Yejin Huh

Quantum Phase Transitions
in d-wave Superconductors
and Antiferromagnetic Kagome Lattices
A dissertation presented
by
Yejin Huh
to
The Department of Physics
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in the subject of
Physics
Harvard University
Cambridge, Massachusetts
August 2011
2011
c - Yejin Huh
All rights reserved.

Thesis advisor Author
Subir Sachdev Yejin Huh
Quantum Phase Transitions in d-wave Superconductors

and Antiferromagnetic Kagome Lattices
Abstract
Strongly correlated systems are of interest due to their exotic collective behavior.
In this thesis we study low energy effective theory and quantum phase transitions of
d-wave superconductors and spin liquids.
First we examine the quantum theory of the spontaneous breaking of lattice rota-
tion symmetry in d-wave superconductors on the square lattice. This is described by
a field theory of an Ising nematic order parameter coupled to the gapless fermionic
quasiparticles. We determine the structure of the renormalization group to all orders
in a 1/Nf expansion, where Nf is the number of fermion spin components. Asymp-
totically exact results are obtained for the quantum critical theory in which, as in the
large Nf theory, the nematic order has a large anomalous dimension, and the fermion
spectral functions are highly anisotropic.
Next we study quantum phase transitions in antiferromagnetic kagome lattices.
Due to the high geometric frustration, this system poses as a good candidate for a
spin liquid with exotic excitations. Here we look at physics of the spinon and vison
sector.
In the spinon sector, we investigate the zero-temperature phase diagram of the
nearest-neighbor kagome antiferromagnet in the presence of Dzyaloshinksii-Moriya
interaction. We develop a theory for the transition between Z2 spin liquids with
bosonic spinons and a phase with antiferromagnetic long-range order. Connections
to recent numerical studies and experiments are discussed.
Finally in the vison sector, we present a projective symmetry group (PSG) analysis
of the spinless excitations of Z2 spin liquids on the kagome lattice. In the simplest
case, vortices carrying Z2 magnetic flux (‘visons’) are shown to transform under the
iii
Abstract iv
48 element group GL(2, Z3 ). Alternative exchange couplings can also lead to a second
case with visons transforming under 288 element group GL(2, Z3 ) × D3 . We study the
quantum phase transition in which visons condense into confining states with valence
bond solid order. The critical field theories and confining states are classified using
the vison PSGs.
Contents
Title Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Citations to Previously Published Work . . . . . . . . . . . . . . . . . . . vii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1 Introduction 1
1.1 d-wave Superconductivity and Nematic Ordering . . . . . . . . . . . . 2
1.2 Phase Transitions in Kagome Lattice Z2 Spin Liquids . . . . . . . . . 5
1.3 Organization of the thesis . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Renormalization group theory of nematic ordering

in d-wave superconductors 14
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Renormalization group analysis . . . . . . . . . . . . . . . . . . . . . 23
2.4 Theory for small v∆ /vF . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 Quantum criticality of the kagome antiferromagnet

with Dzyaloshinskii-Moriya interactions 35
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Mean field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Quantum criticality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.A Microscopic form of the Hamiltonian . . . . . . . . . . . . . . . . . . 51
3.B Dispersion of the lowest excited spinon state . . . . . . . . . . . . . . 52
3.C Condensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
v
Contents vi
4 Vison states and confinement transitions of Z2 spin liquids on the

kagome lattice 56
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Z2 spin liquids and visons . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3 PSG analysis of the fully frustrated Ising model on the dice lattice . . 66
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.A Berry phase of a vison . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.B Effective Ising models for visons on various other lattice geometries . 84
Citations to Previously Published Work
Chapter 2 has been previously published as
“Renormalization group theory of nematic ordering in d -wave supercon-

ductors”, Yejin Huh and Subir Sachdev, Phys. Rev. B, 78, 064512 (2008),
arXiv:0806.0002.
“Quantum criticality of the kagome antiferromagnet with Dzyaloshinskii-

