Quantum Spin Liquids - Review - RMP2017

Download as pdf or txt
Download as pdf or txt
You are on page 1of 50

REVIEWS OF MODERN PHYSICS, VOLUME 89, APRIL–JUNE 2017

Quantum spin liquid states


Yi Zhou
Department of Physics and Zhejiang Institute of Modern Physics,
Zhejiang University, Hangzhou 310027, People’s Republic of China
and Collaborative Innovation Center of Advanced Microstructures, Nanjing 210093, China

Kazushi Kanoda
Department of Applied Physics, University of Tokyo, Hongo 7-3-1, Bunkyo-ku,
Tokyo 113-8656, Japan

Tai-Kai Ng
Department of Physics, Hong Kong University of Science and Technology,
Clear Water Bay Road, Kowloon, Hong Kong, China
(published 18 April 2017)

This is an introductory review of the physics of quantum spin liquid states. Quantum magnetism is
a rapidly evolving field, and recent developments reveal that the ground states and low-energy
physics of frustrated spin systems may develop many exotic behaviors once we leave the regime of
semiclassical approaches. The purpose of this article is to introduce these developments. The
article begins by explaining how semiclassical approaches fail once quantum mechanics become
important and then describe the alternative approaches for addressing the problem. Mainly spin-
1=2 systems are discussed, and most of the time is spent in this article on one particular set of
plausible spin liquid states in which spins are represented by fermions. These states are spin-singlet
states and may be viewed as an extension of Fermi liquid states to Mott insulators, and they are
usually classified in the category of so-called SUð2Þ, Uð1Þ, or Z2 spin liquid states. A review is
given of the basic theory regarding these states and the extensions of these states to include the
effect of spin-orbit coupling and to higher spin (S > 1=2) systems. Two other important approaches
with strong influences on the understanding of spin liquid states are also introduced: (i) matrix
product states and projected entangled pair states and (ii) the Kitaev honeycomb model.
Experimental progress concerning spin liquid states in realistic materials, including anisotropic
triangular-lattice systems [κ-ðETÞ2 Cu2 ðCNÞ3 and EtMe3 Sb½PdðdmitÞ2 2 ], kagome-lattice system
[ZnCu3 ðOHÞ6 Cl2 ], and hyperkagome lattice system (Na4 Ir3 O8 ), is reviewed and compared against
the corresponding theories.

DOI: 10.1103/RevModPhys.89.025003

CONTENTS D. Numerical realization of Gutzwiller projection:


Variational Monte Carlo method and some results 17
I. Introduction 2 1. One-dimensional lattice 19
II. From Semiclassical to Nonlinear-σ Model Approaches for 2. Triangular lattice 19
Quantum Antiferromagnets 3 3. Kagome lattice 20
A. Two-spin problem 3 E. Classification of spin liquid states:
B. Berry’s phase 3 Quantum orders and projective symmetry groups 21
C. Nonlinear-σ model 4 IV. Beyond RVB Approaches 22
1. Topological term 5 A. RVB and its generalization to spin systems
D. Quantum spin chains and the Haldane conjecture 6 with strong spin-orbit coupling 23
1. Integer spin chains 6 B. RVB approach to S > 1=2 systems 24
2. Half-odd-integer spin chains 6 C. Matrix product state and projected entangled
3. Open spin chains and end states 6 pair state 26
E. Higher dimensions and frustrated quantum 1. Valence-bond solids and MPSs in one dimension 27
antiferromagnets 7 2. PEPSs in higher dimensions and beyond 28
III. RVB States 8 D. Kitaev honeycomb model and related issues 29
A. RVB theory and gauge theory 10 V. QSL States in Real Materials 32
B. Uð1Þ gauge fluctuations 13 A. Anisotropic triangular-lattice systems:
1. Mott transition: Relation between Fermi κ-ðETÞ2 Cu2 ðCNÞ3 and EtMe3 Sb½PdðdmitÞ2 2 32
and spin liquids 14 1. κ-ðETÞ2 Cu2 ðCNÞ3 33
C. Z2 spin liquid states 16 2. EtMe3 Sb½PdðdmitÞ2 2 36

0034-6861=2017=89(2)=025003(50) 025003-1 © 2017 American Physical Society


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

B. Kagome-lattice system: ZnCu3 ðOHÞ6 Cl2 37 The limitations of the semiclassical approach when applied
C. Hyperkagome-lattice system: Na4 Ir3 O8 40 to systems with frustrated interactions are discussed in
D. Experimental summary 42 Sec. III, where we introduce the alternative idea of construct-
VI. Summary 44
ing variational wave functions directly. We introduce
Acknowledgments 45
Anderson’s famous idea of the RVB wave function for
Appendix: Path Integral for a Single Spin 45
References 46
spin-1=2 systems and discuss how this can be implemented
in practice. The difficulty of incorporating the SUð2Þ spin
algebra in the usual many-body perturbation theory is noted,
I. INTRODUCTION and the trick of representing spins by particles (fermions or
bosons) with constraints to avoid this difficulty is introduced.
Quantum spin liquid (QSL) states in dimensions of d > 1 The nontrivial SUð2Þ gauge structure in the fermion repre-
have been a long-sought goal in condensed matter physics. sentation of RVB states and the resulting rich structure of the
The general idea is that when acting on spin systems, quantum low-energy effective field theories for these spin states
mechanics may lead to exotic ground states and low-energy [SUð2Þ, Uð1Þ, and Z2 spin liquids] are discussed. An
behaviors that cannot be captured by traditional semiclassical interesting linkage of the Uð1Þ spin liquid state to the
approaches. The difficulty in implementing this idea is that we (metallic) Fermi liquid state through a Mott metal-insulator
have no natural place to start once we have left the comfort transition is introduced.
zone of semiclassical approaches, at least in dimensions larger The difficulty of finding controllable approaches for study-
than 1. Except for a few exactly solvable models, we must rely
ing spin liquid states has led to an extension of the RVB
heavily on numerical or variational approaches to “guess” the
approach and a search for alternative approaches. Some of
correct ground-state wave functions and on a combination of
these approaches are briefly reviewed in Sec. IV, including
sophisticated numerical and analytical techniques to under-
(i) the extension of the RVB approach to include the effect of
stand the corresponding low-energy excitations.
spin-orbit coupling and to higher spin (S > 1=2) systems,
Several excellent reviews are available on QSLs (P. A. Lee,
(ii) the concepts of matrix product states and projected
2008; Balents, 2010) and frustrated magnetism (Diep, 2004;
entangled pair states, and (iii) the Kitaev honeycomb model.
Lacroix, Mendels, and Mila, 2011). This article complements
The main message of this section is that a larger variety of
those mentioned by providing a pedagogical introduction to
exotic spin states become possible when we leave the
this subject and reviews the current status of the field. We
paradigm of spin-1=2 systems with rotational symmetry.
explain, at an introductory level, why sophisticated
approaches are needed to study QSL states, how these The Uð1Þ and Z2 spin liquid states belong to a very small
approaches are implemented in practice, and what new subset of the plausible exotic states once we leave the
physics may be expected to appear. The experimental side paradigm of semiclassical approaches.
of the story and the drawbacks or pitfalls of the theoretical Section V is devoted to a survey of experimental research
approaches are also discussed. We concentrate mainly on on spin liquid states. Special attention is paid to the Uð1Þ spin
spin-1=2 systems and study in detail one particular set of liquid state, on which most experimental efforts have been
plausible spin liquid states that are usually termed resonating focused. The best studied examples are a family of organic
valence-bond (RVB) states. The spins are treated as fermions compounds denoted by κ-ðETÞ2 Cu2 ðCNÞ3 (ET) (Shimizu
in these states, which may be viewed as an extension of Fermi et al., 2003) and PdðdmitÞ2 ðEtMe3 SbÞ (dmit salts) (Itou
liquid states to Mott insulators. They are usually classified in et al., 2008). Both materials are Mott insulators near the
the category of SUð2Þ, Uð1Þ, or Z2 spin liquid states. Because metal-insulator transition and become superconducting (ET)
of the intrinsic limitations of the fermionic RVB approach, or metallic (dmit) under modest pressure. Despite the large
many other approaches to spin liquid states have been magnetic exchange J ≈ 250 K observed in these systems,
developed by others. These approaches often lead to other there is no experimental indication of long-range magnetic
exotic possibilities not covered by the simple fermionic ordering down to a temperature of ∼30 mK. A linear
approach. Two of these approaches are introduced in this temperature dependence of the specific heat and a Pauli-like
article for completeness: (i) matrix product states (MPSs) and spin susceptibility have been found in both materials at low
projected entangled pair states (PEPSs) and (ii) the Kitaev temperature, suggesting that the low-energy excitations
honeycomb model. are spin-1=2 fermions with a Fermi surface (Yamashita
The article is organized as follows. In Sec. II, we introduce et al., 2008; Watanabe et al., 2012). This Fermi-liquid-like
the semiclassical approach to simple quantum antiferromag- behavior is further supported by their Wilson ratios, which are
nets, and we explain the importance of the spin Berry phase close to 1. In addition to ET and dmit salts, the kagome
and how one can include it in a semiclassical description compound ZnCu3 ðOHÞ6 Cl2 (Helton et al., 2007) and the
to obtain the correct theory. In particular, we show how it three-dimensional hyperkagome material Na4 Ir3 O8 (Okamoto
leads to the celebrated Haldane conjecture. The existence et al., 2007) are also considered to be candidates for QSLs
of end excitations as a natural consequence of the with gapless excitations. Experimental surveys on these QSL
low-energy effective theory of these systems is discussed. candidate materials are presented in this article, including their
One-dimensional quantum spin systems are of great interest at thermodynamics, thermal transport, and various spin spectra.
present because they provide some of the simplest realizations We also briefly introduce the discoveries of a few new
of symmetry-protected topological (SPT) phases in strongly materials and discuss the existing discrepancies between
correlated systems. experiments and theories. The paper is summarized in Sec. VI.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-2


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

II. FROM SEMICLASSICAL TO NONLINEAR-σ MODEL L ¼ r × p ¼ r0 r̂ × p; ð4Þ


APPROACHES FOR QUANTUM ANTIFERROMAGNETS
where p ¼ mr0 r̂_ is the momentum and m is the mass. Using
Here we consider simple Heisenberg antiferromagnets on
Eq. (4), we obtain
bipartite lattices (with sublattices A and B) with the
Hamiltonian _
r̂ × L ¼ −r0 p ¼ −mr20 r̂: ð5aÞ
X
H ¼ J Si · Sj ; ð1Þ
We also have
hi;ji

_ ¼0
L ð5bÞ
where J > 0 and hi; ji describe a pair of nearest neighbor sites
in the bipartite lattice. In a bipartite lattice, any two nearest
neighbor sites always belong to different sublattices. S is a (the conservation of angular momentum). Comparing Eqs. (3)
quantum spin with magnitude S ¼ n=2, where n ¼ positive and (5), we find that the equation of motion for two spins is
integer. Examples of bipartite lattices include 1D spin chains, equivalent to the equation of motion for a free rotor if we
2D square or honeycomb lattices, and 3D cubic lattices. identify L → M, r̂ → n̂, and J ¼ I −1 , where I ¼ mr20 is the
moment of inertia of the rotor.
A. Two-spin problem
The quantum Hamiltonian of the free rotor is

The semiclassical approach begins with the assumption 1 2


Hrotor ¼ L ;
that the quantum spins are “close” to classical spins, and 2I
it is helpful to start by first analyzing the corresponding
and its solution is well known. The eigenstates are the
classical spin problem. For simplicity, we start by consider-
spherical harmonics Y lm ðθ; ϕÞ (where θ and ϕ specify the
ing only two classical spins coupled by the Heisenberg
direction of the unit vector r̂) with eigenvalues
interaction
L2 ¼ lðl þ 1Þℏ2 ; Lz ¼ mℏ;
H ¼ JSA · SB ðJ > 0Þ:

The classical spins obey Euler’s equation of motion: and corresponding energies El ¼ lðl þ 1Þℏ2 =2I, where l and
m are integers such that l ≥ 0 and l ≥ jmj. In particular,
∂SAðBÞ LðMÞ ¼ 0 for the ground state of the quantum rotor, but the
¼ JSBðAÞ × SAðBÞ : ð2Þ direction of the vector rðNÞ is completely uncertain
∂t pffiffiffiffiffi
[Y 00 ðθ; ϕÞ ¼ 1= 4π ] as a result of quantum fluctuations,
This equation can be solved most easily by introducing the indicating a breakdown of the classical solution, in which n is
magnetization and staggered magnetism vectors MðNÞ ¼ fixed in the ground state. (Alternatively, one can gain this
SA þ ð−ÞSB , where it is easy to show from Eq. (2) that understanding from the Heisenberg uncertainty principle
hδr̂ihδLi > ℏ. With L ¼ 0 in the ground state, δL ≡ 0 and
∂M ∂N δr̂ → ∞, the direction of the vector r̂ becomes completely
¼ 0; ¼ JM × N; ð3Þ
∂t ∂t uncertain.)
A moment of thought indicates that our mapping of the spin
indicating that classically the staggered magnetization vector problem to the rotor problem cannot be totally correct. What
N rotates around the (constant) total magnetization vector M. happens if SA is an integer spin and SB is a half-odd-integer
Let SAðBÞ ¼ SAðBÞ r̂AðBÞ , where SAðBÞ are the magnitudes of the spin? Elementary quantum mechanics tells us that the ground
spins SAðBÞ and r̂AðBÞ are unit vectors indicating the directions state should carry half-odd-integer angular momentum. The
of SAðBÞ ; then, the classical ground state has r̂A ¼ −r̂B with possibility of such a scenario is missing in our rotor mapping,
M ¼ 0, i.e., the two spins are antiferromagnetically aligned. in which the spin magnitudes SAðBÞ do not appear.
Note that the equation of motion given in Eq. (3) implies that
∂ðN2 Þ=∂t ¼ 0, i.e., the magnitude of N remains unchanged B. Berry’s phase
during its motion. Therefore, if we write N ¼ N n̂, where N is
the magnitude of N and n̂ is the unit vector denoting its The missing piece in our mapping of the two-spin problem
direction, we find that only n̂ changes under the equation of to the rotor model is the Berry’s phase (Berry, 1984), which is
motion given in Eq. (3). carried by spins but is absent in rotors. The correct spin-
The effects of quantum mechanics can be seen most easily quantization rule is recovered only after this piece of physics
by observing that the equation of motion given in Eq. (3) is properly added into the rotor problem. First, let us review
describes the dynamics of a free rotor (a rigid rod with one end the Berry’s phase carried by a single spin.
fixed such that the rod can rotate freely around the fixed end). We recall that for a spin tracing out a closed path C on the
A free rotor can be represented by a vector r ¼ r0 r̂, where surface of the unit sphere, the spin wave function acquires a
r0 ¼ const is the length of the rod and r̂ is the unit radial Berry’s phase γðCÞ ¼ SΩðCÞ, where S is the spin magnitude
vector describing the orientation of the rod. The rod has an and ΩðCÞ is the surface area under the closed path C on the
angular momentum of unit sphere (see Fig. 1). SΩðCÞ can be represented more

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-3


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

The most interesting feature of the monopole spherical


C harmonics is that they allow half-odd-integer angular momen-
r̂ ( t )
tum states (which occur when jΔSj is a half odd integer). The
ground state carries an angular momentum of L ¼ jΔSj and is
ð2jΔSj þ 1Þ-fold degenerate, corresponding to the degeneracy
of a quantum spin of magnitude jΔSj in agreement with the
exact result for the two-spin problem.

C. Nonlinear-σ model

FIG. 1. Berry’s phase with a magnetic monopole. The two-spin problem tells us that there are two important
elements that we must keep track of when a classical spin
problem is replaced with the corresponding quantum spin
conveniently by imagining the spin trajectory as the trajectory
problem: (a) quantum fluctuations, originating from the (non)
of a particle carrying a unit charge moving on the surface of
commutation relation between the canonical coordinates (N)
the unit sphere. In this case, the Berry’s phase is simply the
and momenta (M), and (b) Berry’s phase, which dictates
phase acquired by the charged particle if a magnetic monopole
the quantization of the spins. In the following, we generalize
of strength S [i.e., BðrÞ ¼ ðS=r2 Þr̂] is placed at the center of
the rotor approach to the many-spin systems described by the
the sphere. The Berry’s phase acquired is the magnetic flux
antiferromagnetic (AFM) Heisenberg model, keeping in mind
enclosed by the closed path C.
these two elements.
Let SAM ðrÞ be the vector potential associated with the
Following Haldane (1983a, 1983b), we consider here
monopole, i.e., ∇ × AM ¼ r̂=r2 ; then, in the “charge þ
Heisenberg antiferromagnets on a bipartite lattice described
gauge field” representation, the effect of the Berry’s phase
by the Hamiltonian given in Eq. (1). As in the two-spin
can be described by a vector-potential term in the action:
problem, we introduce the magnetization vectors Mðxi Þ and
Z the staggered magnetization vectors Nðxi Þ such that
SB ¼ ℏSΩðCÞ ¼ ℏS _
dtAM ðr̂Þ · r̂: ð6Þ
SAi ¼ Mðxi Þ þ Nðxi Þ;
This is an example of a Wess-Zumino term for quantum SBi ¼ Mðxi Þ − Nðxi Þ; ð10Þ
particles. A more rigorous derivation of the Wess-Zumino action
is given in the Appendix, where the action for a single spin in a where SAðBÞ denote spins on the AðBÞ sublattices of the
magnetic field is derived via a path integral approach. bipartite lattice. We assume that the ground state of the
We now revisit the two-spin problem. With the Berry’s quantum system is “classical like” with nearly antiparallel
phases included, the Lagrangian of the corresponding rotor spins on two nearest neighboring sites such that Mðxi Þ ≪
problem becomes Nðxi Þ, where both MðxÞ and NðxÞ are very slowly varying
functions in space. (We show that this assumption can be
1 _ 2 þ ℏSA AM ðr̂A Þ · r̂_ A þ ℏSB AM ðr̂B Þ · r̂_ B ; justified in the following section.) The classical equation of
L¼ ðn̂ × n̂Þ ð7Þ
2J motion for the spin at lattice site i is

where N ¼ N n̂ ¼ SA − SB . To simplify the problem, we adopt AðBÞ  X 


∂Si BðAÞ AðBÞ
the semiclassical approximation r̂A ¼ −r̂B in the Berry’s ¼J Sj × Si ; ð11Þ
phase terms, which is a reasonable approximation for states ∂t j¼NNðiÞ
close to the classical ground state. With this approximation,
we obtain where j ¼ NNðiÞ means that j represents the nearest neighbor
sites of i.
1 _ 2 þ ℏΔSAM ðn̂Þ · n̂;
_ Using Eq. (10), after some straightforward algebra and
L→ ðn̂ × n̂Þ ð8Þ
2J taking the continuum limit, we obtain

where n̂ ¼ r̂A and ΔS ¼ SA − SB . The Hamiltonian of the ∂NðxÞ


∼ JzMðxÞ × NðxÞ;
system is ∂t
∂MðxÞ Ja2 2
J ∼− ½∇ NðxÞ × NðxÞ; ð12Þ
HM ¼ ½Π − ℏΔSAM ðn̂Þ2 ; ð9Þ ∂t 2
2
where a is the lattice spacing and z ¼ 2d is the coordination
_ is the canonical momentum of the rotor.
where Π ¼ n̂=J number. We assumed a square (cubic)-type lattice and adopted
HM is the Hamiltonian of a charged particle moving on the slowly varying approximation
the surface of a unit sphere with a magnetic monopole of
strength jΔSj located at the center of the sphere. The Mðxiþ1 Þ þ Mðxi−1 Þ ∼ 2Mðxi Þ;
eigenstates of the Hamiltonian are well known and are called
the monopole spherical harmonics (Wu and Yang, 1976). Nðxiþ1 Þ þ Nðxi−1 Þ ∼ 2Nðxi Þ þ a2 ∂ 2x Nðxi Þ;

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-4


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Z  
etc. in deriving the result. We also assumed MðxÞ to be small ℏS ∂ ∂
ST ∼ d dd x    d Ω(n̂ðxÞ): ð15Þ
and neglected all nonlinear terms in MðxÞ in deriving Eq. (12). 2 ∂ 1x ∂x
To proceed further, we consider the situation in which all
spins have the same magnitude S. Then it is easy to see from ST is sensitive to the boundary conditions (see the following
Eq. (10) that NðxÞ2 þ MðxÞ2 ¼ S2 and NðxÞ · MðxÞ ¼ 0. discussion), and we assume closed (periodic) boundary
Assuming that M ¼ jMðxÞj ≪ N ¼ jNðxÞj ∼ S, we find from conditions in the following. The case of open boundary
pffiffiffi
Eq. (12) that M ∼ ω=zJ and ω ∼ zJaSjkj, where ω and k conditions is discussed later. For periodic boundary condi-
are the frequency and wave vector, respectively, of p the
ffiffiffi tions, it is easy to see that ST is zero unless the integrand has a
fluctuations in N. In particular, M ≪ N when ak ≪ z, nontrivial topological structure.
i.e., when NðxÞ is slowly varying in space. To evaluate ∂ x Ω, we recall that Ωðn̂Þ measures the area on
In the following, we adopt the approximation N ∼ S and the surface of the sphere bounded by the trajectory n̂ðtÞ. Thus,
write NðxÞ ¼ Sn̂ðxÞ, where n̂2 ¼ 1. Eliminating MðxÞ from the variation δΩðn̂Þ due to a small variation in the trajectory δn̂
Eq. (12), we obtain is simply
Z
∂ 2 n̂ðx; tÞ zðSJaÞ2 2
¼ ∇ n̂ðx; tÞ; ð13aÞ δΩðn̂Þ ¼ dtδn̂ · ðn̂ × ∂ t n̂Þ
∂t2 2

corresponding to the following classical action for the vector and


field n̂:
Z Z   
Z Z   2  ℏS ∂ ∂
1 1 ∂ n̂ zJðSaÞ2 ST ¼ dd x dt    n̂ · ðn̂ × ∂ t n̂Þ: ð16Þ
Sσ ¼ dt dd x − ð∇n̂Þ2 ; ð13bÞ 2d ∂ 1x ∂ dx
2 J ∂t 2
The total effective action describing the quantum antiferro-
with the constraint n̂2 ¼ 1. Sσ is the nonlinear-σ model magnet is S ¼ Sσ þ ST .
(NLσM) for the unit vector field n̂ðxÞ. The topological term is nonzero in one dimension and is
Comparing Eqs. (8) and (13b), we see that the NLσM can usually written in the form
be viewed as a continuum model describing coupled rotors
Z
n̂ðxÞ. The first term in the action gives the kinetic energy for ST θ X
the rotors, which we discussed in detail for the two-spin ¼ d2 xεμν n̂ · ð∂ μ n̂ × ∂ ν n̂Þ; ð17aÞ
ℏ 8π μ;ν¼0;1
model. The second term represents the coupling between
nearest neighboring rotors in the lattice spin model. We note
that the term for the coupling between rotors has a magnitude where x0 ¼ t, x1 ¼ x, θ ¼ 2πS, and εμν is the rank-2 Levi-
of ∼S2 and dominates over the kinetic energy in the limit of Cività antisymmetric tensor (Haldane, 1985; Affleck, 1986).
large S. The Pontryagin index
A more systematic derivation of the NLσM starting from Z
1 X
Eq. (10) can be achieved by writing Q¼ d2 xεμν n̂ · ð∂ μ n̂ × ∂ ν n̂Þ ¼ integer ð17bÞ
8π μ;ν¼0;1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
 Mðxi Þ2
Si ¼ ηi Sn̂ðxi Þ 1 −    þ Mðxi Þ;
S  measures how many times the 2½¼ 1ðspaceÞ þ 1ðtimeÞ-
dimensional spin configuration n̂ has wrapped around the
unit sphere. In two dimensions,
where ηi ¼ eiπx and we still have NðxÞ · MðxÞ ¼ 0.
Assuming that MðxÞ is small, we can integrate out MðxÞ Z
ℏθ ∂QðyÞ
in a power series expansion of MðxÞ in the path integral. The ST → dy ¼ 0;
NLσM for n̂ðxÞ is thus obtained to the leading (Gaussian) 2 ∂y
order (Auerbach, 1994).
where QðyÞ is the Pontryagin index that arises from summing
over all spin configurations in the yth column of the two-
1. Topological term dimensional lattice. The sum is zero for smooth spin con-
We next consider the Berry’s phase contribution to the figurations because Q is an integer and thus cannot “change
action. Following the Appendix, the total Berry’s phase smoothly” (Dombre and Read, 1988; Fradkin and Stone,
contribution is 1988; Haldane, 1988b; Wen and Zee, 1988). For the same
X X reason, ST vanishes for any number of dimensions greater
ST ¼ SB ðr̂i Þ ∼ ℏS ð−1Þi Ωðn̂i Þ; ð14Þ than 1. However, one should be cautioned that this conclusion
i i is valid only when we restrict ourselves to smooth spin
R configurations n̂ðx; tÞ when computing ST . The Berry’s phase
where SΩðr̂i Þ ¼ S dtAM ðr̂i Þ · r̂_ i is the Berry’s phase for a may have a nonzero contribution if we also allow singular spin
single spin and ð−1Þi ¼ 1 ð−1Þ for sites on even (odd) configurations in the theory. This is the case in ð2 þ 1ÞD,
sublattices. In the last step, we assumed that the spins are where monopolelike spin configurations are allowed in 3D
almost antiparallel. In the continuum limit, we obtain space (Haldane, 1988b; Read and Sachdev, 1990).

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-5


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

D. Quantum spin chains and the Haldane conjecture It turns out that this naive expectation is valid only in
dimensions of d > 1. In one dimension, the magnetically
We now study the predictions of the effective action for ordered state is not stable because of quantum fluctuations
quantum spin chains. In one dimension, the quantum spin associated with the Goldstone mode (Mermin-Wigner-
chains are described by the path integral Hohenberg theorem), and the ground state is always quantum
Z disordered (Mermin and Wagner, 1966; Hohenberg, 1967),
D½n̂ðx; tÞeði=ℏÞ½Sσ ðn̂ÞþST ðn̂Þ : i.e., a spin liquid state. This result can be shown more
rigorously through a renormalization group (RG) analysis
of the NLσM. We do not go through this analysis in detail;
We first consider the topological term. We note that ST ¼ instead, we simply assume that this is the case and examine its
2ℏπSQ and eði=ℏÞST ¼ ð−1Þ2SQ (Q ¼ integer). In particular, consequences. Readers interested in the RG analysis can
eði=ℏÞST ≡ 1 for integer spin chains, and the Berry’s phase has consult, for example, Polyakov (1975), Brézin and Zinn-
no effect on the effective action. However, eði=ℏÞST ¼ 1 for Justin (1976), and Polyakov (1987).
half-odd-integer spin chains, depending on whether Q is even Physically, this result means that after some renormaliza-
or odd. There is no further distinction between spin chains tion, the ground state of integer spin chains can always be
with different spin values S in ST . This result leads to the first viewed as a product state of local spin singlets, irrespective of
part of the Haldane conjecture, namely, that fundamental the spin magnitude S. The lowest-energy excitations are
differences exist between integer and half-odd-integer spin gapped spin-triplet (L ¼ 1) excitations. This is the Haldane
chains (Haldane, 1988b). To proceed further, we first consider conjecture for integer spin chains.
integer spin chains, where eði=ℏÞST ≡ 1 and the system is
described by the “pure” NLσM Sσ . 2. Half-odd-integer spin chains
The RG analysis cannot be straightforwardly applied to
1. Integer spin chains half-odd-integer spin chains because of the appearance of the
We start by asking the following question: what are the topological term ST . To understand why, let us again take the
plausible ground states described by Sσ ? For this purpose, it is RG to the strong coupling limit and examine what happens in
more convenient to consider a lattice version of Sσ : this case.
To zeroth order, the Hamiltonian of the system consists only
1
Z X1 ∂ n̂i 2 
of the kinetic energy term. However, the rotors are moving
Sσ → dt þ JS2 n̂i · n̂iþ1 ; ð18Þ
2 i
J ∂t under the influence of effective monopole potentials origi-
nating from ST . In particular, all half-odd-integer spin chains
with the corresponding Hamiltonian have the same ST with an effective magnetic monopole
strength of 1=2, corresponding to that of a spin-1=2 chain.
JX In this case, the ground state of a single rotor has an angular
Hσ ¼ ½ðLi Þ2 − S2 n̂i · n̂iþ1 ; ð19Þ momentum of L ¼ 1=2 and is twofold degenerate [see the
2 i
discussion after Eq. (9)]. The total degeneracy of the ground
where Li is the angular momentum operator for the ith rotor. state is 2N , where N is the number of lattice sites. This
The Hamiltonian contains two competing terms, and we enormous degeneracy implies that the coupling between
expect that it may describe two plausible phases, a strong rotors cannot be neglected when we consider the rotor
coupling phase, in which the kinetic energy (first) term Hamiltonian given in Eq. (19), and the strong coupling
dominates, and a weak coupling phase, in which the potential expansion simply tells us that the system behaves like a
energy (second) term dominates. A natural control parameter coupled-spin-1=2 chain (Shankar and Read, 1990).
for this analysis is the spin magnitude S, which dictates the Fortunately, the antiferromagnetic spin-1=2 chain can be
magnitude of the potential energy. In the first case (small S), in solved using the exact Bethe Ansatz technique (Giamarchi,
which the potential energy term is small, we expect that the 2003). The exact Bethe Ansatz solution tells us that the
ground state can be viewed, to a first approximation, as a antiferromagnetic spin-1=2 Heisenberg chain is critical,
product of local spin singlets, i.e., L ¼ 0 states, namely, the ground state has no long-range magnetic order
but has a gapless excitation spectrum. Unlike integer spin
jGi ¼ j0i1 j0i2    j0iN ; chains, where the lowest-energy excitations carry spin S ¼ 1,
the elementary excitation of this system has spin S ¼ 1=2.
where j0ii represents the L ¼ 0 state for the rotor on site i. Combining this with the continuum theory leads to the
The lowest-energy excitations are L ¼ 1 states separated from Haldane conjecture for half-odd-integer spin chains, namely,
the ground state by an excitation gap ∼ℏ2 J. This picture is that they are all critical with elementary S ¼ 1=2 excitations.
believed to be correct as long as the magnitude of the potential
energy term is much smaller than the excitation energy for the 3. Open spin chains and end states
L ¼ 1 state. In the second case, in which the potential energy The Haldane conjecture has been checked numerically for
term dominates (large S), we expect that the ground state is a quantum spin chains with different spin magnitudes and has
magnetically ordered (Néel state) with n̂i ¼ n̂0 at all sites i, been found to be correct in all cases that have been studied
where the excitations are Goldstone modes of the ordered state thus far. One may wonder whether the difference in spin
(spin waves). magnitudes may manifest at all in some low-energy properties

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-6


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

of quantum spin chains. The answer is yes, when we consider strong coupling expansion as before, we find that the system
open spin chains. behaves as an open spin-1=2 chain coupled to two end spins
Recall that we have always assumed periodic boundary with a magnitude of 1=2 þ ð1=2ÞðS − 1=2Þ. The problem of
conditions in deriving ST . In fact, a periodic boundary impurity end spins coupled to a spin-1=2 chain has been
condition is needed to define the Pontryagin index for the analyzed using the bosonization technique, through which it
topological term ST . For an open chain of length L, ST is was found that after the screening induced by the spin-1=2
replaced by (Haldane, 1983a; Affleck, 1990; Ng, 1994) chain (essentially a Kondo effect), a free spin with a
magnitude of ð1=2ÞðS − 1=2Þ is left at each end of the spin
Z
ðoÞ ℏ ∂SB (n̂ðxÞ)
L chain (Eggert and Affleck, 1992). Note that the existence of
ST ¼ dd x end states in half-odd-integer spin chains is rather nontrivial
2 0 ∂x
because the bulk spin excitations are gapless. As a result, the
ℏS
¼ 2πℏSQ þ ½Ω(n̂ðLÞ) − Ω(n̂ð0Þ); ð20Þ end spins at the two ends of a half-odd-integer spin chain are
2
coupled by a term J eff ∼ JS2 =L ln L, where L is the length of
where 2πSQ ¼ θQ is the usual topological θ term that we the spin chain. The excitation energy of the end state is
obtain when Ω(n̂ð0Þ) ¼ Ω(n̂ðLÞ), i.e., when we consider logarithmically lower than the energy of the bulk spin
periodic boundary conditions. An open chain differs from a excitations, which have an energy of ∼J=L (Ng, 1994).
closed chain in the existence of an additional boundary These predictions for open chains and end states based on
Berry’s phase term with an effective spin magnitude of S=2. the NLσM plus topological θ term analysis have been verified
We now examine the effect of this additional Berry’s phase numerically by means of density matrix renormalization
term. First, we consider integer spin chains. Following the group (DMRG) calculations (Qin, Ng, and Su, 1995).
previous discussion, we expect the spin chain to be described
E. Higher dimensions and frustrated quantum antiferromagnets
by the strong coupling limit of the effective Hamiltonian given
in Eq. (19), except that the rotors at the two ends of the spin The NLσM approach to quantum antiferromagnets has been
chain are subjected to monopole potentials of strength S=2, extended to higher dimensions and to frustrated quantum
resulting in effective free spins of magnitude S=2 located at antiferromagnets. For simple antiferromagnets, ST vanishes in
the ends of the spin chain. The two spins are coupled by a term dimensions of d > 1, and we need only consider the NLσM,
J eff ∼ JS2 e−L=ξ when the coupling between rotors is consid- i.e., Sσ . As discussed, Sσ describes two plausible phases, the
ered, where ξ ∼ E−1 g is the correlation length and Eg is the spin weak coupling phase, in which the ground state is antiferro-
gap. These end states can also be understood based on a wave magnetically ordered, and the strong coupling phase, in which
function proposed by Affleck, Kennedy, Lieb, and Tasaki (the the ground state is gapped. The weak coupling phase is
AKLT state) for S ¼ 1 spin chains (Affleck et al., 1987) (see favored for large spin magnitudes S. Various numerical and
Sec. IV) and have been observed experimentally in S ¼ 1 spin analytical studies have consistently demonstrated that the
chain materials (Glarum et al., 1991). In modern terminology, ground state is always Néel ordered for simple quantum
the end states of integer spin chains are a manifestation of SPT antiferromagnets on a 2d square lattice, even for the smallest
order (Gu and Wen, 2009; Chen et al., 2012; Pollmann et al., possible spin value of S ¼ 1=2 (Manousakis, 1991). For this
2012), which manifests itself as a boundary action that is reason, physicists have turned to frustrated spin models to
protected by rotational [SO(3)] symmetry.1 look for exotic spin liquid states.
For half-odd-integer spin chains, the analysis is a bit more The NLσM approach has generated interesting results
ðoÞ when applied to weakly frustrated spin models, where the
complicated. We start by rewriting Eq. (20) for ST as follows
(Ng, 1994): main effect of frustration is to reduce the effective coupling
strength between rotors (for example, J 1 − J 2 models, in
  which a next-nearest neighbor antiferromagnetic coupling is
ðoÞ ℏ 1
ST ¼ 4π Q þ S½Ω(n̂ðLÞ) − Ω(n̂ð0Þ) added to the Heisenberg model on a square lattice). In this
2 2
 case, it has been shown that spin-Peierls order can be obtained
ℏ 1 1 when discontinuous monopolelike spin configurations are
¼ 4π Q þ ½Ω(n̂ðLÞ) − Ω(n̂ð0Þ)
2 2 2 included in the calculation of ST (Read and Sachdev,
   1990). However, the method becomes questionable when
1
þ S − ½Ω(n̂ðLÞ) − Ω(n̂ð0Þ) ; ð21Þ applied to strongly frustrated spin systems, in which effective
2
rotors become difficult to define locally, for example, the
antiferromagnetic Heisenberg model on a kagome lattice.
where we replaced S with 1=2 in the usual topological
Generally speaking, a continuum theory is reliable only if
(Pontryagin index) term and divided the boundary Berry’s
the short-distance physics is captured correctly by the under-
phase term into two parts; the first part, when combined with
lying classical or mean-field theory. A continuum theory
the Pontryagin index term, is the total Berry’s phase con-
becomes unreliable if the short-distance physics it assumes is
tribution for an open S ¼ 1=2 spin chain, and the second part
not correct. This seems to be the case for the NLσM approach
is the additional contribution when S > 1=2. Performing the
when applied to strongly frustrated spin systems. In the
following sections, we consider alternative methods of treat-
For S ¼ 1 chains, the S ¼ 1=2 end states are protected by a
1
ing quantum spin systems, keeping in mind the physics that
weaker Z2 × Z2 symmetry (Chen, Gu, and Wen, 2011a, 2011b). we have previously discussed.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-7


