do the following exercize beforeproceeding further.

Exercise-2

• Define the operator, c = (c1 + c†1), cx = 1

i (c1 −c†1), c

y = (c2 + c†2), cz = 1

i (c2 − c†2). Verify that

they satisfy usual anticommutation relation, theyare self conjugate and when multiplied with itselfgives unity.

• Define σx = icxc, σy = icyc, σz = iczc. Verifythat Pauli matrices satisfy usual commutation rela-tion.

• Check whether the above definition reproducesσxσyσz = ±i for the odd particle states and evenparticle states respectively.

• Find the physical Hilbert space corresponding to thedefinition. Remember that a spin-1/2 particle has amatrix representation by Pauli matrices and has aHilbert space dimension two. However in the aux-iliary variable when we express the spin variable byfermionic operators, we used two fermions. Noweach fermions has a Hilbert space dimension two,thus two fermions together has Hilbert space dimen-sion of four. Thus the initial physical Hilbert spaceof a given spin of dimension two has been mappedto a Hilbert space dimension of four. This causesenlargement of Hilbert space dimensions and ac-counts for unphysical states. The previous exerciseis aimed at having an idea about sub-classificationof this extended Hilbert space where σxσyσz = ±ineeds to be worked out.

C. Quadratic Hamiltonian

To proceed we write the Hamiltonian in terms offermionic operators as discussed above. After insert-ing the relations given in Exercise-2 in Eq. 1(i.e σα

i =icαi ci, α = x, y, z), it reduces to

H =∑

x−link

Jx(icxi,ac

xj,b)ici,acj,b +

∑

y−link

Jy(icyi,ac

yj,b)ici,acj,b

+∑

z−link

Jz(iczi,ac

zj,b)ici,acj,b (4)

In the above equation ‘a’ and ‘b’ denotes the sublatticeindex. We observe that each term in the above Hamilto-nian is quatric in Majorana fermion operators. Generally

such quatric Hamiltonian is quite difficult to solve. How-ever it can be easily noted that operators in the paren-thesis of each term of the above Hamiltonian commutewith the Hamiltonian and commute among themselves.Which means they follow the following commutation re-lation which the interested readers are requested to checkbefore proceeding further.

[icαi,acαj,b, ic

βi,ac

βj,b] = 0, [icαi,ac

αj,b, ici,acj,b] = 0 (5)

In the above equation α, β = x, y, z and (i, j) denotesa pair of nearest neighbour sites joined by a α-type bond.It means that they are conserved quantities as far as thisfermionised Hamiltonian is concerned. This fact makesEq. 4 to be effectively quadratic in Majorana fermions.Let us call, icxi,ac

xj,b = uxi,j for the x-link. Similarly we

define uyi,j and uzi,j on y and z links respectively. It isobvious that, uαi,j = −uαj,i and its eigenvalues can take

value ±1 given the fact (uαi,j)2 = 1. Here we follow the

convention of keeping the indices of the site belonging tothe ‘a’ sub lattice first and then for ‘b’ sub lattice in theexpression of ui,j(fro brevity many times explicit mentionof α in the expression of uαij will be avoided.).

Exercise-3

• Using the definition of fermionization in Exercise-2and definition of Bp as in Eq. 2 show that

Bp =∏

(j,k)ǫboundary(p)

uαj,k. (6)

Using the above replacements for the conserved quan-tities on each bond, the Hamiltonian takes the followingform,

H =∑

<ij>x

Jxuxi,jici,acj,b +

∑

<ij>y

Jyuyi,jici,acj,b

+∑

<ij>z

Jzuzi,jici,acj,b (7)

= Ψ†H([u])Ψ (8)

where Ψ† = (c1, c2, ..., cN ) is an one dimensional row ma-trix of length ‘N ’ and H[u] is a N ×N matrix which de-pends on the values of uij on each link. Now we see thatabove Hamiltonian describes a tight binding Majoranafermion hopping interactions but the hopping matrix el-ements are coupled with conserved operator or fields uαi,jon each bonds. As we mentioned that this operatorshave eigenvalues ±1. These uαi,j are called Z2 gaugefields because of the eigenvalue to be ±. Dependingupon the values of these Z2 gauge fields the eigenvaluesof the system will change. Physically one way to visu-alise is that initially we had quartic Majorana fermioninteraction of the form icαi,ac

αj,bici,acj,b. This process can

be visualise in many ways. For example this can be

5

thought as two Majorana fermion hopping process hap-pening simultaneously such as cαi,a → cαj,b, ci,a → cj,bor cαα,i,a → cj,b, ci,a → cαj,b or similar processes. Alter-natively it can be thought of as two on site interactionshappening simultaneously yielding an energy cost. Thison sight interaction involves the Majorana fermions ata given site only. All possible equivalent explanationexists for such four body interactions and they all aretrue. However the fact that the combination icαi,a → cαj,bis a conserve quantity signify that once eigenvalue oficαi,a → cαj,b is fixed to either +1 or -1 (like an initial

condition) it remains same with time. Lets assume thatthe eigenvalue is fixed at +1(or -1) then the hopping pro-cess ci,a → cj,b with an phase +1( or -1). This also signifythat the process cαi,a → cj,b, ci,a → cαj,b or other possi-ble processes are not allowed by the system. Further inEq. 8, if we transform ci,a → λici,a, we observe that tokeep the Hamiltonian invariant under such transforma-tion uαi,j must change according to uαi,j → λiu

αi,jλi, where

the allowed value of λi is again ±1. THus the Hamilto-nian given in Eq. 8 has a underlying Z2 symmetry18.

Exercise-4

• Derive the Hamiltonian as in Eq. 8

• Prove that [H,uαij ] = 0, [uαij , uβkl], α, β = x, y, z

• Derive the Heisenberg equation of motion for theoperator ici,ac

αj,b and icαi,ac

ai and try to explain it

with your own understanding.

It is to be noted that the new conserved quantities,ui,j , were absent in the original spin Hamiltonian. Inthe original spin Hamiltonian as given in Eq. 1, there isno conserve quantity associated with a bonds. Earlierwe found that there was only one type of conservequantity associated with each hexagonal plaquette onlyand it is given in Eq. 2. Also from exercise 3 (Eq. 6),we found that the Bp is a product of six conservedquantities (ui,j ’s) defined on each bond. Initially ourphysical degrees were the spins represented by a productof two Majorana fermions. The bond conserved quantityui,j is also product of two Majorana fermions but eachMajorana fermion is taken from the two spins attachedat the end of a particular bond. And one can verifythat a uij can not be expressed in terms of spins. Itis the product of uij over the links of a hexagonalplaquette which can be expressed as a product of sixspin operators and called Bp. One can easily verify thatthe square of Bp is one indicating that the eigenvalueof the operator Bp is ±1. We have also seen thateigenvalue of the bond-conserve quantity uij is also±1. As Bp is expressed as a product of six uij we canunderstand that for a given eigenvalue of Bp there aremany choices for the eigenvalues of the participatinguij ’s. Among the total 26 configurations that six uijprovides, half of them yields Bp = 1 and other halfyields −1. For a given eigenvalue of Bp say +1, among

the various combinations, we will observe that a givenuij changes its eigenvalue from +1 to -1 though valueof Bp is fixed to 1. For this reason uijs are called gaugefields analogous to magnetic vector potential and Bp’sare physical observable analogous to Magnetic field.Physically uij can not be measured and its expectationvalue or outcome in experiment will be zero because ofgauge averaging over many combinations that one uijtakes. However eigenvalue of Bp is a physical observable.Thus we will use the phrase that Bp is gauge invariantand uij are not gauge invariant.