Moriya interactions”, Yejin Huh and Lars Fritz and Subir Sachdev, Phys.
Rev. B, 81, 144432 (2010), arXiv:1003.0891.
“Vison states and confinement transitions of Z2 spin liquids on the kagome

lattice”, Yejin Huh and Matthias Punk and Subir Sachdev, arXiv:1106.3330.
Electronic preprints (shown in typewriter font) are available on the Internet at

the following URL:
http://arXiv.org
vii
Acknowledgments
It is a pleasure to thank the many people who have made this thesis possible.
I would like to start by expressing my sincere gratitude to my advisor, Professor
Subir Sachdev. Since first meeting him as a sophomore at Yale, I’ve learned everything
from quantum mechanics, statistical mechanics, to quantum phase transitions from
him. More importantly, I’ve been inspired by his depth of understanding and flow of
ideas. His guidance has helped me in the research and it has been an honor working
with him.
I would also like to thank Professor Jenny Hoffman and Professor Eugene Demler
for serving on my committee along with my advisor. I deeply appreciate all the
advice, insightful comments, and kind words they have given me.
It has been a great experience learning from talented postdocs at Harvard. I
am especially grateful to Lars Fritz and Matthias Punk. I have collaborated with
Lars on a project which became Chapter 3, and with Matthias on what became
Chapter 4. I have enjoyed working with them and learned a lot from each of them.
The exchanges have always been inspiring and stimulating. I would also like to thank
Markus Mueller, Cenke Xu, Erez Berg, and Philipp Strack for the discussions and
advice.
I have appreciated the camaraderie of my fellow students in Subir’s group, Adrian
Del Maestro, Yang Qi, Rudro Biswas, Max Metlitski, Eun Gook Moon and Susanne
Pielawa, all of whom I have become great friends with. I have benefited a great
amount from discussions with them. I’d like to thank Adrian for advice when I first
started out. I would like to especially thank Max for educating me on a lot of field
theory, and Eun Gook for being a wonderful officemate as well as for the great many
discussions and advice.
Sheila Ferguson, William Walker and Heidi Kaye have helped me navigate through
the system. I am grateful for their assistance, which has let me focus on my studies
and research.
I would also like to take this opportunity to thank some wonderful friends I have
made here at Harvard. Jung Ook Hong, Eunmi Chae, Junhyun Lee, Seung-Yeon
Kang, and Jaehoon Lee have all been immense support through my time at Harvard.
viii
Acknowledgments ix
Finally I would like to thank my parents Ick-Ryol Huh and Myoung Soon Kim for
always loving me, supporting me and being my role models. I would like to thank
my sister and best friend, Yesol Huh, for always being someone I can count on.
Dedicated to my father Ick Ryol Huh,
my mother Myoung Soon Kim,
and my sister Yesol Huh.
x
Chapter 1
Introduction
Classifying quantum phases and studying the quantum phase transitions between
them are powerful tools for studying the collective behavior of strongly correlated
materials. Continuous quantum phase transitions are usually associated with a van-
ishing order parameter and diverging physical quantities at the critical point. The
low energy effective theory and exponents for the power law divergence are charac-
teristics of the transition. While quantum phase transitions occur strictly at 0K,
which is experimentally not attainable, they do influence the T > 0 region as well. In
fact, there is a fan-like region in the phase diagram called the quantum critical region
characterized by its large critical fluctuations. Its universal properties are controlled
by the quantum critical point which lies at 0K. (Sachdev, 1999)
Various methods are used to study these quantum phases. In this thesis we will
encounter renormalization group method, mean field theory, projective symmetry
group analysis, and Ginzburg-Landau-Wilson type functionals. Systems that are
studied are d-wave superconductors and antiferromagnetic kagome lattices.
1
Chapter 1: Introduction 2
1.1 d-wave Superconductivity and Nematic Order-