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

III. RVB STATES bond is a spin-singlet state constructed from two S ¼ 1=2
spins at sites i and j, given by
The semiclassical approach, which is based on fluctuations
around a presumed classical (Néel) order, is difficult to apply 1
in frustrated lattice models. The difficulties arise from two ði; jÞ ¼ pffiffiffi ðj↑i ↓j i − j↓i ↑j iÞ; ð22Þ
2
main sources. First, different degenerate or quasidegenerate
and an RVB state is a tensor product of valence-bond states,
classical ground states may exist in a frustrated spin system. It
whose wave function is given by
is difficult to include these quasidegenerate classical ground
states in the NLσM description. Second, the effect of Berry’s X
phases becomes intractable because of the complicated jΨRVB i ¼ aði1 j1 in jn Þ jði1 ; j1 Þ    ðin ; jn Þi; ð23Þ
i1 j1 in jn
(classical) spin trajectory.
The term geometric frustration (or frustration for short) was where ði1 ; j1 Þ    ðin ; jn Þ are dimer configurations covering the
introduced by Gerard Toulouse in the context of frustrated entire lattice (see Fig. 3). The wave function is summed over all
magnetic systems (Toulouse, 1977; Vannimenus and Toulouse, possible ways in which the lattice can be divided into pairs of
1977). Indeed, frustrated magnetic systems had long been lattice sites (i.e., dimers). The quantities aði1 j1 in jn Þ are varia-
studied prior to that time. Early work included a study
tional parameters determined by minimizing the ground-state
conducted Wannier (1950) on the classical Ising model on a
energy of a given Hamiltonian. For a quantum disordered
triangular lattice with antiferromagnetically coupled nearest
antiferromagnet, it was proposed that the valence-bond pairs in
neighbor spins, which serves as the simplest example of
the RVB construction are dominated by short-range pairs,
geometric frustration (Diep, 2004). Because of the AFM
resulting in liquidlike states with no long-range spin order. The
coupling, two nearest neighboring spins A and B tend to be
corresponding spin correlation function hSi ⋅ Sj i in the RVB
antiparallel (see Fig. 2). Then, a third spin C that is a neighbor of
state may be short in range, with a finite correlation length
both A and B is frustrated because its two possible orientations,
[usually called short-range RVB (sRVB)], or may decay with
up and down, both have the same energy. The classical ground
distance following a power law (algebraic spin liquid states).
state has a high level of degeneracy. As a result, we cannot
The state is called a valence-bond solid (VBS) state if a single
choose a classical spin order as the starting point for con-
dimer configuration dominates in the ground state. An alge-
structing the NLσM for the quantum S ¼ 1=2 XXZ model
braic spin liquid state is usually invariant under all symmetry
X ðzÞ ðzÞ X ðxÞ ðxÞ ðyÞ ðyÞ operations allowed by the lattice, whereas a VBS state usually
H ¼ Jz Si Sj þ J ⊥ ðSi Sj þ Si Sj Þ breaks the translational or rotational lattice symmetry.
hi;ji hi;ji
The wave function given in Eq. (23), which is parametrized
by aði1 j1 in jn Þ, has too many variational degrees of freedom
with J z ≫ J⊥ because there exist infinite spin configurations
even after the translational and rotational symmetries of the
with the same classical energy. We note that the spin-spin
wave function are considered and must be simplified for
correlation has been found to decay following a power law at
practical purposes. A solution was proposed by Baskaran,
zero temperature in the exact solution for the classical Ising
Zou, and Anderson (1987), who noted that the Bardeen-
model (Stephenson, 1970).
Cooper-Schrieffer (BCS) states for superconductors are direct
In this case, an alternative approach is a variational wave
product states of spin-singlet Cooper pairs and suggested that
function, in which we essentially must guess the ground-state
good RVB wave functions can be constructed from BCS wave
wave function based on experience or physical intuition. An
functions via a Gutzwiller projection, denoted by PG :
important idea related to this approach is the RVB concept for
Y
spin-1=2 systems suggested by Anderson. The term RVB was jΨRVB i ¼ PG jΨBCS i; jΨBCS i ¼ ðuk þ vk c†k↑ c†−k↓ Þj0i;
first coined by Pauling (1949) in the context of metallic k
materials. Anderson (1973) revived interest in this concept
ð24Þ
when he constructed a nondegenerate quantum ground state
for an S ¼ 1=2 AFM system on a triangular lattice. A valence

A B

?
C

FIG. 2. Geometric frustration. The spin C is frustrated because


either the up or down orientation will give rise to the same energy FIG. 3. A spin-singlet dimer configuration covering a lattice. An
in the AFM Ising limit. RVB state is a superposition of such configurations.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-8


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

where c†k↑ and c†−k↑ are electron creation operators and the that the ground state for such an AFM system is a spin-
numerical coefficients uk and vk are determined from a trial singlet state with positive-definite coefficients in the Ising
BCS mean-field Hamiltonian HBCS through the Bogoliubov– basis fð−1ÞN A↓ jσ 1    σ N ig, where N A↓ is the number of down
de Gennes equations, i.e., the RVB wave function is fixed by spins in sublattice A and N is the number of lattice sites. Using
the parameters determining HBCS . The number of electrons at this result, Liang, Doucot, and Anderson (1988) proposed the
each lattice site may take a value of 0, 1, or 2 in the original use of the following trial ground-state RVB wave function for
BCS wave functions. The Gutzwiller projection PG removes spin-1=2 Heisenberg antiferromagnets on a square lattice:
all wave function components with doubly occupied sites X
from the BCS state and freezes the charge degrees of freedom. jΨLDA i ¼ hði1 − j1 Þ    hðin − jn Þ
A half-filled Mott insulator state is obtained if the total iα ∈A;jβ ∈B

number of electrons is equal to the number of lattice sites. × ð−1ÞN A↓ jði1 ; j1 Þ    ðin ; jn Þi; ð26Þ
We note that the technique of Gutzwiller projection is
currently being widely applied to other mean-field wave where hðrÞ represents a positive-definite function of the bond
functions jΨMF i to study Mott insulating states in length r. This particular wave function can be conveniently
diverse physical systems. Interesting and energetically favor- represented as a Gutzwiller-projected wave function in the
able wave functions are often obtained when jΨMF i is chosen Schwinger boson representation, whereas the representation
properly. of the same wave function in terms of fermions is far from
In addition to representing spins by electrons or fermions, straightforward (Read and Chakraborty, 1989). However, it
one may also use Schwinger bosons to represent spins to has been shown that the projected BCS wave function given in
construct RVB wave functions [see also the discussion after Eq. (24) will satisfy Marshall’s sign rule provided that the
Eq. (27)]. It is easy to recognize that, in general, almost any spatial Fourier transformation of uk and vk (¼ uij and vij )
mean-field wave function jΨMF i can be employed to construct connects only sites in different sublattices in a bipartite lattice
a corresponding spin state as follows: (Yunoki and Sorella, 2006; Li and Yang, 2007).
It was noted by Ma (1988) that the sum of states
jΨSpin i ¼ PG jΨMF i; ð25Þ jði1 ; j1 Þ    ðin ; jn Þi, with iα ∈ A and jβ ∈ B, forms an over-
complete set for spin-singlet states in a bipartite lattice.
where jΨMF i is the ground state of a trial mean-field Because h is a positive function, it can be interpreted as a
Hamiltonian Htrial ðc; c† ; a1 ; …; aN Þ, where c†iσ ðciσ Þ can re- weight factor in a Monte Carlo (MC) simulation based on loop
present either fermions or bosons and a1 ; …; aN are varia- gas statistics. Such a calculation has been performed for large
tional parameters determined by minimizing the energy of the lattices by Liang, Doucot, and Anderson (1988), and a very
parent spin Hamiltonian.2 The invention of Gutzwiller pro- accurate ground-state wave function for the AFM Heisenberg
jection techniques enables us to construct a large variety of model on a square lattice was obtained. The wave function can
variational spin wave functions, of which the best is the one give rise to either long-range or short-range spin correlations
with the lowest energy. depending on the choice of hðrÞ.
The most important difference between the fermion and Once a proper RVB ground-state wave function has been
boson constructions is that they lead to very different sign constructed, the next natural question is what are the low-
energy dynamics, or the elementary excitations on top of the
structures in the spin wave function jΨRVB i. In a bosonic wave
ground states? A natural candidate for excitation is to break a
function, when two spins (note that only spin degrees of
spin-singlet pair in the ground state to form a spin-triplet
freedom remain after Gutzwiller projection) at different sites
excited state with two unpaired spins. For a long-range
are interchanged, the wave function does not change, whereas
magnetically ordered state, it was found that the two unpaired
the wave function does change sign when two spins are
spins will bind together closely in space and that the resulting
interchanged in a fermionic wave function. These different
elementary excitations will be localized spin-triplet excita-
sign structures represent very different quantum entanglement
tions with well-defined energy and momentum. This is
structures in the corresponding RVB wave functions. A
nothing but a spin wave or magnon excitation, as guaranteed
famous example is Marshall’s sign rule (Marshall, 1955)
by the Goldstone theorem. By contrast, for a QSL state with
for the AFM Heisenberg model on a bipartite lattice, where
short-range spin correlation, it was proposed that the two
the Heisenberg exchange exists only between bonds linking
unpaired spins may interact only weakly with each other and
sites in different sublattices. Marshall’s theorem tells us
can be regarded as independent spin-1=2 elementary excita-
tions called spinons (see Fig. 4). The existence of S ¼ 1=2
2
For historical reasons, the fermion representation is also called spinon excitations is one of the most important predictions in
the slave-boson representation, and the Schwinger boson represen- QSLs and is crucial to the experimental verification of QSLs.
tation is also called the slave-fermion representation. In the context of The process through which a spin-1 magnon turns into two
doped Mott insulators, one can decompose the electron annihilation independent spin-1=2 spinons is an example of fractionali-
operator as ciσ ¼ h†i f iσ , where f iσ carries a charge-neutral spin and zation. Whether fractionalization of spin excitations actually
h†i is the (spinless) hole creation operator. If the spinon operator fiσ is occurs in a particular spin system is a highly nontrivial
fermionic, then the charge carrier (h†i ) is a “slave boson,” whereas if question. A systematic way to examine whether fractionali-
the spinon operator is bosonic, then the charge carrier is a “slave zation may occur in a spin model was first proposed by Wen
fermion.” (1989, 1991) based on the concept of confinement or

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-9


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

where α; β ¼ ↑; ↓ are spin indices, f†iα ðf iα Þ is the fermion


creation (annihilation) operator, and σ~ ¼ ðσ 1 ; σ 2 ; σ 3 Þ repre-
sents the Pauli matrices. It is easy to confirm that the three
components of S~ i satisfy the SUð2Þ Lie algebra relation
½Sλi ; Sμj  ¼ iϵλμν Sνi δij , where λ, μ, ν ¼ 1, 2, 3, and ϵλμν is the
antisymmetric tensor. Hence, Eq. (27a) is a representation of
SUð2Þ spins. However, the local Hilbert space for two
fermions contains four Fock states fj0i; f†↑ j0i ¼ j↑i; f †↓ j0i ¼
j↓i; f †↑ f †↓ j0i ¼ j↑↓ig; this is larger than the physical spin
Hilbert space for spin 1=2 ¼ fj↑i; j↓ig, and we need to
impose the single-occupancy constraint
X
f†iα fiα ¼ 1 ð27bÞ
α

to remove the unphysical states to obtain a proper spin


representation. This is what the Gutzwiller projection does.
The construction presented in Eq. (27) is equally applicable
for bosons (the Schwinger boson representation) because
FIG. 4. A spinon excitation on top of an RVB ground state.
the SUð2Þ Lie algebra is independent of the statistics of the
represented particles. In the following, we focus on the
fermion representation approach because it has been found
deconfinement in lattice gauge theory.3 This approach is to be a more fruitful approach for constructing QSLs. Readers
explained in the following section, where the gauge theory who are interested in the Schwinger boson approach can refer
for QSLs is introduced. to Arovas and Auerbach (1988) for details.
There are multiple choices of ffiα g available to represent
A. RVB theory and gauge theory spin operators even once the single-occupancy constraint is
satisfied and the statistics of the particles have been chosen.
This section presents a brief survey of how RVB theory is For example, a new set of ffiα g can be obtained through an
implemented in practice, especially how low-energy effective Uð1Þ gauge transformation:
field theories for QSL states are constructed, which is crucial
for characterizing QSLs. We discuss a few common examples f iα → f0iα ¼ eiθðiÞ fiα :
of QSLs and define the SUð2Þ, Uð1Þ, and Z2 spin liquid states.
The nature of the Uð1Þ QSL state is then further illuminated It is easy to verify that ff 0iα g forms another representation of
by relating it to a Fermi liquid state through a Mott metal- spin operators by replacing fiα with f0iα in Eq. (27), inde-
insulator transition. We shall see that analytical approaches pendent of whether the f’s are fermions or bosons. This
have strong limitations and should be complemented by multiplicity is called gauge redundancy or gauge symmetry in
numerical approaches in practice. the literature. We call it gauge redundancy here because
One complication associated with the RVB construction is symmetry usually refers to situations in which there are
that there exist, in general, different mean-field states jΨMF i multiple physically distinct states with the same properties,
that correspond to the same RVB spin wave function after e.g., there is a degeneracy in energy. However, the gauge
degree of freedom we discuss here is not a “real” symmetry
Gutzwiller projection. This redundancy originates from the
among different physical states. Here two gauge-equivalent
enlarged Hilbert space in the boson and fermion representa-
states are the same state in the spin Hilbert space. They just
tion for spins and is called gauge redundancy or gauge
“look” different when they are represented by particles that
symmetry. Gutzwiller projection removes this redundancy,
live in an enlarged Hilbert space. There is no way to
resulting in a unique state in spin Hilbert space. To see how distinguish them physically (Wen, 2002).
this occurs, we consider the fermion representation of S ¼ The gauge redundancy in the fermion representation of S ¼
1=2 spin operators (Abrikosov, 1965; Baskaran, Zou, and 1=2 spins extends beyond Uð1Þ. There exists an additional
Anderson, 1987; Baskaran and Anderson, 1988): SUð2Þ gauge structure because of the particle-hole symmetry in
the fermion representation, which is absent in the Schwinger
1 X † boson representation. An elegant way of showing this SUð2Þ
S~ i ¼ fiα σ~ αβ f iβ ; ð27aÞ
2 αβ
gauge structure was suggested by Affleck, Zou et al. (1988),
who introduced the following 2 × 2 matrix operator:
!
3 f↑ f†↓
This criterion for fractionalization works only in dimensions Ψ¼ : ð28Þ
d > 1. In one dimension, gauge fields are always confining, while f↓ −f†↑
spinons appear in energy spectrum as the gapless spin-1=2 excita-
tions of the quantum antiferromagnet Heisenberg model (Mudry and It is straightforward to show that the spin operator can be
Fradkin, 1994a, 1994b). reexpressed in terms of Ψ as

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-10


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

S~ i ¼ trðΨ†i σ~ Ψi Þ: ð29Þ with


The single-occupancy condition given in Eq. (27b) also leads to X X 
3
the identities Sðf; f̄; aÞ ¼ Sðf; f̄Þ − i a0 f̄iα fiα − 1
α
fi↑ fi↓ ¼ f †i↑ f †i↓ ¼ 0:
i
ð30aÞ  X 
Together with Eq. (30a), Eq. (27b) can be rewritten in the þ ða10 þ ia20 Þ ϵαβ fiα fiβ þ H:c: : ð37Þ
αβ
following compact vector form:
It is generally believed (but has not been proven) that the
trðΨi σ~ Ψ†i Þ ¼ 0: ð30bÞ partition function Z will remain invariant under a Wick
We now consider the following SUð2Þ gauge transforma- rotation of the fields al0 in the path integral, namely, we
tion of Ψ: can replace ial0 with al0 . Then, the action becomes
 X X 
Ψi → Ψ0i ¼ Ψi W i ; W i ∈ SUð2Þ: ð31Þ Sðf; f̄; aÞ ¼ Sðf; f̄Þ − a30 f̄iα fiα − 1
i α
The spin operator S~ i in Eq. (29) remains invariant under this  X 
transformation because W i W †i ¼ 1. The single-occupancy þ ða10 þ ia20 Þ ϵαβ fiα fiβ þ H:c: : ð38Þ
αβ
constraint given in Eq. (30b) is also invariant because
W i σ~ W †i represents a rotation of vector σ~ but all components The action given in Eq. (38) serves as the starting point
of trðΨi σ~ Ψ†i Þ are zero, i.e., Ψi → Ψ0i ¼ Ψi W i is also a valid for theoretical analysis. The path integral is difficult to solve,
representation for S ¼ 1=2 spins. and approximate methods are generally needed. We start with
We show now how RVB theory is implemented in an a mean-field theory in which we assume that the path integral
analytical fermionic approach. For concreteness, we consider is dominated by saddle points characterized by equal-time
P P
an AFM Heisenberg model on a lattice: expectation values of the operators α f †iα f iα , αβ ϵαβ fiα fiβ ,
X and al0 ðiÞ:
H¼J S~ i · S~ j ; ð32Þ
hiji
X X
χ ij ¼ hf †iα f jα i; Δij ¼ ϵαβ hfiα fjβ i; al0 ¼hal0 ðiÞi;
α αβ
where hiji denotes a nearest neighbor bond and J > 0. The
ð39Þ
spin exchange S~ i · S~ j can be written in terms of fermionic
(spinon) operators:
where ϵαβ is the totally antisymmetric tensor (ϵ↑↓ ¼ 1), l ¼ 1,
1 X † 2, 3. It is easy to verify that χ ij and Δij satisfy the relations
S~ i · S~ j ¼ ð2f iα f iβ f †jβ fjα − f†iα fiα f †jβ fjβ Þ; ð33Þ χ ij ¼ χ ji and Δij ¼ Δji . Note that any time-dependent fluc-
4 αβ
tuations in Δij ; χ ij and al0 ðiÞ are ignored in mean-field theory.
With these approximations, we arrive at the following mean-
where we used the relation σ~ αβ · σ~ α0 β0 ¼ 2δαβ0 δα0 β − δαβ δα0 β0 .
field Hamiltonian:
The constraint given in Eq. (27b) or (30b) can be imposed by
inserting delta functions into the imaginary-time path integral. X  X X 
3 † † †
The corresponding partition function is HMF ¼ − J χ ji fiα fjα þ Δij ϵαβ f iα fjβ þ H:c
hiji
8 α αβ
Z Y X  
Z ¼ D½f; f̄ exp½−Sðf; f̄Þ δ f̄iα fiα − 1 − jχ ij j2 − jΔij j2
i α
X  X 
X X 
×δ ϵαβ fiα fiβ δ ϵαβ f̄iα f̄iβ ; ð34Þ þ a30 f†iα fiα − 1
αβ αβ i α
 X 
where the action Sðf; f̄Þ is given by þ ða10 þ ia20 Þ ϵαβ f iα fiβ þ H:c: ; ð40Þ
αβ
Z X 
β
Sðf; f̄Þ ¼ dτ f̄iα ∂ τ fiα − H : ð35Þ where χ ij , Δij , and al0 are determined by minimizing the
0 iα
ground-state energy with the exact constraint condition (27b)
The delta functions can be represented by the integration over replaced with the average constraint
X †
R a0 ðiÞ on all sites i, l ¼ 1, 2, 3. Using the
l
real auxiliary fields
hfiα fiα i ¼ 1: ð41Þ
relation δðxÞ ¼ ðdk=2πÞeikx , we obtain α
Z
Z¼ D½f; f̄; a exp½−Sðf; f̄; aÞ; ð36Þ The spin exchange term S~ i · S~ j in Eq. (33) can be evaluated
within the mean-field assumption using the Wick theorem.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-11


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Maintaining spin rotation invariance in the calculation, we becomes explicit if we introduce a doublet field
obtain ψ ¼ ðf↑ ; f †↓ ÞT and a 2 × 2 matrix
 
hS~ i · S~ j i ¼ −38ðχ ij χ ij þ Δij Δij Þ ð42Þ χ ij Δji
uij ¼ :
Δij −χ ji
in mean-field theory.
Physically, the mean-field theory is equivalent to assuming The mean-field Hamiltonian (40) can be written in a compact
that the ground state of the spin system is given by a mean- manner as
field wave function jΨMF i without Gutzwiller projection. The X 3 1 
spin exchange energy (42) evaluated in this way is usually not HMF ¼ J Trðu†ij uij Þ − ðψ †i uij ψ j þ H:c:Þ
a good estimate of the energy of the real spin wave function. In hiji
8 2
practice, this mean-field theory provides an effective way to X
þ al0 ψ †i τl ψ i ; ð44Þ
obtain a BCS Hamiltonian to construct a Gutzwiller-projected
i;l
wave function. Whether the spin wave function obtained
through Gutzwiller projection is a good wave function for the where the τl , l ¼ 1, 2, 3, are the Pauli matrices. From Eq. (44)
spin Hamiltonian can be tested only by evaluating the energy we can see that the Hamiltonian HMF is invariant under a local
of the wave function numerically (see Sec. III.D). SUð2Þ transformation W i :
In the following section, we assume that the Gutzwiller-
projected wave function PG jΨMF i is a sufficiently good ψ i → Wiψ i; uij → W i uij W †j : ð45Þ
starting point to locate the true ground state of the spin
Hamiltonian. In this case, we expect that the ground and This SUð2Þ gauge transformation is the same as that in
low-energy states constructed from HMF are adiabatically Eq. (31), where Ψ ¼ ðψ; iσ 2 ψ † ÞT .
connected to the corresponding Gutzwiller-projected wave Because of this SUð2Þ gauge structure, if we regard the
functions and that we may construct an effective low-energy Ansatz ðuij ; al0 τl Þ as labeling a physical spin wave function
Hamiltonian or Lagrangian of the spin system from fluctua- ðu ;al τl Þ ðu ;al τl Þ
jΨspin
ij 0
i ¼ PG jΨMFij 0 i, then such a label is not a one-to-
tions around H MF through the usual path integral technique.
The fluctuations in Δij ; χ ij and al0 ðiÞ describe spin-singlet one label. Two Ansätze related by an SUð2Þ gauge trans-
excitations and are usually called gauge fluctuations. Before formation, ðuij ; al0 τl Þ and ðu0ij ; a0l0 τl Þ ¼ (Wðuij Þ; Wðal0 τl Þ),
discussing gauge fluctuations, we first discuss the effect of label the same physical spin wave function:
gauge redundancy on the mean-field states.
(Wðuij Þ;Wðal0 τl Þ) ðu ;al0 τl Þ
To illustrate, we consider two mean-field QSL states with jΨspin ðfαi gÞi ¼ PG jΨMF i ¼ PG jΨMFij i; ð46Þ
different structures of the mean-field parameters fχ ij ; Δij ;
al0 ðiÞg. We place the states on a simple square lattice. The first where Wðuij Þ ¼ W i uij W †j and W(al0 ðiÞτl ) ¼ W i al0 ðiÞτl W †i ,
state is the uniform RVB state with W i ∈ SUð2Þ. The uniform RVB state and the zero-flux state
discussed denote the same physical spin state because they are
χ ij ¼ 0; related by a gauge transformation,

Δ; NN bonds;  
Δij ¼ π
0; others; W i ¼ exp i τ2 :
4
al0 ¼0 ðl ¼ 1; 2; 3Þ: ð43aÞ
More generally, the existence of gauge redundancy implies
The second example considered is the zero-flux state given by that the low-energy fluctuations in spin systems have a similar
redundancy. To measure gauge fluctuations, we introduce the

χ; NN bonds; loop variables
χ ij ¼
0; others;
PðCi Þ ¼ uij ujk    uli ;
Δij ¼ 0;
al0 ¼ 0 ðl ¼ 1; 2; 3Þ: ð43bÞ where i; j; k; …; l denote a loop of lattice sites that passes
through site i. PðCi Þ measures gauge fluxes and has the
Δ and χ are real numbers. We show that irrespective of their general form
very different appearances, these two mean-field Ansätze
actually give rise to the same spin state after Gutzwiller PðCi Þ ¼ AðCi Þτ0 þ BðCi Þ · ~τ;
projection. The two states are gauge equivalent because they
can be transformed into each other through a proper gauge where τ0 is the identity matrix and ~τ ¼ fτ1 ; τ2 ; τ3 g represents
transformation. the Pauli matrices, AðCi Þ and BðCi Þ measure the Uð1Þ and
The Hamiltonian given in Eq. (40) retains a local SUð2Þ SUð2Þ components, respectively, of the gauge flux. For a
structure, which originates from the gauge redundancy in translationally invariant mean-field state, we can find a gauge
the fermion representation of spin. This local SUð2Þ symmetry with BðCi Þ ¼ n̂BðCi Þ, where AðCi Þ and BðCi Þ are

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-12


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

proportional to the area of the loop. Under a gauge trans- Upon writing χ ij ¼ χeiaij , where aij denotes phase fluctua-
formation, tions, it is straightforward to see that

PðCi Þ → W i PðCi ÞW †i ; PðCi Þ ∝ exp ½iΦðCi Þτ3 ;

and the “direction” of n̂ changes. The presence of gauge where ΦðCi Þ ¼ ðaij þ ajk þ    þ ali Þ is the total Uð1Þ gauge
redundancy means that we may perform gauge transforma- flux enclosed by the loop, i.e., the phase fluctuations of χ ij
tions to change the “local” directions of n̂, but the physical represent one component of the SUð2Þ gauge fluctuations.
spin state remains unchanged. The effective Lagrangian describing these low-energy
For a given mean-field state, it is useful to distinguish phase fluctuations is
between two kinds of gauge transformations: those that
X  
change the mean-field Ansatz fuij ; al0 ðiÞg and those that do ð0Þ 3X X
L ¼ f̄ iα ð∂ τ − a0 Þfiα þ Jχe iaji
f̄iα fjα þ H:c: ;
not. The latter constitute a subgroup of the original SUð2Þ iα
8 hiji α
symmetry called an invariant gauge group (IGG) (Wen, 2002):
ð48Þ
IGG ≡ fW i jW i uij W †j ¼ uij ; W i ∈ SUð2Þg: ð47Þ and the corresponding Lagrangian in the continuum limit is
Z X
It can be shown rather generally that for a stable QSL state, ð0Þ
physical gapless gauge excitations exist only for those L ¼ d~r f̄α ð~rÞð∂ τ − a0 Þfα ð~rÞ
α
fluctuations belonging to the IGG of the corresponding
1
mean-field Ansatz. Therefore, it is important to understand þ f̄ ð~rÞð−i▿ þ a~ Þ2 f α ð~rÞ; ð49Þ
the structure of the IGGs in spin liquid states. Within the 2m α
fermionic SUð2Þ formalism, there are only three plausible
where m is the effective mass for the spinon energy
kinds of IGG: SUð2Þ, Uð1Þ, and Z2 . We call the corresponding
dispersion determined by Jχ and the vector field a~ ð~rÞ is
spin liquids SUð2Þ, Uð1Þ, and Z2 spin liquids. SUð2Þ spin
given by the lattice gauge field aij through
liquids have BðCi Þ ¼ 0 with IGG ¼ SUð2Þ. They are rather
unstable because of the existence of a large amount of gapless  
SUð2Þ gauge field fluctuations. Uð1Þ spin liquids have BðCi Þ ~ri þ ~rj
aij ¼ ð~ri − ~rj Þ · a~ : ð50Þ
pointing in only one direction for all loops Ci . The con- 2
densation of fluxes in one direction provides an Anderson-
Higgs mechanism for SUð2Þ fluxes in directions Thus, the low-energy effective field theory describes non-
perpendicular to BðCÞ and turns the IGG into Uð1Þ. The relativistic spin-1=2 fermions (spinons) coupled to the Uð1Þ
low-energy fluctuations are Uð1Þ gauge field fluctuations. Z2 gauge field (a0 ð~rÞ; a~ ð~rÞ) in the continuum limit.
spin liquids have BðCi Þ pointing in different directions for The other spin liquid state we introduce here is the π-flux
different loops that pass through the same site i. The gauge state (Affleck and Marston, 1988; Kotliar, 1988) on a square
fluctuations are all gapped because the Anderson-Higgs lattice given by Δij ¼ al0 ¼ 0 and
mechanism now applies to fluxes in all directions. A few 
examples of mean-field Ansätze for these three types of spin χ; μ ¼ x;
χ i;iþμ̂ ¼ ð51Þ
liquid states are presented in the following sections. iχð−1Þ ; μ ¼ y:
ix

B. Uð1Þ gauge fluctuations It is easy to see that PðCi Þ ∝ exp ðiπτ3 Þ per square plaquette in
the mean-field Ansatz, i.e., the π-flux state has IGG ¼ Uð1Þ
We briefly discuss the Uð1Þ gauge theory in regard to two and is a Uð1Þ spin liquid.
examples of spin liquids that are believed to exist in nature The zero-flux and π-flux states are physically distinct states
(see Sec. V). The first example is the zero-flux state given in because of their different IGGs. Their mean-field spinon
Eq. (43b), for which Δij ¼ al0 ¼ 0 and χ ij ¼ χ in the mean- dispersions are also qualitatively different. The zero-flux state
field Ansatz. has a mean-field dispersion of E0 ðkÞ ~ ¼ −Jχðcos kx þ cos ky Þ,
It is easy to see that BðCi Þ ≡ 0 and that the IGG of such a qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
whereas the π-flux state has Eπ ðkÞ~ ¼ Jχ cos2 kx þ cos2 ky
QSL is SUð2Þ, i.e., the zero-flux state describes a SUð2Þ spin
liquid. The low-energy fluctuations are SUð2Þ gauge fluctua- with a reduced Brillouin zone. The continuum theory
tions. Here we do not consider the full SUð2Þ gauge fluctua- describes nonrelativistic fermions with a large Fermi surface
tions; we consider only the phase fluctuations of χ ij , i.e., Uð1Þ in the zero-flux state and describes Dirac fermions with four
gauge fluctuations. The consideration of only Uð1Þ gauge Fermi points [k ¼ ðπ=2; π=2Þ] in the π-flux state (Affleck
fluctuations for the zero-flux state can be justified in a slave- and Marston, 1988). The effective continuum theory for the
rotor theory for the Hubbard model (Lee and Lee, 2005) or in π-flux state has the form
a phenomenological Landau Fermi-liquid-type approach for X
spin liquid states near the metal-insulator transition (see LðπÞ ¼ ½ψ̄ þσ ð∂ μ − iaμ Þτμ ψ þσ þ ψ̄ −σ ð∂ μ − iaμ Þτμ ψ −σ ; ð52Þ
Sec. III.B.1). μσ