Technicality

Before proceeding further we elaborate on the Hamil-tonian represented in Eq. 8. The motivation is thatphysics is not always a story telling. Many truth of agiven physical system can be understood by an intelli-gent mind without going into the details of mathematics.But very often, the situations become so complex for amany body system that one needs to rely on mathemat-ics. In Kitaev model such mathematics become exactbeautifully and this motivates us to explain in detail theconsequences of mapping of original spin problem into aMajorana fermion hopping problem coupled with Z2 con-served gauge field. Such an exact description may hap-pen for other system approximately and we believe thata thorough understanding of it in the context of Kitaevmodel will help the reader to extend their imaginationeasily to explore the intricacies of other complex system.Lets consider a finite Honeycomb lattice such that it hasN1 dimer or z-bond in the ~a1 direction and N2 dimer orz-bonds in ~a2 direction. Thus we have a system of Nspins with N = 2N1N2. Because there are two spins ina given z-bonds. Total number of bonds of a particulartype say x, y, z are equal and they are N1N2. Thuswe have a total number of bonds Nb = 3N1N2 = 3N/2.The original spin Hamiltonian of N spins is defined in2N = M dimensional Hilbert space. This means thatwe are to solve a M × M matrix with M real eigen-values and eigenstates. Now while we had employed afermionization procedure where at each site there aretwo fermions implying that each site is associated witha Hilbert space dimension of four yielding total Hilbertspace of 4N = (22)N = 2N2N = M1 = 2M which meanswe have now enlarged Hilbert space yielding M1 numberof eigenstates/eigenvalue which is much more than theoriginal Hilbert space. Thus though the original problemin spin-space was dimension 2N , after fermionisation nowwe have a problem of dimension 2N × 2N . This situationis depicted in Fig. 2.

Let us try to find another connection which will explainthe relations between the eigenstates of the original spinproblem given by Eq. 1 or Eq. 8 and Majorana fermionhopping problem given by Eq. 4. In the spin space we

6

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

����������������������������

����������������������������

����������������������������

����������������������������

���������������������

���������������������

���������������������

���������������������

���������������������

���������������������

���������������������

���������������������

1

2

3

Majorana Fermionisation

Original spin system N spinof

Hilbert space dimension

PSfrag replacements

2N2N

2N

2N

2N

2N

2N2N

2N

2N

FIG. 2. The big circle at the top denotes the original Hilbertspace dimension of 2N . At the lower the 2N copies of the orig-inal Hilbert space is shown. This happens due to Majoranafermionisation using four Majorana fermions at a given site.The description of Kitaev model in each of these 2N copies areidentical up to an gauge transformations which would connectone gauge copy to another.

PSfrag replacements N ×N

N ×N

N ×N

N ×N

FIG. 3. For a system of ‘N ’ sites we have a matrix of dimen-sion N × N for a given realisation of uij on each link. Suchmatrix is shown by the small coloured matrix. For a givenN sites, there are 3N/2 bonds yielding a total possibility of

23N/2 such configuration. Thus the dimension of the outerbig matrix is 23N/2 × 23N/2. For more detail on the countingon the Majorana fermion and its relation to complex fermionssee text below.

must have M = 2N number of eigenstates and eigenval-ues. How do we obtain that from Eq. 8 and what is theconsequences of enlargement of Hilbert space dimension.This will be explained here. For a given distributionof gauge fields on the bonds of honeycomb lattice, wehave a matrix of dimension N which has N eigenvaluesand N single particle Majorana fermion eigenstates. Ifone diagonalizes H[u] in Eq. 8, we obtain λ[u]i, i = 1, Nwith eigenvectors |λ[u]〉i =

∑

i γ[u]ici. However theseMajorana fermions are to be regrouped to yield N/2complex fermions which have a definite occupationnumber representation19. From this N/2 single particleeigenstates one can obtain 2N/2 many-body statesby constructing eigenstates having arbitrary particlenumber. The energy eigenvalues for such many bodystates can be written as

∑

i λ[u]i. One can easily note

that there are 2N/2 such many body eigenstate which

can be obtained by∑

m

(N2

m

)

. However we rememberthat such description is true for a given distribution ofconserved quantities on every bonds. This situation isdepicted in Fig. 3 where the dimension of the outer big

matrix is N2 × 3N

2 = 3N2

4 . Inside this big matrix thesmaller diagonal matrix represent a certain distributionof conserved gauge field and the dimension of thissmaller matrix is N yielding N/2 complex single particlecomplex fermionic eigenstate which in turn yield 2N/2

many body state. Now we already mentioned thereare in total Nb = 3N/2 number of bonds yielding intotal 23N/2 such combinations. This means, in referenceto Fig. 3, we have 23N/2 such small diagonal matrix.Each of this smaller matrix yields an eigenstates of2N/2 number. Thus the total number of eigenstatesis 2N/223N/2 = 2N2N which matches with our earliercounting.

Earlier we mentioned that uij ’s are not gauge invari-ant quantities and they are not physical observable. Thegauge invariant physical observables are the plaquetteconserve quantity Bp. There are many combinations ofuij which yields a configurations of Bp for each plaquette.And it is important to note that the energy eigenvalueonly depends on the distribution of Bp not uij . If twoconfigurations of uij yields same distributions of Bp, theyshould have identical eigenvalues. In reference to Fig. 3,this means that there are many gauge equivalent smallermatrix which gives identical distributions of Bp and haveidentical energy eigenvalues. We leave it to the interestedreader to check themselves with the following exercise.

Exercise-5

• Calculate the number of ways in which one can haveBp = 1 for all the plaquette. Is this number samefor having a random configuration of Bp for all theplaquette.

• In Fig. 4, A and B we have given two configura-tions for Bp=1 for all the plaquette. Following theprocedures of Ref.7, Sec-5, construct the eigenval-ues and check that they yields identical spectrum.Construct similar distributions of gauge field suchthat it gives Bp = −1 for all the plaquette. Cal-culate the spectrum as done for BP = 1 and findwhich configuration has minimum ground state en-ergy. Explain your finding.

Lieb Theorem

We observe from Eq. 8 that Hamiltonian is functionalof configurations of conserve quantity uij defined on eachbonds. Such Hamiltonian can be represented by block di-agonal form in the eigenstate of uij as represented by thesmall square matrix in Fig. 3. Each sub-block refer to

7

a certain distribution of uij and we also explained be-fore that each block corresponds to a certain distributionof gauge invariant conserve quantity Bp for each plaque-tte. There are many distinct configurations of uij whichyields a unique configurations of Bp. We also know thateigenvalue of Bp is ±1. Now the important question isthe following. Does the ground state obtained from eachsub-square block in Fig. 3 yields same energy or doesit depends on the distribution of uij or does it dependson the distribution of Bp. As the uij is not a gaugeinvariant object and physically not observable, the en-ergy is not directly dependent on the distribution of uijrather it depends on the distribution of Bp. The nextquestion is which distribution of Bp yields the absoluteminima. From a very remarkable theorem20 by E. Lieb.we know that uniform configuration of Bp = 1 for eachhexagonal plaquette yields the global minima. This isobtained by fixing ui,j = 1 for every link (there are manyother configurations which yields Bp = 1, however eachof them would yield identical ground state energy). Thishas also been confirmed by Kitaev numerically7. For theuniform choices of ui,j(which corresponds to global min-ima ) we can easily diagonalise the Hamiltonian and getthe ground state wave function.

Physical wave function and projection

operator

With the discussion of the foregoing paragraph, let uscall the wave function obtained by diagonalizing Eq. 8with uniform configuration of uij = 1(yielding Bp = 1for each hexagonal plaquette) as |ψ〉ext. We have deliber-ately added the subscript ‘ext’ to remind the fact that theabove wave function is obtained in the extended Hilbertspace. One may question whether the ground state en-ergy obtained in such a way is the true ground stateenergy which should have been obtained in the physicalHilbert space. Moreover what about the wave functionit self? Because to calculate other physical observablewe need the true ground state belonging to the physicalHilbert space. Otherwise |ψ〉ext is not of much useful.Now it is time to discuss this issue. Whenever we makean operation which takes us from an Hilbert space H>

with more number of states to another Hilbert space H<

with less number of states such that some states are ex-cluded in H<, we need an projection operator. If anystates Ψ> belongs to H>, the corresponding states inH< after projection is obtained as Ψ< = PΨ> where Pis the projection operator. The job of the projection op-erator is to remove the unphysical states and keep onlythe physical state in the expansion of Ψ> = γi|i〉 whereγi is complex coefficient and |i〉 is the normalized basisvector belonging to H>. Actually |i〉 can be decomposedinto two groups |i<〉 and |i>〉 such that |i<〉 constitutethe normalized basis vector of H<. The job of P is toremove or annihilate the states |i>〉 such that PΨ> in-volves only the states belonging to H<.

|ψ〉phy = P |ψ〉ext (9)

PSfrag replacements A B

C D

++

+

+ ++

++

+++

+

+++

++

+ +++

+++

+

+

++ ++

+− −

−−

−−

−

1

11

1

11

1

11

1

11

2

2

2

2

2

2

2

2

3

3

3

3

3

3

3

3

4 4

4

4 4

4

4 4

4

4 4

4

5555

5555

66

66

77

77

8888

8888

FIG. 4. In the panel A and B, two choices of conserved Z2

gauge fields are shown to yield identical distribution of Bp.For C and D two equivalent Z2 gauge field choices are to befound to yield Bp = −1 for all the plaquette.