ing
Since its discovery in 1986, many different families of cuprates have been discov-
ered. Some examples are LSCO (La2−x Srx CuO2 ), BSCCO (Bi2 Sr2 Can Cun+1 O2n+6 ),
and YBCO (YBa2 Cu3 O6+x ). Structurally all these materials have Copper oxide layers
in common, and it is believed that these CuO2 planes are responsible for the physics
of superconductivity. Electronic states near the Fermi level of high Tc cuprates derive
from the hybridized d and p orbitals of copper and oxygen. Layers of Copper oxide
are only very weakly coupled due to the large c/a ratio of the cuprate unit cell. This
in turn means that the size of the Brillouin zone in the kz direction is very small
and that a cuprate is a quasi-2 dimensional system with the symmetry properties
1
of a square lattice. In undoped parent materials of the cuprate, 2
spins are located
at the sites of copper, forming a square lattice of spin 12 s. The ground state of this
material is an antiferromagnetic Mott Insulator. Similarities between the antiferro-
magnetic low-temperature state of the undoped materials and the superconducting
state that emerges upon doping, primarily the dx2 −y2 orbital state of the Cu2+ ions,
suggest that electron-electron interactions are more significant than electron-phonon
interactions in cuprates, making the superconductivity unconventional. The Fermi
surface of these materials is shown in Fig 1.1. d-wave superconductors have an order
parameter, or the superconducting gap, proportional to cos kx − cos ky which vanish
along kx = ±ky . On the Fermi surface, there are four special points where the gap
vanishes. At these four nodal points of the Fermi surface, we can have excitations
with arbitrarily low energies.
Neutron scattering experiments on under-doped YBa2 Cu3 O6.45 (Tc =35K) Hinkov
et al. (2004, 2007, 2008) suggest nematic ordering. Upon measuring the spin-spin
correlation function, they have observed spontaneous onset of one dimensional order
below 150K. The anisotropy increased at lower temperatures and with decreasing
excitation energy. Spin excitation spectra are shown in Fig. 1.2. It can be seen that
the anisotropy comes from incommensurate peaks symmetrically displacing from Neel
k
y
k
x
2 1
q2 q1
3 4
Figure 1.1: Fermi surface and nodal points of a d-wave superconductor. A typical
large fermi surface is illustrated. The superconducting gap, ∆(k) = ∆0 (cos kx −
cos ky ), vanishes at four nodal points on the fermi surface. Excitations with arbitrarily
low energies are allowed and thus these nodal points play an important role in the
low energy physics.
order wave vector QAF = (π, π).

This was not recognized in previous experiments due to crystal twinning. In
cuprates, subtle crystallographic distortions can serve as aligning fields for symmetry-
broken electronic phases, which will otherwise be averaged out. YBCO has natural
slight anisotropy which serves as the aligning field. The parent material of YBCO
is YBa2 Cu3 O6 which is a Mott insulator which can be hole-doped by adding oxygen
atoms interstitially. This creates CuO chains only in the b direction which is unique
to this type of cuprate. Thus the a and b axes are no longer equivalent. This leads
to slightly, about 1%, different lattice constants. Usually YBCO is twinned when
doped, which means that there are patches with a in x axis and b in the y axis and
patches rotated by 90 degrees. In Hinkov et al. (2004, 2007, 2008) experiments, they
detwinned the material to observe the anisotropy. (by applying mechanical uniaxial
mechanical stress while flowing oxygen.)
Vojta et al. (Vojta et al., 2000c,b, 2008) have studied possible ordering that
could occur from a d-wave superconductor. Seven inequivalent order parameters were
Figure 1.2: (Hinkov et al., 2008) Spin excitation spectra at T = 5K at 3, 7, and

50meV (energy transfer) are plotted and wave vectors are in reciprocal lattice units.
Here we can see that spin excitations occur near Neel order wave vector QAF = (π, π).
At high excitation energies the spectrum is isotropic while at lower excitation energies,
anisotropy increases with decreasing energy. D shows the spectrum that would been
measured in the presence of crystal twinning. Previously the anisotropy had not been
measured due to crystal twinning. E shows intensity plots along two different axes a
and b. Along a, there are two incommensurate peaks symmetrically displaced from
QAF , while along b, the distribution is commensurate.
found, involving charge density waves and various subsidiary ordering, shifting the
four nodal points. In some cases the C4 symmtry was broken. In this thesis, we will
focus on the appearance of nematic ordering, in which the four-fold rotation breaks
into a two-fold rotation symmetry. This is the appearance of s-wave pairing. Initial
renormalization group analysis by Vojta et al. (2000c,b, 2008) of nematic ordering at
T = 0 found runaway flow to strong coupling in a computation based on an expansion
in (3 − d), where d is the spatial dimensionality. More recently, Kim et al. (2008)
argued that a second-order quantum phase transition existed in the limit of large
Nf , where Nf is the number of spin components. The physical case corresponding to
Nf = 2. In Chapter 2 we will present a different RG analysis using the framework of
1 1
the Nf
expansion. We will find a fixed point that does indeed exist at the order Nf
,
describing a second order transition.
1.2 Phase Transitions in Kagome Lattice Z2 Spin