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-13


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

where μ ¼ 0, 1, 2. The two-component Dirac spinor fields This implies that a Uð1Þ spin liquid is stable at the physical
ψ σ describe two inequivalent Dirac nodes in the spinon value of N ¼ 2 (Nogueira and Kleinert, 2005). Moreover, by
spectrum (Affleck and Marston, 1988). The two effective low- mapping the spinon Fermi surface in ð2 þ 1ÞD to an infinite
energy Lagrangians Lð0Þ and LðπÞ describe two different types set of (1 þ 1)-dimensional chiral fermions, S.-S. Lee (2008)
of spin liquid states that are believed to exist in nature. We argued that an instanton has an infinite scaling dimension for
discuss these states again in Sec. V. any N > 0. Therefore, the QSL phase is stable against
The continuum action L serves as the starting point for instantons, and the noncompact Uð1Þ gauge theory is a good
studying the stability and low-energy properties of spin liquid low-energy description.
states. Integrating out the fermion fields (at each momentum We note that mechanisms other than confinement arising
shell) gives rise to a Maxwellian potential energy term in the from gauge fluctuations may also lead to the instability of
gauge field: Uð1Þ QSLs, such as Amperean pairing (Lee, Lee, and Senthil,
2007) and spin-triplet pairing (Galitski and Kim, 2007)
1 between spinons.
ð▿ × a~ Þ2 ;
2g2 ðΛÞ A nontrivial prediction of the Uð1Þ gauge theory for spin
liquids is that it leads to charge excitations with a soft gap
where gðΛÞ is a running gauge coupling constant in the sense (Ng and Lee, 2007), which can be detected by means of their
of renormalization group theory, which depends on the energy ac conductivities σðωÞ. It has been predicted that σðωÞ ∼ ωα in
or momentum scale Λ. If gðΛÞ → 0 in the low-energy and these spin liquid states, with α ∼ 3.33 in a nonrelativistic spin
long-wavelength limit of Λ → 0, then the gauge fluctuations liquid and α ¼ 2 in a Dirac fermion spin liquid (Potter,
become increasingly weak. The corresponding interaction Senthil, and Lee, 2013). It is expected that this soft gap
between two fermions becomes too weak to bind them and the related charge fluctuations will manifest themselves
together, and the elementary excitations in the spin system most clearly when the system is near the metal-insulator
are spin-1=2 fermionic excitations called spinons. This phe- transition (see Sec. III.B.1).
nomenon is called deconfinement, and the ground state is a Because charge fluctuations will manifest themselves near
filled Fermi sea of spinons. By contrast, if gðΛÞ → ∞ as the metal-insulator transition, spin liquids in “weak” Mott
Λ → 0, then two spinons will always be confined together to insulators become an interesting topic (Senthil, 2008;
form a magnon. This phenomenon is called confinement. In Podolsky et al., 2009; Grover et al., 2010) for investigation.
this case, the mean-field QSL ground state breaks down into a To study the effect of charge fluctuations near the metal-
spin-ordered state because of the strong gauge fluctuations, insulator transition, Lee and Lee (2005) began with the
and magnon excitations are recovered in this ordered state. Hubbard model and developed a Uð1Þ gauge theory with
It is not exactly clear which kinds of mean-field QSL states the help of the slave-rotor representation (Florens and
are stable against gauge fluctuation. It is generally believed Georges, 2004). A number of physical phenomena, including
that Z2 QSL states are stable because Z2 (Ising) gauge theories transport properties (Nave and Lee, 2007) and the Kondo
are deconfining (Fradkin and Shenker, 1979), whereas SUð2Þ effect (Ribeiro and Lee, 2011), have been studied using this
QSL states are unstable because of the presence of large gauge framework. Charge fluctuations correspond to higher-order
fluctuations. The case of Uð1Þ QSL states is more nontrivial. spin ring-exchange terms in terms of the spin Hamiltonian
The SUð2Þ gauge group and the corresponding gauge fields (Misguich et al., 1998; Yang et al., 2010).
are compact in spin liquid states. To reflect the compactness of
the Uð1Þ gauge group, one must replace the electromagnetic 1. Mott transition: Relation between Fermi and spin liquids
field tensor F2μν with 2ð1 − cos Fμν Þ. This periodic version of Zhou and Ng (2013) proposed a different way to understand
Uð1Þ gauge theory is called compact Uð1Þ gauge theory. Uð1Þ spin liquids near the Mott transition. They proposed that
A pure compact Uð1Þ gauge theory always gives rise to spin liquids near the Mott transition can be regarded as “Fermi
confinement in two dimensions (Polyakov, 1975, 1977), liquids” with a constraint imposed on the current operator. For
but whether deconfinement is possible in the presence of a isotropic systems, they observed that the charge current
matter field is an open question. Herbut et al. (2003) carried by quasiparticles,
argued that the theory is always confining in the presence
of a Fermi surface or nodal fermions (Herbut and Seradjeh,  
m Fs1 ð0Þ
2003). Their conclusion depends on an approximate effective J¼  1þ J ; ð53aÞ
m d
action for the gauge field obtained by integrating out the
fermions to the lowest order. However, this approximation is
questionable for gapless fermions. Indeed, Hermele et al. is renormalized by the Landau parameter Fs1 in Fermi liquid
(2004) proved that when the spin index is generalized to N theory, but the thermal current,
flavors, deconfinement arises in the case of 2N two-
component Dirac fermions coupled to complex Uð1Þ gauge m ð0Þ
JQ ¼ J ; ð53bÞ
fields for sufficiently large N, thus providing a counter m Q
example to confinement. Further renormalization group
analysis for compact quantum electrodynamics in ð2 þ 1ÞD is not, where Jð0Þ and JQ ð0Þ are the charge and thermal
shows that deconfinement occurs when N > N c ¼ currents, respectively, carried by the corresponding noninter-
36=π 3 ≃ 1.161, where N is the number of fermion replicas. acting fermions and d is the number of dimensions of the

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-14


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

system. For systems with Galilean invariance, the charge The Lagrangian presented in Eqs. (55) and (57) can be
current carried by quasiparticles is not renormalized, and rewritten in the standard form of Uð1Þ gauge theory by noting
m =m ¼ 1 þ Fs1 =d (Baym and Pethick, 2004). However, this that in this representation, the fermion current is given by
is not valid in general for electrons in crystals, where Galilean
invariance is lost. In this case, m =m ≠ 1 þ Fs1 =d, and the −i X † n
j¼ ½ψ σ ∇ψ σ − ð∇ψ †σ Þψ σ  −  a;
charge current carried by quasiparticles is renormalized 2m σ m
through quasiparticle interaction. In the special case in which
R
1 þ Fs1 =d → 0 while m =m remains finite, J → 0, suggesting where ψ σ ðrÞ ¼ e−ik·r ckσ is the Fourier transform of ckσ . The
that the fermionic system is in a special state wherein spin-1=2 Lagrangian can be written as
quasiparticles do not carry charge due to interaction but still
carry entropy. This is exactly what one expects for spinons in XZ    

QSL states. L¼ dd r ψ †σ i − φ ψ σ − Hðψ †σ ; ψ σ Þ þ Lðφ; aÞ;
σ
∂t
Baym and Pethick noted that the limit of 1 þ Fs1 =d → 0 is a
singular point in Fermi liquid theory and that higher-order ð58aÞ
q- and ω-dependent terms should be included in the Landau
interaction to ensure that finite results are obtained when where
calculating physical response functions. Expanding at small q
and ω, they obtained 1
Hðψ †σ ; ψ σ Þ ¼ jð∇ − iaÞψ σ j2 ð58bÞ
2m
1þ Fs1 ðq; ωÞ=d
∼ α − βω2 þ γ t q2t þ γ l q2l ; ð54Þ and
Nð0Þ
Z    
where qt ∼ ∇× and ql ∼ ∇ are associated with the transverse 1 n d 2 Nð0Þ 2
Lðφ; aÞ ¼ dd r 1 þ a þ φ : ð58cÞ
(curl) and longitudinal (gradient) parts, respectively, of the 2 m Fs1 Fs0
small-~q expansion. In a QSL state, α ¼ 0. They found that to
ensure that the system is in an incompressible (insulator) state, Using Eq. (54), they found that in the small-q limit, the
it is necessary to have γ l ¼ 0. transverse part of Lðφ; aÞ in the spin liquid state is given by
To show that this phenomenology actually describes Z   2 
fermionic spin liquids with Uð1Þ gauge fluctuations, Zhou n ∂a
Lt ðφ; aÞ ¼ − dd r β − γ t ð∇ × aÞ2 : ð59Þ
and Ng (2013) considered a Landau Fermi liquid with 2m ∂t
interaction parameters of Fs0 ðqÞ and Fs1 ðqÞ only. The long-
wavelength and low-energy dynamics of the Fermi liquid are The Lagrangian as expressed in Eq. (58) together with Eq. (59)
described by the following effective Lagrangian: is the standard Lagrangian used to describe QSLs with Uð1Þ
gauge fluctuations. The analysis can be rather straightfor-
X 

 
wardly generalized to a Uð1Þ spin liquid with Dirac fermion
Leff ¼ c†kσ 0 †
i − ξk ckσ − H ðc ; cÞ ; ð55Þ
k;σ
∂t dispersion. The appearance of a soft charge gap in Uð1Þ spin
liquids can be understood from the phenomenological form of
where c†kσ ðckσ Þ is the spin-σ fermion creation (annihilation) Fs1 ðq; ωÞ as expressed in Eq. (54), which suggests that the
operator with momentum k and quasiparticles carry vanishing charges only in the limit of
q; ω → 0. The appearance of a nonvanishing β in Eq. (54) leads
  to an ac conductivity σðωÞ with a power-law form. This picture
1 X Fs1 ðqÞ
H0 ðc† ;cÞ ¼ jðqÞ · jð−qÞ þ F s
0 ðqÞnðqÞnð−qÞ is very different from theories of spin liquid states that start
2Nð0Þ q v2F
from simple spin models in which charge fluctuations are
ð56Þ absent at all energy scales and suggests that charge fluctuations
are important in regions near the Mott transition. We note that
describes the current-current and density-density interactions charge fluctuations can be (partially) incorporated into the spin
between quasiparticles (Larkin, 1964; Leggett, 1965), where models through ring-exchange terms.
q ¼ ðq; ωÞ and vF ¼ ℏkF =m is the Fermi velocity. The close relationship between Fermi liquids and spin
The current and density interactions can be decoupled by liquid states suggests an alternative picture of the Mott metal-
introducing fictitious gauge potentials a and φ (Hubbard- insulator transition with respect to that put forward by
Stratonovich transformation) as follows: Brinkman and Rice (1970), who proposed that a metal-
insulator (Mott) transition is characterized by a diverging
X 
1 n d Nð0Þ 2

effective mass m =m → ∞ and an inverse compressibility
H0 ðc† ; cÞ → j · a þ nφ − 2
a þ s φ ; κ → 0 at the Mott transition point, with a correspondingly
q
2 m Fs1 ðqÞ F0 ðqÞ
vanishing quasiparticle renormalization weight Z ∼ m=
ð57Þ m → 0. The diverging effective mass and vanishing quasi-
particle weight imply that the Fermi liquid state is destroyed at
where n is the fermion density. The equality dðn=m Þ ¼ the Mott transition and that the Mott insulator state is distinct
Nð0Þv2F was used in formulating Eq. (57). from the Fermi liquid state on the metal side.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-15


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

The phenomenology described here suggests an alter- into a gapped QSL at low temperatures of T < T c ðUÞ on the
native picture in which the Fermi surface is not destroyed, insulating side. The nature of the low-temperature QSLs
but the Landau quasiparticles are converted into spinons depends on the microscopic details of the system and cannot
ð1 þ Fs1 =d → 0Þ at the Mott transition. In particular, the be determined based on the above phenomenological
effective mass m =m may not diverge at the metal-insulator considerations.
transition, although Z → 0. A schematic phase diagram for the
Mott (metal-QSL) transition is presented in Fig. 5 for a generic C. Z2 spin liquid states
Hubbard-type Hamiltonian with a hopping integral t and an
on-site Coulomb repulsion U. The system is driven into a Mott An example of a Z2 spin liquid state was first constructed by
insulator state at zero temperature at U ¼ U c , where Wen (1991) for a J1 − J 2 Heisenberg model on a square
1 þ Fs1 ðU > Uc Þ=d ¼ 0. This picture suggests that a Uð1Þ lattice, where J 1 and J2 are the nearest neighbor and next-
spin liquid state is likely to exist in an insulator close to the nearest neighbor Heisenberg interactions, respectively. Wen
Mott transition. considered the mean-field Ansatz
The point 1 þ Fs1 =d ¼ 0 is a critical point in Fermi liquid  
theory called the Pomeranchuk point. The Fermi surface is χ 0
ui;iþμ̂ ¼ ; ð60aÞ
unstable with respect to deformation when 1 þ Fs1 =d < 0. The 0 −χ
criticality of this point implies that the QSLs obtained in this
way are marginally stable because of large critical fluctua- where μ̂ ¼ x̂; ŷ, and
tions. A similar conclusion can be drawn from Uð1Þ gauge
theory by analyzing the Uð1Þ gauge fluctuations. As a result,  
0 Δ0  iΔ1
QSLs with large Fermi surfaces are, in general, susceptible to ui;ix̂þŷ ¼ ui;i∓x̂−ŷ ¼ ; ð60bÞ
the formation of other, more stable QSLs at lower temper- Δ0 ∓ iΔ1 0
atures, such as Z2 QSLs or VBS states that gap out part of or
the entire Fermi surface. This is indicated schematically in the where χ, Δ0 , and Δ1 are nonzero real numbers; a2;3
0 ¼ 0, and
phase diagram shown in Fig. 5(b), where the system is driven a10 ≠ 0. It is easy to check PðCÞ for two loops: C1 ¼ i →
i þ x̂ → i þ x̂ þ ŷ → i and C2 ¼ i → i þ ŷ → i þ ŷ − x̂ → i.
We obtain
1
m (a) PðC1 Þ ¼ χ 2 ðΔ0 τ1 þ Δ1 τ2 Þ ð61aÞ
m*
Z T 0 and

PðC2 Þ ¼ −χ 2 ðΔ0 τ1 − Δ1 τ2 Þ; ð61bÞ


QSL
metal insulator which demonstrates that BðC1 Þ ≠ BðC2 Þ and that the spin
0 liquid state described is a Z2 spin liquid state. The mean-field
U/t ground state describes a half-filled spinon band with a band
1 F1s / d 0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
~ 2 þ ε2 ðkÞ
dispersion given by E ðkÞ ¼  ε1 ðkÞ ~ 2 þ ε3 ðkÞ
~ 2,
T (b) where

critical region
ε1 ðkÞ
~ ¼ 2J 1 χðcosðkx Þ þ cosðky ÞÞ;
~ ¼ 2J 2 Δ0 ½cosðkx þ ky Þ þ cosðkx − ky Þ þ a1 ;
ε2 ðkÞ
QSL with a spinon FS 0
Femi liquid
Tc (U ) ε3 ðkÞ
~ ¼ 2J 2 Δ1 ½cosðkx þ ky Þ − cosðkx − ky Þ:
gapped phase
0
1 F1s (U ) / d 0 U /t Note that the spinon spectrum is fully gapped.
Many other examples of Z2 spin liquid states have been
FIG. 5. (a) Schematic zero-temperature phase diagram for the constructed in the literature. For instance, a nodal gapped Z2
Mott transition. U is the strength of the Hubbard interaction, and t spin liquid state was proposed by Balents, Fisher, and Nayak
is the hopping integral. The electron quasiparticle weight and the (1998) and Senthil and Fisher (2000). The corresponding
quasiparticle charge current ∼1 þ Fs1 =d vanish at the critical mean-field Ansatz includes nearest neighbor and next-nearest
point, whereas the effective mass remains finite. (b) Schematic neighbor hopping as well as d-wave pairing on nearest
phase diagram showing finite-temperature crossovers and pos- neighbor bonds on the square lattice:
sible instability toward gapped phases at lower temperatures.
There exists a (finite-temperature) critical region around U c  
χ1 Δ
where the phenomenology is not applicable. From Zhou and ui;iþx̂ ¼ ; ð62aÞ
Ng, 2013. Δ −χ 1

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-16


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

 
χ1 −Δ
ui;iþŷ ¼ ; ð62bÞ
−Δ −χ 1

and
 
χ2 0
ui;ix̂ŷ ¼ ; ð62cÞ
0 −χ 2

where χ 1 , χ 2 , and Δ are nonzero real numbers; a1;2


0 ¼ 0, and
a30 ≠ 0. The spinon dispersion is given by E ðkÞ ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 εðkÞ ~ 2 þ ΔðkÞ ~ 2 , where
FIG. 6. A Z2 vortex created by flipping the signs of the uij on
the bonds cut by the dashed line (indicated by thick lines).
~ ¼ 2J1 χ 1 ½cosðkx Þ þ cosðky Þ þ 2J 2 χ 2 ½cosðkx þ ky Þ
εðkÞ
þ cosðkx − ky Þ þ a30 ;
~ ¼ 2J1 Δ½cosðkx Þ − cosðky Þ þ a3 ; novel properties from its QSL parent, and new phenomena
ΔðkÞ 0 may also emerge. For instance, it was proposed that doping a
kagome system can give rise to an exotic superconductor with
and is found to be gapless at four k~ points with a linear an hc=4e-quantized flux (as opposed to the usual hc=2e
dispersion. Thus, this spin liquid is a Z2 nodal spin liquid. We quantization) (Ko, Lee, and Wen, 2009).
reiterate that Z2 spin liquid states are expected to be the most
stable because the SUð2Þ gauge fields are gapped and the D. Numerical realization of Gutzwiller projection:
fermionic spins are interacting only through short-range Variational Monte Carlo method and some results
interactions.
It was observed by Wen (1991) that in addition to spinons, The theories of QSL states rely heavily on the reliability of
a soliton-type excitation exists in a Z2 spin liquid. Gutzwiller-projected wave functions. In this section, we
This excitation is nothing but a π flux in the Z2 gauge field, discuss how Gutzwiller projection is performed numerically
called a “Z2 vortex.” This Z2 vortex can be described by a new in practice and how the physical observables can be evaluated
mean-field Ansatz, using a Monte Carlo method for a given projected wave
function jΨRVB i ¼ PG jΨMF i.
Two types of mean-field Ansatz are frequently used in
u~ ij ¼ uij Θij ; constructing QSL states. The first one contains only (fer-
mionic) spinon hopping terms χ, and the mean-field ground
where Θij ¼ 1 generates a π flux on a lattice. One possible state is a half-filled Fermi sea. The second one includes both
choice of Θij is illustrated in Fig. 6, where Θij ¼ −1 on the hopping terms and pairing terms Δ, and the mean-field
bonds cut by the dashed line and Θij ¼ 1 on the other bonds. ground state is a BCS-type state with a fermion energy gap.
An interesting consequence of such a Z2 vortex is that the These two types of wave functions describe Uð1Þ and Z2
statistics of a spinon can be changed from bosonic to fermionic spin liquid states, respectively, with the proper choice of
and vice versa if it is bound to a vortex. Therefore, Z2 spin hopping and pairing parameters. For a given spin
liquids may contain charge-neutral spin-1=2 spinons with both Hamiltonian, we can determine these hopping and pairing
bosonic and fermionic statistics (Ng, 1999). The dynamics of parameters by optimizing the ground-state energy. Therefore,
Z2 vortices can give rise to interesting physical consequences this approach is called the variational Monte Carlo (VMC)
(Ng, 1999; Qi, Xu, and Sachdev, 2009). method.
It is worth noting that the J 1 − J 2 model on a square lattice For a projected Fermi sea state, the mean-field ground-state
has been well studied. The lowest-energy Z2 spin liquid state wave function on a lattice with N sites can be constructed by
is a nodal spin liquid with four Dirac points (Capriotti et al., filling the N lowest states in the mean-field band:
2001; Hu et al., 2013), labeled as Z2Azz13 in the projected
group symmetry classification scheme (Wen, 2002), which we YY
N=2

discuss in Sec. III.E. This nodal Z2 spin liquid is energetically jΨFS i ¼ ψ †kσ j0i;
σ k¼1
competitive with calculations performed using the DMRG
(Jiang, Yao, and Balents, 2012; Gong et al., 2014) and PEPS
where σ ¼ ↑; ↓ is the spin index and the states are sorted in
(Wang et al., 2013) approaches.
Relation to superconductivity: RVB theory was developed order of ascending energy E1 ≤    ≤ EN=2 < EF. ψ †kσ creates
not only for QSLs but also for high-T c superconductivity an eigenstate in the mean-field band and can be expressed as
(Anderson, 1987). It is generally believed that Z2 spin liquid X
states may become superconductors upon doping (Lee, ψ †kσ ¼ ak ðiÞc†iσ ;
Nagaosa, and Wen, 2006). The superconducting state inherits i

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-17


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

X
where each value of i denotes a site and c†iσ is a local fermion PG jΨBCS i ¼ sgnði1 ; …; iN=2 ; j1 ; …; jN=2 Þ
creation operator. The eigenstate wave function ak ðiÞ does not fσ i g
depend on the spin index σ for spin-singlet states because of × det½wði1 ; …; iN=2 ; j1 ; …; jN=2 Þ
the spin rotational symmetry. More explicitly,
× jσ 1 ; …; σ N i; ð68Þ

N=2 X
YY N  where jσ 1 ; …; σ N i is a state in the Ising basis with N=2
jΨFS i ¼ ai ðjÞc†jσ j0i; ð63Þ up spins located at sites i1 ; …; iN=2 and N=2 down spins
σ i¼1 j¼1
located at sites j1 ; …; jN=2 and wði1 ; …; iN=2 ; j1 ; …; jN=2 Þ is
an N=2 × N=2 matrix given by
and the Gutzwiller-projected wave function can be written in 0 W
terms of the product of three factors: i1 j 1  W i1 jN=2 1
B .. C
wði1 ; …; iN=2 ; j1 ; …; jN=2 Þ ¼ B
@  .    A:
C
X
PG jΨFS i ¼ sgnfi1 ;…;iN=2 ;j1 ;…;jN=2 gdet½Aði1 ;…;iN=2 Þ W iN=2 j1    W iN=2 jN=2
fσ i g
ð69Þ
× det½Aðj1 ;…;jN=2 Þjσ 1 ;…;σ N i; ð64Þ
A key observation regarding these two projected wave
where jσ 1 ; …; σ N i is a state in the Ising basis with N=2 up functions, Eqs. (64) and (68), is that both of them can be
spins located at sites i1 ; …; iN=2 and N=2 down spins located written as a determinant or as a product of two determinants.
This allows us to numerically evaluate a projected wave
at sites j1 ; …; jN=2 ; sgnfi1 ; …; iN=2 ; j1 ; …; jN=2 g is the sign of
function. For a large system, the number of degrees of
the permutation P ¼ fi1 ; …; iN=2 ; j1 ; …; jN=2 g, and
freedom increases exponentially with the system size. In this
Aði1 ; …; iN=2 Þ is an N=2 × N=2 matrix given by case, the Monte Carlo method is applied to evaluate the
energy, magnetization, and spin correlation for these projected
0 1 wave functions (Horsch and Kaplan, 1983; Gros, 1989). Next
a1 ði1 Þ  a1 ðiN=2 Þ
we briefly describe how the MC method works. Readers who
B .. C
Aði1 ; …; iN=2 Þ ¼ B
@  . 
C:
A ð65Þ are interested in the details can see Gros (1989).
The expectation value of an operator Θ in a system with the
aN=2 ði1 Þ    aN=2 ðiN=2 Þ spin wave function jΨi can be written as

hΨjΘjΨi X hΨjαihβjΨi
A BCS-type mean-field ground state with spin-singlet hΘi ¼ ¼ hαjΘjβi ; ð70Þ
hΨjΨi α;β
hΨjΨi
pairing can be written as
P where the spin configurations jαi and jβi are states in the Ising
ð1=2Þ W ij ðc†i↑ c†j↓ −c†i↓ c†j↑ Þ
jΨBCS i ¼ e i;j j0i; ð66Þ basis with N=2 up spins and N=2 down spins. This sort of
expectation value is recognized to be amenable to a MC
where i and j are site indices and W ij ¼ W ji for fermionic evaluation (Horsch and Kaplan, 1983). The expectation value
spin-singlet pairing. For a system with lattice translational expression given in Eq. (70) can be rewritten as
symmetry, W ij can be written explicitly as
XX hαjΘjβihβjΨi jhαjΨij2 X
X vk hΘi ¼ ¼ fðαÞρðαÞ;
hαjΨi hΨjΨi
W ij ¼ − e−ik·ðRi −Rj Þ ; α β α
uk
k ð71Þ

where uk and vk are given in the BCS form. In the more with
general situation in which lattice translational symmetry is
lost, the W ij ’s are determined from the Bogoliubov–de X hαjΘjβihβjΨi jhαjΨij2
Gennes equations. Gutzwiller projection retains only states fðαÞ ¼ ; ρðαÞ ¼ :
β
hαjΨi hΨjΨi
with a number of electrons equal to the number of lattice
sites and removes all terms with more than one electron per
It follows that
site, i.e.,
X
X N=2 ρðαÞ ≥ 0; ρðαÞ ¼ 1:
jΨRVB i ¼ PG i<j
W ij c†i↑ c†j↓ j0i: ð67Þ α

Note that for a “local operator” Θ (e.g., Θ ¼ S~ i · S~ j ) and a


In the spin representation, the projected BCS state can be given spin configuration jαi, only a limited number of
written as “neighbor” configurations jβi give rise to a nonvanishing

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-18


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

hαjΘjβi. As noted by Horsch and Kaplan (1983), the compu- TABLE I. Comparison of ground-state energy and spin suscep-
tation time for the ratio hβjΨi=hαjΨi is of OðN 2 Þ. Therefore, tibility in one dimension. The first row shows the results for the
hΘi can be evaluated by means of a random walk through spin projected Fermi sea. The second row shows the results for the exact
ground state of the Heisenberg model. From Gros, 1989.
configuration space with weight ρðαÞ. As in the standard MC
method, the probability Tðα → α0 Þ of transitioning from one hS~ i · S~ iþ1 i χ
configuration α to another configuration α0 can be chosen as
follows: Gutzwiller −0.442 118 (Gebhard 0.058  0.008 (Gros, Joynt,
and Vollhardt, 1987) and Rice, 1987)
 Exact −0.443 147 (Lieb 0.0506 (Griffiths, 1964)
1; ρðα0 Þ ≥ ρðαÞ; and Wu, 1968)
Tðα → α0 Þ ¼
ρðα0 Þ=ρðαÞ; ρðα0 Þ < ρðαÞ:

The new configuration α0 is accepted with probability down spins, respectively, in the wave function. The lowest-
Tðα → α0 Þ. energy state in the subspace with Sz ¼ m is given by
Because hαjΨi is either a determinant or a product of two
Y Y
determinants, the computation time for hαjΨi is of OðN 3 Þ. PG jΨm i ¼ PG ψ †k↑ ψ †k↑ j0i; ð72Þ
The computational resource consumption for the MC weight jkj≤kF↑ jkj≤kF↓
factor Tðα → α0 Þ is not too high, and consequently, this MC
method is feasible for Gutzwiller projection. Moreover, the where kFσ ¼ πðN σ − 1Þ=N ¼ πðN σ − 1Þ=ðN ↑ þ N ↓ Þ. With
computation time of the ratio Tðα → α0 Þ can be reduced to the help of this trial wave function, the spin susceptibility
OðN 2 Þ if the corresponding matrix Aðα0 Þ or wðα0 Þ in Eq. (65) χ can be calculated (Gros, Joynt, and Rice, 1987). It was found
or (69) differs from AðαÞ or wðαÞ by only one row or column. that χ is close to the value obtained from the exact solution
This can be achieved by properly choosing the spin update (Griffiths, 1964). The numerical results are summarized in
procedure, e.g., the interchange of two opposite spins. This Table I.
algorithm was first introduced by Ceperley, Chester, and
Kalos (1977) for the MC evaluation of a fermionic trial wave
function. 2. Triangular lattice
As a variational method, the VMC method not only yields Historically, the AFM spin-1=2 Heisenberg Hamiltonian on
an upper bound on the ground-state energy for a spin a triangular lattice was the first model to be proposed for the
Hamiltonian but also provides detailed information on the microscopic realization of a spin liquid ground state (Fazekas
trial ground state. This information is useful for understand- and Anderson, 1974). However, the minimum-energy con-
ing the nature of the ground-state wave function. In the figuration for the classical Heisenberg model on a triangular
remainder of this section, we discuss some numerical results lattice is well known to be the 120° Néel state. There has been
regarding Gutzwiller-projected wave functions on one- and a long-standing debate regarding whether the frustration
two-dimensional frustrated lattices. together with quantum fluctuations could destroy the long-
range 120° Néel order, leading to a spin liquid state. Many trial
1. One-dimensional lattice wave functions have been proposed as the ground state of the
One-dimensional systems usually serve as benchmarks for nearest neighbor Heisenberg model on a triangular lattice,
comparison because exact solutions are often available. It including a chiral spin liquid state (Kalmeyer and Laughlin,
turns out that PG jΨFS i, which is gauge equivalent to 1987) and 120°-Néel-order states with quantum mechanical
PG jΨBCS i in one dimension, is an excellent trial wave function corrections (Huse and Elser, 1988; Sindzingre, Lecheminant,
for the ground state of the one-dimensional Heisenberg model. and Lhuillier, 1994). Capriotti, Trumper, and Sorella (1999)
The energy for PG jΨFS i is higher than that of the exact ground utilized the Green’s function Monte Carlo method with the
state by only 0.2% (Gebhard and Vollhardt, 1987; Gros, Joynt, stochastic reconfiguration technique to obtain the state of the
and Rice, 1987; Yokoyama and Shiba, 1987). The spin-spin model with the lowest energy (to our knowledge, the ground-
correlation decays following a power law at large distances, state energy per site is 0.5458  0.0001), which exhibits 120°
long-range Néel order. More recently, the three-sublattice 120°
hS~ i · S~ iþr i ∼ ð−1Þr =jrj, consistent with the results obtained
Néel order was further confirmed by DMRG (White and
through bosonization (Luther and Peschel, 1975). Indeed, it
Chernyshev, 2007).
was shown that this Gutzwiller-projected wave function is the
It thus seemed that for a triangular lattice, the possibility of
exact ground state of the Haldane-Shastry model (Haldane,
a spin liquid state had been ruled out. However, the story
1988a; Shastry, 1988),
continues. It was found that a four-spin ring exchange
stabilizes the projected Fermi sea state against a long-range
JX N XN −1
1
HH-S ¼ S~ · S~ ; AFM state (Motrunich, 2005). Because multispin ring
2 i¼1 r¼1 sin2 ðπr=NÞ i iþr exchange reflects the charge fluctuations in the vicinity of
the Mott transition, this result provides theoretical support for
which describes an AFM Heisenberg chain with long-range the search for spin liquid states in a Mott insulating state close
coupling (a periodic version of 1=r2 exchange). to the metal-insulator transition.
Excited states with Sz ¼ m ¼ ðN ↑ − N ↓ Þ=2 can also be The model Hamiltonian that contains both nearest neighbor
constructed, where N ↑ and N ↓ are the numbers of up and Heisenberg exchange and four-spin ring exchange is

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-19


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Quantum spin liquid spins of a triangle at 120° from each other in a plane, but it does
AF not fix the relative orientation of the plane of one triad with
??? Projected Fermi sea respect to the planes of the triads on neighboring triangles.
These degrees of freedom lead to a continuous local degeneracy
~ 0 .14 J ring / J
of the ground states. Note that this degeneracy exists even if we
restrict ourselves to coplanar spin states. Two of the simplest
FIG. 7. Variational phase diagram for the Hamiltonian presented
examples (Sachidev, 1992) are the three sublattice planar states
in Eq. (73). From Motrunich, 2005. pffiffiffiffiffiffipffiffiffi
shown in Fig. 8 for the q ¼ 0 and 3× 3 ordered states.
The large classical ground-state degeneracy must be lifted
by quantum fluctuations. The nature of the ground state for the
quantum model is highly speculative because of the enormous
degeneracy in the classical model. Many arguments have been
presented in the literature regarding what kind of ground state
ð73Þ is favored, and this issue is still under debate (Diep, 2004).
In the following, we discuss the Uð1Þ QSL state, which is one
of the promising candidates for the ground state of a spin-1=2
where P12 ¼ 2S~ 1 · S~ 2 þ 1=2 interchanges the two spins at
Heisenberg antiferromagnet on a kagome lattice.
sites 1 and 2 and the four-spin exchange operators satisfy the
Inspired by neutron-scattering experiments on herbertsmi-
following relations: P†1234 ¼ P4321 and P1234 þ P4321 ¼ thite, ZnCu3 ðOHÞ6 Cl2 , Ran et al. (2007) constructed a series
P12 P34 þ P14 P23 − P13 P24 þ P13 þ P24 − 1. of variational wave functions of Uð1Þ spin liquids on a
By comparing the trial energies of the AFM-ordered states kagome lattice. The corresponding mean-field Ansatz involves
proposed by Huse and Elser (1988) with those of various only fermionic spinon hopping on nearest neighbor bonds:
fermionic spin liquid states, Motrunich (2005) found that the
ring-exchange term favors a spin liquid ground state over the X
HMF ¼ J ðχ ij f†jσ fiσ þ H:c:Þ;
AFM-ordered state. The results are summarized in Fig. 7. For
hijiσ
small ring exchange, i.e., J ring =J ≲ 0.14, the ordered states are
of lower energy. However, for J ring =J ≳ 0.14, spin liquid states where the complex field χ ij lives on the links between two
are energetically favored. For larger values of J ring =J≳ neighboring sites. For a kagome lattice, the mean-field states
0.30.35, the optimal spin liquid state is the projected are characterized by the Uð1Þ gauge fluxes through the
Fermi sea state. In the intermediate regime, optimized wave triangles and hexagons. Large-N expansion suggests several
functions with extended anisotropic s-wave, dx2 −y2 , and
dx2 −y2 þ idxy spinon pairings have similar energies.
Recently, a novel Z2 spin liquid state on a triangular lattice
q 0
A A
was proposed, where the paired fermionic spinons preserve all
symmetries of the system and the system has a gapless
A C B C B C
excitation spectrum with quadratic bands that touch at
q ¼ 0. It was shown through the VMC method that this Z2
spin liquid state has a highly competitive energy when J ring =J A A A
is realistically large (Mishmash et al., 2013).
A C B C B C B
3. Kagome lattice
Unlike the case of a triangular lattice, the classical A A
Heisenberg model on a kagome lattice has an infinite number
of degenerate ground states that are connected to one another by B C
continuous local distortions of the spin configuration (Villain 3x 3
A A
et al., 1980). This property holds on any lattice with corner-
sharing units, such as checkerboard, kagome, and pyrochlore
A C B A C B
lattices (Moessner and Chalker, 1998). For instance, on a
kagome lattice formed by corner-sharing triangles, the nearest
neighbor Heisenberg Hamiltonian can be written as the sum of B C A
the squares of the total spins S~ △ ¼ S~ 1 þ S~ 2 þ S~ 3 of individual
triangles that share only one vertex: A C B A C B
X
H¼J ðS~ △ Þ2 : A B

pffiffiffi pffiffiffi
FIG. 8. Two classical planar Néel states (q ¼ 0 and 3 × 3)
Classical ground states are obtained whenever S~ △ ¼ 0. This on a kagome lattice. A, B, and C specify three coplanar spin
triangle rule fixes the relative orientations of the three classical orientations with intersection angles of 120°.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-20