Now we try to understand the projection operator andits expression in the context of Majorana fermionisa-tion in a little depth and we find a useful connectionamong the various copies of the Hilbert space that wementioned earlier. Lets | ↑〉 and | ↓〉 be the eigenstatesof σz with eigenvalue ±1. The action of σx and σy onthese two states are defined as σx| ↑ (↓)〉 = | ↓ (↑)〉 andσy| ↑ (↓)〉 = i(−i)| ↓ (↑)〉. Now the complex fermionci, c

xi,a, c

yi,a, c

zi,a that has been used to define the Ma-

jorana fermionisation of spin operators has the states

|00〉, |10〉 = c†1|00〉, |01〉 = c†2|00〉, |11〉 = c†1c†2|00〉.

It is clear that the original spin states needs to bemapped to these four states and there is an enlargementof states. To understand the mapping let us recall thatD = σxσyσz = i is an identity which must be hold for anystates. If we calculateD according to the definition given

in Exercise-2 one finds D = i(1− 2c†1c1)(1− 2c†2c2). Nowwe see that D = i holds true only for the states |00〉 and|11〉. Thus any physical states should have the generalrepresentation Ψph = a00|00〉+ a11|11〉. But while work-ing on the extended Hilbert space we encounter statesΨext = a00|00〉+ a11|11〉+ a01|01〉+ a10|10〉. How do weget rid of the unphysical states |01〉 and |10〉. Note thatthe operator P = (1 + D)/2 acting on this unphysicalstates yields zero and keeps the physical states as it is.Thus it is straightforward to check that PΨext = Ψph.Now this is about a given site and we call this projectionoperator Pi = (1 + Di)/2. The above explanation canbe extended to all the sites of the entire system and thetotal projection operator is defined as,

P =∏

iǫall sites

(1 +Di)

2(10)

Before, we go for a formal solution we advise the readerto go through the following exercise which will strengthentheir familiarity with the concept of Majorana fermioni-

8

sation.

Exercise-6

• Find the eigenstates of the operator c, cx, cy, cz de-fined in Exercise 2.

• Find the eigenstates of the operator σx, σy, σz inthe Fock space of four states described above and seethe consequences of projection operators on them.

D. The ground state

We have already argued that it is the uniform config-uration of Bp = 1 which contains the global minima ofthe spectrum. Here we consider the choice uij =1 foreach link which is one of the realizations of Bp = 1 foreach plaquette. After doing that the Majorana fermionhopping Hamiltonian given in Eq. 1 reduces to a trans-lational invariant Hamiltonian facilitating easy solutionusing Fourier transformations. The translational invari-ant Hamiltonian is given by,

H=∑

x−link

Jxici,acj,b +∑

y−link

Jyici,acj,b +

∑

z−link

Jzici,acj,b (11)

To solve the above Hamiltonian we define the followingFourier transformations for the Majorana fermions,

ci,a(b) =∑

k

1√MN

ei~k.~rck,a(b). (12)

Here we have taken a lattice with M and N unit cellsin the directions of ~a1 and ~a2 respectively as shown in

Fig. 1. Here ~r = m~a1 + n~a2 and ~k = pM~b1 +

qN~b2, where

~b1,2 are the reciprocal lattice vectors are given by,

~b1 =4π√3(

√3

2ex +

1

2ey) ; ~b2 =

4π√3ey (13)

Here ‘p’ and ‘q’ varies from −M/2 to M/2 and −N/2 toN/2 respectively. The above discussion defines the Bril-

louin zone. We notice that the property c†i = ci implies

ck = c†−k. After performing the Fourier transformation,we get the Hamiltonian in k-space as follows,

H =∑

kǫHBZ

(c†k,ac†k,b)

(

0 if∗k

−ifk 0

)(

ck,ack,b

)

(14)

In the above equation ‘HBZ’ stands for half Brillouin

zone. Note that for the condition ck = c†−k, all the Ma-jorana modes are not independent. Their is a specificway a Majorana fermion with positive momentum is re-lated with a Majorana fermion with a negative momen-

tum with the same magnitude. Equivalently the ck = c†−k

FIG. 5. In the above we have plotted the first Brillouin zone

for the honeycomb lattice ~b1 and ~b2 denotes the reciprocallattice vector as defined in Eq. 13.

implies that annihilating a Majorana fermion with mo-mentum k is same as creating a Majorana fermion withmomentum −k. The spectral function fk is given by,

fk = Jz + Jxe−ik1 + Jye

−ik2 (15)

In above expression ‘k1’ and ‘k2’ are the components of ~kalong ‘x’ bond and ‘y’ bond respectively. They are givenby,

k1 = ~k.n1 ; k2 = ~k.n2 (16)

n1 =1

2ex +

√3

2ey ; n2 =

−12ex +

√3

2ey (17)

Here n1 and n2 are the unit vector along the ‘x’ and ‘y’bond respectively. The Hamiltonian given in Eq. 14 canbe diagonalised easily with the following unitary trans-formation given below,

(

ck,ack,b

)

=1√2

(

vk −vk1 1

)(

ηkξk

)

, (18)

with vk = if∗k/|fk|. The diagonalised Hamiltonian is

given by,

H =∑

k

Ek(η†kηk − ξ

†kξk), (19)

where Ek = |fk| is the quasi particle energy associatedwith new field operators αk and βk. The ground state isobtained by filling up all the negative energy states of βkquasi particles and can be written as,

|G〉 = Πk,HBZβ†k|0〉, (20)

where |0〉 represents the vacuum state such thatαk|0〉 = βk|0〉 = 0. Here the summation is over the halfBrillouin zone as explained before. At this point it isimportant whether the spectrum is gapped or not whichimplies that if we want to create an excitation over theground state do we require finite energy or not. If Ek = 0

9

for some values of ‘k’ we need no finite energy to createan excitation over the ground state and the system iscalled gapless system. Whether a system is gapless ornot has much bearing to the thermodynamic quantitiesof the system as it determines how the one part of thesystem responses due to disturbances at some other part.To find whether the spectrum is gapless or not we solvefor Ek = 0 which implies fk = 0. It turns out that the,fk = 0 has solutions if and only if |Jx|, |Jy|, |Jz | satisfythe following triangle inequalities:

|Jx| ≤ |Jy|+ |Jz|, |Jy| ≤ |Jx|+ |Jz|, |Jz| ≤ |Jx|+ |Jy|(21)

Phase Diagram

B

A

AA

Gapless

Phase

Gapped

Phase

PSfrag replacements

Jx = 0Jy = 0

Jz = 0

FIG. 6. Phase diagram for Kitaev model in the parame-ter space. A point in the above triangle describes relativemagnitudes of Jx, Jy , Jz. Three sides of the triangle describeJx = 0, Jy = 0 and Jz = 0 as given in the figure. The region‘A’ is gapped and the region ‘B’ is gapless. The gapless regionacquires a gap in the presence of Magnetic field.

The above inequalities as given in Eq. 21 can be repre-sented as a point inside a equilateral triangle which hasbeen shown in Fig. 6.

In Fig. 7 and Fig. 8, we plotted how the spectrum lookslike for gapless and gapped phase respectively.