Liquids
Frustrated magnets are consisted of localized magnetic moments, or spins, inter-
acting with competing exchange interactions that cannot be simultaneously satisfied.
A well-known simple example of a geometrically frustrated magnet is a triangle of
antiferromagnetic Ising spins, which exhibits frustration even at the level of three
sites. It can be easily verified that there is no way of satisfying all three bonds, and
that there are six ground states, each with two satisfied bonds and one frustrated
bond.
Macroscopic number of ground states in a lattice leads to strong fluctuations in
these systems. Both classical and quantum fluctuations may be present. Unlike
classical fluctuations which at lower temperatures become more sluggish, eventually
freezing out of equilibrium, quantum fluctuations persist down to 0K. In this thesis
1
we will focus on systems with spin S = 2
that have strong quantum fluctuations
which connect different ground states.
Interest in frustrated magnets are largely connected to spin liquids. Quantum
spin liquids have such strong quantum fluctuations that it has no magnetic order
down to 0K. It has a non-magnetic ground state although it is built from local
moments. Valance bonds, having the same property, is a good example of building
blocks that may describe the system. Resonating valance bonds (RVB) were first
proposed by Anderson (1973) as ground state wave functions for spin liquids. RVB
state is a superposition of electron pairs that form singlet bonds. This is a liquid-like
state that does not break spin rotation symmetry or any lattice symmetry. If the
RVB state is frozen into a specific configuration that breaks lattice symmetry, it is a
valance bond solid (VBS).
Candidates for spin liquids include geometrically frustrated lattices such as tri-
angular, kagome, and pyrocloro lattices. Some examples of specific systems are or-
ganic compounds EtMe3 Sb[Pd(dmit)2 ]2 (triangular) and κ-(BEDT-TTF)2 Cu2 (CN)3
(triangular), non-organic compounds herbertsmithite (kagome) and Na4 Ir3 O8 (hyper-
kagome).
Unlike organic materials which have a small charge gap, kagome antiferromagnets
have a large charge gap. Charges are well localized and therefore the Heisenberg
Hamiltonian is a good description of the system. Herbertsmithite is known to evade
ordering down to temperatures of 35 mK. The material does not show signs of struc-
tural distortions or anisotropy. (Helton et al., 2007; Ofer et al., 2006; Mendels
et al., 2007; Imai et al., 2008) Substitutional disorder and Dzyaloshinski-Moriya in-
teractions are present. (Olariu et al., 2008; Gregor and Motrunich, 2008; Rozenberg
and Chitra, 2008; Lee et al., 2007; Rigol and Singh, 2007; Zorko et al., 2008; Ofer
and Keren, 2009) Effect of Dzyaloshinski-Moriya interactions on quantum phases of
kagome antiferromagents are the focus of Chapter 3.
On the theoretical side, the most recent evidence (Nikolic and Senthil, 2003; Singh
and Huse, 2007; Jiang et al., 2008; Evenbly and Vidal, 2010; Poilblanc et al., 2010;
Schwandt et al., 2010) on the nearest-neighbor antiferromagnet points consistently
to a ground state with a spin gap of 0.05J and VBS order. The pattern of the VBS
order is quite complex with a large unit cell, but was anticipated by Marston and
Zeng (Marston and Zeng, 1991) by an application of the VBS selection mechanism
described in the 1/N expansion of the SU(N ) antiferromagnet. (Read and Sachdev,
1989a)
Below I will briefly present some theoretical tools that are useful in studying Z2
spin liquids. I will start with the reviewing the Schwinger boson method, discussing
the emergent Z2 gauge field, and will outline a mapping between a frustrated magnet
and a fully frustrated Ising model on the dual lattice. These will be used in Chapters 3
and 4.
1.2.1 Schwinger Boson

The Hamiltonians we are interested are the standard antiferromagnetic Heisenberg
Hamiltonian,
X
H= Jij Si · Sj (1.1)
ij
where Jij are nearest neighbor antiferromagnetic exchange interaction between spin
S quantum spins Si and Sj , each located at sites i and j. We describe SU(2) states
using spin 1
2
Schwinger bosons created by operators b†iα , with α =↑ / ↓. (Arovas and
Auerbach, 1988; Auerbach and Arovas, 1988)
1
|S, mi ≡ p (b†i↑ )S+m (b†i↓ )S−m |0i (1.2)
(S + m)!(S − m)!
It follows that
b†iα bαi = nb = 2S
1
Sia = b†iα σβaα bβi . (1.3)
2
Here we define a singlet operator between sites i and j as
X
Qij = σσ0 biσ bjσ0 (1.4)
σσ 0
with a fully anti-symmetric tensor.

!
0 1
= (1.5)
−1 0
We can describe the Hamiltonian as

J X †
H=− (Q Qij − 2S 2 ) (1.6)
2 <ij> ij
where J = Jij > 0 for nearest neighbors i and j.