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

candidate mean-field states (Marston and Zeng, 1991; 2013; Iqbal, Poilblanc, and Becca, 2014) found that the
Hastings, 2000): (i) VBS states, which break translation gapless Uð1Þ-Dirac spin liquid is competitive with gapped
symmetry; (ii) a spin liquid state (SL-½π=2; 0) with a flux Z2 spin liquids. By performing a finite-size extrapolation of
of þπ=2 through each triangle on the kagome lattice and zero the ground-state energy, they obtained an energy per site of
flux through the hexagons, which is a chiral spin liquid state E=J ¼ −0.4365ð2Þ, which is within three error bars of the
that breaks time-reversal symmetry; (iii) a spin liquid state estimates obtained using the DMRG method. In summary, the
(SL-½π=2; 0) with staggered π=2 fluxes through the trian- Uð1Þ-Dirac spin liquid state has proven to be a good candidate
gles (þπ=2 through up triangles and −π=2 through down for describing a critical phase on a kagome lattice.
triangles) and zero flux through the hexagons; (iv) a spin
liquid state (SL-½π=2; π) with a flux of þπ=2 flux through
E. Classification of spin liquid states: Quantum orders
each triangle and a flux of π through each hexagon; (v) a and projective symmetry groups
uniform RVB spin liquid state (SL-[0, 0]) with zero flux
through both triangles and hexagons, which has a spinon The use of Gutzwiller-projected wave functions can be
Fermi surface; and (vi) a Uð1Þ-Dirac spin liquid state made more systematic by using a powerful approach based on
(SL-½0; π) with zero flux through the triangles and a flux classifying spin liquid states according to their symmetry
of π through each hexagon, which has four flavors of two- properties. For classical systems, it was observed by Landau
component Dirac fermions. that symmetry is a universal property shared by all macro-
By performing VMC calculations on 8 × 8 × 3 and 12 × scopic states within the same phase, irrespective of micro-
12 × 3 lattices, Ran et al. (2007) found that the Uð1Þ-Dirac scopic details. Consequently, the symmetry (or broken
spin liquid state (SL-½0; π) has the lowest energy among states symmetry) associated with classical order parameters serves
(i)–(vi) listed after Gutzwiller projection, with a ground-state as a powerful tool for characterizing different classical phases.
energy of −0.429J per site. Note that there is no tunable This approach can be generalized to quantum spin systems
parameter in this Uð1Þ-Dirac spin liquid state. This energy is described by Gutzwiller-projected wave functions, with addi-
remarkably favorable because the value is very close to the tional constraints.
exact diagonalization result when extrapolated to the thermo- For spin liquid states described by Gutzwiller-projected
dynamic limit. A comparison among the ground-state energies wave functions, one might expect that the quantum phases
determined using this VMC method and other numerical could be classified according to the symmetry properties of the
methods is presented in Table II. They also found that the mean-field Ansatz ðuij ; al0 τl Þ. However, the usual classical
Uð1Þ-Dirac spin liquid state is stable against VBS ordering
symmetry group (SG) is insufficient for classifying these
and chiral spin liquid states with fluxes of θ through the
states for two reasons: (i) Because of the gauge redundancy,
triangles and π − 2θ through the hexagons. The spin corre-
different mean-field descriptions exist for the same QSL state.
lation functions exhibit algebraic decay with distance because
For instance, the uniform RVB state and the zero-flux state
of the Dirac nodes in the spinon spectrum.
correspond to the same spin state, and the d-wave RVB state
We note that exact diagonalization (Leung and Elser, 1993;
on a square lattice is also the π-flux state. (ii) QSL states may
Lecheminant et al., 1997; Mila, 1998; Waldtmann et al., 1998)
have inherent (phase) structures contained in the mean-field
and DMRG calculations (Jiang, Weng, and Sheng, 2008; Yan,
Huse, and White, 2011; Depenbrock, McCulloch, and Ansatz ðuij ; al0 τl Þ that cannot be fully distinguished based on
Schollwöck, 2012; Jiang, Wang, and Balents, 2012) strongly the SG constructed for classical systems. To address this issue,
indicate the existence of a spin gap and seem to rule out the Wen (2002) proposed a new mathematical object called a
Uð1Þ-Dirac spin liquid scenario. However, this disagreement projective symmetry group (PSG), which generalizes
may be a finite-size effect. The applicability of exact diag- Landau’s approach and has now become an important tool
onalization is limited to very small lattices of up to 36 sites, in studying QSLs and the quantum phase transitions between
and the maximum cylinder circumference used in the DMRG different QSL states.
approach is only 17 lattice spacings. Recently, through the Wen proposed that the symmetry of the mean-field Ansatz
combination of the Lanczos algorithm for projected fermionic ðuij ; al0 τl Þ is a universal property and serves as a kind of
wave functions with the Green’s function Monte Carlo “quantum number” that can be used to characterize quantum
technique, Iqbal, Becca, Sorella, and Poilblanc (Iqbal et al., orders in QSLs. The macroscopic properties of the Ansatz are
characterized by its PSG. An element of a PSG is a combined
operation consisting of a symmetry transformation U followed
TABLE II. Comparison of the ground-state energies (in units of J)
by a local gauge transformation GU ðiÞ. The PSG of a given
determined using different methods for the nearest neighbor
Heisenberg model on a kagome lattice. In the VMC method, the mean-field Ansatz consists of all combined operations that
Uð1Þ-Dirac spin liquid state (SL-½0; π) is used. leave the Ansatz unchanged, i.e.,

Method Energy per site


PSG ≡ fGU jGU Uðuij Þ ¼ uij ; GU ðiÞ ∈ SUð2Þg; ð74Þ
Exact diagonalization −0.43 (Waldtmann et al., 1998)
Coupled cluster −0.4252 (Farnell, Bishop,
method and Gernoth, 2001) where Uðuij Þ ¼ u~ ij ≡ uUðiÞ;UðjÞ , GU Uðuij Þ ≡ GU ðiÞu~ ij G†U ðjÞ,
Spin-wave variational −0.419 (Arrachea, Capriotti, U generates the symmetry transformation (SG), and GU is the
method and Sorella, 2004) associated gauge transformation. From this definition, it is
VMC method −0.429 (Ran et al., 2007)
easy to see that

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-21


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

SG ≡
PSG
: uij ¼ iρij W ij ; ρij ¼ real number; W ij ∈ SUð2Þ:
IGG
ð76Þ
The PSGs of two mean-field Ansätze related by
a gauge transformation W are also related. From We end with a brief discussion of an issue related to
techniques for the classification of PSGs. For any two given
WGU Uðuij Þ ¼ Wðuij Þ, where Wðuij Þ ≡ W i uij W †j , we obtain
symmetry transformations, their corresponding PSG elements
WGU UW −1 Wðuij Þ ¼ Wðuij Þ. Therefore, if GU U belongs to must satisfy certain algebraic relations determined by the
the PSG of the mean-field Ansatz uij , then WGU UW −1 symmetry transformations. Solving these equations allows us
belongs to the PSG of the gauge-transformed Ansatz to construct a PSG of a type called an algebraic PSG. The
Wðuij Þ. We see that the gauge transformation GU associated name algebraic PSG is introduced to distinguish such PSGs
with the transformation U changes in the following way under from the invariant PSGs defined earlier. Any invariant PSG is
an SUð2Þ gauge transformation W: an algebraic PSG; however, an algebraic PSG is not neces-
sarily an invariant PSG unless there exists an Ansatz such that
GU ðiÞ → WðiÞGU ðiÞW(UðiÞ)† : ð75Þ the algebraic PSG is the total symmetry group of that Ansatz.
To provide an example, we again consider translations. The
Wen proposed that mean-field Ansätze with different PSGs two translation elements T x and T y satisfy the following
belong to different classes of QSL states, just as classical relation:
states with different SGs belong to different classical phases.
As examples, we consider the PSGs of the zero-flux state T x T y T −1 −1
x T y ¼ 1: ð77Þ
given in Eq. (43b) and the π-flux state given in Eq. (51) on a
square lattice. For illustration, let us consider the PSG From the definition of a PSG, we find that the two PSG
associated with translational symmetry. First, we consider elements Gx T x and Gy T y must satisfy the algebraic relation
the zero-flux state. The mean-field Ansatz given in Eq. (43b) is
invariant under the translation transformations T x ði → i þ x̂Þ Gx T x Gy T y ðGx T x Þ−1 ðGy T y Þ−1
3
and T y ði → i þ ŷÞ and the gauge transformation GðθÞ ¼ eiθτ . ¼ Gx T x Gy T y T −1 −1 −1 −1
x Gx T y Gy
The elements of the PSG have the form GU U; GU ¼ GðθÞ
¼ Gx ðiÞGy ði − x̂ÞG−1 −1
x ði − ŷÞGy ðiÞ ∈ G; ð78Þ
and U ¼ ðT x Þn ðT y Þm , where n and m are arbitrary integers.
The π-flux state is different. The mean-field Ansatz given in where we denote the IGG by G. Each solution (Gx T x ; Gy T y ) of
Eq. (51) breaks translational symmetry in the x direction Eq. (78) is an algebraic PSG for T x and T y . By adding other
because of the odd number of lattice sites. Thus, we naively
symmetry transformations, we can find and classify all
expect that the PSG should consist of elements GU U with
algebraic PSGs associated with a given symmetry group.
GU ¼ GðθÞ and U ¼ ðT x Þ2n ðT y Þm . However, this is incor- Because an invariant PSG is always an algebraic PSG, we can
rect because the two mean-field Ansätze, check whether an algebraic PSG is an invariant PSG by
 constructing an explicit Ansatz uij . If an algebraic PSG
χ; μ ¼ x; supports an Ansatz uij with no additional symmetries, then
χ i;iþμ̂ ¼
iχð−1Þix ; μ ¼ y; it is an invariant PSG. Through this method, we can classify
symmetric spin liquids in terms of PSGs.
and Wen (2002) utilized PSGs to classify QSL states with spin
rotational symmetry, time-reversal symmetry, and all lattice

χ; μ ¼ x; symmetries on a square lattice. Later, the PSG classification
χ i;iþμ̂ ¼ ix þ1 approach for symmetric QSLs was applied to triangular (Zhou
iχð−1Þ ; μ ¼ y;
and Wen, 2002), star (Choy and Kim, 2009), and kagome (Lu,
Ran, and Lee, 2011) lattices. The PSG classification scheme
are actually related by a gauge transformation W i ¼ ð−1Þiy τ0
can also be generalized to bosonic QSL states (Wang and
and correspond to the same physical spin state. As a result, the
Vishwanath, 2006; Wang, 2010b) and to QSL states that break
transformations GU0 U0 with GU0 ¼ GðθÞð−1Þiy τ0 and U 0 ¼
spin rotational symmetry and/or time-reversal symmetry (Kou
ðT x Þ2nþ1 ðT y Þm are also elements of the PSG for the π-flux and Wen, 2009; Bieri, Lhuillier, and Messio, 2016).
state. The zero-flux state and the π-flux state have different
PSGs and therefore belong to different classes of Uð1Þ QSL IV. BEYOND RVB APPROACHES
states.
More generally, other lattice symmetry operations (reflec- There are many reasons to go beyond the simple RVB
tions and rotations), such as the parity transformations approach for S ¼ 1=2 spin systems, for example, the discov-
Pxy (ðix ; iy Þ → ðiy ; ix Þ) and Pxȳ (ðix ; iy Þ → ð−iy ; −ix Þ) on a ery of a plausible spin liquid state in a spin-S ¼ 1 system
square lattice, the spin rotation transformation and the (Zhou et al., 2011) and the rise in interest in Mott insulators in
time-reversal transformation, are also considered when con- systems with strong spin-orbit coupling where rotational
structing PSGs, in addition to translations. The spin rotational symmetry is broken and the ground state cannot be a pure
symmetry of spin liquid states requires the mean-field Ansatz spin singlet (Jackeli and Khaliullin, 2009). What is the nature
to take the form of the spin liquid states in these systems? More importantly,

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-22


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

we are interested in the possibility of exotic spin liquid states at T > T c , where there exists a finite spinon Fermi surface,
beyond the RVB description, where the elementary excitations and that a spinon pairing gap characterized by Hpair opens up
may possess exotic properties beyond the simple spinon at T < T c . The power-law behavior CV ∝ T 2 that is observed
picture. at low temperatures of T < T c indicates that the gap has line
We introduce some of these developments in this section. nodes on the Fermi surfaces. To determine the pairing
We start by introducing the generalization of the RVB symmetry, Zhou et al. noted that a group theoretical analysis
approach to spin systems with strong spin-orbit coupling indicates that a spin-triplet pairing state on a cubic lattice can
and to S > 1=2 spin systems in Secs. IV.A and IV.B, followed create only full or point nodal gaps (Sigrist and Ueda, 1991),
by the introduction of matrix product states and projected which seems to imply singlet pairing. However, because of the
entangled pair states in Sec. IV.C, which are completely broken inversion symmetry on a hyperkagome lattice (Hahn,
different ways of constructing spin wave functions compared 1996), the spin-singlet and spin-triplet pairing states are, in
with the RVB approach. We end this section with an general, mixed together in the presence of spin-orbit coupling
introduction to the Kitaev honeycomb model, which repre- (Gor’kov and Rashba, 2001; Frigeri et al., 2004).4
sents yet another different approach to constructing spin wave In terms of the d vector, the gap function Δαβ ðkÞ
functions in a system with strong spin-orbit coupling with (α; β ¼ ↑; ↓) has the general matrix form (Leggett, 1975),
exotic properties beyond the simple spinon picture.
ΔðkÞ ¼ i½d0 ðkÞσ 0 þ dðkÞ · σσ y ; ð79Þ
A. RVB and its generalization to spin systems with strong
spin-orbit coupling
and the spinon pairing must be a singlet or a singlet-with-
Strong spin-orbit coupling may cause interesting exper- triplet admixture because of spin-orbit coupling in order to
imental consequences that are absent in systems with spin have line nodes (Zhou et al., 2008).
rotational symmetry. An example suggested by Zhou et al. We now consider the spin susceptibility of such mixed
(2008) is presented here, in which strong spin-orbit coupling states. Zhou et al. showed that if both singlet and triplet
in Ir atoms is used to explain the anomalous behavior of the pairings are present and the spin-orbit scattering is much
Wilson ratio observed in Na4 Ir3 O8 , which was experimentally weaker than the pairing gap Δ, then the k-dependent electronic
proposed (Okamoto et al., 2007) as the first candidate for a 3D contribution to the spin susceptibility is given by
QSL on a hyperkagome lattice with fermionic spinons.
χ ii ðkÞ d d þ di di d0 d0 þ di di
Although the Curie-Weiss constant is estimated to be as ¼ 1 − 0 0 þ Yðk; TÞ;
large as θW ∼ 650 K in Na4 Ir3 O8 , indicating strong AFM χ N ðkÞ d0 d0 þ d · d d0 d0 þ d · d
coupling, there is no observed thermodynamic and magnetic
anomaly indicative of long-range spin ordering down to 2 K. where i ¼ x, y, z; χ N is the normal state contribution at Δ ¼ 0,
The specific heat ratio γ ¼ CV =T shows a rather sharp peak at and Yðk; TÞ is the k-dependent Yosida function (Leggett,
a temperature of T c ∼ 20 K, indicating the existence of a 1975). Under the assumption that the d vector is pinned by the
phase transition or crossover at T c . By contrast, the spin lattice, for a polycrystalline sample, one must average over all
susceptibility χðTÞ is nearly independent of temperature for all spatial directions, resulting in
temperatures T ≪ θW . Using the experimental values of the  
spin susceptibility χ and the specific heat ratio γ at the specific χs 2 2 jd0 j2 1 2 jd0 j2
¼ − þ þ YðTÞ; ð80Þ
heat peak at ∼20 K, for T > T c, the Wilson ratio RW ¼ χ N 3 3 jd0 j2 þ jdj2 3 3 jd0 j2 þ jdj2
π 2 k2B χ=3μ2B γ of the material is 0.88, which is very close to that
of a Fermi gas where RW is unity. Therefore, for a wide range where YðTÞ is the (spatially averaged) Yosida function, which
of temperatures T c < T < θW , the system seems to behave as vanishes at zero temperature; χ s is the spin susceptibility
a Fermi liquid of spinons. Below T c, the specific heat below T c, and χ N is the Pauli spin susceptibility in the normal
decreases to zero as CV ∼ T 2 , suggesting a line nodal gap state. Therefore, χ s =χ N reduces to
in the low-lying quasiparticle spectrum. However, this picture
needs to be reconciled with the observation that the spin 2 2 jd0 j2

susceptibility χ remains almost constant, resulting in an 3 3 jd0 j2 þ jdj2
anomalously large Wilson ratio of RW ≫ 1 at temperatures
of T < T c . at zero temperature. If the spin-triplet pairing dominates, then
The spins in Na4 Ir3 O8 originate from the low-spin 5d5 Ir4þ χ s =χ N → 2=3, whereas if the spin-singlet pairing dominates,
ions, which form a 3D network in the form of a corner-sharing
hyperkagome lattice. Chen and Balents (2008) suggested that 4
In general, for a many-spin system in which spin rotational
because of the large atomic number, the spin-orbit coupling in symmetry is broken, the spin-S ¼ 0 state(s) will mix with spin-S ≥ 1
Ir atoms is expected to be strong. In the following section, we states even in the presence of spatial inversion symmetry. The only
explain the anomalous Wilson ratio based on a modified RVB exception is the two-spin system, in which inversion symmetry
spin liquid picture in which both spin-singlet and spin-triplet provides a good quantum number that separates the spin-singlet state
pairings exist in the spin-pairing wave function. from the spin-triplet states. Because the RVB approach begins from
Based on the experimental observations discussed earlier, mean-field spin wave functions that are superpositions of two-spin
Zhou et al. (2008) proposed that a simple spinon hopping pairing states, broken inversion symmetry is needed for the con-
Hamiltonian H0 determines the physics of the spin liquid state struction of mixed spin-singlet and spin-triplet states.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-23


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

then χ s =χ N → 0. However, neither of these cases is observed where J ⊥ ; J z > 0. This model can be mapped to the isotropic
in experiments; instead, χ changes only negligibly below T c (XXX) Heisenberg model with the DM interaction,
(Okamoto et al., 2007). This suggests that strong spin-orbit
X
coupling is needed to explain the absence of a marked change D · ðSi × Siþ1 Þ;
in χ below T c ∼ 20 K. i
It is well known that in conventional BCS singlet super-
conductors, the Knight shift, which is proportional to the Pauli in one dimension with open boundary conditions through the
paramagnetic susceptibility, changes very little below T c for transformation
heavy elements such as Sn and Hg (Androes and Knight,
1959). It is understood that this is caused by the destruction of  X nθ 
spin conservation due to the spin-orbit coupling. A clear U ¼ exp −i Szn
n
2
explanation was presented by Anderson (1959) using the
notion of time-reversed pairing states. We first consider the
imaginary part of the spin response function χ 00 ðq; ωÞ. If with cos θ ¼ Jz =J ⊥ and D ¼ J ⊥ sin θ, where U † H XXZ U ¼
the total spin is conserved, then the dynamics are diffusive and HJþDM , with HJ denoting the isotropic Heisenberg model
χ 00 ðq; ωÞ will have a central peak in ω space with a width of with interaction J.
Dq2 , which goes to zero as q → 0. Superconductivity gaps out Trial mean-field wave functions with the general pairing
all low-frequency excitations, thus removing this central peak.
By the Kramers-Kronig relation, the real part χ 0 ðq ¼ 0; ΔðkÞ ¼ i½d0 ðkÞσ 0 þ dðkÞ · σσ y
ω ¼ 0Þ vanishes in the superconducting ground state. In
the presence of spin-orbit coupling, the total spin is not are being considered for the construction of the corresponding
conserved but rather decays with a lifetime τs . In this case, Gutzwiller-projected wave functions. The trial ground-state
χ 00 ðq ¼ 0; ωÞ has a central peak with a width of 1=τs . The wave functions have the best energy when the d vector has the
superconducting gap (formed by a pair of time-reversal states) form d0 ¼ 0 and dðkÞ ¼ dz ẑ ¼ iΔ sin k for J z > J ⊥ (Ising
Δ cuts a hole in χ 00 ðωÞ for ω < Δ but leaves the ω ≫ Δ region regime), whereas the preferred form is d0 ¼ 0 and dðkÞ ¼
intact, consistent with the physical expectation that the high- dy ŷ ¼ Δ sin k for J z < J ⊥ (planar regime). The overlap
frequency region should be unaffected by pairing. By the between the trial ground-state wave function and the exact
Kramers-Kronig relation, χ 0 will be reduced, but if the spin- ground-state wave function obtained through exact diagonal-
orbit coupling is sufficiently strong that ization is better than 95% in all cases that have been
considered. Notably, the pairing state with dðkÞ ¼ dy ŷ ¼
1
≫ Δ; ð81Þ Δ sin k does not conserve Stotz and is not considered in the
τs classification scheme used by Dodds, Bhattacharjee, and Kim
(2013).
then the reduction will be small, i.e.,
χs B. RVB approach to S > 1=2 systems
¼ 1 − OðΔτs Þ:
χN
Historically, the search for spin liquid states has been
Equation (81) is the strong spin-orbit coupling condition that focused on spin-1=2 systems because such systems have
is required to have very little change in the spin susceptibility the strongest quantum mechanical fluctuation effects (see
below T c. We emphasize that the criterion for discriminating Sec. II) when the unfrustrated Heisenberg model is consid-
strong from weak spin-orbit coupling that is given by Eq. (81) ered. The situation is different when we consider spin systems
is completely different from the usual criterion, which with frustrated interactions (Chandra and Doucot, 1988). In
compares the spin-orbit energy λ with the splitting of the this case, it is not obvious whether a spin liquid state is more
t2g levels E3 (Chen and Balents, 2008). Another way to likely to exist in systems of lower spin. In fact, it was recently
explain the large Wilson ratio observed in Na4 Ir3 O8 was found that gapless spin liquid states may exist in a two-
provided by Chen and Kim (2013), in which strong spin-orbit dimensional spin-1 compound Ba3 NiSb2 O9 under high pres-
coupling is still essential. sure (Cheng et al., 2011). In this section, we examine how we
From a theoretical perspective, the PSG classification can construct spin liquid states for S > 1=2 systems by
scheme has been applied to classify the spin liquid states generalizing the RVB approach developed for S ¼ 1=2
on a kagome lattice with the Dzyaloshinskii-Moriya (DM) systems. Note that there are multiple possible methods of
interaction (Dodds, Bhattacharjee, and Kim, 2013). More generalization. For example, Greiter and Thomale (2009)
recently, to test the validity of the RVB approach in con- constructed a chiral spin liquid state using a fractional
structing wave functions for spin systems with strong spin- quantum Hall wave function, whereas Xu et al. (2012)
orbit coupling, Sze, Zhou, and Ng (2016) applied the constructed a spin liquid state for an S ¼ 1 system by
Gutzwiller-projected wave function of fermion pairing states representing a spin of 1 as the sum of two S ¼ 1=2 spins.
to study the S ¼ 1=2 anisotropic Heisenberg (XXZ) chain Liu, Zhou, and Ng (2010a, 2010b) developed an alternative
X X approach in which a spin S is represented by 2S þ 1 fermions.
H ¼ Jz Szi Sziþ1 þ J ⊥ ðSxi Sxiþ1 þ Syi Syiþ1 Þ; ð82Þ In the following section, we consider this last approach, and
i i we demonstrate the existence of fundamental differences

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-24


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

between half-odd-integer spin and integer spin systems in this We now examine the internal symmetry group associated
approach. with the redundancy in the fermion representation. The
We begin with the fermion representation of general spins. internal symmetry group is different for integer and half-
To generalize the fermion representation of S ¼ 1=2 spins to odd-integer spins; it is Uð1Þ⊗Z ¯ 2 ¼ feiσz θ ; σ x eiσz θ ¼
−iσ z θ
an arbitrary spin S, Liu, Zhou, and Ng (2010a, 2010b) e σ x ; θ ∈ Rg for the former and SUð2Þ for the latter.
introduced 2S þ 1 species of fermionic operators cm that The reason for this difference can be qualitatively understood
satisfy anticommutation relations, as follows: Note that C and C̄ are not independent. The
operators in the internal symmetry group “mix” the two
fcm ; c†n g ¼ δmn ; ð83Þ fermion operators in the same row of C and C̄, i.e., cS and
c†−S . For integer spins, c0 and ð−1ÞS c†0 will be “mixed.” For
where m; n ¼ S; S − 1; …; −S. The spin operator can be
fc0 ; c†0 g ¼ 1 to remain invariant, there are only two possible
expressed in terms of these operators as follows:
methods of “mixing”: one is a Uð1Þ transformation, and the
Ŝ ¼ C† IC; other is interchanging the two operators. These operations
form the Uð1Þ⊗Z ¯ 2 group. For half-odd-integer spins, the pair
where C ¼ ðcS ; cS−1 ; …; c−S ÞT and I a (a ¼ x, y, z) is (c0 ; ð−1ÞS c†0 ) does not exist, and the symmetry group is the
a ð2S þ 1Þ × ð2S þ 1Þ matrix whose matrix elements are maximum SUð2Þ group. Thus, the difference between integer
given by and half-odd-integer spins is a fundamental property of the
fermion representation.
I amn ¼ hS; mjSa jS; ni: Now let us see how the constraint expressed in Eq. (86)
transforms under the symmetry groups. For S ¼ 1=2, the
It is straightforward to show that the resulting spin operator Ŝ
constraint given in Eq. (86) is invariant under the trans-
satisfies the SUð2Þ angular momentum algebra. Under a
formation ψ → ψW because the right-hand side vanishes (as a
rotational operation, C is a spin-S “spinor” transforming as
result of the particle-hole symmetry of the Hilbert space). For
Cm → DSmn Cn and Ŝ is a vector transforming as Sa → Rab Sb ;
integer spins, if W ¼ eiσz θ , then Wσ z W † ¼ σ z , and Eq. (86) is
here DS is the (2S þ 1)-dimensional irreducible representation
invariant. If W ¼ σ x eiσz θ , then Wσ z W † ¼ −σ z , meaning that
of the SUð2Þ group generated by I, and R is the adjoint
the “particle” picture [þ sign in Eq. (86)] and the “hole”
representation.
picture [− sign in Eq. (86)] are transformed into each other.
As in the S ¼ 1=2 case, a constraint that there must be only
For a half-odd-integer spin with S ≥ 3=2, W ∈ SUð2Þ is a
one fermion per site is needed to project the fermionic system
rotation, and we may extend the constraint into a vector form
into the proper Hilbert space representing spins, i.e.,
in a manner similar to the S ¼ 1=2 case, such that Eq. (86)
becomes
ðN̂ i − N f Þjphyi ¼ 0; ð84Þ

where i is the site index and N f ¼ 1 (the particle picture, one Trðψ σ~ ψ † Þ ¼ (0; 0; ð2S − 1Þ)T : ð87Þ
fermion per site). Alternatively, it is straightforward to show
that the constraint N f ¼ 2S (the hole picture, one hole per site) Under the group transformation ψ → ψW,
equivalently represents a spin. The N f ¼ 1 representation can
be mapped to the N f ¼ 2S representation via a particle-hole Trðψ σ~ ψ † Þ → ðR−1 Þ(0; 0; ð2S − 1Þ)T ; ð88Þ
transformation. For S ¼ 1=2, the particle picture and the hole
picture are identical, reflecting an intrinsic particle- where Wσ a W † ¼ Rab σ b , a; b ¼ x, y, z, i.e., R is a 3 × 3
hole symmetry of the underlying Hilbert space, which is matrix representing a 3D rotation. The transformed constraint
absent for S ≥ 1. represents a new Hilbert subspace, which is still a (2N þ 1)-
Following Affleck, Zou, Hsu, and Anderson (1988), dimensional irreducible representation of the spin-SUð2Þ
Liu, Zhou, and Ng (2010a) introduced another spinor algebra. Any measurable physical quantity, such as the spin
C̄ ¼ (c†−S ; −c†−Sþ1 ; c†−Sþ2 ; …; ð−1Þ2S c†S )T , whose components S, remains unchanged in this new Hilbert space. Therefore, for
can be written as C̄m ¼ ð−1ÞS−m c†−m , where the index m half-odd-integer spins (S ≥ 3=2), there exist infinitely many
runs from S to −S as for C. Upon combining C and C̄ into a ways of imposing the constraint that gives rise to a Hilbert
ð2S þ 1Þ × 2 matrix ψ ¼ ðC; C̄Þ, it is straightforward to see subspace representing a spin. However, for integer spins, there
that the spin operators can be reexpressed as exist only two possible constraint representations.
The fermion representation can be used to construct mean-
field Hamiltonians for spin models with arbitrary spins after
Ŝ ¼ 12 Trðψ † IψÞ ð85Þ
the spin-spin interaction is written down in terms of fermion
operators. For the spin-1=2 case, the Heisenberg interaction
and that the constraint can be expressed as
can be written as (see Sec. III)
Trðψσ z ψ † Þ ¼ 2S þ 1 − 2N f ¼ ð2S − 1Þ; ð86Þ
Ŝi · Ŝj ¼ −18 Tr∶ðψ †i ψ j ψ †j ψ i Þ∶ ¼ −14∶ðχ †ij χ ij þ Δ†ij Δij Þ∶; ð89Þ
where the þ sign implies N f ¼ 1 and the − sign
implies N f ¼ 2S. where

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-25


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

χ ij ¼ C†i Cj ; Δij ¼ C̄†i Cj : ð90Þ rearranging the spins such that the system is a product of spin-
singlet ground states for N − 1 of the sites plus a single spin-1
The definitions of χ ij and Δij in this form can be extended to spinon. This excitation is a nonperturbative, topological
arbitrary spins. The only difference is that for an integer spin, excitation that cannot be achieved by simply Gutzwiller
projecting a BCS excited state in the RVB construction. It
χ ji ¼ χ †ij and Δji ¼ −Δij , whereas for a half-odd-integer spin,
was demonstrated by Liu, Zhou, and Ng (2014) that the
χ ji ¼ χ †ij and Δji ¼ Δij . The parity of the pairing term Δij construction of these two kinds of excitations gives rise to the
differs for integer and half-odd-integer spins (Liu, Zhou, and so-called one-magnon and two-magnon excitation spectra
Ng, 2010a). For S ¼ 1, it can be shown, after some straight- in the Haldane phase of the S ¼ 1 bilinear-biquadratic
forward algebra, that the Hamiltonian can be written as (Liu, Heisenberg model. Similar construction approaches are not
Zhou, and Ng, 2010a) possible for S ¼ 1=2 systems.