If the inequalities are strict, there are exactly two so-lutions: k = ±q∗, one in each HBZ . The region de-fined by inequalities in Eq. 21 is the shaded region B inFig. 6; this phase is gapless. The region marked by Ais gapped. The low energy excitations are different inthese two phases. In the gapless phase the low energyexcitations are the Majorana fermions but in the gappedphases the low energy excitations corresponds to the ver-tex excitations which corresponds to the excitations ofBp, the conserved quantities. In the presence of mag-netic field the phase B acquires a gap. These two regionsare topologically distinct as indicated by spectral Chernnumber which is zero for phase A and one for the phaseB7. We have argued that as the projection operator Pcommutes with the Hamiltonian, the solution obtained

PSfrag replacements

-2-2

Ek

2

2

kx

ky

-5

0

0 5

FIG. 7. Spectrum Ek as defined in Eq. 19 has been plottedabove for Jx = Jy = Jz = 1. We observe that there are sixpoints in the Brillouine zone where Ek vanishes resulting agapless spectrum. The dispersion near this gapless points arealso linear.

PSfrag replacements

-4

Ek

2

2

kx

ky

-2

-2

0

0

4-5 5

FIG. 8. In the above Ek has been plotted for Jx = Jy =1, Jz = 2.2. We note that there is a gap between valenceband and conduction band. The spectrum near the minimumfor conduction band is quadratic.

in the extended Hilbert space is exact. One can indeedshow that there exist a non zero projections. But thismethod of solving gives eigenstates of the Hamiltonianas well as the eigenstates of the conserved quantities interms of Majorana fermions whose occupation number isnot well defined. In the next section, we would extendthe Majorana fermionisation in an useful way such thateigenfunctions of the Hamiltonian as well as uαij is repre-sented by the usual occupation number representation ofcomplex fermions19.

III. ORDER PARAMETER

We have seen that the initial spin-Hamiltonian givenin Eq. 1 is reduced to an effective quadratic fermionicHamiltonian as given in Eq. 4. The initial physical ob-ject was spin 1/2 magnetic moment. For a two spinsbelonging to two different sites, spin-angular momentumcommute with each other if we express them by suitablerepresentation. For spin-1/2 particle, the Pauli matri-ces are a faithful representations. However, the effectiveFermionic Hamiltonian consists of Majorana fermionswhich aniticommutes. This conversion of effective de-grees of freedom is known as emergent degrees of freedomdue to interactions. However there is a definite connec-tions between them. The eigenstates either obtained by

10

diagonalizing the original spin-Hamiltonian or the effec-tive Majorana Hamiltonian has one to one correspon-dence and they have identical eigenvalue spectrum. Indealing with physical systems, generally one is interestedwith the ground state wave functions at low temperature.Now we must ask what characterizes the ground states,what is the properties of the ground states of differentphases, that will differentiate one phases from another.various order parameters are used to determine a certainphases and distinguish from other. The magnetizationor spin-spin correlation between the effective degrees offreedom is such a measure. In many occasion such cor-relation functions can only be calculated approximately.However the beauty of Kitaev model renders us to com-pute the correlation functions exactly21. Mathematicallythe correlation between two observables or operators O1

and O2 is expressed as < O1O2 > where < ... > denotesground state expectation value at zero temperature6. Atfinite temperature it means a thermal average. Physi-cally it means what is joint probability that if the oper-ator O1 takes value O1 and the operator O2 takes valueO2. Apart from correlation function magnetization isalso used to characterizes phases of magnetic and inter-acting spin systems. Magnetization is defined as the av-erage value of magnetic moment in a system and at lowtemperature for quantum mechanical system it is defined

as < ~M > where ~M = 1N

∑

i ~mi and the angular bracketimplies expectation with respect to ground state. Here~m is the magnetic moment at a given site. Now we fol-low an exact calculation of magnetization and two spincorrelation function to find out what kind of order is ex-hibited by the ground state wave function. To do that wefirst discuss an extension of Kitaev’s Majorana fermioni-sation such that the mathematical steps to calculate thespin-spin correlation function or magnetization is verystraightforward.

A. Bond fermion formalism

We have seen in Sec. II B that two complex fermionsyield four Majorana fermions. Each complex fermion canbe rewritten into two Majorana fermions. Now to facil-itate the easy computation of spin-spin correlations weinvert the above procedure by regrouping two differentMajorana fermions to define a complex fermion. We haveseen that at every link there has been one conserve quan-tity named uαi,j made out of the Majorana fermion cαi,aand cαj,b. Here ‘i’ and ‘j’ denotes the two sites of a bond,‘a’ and ‘a’ denotes sub-lattice indices and ‘α’ denotes aspecific bond(α = x, y, z). We regroup these two Majo-rana fermions to define a complex fermion named χ〈ij〉αwhich lives on the bond joining sites ‘i’ and ‘j’. We callthis procedure as bond fermion formalism. From nowon we follow the convention that the site ‘i’ in the bond〈ij〉α belongs to a sub-lattice and the site ‘j’ belongs tob sub-lattice. Also from now on we do not mention thesub-lattice index ‘a’ and ‘b’ explicitly. We define complex

fermions on each bond as,

χ〈ij〉α =1

2

(

cαi + icαj)

(22)

χ†〈ij〉α =

1

2

(

cαi − icαj)

(23)

c

c

cc

cc χy χx

χz

χx χy

c2

c3

y

z

y

y

x

x

21

6

54

3

c3

c3

c3

c2c2

c2x

x y

y

z

z2

3

FIG. 9. Elementary hexagon and ‘bond fermion’ construction.A spin is replaced with four Majorana fermions (c, cx, cy , cz).Bond fermion χ〈23〉 for the bond joining site 2 and site 3 isshown . Spin operators are also defined.

For example with reference to the Fig. 9, for the z-bond joining site 2 and site 3, and for the y-bond joiningsite 1 and site 2, we define,

χ〈23〉z= (cz2 + icz3) (24)

χ〈12〉y= (cy1 + icy2) (25)

Then it follows that for the site ‘2’ and ‘3’ the σz operatorbecomes,

σz2= ic2(χ〈23〉z + χ†

〈23〉z) (26)

σz3= c2(χ〈23〉z − χ†

〈23〉z) (27)

Below we write the result of this re-fermionisation fora bond of type ‘α’ joining site ‘i’ and ‘j’,

χ〈ij〉α =1

2

(

cαi + icαj)

(28)

σαi = ici

(

χ〈ij〉α + χ†〈ij〉α

)

(29)

σαj = icj

(

χ〈ij〉α − χ†〈ij〉α

)

(30)

It is clear that three components of a spin operator ata given site gets connected to three different χ fermionsdefined on the three different bonds emanating from it.The bond variables are related to the number operators of

these fermions, u〈ij〉α ≡ icαi cαj = 2χ†

〈ij〉αχ〈ij〉α − 1. Thus

the effective picture is understood easily from the Fig. 9.We identify a χ fermion on every bond whose occupa-tion number can be zero or one. This occupation num-ber determines the value of u〈ij〉 on that bond. But thesefermions are conserved and serve as an effective Z2 gaugepotential for hopping ‘c’ fermions. As χ fermions are con-served, all eigenstates can therefore be chosen to have adefinite χ fermion occupation number. The Hamiltonianis then block diagonal in occupation number represen-tation, each block corresponding to a distinct set of χ

11

fermion occupation numbers for every bonds. Thus alleigenstates in the extended Hilbert space take the fol-lowing factorised form,

|Ψ〉 = |MG ;G〉 ≡ |MG〉|G〉 (31)

with χ†〈ij〉αχ〈ij〉α |G〉 = n〈ij〉α |G〉 (32)

where n〈ij〉α = (u〈ij〉α + 1)/2 and |MG〉 is a manybody eigenstate in the matter sector determined by ‘c’fermions, corresponding to a given Z2 field configurationdetermined by |G〉.

π πi

j��

PSfrag replacements

|ψ〉

σzi

|ψ′〉

FIG. 10. How a spin fractionalises into two static π fluxesand a dynamic Majorana fermion is shown. |ψ〉 is a statewith zero flux. We apply σz

i where site ‘i’ is connected withsite ‘j’. As a result we get a state |ψ′〉 with two static π fluxesat the plaquette sharing bond 〈ij〉 and a dynamic Majoranafermion represented by black circle.