In the next step we generalize the fields to N flavors. SU(2N ) turns out to be too
large and there is no generalization for the antisymmetric tensor for N > 1. Sp(N )
is the correct generalization which preserves the singlet formation. (Read and
Sachdev, 1991) This leads to
1 X
Qij = σσ0 bm m
iσ bjσ 0
N m,σ,σ0
J X αβ † †
H = − (J biα bjβ )(Jγδ biγ bjδ ) (1.7)
2N <ij>
up to an additive constant, where J is the Sp(N ) generalization of .

In Chapter 3 we are interested in the the limit of N → ∞ while keeping the ratio
nb
κ= (1.8)
N
fixed. We will study phase transitions occuring while tuning κ.
1.2.2 Z2 Spin Liquid

In the following I will motivate the Z2 spin liquid starting from the Schwinger
boson Hamiltonian discussed in section 1.2.1 using projective symmetry group (PSG)
analysis following Wang and Vishwanath (2006). The mean field Hamiltonian is
J X †
HM F = − (q Qij + Q†ij qij − qij∗ qij ). (1.9)
2 <ij> ij
In the Schwinger boson description of spins there is a local U (1) transformation,
brσ → eiφ(r) brσ , (1.10)
which leaves all the observables invariant as can be seen in Eq. 1.3. The mean-field
ansatz transforms as,
qij → e−iφ(i)−iφ(j) qij . (1.11)
As defined by Wen (2002), invariant gauge group (IGG) is the subgroup of PSG
which keeps the ansatz invariant. This is an artificial gauge symmetry that rises
from defining fractionalized particles. These do not have a corresponding physical
symmetry and describe the low energy gauge group. Assuming a non-zero qij , for
bipartite lattices a transformation of brσ → e±iφ brσ with opposite signs on different
sublattices, keeps the ansatz invariant and therefore the IGG is U(1). However,
for non-bipartite frustrated lattices, such as the antiferromagnetic kagome lattice,
the only transformations that satisfy this condition are the identity operation and
brσ → −brσ . The IGG is Z2 . Therefore the low energy effective theory (fluctuations
about the mean field state) of an antiferromagnetic kagome lattice will be described
by an emergent Z2 gauge field.
Another approach we can take is to realize that the frustration introduces a charge
−2 Higgs scalar, coming from the non-collinear wave vector which minimizes the
energy (coplanar critical mode), which reduces the U(1) gauge symmetry down to
Z2 . (Sachdev, 2010) In the case with large spin gap, we may integrate out the spin
degrees of freedom and we will be left with a pure Z2 gauge degree of freedom.
1.2.3 Quantum Dimer Model

Low energy singlet excitations of a quantum paramagnet can be captured using
a quantum dimer model. RVB states that involve only nearest neighbor pairing are
described using a quantum dimer model. A quantum dimer model on a square lattice
can be written by
X
HQDM = −tT̂ + v V̂ = (−t(| =ihk | + | kih= |) + v(| =ih= | + | kihk |)) . (1.12)

V̂ is the potential term while T̂ describes the motion. The kinetic term flips
the dimers from one configuration to another. Only flippable even loops, square
plaquettes in the square lattice and hexagons, diamonds, and stars etc. for kagome
lattices are allowed. (for a full classification of of flippable loops on kagome lattices,
see Nikolic and Senthil (2003))
This quantum dimer model can be derived as a perturbation expansion of the

SU(2) Heisenberg magnet where orthonormalized dimer coverings form a basis and
overlaps between dimer coverings give rise to the kinetic term. (Rokhsar and Kivel-
son, 1988) Another method is to start from the Schwinger boson method discussed
in Section 1.2.1. Instead of keeping κ fixed, we can fix nb and take the N → ∞
limit. This gives rise to the quantum dimer model with hard core constraint. Dimer
1
resonance term comes in order N
. (Affleck and Marston, 1988)
1.2.4 Ising Gauge Theory