Ŝi · Ŝj ¼ −12 Tr∶ðψ †i ψ j ψ †j ψ i Þ∶ ¼ −∶ðχ †ij χ ij þ Δ†ij Δij Þ∶: ð91Þ C. Matrix product state and projected entangled pair state

However, for S > 1, we cannot write the spin-spin inter- In this section, we discuss two approaches to spin liquid
states that have completely different starting points from those
action Ŝi · Ŝj in terms of χ ij and Δij alone. In the case of
of the RVB, or Gutzwiller-projected mean-field theory,
S ¼ 3=2, triplet hopping and pairing terms must be introduced approach we discussed in Sec. III. We begin with MPSs
to represent the Heisenberg interaction. Generally speaking, and PEPSs, which represent another popular class of varia-
quintet and higher multipolar hopping and pairing operators tional wave functions that are currently being applied to spin
are needed to represent the Heisenberg Hamiltonian when S systems. Translationally invariant MPSs in spin chains were
becomes larger (Liu, Zhou, and Ng, 2010a). In the following, first constructed and studied by Fannes, Nachtergaele, and
we restrict ourselves to S ¼ 1 systems. Werner (1992) as an extension of the AKLT state (Affleck
In this case, the mean-field Hamiltonians are BCS-type et al., 1987); in this context, they called them finitely
Hamiltonians, as in the case of S ¼ 1=2 spins. The physical correlated states. The term MPS was coined by Klümper,
spin wave function can be obtained by applying Gutzwiller Schadschneider, and Zittartz (1993), who extended the AKLT
projection to the mean-field ground state. There are two major state in a different way. Later, Östlund and Rommer (1995)
differences between S ¼ 1 and S ¼ 1=2 spin systems: realized that the state resulting from DMRG (White, 1992) can
(1) Because of the different internal symmetry group be written as an MPS. This approach is very successful for
[Uð1Þ⊗Z ¯ 2 ], S ¼ 1 spin liquid states are of either the Uð1Þ one-dimensional systems and can be generalized to systems of
or Z2 type. There are no SUð2Þ spin liquid states for integer two (or more) dimensions.
spin systems in the fermionic construction. Therefore, we First, let us consider the quantum wave function of a one-
expect that in general spin liquid states for integer spin dimensional spin system that is translationally invariant with a
systems, if they exist, are more stable against gauge fluctua- local Hamiltonian H. The wave function can be generally
tions. (2) The difference in parity of the pairing terms leads to expressed as
interesting possibilities for obtaining topological spin liquid
X
states in S ¼ 1 systems that are not easy to realize in S ¼ 1=2 jΨi ¼ ϕðs1 ; s2 ; …; sN Þjs1 ; s2 ; …; sN i; ð92Þ
systems (Liu, Zhou, and Ng, 2010b; Bieri et al., 2012). This s1 ;s2 ;…;SN
difference leads to the existence of a Haldane phase in the
bilinear-biquadratic Heisenberg spin chain in the fermionic where js1 ; s2 ; …; sN i represents a spin configuration with
description (Liu et al., 2012). spins si on sites i ¼ 1; 2; …; N and ϕðs1 ; s2 ; …; sN Þ is the
Finally, we note the existence of a fundamental difference in amplitude of the spin configuration in the quantum state jΨi.
the excitation spectrum of an S ¼ 1 spin system compared Because of the spin-spin interaction, spin configurations at far
with that of an S ¼ 1=2 system, under the assumption that the away sites are generally correlated, and we cannot write
ground states are spin singlets. For an integer spin system, we ϕðs1 ; s2 ; …; sN Þ ¼ ϕ0 ðs1 Þϕ0 ðs2 Þ    ϕ0 ðsN Þ in general. The
can form spin-singlet states in a lattice with either an even or MPS approach is a powerful method of constructing wave
an odd number of lattice sites N, as long as N > 1, whereas for functions with nonlocal quantum correlations. The trick is to
a half-odd-integer spin system, spin-singlet states can be extend the direct product wave function ϕðs1 ; s2 ; …; sN Þ ¼
formed only in a lattice with an even number of sites. In ϕ0 ðs1 Þϕ0 ðs2 Þ    ϕ0 ðsN Þ to matrix products.
the RVB approach, angular momentum L ¼ 1 excitations of More explicitly, we associate a matrix As with each spin
the system are formed by Gutzwiller projecting the excited state s; then the wave function amplitude ϕðs1 ; s2 ; …; sN Þ can
states in BCS theory, i.e., by breaking a pair of spin singlets in be written as
the BCS ground state. The resulting excited state consists of
two excited spinons, which are S ¼ 1=2 objects for spin-1=2 ϕðs1 ; s2 ; …; sN Þ ¼ TrfAs1 ½1As2 ½2    AsN ½Ng; ð93Þ
systems but are S ¼ 1 objects for spin-1 systems. In an S ¼ 1
spin liquid, these two S ¼ 1 spinons together form an L ¼ 1 where the trace is used to impose the periodic boundary
excitation. condition. As an example, we consider an S ¼ 1=2 two-spin
There is, however, another method of forming an L ¼ 1 system and choose A↑ ¼ σ z and A↓ ¼ σ x , where the σ’s are
excitation in a spin-1 spin liquid. Beginning from a lattice Pauli matrices. It is easy to see that in this case ϕð↑; ↑Þ ¼
system with N sites, we may form an L ¼ 1 excitation by ϕð↓; ↓Þ ≠ 0 and ϕð↑; ↓Þ ¼ ϕð↓; ↑Þ ¼ 0. A different choice of

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-26


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

A↑ ¼ σ þ and A↓ ¼ σ − yields ϕð↑; ↓Þ ¼ ϕð↓; ↑Þ ≠ 0 and spin S ¼ 0 or singlet state and preserving only the spin S ¼ 1
ϕð↑; ↑Þ ¼ ϕð↓; ↓Þ ¼ 0. The correlation between the different or triplet states.
spin states on the two sites is determined by the matrix As that For every adjacent pair of S ¼ 1 spins, two of the four
is chosen to link the sites. Extending the construction to more constituent S ¼ 1=2 spins are projected into a state with a total
than two sites, one sees that the choice of the matrices Aσ spin of zero by the valence bond. Therefore, the pair of S ¼ 1
determines the quantum entanglement structure of the wave spins is forbidden from existing in a combined spin S ¼ 2
function. state. This condition can be realized by considering a
When the MPSs are treated as variational wave functions, Hamiltonian that is a sum of projectors Pi;iþ1 that projects
one may determine the number of variational parameters in the the pairs of S ¼ 1 spins from the 1 ⊗ 1 ¼ 2 ⨁ 1 ⨁ 0 space
wave functions by means of a simple counting argument. The into the spin S ¼ 2 subspace,
number of parameters P appearing in an MPS wave function X
in the form of Eq. (93) depends on the size of the matrix A and HAKLT ¼ Pi;iþ1 : ð94aÞ
the number of available states S per site. In general, P ∼ S × i
M2 for an M × M matrix as long as P < SN , where N is the
number of sites in the system. Thus, MPS wave functions are Because the projection operators Pi;iþ1 are positive semi-
generally variational wave functions with a large number of definite, the ground state satisfies HAKLT jΨG i ¼ 0 and is
built-in variational parameters. As the dimension M → ∞, simply the AKLT state. The projection operator Pi;iþ1 can be
MPSs can represent any quantum state of the many-body written in terms of spin-1 operators as follows (Affleck et al.,
Hilbert space with arbitrary accuracy. In practice, the low- 1987):
energy states of gapped local Hamiltonians in one dimension
can be efficiently represented by MPSs with a finite value of Pi;iþ1 ¼ 13 þ 12ðSi · Siþ1 Þ þ 16ðSi · Siþ1 Þ2 : ð94bÞ
M (Verstraete and Cirac, 2006; Hastings, 2007). The DMRG
method (White, 1992) and its generalizations (Schollwöck, The AKLT state is important because it is an explicit spin
2005) can be viewed as systematic approaches for construct- wave function that realizes the Haldane phase for integer spins
ing MPS variational wave functions as the size of the system (see Sec. II). In particular, it is easy to see from Fig. 9 that an
gradually increases. unpaired S ¼ 1=2 spin will be left at each end of the spin
The MPS construction can be extended in several ways. chain, which is a realization of the end state discussed in
First, it can be extended to higher dimensions by replacing the Sec. II for S ¼ 1 Heisenberg spin chains. In the following, we
matrices A (¼ rank-2 tensors) with higher-rank tensors T. show how the AKLT state can be written as an MPS state.
These wave functions are presently known as PEPSs The AKLT state can be constructed in two steps. First, we
(Verstraete and Cirac, 2004a, 2004b). Second, the local split each site i in the spin-1 chain into two sites iL and iR ,
correlation or entanglement between a pair of sites in a thereby forming a spin-1=2 chain with 2N sites, as in Fig. 9
PEPS can be generalized to a cluster (or simplex), resulting (where N is the number of sites in the parent spin-1 chain) and
in states called projected entangled simplex states (PESSs) construct a dimerized chain in which the spins at sites iR and
(Xie et al., 2014). A representative example of a PESS is the i þ 1L (i ¼ 1; 2; …; N) are joined by a valence bond [see
simplex solid state proposed by Arovas (2008). Eq. (22)]. The singlet bond between sites iR and i þ 1L can be
written as
1. Valence-bond solids and MPSs in one dimension X
The physics of an MPS or PEPS wave function is encoded ði; i þ 1Þ ¼ Rσ iR ;σ iþ1L jσ iR ijσ iþ1L i; ð95Þ
σ iR ;σ iþ1L
in the tensors linking neighboring spin states. In general, these
link tensors can be optimally constructed using the DMRG
approach or tensor-based renormalization methods (Cirac and where σ ¼ ↑; ↓ and the Rσσ 0 are the components of a 2 × 2
Verstraete, 2009). In this section, we discuss a simple example matrix:
of tensors that represents a particular class of spin states called !
0 p1ffiffi
VBS states. To begin, we introduce a well-known example of 2
R¼ : ð96Þ
a valence-bond solid state—the Affleck-Kennedy-Lieb-Tasaki − p1ffiffi2 0
state (Affleck et al., 1987).
The AKLT state is an example of a VBS state in which only In this representation, the wave function of the dimerized spin-
one spin-singlet configuration is allowed in the wave function 1=2 chain can be written as
given in Eq. (23). It is a one-dimensional VBS state con-
structed for a S ¼ 1 spin chain, represented pictorially in
Fig. 9, where each gray bond represents a spin singlet formed iL iR i + 1L i + 1R
by two S ¼ 1=2 spins, i.e., Eq. (22). Each lattice site
is connected to two other sites by two valence bonds and
is occupied by two S ¼ 1=2 spins. The AKLT wave function is
formed by projecting the spin-1=2 ⊗ 1=2 ¼ 1 ⨁ 0 quartet
states into the spin-S ¼ 1 triplet states. This is represented singlet spin S = 1 projector
graphically in Fig. 9 by the circles, which represent projection
operators tying together two S ¼ 1=2 spins, projecting out the FIG. 9. A valence-bond solid construction of the AKLT state.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-27


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

X
jΨi ¼ Rσ1R σ2L    RσN−1R σNL jσ 1R ; …; σ N L i: ð97Þ (Affleck, Kennedy et al., 1988). These states can be written as
σ 1R ;…;σ N L PEPSs in their respective lattices.
For instance, on a square lattice with a coordination number
Note that this state is a direct product state of S ¼ 1=2 RVB of 4, a generic PEPS wave function can be written in terms of
singlet pairs with the two end spins (σ 1L and σ N R ) unspecified. rank 4 tensors as follows:
Next we project the two S ¼ 1=2 spins at sites iL and iR to X
the spin-1 states j1; mi (m ¼ 0; 1) with jΨi ¼ ϕð½sij Þj½sij i; ð103aÞ
½sij 
1
j1; 1i ¼ j↑↑i; j1; 0i ¼ pffiffiffi ðj↑↓i þ j↓↑iÞ; where i; j ¼ 1; …; N for an N × N system, ½sij  ¼
2
ðs11 ; …; s1N ; s21 ; …; s2N ; …; sN1 ; …; sNN Þ denotes a spin con-
j1; −1i ¼ j↓↓i: ð98Þ figuration, and

This projection can be expressed in terms of three matrices ϕð½sij Þ ¼ Tr½T s11    T s1N T s21    T sNN ; ð103bÞ
M0;1 , where
s
X where the T s ’s are rank 4 tensors with components T lrud ij
,
j1; mi ¼ 0
σσ 0 jσijσ i
Mm ð99Þ where sij is the physical spin index; l, r, u, and d represent
σ;σ 0 links connected to the tensors at the left, right, up, and down
neighboring sites ði − 1; jÞ, ði þ 1; jÞ, ði; j − 1Þ, and
with ði; j þ 1Þ, respectively; and “Tr” means tensor contraction.
  The above mathematical expression of tensor contraction is
1 0 usually represented by diagrams such as those shown in
M1 ¼ ; ð100aÞ
0 0 Fig. 10 for a square lattice, where connected lines represent
the contraction of tensors with the same index and open lines
 
0 0 represent the physical spin states sij ¼ −S; …; S.
−1
M ¼ ; ð100bÞ As an example, a spin S ¼ 2 AKLT state on a square lattice
0 1
can be written in PEPS form as shown in Fig. 11. The tensors
and T s can be obtained using the VBS construction with the
tensors R and Ms , as in one dimension. The tensor R is still
0 p1ffiffi  defined by Eq. (96). The tensors Ms , s ¼ 0; 1; 2, project a
2
M ¼ 0
: ð100cÞ state consisting of four S ¼ 1=2 spins in the auxiliary Hilbert
p1ffiffi 0 space 12 ⊗ 12 ⊗ 12 ⊗ 12 ¼ 2 ⨁ 1 ⨁ 0 into the physical S ¼ 2
2
spin space, whose components are given by
Thus, the AKLT state can be written as
X
jΨAKLT i ¼ ϕAKLT ðs1 ; …; sN Þjs1 ; s2 ; …; sN i; ð101Þ
s1 ;s2 ;…;sN

where si ¼ 0; 1 and

ϕAKLT ðs1 ; …; sN Þ
X sij u
¼ ½Msσ11L σ1R Rσ 1R σ2L × M sσ22L σ 2R    Rσ N−1R σ NL 
σ 1R ;…;σ N L l TT r
d
¼ ½As1 As2    AsN σ 1 σ : ð102aÞ
L NR

Here As ¼ Ms R, and σ 1L ; σ N R ¼ ↑; ↓ correspond to four


degenerate ground states on an open chain. Imposing the
periodic boundary condition gives rise to a nondegenerate
ground state with

ϕAKLT ðs1 ; …; sN Þ ¼ Tr½As1 As2    AsN : ð102bÞ


FIG. 10. Graphical representation of a PEPS in terms of
contracted tensors (tensor network). Each box denotes a tensor
2. PEPSs in higher dimensions and beyond sij
T with components T lrud at site ij, where l, u, r, and d are tensor
The AKLT construction can be extended to construct other indices related to left, right, up, and down bonds, respectively,
types of VBS states and states in higher dimensions. linking to their neighbors; the open lines represent the physical
Straightforward examples include S ¼ 2 VBS states on a spin states sij , and the connected lines represent the contraction of
square lattice and S ¼ 3=2 VBS states on a honeycomb lattice the tensors.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-28


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

D. Kitaev honeycomb model and related issues

It was previously believed that spin rotational symmetry is


essential for a QSL state that supports fractional spinon
excitations. If the spin rotational symmetry is broken, the
system tends to approach an ordered state. Kitaev (2006)
provided a counterexample to this belief through an unusual,
exactly solvable model in two dimensions with strong spin-
orbit coupling, which destroys the spin rotational symmetry,
but in which deconfined spinons nevertheless exist on top of
the QSL ground states. This famous model is now called the
Kitaev honeycomb model. In this section, we briefly review
the Kitaev honeycomb model to see how exotic ground states
and low-energy excitations emerge from this model with
broken rotational symmetry. The possibility of the realization
of Kitaev-like models in realistic materials is also discussed.
u
Kitaev considered a spin-1=2 model on a honeycomb lattice
l with spin-orbit coupling (Kitaev, 2006). He divided all nearest
T r R M neighbor bonds in the honeycomb lattice into three types,
d
called x links, y links, and z links as shown in Fig. 12. The
FIG. 11. The VBS construction of an S ¼ 2 AKLT state on a
Hamiltonian is given as follows:
square lattice and the corresponding tensors. X X X
H ¼ −J x K ij − Jy K ij − J z K ij ; ð106Þ
x link y link z link
M sσl σr σu σd ¼ hsjσ l σ r σ u σ d i; ð104Þ
where K ij is defined as
where σ l ; σ r ; σ u ; σ d ¼ ↑; ↓. The tensor T is given by 8
>
> σ x σ x ; if ði; jÞ is a x link;
X < i j
y y
T sσ l σ r σu σ d ¼ Msσ l0 σr σu0 σd Rσ l σ l0 Rσ u σ u0 : ð105Þ K ij ¼ σ i σ j ; if ði; jÞ is a y link; ð107Þ
>
>
σ l0 ;σ u0 : σ z σ z ; if ði; jÞ is a z link:
i j

The tensor product state constructed from the above T s ’s give Note the strong anisotropy in the spin-spin couplings K ij .
rise to the S ¼ 2 AKLT state on a square lattice. We first consider the following loop operators W p defined
The VBS construction can be further extended by “frac- for a hexagonal loop:
tionalizing” the spins in more exotic ways (for example, using
the Majorana fermion representation of spins). In this way, we
W p ≡ σ x1 σ y2 σ z3 σ x4 σ y5 σ z6 ¼ K 12 K 23 K 34 K 45 K 56 K 61 ; ð108Þ
can write the toric code model (Kitaev, 2003) in the PEPS
form as well as the Kitaev honeycomb model (Kitaev, 2006) where p is used to label the lattice plaquettes (hexagons), as
(with a residual fermionic degree of freedom at each site; see shown in Fig. 13. It is easy to verify that ½W p ; K ij  ¼ 0;
Sec. IV.D). The relation between RVB states and PEPSs has
therefore, ½H; W p  ¼ 0. Hence, the W p ’s serve as good
also been exploited to show that some RVB states can be
quantum numbers for the Hamiltonian given in Eq. (106),
written as PEPSs (Verstraete et al., 2006; Schuch et al., 2012;
Poilblanc and Schuch, 2013; Wang et al., 2013). However, the
general relation between RVB states and PEPSs remains x y x y x y
unclear.
The PEPS construction provides a way to describe entan- z z z z
glement among local spins based on the construction of local
pairs, and its application to geometrically frustrated lattices is x y x y x y x y
limited. To overcome this limitation, researchers have
extended the pair construction procedure to consider entan- z z z z z
glement between more than two sites, say, a cluster or a
y x y x y x y x
simplex, to construct projected states. These projected
entangled simplex states form the basis for more elaborate
numerical approaches (Xie et al., 2014). Combined with z z z z
numerical techniques (tensor-based renormalization), these y y y
tensor-network methods now provide an alternative means of
constructing variational wave functions. Interested readers can
refer to Verstraete, Murg, and Cirac (2008), Cirac and FIG. 12. Kitaev honeycomb model. x, y, and z denote three
Verstraete (2009), and Orus (2014) for details. types of links in the honeycomb lattice.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-29


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

z
The Majorana representation without constraints is redundant
y 3
x and enlarges the physical spin Hilbert space. Note that D2 ¼ 1
2 4 and that D has two eigenvalues, D ¼ 1, thereby splitting
p Wp the local Hilbert space into two sectors. The physical spin
1 5 Hilbert space corresponds to the sector with all Dj ¼ 1.
x y
6 Therefore, the physical spin wave function jΨspin i can be
obtained from the Majorana fermion wave function jΨMajorana i
z
through the projection
y y
FIG. 13. Loop operator W p ¼ σ x1 σ 2 σ z3 σ x4 σ 5 σ z6 on a lattice
plaquette (hexagon).
Y 1 þ Dj
jΨspin i ¼ jΨMajorana i; ð113Þ
j
2
and the total Hilbert space for spins can be divided into a direct
product of sectors that are eigenspaces of fW p g. However, the which retains the Dj ≡ 1 sector and removes all other sectors
eigenvalue problem cannot be completely solved by determin- in the enlarged Hilbert space. Note that
ing the eigenspaces of fW p g. Each W p has only two
1 þ Dj
eigenvalues, wp ¼ 1. Each plaquette contains six sites, ¼ nj↑ þ nj↓ − 2nj↑ nj↓
2
and each site is shared by three plaquettes. Therefore, the
number of plaquettes is given by m ¼ N=2, where N is the and that Eq. (113) is nothing but the Gutzwiller projection. In
number of sites. It follows that the dimension of each addition, note that Dj serves as a Z2 gauge transformation in
eigenspace of fW p g is 2N =2m ¼ 2N=2 , i.e., splitting the the enlarged Hilbert space (Dj bαj Dj ¼ −bαj , Dj cj Dj ¼ −cj )
Hilbert space into eigenspaces of fW p g cannot solve the and commutes with the spin operators (½Dj ; σ αj  ¼ 0,
eigenvalue problem completely. α ¼ x, y, z) and thus with the Hamiltonian. As a result, the
Kitaev realized that to solve the model Hamiltonian given in Gutzwiller projection is “trivial” in the sense that
Eq. (106), spins can be written in terms of four Majorana
fermions, because a Majorana fermion can be viewed as the Y 1 þ Dj
real or imaginary part of a complex fermion. To illustrate this jΨMajorana i
j
2
approach, we rewrite the complex fermions f↑ and f↓ in
Eq. (27) in terms of four Majorana fermions c1 , c2 , c3 , and c4 :
is an eigenstate of H in the projected Hilbert space as long as
jΨMajorana i is an eigenstate of H in the “unprojected” Hilbert
f↑ ¼ 12ðc1 þ ic2 Þ; f†↑ ¼ 12ðc1 − ic2 Þ;
space and
f↓ ¼ 12ðc3 þ ic4 Þ; f†↓ ¼ 12ðc3 − ic4 Þ; ð109aÞ
Y 1 þ Dj
jΨMajorana i ≠ 0.
where the operators cα (α ¼ 1, 2, 3, 4) are Hermitian and j
2
satisfy
In the Majorana fermion representation, K ij in Eq. (107)
cα cβ þ cβ cα ¼ 2δαβ : ð109bÞ becomes

Thus, the three spin components read σ x ¼ ði=2Þ× K ij ¼ −iðibαi bαj Þci cj ; ð114Þ
ðc1 c4 − c2 c3 Þ, σ y ¼ ði=2Þðc3 c1 − c2 c4 Þ, and σ z ¼ ði=2Þ×
ðc1 c2 − c3 c4 Þ. The single-occupancy condition f†↑ f↑ þ where α ¼ x, y, z depends on the type of link ðijÞ. The
operator ibαi bαj is Hermitian, and we denote it by ûij ¼ ibαi bαj .
f†↓ f↓ ¼ 1 (and f †↑ f †↓ ¼ f↑ f↓ ¼ 0) becomes
Thus, we write

c1 c2 þ c3 c4 ¼ c1 c3 þ c2 c4 ¼ c1 c4 þ c3 c2 ¼ 0; ð110Þ iX
H¼ Â c c ; ð115aÞ
4 hj;ki jk j k
which can be simplified to the single equation c1 c2 c3 c4 ¼ 1.
Using these constraints, the spin operators can be written as with
σ x ¼ ic1 c4 , σ y ¼ −ic2 c4 , and σ z ¼ −ic3 c4 . Rewriting
bx ¼ c1 , by ¼ −c2 , bz ¼ −c3 , and c ¼ c4 , we arrive at the Âjk ¼ 2JαðjkÞ ûjk ; ûjk ¼ ibαj bαk ; ð115bÞ
Kitaev representation
where hj; ki denotes nearest neighbor links on the honeycomb
σ x ¼ ibx c; σ y ¼ iby c; σ z ¼ ibz c; ð111Þ lattice and, by definition, ûjk ¼ −ûkj and Âjk ¼ −Âkj . The
Hamiltonian structure in this Majorana fermion representation
with the constraint is shown schematically in Fig. 14. Note that ½H; ûjk  ¼ 0 and
½ûjk ; ûj0 k0  ¼ 0. The enlarged Hilbert space of Majorana
D ≡ bx by bz c ¼ 1: ð112Þ fermions can be decomposed into common eigenspaces of

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-30


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Majorana fermions
J x , J y , and Jz . For instance, if J z is replaced with −J z , we can
compensate for this sign change by changing the signs of the
variables ujk for all z links using the gauge operator Dj,
cj
leaving the values of Ajk and wp unchanged. Therefore, as far
z
b j as solving for the ground-state energy and the excitation
ujk spectrum is concerned, the signs of the exchange constants J
bk
z do not matter. However, such a sign change does affect other
spins measurable physical quantities.
ck
Second, it was proven by Lieb (1994) and numerically
investigated by Kitaev himself that the ground state of the
Majorana system is achieved when the system is in the vortex-
free configuration, namely, wp ¼ 1 for all plaquettes p. In this
FIG. 14. Graphic representation of the four-Majorana-fermion vortex-free configuration, one can solve for the (fermionic)
decomposition of the Hamiltonian expressed in Eq. (106). energy spectrum of the Hamiltonian by directly Fourier
transforming Eq. (118) to obtain
fûjk g indexed by the corresponding eigenvalues ujk ¼ 1.
ϵq ¼ jJx eiq·a þ J y eiq·b þ J z j; ð121Þ
Thus, the Hamiltonian in the invariant subspace indexed by
u ¼ fujk g becomes pffiffiffi pffiffiffi
where a ¼ ð1=2; 3=2) and b ¼ ð−1=2; 3=2) are two basis
iX vectors in the xy coordinates. The fermionic spectrum may or
Hu ¼ A c c ; A ¼ 2J αðjkÞ ujk ; ð116Þ may not be gapped, depending on whether a solution to the
4 hj;ki jk j k jk
equation ϵq ¼ 0 exists. ϵq ¼ 0 has a solution if and only if
jJx j, jJ y j, and jJ z j satisfy the triangle inequalities:
where we replaced Âjk and ûjk with their eigenvalues.
Note that ujk → −ujk upon the Z2 gauge transformation jJx j ≤ jJ y j þ jJz j; jJ y j ≤ jJ z j þ jJ x j; jJ z j ≤ jJ x j þ jJ y j:
ujk → Dj ujk Dj , and it is more convenient to classify the
eigenstates of H in terms of the gauge-invariant loop operator ð122Þ
Wðj0 ; …; jn Þ ¼ K jn jn−1    K j1 j0 , which can be written as
As a result, two phases exist in the system of Majorana
Y
n  fermions on the honeycomb lattice, with the phase diagram
Wðj0 ; …; jn Þ ¼ −iûjs js−1 cn c0 : ð117Þ shown in Fig. 15. The first phase, called the A phase, is gapped
s¼1 and contains three subphases (Ax , Ay , and Az ) in the phase
diagram. The second, called the B phase, is gapless. In the A
The closed-loop operator W p [see Eq. (108)] is gauge
phase, for example, in the Az subphase, the Hamiltonian
invariant under the Z2 transformation because cn ¼ c0 , and expressed in Eq. (106) can be mapped to the Kitaev toric code
the gauge-invariant quantities w ¼ fwp g can be used instead model in the limit jJ x j; jJ y j ≪ jJ z j, and the phase hosts
of u ¼ fujk g to parametrize the eigenstates, i.e., Abelian anyonic excitations. The B phase acquires an energy
iX
gap in the presence of a magnetic field. Interestingly, it hosts
Hw ¼ A cc: ð118Þ stable non-Abelian anyons when the energy gap is opened up
4 hj;ki jk j k by a magnetic field. The B phase is an attractive state in the
context of topological quantum computation. Interested read-
For a given set of Aij fixed by fwp g, the quadratic Hamiltonian ers can refer to the recent review article by Nayak et al. (2008)
as expressed in Eqs. (116) and (118) can be diagonalized into for details.
the following canonical form:

iX X  † 1
 Jz 1, Jx Jy 0
Hcanonical ¼ ϵm c0m c00m ¼ ϵm fm f m − ; ð119Þ
2 m m
2
Az
where ϵm ≥ 0, c0m and c00m are normal Majorana modes, and
f†m ¼ ð1=2Þðc0m − ic00m Þ and fm ¼ ð1=2Þðc0m þ ic00m Þ are the
corresponding complex fermion operators. The ground state B
of the Majorana system has an energy of Ax Ay
Jy 1,
1X
Jx 1,
E¼− ϵ : ð120Þ Jy Jz 0 Jz Jx 0
2 m m
FIG. 15. Phase diagram of the Kitaev honeycomb model. The
We now discuss the system of Majorana fermions on the triangle is the section of the positive octant (J x ; J y ; J z ≥ 0) that
honeycomb lattice. First, we note that the global ground-state lies in the plane J x þ J y þ J z ¼ 1. The A phase contains three
energy does not depend on the signs of the exchange constants gapped subphases. The B phase is gapless.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-31


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

In addition to the elegant Majorana decomposition method Khaliullin (2009) and Chaloupka, Jackeli, and Khaliullin
pioneered by Kitaev, other insightful approaches to the Kitaev (2010) that a generalization of the Kitaev honeycomb model
honeycomb model also exist. For instance, Feng, Zhang, and may indeed arise in layered honeycomb lattice materials in the
Xiang (2007) and Chen and Nussinov (2008) found that the presence of strong spin-orbit coupling. They showed that in
original Kitaev honeycomb model can be exactly solved with certain iridate magnetic insulators (A2 IrO3 , A ¼ Li, Na), the
the help of the Jordan-Wigner transformation. This approach effective low-energy Hamiltonian for the effective J eff ¼ 1=2
provides a topological characterization of the quantum phase iridium moments is given by a linear combination of the
transition from the A phase to the B phase. A nonlocal string AFM Heisenberg model (HH ) and the Kitaev honeycomb
order parameter can be defined in one of these two phases model (H K ),
(Feng, Zhang, and Xiang, 2007; Chen and Nussinov, 2008). In
the appropriate dual representations, these string order param- H ¼ ð1 − αÞH H þ 2αHK ; ð123Þ
eters become local order parameters after some singular
transformation, and a description of the phase transition in where α, expressed in terms of the microscopic parameters,
terms of Landau’s theory of continuous phase transitions determines the relative strength of the Heisenberg and Kitaev
becomes applicable (Feng, Zhang, and Xiang, 2007). The interactions. Interestingly, the Kitaev honeycomb model can
Jordan-Wigner transformation also enables a fermionization also be realized as the exact low-energy effective Hamiltonian
of the Kitaev honeycomb model, allowing it to be mapped to a of a spin-1=2 model with spin rotational and time-reversal
p-wave-type BCS pairing problem. The spin wave function symmetries (Wang, 2010a). The Heisenberg-Kitaev model
can be obtained from the fermion model, and the anyonic (123) exhibits a rich phase diagram. Readers who are
character of the vortex excitations in the gapped phase also has interested in these developments can refer to, for example,
an explicit fermionic construction (Chen and Nussinov, 2008). Chaloupka, Jackeli, and Khaliullin (2010), Jiang et al. (2011),
The Kitaev honeycomb model can also be understood Kimchi and You (2011), Reuther, Thomale, and Trebst (2011),
within the framework of fermionic RVB theory. Both confine- Price and Perkins (2012), Schaffer, Bhattacharjee, and Kim
ment-deconfinement transitions from spin liquids to AFM or (2012), Singh et al. (2012), Yu et al. (2013), Kimchi
stripy AFor FM phases and topological quantum phase and Vishwanath (2014), and Lee et al. (2014) for details.
transitions between gapped and gapless spin liquid phases A comprehensive review on this topic was also given by
can be described within the framework of Z2 gauge theory Nussinov and van den Brink (2013).
(Baskaran, Mandal, and Shankar, 2007; Mandal et al., 2011;
Mandal, Shankar, and Baskaran, 2012). V. QSL STATES IN REAL MATERIALS
Exact diagonalization has been applied to study the Kitaev
honeycomb model on small lattices (Chen, Wang, and Das Experimental studies of interacting spins in geometrically
Sarma, 2010). Perturbative expansion methods have been frustrated lattices aim at identifying nontrivial and exotic
developed to study the gapped phases of the Kitaev honey- ground states. Among these ground states, spin liquid states
comb model and its generalization (Dusuel et al., 2008; have been sought ever since the proposal of the RVB state
Schmidt, Dusuel, and Vidal, 2008; Vidal, Schmidt, and (Anderson, 1973). This issue has been intensively debated in
Dusuel, 2008). Several papers (Lee, Zhang, and Xiang, the context of the spin states behind the high-T c super-
2007; Yu, 2008; Yu and Wang, 2008; Kells, Slingerland, conductivity of cuprates. However, before this century, there
and Vala, 2009) have noted the existence of an analogy was no direct observation of spin liquid states. The situation
between the Z2 vortices in the Kitaev honeycomb model and changed in 2003, when an organic Mott insulator with a
the vortices in p þ ip superconductors. quasitriangular lattice was found to exhibit no magnetic
Enormous efforts have been devoted to searching for ordering even at tens of mK, 4 orders of magnitude lower
exactly solvable generalizations of the Kitaev honeycomb than the energy scale of the exchange interactions (Shimizu
model. It has been proposed that the exact solvability will not et al., 2003). The low-temperature state is most likely a form
be spoiled when the fermion gap is opened for the non- of the sought-after spin liquids. Since then, what can be called
Abelian phase (Lee, Zhang, and Xiang, 2007; Yu and Wang, spin liquids have been successively reported for quasitrian-
2008). Generalizations to other lattice models and even to gular, kagome, and hyperkagome lattices. In this section, we
three dimensions have also been developed (Yang, Zhou, and review the experimental studies mainly with respect to the
Sun, 2007; Yao and Kivelson, 2007; Baskaran, Santhosh, and magnetic and thermodynamic properties of the materials for
Shankar, 2009; Nussinov and Ortiz, 2009; Ryu, 2009; Wu, which sound experimental data have been accumulated in
Arovas, and Hung, 2009; Yao, Zhang, and Kivelson, 2009; discussing the presence of spin liquids.
Tikhonov and Feigel’man, 2010; Lai and Motrunich, 2011;
Yao and Lee, 2011). Nontrivial emergent particles, such as A. Anisotropic triangular-lattice systems: κ-ðETÞ2 Cu2 ðCNÞ3
chiral fermions (Yao and Kivelson, 2007), have been con- and EtMe3 Sb½PdðdmitÞ2 2
structed in these exactly solvable lattice models. These
developments have significantly advanced our understanding Both are half-filled band systems with anisotropic triangular
of emergent phenomena based on solvable models in dimen- lattices, which are isosceles for κ-ðETÞ2 Cu2 ðCNÞ3 and three
sions greater than 1. different laterals for EtMe3 Sb½PdðdmitÞ2 2 (Kanoda, 2006;
The exotic properties of the Kitaev honeycomb model have Kanoda and Kato, 2011; Kato, 2014). At ambient pressure,
motivated researchers to search for realizations of this model they are Mott insulators; however, the spins are not ordered at
in realistic materials. It has been demonstrated by Jackeli and low temperatures on the order of tens of mK. A noticeable

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-32


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

(a) (C)

-t, t’ (meV)
(b)
t’

Cu[N(CN)2]Cl Cu(NCS)2 Cu2(CN)3

FIG. 17. (a) In-plane structure of the ET layer in κ-ðETÞ2 X.