Now we discuss the results of the above transforma-tions and find that it brings an immediate visualizationof the spin-operators. We observe form Eq. 28 that theeffect of σα

i on any eigenstate becomes very clear. Whenthe spin-operator acts on any eigenstate, in addition toadding a Majorana fermion at site ‘i’, it changes the bondfermion number from zero to one and vice versa (equiv-alently, u〈ij〉α → − u〈ij〉α), at the bond 〈ij〉α. The endresult is that one π flux is added to each of the two pla-quettes that are shared by the bond 〈ij〉a (Fig. 10).Mathematically the action of any spin-operator on anyeigenstate can thus be written as,

σαi = ici

(

χ〈ij〉α + χ†〈ij〉α

)

→ ici π1〈ij〉α π2〈ij〉α(33)

with π1〈ij〉α and π2〈ij〉α defined as operators that createsadditional π fluxes to two adjacent plaquettes shared bythe bond 〈ij〉α (Fig. 10). Now it is easy to understandthe action of one more spins which is connected with theprevious ones. It yields π2

1〈ij〉α = 1, since adding two π

fluxes is equivalent to adding (modulo 2π) zero flux. Thissignify that while action of single spin operator createsthe gauge fermion occupation number to change (eitherdecrease or increase by one), gauge fermion occupationnumber can be brought back to initial values by action ofa the same spin or a different neighbouring spins. Onlycriteria is that the spin angular component of both thespins has to be same and this is determined by the natureof the bonds they are connected with. It is now straight-forward to understand that two states with different fluxconfigurations has vanishing overlap as they belong todifferent distribution of χ fermion occupation numbers.

mathematically this implies that,

〈G|G′〉 = δnG ,nG′ , (34)

where nG and nG′ represent the distribution of χfermions for the state |G〉 and |G′〉 respectively. Thisobservation will be extremely helpful to computespin-spin correlations exactly. Not only two -spin cor-relation function, magnetization can also be calculatedexactly. Apart from two spin correlation functionsother multi spin correlations can be calculated withstraight forward generalisations of this fact that for anyspin-spin correlator to be non-zero the first necessaryconditions is that the simultaneous action of all thespin operators on the ground state must not change theflux configurations or equivalently must not alter thegauge fermion occupation number of the bonds. Thisfact can be extended to all the eigenstates of the Ki-taev model as well which is remarkable for Kitaev model.

B. Magnetisation

Magnetization ~M is an important physical quantitywhich is easily measurable and can be controlled experi-mentally as well. A particular component of magnetisa-tion is defined as below,

Mα=1

N∑

i

〈σαi 〉 (35)

In the above 〈σαi 〉 denotes the expectation value with

respect to ground state. For ferromagnetic state it canbe found that at least one component of 〈σα

i 〉 is non-zero at every site and it is identical for every site. Onthe other hand for anti-ferromagnetic state 〈σα

i 〉 are alsonon-zero at every site however their value is opposite atdifferent sites and makes some pattern depending on theunderlying structure. In Fig. 11 we have shown suchferromagnetic and antiferromagnetic structure in squareand honeycomb lattice. We can see easily that for antiferromagnetic state Mα is zero though for all the siteshaving red spins are having opposite magnetization ofblue spins. At a given site the average value of spin mo-mentum is not zero. In the lower panel we have showna different state where at each shaded green region de-fined on a pair of sites has the following singlet state|s〉 = 1√

2(| ↑↓〉 − | ↓↑〉). For such a state average value

of σαi = 0 at any site22–24. This state is fundamentally

different than the anti ferromagnetic state. For the anti-ferromagentic state at a given site σα

i is not zero. Now letus see what is the value of 〈σα

i 〉 = 0 for the ground stateof the Kitaev model. From the definition of spin-operatoras expressed in Eq. 33 we see that action of a spin on theground state creates two additional flux in the groundstate as shown in Fig. 10, in addition it also adds a Ma-jorana fermion to the ground state. Mathematically this

12

is expressed as

σαi |G〉|MG〉 = ci|Giα〉|MG〉 (36)

Now as the different flux configuration state is mutu-ally orthogonal due to different occupation number of theconserved χ fermion, we obtain,

〈G|〈MG |σαi |G〉|MG〉 = 〈G|〈MG |ci|Giα〉|MG〉 = 0(37)

because 〈G|Giα〉 = 0 following Eq. 34. Thus we see thatthe magnetization is zero for Kitaev model. It is zero forevery site unlike the AFM state where the magnetizationat a given site is not zero but when averaged over thesystem it is zero. For spin singlet also the magnetizationis zero at a given site. However there is an importantdifference between the AFM state, and the spin-singletstate shown in Fig. 11. For AFM state 〈σz

i σzj 〉 = ±1

depending on weather the site ‘i’ and ‘j’ are both arealigned in the same direction or in opposite direction.Note that it does not depends on the distance betweenthe two sites. It is said that the state has a long rangeordered state. For singlet state 〈σz

i σzj 〉 is not zero if the

site ‘i’ and ‘j’ both belongs to same singlet otherwise it iszero. Thus there is no long range correlation in the singletproduct state. However for the singlet state 〈σα〉 is zerowhich bears similarity with Kitaev model. However thereare important differences between the singlet state andKitaev model which will be established once we calculatethe two spin correlation function. With the knowledge offew magnetic states as discussed above, we now move onto calculate the spin-spin correlation function of Kitaevmodel which would establish its spin-liquid nature.

Exercise-7

Consider a spin-silglet state formed between site 1 andsite 2 as expressed by |s〉 = 1

2 (| ↑1〉| ↓2〉 − | ↓1〉| ↑2〉)where | ↑i〉 and | ↓i〉 refers the spin at a site ‘i’ ori-ented along positive and negative z axis respectively. Dothe following questions.

• Calculate 〈s|σz1σ

z2 |s〉

• Using the transformation | ↑i〉 =1√2(| →i〉+ | ←i〉) , and| ↓i〉 = 1√

2(| →i〉 − | ←i〉)

where | →i〉and| ←i〉 denotes the spin-orientedalong x-axis, express |s〉 in terms of | →i〉and| ←i〉with i = 1, 2.

• Now calculate 〈s|σx1σ

x2 |s〉 and show that it is non-

zero. Similarly find 〈s|σy1σ

y2 |s〉 and comment on the

symmetry property of the state |s〉

From the above exersize it must be clear that spin-singletproduct state is an unique state with no long range corre-lation but short range nearest neighbour correlation existwith 〈σα

i σαj 〉 non-zero if ‘i’ and ‘j’ are nearest neighbour

and they belong to a given dimer.

FIG. 11. In the top panel we have shown anti-feromagneticspin configurations in square lattice and in honeycomb lattice.In the below we have shown a dimer state configurations onsquare and honeycomb lattice where the green ellipses repre-sents a dimer. The particular arrangement of dimers to coverthe every site of the lattice is not unique and there exist al-ternative arrangement as well. The alternative arrangementsoften are topologically different.

C. Two-spin correlations, fractionalization,de-confinement

Here we give an outline of computation of spin-spincorrelation functions in Kitaev model which is valid forany region of the phase diagram. The derivation here isgiven in the extended Hilbert space but can be easily ex-tended without any difficulties to physical Hilbert space.This happens because after the implementation of projec-tion operator to ground state wave function obtained inextended Hilbert space one obtains many daughter statesdiffered with respect each other only in gauge sector byfermion occupation number in certain manner so that ac-tion of two spins on any of them still creates orthogonalstates and mutual overlap between such states are againvanishes. The above fact is expressed technically as thefollowing. Because the spin operators are gauge invari-ant (commute with the projection operator) the result ofcomputation of spin-spin correlation function in the ex-tended Hilbert space yields no problem at all. First tobegin with, we consider the two spin dynamical correla-tion functions, in an arbitrary eigenstate of the KitaevHamiltonian in some fixed gauge field configuration G,

Sαβij (t) = 〈MG |〈G|σα

i (t)σβ(0)j |G〉|MG〉 (38)

Here A(t) ≡ eiHtAe−iHt is the Heisenberg representationof an operator A, keeping ~ = 1. Physically the quantity