Since the low energy theory is an Ising gauge theory, we can write the effective
Hamiltonian as
X XY
HGDM = Γ σx − κ σz (1.13)
−
P
where Ising gauge fields σ live on the links of the lattice, − is a sum over all links and
P
sums over closed loops. All loops are allowed due to gauge symmetry, although
the shorter loops are more important. In the case of the kagome lattice,
X XY XY
H=Γ σ x − κ3 σ z − κ6 σz (1.14)
− 4 4 7 7
This Hamiltonian has a gauge invariance under flipping all spins in the σ z basis
emanating from a site. The transformation that generates this is
Y
GIGT = σx (1.15)
+
Ising gauge theory Eq. 1.13 exhibits a confined phase at Γ κ, in which the low
energy excitations are elementary loops of σ x = 1 that cost energy 2Γ per length of
the loop. On the other hand, when Γ κ, the loops proliferate with no tension and
drives deconfinement.
We can map the above Ising gauge theory to a quantum dimer model under a
certain limit following Moessner et al. (2001). We start by interpreting link variables
(gauge fields) σ x = ±1 as the presence/absense of dimer on the link. σ z flips between
σ x = ±1 states. Then we can see that the σ x term is the potential term while σ z acts
as the kinetic term of the dimers. Only flippable loops are allowed in summation for
the dimer model. In a kagome lattice these are even loops containing one hexagon,
such as hexagons, diamonds, which may have different κ values. We can break up
these loops into elementary plaquettes which are hexagons and triangles. It can be
easily seen that flipping each flippable plaquette can be written as flipping a hexagon
and even number of bond-sharing triangles. Here we will treat hexagons and triangles
as independent plaquettes that we may flip and project out the states that are not
allowed.
Assuming that even number of links emanate from a site, we need GIGT = 1 for
cases where we have an even number of dimers on each site, while we should impose
GIGT = −1 for odd number of dimers. Since we are interested in the case where there
is one dimer emanating from all the sites, we will focus on GIGT = −1. To complete
the mapping to the QDM, we need to project out the cases where there are more
than one dimer (such as 3 or 5) from any site. We can reach this by taking the limit
Γ → ∞. Therefore we can see that the Ising gauge theory maps exactly on to the
QDM deep in the confined phase. This mapping will be used in Chapter 4 to describe
different confined phases that appear.
1.2.5 Duality of Ising Gauge Theory and Ising Model in

d=2+1
Here I will review the duality mapping of 2+1d IGT to Ising model. (For an in
depth review, see Kogut (1979).) Define at the center of plaquettes (p),
Y Y Y
µxp ≡ σ z , µzp ≡ − σ x , µyp ≡ iµx µz , = −1, (1.16)
p ;p 3
Q Q
where p is the product of the σ’s around the plaquette p and ;p is the product of
σ’s intersecting any path from infinity to p. is the external gauge defined on the dual
lattice. Notice that without defined as above the definition of µz is path-dependent
where the phase acquired in a loop C is π×(direct sites enclosed in loop C). This
condition comes from the requirement GIGT = −1 and in the case of GIGT = 1 we do
not need .
It can be easily checked that the µ’s satisfy the usual commutation relation of
Pauli matrices. Using this definition we can express the Hamiltonian in Eq. 1.14 as
X X X
Hdual = −Γ pp0 µzp µzp0 − κ3 µxp − κ6 µxp
p,p0 3 6
X X X
= − Jij µzp µzp0 − κ3 µxp − κ6 µxp (1.17)
<ij> 3 6
where the product of the bonds around each elementary plaquette is negative.
Y
sgn(Jij ) = −1. (1.18)
3
This is a fully frustrated Ising model in the dual dice lattice. µ defined on the
dual lattice are vortices in the Z2 gauge field called visons. µzp has the physical
interpretation of creating/annhilating a vison at dual site p, while the transverse field
term serves as the kinetic term which interchanges frustrated bonds (−Jµµ > 0) and
satisfied bonds (−Jµµ < 0) emanating from the site.
Visons have minimal long range coupling with spinons and chagons such that
the spinon (or chargon) gains a phase π when it goes around a vison. We will see
the case of the phase gain of spinon around a vison explicitly in Chapter 4. Thus
when the visons are gapped, spinons and chargons are deconfined and may propagate
freely. This is the fractionalized state. However when the visons condense, they drive
confinement, such that spinons and chargons are no longer well defined quasi-particles.
1.3 Organization of the thesis