It is modeled to (b) an anisotropic triangular lattice. (c) First
FIG. 16. Temperature-pressure phase diagram of the spin liquid principles calculations of transfer integrals in κ-ðETÞ2 X for X ¼
compound with a quasitriangular lattice κ-ðETÞ2 Cu2 ðCNÞ3 , Cu½NðCNÞ2 Cl, CuðNCSÞ2 , and Cu2 ðCNÞ3 ; squares (Nakamura
which undergoes a Mott transition at moderate pressure. From et al., 2009), circles (Kandpal et al., 2009), and triangles
Kurosaki et al., 2005. (Koretsune and Hotta, 2014).

feature of both systems is that they undergo Mott transitions at tight-binding calculation (Mori et al., 1984; Komatsu et al.,
moderate pressures 0.4 GPa for κ-ðETÞ2 Cu2 ðCNÞ3 (Komatsu 1996; Mori, Mori, and Tanaka, 1999) or the first principles
et al., 1996; Kurosaki et al., 2005; Furukawa et al., 2015a) and calculation (Kandpal et al., 2009; Nakamura et al., 2009;
0.5 GPa for EtMe3 Sb½PdðdmitÞ2 2 (Kato et al., 2007). (Note Koretsune and Hotta, 2014). Nevertheless, one can see that the
that these pressure values indicate pressures applied at room values have clear systematic variation in terms of anion X, as
temperature and are reduced by approximately 0.2 GPa at low shown in Fig. 17, where the values of t and t0 are calculated via
temperatures.) The temperature-pressure phase diagram of the latter method: the t0 =t value of κ-ðETÞ2 Cu½NðCNÞ2 Cl is
κ-ðETÞ2 Cu2 ðCNÞ3 is depicted in Fig. 16. A spin liquid phase 0.75 (the MO-based calculations) and 0.44–0.52 (first prin-
resides in proximity to the Mott transition; this feature appears ciples calculations), while that of κ-ðETÞ2 Cu2 ðCNÞ3 is 1.06
to be a key to the stability of spin liquids and can be closely and 0.80–0.99, respectively, suggestive of high geometrical
linked to the metal-insulator transition (Senthil, 2008; Zhou frustration.
and Ng, 2013). According to the numerical studies of the The temperature dependence of the spin susceptibility χ of
anisotropic triangular-lattice Hubbard model, the ground states κ-ðETÞ2 Cu2 ðCNÞ3 differs from that of the less frustrated
near to the Mott transition are controversial (Morita, Watanabe, compound κ-ðETÞ2 Cu½NðCNÞ2 Cl, as seen in Fig. 18
and Imada, 2002; Kyung and Tremblay, 2006; Watanabe et al., (Shimizu et al., 2003). An abrupt upturn at 27 K in the latter
2008; Tocchio et al., 2013; Laubach et al., 2015), implying that is a manifestation of the antiferromagnetic transition, with a
spin liquid and magnetic phases are competing very closely and slight spin canting of approximately 0.3° (Miyagawa et al.,
can be easily imbalanced by a small perturbation. 1995). However, κ-ðETÞ2 Cu2 ðCNÞ3 has no anomaly in χðTÞ.
Its overall behavior features a broad peak, which is reconci-
1. κ-ðETÞ2 Cu2 ðCNÞ3 led by the triangular-lattice Heisenberg model with an exchange
κ-ðETÞ2 Cu2 ðCNÞ3 is a layered compound, where κ-ðETÞ2 X interaction of J ∼ 250 K. In contrast to κ-ðETÞ2 Cu½NðCNÞ2 Cl,
has a variety of anions X and ET is bis(ethylenedithio) the magnetic susceptibility of κ-ðETÞ2 Cu2 ðCNÞ3 may be
tetrathiafulvalene (Komatsu et al., 1996). κ-ðETÞ2 X is com- described by the Heisenberg model because it is situated farther
posed of the ET layers with 1=2 hole per ET and the layers of from the Mott boundary, while κ-ðETÞ2 Cu½NðCNÞ2 Cl under-
monovalent anions X− , which have no contribution to the goes a Mott transition at a low pressure (25 MPa) as it is about to
electronic conduction or magnetism. In the ET layer, strong enter a metallic state (Lefebvre et al., 2000; Kagawa, Miyagawa,
ET dimers are formed ðETÞ2 , each of which accommodates a and Kanoda, 2005). There is no indication of magnetic ordering
hole in an antibonding orbital of the highest occupied in the susceptibility of κ-ðETÞ2 Cu2 ðCNÞ3 , at least down to 2 K,
molecular orbital of the ET. As the antibonding band is half the lowest temperature measured. Furthermore, no Curie-like
filled and the Coulomb repulsive energy is comparable to the upturn can be identified; the concentration of Cu2þ impurity
band width, the family of κ-ðETÞ2 X is a good model system to spins detected by electron spin resonance is estimated to be less
study Mott physics (Kino and Fukuyama, 1995; Kanoda, than 0.01% for κ-ðETÞ2 Cu2 ðCNÞ3 (Shimizu et al., 2006).
1997a, 1997b; Shimizu et al., 2006; Powell and McKenzie, The detailed spin states can be examined by performing
2011). The estimates of the transfer integrals between the nuclear magnetic resonance (NMR) measurements, which
adjacent antibonding orbitals on the isosceles triangular probe the static and dynamical hyperfine fields at the nuclear
lattices t and t0 are around 50 meV, depending on the method sites. Figure 19 shows the single-crystal 1 H NMR spectra for
of calculation, e.g., either the molecular-orbital- (MO-) based the two compounds (Shimizu et al., 2003). A clear line splitting

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-33


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

in κ-ðETÞ2 Cu½NðCNÞ2 Cl at 27 K is evidence for commensu-


rate antiferromagnetic ordering, with the moment estimated to
be 0.45μB per ET dimer in separate 13 C NMR studies
(Miyagawa, Kanoda, and Kawamoto, 2004). However, the
spectra for κ-ðETÞ2 Cu2 ðCNÞ3 show neither a distinct broad-
ening nor splitting down to 32 mK, which is 4 orders of
magnitude lower than the J value of 250 K. This indicates the
absence of long-range magnetic ordering. The absence of
ordering is also corroborated by zero-field μSR experiments
(Pratt et al., 2011). The nuclear spin-lattice relaxation rate
1=T 1 , which probes the spin dynamics, behaves similarly at the
1
H and 13 C sites. Figure 20 shows 1=T 1 at the 13 C sites, which
decreases monotonically with a square-root temperature
dependence down to 10 K and exhibits a diplike anomaly at
approximately 6 K (Shimizu et al., 2006). Below 6 K, 1=T 1
levels off down to 1 K or lower, followed by a steep decrease
approximated by T 3=2 at even lower temperatures. The two
anomalies at 6 and 1.0 K are obvious. However, they are not so
sharp as to be considered as phase transitions. Because of the
large hyperfine coupling of the 13 C sites located in the central
part of ET, an electronic inhomogeneity gradually developing
on cooling is captured by spectral broadening, which is
enhanced at approximately 6 K and saturates below 1 K
(Kawamoto et al., 2006; Shimizu et al., 2006). The detailed
FIG. 18. Magnetic susceptibility of polycrystalline NMR (Shimizu et al., 2006) and μSR (Pratt et al., 2011)
κ-ðETÞ2 Cu2 ðCNÞ3 and κ-ðETÞ2 Cu½NðCNÞ2 Cl. The core diamag- measurements point to the field-induced emergence of stag-
netic susceptibility has already been subtracted. The solid and geredlike moments, which is distinct from the conventional
dotted lines represent the results of the series expansions of the magnetic order. A separate μSR study (Nakajima et al., 2012)
triangular-lattice Heisenberg models using [6=6] and [7=7] Padé suggests a phase separation. The degree of inhomogeneity in
approximations with J ¼ 250 K. The susceptibility of the 13 C relaxation curve, which is characterized by the
κ-ðETÞ2 Cu2 ðCNÞ3 below 30 K is expanded in the inset. From deviation of the exponent in the stretched-exponential fitting
Shimizu et al., 2003. of the relaxation curve (see the inset of Fig. 20), increases below

13
FIG. 20. C nuclear spin-lattice relaxation rate for a single crystal
of κ-ðETÞ2 Cu2 ðCNÞ3 . The open triangles and circles represent the
relaxation rates of two separated lines coming from two non-
equivalent carbon sites in an ET. At low temperatures below 5 K,
the two lines merge and are not distinguished. The inset shows the
FIG. 19. 1 H NMR spectra for single crystals of exponent in the stretched-exponential fitting to the relaxation
κ-ðETÞ2 Cu2 ðCNÞ3 and κ-ðETÞ2 Cu½NðCNÞ2 Cl. From Shimizu curves of the whole spectra, whose relaxation rates are plotted
et al., 2003. using closed diamonds. From Shimizu et al., 2006.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-34


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

5–6 K (Shimizu et al., 2006). The 1 H relaxation curve also conductivity divided by the temperature tends to vanish with
starts to bend at the much lower temperatures, e.g., below 0.4 K, decreasing temperature, as shown in Fig. 22. The gap, if one is
and fits to a roughly equally weighed sum of two exponential present, is estimated to be 0.43 K, which is quite small
functions, the 1=T 1 ’s of which are proportional to T and T 2 . compared with the exchange energy of 250 K. The extremely
No appreciable field dependence of the 13 C relaxation rate is small gap may indicate a gapped Z2 spin liquid located near a
observed between 2 and 8 T. There is no experimental quantum critical point. The discrepancy between the thermal
indication of a finite excitation gap in any of the magnetic transport and NMR and specific heat data remains an open
measurements. issue. It may be attributed to the Anderson localization of
Thermodynamic investigations were conducted by means of spinons.
the specific heat and thermal conductivity measurements. The 6-K anomaly in the NMR spectrum and relaxation rate
Figure 21 shows the specific heat for κ-ðETÞ2 Cu2 ðCNÞ3 and also manifests itself in the specific heat (Yamashita et al.,
several Mott insulators with antiferromagnetic spin ordering 2008) and thermal conductivity (Yamashita et al., 2009) as a
(Yamashita et al., 2008). For all of the antiferromagnetic hump and a shoulder, respectively, indicating that the anomaly
materials, the electronic specific heat coefficient γ is vanishing is thermodynamic, as well as magnetic. However, the thermal
as expected for insulators. For the κ-ðETÞ2 Cu2 ðCNÞ3 spin expansion coefficient shows a cusp (Manna et al., 2010) and
liquid system, however, the extrapolation of the C=T vs T 2 line the ultrasonic velocity shows a diplike minimum, signifying
to absolute zero yields γ ¼ 12–15 mJ=K2 mol. The linearity lattice softening at approximately 6 K (Poirier et al., 2014). In
holds down to 0.3 K, below which a nuclear Schottky view of these results, this anomaly is likely associated with
contribution overwhelms the electronic contribution to C. spin-lattice coupling. Instabilities of the spinon Fermi surfaces
The finite value despite the Mott insulating state is a marked (Lee and Lee, 2005; Galitski and Kim, 2007; Grover et al.,
feature of spin liquids and suggests fermionic excitations in the 2010; Zhou and Lee, 2011) are among the possible origins of
spin degrees of freedom. Interestingly, the low-temperature the anomaly.
susceptibility and the γ value give the Wilson ratio on the order Although the spin liquid is insulating, anomalous charge
of unity. A spinon Fermi sea is an intriguing model for this dynamics are suggested for the low-energy optical and
phenomenon (Motrunich, 2005). However, neither the Uð1Þ dielectric responses. The optical gap for κ-ðETÞ2 Cu2 ðCNÞ3
is much smaller than that for κ-ðETÞ2 Cu½NðCNÞ2 Cl, although
spin liquid, where C follows T 2=3 scaling, nor the Z2 spin
the former system is situated farther from the Mott transition
liquid, where C is gapped, reconciles the observed features in
than the latter (Kézsmárki et al., 2006). It is proposed that
their original forms. Randomness may be an optional param-
gapless spinons are responsible for low-energy optical absorp-
eter to modify the temperature dependence. Another interesting
tion inside the Mott gap (Ng and Lee, 2007). The dielectric
feature is the field insensitivity up to 8 T, which appears
(Abdel-Jawad et al., 2010), microwave (Poirier et al., 2012),
incompatible with the Uð1Þ spin liquid states with Dirac cones.
and terahertz (Itoh et al., 2013) responses are enhanced at low
Thermal transport measurements result in somewhat con-
temperatures. The possible charge-imbalance excitations
troversial consequences (Yamashita et al., 2009). The thermal
within the dimer are theoretically proposed (Hotta, 2010;
Naka and Ishihara, 2010; Dayal et al., 2011). Relaxorlike

FIG. 21. Low-temperature specific heat Cp of


κ-ðETÞ2 Cu2 ðCNÞ3 for several magnetic fields up to 8 T in
Cp =T vs T 2 plots. Those of antiferromagnetic insulators FIG. 22. Low-temperature thermal conductivity κ of
κ-ðETÞ2 Cu½NðCNÞ2 Cl, deuterated κ-ðETÞ2 Cu½NðCNÞ2 Br, and κ-ðETÞ2 Cu2 ðCNÞ3 (samples A and B) in κ=T vs T 2 plots. Sample
β-ðETÞ2 ICl2 are also plotted for comparison. From Yamashita et A was investigated at 10 T applied perpendicular to the basal
al., 2008. plane, as well as at 0 T. From Yamashita et al., 2009.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-35


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

dielectric, transport, and optical properties are discussed in


terms of coupling with disordered anion layers (Pinterić et al.,
2014; Dressel et al., 2016).

2. EtMe3 Sb½PdðdmitÞ2 2
This compound is a member of the A½PdðdmitÞ2 2 family of
materials, which contains a variety of monovalent cations such
as Aþ ¼ Etx Me4−x Zþ (Et ¼ C2 H5 , Me ¼ CH3 , Z ¼ N, P, As,
Sb, and x ¼ 0, 1, 2), where dmit is a 1,3-dithiole-2-thione-4,5-
dithiolate (Kato, 2014) and A½PdðdmitÞ2 2 is a layered system
composed of conducting PdðdmitÞ2 layers and insulating A
layers. In the conducting layers, PdðdmitÞ2 is strongly
dimerized as in κ-ðETÞ2 X, whereas the ½PdðdmitÞ2 2 dimer
accepts an electron from cation Aþ instead of the hole in ETþ 2.
A prominent feature of the A½PdðdmitÞ2 2 family is that the
transfer integrals of the three laterals in the triangular lattice
can be finely tuned via chemical substitution of Aþ ¼
Etx Me4−x Zþ (Kato, 2014). Their first principles calculations
are shown in Fig. 23 (Tsumuraya et al., 2013) The spin liquid
material EtMe3 Sb½PdðdmitÞ2 2 is in a region where the three
transfer integrals are equalized. As expected, the materials
situated outside of this region have antiferromagnetic ground
states. The alloying of the boundary materials offers the
chance to study possible critical regions between spin liquids FIG. 24. Magnetic susceptibility of an antiferromagnet
Me4 Sb½PdðdmitÞ2 2 , a spin liquid EtMe3 Sb½PdðdmitÞ2 2 , and a
and ordered states (Kato, 2014). There is a charge-ordered
charge-ordered insulator Et2 Me2 Sb½PdðdmitÞ2 2. The core diamag-
material near the spin liquid, suggesting that the charge cannot
netic susceptibility has already been subtracted. From Kato, 2014.
always be assumed to be separate degrees of freedom from the
spin physics.
Next we review the properties of EtMe3 Sb½PdðdmitÞ2 2 and 2014), which is reminiscent of κ-ðETÞ2 Cu2 ðCNÞ3 . The fitting
other related materials. of the triangular-lattice Heisenberg model to the data yields an
The magnetic susceptibility of EtMe3 Sb½PdðdmitÞ2 2 shows exchange interaction of 220 to 280 K, which is nearly the
a broad peak at approximately 50 K and points to a finite value same as for κ-ðETÞ2 Cu2 ðCNÞ3. Also shown are the suscep-
in the low-temperature limit without any anomaly down to tibilities of antiferromagnetic and charge-ordered insulators,
2 K, as shown in Fig. 24 (Kanoda and Kato, 2011; Kato, which exhibit small kink signaling of magnetic ordering and a
sudden decrease indicative of a spin gapful state, respectively,
(a) despite their similar behaviors at high temperatures (Tamura
and Kato, 2002). This indicates that the diversity in the ground
states is an outcome of low-energy physics, while the same
diversity is not distinguished at high-energy scales.
The 13 C NMR captures no signature of magnetic ordering
down to 20 mK, although a slight broadening equivalent to the
(b) broadening for κ-ðETÞ2 Cu2 ðCNÞ3 is observed at low temper-
atures (Itou et al., 2010). The temperature dependence of the
13
C nuclear spin-lattice relaxation rate is shown in Fig. 25
(Itou et al., 2010). It exhibits a nonmonotonic temperature
dependence. At low temperatures below 1 K, it follows a T 2
dependence, suggesting no finite gap. However, the power of 2
implies a complicated nodal gap, which is not consistent with
the finite susceptibility value and the thermodynamic mea-
surements described later. Furthermore, 1=T 1 forms a shoulder
or a kink at approximately 1 K and becomes moderate in
temperature dependence above 1 K. The kink temperature
increases for higher magnetic fields or frequencies. The
FIG. 23. First principles calculations of (a) bandwidth W and relaxation curve becomes a nonsingle exponential curve below
(b) transfer integrals in A½PdðdmitÞ2 2 for various cations A. The 10 K but reverses below 1 K, indicating that the inhomogeneity
PdðdmitÞ2 layers are modeled to triangular lattices characterized by increases below 10 K (Itou et al., 2010, 2011). The reversal
transfer integrals tB , tS , and tr . t3 is the interlayer transfer integral. at 1 K can be an indication of either a recovery in the
AF, QSL, and CO stand for antiferromagnet, quantum spin liquid, homogeneity below 1 K or the microscopic nature of the
and charge-ordered insulator. From Tsumuraya et al., 2013. inhomogeneity, which is subject to spin-diffusion averaging of

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-36


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

13 C nuclear spin-lattice relaxation rate 1=T 1 of


FIG. 25. FIG. 27. Low-temperature thermal conductivity κ of EtMe3 Sb ×
EtMe3 Sb½PdðdmitÞ2 2 . The inset shows the 1=T 1 T vs T plots. ½PdðdmitÞ2 2 (dmit-131) in κ=T vs T 2 and κ=T vs T (inset) plots.
The circles indicate the values determined from the stretched- The data of other insulators, i.e., Et2 Me2 Sb½PdðdmitÞ2 2 (dmit-
exponential fitting to the relaxation curves and the squares denote 221, nonmagnetic) and κ-ðETÞ2 Cu2 ðCNÞ3 , are also plotted for
the values determined from the initial decay slopes of the comparison. From Yamashita et al., 2010.
relaxation curves. From Itou et al., 2010.

the heterogeneous relaxation time that is longer at lower The linearity of C=T vs T 2 in EtMe3 Sb½PdðdmitÞ2 2
temperatures. The 1-K relaxation-rate anomaly in is extrapolated to a zero kelvin to give a finite value of γ,
Et2 Me2 Sb½PdðdmitÞ2 2 may be compared to the broad anomaly whereas other Mott insulators appear to have vanishing γ, as
around nearly the same temperature for κ-ðETÞ2 Cu2 ðCNÞ3. expected for conventional insulators. There is no field depend-
However, they appear different with respect to field (or ence in C=T in EtMe3 Sb½PdðdmitÞ2 2 up to 8 T, as in
frequency) dependence and spatial scale of inhomogeneity. κ-ðETÞ2 Cu2 ðCNÞ3 . The thermal conductivity results are con-
The thermodynamic measurements are indicative of fer- sistent with the specific heat results, as seen in Fig. 27, where
mionic low-energy excitations. Figure 26 shows the temper- the low-temperature κ=T value for EtMe3 Sb½PdðdmitÞ2 2 is as
ature dependence of the specific heat (Yamashita et al., 2011). high as 0.2 WK−2 m in the zero-kelvin limit, implying the
presence of gapless thermal transporters with fermionic sta-
tistics (Yamashita et al., 2010). The mean free path for thermal
transport is estimated to be of the order of 1 μm. κ is enhanced
by the application of a magnetic field above a threshold value,
suggesting that the gapped excitations coexist with the gapless
excitations (Yamashita et al., 2010).

B. Kagome-lattice system: ZnCu3 ðOHÞ6 Cl2

The kagome lattice is constructed by using corner-sharing


triangles in contrast to the edge sharing in the triangular
lattices, as shown in Fig. 8. Thus, the spin states in the kagome
lattice have larger degeneracy than those in the triangular
lattices, leading to high potential for hosting a spin liquid.
Actually, the theoretical perspective of seeking a spin liquid is
more promising for the kagome-lattice Heisenberg model than
for the triangular lattice (Sachidev, 1992; Lecheminant et al.,
FIG. 26. Low-temperature specific heat Cp of EtMe3 Sb × 1997; Mila, 1998; Misguich and Lhuillier, 2004). Among
½PdðdmitÞ2 2 for several magnetic fields up to 10 T in Cp =T several candidates for the kagome spin systems, we select a
vs T 2 plots. The data of other insulating systems, i.e., spin-1=2 system ZnCu3 ðOHÞ6 Cl2 , which is known as her-
Et2 Me2 As½PdðdmitÞ2 2 , EtMe3 As½PdðdmitÞ2 2 , and EtMe3 P × bertsmithite, whose magnetism has been extensively inves-
½PdðdmitÞ2 2 , are also plotted for comparison. A large upturn tigated. This is a member of a family of materials with variable
below 1 K is probably attributable to the rotational tunneling of compositions, i.e., Znx Cu4−x ðOHÞ6 Cl2 (0 < x < 1). As an
Me groups. The low-temperature data are expanded in the inset. end material, Cu4 ðOHÞ6 Cl2 has a distorted pyrochlore lattice
From Yamashita et al., 2011. of S ¼ 1=2 Cu2þ spins, whereas the other end material

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-37


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

ZnCu3 ðOHÞ6 Cl2 has a two-dimensional (a − b plane) perfect


kagome lattice of Cu2þ spins separated by different crystallo-
graphic sites occupied by Zn2þ (Shores et al., 2005). The
structural symmetry changes across x ¼ 0.33, above which
Cu2þ partially occupies the Zn sites in addition to the kagome
lattice. There is an argument for the mixture of Zn in the
kagome sites in ZnCu3 ðOHÞ6 Cl2 . Magnetic susceptibility
(Bert et al., 2007) and specific heat (de Vries et al., 2008)
suggest that approximately 6% of the kagome sites are
replaced by nonmagnetic Zn. The same amount of Cu is
assumed to invade the nominal Zn sites. Thus, significant
effort has been made to extract the intrinsic properties of the
kagome lattice from the experimental data.
Experimental evidence for the absence of magnetic order- FIG. 29. Temperature dependence of the inverse magnetic
ing in ZnCu3 ðOHÞ6 Cl2 can be obtained from μSR experiments susceptibility χ −1 of ZnCu3 ðOHÞ6 Cl2 . The line denotes a Cu-
(Mendels et al., 2007). The relaxation profile shows no rie-Weiss fit. Inset: ac susceptibility (at 654 Hz) at low temper-
internal field down to 50 mK. The experiments for a wide atures. From Helton et al., 2007.
range of x found that the absence of an internal field was
persistent in a certain range below x ¼ 1 (see Fig. 28)
(Mendels et al., 2007). The magnetic susceptibility exhibits and 17 O sites are broad (Imai et al., 2008; Olariu et al., 2008),
a Curie-Weiss behavior at high temperatures above 100 K, as reflecting the inhomogeneous local fields, supposedly due to
shown in Fig. 29 (Helton et al., 2007). The Weiss temperature the Zu/Cu mixture. However, the smallest shift value in the
is ∼300 K, which implies an antiferromagnetic exchange broad 35 Cl spectrum follows a Curie-Weiss law down to 25 K,
interaction of J ¼ 17 meV. The dc and ac magnetic suscep- followed by a decrease at lower temperatures (Imai et al.,
tibilities indicate no magnetic ordering down to 0.1 and 2008). This is considered to indicate intrinsic magnetism for
0.05 K, respectively, which is 4 orders of magnitude lower the kagome lattice (Imai et al., 2008). The 17 O probes the
than J (Helton et al., 2007). The susceptibility increases kagome sites more preferentially than the nominal Zn sites
progressively at lower temperatures. Two mechanisms are due to larger hyperfine coupling with the kagome sites (Olariu
possible. First, impurities from Cu/Zn intersite mixing can et al., 2008). The 17 O NMR spectra were decomposed into
give a Curie-like upturn. Second, Dzyaloshinsky-Moriya two components. One is from the 17 O sites coordinated by two
interactions may be present between the adjacent sites with Cu2þ ions, while the other is from the 17 O sites coordinated by
broken inversion symmetry, as in the kagome lattice (Rigol a Cu2þ and a Zu2þ in the kagome plane. The relative fraction
and Singh, 2007). The high-field magnetization measurements of the two components was consistent with a 6% Zn
suggest that the former is mainly responsible for the increasing admixture. The NMR shifts of the respective components,
susceptibility (Bert et al., 2007). as shown in Fig. 30, are considered to be local susceptibilities
NMR, which probes magnetism in a site-selective manner,
was informative particularly for this material because the
analysis of spectra allows one to distinguish the intrinsic
magnetism from the extrinsic one. The NMR spectra at 35 Cl

17
FIG. 30. O NMR shift of two lines (M and D) decomposed
from the observed spectra for a powder of ZnCu3 ðOHÞ6 Cl2 .
The M and D lines are considered to come from the oxygen
sites depicted in the inset. The solid (red) curve represents
FIG. 28. Temperature variation of the spin-frozen fraction deter- the trace of one-half of the value of the M line. The sketch in
mined by muon spin rotation experiments for Znx Cu4−x ðOHÞ6 Cl2. the lower left corner illustrates the environment of a Zn
The inset shows the x dependence of the spin-frozen fraction at a low substituted on the Cu kagome plane, and thick lines represent
temperature. From Mendels et al., 2007. Cu-Cu dimers. From Olariu et al., 2008.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-38


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

at Cu sites with and without Zn2þ at the neighboring sites


(Olariu et al., 2008). Both decrease below 50 K and saturate to
finites values, indicating the gapless nature of the spin
excitations. The low-temperature decrease in the shift at the
Cu site with a Zn neighbor is in contrast to the enhancement
commonly observed in the neighborhood of nonmagnetic
impurities (Olariu et al., 2008). This behavior also suggests
that the Curie-like upturn in the bulk susceptibility at low
temperatures is not from the kagome plane. For the NMR
relaxation rate, all of the O, Cl, and Cu nuclear spins exhibit
power laws against temperature down to 0.47 K for O and 2 K
or lower for Cl and Cu, indicating a gapless spin liquid (see
Fig. 31) (Imai et al., 2008; Olariu et al., 2008). Although the
power somewhat depends on the nuclear site, the relaxation
profile is overall nuclear site insensitive, which is filtered by
the nuclear site-specific form factor determined by its location
relative to the kagome lattice, suggesting nondispersive spin
dynamics. Otherwise, the temperature profile of the relaxation
rate would be site dependent (Olariu et al., 2008). This feature
is potentially relevant to the spinon excitation with the
continuum. More recently, NMR experiments performed at
low temperatures have revealed an anomaly in the relaxation
rate at a temperature depending on the applied field, which
may signify field-induced spin freezing (Jeong et al., 2011).
Recently, a 17 O NMR experiment performed with a single
crystal found different features from those observed so far in
the powder samples (Fu et al., 2015). According to the
analysis of NMR spectra, there is no significant contamination
of Zn in the Cu sites within the kagome plane, and the Knight
shift shows appreciable temperature and field dependences,
suggesting a spin gap of the order of 10 K, as shown in
Fig. 32, in contradiction with the consequences of the earlier
NMR and neutron experiments.
The low-temperature specific heat was investigated under FIG. 32. (a) Temperature dependence of 17
O Knight shift and
external magnetic fields (Helton et al., 2007; de Vries et al., (b) the field dependence of the spin gap deduced from the Knight
2008). As shown in Fig. 33(a) (Helton et al., 2007), there is an shift for a single-crystal ZnCu3 ðOHÞ6 Cl2 . From Fu et al., 2015.
enormous field dependence. The temperature dependence at a
zero field is approximated by a power law T α with α unity or
smaller [see Fig. 33(b)]. The broad peak present even at a zero contribution associated with a field-induced gap, the data
field is shifted to higher temperatures under higher fields. for different fields were analyzed in detail to reveal the
Assuming that the field-dependent peak is a Schottky intrinsic specific heat of the kagome lattice (de Vries et al.,
2008). The deduced Schottky component is consistent with
Zeeman splitting of the 6% Cu impurities in the Zn sites at
higher fields, and the intrinsic C=T follows a power law T α
with α ¼ 1.3 as the best estimate, suggesting gapless excita-
tions (Helton et al., 2007; de Vries et al., 2008; Shaginyan,
Msezane, and Popov, 2011).
Neutron-scattering experiments, which are capable of
profiling spin excitations with respect to momentum and
energy transfers, are available for herbertsmithite. One of the
key issues of elementary excitations in spin liquids is the
possible fractionalization of S ¼ 1 spin excitations into S ¼
1=2 spinons, which could manifest themselves as a continuum
in the spin excitation spectrum, i.e., dynamic structure factor
SðQ; ωÞ, where Q and ω are momentum transfer and energy
transfer divided by ℏ, respectively. Such a continuum is
17
FIG. 31. O, 63 Cu, and 35 Cl nuclear spin-lattice relaxation rates observed in a highly anisotropic triangular-lattice system
1=T 1 for a powder of ZnCu3 ðOHÞ6 Cl2 . The inset shows 17 O 1=T 1 Cs2 CuCl4 (J 0 =J ∼ 3 and J0 ∼ 0.34 meV in Fig. 17), although
vs 1=T plots. From Olariu et al., 2008. it undergoes a magnetic transition into a spin-spiral order at

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-39


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

FIG. 35. Contour plot of dynamical structure factor, Smag


(Q, ω), integrated over 1 ≤ ℏω ≤ 9 meV for a single crystal
ZnCu3 ðOHÞ6 Cl2 at 1.6 K. The intense scattering is extended in a
green-colored region, without peaking at any specific points.
From Han et al., 2012.

inelastic scattering experiments exhibit no excitation gap at


least down to 0.1 meV, which corresponds to ∼J=170, and
insensitivity of the scattering strength to Q, indicating gapless
and local natures of spin fluctuations (Helton et al., 2007).
Furthermore, the scattering intensity is only weakly dependent
FIG. 33. (a) Specific heat C of ZnCu3 ðOHÞ6 Cl2 in various on ω up to 25 meV and temperature up to 120 K and shifts
applied fields. The inset shows C over a wider temperature range toward lower Q as temperature is increased (de Vries et al.,
in applied fields of 0 T (squares) and 14 T (stars). (b) C in a zero 2009). Some of the results are displayed in Fig. 34. All these
field at low temperatures. The lines represent power-law fits. features are suggestive of a continuum in spin excitations and
From Helton et al., 2007. the persistence of the short-range nature of spin correlations
even at low temperatures. Recent experiments on a large
0.62 K (Coldea et al., 2001; Coldea, Tennant, and Tylczynski, single crystal have succeeded in capturing the continuum
2003). Several features signifying the continuum are found nature, as seen in the dark (green) area in Fig. 35. The
via neutron experiments of herbertsmithite, which were first momentum profile of the excitation intensity (dynamic struc-
performed for polycrystalline or powder samples. The ture factor) SðQ; ωÞ is approximately reproduced by the
simulated structure factor of uncorrelated dimer singlets,
Smag (mBarns st. fu. )

which indicates the short-ranged spin correlations at least


-1

(a) D7 Smag, 4 K
8
D7 Smag, 10 K down to 1.6 K (Han et al., 2012). The short-range nature that
-1

6 singlet-triplet form fact.


+ single ionic term (6%)
persists even at low temperatures, as suggested by the powder
4 experiments as well, is generally in favor of a gapped state,
Stot (mBarns st. meV form. unit )

(b)
-1

2K, IN4, ε = 6±2 meV 2.5


2 2K, MARI, ε = 6±2 meV whereas there is no indication of a spin gap down to 0.25 meV
2K, MARI, ε = 22±2 meV
singlet-triplet form. fact.
at any Q values in the excitation spectra (Han et al., 2012). It
0 2
0 0.5 1 1.5
-1
2 2.5 is puzzling that the spin dynamic correlation exhibits short-
-1

Q(Å )
range RVB nature while the spectrum is gapless. One
8 (c) -1
Q = 1.3±0.2 A 1.5
possibility is that the herbertsmithite is in a Z2 spin liquid
Stot (arb. units)

-1

6 in close proximity to a critical point to the Uð1Þ Dirac liquid,


1
4 2K MARI as indicated by some recent numerical works (Li, 2016),
2K IN4 although the true ground state of the isotropic Heisenberg
2 60K IN4
0.5
120K IN4 model on a kagome lattice is still under debate (Iqbal,
0
3 6 9 12 1 1.5 2 2.5 Poilblanc, and Becca, 2016).
ε (meV) -1
Q(Å )

FIG. 34. (a) Instantaneous magnetic correlations at 4 and 10 K C. Hyperkagome-lattice system: Na4 Ir3 O8
for a time scale corresponding to approximately 6.5 meV. The
solid lines are a guide to the eye. (b) The Q dependence in the
The hyperkagome lattice is a three-dimensional network of
dynamic correlations with the energy integration interval indi- corner-sharing triangular lattices. In Na4 Ir3 O8 , the Ir4þ ion
cated in the legend. The dotted lines in (a) and (b) are the structure with 5d5 electrons likely takes on a low-spin state. These ions
factors for dimerlike AF correlations. The dashed line, a single- locate on the corners, forming a S ¼ 1=2 hyperkagome lattice
ion contribution corresponding to the 6% antisite spins in this (Okamoto et al., 2007). The resistivity of the ceramic sample
system, is added. (c) The energy and temperature dependence at is 10 Ω cm at room temperature. The samples are semi-
Q ¼ 1.3 Å−1 . D7, IN4, and MARI in the legends stand for the conducting, with a charge transport gap of 500 K, implying
types of spectrometers used. From de Vries et al., 2009. the proximity of this system to the Mott transition, which is

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-40


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

different from the kagome materials previously reviewed The electronic (magnetic) contribution to the specific heat
(Okamoto et al., 2007). A connection between the spin liquid of Na4 Ir3 O8 , as shown in Fig. 36(b), has a broad peak at 20 K.
and the metal-insulator transition, similar to the case of However, no anomaly signifying magnetic ordering is appar-
κ-ðETÞ2 Cu2 ðCNÞ3 , is shown (Podolsky et al., 2009). A ent (Okamoto et al., 2007). The magnetic entropy estimated
distinct feature of Na4 Ir3 O8 among spin liquid candidates by integrating the C=T in Fig. 36(b) reaches 70%–80% of
is its large spin-orbit coupling, which introduces additional R ln 2 (¼ 5.7 J=mol K) at 100 K, a much lower temperature
interest to the physics of spin liquids (Chen and Balents, 2008; than the Weiss temperature of ∼600 K, which features
Zhou et al., 2008). Several theoretical studies propose that frustrated magnetism. The C=T is characterized by a curious
Na4 Ir3 O8 is a 3D QSL with fermionic spinons (Lawler et al., T 2 dependence at the lowest temperatures. The γ term, when
2008; Zhou et al., 2008). present, appears on the order of 1 mJ=K2 mol Ir. Recent
Figure 36(a) shows the magnetic susceptibility of Na4 Ir3 O8 , experiments extended down to 500 m K have found that Cm =T
which weakly increases with decreasing temperature, as char- is well approximated by a form of γ þ βT 2.4 with γ ¼
acterized by the Curie-Weiss temperature of −650 K (Okamoto 2.5 mJ=K2 mol Ir (Singh et al., 2013). As seen in the inset
et al., 2007). This implies an antiferromagnetic interaction of of Fig. 36(b), the applied field has no influence on the specific
hundreds of kelvin. There is no clear indication of magnetic heat, at least up to 12 T.
ordering at least down to 2 K, whereas a small anomaly The temperature dependence of thermal conductivity is
reminiscent of spin glass observed in the magnetization history shown in Fig. 37 (Singh et al., 2013). At low temperatures
against the field-temperature variation is attributed to a small down to 75 mK, κ=T is linear in T 2 . The κ=T value
fraction of the total spins (Okamoto et al., 2007). extrapolated to T ¼ 0 is 6.3 × 10−2 mW=K2 m, which is a
vanishingly small value, compared with the value of
EtMe3 Sb½PdðdmitÞ2 2 , 0.2 W=K2 m in Fig. 27. The suppres-
(a) sion of the κ=T value by the extrinsic grain-boundary effect is
not ruled out (Singh et al., 2013). The feature that γ is
diminished and κ=T is vanishing at low temperatures, while
both are sizable at high temperatures of the order of kelvin,
appears to be in accordance with a theoretical picture of
spinon Fermi surfaces that undergo a pairing instability at low
temperatures (Zhou et al., 2008). In this context, the magnetic
susceptibility, remaining large even at low temperatures, can
be due to the large spin-orbit interactions of Ir (Zhou
(b) et al., 2008).
The substitution of nonmagnetic Ti4þ ions at Ir sites will
give rise to a Curie-like tail in the spin susceptibility curve
(Okamoto et al., 2007), similar to Zn substitution for Cu in
high-T c cuprates, indicating an RVB spin background. The
scaling analysis of magnetic Gruneisen parameters is sugges-
tive of the proximity of Na4 Ir3 O8 to a zero-field quantum
critical point (Singh et al., 2013).