Sαβij (t) in Eq. 38 gives the joint probability amplitude

of finding a spin at ‘j’ (at time zero) along ‘α’ axis andof finding another spin at ‘i’ to be along ‘β’ axis. As

13

discussed before we write the action of spin operator onany eigenstate as,

σβ(0)j |G〉|MG〉 = cj(0)|Giβ〉|MG〉 (39)

σα(t)i |G〉|MG〉 = ei(H−E)tcj(0)|Giα〉|MG〉 (40)

where, |Giα(jβ)〉 denote the states with extra π fluxesadded to G on the two plaquette adjoining the bond inα or β directions. It means that if α = x, then the twoplaquettes are created adjoining the site ‘i’ to another site‘k’ joined by x-bonds. Similar explanation goes for theβ indices as well. In the above equation E is the energyeigenvalue of the eigenstate |G〉|MG〉. Since the Z2 fluxeson each plaquette is a conserved quantity due to the factthe it is determined by the occupation number of gaugefermions (χ fermions) on the links, it is obvious that thecorrelation function in Eq. 38 which is the overlap of thetwo states in equations 39, 40 is zero unless the spins areon neighbouring sites indicating that for non-vanishingvalues of correlation function we must have ‘j’ as thenearest neighbour site of the site ‘i’. Also note that thecomponents α and β must be identical so that the effect of

σαi and σβ

j is to create a pair of flux and then annihilateit bringing the flux configuration same as before. Theabove simple observations says that the dynamical spin-spin correlation has the form,

Sαβij (t)= g〈ij〉α(t)δα,β , for i, j nearest neighbours (41)

= 0 otherwise. (42)

Computation of gij(0) is straightforward in any eigen-state |MG〉. For the ground state where conserved Z2

charges are unity at all plaquette, we have an analyt-ical expressions of the ground state wave functions asgiven in Eq. 20. For simplicity we provide the expressionfor equal-time correlation by having t = 0 without lossof generality. For correlation function at different timeinterested readers are requested to consult the originalarticle21. Using Eq. 39 and Eq. 40, the equal time 2-spin correlation function for the bond 〈ij〉α reduces tothe following simple expectation value :

〈σαi σ

αj 〉 ≡ Sαα

〈ij〉α(0) = 〈MG |〈G|cicj |G〉|MG〉 (43)

In the above equation we have omitted the time indexfor ci and cj operator for simplicity. One can simplifythe expressions in Eq. 43 by using the Fourier transfor-mations of ci and cj operators and then employing theorthogonal transformations for ck operators as given inEq. 18, we obtain the following final expressions for thetwo-spin equal time correlation function,

〈σαi σ

αj 〉 ≡ Sαα

〈ij〉α(0) =

√3

16π2

∫

BZ

cos θ(k1, k2)dk1dk2

Where cos θ(k1, k2) =ǫkEk

, Ek =√

(ǫ2k +∆2k), in the Bril-

louin zone. ǫk = 2(Jx cos k1 + Jy cos k2 + Jz), ∆k =2(Jx sin k1 + Jy sin k2), k1 = k.n1, k2 = k.n2 and

Jx , , JzJy 01= = = 0

0.40.8

0.1

0.30.6

0.9

0.2

0.95

Jx ,Jy= =0 1 Jz= 0

Jx ,Jy 0= = Jz=0 1,

,

FIG. 12. Contour plot of nearest-neighbour z−z correlation inthe parameter space. We note that z−z correlation increasesas magnitude of Jz increases.

PSfrag replacements

0.2

0.2

Jz 1

1

0.8

0.8

0, 0

〈σzi σ

zj 〉

FIG. 13. In the above we plot the 〈σ1zσ2z〉 on a give z-bonds.We have taken Jx = Jy = 0.5 and vary Jz from 0 to 1. Asexpected for Jz = 0 correlation is zero and for large values itsaturates to 1.

n1,2 = 12ex±

√32 ey are unit vectors along x and y bonds.

At the point, Jx = Jy = Jz, we get Sαα〈ij〉α (0) = −0.52.

It is of interest to examine how this non-vanishing twospin correlation depends on the parameters of the Hamil-tonian mainly on Jx, Jy, Jz The Fig. 12 shows how thenearest neighbour z− z correlations varies in the param-eter ( Jx, Jy, Jz ) space. As we have calculated the abovetwo spin-correlation function for two spins connected bya z-bond, we observe that as we increase the magnitudeof Jz, the value of correlation also increases. In particulartwo limiting cases in the parameter values is worth ex-amining. Consider the case Jz = 0. Looking at the Fig. 1we observe that on this limit, the honeycomb lattice re-duces as collection of decoupled chains. Thus the twospins which have been joined by an interaction are nowwithout any interaction between them, though the spinsmaintain their original interactions with nearest neigh-bour spins in the respective chains. As there is no in-teraction between these two spins, we expect that therewill be no correlation between them and from Fig. 13, wedo find that indeed the correlation between them is zero.The other limiting cases is Jz = ∞ or Jx = Jy = 0, inthis limiting case the two spins connected by a ‘z’ bondsdoes not interact with any other spins except the oneconnected by Jz bonds. Thus one expect that the cor-relation will be stronger and reach the saturation valueone as shown in Fig. 13. Another interesting fact to notethat there is little change in slope as we increase fromgapped region to gapless region.

Fractionalization and De-confinement: We note

14

that here we have only discussed the derivation of equaltime correlation function which means that in Eq. 38both σ’s have identical argument for time. We foundthat the spin-spin correlation function is short-range andalso bond dependent. This property of short-range andbond dependent nature of correlation function remainsunchanged for different time correlation function as well.However the derivation and the exact expression requiresa little mathematical digression and for that we do notpresent that in this article. Instead we discuss the issuewithout going into technical details to meet the inquis-itiveness of the reader. Let us try to understand thephysical consequences of Eq. 38, Eq. 39 and ,Eq. 40. Ther.h.s of Eq. 38 involves inner product of two parts which

are σβ(0)j |G〉|MG〉 and 〈MG |〈G|σα

i (t) which are shown inEq. 39 and Eq. 40. The meaning of Eq. 39 is easy to un-derstand which tells what happens when a given spin actson an eigenstate. It creates two fluxes in the gauge sector|G〉 of the eigenstate and also adds a Majorana fermionin the matter sector( |MG〉). Let us try to expand themeaning of Eq. 40 which represents how the action ofa given spin on an eigenfunction evolves as function oftime. To this end we yield the intermediate steps reach-ing Eq. 40 as follows.

σαi (t)|G〉|MG〉

= eiHtσαi (0)e

−iHt|G〉|MG〉= eiHtσα

i (0)e−iEt|G〉|MG〉

= ei(H−E)tσαi (0)|G〉|MG〉

= ei(H−E)tci(0)|Gi,α〉|MG〉. (44)

In the first and second steps we have used the defini-tion of time evolution of an operator and applied eigen-state condition with energy E respectively. The thirdstep is a simple rearrangement and fourth step imple-ments the effect of spin σα

i which is equivalent to createtwo fluxes in the |G〉 (to have |Gi,α〉) and also addinga Majorana fermion ci in the matter sector |MG . NowWe need to consider the effect of the exponential opera-tor ei(H−E)t on this. Remember that from Eq. 36 thatci(0)|Gi,α〉|MG〉 is not an eigenstate of H . The situationis complicated for two reason. Firstly |Gi,α〉 makes allthe conserved quantity ’u’ to be 1 except on a bond con-necting site ‘i’. Secondly ci(0)〉|MG〉 is not an eigenstatefor such distribution of u’s in Eq. 8. In general if a stateis not an eigenstate of an Hamiltonian then the actionof Hamiltonian on it results into superposition of manyother eigenstates. Thus the action σα

i (t)|G〉|MG〉 can bemathematically represented as follows,

σαi (t)|G〉|MG〉

= e−iEt(

∑

cj(t)Aij |MG〉j)

|Gi,α〉, (45)

where |MG〉j represents ‘j’the eigenstate in the presenceof two fluxes. In each of these |MG〉j , the fluxes arein identical position, they are static but the Majorana

fermion added to each of this eigenstate |MG〉j are in dif-ferent position in general. This effect is a remarkable ef-fect and signifies that in essence a given spin is composedof two objects, a Mojorana fermion and a two flux. Whilethe fluxes do not move, the Majorana fermion moves astime evolves and it is not confined in a given region un-like static flux pairs. This implies that physically a spinis fractionalized into two objects. The fact that Majo-rana fermion is moving gives rise to a de-confinementphenomena. In Fig. 14, we depict pictorially the phe-nomena of fractionalization and de-confinement. The ex-act formula for σα

i (t)|G〉|MG〉 can be seen in the originalmanuscript21.

i_ Ht e

iππ

ππ

i=

PSfrag replacements

Aij(t)

kk

j

j

j

j

Majorana Fermion

FIG. 14. Time evolution and fractionisation of a spin flip at t= 0 at site ‘i’, into a π-flux pair and a propagating Majoranafermion.