In Chapter 2 we examine the quantum theory of the spontaneous breaking of
lattice rotation symmetry in d-wave superconductors on the square lattice. This is
described by a field theory of an Ising nematic order parameter coupled to the gapless
fermionic quasiparticles. We determine the structure of the renormalization group to
all orders in a 1/Nf expansion, where Nf is the number of fermion spin components.
Asymptotically exact results are obtained for the quantum critical theory in which,
as in the large Nf theory, the nematic order has a large anomalous dimension, and
the fermion spectral functions are highly anisotropic.
In Chapter 3 we investigate the zero-temperature phase diagram of the nearest-

neighbor kagome antiferromagnet in the presence of Dzyaloshinksii-Moriya interac-
tion. We develop a theory for the transition between Z2 spin liquids with bosonic
spinons and a phase with antiferromagnetic long-range order. Connections to recent
numerical studies and experiments are discussed.
In Chapter 4 we present a projective symmetry group (PSG) analysis of the spin-
less excitations of Z2 spin liquids on the kagome lattice. In the simplest case, vortices
carrying Z2 magnetic flux (visons) are shown to transform under the 48 element
group GL(2, Z3 ). Alternative exchange couplings can also lead to a second case with
visons transforming under 288 element group GL(2, Z3 ) × D3 . We study the quantum
phase transition in which visons condense into confining states with valence bond
solid order. The critical field theories and confining states are classified using the
vison PSGs.
Chapter 2
Renormalization group theory of

nematic ordering
in d-wave superconductors
2.1 Introduction
There has been a great deal of research on the onset of a variety of competing
orders in the hole-doped cuprate superconductors. In Ref. Vojta et al. (2000c,b, 2008),
a classification of spin-singlet order parameters at zero momentum was presented:
such orders are able to couple efficiently to the gapless nodal quasiparticle excitations
of a d-wave superconductor. Our focus in the present paper will be on one such order
parameter: ‘nematic’ ordering in which the square lattice (tetragonal) symmetry of
the d-wave superconductor is spontaneously reduced to rectangular (orthorhombic)
symmetry (Kivelson et al., 1998). This transition is characterized by an Ising order
parameter, but the quantum phase transition is not in the usual Ising universality
class (Vojta et al., 2000c,b, 2008) because the coupling to the gapless fermionic
quasiparticles changes the nature of the quantum critical fluctuations. Our work
is motivated by recent neutron scattering observations (Hinkov et al., 2004, 2007,
2008) of a strongly temperature (T ) dependent susceptibility to nematic ordering in
14
Chapter 2: Renormalization group theory of nematic ordering
2 1
Figure 2.1: Nodal points of the d-wave superconductor in the square lattice Brillouin
zone. The 2-component Ψ1,2 fermions are in the vicinity of the labelled nodal points
and their partners at diagonally opposite points.
detwinned crystals of YBa2 Cu3 O6.45 .

An initial renormalization group (RG) analysis (Vojta et al., 2000c,b, 2008) of
nematic ordering at T = 0 found runaway flow to strong-coupling in a computation
based on an expansion in (3 − d) where d is the spatial dimensionality. More recently,
Ref. Kim et al. (2008) argued that a second-order quantum phase transition existed
in the limit of large Nf , where Nf is the number of spin components (the physical
case corresponding to Nf = 2). We refer the reader to Ref. Kim et al. (2008) for
further discussion on the physical importance and experimental relevance of nematic
ordering in d-wave superconductors.
Here we will present a RG analysis using the framework of the 1/Nf expansion. We
will find that a fixed point does indeed exist at order 1/Nf , describing a second-order
quantum transition associated with the onset of long-range nematic order. However,
the scaling properties near the fixed point have a number of unusual properties, con-
trolled by the marginal flow of a “dangerously” irrelevant parameter. This parameter
is v∆ /vF , where v∆ and vF are the velocities of the nodal fermions parallel and per-
pendicular to the Fermi surface (see Fig. 2.1). We will show that the fixed point
has (v∆ /vF )∗ = 0 and so the transition is described by an “infinite anisotropy” away
from the “relativistic” fixed points found for other competing orders (Vojta et al.,
2000c,b, 2008). It is important to note, however, that even though the fixed point
has (v∆ /vF )∗ = 0, it is not described by an effectively one-dimensional theory of a
straight Fermi surface; a fully two-dimensional theory is needed as is discussed in
more detail in Section 2.4. The approach to this fixed point is logarithmically slow,
and physical properties have to be computed at finite v∆ /vF . Our main results for
the RG flow of the velocities are in Eqs. (2.25) and (2.26), where C1,2,3 are functions
only of v∆ /vF which are specified in Eq. (2.35); for v∆ /vF 1, the equations reduce
to the explicit forms in Eqs. (2.39-2.41). These (and related) equations can be inte-
grated in a standard manner to yield the dependence of observables on temperature
and deviation from the nematic critical point, and the results are in Figs. 2.6, 2.7,
and 2.8.
The existence of a fixed point at (v∆ /vF )∗ = 0 suggests that we analyze the theory
directly in the limit of small v∆ /vF , without an appeal to an expansion in powers of
1/Nf . The nature of the small v∆ /vF limit is quite subtle, and has to be taken with
care: it will be described in Section 2.4. We argue there that the fluctuations of the
nematic order are controlled by a small parameter which is ∼ v∆ /(Nf vF ), and not
1/Nf alone. Consequently, we believe our computations are controlled in the limit of
small v∆ /vF even for Nf = 2. Indeed, the results in Eqs. (2.39-2.41), and many other
related results, are expected to be exact as we approach the quantum critical point.
The outline of the remainder of this paper is as follows. The field theory for the
nematic ordering transition will be reviewed in Section 2.2, along with a discussion of
the 1/Nf expansion. The RG analysis to order 1/Nf will be presented in Section 2.3.
Finally, higher order corrections in 1/Nf , and the nature of the small v∆ /vF limit will
be discussed in Section 2.4
2.2 Field theory