(c)

FIG. 36. (a) Temperature dependence of the inverse magnetic


susceptibility χ −1 of polycrystalline Na4 Ir3 O8 under 1 T. The
inset shows magnetic susceptibility χ in various fields up to 5 T;
for clarity, the curves are shifted by 3 × 10−4, 2 × 10−4 , and 1 ×
10−4 emu=mol Ir for 0.01, 0.1, and 1 T data, respectively.
(b) Magnetic specific heat Cm divided by temperature T of
polycrystalline Na4 Ir3 O8 . To estimate Cm , data for Na4 Sn3 O8 is
used as a reference of the lattice contribution. The inset shows FIG. 37. Thermal conductivity κ of Na4 Ir3 O8 in κ=T vs T 2 plots
Cm =T vs T in various fields up to 12 T. (c) Magnetic entropy. for magnetic fields of 0 and 5 T. The inset shows the low-
From Okamoto et al., 2007. temperature part of the data. From Singh et al., 2013.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-41


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Recent μSR (Dally et al., 2014) and NMR experiments D. Experimental summary
(Shockley et al., 2015) have found some indications that are
not in accordance with these claims. Both probes detected Because of intensive experimental studies, unconventional
the emergence of local fields signifying the freezing of thermodynamic and magnetic properties that evoke spin
moments at low temperatures, as shown in Fig. 38. The liquids have been found in several materials with anisotropic
muons are revealed to sense an inhomogeneous local field of triangular lattices, kagome lattices, and hyperkagome lattices
electronic origin that appears at 6 K, where the irreversibility as seen. These materials exhibit no indications of conventional
in magnetization occurs and levels off to 70 G on average, magnetic ordering. Their magnetic and thermodynamic prop-
which may correspond to 0.5μB on Ir. It is suggested, erties are summarized in Table III. It appears that the gapless
however, that the spin correlation is short ranged (of the nature is a property that a class of frustrated lattices con-
order of 1 unit cell) and quasistatic in that the slow dynamics structed with triangles possesses, although the thermal con-
captured by the relaxation rate persist down to 20 mK. The ductivity of κ-ðETÞ2 Cu2 ðCNÞ3 suggested a small excitation
quasistatic nature is also seen in the S ¼ 1 triangular-lattice gap 3 orders of magnitude smaller than J. Recent NMR work
system NiGa2 S4 (Nakatsuji et al., 2005; MacLaughlin et al., on herbertsmithite insists on gapped spin excitations, and
anomalous quasistatic spin freezing has recently been revealed
2008). 17 O and 23 Na NMR lines show broadening, which is
by μSR and NMR studies of the hyperkagome system. This
roughly scaled to the μSR results at low temperatures, as
feature and the successful observation of fractionalized
seen in Fig. 38; the moment is estimated at 0.27μB on Ir. The
excitations in a kagome lattice (Han et al., 2012) tempt
NMR line profile also suggests inhomogeneous spin freez-
one to think about spinons as promising elementary excita-
ing and slow dynamics persisting down to low temperatures
tions in spin liquids. How to detect the spinon Fermi surfaces,
although the temperature dependences of the relaxation rates
if they exist, is a focus—smoking-gun experiments are
on the muon and 23 Na differ. Noticeably, the 23 Na relaxation
awaited.
rate exhibits a peak indicative of the critical slowing down at
As seen in Table III, several experimental characteristics are
approximately 7.5 K despite no anomaly in specific heat. seemingly inconsistent within given materials; understanding
The nature and origin of these anomalous properties are not the apparently contradicting data in a consistent way requires
clear at present; however, it is likely that disorder plays a clarification of the nature of the spin states. One of the key
vital role in this system, which can host configurationally issues may be the randomness present in real materials. In
degenerate phases with fluctuating order (Dally et al., 2014). particular, it has long been recognized that the effect of
Considering that muon, 17 O, and 23 Na captured the behavior inevitable Zn/Cu admixtures in herbertsmithite has to be
of the majority of spins in the sample, the disorder effect, if separated from the intrinsic magnetism. More recently, the
any, is such that it is not restricted to finite areas but issue of inhomogeneous quasistatic spin correlation with slow
extended over the system, being reminiscent of the quantum dynamics in the hyperkagome-lattice system has emerged as a
Griffiths effect given the inhomogeneity and slow dynamics. consequence of disorder. Theoretically, it was proposed that as
randomness is intensified, the 120-degree Néel order in the
triangular-lattice Heisenberg model is changed to a sort of
3.90 3.95 4.00 4.05 random singlets but not spin glass state. It is intriguing that
0.5 4 randomness appears to enhance the quantum nature because
Magnitude (A.U.)

Cu the singlet is a purely quantum state (Watanabe et al., 2014;


0.4 Shimokawa, Watanabe, and Kawamura, 2015). In the case of
3 kagome lattices, it was theoretically suggested that disorder
could lead to a valence-bond glass state (Singh, 2010) or a
ΔH (kOe)

17

0.3 Frozen gapless spin liquid state (Kawamura, Watanabe, and


ΔH (kOe)

2 Shimokawa, 2014; Shimokawa, Watanabe, and Kawamura,


0.2 6.90 6.95 7.00 7.05 2015). Furthermore, a recent NMR experiment on an organic
23

Field (T) Mott insulator, i.e., κ-ðETÞ2 Cu½NðCNÞ2 Cl, found that the
Na 4T 1 antiferromagnetic ordering in the pristine crystal, when
0.1 Na 7T irradiated by x rays, disappears. Spin freezing, spin gap,
O 6.5T
μSR and critical slowing down are not observed, but gapless spin
0.0 0 excitations emerge, suggesting a novel role of disorder that
0 5 7.2K 10 15 20
brings forth a QSL from a classical ordered state (Furukawa
Temperature (K)
et al., 2015b). Whether the randomness is fatal or vital to the
FIG. 38. The line width (FWHM) of Gaussian-broadened O 17 physics of a QSL is a nontrivial issue to be resolved.
23
and Na NMR spectra and the mean value of the distributed local The development of new materials, although not addressed
fields detected based on μSR (Dally et al., 2014). For the NMR in this review, is under way. Among them is a new type of
line width, its deviation from the value at 15 K is plotted. Inset: hydrogen-bonded κ-H3 ðCat-EDT-TTFÞ2 with a triangular
23 Na spectra at 78.937 MHz for 7 T (empty circles) and lattice of one-dimensional anisotropy (Isono et al., 2013)
45.046 MHz for 4 T (solid line) with the horizontal axis shifted and κ-ðETÞ2 Ag2 ðCNÞ3 , an analog of κ-ðETÞ2 Cu2 ðCNÞ3
by 3.005 T at 1.3 K. The square blue line shows the expected (Saito, 2014). Another compound with a hyperkagome lattice
powder pattern of the spectrum with every Ir site carrying the structure, i.e., PbCuTe2 O6 , with Curie-Weiss temperature
same moment. From Shockley et al., 2015. θ ¼ −22 K is also proposed to be a spin liquid candidate

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-42


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

TABLE III. Spin liquid materials summary.


Triangular, Triangular Kagome Hyperkagome,
Material κ-ðETÞ2 Cu2 ðCNÞ3 M½PdðdmitÞ2 2 ZnCu3 ðOHÞ6 Cl2 Na4 Ir3 O8
Susceptibility A broad peak at 60 K, A broad peak at 50 K, Curie-Weiss at high T Curie-Weiss
finite at 2 K, J ¼ 250 K finite at 2 K, ΘW ¼ −300 K, J ¼ 230 K, ΘW ¼ −650 K
(Shimizu et al., 2003) J ¼ 220–280 K upturn at low T due to (de Vries et al., 2009;
(Kato, 2014) impurity sites Han et al., 2012)
(Bert et al., 2007;
Helton et al., 2007)
Specific heat Gapless, γ ¼15 mJ=K2 mol, Gapless, Gapless, C ∼ T α , α ¼ 1.3 Gapless, C ∼ T 2
field independent γ ¼ 20 mJ=K2 mol, at high fields (de Vries et al., 2009),
(Yamashita et al., 2008) field independent (de Vries et al., 2008) C ∼ γT þ βT 2.4 ,
(Yamashita et al., 2011) γ ¼ 2 mJ=K2 mol
(Han et al., 2012),
field independent
(Okamoto et al., 2007;
Singh et al., 2013)
Thermal Gapped, Δ ¼ 0.46 K Gapless, finite κ=T Vanishingly small κ=T
conductivity (Yamashita et al., 2009) (Yamashita et al., 2010) (Singh et al., 2013)
NMR shift Not precisely resolved Not precisely resolved High-T, broad peak 17 O shift—scales to χ
bulk
(Shimizu et al., 2006) (Itou et al., 2010) at at 50–60 K for 17 O in 100–300 K but
(Olariu et al., 2008; levels off below 80 K
Fu et al., 2015), at (Shockley et al., 2015),
17
25–50 K for 35 Cl O, 23 Na inhomogeneous
(Imai et al., 2008), line broadening at low T
low-T gapless: finite value (Shockley et al., 2015)
(Olariu et al., 2008)
gapped: Δ ∼ 10 K
(Fu et al., 2015)
NMR 1=T 1 Inhomogeneous 1=T 1 , Inhomogeneous 1=T 1 , 1=T 1 ∼ T α , α ∼ 0.73 23
Na 1=T 1 , a peak
power law, power law, for 17 O (Olariu et al., formation typical
1
H 1=T 1 ; ∼T= ∼ T 2 at 13
C 1=T 2 at < 0.5 K 2008), α ∼ 0.5 for of critical slowing
T < 0.3 K (stretched exponential) 63
O (Imai et al., 2008), down at 7.5 K
(two components) (Itou et al., 2010) field-induced spin freezing (Shockley et al.,
(Shimizu et al., 2003), (Jeong et al., 2011) 2015)
13 C 1=T ; ∼1=T 1.5
1
at T < 0.2 K
(stretched exponential)
(Shimizu et al., 2006)
μSR No internal field at 0 T No internal field at 0 T Emergence of distributed
(Pratt et al., 2011; (Mendels et al., 2007) local fields below 6 K,
Nakajima et al., 2012) quasistatic short-ranged
spin freezing with
slow dynamics
(Dally et al., 2014)
Neutron Powders ∼gapless
(< 0.1 meV)
(Helton et al., 2007;
de Vries et al., 2009),
single crystal ∼gapless
(< 0.25 meV)
(Han et al., 2012), continuum
in dynamic structure factor
(Helton et al., 2007;
de Vries et al., 2009;
Han et al., 2012)

(Koteswararao et al., 2014; Khuntia et al., 2016). The To remedy this situation, theorists have proposed new
entanglement of additional degrees of freedom with quantum experiments to identify QSLs through identifying nontrivial
spins may be another direction for future studies; e.g., properties of spinons and gauge fields. For example, power-
Ba3 CuSb2 O9 is proposed to host a spin-orbital coupled liquid law ac conductivity inside the Mott gap has been noted (Ng
state (Zhou et al., 2011; Nakatsuji et al., 2012). and Lee, 2007). A giant-magnetoresistance-like experiment
It should be emphasized that the identification of QSL was proposed to measure mobile spinons through oscillatory
experimentally is an important and challenging task. As a coupling between two ferromagnets via a QSL spacer
“featureless” Mott insulator, there exists no simple magnetic (Norman and Micklitz, 2009). The thermal Hall effect in
order for identifying QSL states, and so far there exists only insulating quantum magnets was proposed as a probe for the
indirect experimental evidence for mobile fermionic spinons thermal transport of spinons, where different responses were
in some candidate compounds as previously discussed. used to distinguish between magnon and spinon transports

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-43


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

(Katsura, Nagaosa, and Lee, 2010). Raman scattering was VI. SUMMARY
proposed as a signature to probe the Uð1Þ QSL state (Ko et al.,
2010). It was also proposed that the spinon lifetime and mass In this review, we provided a pedagogical introduction to
as well as gauge fluctuations can be measured through a sound the subject of QSLs and reviewed the current status of the
attenuation experiment (Zhou and Lee, 2011), and neutron field. We first discussed the semiclassical approach to simple
scattering can be used to detect scalar spin chirality fluctua- quantum antiferromagnets. We explained how it leads to the
tions in the kagome system (Lee and Nagaosa, 2013). Low- Haldane conjecture in one dimension and why it fails for
energy electron spectral functions were evaluated for future frustrated spin models. We then focused on spin-1=2 systems
angle-resolved photoemission spectroscopy (ARPES) experi- with spin rotational symmetry and introduced the RVB
ments (Tang, Fisher, and Lee, 2013) and it was proposed that concept and the slave-particle plus Gutzwiller-projected wave
spin current flow through a metal-QSL-metal junction can be function approaches. We explained the technical difficulties
used to distinguish different QSLs (Chen et al., 2013). More associated with these approaches and why slave-particle
recently, it was suggested that there exists a long-life surface approaches naturally led to gauge theories for spin liquid
plasmon mode propagating along the interface between a states. The natures of SUð2Þ, Uð1Þ, and Z2 spin liquid states
linear medium and a QSL with a spinon Fermi surface at were explained, and the extensions of the approach to systems
frequencies above the charge gap, which can be detected by with spin-orbit coupling and S > 1=2 systems were intro-
the widely used Kretschmann-Raether three-layer configura- duced. We explained that because of the intrinsic limitations
tion (Ma and Ng, 2015). of the analytical slave-particle approach, many alternative
However, there exists important discrepancies between approaches to spin liquid states were developed, both num-
existing experiments and theories in some of the previously erically and analytically. These approaches comple-
experiments. mented each other and often led to exotic possibilities not
(1) Specific heat: Using the one-loop calculation supple- covered by the simple fermionic slave-particle approach. The
mented by the scaling analysis (Lee and Nagaosa, 1992; experimental side of the story was also introduced with a
Polchinski, 1994), it was found that the strong coupling review on the properties of several candidate spin liquid
between the Uð1Þ gauge field and the spinon Fermi surface materials, including anisotropic triangular-lattice systems
[κ-ðETÞ2 Cu2 ðCNÞ3 and EtMe3 Sb½PdðdmitÞ2 2 ], kagome-
leads to the T 2=3 correction to the temperature dependence of
lattice systems [ZnCu3 ðOHÞ6 Cl2 ], and hyperkagome lattice
specific heat in Uð1Þ gauge theory. This predicted T 2=3 systems (Na4 Ir3 O8 ). We noted several outstanding difficulties
behavior has never been observed in experiments. Instead, with attempts to explain experimental results using existing
linear, Fermi-liquid-like specific heat is found to exist in a wide theories. These difficulties indicated that the field of QSLs is
range of temperatures in both organic materials κ-ET and dmit. still wide open and immature and that important physics may
Some theories exist that try to explain this missing singular still be missing in our present understanding of QSLs.
T 2=3 specific heat. For instance, Z4 and Z2 spin liquid states While keeping the discussion at an introductory level, we
with a spinon Fermi surface were proposed (Barkeshli, Yao, were not able to cover many important developments in the
and Kivelson, 2013) as well as Z2 spin liquid states with study of spin liquid states, and many technical details were
quadratic touched spinon bands (Mishmash et al., 2013). neglected, both theoretically and experimentally. For example,
However, all these proposals require fine-tuned parameters. the important techniques of renormalization groups and
A more natural way of explaining existing experiments is still conformal field theory were not addressed. We also did not
missing. discuss in detail the many developments related to MPSs and/
(2) Thermal Hall effect: Katsura, Nagaosa, and Lee (2010) or PEPSs and the corresponding numerical DMRG technique,
theoretically investigated the thermal Hall effect induced by an the understanding of spin systems with broken rotational
external magnetic field in a Uð1Þ spin liquid with a spinon symmetry following the discovery of the Kitaev state, and the
Fermi surface and predicted measurable electronic contribu- spin liquid physics of S > 1=2 systems. The role of topology
tions. Their predicted sizable thermal Hall effect has never in spin liquid states was not touched upon except as it is
been observed in an experiment on dmit compounds relevant to examples of spin liquid states. These are rapidly
(Yamashita et al., 2010). This contradiction between experi- evolving areas in which new discoveries are expected.
ment and theory remains unsolved, although an explanation In the following, we outline a few other topics that are
that depends on fine-tuned parameters was proposed neglected but have either played important historical roles in
(Mishmash et al., 2013). the development of the field of QSLs or shed light on future
(3) Power-law ac conductivity: A power-law ac conduc- research:
tivity inside the Mott gap was proposed by Ng and Lee (2007). Quantum dimer models: Quantum dimer models (QDMs)
Indeed, power-law behavior σðωÞ ∼ ωα has been observed in are a class of models defined in the Hilbert space of nearest
both κ-ET (Elsässer et al., 2012) and herbertsmithite (Pilon neighbor valence-bond (or dimer) coverings over a lattice
et al., 2013). However, the power α observed in both instead of the spin Hilbert space (Rokhsar and Kivelson,
compounds is smaller than the predicted value, indicating 1988). QDMs can be obtained in certain large-N limits of
that there exist more in-gap electronic excitations than those SUðNÞ or SpðNÞ antiferromagnets (Read and Sachdev, 1989)
predicted in the Uð1Þ gauge theory. and provide a simplified description of RVB states. This
Thus, despite all the theoretical efforts, the understanding simplification allows researchers to proceed further in ana-
and finding of realistic “smoking-gun” evidence for QSLs lytical treatments because of the close relations that arise with
remains the greatest challenge in the study of QSLs. classical dimer problems, Ising models, and Z2 gauge theory

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-44


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

(Fisher, 1961; Kasteleyn, 1961, 1963; Moessner, Sondhi, and To conclude, the field of QSLs is still wide open, both
Fradkin, 2001; Misguich, Serban, and Pasquier, 2002; theoretically and experimentally. The major difficulty in
Moessner and Sondhi, 2003). However, by construction, understanding QSLs is that they are intrinsically strongly
QMDs focus on the dynamics in the spin-singlet subspace correlated systems, for which no perturbative approach is
and ignore spin-triplet excitations. Therefore, they are not available. Theorists have been using all of the available tools
directly relevant to spin systems in which the magnetic as well as inventing new theoretical tools to understand QSLs
excitations are gapless. with the hope that novel emerging phenomena not covered by
An advantage of QDMs is that some QDMs are exactly perturbative approaches can be uncovered. Thus far, there
solvable (Misguich, Serban, and Pasquier, 2002; Yao and have been a few successes, and new experimental discoveries
Kivelson, 2012). Thus, many issues related to QSLs that are and theoretical ideas are rapidly emerging. However, a basic
difficult to address, such as spinon deconfinement, Z2 mathematical framework that can be used to understand QSLs
vortices, and topological order, can be addressed explicitly systematically is still lacking. We expect that more new
in QDMs. Interestingly, some spin-1=2 Hamiltonians give rise physics will be discovered in QSLs, posing a challenge to
to sRVB ground states defined in the dimer Hilbert space both theorists and experimentalists to construct a basic
when the relationship between the spin and dimer configu- framework for the understanding of QSLs.
rations is properly chosen (Fujimoto, 2005; Seidel, 2009;
Cano and Fendley, 2010). Readers who are interested in ACKNOWLEDGMENTS
further details on QDMs can refer to Diep (2004), Chapter. 5.5
and Lacroix, Mendels, and Mila (2011), Chapter 17. Y. Z. and T. K. N. thank Patrick A. Lee, Zheng-Xin Liu,
Chiral spin liquids: QSL states that break the parity (P) and Naoto Nagaosa, Shaojin Qin, Zhaobin Su, Hong-Hao Tu, Xiao-
time-reversal (T) symmetries while conserving the spin rota- Gang Wen, Zheng-Yu Weng, Tao Xiang, Fu-Chun Zhang, and
tional symmetry were proposed by Kalmeyer and Laughlin Guang-Ming Zhang for their close collaboration on related
(1987, 1989). These states are called chiral spin liquids. issues over the years. K. K. is grateful to T. Furukawa, H.
Kalmeyer and Laughlin suggested that some frustrated Hashiba, H. Kasahara, H. Kobashi, Y. Kurosaki, M. Maesato,
Heisenberg antiferromagnets in 2D can be described by K. Miyagawa, M. Poirier, F. Pratt, G. Saito, and Y. Shimizu, for
bosonic fractional quantum Hall wave functions. Soon after- their collaboration on the topic of spin liquids. We benefited
ward, Wen, Wilczek, and Zee (1989) introduced a generic greatly from discussions with our colleagues Yan Chen,
method of describing chiral spin liquids. They suggested that Yin-Chen He, Bruce Normand, Fa Wang, Cenke Xu, and
chiral spin states can be characterized in terms of the spin Hong Yao. Y. Z. is supported by the National Key Research and
chirality E123 ¼ S~ 1 · ðS~ 2 × S~ 3 Þ, defined for three different spins Development Program of China (No. 2016YFA0300202), the
S~ 1 , S~ 2 , and S~ 3 . The expectation value of the spin chirality in National Basic Research Program of China under Grant
fermionic RVB theory is given by hE123 i ¼ 12 Imhχ 12 χ 23 χ 31 i, No. 2014CB921201, and the National Natural Science
where χ ij are the short-range order parameters defined Foundation of China under Grant No. 11374256. He also
in Eq. (39). acknowledges the hospitality of the Max Planck Institute for the
Exactly solvable Hamiltonians hosting both gapful chiral Physics of Complex Systems in Dresden, where this review
spin liquid states (Laughlin, 1989; Schroeter et al., 2007; Yao article was finalized. K. K. is partially supported by the JSPS
and Kivelson, 2007; Thomale et al., 2009) and gapless chiral KAKENHI under Grants No. 20110002, No. 25220709, and
spin liquids (Chua, Yao, and Fiete, 2011) have been found. No. 24654101, the National Science Foundation under Grant
There is also numerical evidence for chiral spin liquids on No. PHYS-1066293, and by the hospitality of the Aspen Center
some 2D frustrated lattices (Sorella et al., 2003; Bauer et al., for Physics. T. K. N. acknowledges support from Hong Kong
2013; Nielsen, Sierra, and Cirac, 2013; Gong, Zhu, and Research Grants Council through Grant No. 603913.
Sheng, 2014; He and Chen, 2014; He, Sheng, and Chen,
2014; Gong et al., 2015; Zhu, Gong, and Sheng, 2015). It was APPENDIX: PATH INTEGRAL FOR A SINGLE SPIN
suggested that the statistics of spinons in these chiral spin
liquid states can be non-Abelian; see, e.g., Yao and Kivelson We consider the path integral for a single spin S in a
(2007) and Greiter and Thomale (2009). magnetic field B (H ¼ S · B) in the coherent state represen-
Characterizing spin liquid states numerically: Because of tation. Spin coherent states are defined as
rapid advancements in the power of numerical approaches to
spin models, the characterization of spin liquid states for Ŝjni ¼ Snjni;
specific spin models from numerical data has become a
rapidly evolving field. In addition to the MPS and/or PEPS
where Ŝ is the spin operator. The path integral can be derived
approach and the corresponding numerical DMRG technique,
by using the identity operator
Tang and Sandvik (2013) developed a quantum Monte Carlo
method of characterizing spinon size and confinement length  Z Z
2S þ 1 3 2
in quantum spin systems, which allows the spinon confine- I¼ d nδðn − 1Þjnihnj ¼ Dnjnihnj
ment-deconfinement issue to be studied numerically. Another 4π
important achievement is the use of entanglement entropy to ðA1aÞ
characterize QSL states. Interested readers can consult Grover,
Zhang, and Vishwanath (2013) for a brief review. and the corresponding inner product

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-45


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

 S
1 þ n1 · n2 The classical action of the system in real time is given by
hn1 jn2 i ¼ eiSΦðn1 ;n2 ;n0 Þ ; ðA1bÞ
2 Z T
Scl ¼ SΩ(nðtÞ) − S dtB · nðtÞ; ðA7aÞ
where n0 is a fixed unit vector and is usually chosen to be 0
n0 ¼ ẑ, Φðn1 ; n2 ; n0 Þ is the area of the spherical triangle with
vertices n1 , n2 , and n0 , and SΦ is the Berry’s phase acquired and the classical equation of motion δScl =δ0 nðtÞ ¼ 0 leads to
by a particle traveling through a loop formed by the edges of the Euler equation of motion
the spherical triangle.
The partition function Z ¼ e−βH can be written as a path n × ½ðn × ∂ t nÞ − B ¼ 0; ðA7bÞ
integral using the standard procedure:
where we used the result that a small variation δn leads to a
change in ΩðC½nÞ that is given by
Z¼ lim ðe−δtH ÞN t
N t →∞;δt→0 Z
 Z  β
δΩ(nðtÞ) ¼ dtδnðtÞ · ½nðtÞ × ∂ t nðtÞ:
¼ lim Πj¼1 Dnj ðΠNj¼1
Nt t
hnj je−iδtH jnjþ1 iÞ; ðA2Þ 0
N t →∞;δt→0

with the periodic boundary condition jnð0Þi ¼ jnðβÞi. REFERENCES


In the limit δt → 0, we may approximate
Abdel-Jawad, M., I. Terasaki, T. Sasaki, N. Yoneyama, N. Kobayashi,
Y. Uesu, and C. Hotta, 2010, Phys. Rev. B 82, 125119.
hnj je−iδtH jnjþ1 i Abrikosov, A. A., 1965, Physics 2, 5 [http://www.itp.ac.ru/en/
∼ hnj jnjþ1 i − δthnj jHjnjþ1 i persons/abrikosov‑alexey‑alekseevich/].
  Affleck, I., 1986, Nucl. Phys. B 265, 409.
hnj jHjnjþ1 i Affleck, I., 1990, Field Theory Methods and Quantum Critical
∼ hnj jnjþ1 i 1 − δt
hnj jnjþ1 i Phenomena, edited by E. Brezin and J. Zinn-Justin (North-Holland,
  Amsterdam).
1 þ nj · njþ1 S
∼ eiSΦðnj ;njþ1 ;n0 Þ ð1 − δtSB · nt Þ; ðA3Þ Affleck, I., T. Kennedy, E. Lieb, and H. Tasaki, 1988, Commun.
2 Math. Phys. 115, 477.
Affleck, I., T. Kennedy, E. H. Lieb, and H. Tasaki, 1987, Phys. Rev.
which is valid to first order in δt. In deriving the last equality Lett. 59, 799.
in Eq. (A3), we made use of the result hnjŜ ¼ hnjn. Affleck, I., and J. B. Marston, 1988, Phys. Rev. B 37, 3774.
Furthermore, we note that Affleck, I., Z. Zou, T. Hsu, and P. W. Anderson, 1988, Phys. Rev. B
38, 745.
 S Anderson, P., 1973, Mater. Res. Bull. 8, 153.
1 þ nj · njþ1 2
∼ eSδt∂ t ½nðtÞ ¼ eð0Þ
Sln½1þðδt=2ÞnðtÞ·∂ t nðtÞt¼tj
∼e Anderson, P. W., 1959, Phys. Rev. Lett. 3, 325.
2 Anderson, P. W., 1987, Science 235, 1196.
ðA4Þ Androes, G. M., and W. D. Knight, 1959, Phys. Rev. Lett. 2, 386.
Arovas, D. P., 2008, Phys. Rev. B 77, 104404.
Arovas, D. P., and A. Auerbach, 1988, Phys. Rev. B 38, 316.
to leading order in δt. Therefore,
Arrachea, L., L. Capriotti, and S. Sorella, 2004, Phys. Rev. B 69,
224414.
hnj je−iδtH jnjþ1 i ∼ eiSΦðnj ;njþ1 ;n0 Þ−δtSB·nt ðA5Þ Auerbach, A., 1994, Interacting Electrons and Quantum Mangetism
(Springer-Verlag, New York).
and Balents, L., 2010, Nature (London) 464, 199.
Balents, L., M. P. A. Fisher, and C. Nayak, 1998, Int. J. Mod. Phys. B
Z Rβ 12, 1033.
Z∼ DnðtÞeiSΩ(nðtÞ)−S 0
dtB·nðtÞ
; ðA6Þ Barkeshli, M., H. Yao, and S. A. Kivelson, 2013, Phys. Rev. B 87,
140402.
Baskaran, G., and P. W. Anderson, 1988, Phys. Rev. B 37, 580.
where Baskaran, G., S. Mandal, and R. Shankar, 2007, Phys. Rev. Lett. 98,
Z  Z  247201.
Baskaran, G., G. Santhosh, and R. Shankar, 2009, arXiv:0908.1614.
DnðtÞ ¼ lim ΠNj¼1
t
Dnj
N t →∞;δt→0 Baskaran, G., Z. Zou, and P. Anderson, 1987, Solid State Commun.
63, 973.
and Bauer, B., B. P. Keller, M. Dolfi, S. Trebst, and A. W. W. Ludwig,
2013, arXiv:1303.6963.
X Baym, G., and C. Pethick, 2004, Landau Fermi-Liquid Theory:
Ω(nðtÞ) ¼ Φðnj ; njþ1 ; n0 Þ Concepts and Applications (Wiley-VCH Verlag GmbH & Co.
j KGaA, Weinheim).
Berry, M. V., 1984, Proc. R. Soc. A 392, 45.
is the total area on the surface of the unit sphere covered by the Bert, F., S. Nakamae, F. Ladieu, D. L’Hôte, P. Bonville, F. Duc, J.-C.
(closed) path swept out by the spin nðtÞ from t ¼ 0 to t ¼ β. Trombe, and P. Mendels, 2007, Phys. Rev. B 76, 132411.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-46