Now in comparison with the singlet state we discussedbefore we find a number of differences. The similarity be-tween the two is that in both the cases the two spin corre-lation function is short range i.e it exists only for nearestneighbour bonds only. However there are a number ofimportant differences. For example for singlet state, thecorrelation function is non-zero if the two sites belongto a singlet only and for a given singlet 〈σα

i σαj 〉 exist for

α = x, y, z. However for Kitaev model for any two near-est neighbour correlation function is non-zero but theyare anisotropic in the sense that for a ‘α’ type bonds only〈σα

i σαj 〉 is non-zero and for other component the spin-spin

correlation function is zero.

D. Multispin Correlation function,topologicaldegeneracy

We have seen that for Kitaev model magnetization iszero at every site. Also two spin correlation function is

1

2 3

45 6

xy

z

1

23

45

6

FIG. 15. Multispin correlation function exist for Kitaevmodel. For the sites 1 to 6, 〈σz

1σy2σy3σy4σz5σ

y6〉 is non-zero.

For detail see text below.

15

23

4 6 81

12

1413

1211

9

81

23

45

67

1 5 7

1413

10 9

8

10

11

FIG. 16. A cartoon picture of torus geometry for a 4 × 4hexagon has drawn. The numerics show the specific way theboundaries are connected with each other. The above geom-etry is not unique and different geometry of lattice exist fortorus realization.

extremely short range such that it does not extend be-yond nearest neighbour. It is natural to ask then whatis the difference between a normal paramagnetic or otherdisordered magnetic state for which long range two spincorrelation is zero. We will now discuss that. Thoughthe two spin correlation exists only for nearest neighbourspins there is underlying long range correlation betweenspins such that spins at long distances are entangledunlike paramagnetic or other disordered magnetic statewhere such long range correlation does not exist. We willalso compare it with the singlet state described beforeand find a very important differences. The existence oflong range multispin correlation function depends againon the creation and annihilation of flux configurations in|G〉 such that the final and the initial flux configurationsremains same. We note that this was the main reason fornon-vanishing two spin correlations. In the right panel ofFig. 15, we have drawn a cartoon of Kitaev model wherex, y and z type of bonds are shown. Various numerics 1to 6 denote many spins for which we wish to show thata non-vanishing multispin correlation function exist. Wenotice that for bond 1− 2, σz

1σz2 does not change the flux

configurations in |G〉, similarly or bond 2− 3, σx2σ

x3 does

not change the flux configuration in |G〉. Thus for thesite 1 − 2− 3, σz

1σz2σ

x2σ

x3 ∼ σz

1σy2σ

x3 does not change the

flux configurations in |G〉 and this three spin correlator isnon-zero. Thus for the spin 1 to 6 the following multispincorrelation is non-zero,

S1−6 = σz1σ

y2σ

y3σ

y4σ

z5σ

y6 (46)

Thus for any path of arbitrary length there is a non-vanishing multispin correlation associated with it. Itshows that all the spins irrespective of the distance be-tween them are really entangled or depends on each otherin a specific way. This is not true for ordinary paramag-netic or disordered spin configurations. Also the abovemultispin correlation is not present in the singlet stateas shown in the left panel because there is no correlationexist between spin 2 and spin 3. However there couldbe an alternative singlet covering such that singlet stateis formed on bonds 2 − 3, 4 − 5 etc. If we consider the

superposition of this two states, we might have a mul-tispin correlation non-zero. Thus though Kitaev modeland the spin-singlet state both has short range two spin-correlation of different nature as well as of long rangemultispin correlation of different nature. The long rangemultispin correlations or string operators lead to anothernovel concept which is knew as topological order. Theword topology here is used to denote the fact that themovement of a given spin really depends on each otherand every spin is entangled every other spin in a partic-ular manner. This also says that a spin can not be ar-bitraily oriented and it must conform to certain globalpattern or the spin-orientations of other spins. Suchglobal pattern of spin configurations of all the spin mightlead some global constraint such that certain patternsare mutually exclusive leading to an existence of longrange string order parameter which takes distinct valuesfor each different pattern of spin configuration. We willnow explain this fact for Kitaev model defined on torus ofhoneycomb lattice. Torus means two dimensional latticesuch that it is periodically connected in both the direc-tion. In Fig. 16, we have shown such periodic boundarycondition applied to a 4×4 arrangement of hexagons. Ascan be seen that the lower and upper sides are connectedas well as the left and the right sides. Note that each onedimensional edges are periodically connected. Now weevaluate some numbers which the reader are requested tocross-check of this 4 × 4 arrangement of hexagon having32 sites. For a general N ×N hexagons, there are totalnumber of sites M = 2N2. Total number of hexagonis also M . We know that associated with each hexagonthere is a conserve quantity called Bp. Thus there arein total M number of Bp. Now one can check that fortorus geometry once all the Bp are multiplied together itbecomes unity such that,

∏

p=1,M

Bp = 1 (47)

This tells that all the Bp is not independent and thereare in total M − 1 independent Bp. It can be shownthat there is a conserve quantity associated with everyclosed loop, but they are not independent as they canbe constructed by a product of certain Bp’s. Howeverthere two global loop operator which extends from oneend to another which can not be expressed as a productof Bp and there are two such loop operators which wecall W1 and W2. These are analogous to Wilson loopoperator in quantum field theory26,27. For example takea horizontal zig-zag line constituted out of alternating xand y bonds. the product of σz for all the sites on thatline is a conserved quantity.

W1 =∏

x−y chain

σzi (48)

Similarly for a x− z chain extending from one end to an-other end, we can constructW2 which is conserved. Thustotal number of conserved quantity becomes M+1. Now

16

0

1

2

3

N−1

N

FIG. 17. Various red lines represent the spin states with evennumber of spin in the +z direction. The blue lines show oddnumber of spins in the −z direction. The red states contributeto W1 = 1 and the blue states contribute to W1 = −1.

it can be checked that W 2i = 1 implying that eigenval-

ues of Wi is ±1. The presence of these two additionalindependent conserved quantities imply that there willbe four different eigenstates of certain identical configu-rations of Bp but with different eigenvalues of Wi. Thesefour eigenfunctions can be represented as,

|Ψ1〉 = |MG〉|G(Bp);W1 = 1,W2 = 1〉 (49)

|Ψ2〉 = |MG〉|G(Bp);W1 = −1,W2 = 1〉 (50)

|Ψ3〉 = |MG〉|G(Bp);W1 = 1,W2 = −1〉 (51)

|Ψ4〉 = |MG〉|G(Bp);W1 = −1,W2 = −1〉 (52)

These Wi manifests a hidden topological order which wewill try to understand by considering how many ways W1

defined in Eq. 48 can be made one for one dimensionalarrangement of spins. In Fig. 17, we have shown by redline the states for which even number of spins are alignedin postive z-direction. For a one dimensional chain of Nsites there are 2N−1 such states. Similarly by blue line,we represents states with odd number of spins orientedalong negative z-axis. All the red states in a given x− yline contribute to the wave function to make W1 = 1 andsuch states can not be converted to an eigenfunction withW1 = −1 by local flipping of a given spin or a number ofspins. For that we need to convert all the red states tothe blue one. It is non-local transformation in the Hilbertspace and thus topological. However the energy does notdepends on the eigenvalues of Wi, it only depends on theflux configurations Bp. Thus all the states |Ψi〉, i = 1, 4has identical energy in the thermodynamic limit25. Thisis a manifestation of topological degeneracy.

IV. DISCUSSION

In this pedagogical paper, we have tried to explainfew aspects of Kitaev model7. We have explained indetail how the Majorana fermionisation is used to solvethe model. By solving we mean finding the eigenvalueand eigenvectors. All the eigenvectors and eigenvaluesof the Kitaev model can be found in principle. This is adistinct feature which makes it stand apart from otherexactly solvable model where in some cases we have onlythe ground state28. It may be noted that Kitaev modelcan be defined in any lattices with co-ordination numberof three and exact solution can be obtained29,30. The

Kitaev model is reduced to a Z2 gauge theory and in theground state sector all the Z2 gauge fields are such thattheir product over a plaquette becomes unity. Uniformchoice of Z2 gauge fields for every link is convenientchoice for such configuration. The spectrum, phasediagram and calculation of correlation function has beenshown explicitly and elaborately. Notable feature of thephase diagram is that it consists of gapless and gappedphase both. In the gapless phase near the region wherethe spectrum becomes zero, the energy varies linearlywith momentum. On the hand where the spectrum isgapped, the energy varies as quadratically with momen-tum. The correlation function calculation shows thattwo spin correlation function is short range and bonddependent. This property of correlation function is truefor both static and dynamic correlation function. Theshort range nature of correlation function indicates thatKitaev model is an example of quantum spin liquid statewhere there is no long range order between two spinsas well there is no average magnetization i.e averagevalue of any spin operator is zero. We have shownthat the action of a spin operator on any eigenstateis two fold. Firstly it changes the occupation numberof a Z2 gauge field on a particular bond by changingthe value of Bp in the adjacent plaquette. Secondlyat also adds a Majorana fermion at a given site thespin belongs to. However the time evolution of thestate shows that the pair of flux created does not movebut the Majorana fermion added can move and is notconfined or attached to that site only. This phenomenais called fractionalization of spins into static fluxes anddynamic Majorana fermion. The moving Majoranafermion defines a de-confinement phase which is gaplessone with linear spectrum at low energy.

Moving further we have shown how the long rangeor multispin entanglement exist in an eigenfunction byshowing non-zero multispin correlation function and ex-plained how it is different than other magnetic state withshort range two-spin correlation. Existence of topologicaldegeneracy of eigenfunction in thermodynamic limit hasbeen explained schematically. Historically Kitaev modelwas conceived mainly with the aim of implementation toquantum computations due to the existence of topolog-ical order and non-trivial topological excitations calledanyons. It may be noted that the gapped phase of Ki-taev model realizes abelian anions and gapless phase real-izes non-abelian anyons. The braiding properties of theseanyons in relation to realizing quantum gates has beendiscussed in detail7,31. Kitaev model is exactly solvablein the sense that all the eigenfunction and eigenvaluescan be obtained formally. The exact solvability is lostonce other more conventional spin-spin interaction suchas Ising or Heisenberg interaction32–34 is added to Ki-taev model. In this case the Z2 gauge field defined oneach bond is no longer remains conserved. The flux op-erator Bp also does not commute with the Hamiltonianany more. Physically this means that the fractionali-

17

sation of spins into Majorana fermion and static fluxesare not possible no longer. Because the fluxes also movealong with the Majorana fermions though their dynam-ics may be at different scale. Though the Kitaev modelwas proposed with a theoretical foundation for quantumcomputations, it was rather surprising that this unusualinteractions happens to exist in certain materials such asIridiate and RuCl3 etc35–37. Unfortunately along withKitaev interaction, there exists some other interactions

such as Heisenberg type interactions. Presently intenseresearch is underway to explore various aspects Kitaevmodel such as fractionalisation, de-confinement to con-finement transition, detection of spin-liquid state etc. In-terested reader are requested to go through some of therecent developments38 and the references there in.

Acknowledgement : A. M. Jayannavar acknowl-edges DST, India for JCBose fellowship. S M acknowl-edges G. Baskaran and R. Shankar for their encouragingpresence throughout the journey involving Kitaev model.

∗ saptarshi@iopb.res.in† jayan@iopb.res.in1 David L. Goodstein, States of Matter, Dover Publications

(2002)2 Xie Chen, Zheng-Cheng Gu, Zheng-Xin Liu and Xiao-

Gang Wen, Science 338 (6114), 1604-1606(2012).3 Leon Balents, Nature, 464(7286): 199–208 (2010).4 Hong Yao and Steven A. Kivelson, Phys. Rev. Lett 108,

247206 (2012).5 L. D. Landau, E. M. Lifschitz, Statistical Physics: Course

of Theoretical Physics Vol 5 (Pergamon, London, 1958).6 J. J. Sakurai, Modern Quantum Mechanics, Addison-

Wesley Publishing Company (1994).7 A. Y Kitaev, Ann. Phys. 321, 2 (2006).8 Stephen M Winter et al 2017 J. Phys.: Condens. Matter

29 493002.9 Neil W. Ashcroft and N. David Mermin, Solid State

Physics, HARCOURT ASIA PTE LTD, 2001.10 G. Baskaran, Diptiman Sen and R. Shankar Phys. Rev. B

78, 115116 (2008)11 J.E. House, Fundamentals of Quantum Mechanics, Aca-

demic Press (2018)12 A. Tsvelik, Quantum Field Theory in Condensed Matter

Physics. Cambridge University Press, Cambridge, (1995).13 P. Jordan and E. Wigner, Z. Physik 47, 631 (1928).14 Pierre Pfeuty, Annals of Physics, 57, 79-90(1970).15 E. Lieb, T. Schultz, D. Mattis, Ann. of Phys., 16, 407

(1961).16 A Yu Kitaev 2001 Phys.-Usp. 44 131 (2001).17 H. Bethe, Z. Phys., 71 , 205 (1931).18 Xiao-Gang Wen, Phys. Rev. B, 65, 165113(2002).19 Henrik Bruus and Karsten Flensberg, Many-body Quan-

tum Theory in Condensed Matter Physics, Oxford Univer-sity Press, 2007.

20 E. H. Lieb, Phys. Rev. Lett. 73, 2158 (1994).21 G. Baskaran, S. Mandal, and R. Shankar, Phys. Rev. Lett.

98 , 247201 (2007).22 R. Moessner and S. L. Sondhi, Phys. Rev. Lett 86, 1881

(2000).23 Michael Kamfor et al J. Stat. Mech. (2010) P08010.24 Alain Verberkmoes and Bernard Nienhuis, Phys. Rev. Lett

83, 3986 (1999).25 S Mandal, R Shankar and G Baskaran, J. Phys. A: Math.

Theor. 45 (2012) 335304.26 R. Giles, “Reconstruction of gauge potentials from Wilson

loops,” Phys. Rev. D, 24, 2160-2168 (1981).27 Wilson, K. "Confinement of quarks". Physical Review D.

10 (8): 2445(1974).28 C. K. Majumdar and D. P. Ghosh, J. Math. Phys. 10, 1388

(1969).29 Saptarshi Mandal and Naveen Surendran, Phys. Rev. B

79, 024426 (2009).30 Saptarshi Mandal, and Naveen Surendran, Phys. Rev. B

90, 104424 (2014).31 A.Y Kitaev, Ann. Phys. 303, 2 (2003).32 S. Mandal et al, Phys. Rev. B 84, 155121 (2011).33 Johannes Knolle, Subhro Bhattacharjee, and Roderich

Moessner, Phys. Rev. B 97, 134432 (2018).34 Robert Schaffer, Subhro Bhattacharjee, and Yong Baek

Kim Phys. Rev. B 86, 224417 (2012).35 G. Jackeli and G. Khaliullin, Phys. Rev. Lett., 102, 017205

(2009).36 Jiri Chaloupka, 1,2 George Jackeli, 2, and Giniyat Khali-

ullin 2, Phys. Rev. Lett 105, 027204 (2010).37 A. Banerjee, et al, Nature Materials, vol 15, page 733,

(2016).38 Hidenori Takagi, Tomohiro Takayama, George Jackeli and

Giniyat Khaliullin, arXiv 1903.08081 ( Accepted for pub-lication in Nature Reviews Physics).