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Bieri, S., C. Lhuillier, and L. Messio, 2016, Phys. Rev. B 93, Farnell, D. J. J., R. F. Bishop, and K. A. Gernoth, 2001, Phys. Rev. B
094437. 63, 220402.
Bieri, S., M. Serbyn, T. Senthil, and P. A. Lee, 2012, Phys. Rev. B 86, Fazekas, P., and P. Anderson, 1974, Philos. Mag. 30, 423.
224409. Feng, X. Y., G. M. Zhang, and T. Xiang, 2007, Phys. Rev. Lett. 98,
Brézin, E., and J. Zinn-Justin, 1976, Phys. Rev. B 14, 3110. 087204.
Brinkman, W. F., and T. M. Rice, 1970, Phys. Rev. B 2, 4302. Fisher, M. E., 1961, Phys. Rev. 124, 1664.
Cano, J., and P. Fendley, 2010, Phys. Rev. Lett. 105, 067205. Florens, S., and A. Georges, 2004, Phys. Rev. B 70, 035114.
Capriotti, L., F. Becca, A. Parola, and S. Sorella, 2001, Phys. Rev. Fradkin, E., and S. H. Shenker, 1979, Phys. Rev. D 19, 3682.
Lett. 87, 097201. Fradkin, E., and M. Stone, 1988, Phys. Rev. B 38, 7215.
Capriotti, L., A. E. Trumper, and S. Sorella, 1999, Phys. Rev. Lett. Frigeri, P. A., D. F. Agterberg, A. Koga, and M. Sigrist, 2004, Phys.
82, 3899. Rev. Lett. 92, 097001.
Ceperley, D., G. V. Chester, and M. H. Kalos, 1977, Phys. Rev. B 16, Fu, M., T. Imai, T.-H. Han, and Y. S. Lee, 2015, Science 350, 655.
3081. Fujimoto, S., 2005, Phys. Rev. B 72, 024429.
Chaloupka, J. c. v., G. Jackeli, and G. Khaliullin, 2010, Phys. Rev. Furukawa, T., K. Miyagawa, H. Taniguchi, R. Kato, and K. Kanoda,
Lett. 105, 027204. 2015a, Nat. Phys. 11, 221.
Chandra, P., and B. Doucot, 1988, Phys. Rev. B 38, 9335. Furukawa, T., K. Miyagawa, H. Taniguchi, R. Kato, and K. Kanoda,
Chen, C. Z., Q. F. Sun, F. Wang, and X. C. Xie, 2013, Phys. Rev. B 2015b, Phys. Rev. Lett. 115, 077001.
88, 041405. Galitski, V., and Y. B. Kim, 2007, Phys. Rev. Lett. 99, 266403.
Chen, G., and L. Balents, 2008, Phys. Rev. B 78, 094403. Gebhard, F., and D. Vollhardt, 1987, Phys. Rev. Lett. 59, 1472.
Chen, G., and Y. B. Kim, 2013, Phys. Rev. B 87, 165120. Giamarchi, T., 2003, Quantum Physics in One Dimension (Oxford
Chen, H. D., and Z. Nussinov, 2008, J. Phys. A 41, 075001. University Press, New York).
Chen, H. D., B. Wang, and S. Das Sarma, 2010, Phys. Rev. B 81, Glarum, S. H., S. Geschwind, K. M. Lee, M. L. Kaplan, and J.
235131. Michel, 1991, Phys. Rev. Lett. 67, 1614.
Chen, X., Z. C. Gu, Z. X. Liu, and X. G. Wen, 2012, Science 338, Gong, S.-S., W. Zhu, L. Balents, and D. N. Sheng, 2015, Phys. Rev.
1604. B 91, 075112.
Chen, X., Z. C. Gu, and X. G. Wen, 2011a, Phys. Rev. B 83, 035107. Gong, S.-S., W. Zhu, and D. Sheng, 2014, Sci. Rep. 4, 6317.
Chen, X., Z. C. Gu, and X. G. Wen, 2011b, Phys. Rev. B 84, 235128. Gong, S.-S., W. Zhu, D. N. Sheng, O. I. Motrunich, and M. P. A.
Cheng, J. G., G. Li, L. Balicas, J. S. Zhou, J. B. Goodenough, C. Xu, Fisher, 2014, Phys. Rev. Lett. 113, 027201.
and H. D. Zhou, 2011, Phys. Rev. Lett. 107, 197204. Gor’kov, L. P., and E. I. Rashba, 2001, Phys. Rev. Lett. 87, 037004.
Choy, T. P., and Y. B. Kim, 2009, Phys. Rev. B 80, 064404. Greiter, M., and R. Thomale, 2009, Phys. Rev. Lett. 102, 207203.
Chua, V., H. Yao, and G. A. Fiete, 2011, Phys. Rev. B 83, 180412. Griffiths, R. B., 1964, Phys. Rev. 133, A768.
Cirac, J. I., and F. Verstraete, 2009, J. Phys. A 42, 504004. Gros, C., 1989, Ann. Phys. (N.Y.) 189, 53.
Coldea, R., D. A. Tennant, A. M. Tsvelik, and Z. Tylczynski, 2001, Gros, C., R. Joynt, and T. M. Rice, 1987, Phys. Rev. B 36, 381.
Phys. Rev. Lett. 86, 1335. Grover, T., N. Trivedi, T. Senthil, and P. A. Lee, 2010, Phys. Rev. B
Coldea, R., D. A. Tennant, and Z. Tylczynski, 2003, Phys. Rev. B 68, 81, 245121.
134424. Grover, T., Y. Zhang, and A. Vishwanath, 2013, New J. Phys. 15,
Dally, R., T. Hogan, A. Amato, H. Luetkens, C. Baines, J. Rodriguez- 025002.
Rivera, M. J. Graf, and S. D. Wilson, 2014, Phys. Rev. Lett. 113, Gu, Z. C., and X. G. Wen, 2009, Phys. Rev. B 80, 155131.
247601. Hahn, T., 1996, International tables for crystallography, Volume A,
Dayal, S., R. T. Clay, H. Li, and S. Mazumdar, 2011, Phys. Rev. B 83, Space-Group Symmetry (Kluwer Academic Publishers, Dordrecht,
245106. Netherlands), Vol. A, 4th ed.
Depenbrock, S., I. P. McCulloch, and U. Schollwöck, 2012, Phys. Haldane, F., 1985, J. Appl. Phys. 57, 3359.
Rev. Lett. 109, 067201. Haldane, F. D. M., 1983a, Phys. Lett. A 93, 464.
de Vries, M. A., K. V. Kamenev, W. A. Kockelmann, J. Sanchez- Haldane, F. D. M., 1983b, Phys. Rev. Lett. 50, 1153.
Benitez, and A. Harrison, 2008, Phys. Rev. Lett. 100, 157205. Haldane, F. D. M., 1988a, Phys. Rev. Lett. 60, 635.
de Vries, M. A., J. R. Stewart, P. P. Deen, J. O. Piatek, G. J. Nilsen, Haldane, F. D. M., 1988b, Phys. Rev. Lett. 61, 1029.
H. M. Rø nnow, and A. Harrison, 2009, Phys. Rev. Lett. 103, Han, T.-H., J. S. Helton, S. Chu, D. G. Nocera, J. A. Rodriguez-
237201. Rivera, C. Broholm, and Y. S. Lee, 2012, Nature (London) 492,
Diep, H. T., 2004, Frustrated spin systems (World Scientific, 406.
Singapore). Hastings, M. B., 2000, Phys. Rev. B 63, 014413.
Dodds, T., S. Bhattacharjee, and Y. B. Kim, 2013, Phys. Rev. B 88, Hastings, M. B., 2007, J. Stat. Mech. 08, P08024.
224413. He, Y. C., and Y. Chen, 2014, arXiv:1407.2740.
Dombre, T., and N. Read, 1988, Phys. Rev. B 38, 7181. He, Y. C., D. Sheng, and Y. Chen, 2014, Phys. Rev. Lett. 112,
Dressel, M., P. Lazić, A. Pustogow, E. Zhukova, B. Gorshunov, J. A. 137202.
Schlueter, O. Milat, B. Gumhalter, and S. Tomić, 2016, Phys. Rev. Helton, J. S., et al., 2007, Phys. Rev. Lett. 98, 107204.
B 93, 081201. Herbut, I. F., and B. H. Seradjeh, 2003, Phys. Rev. Lett. 91, 171601.
Dusuel, S., K. P. Schmidt, J. Vidal, and R. L. Zaffino, 2008, Phys. Herbut, I. F., B. H. Seradjeh, S. Sachdev, and G. Murthy, 2003, Phys.
Rev. B 78, 125102. Rev. B 68, 195110.
Eggert, S., and I. Affleck, 1992, Phys. Rev. B 46, 10866. Hermele, M., T. Senthil, M. P. A. Fisher, P. A. Lee, N. Nagaosa, and
Elsässer, S., D. Wu, M. Dressel, and J. A. Schlueter, 2012, Phys. Rev. X. G. Wen, 2004, Phys. Rev. B 70, 214437.
B 86, 155150. Hohenberg, P. C., 1967, Phys. Rev. 158, 383.
Fannes, M., B. Nachtergaele, and R. F. Werner, 1992, Commun. Horsch, P., and T. A. Kaplan, 1983, J. Phys. C 16, L1203.
Math. Phys. 144, 443. Hotta, C., 2010, Phys. Rev. B 82, 241104.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-47


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Hu, W.-J., F. Becca, A. Parola, and S. Sorella, 2013, Phys. Rev. B 88, Kitaev, A., 2003, Ann. Phys. (Amsterdam) 303, 2.
060402. Kitaev, A., 2006, Ann. Phys. (Amsterdam) 321, 2.
Huse, D. A., and V. Elser, 1988, Phys. Rev. Lett. 60, 2531. Klümper, A., A. Schadschneider, and J. Zittartz, 1993, Europhys.
Imai, T., E. A. Nytko, B. Bartlett, S. M. P., and D. G. Nocera, 2008, Lett. 24, 293.
Phys. Rev. Lett. 100, 077203. Ko, W. H., P. A. Lee, and X. G. Wen, 2009, Phys. Rev. B 79, 214502.
Iqbal, Y., F. Becca, S. Sorella, and D. Poilblanc, 2013, Phys. Rev. B Ko, W. H., Z. X. Liu, T. K. Ng, and P. A. Lee, 2010, Phys. Rev. B 81,
87, 060405. 024414.
Iqbal, Y., D. Poilblanc, and F. Becca, 2014, Phys. Rev. B 89, 020407. Komatsu, Y., N. Matsukawa, T. Inoue, and G. Saito, 1996, J. Phys.
Iqbal, Y., D. Poilblanc, and F. Becca, 2016, arXiv:1606.02255. Soc. Jpn. 65, 1340.
Isono, T., H. Kamo, A. Ueda, A. Takahashi, K. Nakao, R. Kumai, H. Koretsune, T., and C. Hotta, 2014, Phys. Rev. B 89, 045102.
Nakano, K. Kobayashi, Y. Murakami, and H. Mori, 2013, Nat. Koteswararao, B., et al., 2014, Phys. Rev. B 90, 035141.
Commun. 4, 1344. Kotliar, G., 1988, Phys. Rev. B 37, 3664.
Itoh, K., H. Itoh, M. Naka, S. Saito, I. Hosako, N. Yoneyama, S. Kou, S. P., and X. G. Wen, 2009, Phys. Rev. B 80, 224406.
Ishihara, T. Sasaki, and S. Iwai, 2013, Phys. Rev. Lett. 110, Kurosaki, Y., Y. shimizu, K. Miyagawa, and K. Kanoda, 2005, Phys.
106401. Rev. Lett. 95, 177001.
Itou, T., A. Oyamada, S. Maegawa, and R. Kato, 2010, Nat. Phys. 6, Kyung, B., and A.-M. S. Tremblay, 2006, Phys. Rev. Lett. 97,
673. 046402.
Itou, T., A. Oyamada, S. Maegawa, M. Tamura, and R. Kato, 2008, Lacroix, C., P. Mendels, and F. Mila, 2011, Introduction to
Phys. Rev. B 77, 104413. Frustrated Magnetism: Materials, Experiments, Theory, Vol. 164
Itou, T., K. Yamashita, M. Nishiyama, A. Oyamada, S. Maegawa, K. (Springer, New York).
Kubo, and R. Kato, 2011, Phys. Rev. B 84, 094405. Lai, H.-H., and O. I. Motrunich, 2011, Phys. Rev. B 84, 085141.
Jackeli, G., and G. Khaliullin, 2009, Phys. Rev. Lett. 102, 017205. Larkin, A. I., 1964, Sov. Phys. JETP 19, 1478 [http://jetp.ac.ru/
Jeong, M., F. Bert, P. Mendels, F. Duc, J. C. Trombe, M. A. de Vries, cgi‑bin/dn/e_019_06_1478.pdf].
and A. Harrison, 2011, Phys. Rev. Lett. 107, 237201. Laubach, M., R. Thomale, C. Platt, W. Hanke, and G. Li, 2015, Phys.
Jiang, H. C., Z. C. Gu, X. L. Qi, and S. Trebst, 2011, Phys. Rev. B 83, Rev. B 91, 245125.
245104. Laughlin, R. B., 1989, Ann. Phys. (N.Y.) 191, 163.
Jiang, H. C., Z. Wang, and L. Balents, 2012, Nat. Phys. 8, 902. Lawler, M. J., A. Paramekanti, Y. B. Kim, and L. Balents, 2008,
Jiang, H. C., Z. Y. Weng, and D. N. Sheng, 2008, Phys. Rev. Lett. Phys. Rev. Lett. 101, 197202.
101, 117203. Lecheminant, P., B. Bernu, C. Lhuillier, L. Pierre, and P. Sindzingre,
Jiang, H.-C., H. Yao, and L. Balents, 2012, Phys. Rev. B 86, 024424. 1997, Phys. Rev. B 56, 2521.
Kagawa, F., K. Miyagawa, and K. Kanoda, 2005, Nature (London) Lee, D. H., G. M. Zhang, and T. Xiang, 2007, Phys. Rev. Lett. 99,
436, 534. 196805.
Kalmeyer, V., and R. B. Laughlin, 1987, Phys. Rev. Lett. 59, 2095. Lee, E. K. H., R. Schaffer, S. Bhattacharjee, and Y. B. Kim, 2014,
Kalmeyer, V., and R. B. Laughlin, 1989, Phys. Rev. B 39, 11879. Phys. Rev. B 89, 045117.
Kandpal, H. C., I. Opahle, Y.-Z. Zhang, H. O. Jeschke, and R. Lee, P. A., 2008, Science 321, 1306.
Valenti, 2009, Phys. Rev. Lett. 103, 067004. Lee, P. A., and N. Nagaosa, 1992, Phys. Rev. B 46, 5621.
Kanoda, K., 1997a, Physica C (Amsterdam) 282–287, 299. Lee, P. A., and N. Nagaosa, 2013, Phys. Rev. B 87, 064423.
Kanoda, K., 1997b, Hyperfine Interact. 104, 235. Lee, P. A., N. Nagaosa, and X.-G. Wen, 2006, Rev. Mod. Phys. 78,
Kanoda, K., 2006, J. Phys. Soc. Jpn. 75, 051007. 17.
Kanoda, K., and R. Kato, 2011, Annu. Rev. Condens. Matter Phys. 2, Lee, S. S., and P. A. Lee, 2005, Phys. Rev. Lett. 95, 036403.
167. Lee, S. S., P. A. Lee, and T. Senthil, 2007, Phys. Rev. Lett. 98,
Kasteleyn, P., 1961, Physica (Utrecht) 27, 1209. 067006.
Kasteleyn, P., 1963, J. Math. Phys. (N.Y.) 4, 287. Lee, S.-S., 2008, Phys. Rev. B 78, 085129.
Kato, R., 2014, Bull. Chem. Soc. Jpn. 87, 355. Lefebvre, S., P. Wzietek, S. Brown, C. Bourbonnais, D. Jerome, C.
Kato, R., A. Tajima, A. Nakao, A. Tajima, and M. Tamura, 2007, Meziere, M. Fourmigue, and P. Batail, 2000, Phys. Rev. Lett. 85,
Multifunctional Conducting Molecular Materials (RSC, Cambridge), 5420.
p. 32. Leggett, A. J., 1965, Phys. Rev. 140, A1869.
Katsura, H., N. Nagaosa, and P. A. Lee, 2010, Phys. Rev. Lett. 104, Leggett, A. J., 1975, Rev. Mod. Phys. 47, 331.
066403. Leung, P. W., and V. Elser, 1993, Phys. Rev. B 47, 5459.
Kawamoto, A., Y. Honma, K. Kumagai, N. Matsunaga, and K. Li, T., 2016, arXiv:1601.0216.
Nomura, 2006, Phys. Rev. B 74, 212508. Li, T., and H.-Y. Yang, 2007, Phys. Rev. B 75, 172502.
Kawamura, H., K. Watanabe, and T. Shimokawa, 2014, J. Phys. Soc. Liang, S., B. Doucot, and P. W. Anderson, 1988, Phys. Rev. Lett. 61,
Jpn. 83, 103704. 365.
Kells, G., J. K. Slingerland, and J. Vala, 2009, Phys. Rev. B 80, Lieb, E. H., 1994, Phys. Rev. Lett. 73, 2158.
125415. Lieb, E. H., and F. Y. Wu, 1968, Phys. Rev. Lett. 20, 1445.
Kézsmárki, I., Y. Shimizu, G. Mihály, Y. Tokura, K. Kanoda, and G. Liu, Z. X., Y. Zhou, and T. K. Ng, 2010a, Phys. Rev. B 82, 144422.
Saito, 2006, Phys. Rev. B 74, 201101. Liu, Z. X., Y. Zhou, and T. K. Ng, 2010b, Phys. Rev. B 81, 224417.
Khuntia, P., F. Bert, P. Mendels, B. Koteswararao, A. V. Mahajan, M. Liu, Z. X., Y. Zhou, and T. K. Ng, 2014, New J. Phys. 16, 083031.
Baenitz, F. C. Chou, C. Baines, A. Amato, and Y. Furukawa, 2016, Liu, Z. X., Y. Zhou, H. H. Tu, X. G. Wen, and T. K. Ng, 2012, Phys.
Phys. Rev. Lett. 116, 107203. Rev. B 85, 195144.
Kimchi, I., and A. Vishwanath, 2014, Phys. Rev. B 89, 014414. Lu, Y. M., Y. Ran, and P. A. Lee, 2011, Phys. Rev. B 83, 224413.
Kimchi, I., and Y. Z. You, 2011, Phys. Rev. B 84, 180407. Luther, A., and I. Peschel, 1975, Phys. Rev. B 12, 3908.
Kino, H., and H. Fukuyama, 1995, J. Phys. Soc. Jpn. 64, 2726. Ma, M., 1988, Phys. Rev. B 38, 6813.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-48


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Ma, Y.-F., and T.-K. Ng, 2015, Phys. Rev. B 91, 075106. Okamoto, Y., M. Nohara, H. Aruga-Katori, and H. Takagi, 2007,
MacLaughlin, D. E., Y. Nambu, S. Nakatsuji, R. H. Heffner, L. Shu, Phys. Rev. Lett. 99, 137207.
O. O. Bernal, and K. Ishida, 2008, Phys. Rev. B 78, 220403. Olariu, A., P. Mendels, F. Bert, F. Duc, J. C. Trombe, M. A. de Vries,
Mandal, S., S. Bhattacharjee, K. Sengupta, R. Shankar, and G. and A. Harrison, 2008, Phys. Rev. Lett. 100, 087202.
Baskaran, 2011, Phys. Rev. B 84, 155121. Orus, R., 2014, Ann. Phys. (Amsterdam) 349, 117.
Mandal, S., R. Shankar, and G. Baskaran, 2012, J. Phys. A 45, Östlund, S., and S. Rommer, 1995, Phys. Rev. Lett. 75, 3537.
335304. Pauling, L., 1949, Proc. R. Soc. A 196, 343.
Manna, R. S., M. de Souza, A. Brühl, J. A. Schlueter, and M. Lang, Pilon, D. V., C. H. Lui, T. H. Han, D. Shrekenhamer, A. J. Frenzel,
2010, Phys. Rev. Lett. 104, 016403. W. J. Padilla, Y. S. Lee, and N. Gedik, 2013, Phys. Rev. Lett. 111,
Manousakis, E., 1991, Rev. Mod. Phys. 63, 1. 127401.
Marshall, W., 1955, Proc. R. Soc. A 232, 48. Pinterić, M., et al., 2014, Phys. Rev. B 90, 195139.
Marston, J., and C. Zeng, 1991, J. Appl. Phys. 69, 5962. Podolsky, D., A. Paramekanti, Y. B. Kim, and T. Senthil, 2009, Phys.
Mendels, P., F. Bert, M. A. de Vries, A. Olariu, A. Harrison, F. Duc, Rev. Lett. 102, 186401.
J. C. Trombe, J. S. Lord, and A. a. C. B. Amato, 2007, Phys. Rev. Poilblanc, D., and N. Schuch, 2013, Phys. Rev. B 87, 140407.
Lett. 98, 077204. Poirier, M., M. de Lafontaine, K. Miyagawa, K. Kanoda, and Y.
Mermin, N. D., and H. Wagner, 1966, Phys. Rev. Lett. 17, 1133. Shimizu, 2014, Phys. Rev. B 89, 045138.
Mila, F., 1998, Phys. Rev. Lett. 81, 2356. Poirier, M., S. Parent, A. Cote, K. Miyagawa, K. Kanoda, and Y.
Misguich, G., B. Bernu, C. Lhuillier, and C. Waldtmann, 1998, Phys. Shimizu, 2012, Phys. Rev. B 85, 134444.
Rev. Lett. 81, 1098. Polchinski, J., 1994, Nucl. Phys. B 422, 617.
Misguich, G., and C. Lhuillier, 2004, Frustrated spin systems, edited Pollmann, F., E. Berg, A. M. Turner, and M. Oshikawa, 2012, Phys.
by T. H. Diep (World Scientific, Singapore). Rev. B 85, 075125.
Misguich, G., D. Serban, and V. Pasquier, 2002, Phys. Rev. Lett. 89, Polyakov, A., 1977, Nucl. Phys. B 120, 429.
137202. Polyakov, A., 1987, Gauge Fields and Strings (Contemporary Con-
Mishmash, R. V., J. R. Garrison, S. Bieri, and C. Xu, 2013, Phys. cepts in Physics) (Harwood Academic Publishers, Switzerland).
Rev. Lett. 111, 157203. Polyakov, A. M., 1975, Phys. Lett. B 59, 79.
Miyagawa, K., K. Kanoda, and A. Kawamoto, 2004, Chem. Rev. Potter, A. C., T. Senthil, and P. A. Lee, 2013, Phys. Rev. B 87,
104, 5635. 245106.
Miyagawa, K., A. Kawamoto, Y. Nakazawa, and K. Kanoda, 1995, Powell, B. J., and R. H. McKenzie, 2011, Rep. Prog. Phys. 74,
Phys. Rev. Lett. 75, 1174. 056501.
Moessner, R., and J. T. Chalker, 1998, Phys. Rev. Lett. 80, 2929. Pratt, F. L., P. J. Baker, S. J. Blundell, T. Lancaster, S. Ohira-
Moessner, R., and S. L. Sondhi, 2003, Phys. Rev. B 68, 054405. Kawamura, C. Baines, Y. Shimizu, K. Kanoda, I. Watanabe, and
Moessner, R., S. L. Sondhi, and E. Fradkin, 2001, Phys. Rev. B 65, G. Saito, 2011, Nature (London) 471, 612.
024504. Price, C. C., and N. B. Perkins, 2012, Phys. Rev. Lett. 109, 187201.
Mori, T., A. Kobayashi, Y. Sasaki, H. Kobayashi, G. Saito, and H. Qi, Y., C. Xu, and S. Sachdev, 2009, Phys. Rev. Lett. 102, 176401.
Inokuchi, 1984, Bull. Chem. Soc. Jpn. 57, 627. Qin, S., T. K. Ng, and Z. B. Su, 1995, Phys. Rev. B 52, 12844.
Mori, T., H. Mori, and S. Tanaka, 1999, Bull. Chem. Soc. Jpn. 72, Ran, Y., M. Hermele, P. A. Lee, and X. G. Wen, 2007, Phys. Rev.
179. Lett. 98, 117205.
Morita, H., S. Watanabe, and M. Imada, 2002, J. Phys. Soc. Jpn. 71, Read, N., and B. Chakraborty, 1989, Phys. Rev. B 40, 7133.
2109. Read, N., and S. Sachdev, 1989, Nucl. Phys. B 316, 609.
Motrunich, O. I., 2005, Phys. Rev. B 72, 045105. Read, N., and S. Sachdev, 1990, Phys. Rev. B 42, 4568.
Mudry, C., and E. Fradkin, 1994a, Phys. Rev. B 50, 11409. Reuther, J., R. Thomale, and S. Trebst, 2011, Phys. Rev. B 84,
Mudry, C., and E. Fradkin, 1994b, Phys. Rev. B 49, 5200. 100406.
Naka, M., and S. Ishihara, 2010, J. Phys. Soc. Jpn. 79, 063707. Ribeiro, P., and P. A. Lee, 2011, Phys. Rev. B 83, 235119.
Nakajima, S., et al., 2012, J. Phys. Soc. Jpn. 81, 063706. Rigol, M., and R. R. P. Singh, 2007, Phys. Rev. Lett. 98,
Nakamura, K., Y. Yoshimoto, T. Kusugi, R. Arita, and I. M., 2009, J. 207204.
Phys. Soc. Jpn. 78, 083710. Rokhsar, D. S., and S. A. Kivelson, 1988, Phys. Rev. Lett. 61, 2376.
Nakatsuji, S., Y. Nambu, H. Tonomura, O. Sakai, S. Jonas, C. Ryu, S., 2009, Phys. Rev. B 79, 075124.
Broholm, H. Tsunetsugu, Y. Qiu, and Y. Maeno, 2005, Science Sachidev, S., 1992, Phys. Rev. B 45, 12377.
309, 1697. Saito, G., 2014, unpublished.
Nakatsuji, S., et al., 2012, Science 336, 559. Schaffer, R., S. Bhattacharjee, and Y. B. Kim, 2012, Phys. Rev. B 86,
Nave, C. P., and P. A. Lee, 2007, Phys. Rev. B 76, 235124. 224417.
Nayak, C., S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, Schmidt, K. P., S. Dusuel, and J. Vidal, 2008, Phys. Rev. Lett. 100,
2008, Rev. Mod. Phys. 80, 1083. 057208.
Ng, T. K., 1994, Phys. Rev. B 50, 555. Schollwöck, U., 2005, Rev. Mod. Phys. 77, 259.
Ng, T. K., 1999, Phys. Rev. Lett. 82, 3504. Schroeter, D. F., E. Kapit, R. Thomale, and M. Greiter, 2007, Phys.
Ng, T. K., and P. A. Lee, 2007, Phys. Rev. Lett. 99, 156402. Rev. Lett. 99, 097202.
Nielsen, A. E., G. Sierra, and J. I. Cirac, 2013, Nat. Commun. 4, Schuch, N., D. Poilblanc, J. I. Cirac, and D. Pérez-García, 2012,
2864. Phys. Rev. B 86, 115108.
Nogueira, F. S., and H. Kleinert, 2005, Phys. Rev. Lett. 95, Seidel, A., 2009, Phys. Rev. B 80, 165131.
176406. Senthil, T., 2008, Phys. Rev. B 78, 045109.
Norman, M. R., and T. Micklitz, 2009, Phys. Rev. Lett. 102, 067204. Senthil, T., and M. P. A. Fisher, 2000, Phys. Rev. B 62, 7850.
Nussinov, Z., and G. Ortiz, 2009, Phys. Rev. B 79, 214440. Shaginyan, V. R., A. Z. Msezane, and K. G. Popov, 2011, Phys. Rev.
Nussinov, Z., and Z. van den Brink, 2013, arXiv:1303.5922. B 84, 060401.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-49


Yi Zhou, Kazushi Kanoda, and Tai-Kai Ng: Quantum spin liquid states

Shankar, R., and N. Read, 1990, Nucl. Phys. B 336, 457. Wang, L., D. Poilblanc, Z. C. Gu, X. G. Wen, and F. Verstraete, 2013,
Shastry, B. S., 1988, Phys. Rev. Lett. 60, 639. Phys. Rev. Lett. 111, 037202.
Shimizu, Y., K. Miyagawa, K. Kanoda, M. Maesato, and G. Saito, Wannier, G. H., 1950, Phys. Rev. 79, 357.
2003, Phys. Rev. Lett. 91, 107001. Watanabe, D., et al., 2012, Nat. Commun. 3, 1090.
Shimizu, Y., K. Miyagawa, K. Kanoda, M. Maesato, and G. Saito, Watanabe, K., H. Kawamura, H. Nakano, and T. Sakai, 2014, J. Phys.
2006, Phys. Rev. B 73, 140407. Soc. Jpn. 83, 034714.
Shimokawa, T., K. Watanabe, and H. Kawamura, 2015, Phys. Rev. B Watanabe, T., H. Yokoyama, Y. Tanaka, and J. Inoue, 2008, Phys.
92, 134407. Rev. B 77, 214505.
Shockley, A. C., F. Bert, J.-C. Orain, Y. Okamoto, and P. Mendels, Wen, X. G., 1989, Phys. Rev. B 39, 7223.
2015, Phys. Rev. Lett. 115, 047201. Wen, X. G., 1991, Phys. Rev. B 44, 2664.
Shores, M. P., E. A. Nytko, B. M. Bartlett, and D. G. Nocera, 2005, J. Wen, X. G., 2002, Phys. Rev. B 65, 165113.
Am. Chem. Soc. 127, 13462. Wen, X. G., F. Wilczek, and A. Zee, 1989, Phys. Rev. B 39, 11413.
Sigrist, M., and K. Ueda, 1991, Rev. Mod. Phys. 63, 239. Wen, X. G., and A. Zee, 1988, Phys. Rev. Lett. 61, 1025.
Sindzingre, P., P. Lecheminant, and C. Lhuillier, 1994, Phys. Rev. B White, S. R., 1992, Phys. Rev. Lett. 69, 2863.
50, 3108. White, S. R., and A. L. Chernyshev, 2007, Phys. Rev. Lett. 99,
Singh, R. R. P., 2010, Phys. Rev. Lett. 104, 177203. 127004.
Singh, Y., S. Manni, J. Reuther, T. Berlijn, R. Thomale, W. Ku, Wu, C., D. Arovas, and H.-H. Hung, 2009, Phys. Rev. B 79, 134427.
S. Trebst, and P. Gegenwart, 2012, Phys. Rev. Lett. 108, Wu, T. T., and C. N. Yang, 1976, Nucl. Phys. B 107, 365.
127203. Xie, Z., J. Chen, J. F. Yu, X. Kong, B. Normand, and T. Xiang, 2014,
Singh, Y., Y. Tokiwa, J. Dong, and P. Gegenwart, 2013, Phys. Rev. B Phys. Rev. X 4, 011025.
88, 220413. Xu, C., F. Wang, Y. Qi, L. Balents, and M. P. A. Fisher, 2012, Phys.
Sorella, S., L. Capriotti, F. Becca, and A. Parola, 2003, Phys. Rev. Rev. Lett. 108, 087204.
Lett. 91, 257005. Yamashita, M., N. Nakata, Y. Kasahara, T. Sasaki, N. Yoneyama, N.
Stephenson, J., 1970, J. Math. Phys. (N.Y.) 11, 420. Kobayashi, S. Fujimoto, T. Shibauchi, and Y. Matsuda, 2009, Nat.
Sze, W. P., Y. Zhou, and T.-K. Ng, 2016, unpublished. Phys. 5, 44.
Tamura, M., and R. Kato, 2002, J. Phys. Condens. Matter 14, L729. Yamashita, M., N. Nakata, Y. Senshu, M. Nagata, H. M. Yamamoto,
Tang, E., M. P. A. Fisher, and P. A. Lee, 2013, Phys. Rev. B 87, R. Kato, T. Shibauchi, and Y. Matsuda, 2010, Science 328, 1246.
045119. Yamashita, S., Y. Nakazawa, M. Oguni, Y. Oshima, H. Nojiri, Y.
Tang, Y., and A. W. Sandvik, 2013, Phys. Rev. Lett. 110, 217213. Shimizu, K. Miyagawa, and K. Kanoda, 2008, Nat. Phys. 4, 459.
Thomale, R., E. Kapit, D. F. Schroeter, and M. Greiter, 2009, Phys. Yamashita, S., T. Yamamoto, Y. Nakazawa, M. Tamura, and R. Kato,
Rev. B 80, 104406. 2011, Nat. Commun. 2, 275.
Tikhonov, K. S., and M. V. Feigel’man, 2010, Phys. Rev. Lett. 105, Yan, S., D. Huse, and S. White, 2011, Science 332, 1173.
067207. Yang, H. Y., A. M. Läuchli, F. Mila, and K. P. Schmidt, 2010, Phys.
Tocchio, L. F., H. Feldner, F. Becca, V. R., and C. Gros, 2013, Phys. Rev. Lett. 105, 267204.
Rev. B 87, 035143. Yang, S., D. L. Zhou, and C. P. Sun, 2007, Phys. Rev. B 76, 180404.
Toulouse, G., 1977, Commun. Phys. 2, 115. Yao, H., and S. A. Kivelson, 2007, Phys. Rev. Lett. 99, 247203.
Tsumuraya, T., H. Seo, H. Tsuchiizu, R. Kato, and T. Miyazaki, Yao, H., and S. A. Kivelson, 2012, Phys. Rev. Lett. 108, 247206.
2013, J. Phys. Soc. Jpn. 82, 033709. Yao, H., and D. H. Lee, 2011, Phys. Rev. Lett. 107, 087205.
Vannimenus, J., and G. Toulouse, 1977, J. Phys. C 10, L537. Yao, H., S.-C. Zhang, and S. A. Kivelson, 2009, Phys. Rev. Lett. 102,
Verstraete, F., and J. I. Cirac, 2004a, arXiv:cond-mat/0407066. 217202.
Verstraete, F., and J. I. Cirac, 2004b, Phys. Rev. A 70, 060302.. Yokoyama, H., and H. Shiba, 1987, J. Phys. Soc. Jpn. 56, 3570.
Verstraete, F., and J. I. Cirac, 2006, Phys. Rev. B 73, 094423. Yu, Y., 2008, Nucl. Phys. B 799, 345.
Verstraete, F., V. Murg, and J. Cirac, 2008, Adv. Phys. 57, 143. Yu, Y., L. Liang, Q. Niu, and S. Qin, 2013, Phys. Rev. B 87, 041107.
Verstraete, F., M. M. Wolf, D. Perez-Garcia, and J. I. Cirac, 2006, Yu, Y., and Z. Wang, 2008, Europhys. Lett. 84, 57002.
Phys. Rev. Lett. 96, 220601. Yunoki, S., and S. Sorella, 2006, Phys. Rev. B 74, 014408.
Vidal, J., K. P. Schmidt, and S. Dusuel, 2008, Phys. Rev. B 78, Zhou, H. D., E. S. Choi, G. Li, L. Balicas, C. R. Wiebe, Y. Qiu, J. R. D.
245121. Copley, and J. S. Gardner, 2011, Phys. Rev. Lett. 106, 147204.
Villain, J., R. Bidaux, J.-P. Carton, and R. Conte, 1980, J. Phys. Zhou, Y., and P. A. Lee, 2011, Phys. Rev. Lett. 106, 056402.
France 41, 1263. Zhou, Y., P. A. Lee, T. K. Ng, and F. C. Zhang, 2008, Phys. Rev. Lett.
Waldtmann, C., H. U. Everts, B. Bernu, C. Lhuillier, P. Sindzingre, P. 101, 197201.
Lecheminant, and L. Pierre, 1998, Eur. Phys. J. B 2, 501. Zhou, Y., and T. K. Ng, 2013, Phys. Rev. B 88, 165130.
Wang, F., 2010a, Phys. Rev. B 81, 184416. Zhou, Y., and X.-G. Wen, 2002, arXiv:cond-mat/0210662.
Wang, F., 2010b, Phys. Rev. B 82, 024419. Zhu, W., S. S. Gong, and D. N. Sheng, 2015, Phys. Rev. B 92,
Wang, F., and A. Vishwanath, 2006, Phys. Rev. B 74, 174423. 014424.

Rev. Mod. Phys., Vol. 89, No. 2, April–June 2017 025003-50

You might also like