We begin by review the field theory for the nematic ordering transition introduced
in Ref. Vojta et al. (2000c,b, 2008).
The action for the field theory, S, has three components
S = SΨ + Sφ0 + SΨφ . (2.1)
The first term in the action, SΨ , is simply that for the low energy fermionic
excitations in the dx2 −y2 superconductor. We begin with the electron annihilation
operator cqa at momentum q and spin a =↑, ↓. We will shortly generalize the theory
to one in which a = 1 . . . Nf , with Nf an arbitrary integer. We denote the cqa
operators in the vicinity of the four nodal points q = (±K, ±K) by (anti-clockwise)
†
f1a , f2a , f3a , f4a . Now introduce the 2-component Nambu spinors Ψ1a = (f1a , εab f3b )
†
and Ψ2a = (f2a , εab f4b ) where εab = −εba and ε↑↓ = 1 [we will follow the convention of
writing out spin indices (a, b) explicitly, while indices in Nambu space will be implicit].
Expanding to linear order in gradients from the nodal points, the Bogoliubov action
for the fermionic excitations of a d-wave superconductors can be written as (see
Fig. 2.1)
Nf
d2 k X X †
Z
SΨ = 2
T Ψ1a (−iωn + vF kx τ z + v∆ ky τ x ) Ψ1a
(2π) ω a=1
n
Nf
d2 k X X †
Z
+ T Ψ (−iωn + vF ky τ z + v∆ kx τ x ) Ψ2a . (2.2)
(2π)2 ω a=1 2a
n
Here ωn is a Matsubara frequency, τ α are Pauli matrices which act in the fermionic
particle-hole space, kx,y measure the wavevector from the nodal points and have been
rotated by 45 degrees from qx,y co-ordinates, and vF , v∆ are velocities. The sum over
a in Eq. (2.2) can be considered to extend over an arbitrary number Nf .
The second term, Sφ0 describes the effective action for the Ising nematic order
parameter, which we represent as a real field φ. Considering only the contribution to
its action generated by high energy electronic excitations, we obtain only the analytic
terms present in the quantum Ising model:
c2
Z h1 r u0 i
Sφ = d2 xdτ (∂τ φ)2 + (∇φ)2 + φ2 + φ4 ;
0
(2.3)
2 2 2 24
here τ is imaginary time, c is a velocity, r tunes the system across the quantum critical
point, and u0 is a quartic self-interaction.
The final term in the action, SΨφ couples the Ising nematic order, φ, to the nodal
fermions, Ψ1a , Ψ2a . This can be deduced by a symmetry analysis (Vojta et al.,
2000c,b, 2008), and the important term is a trilinear “Yukawa” coupling
Z h Nf i
X † x † x
2
SΨφ = d xdτ λ0 φ Ψ1a τ Ψ1a + Ψ2a τ Ψ2a , (2.4)
a=1
where λ0 is a coupling constant.

We can now see that the action S describes the couplings between the φ bosons
and the Ψ fermions, both of which have a ‘relativistic’ dispersion spectrum. However,
the velocity of ‘light’ in their dispersion spectrums, c, vF , v∆ , are not all equal. If we
choose equal velocities with vF = v∆ = c, then the decoupled theory SΨ + Sφ0 does
have a relativistically invariant form. However, even in this case, the fermion-boson
coupling SΨφ is not relativistically invariant: the coupling matrix τ x in Eq. (2.4)
breaks Lorentz symmetry, and this feature will be crucial for our analysis. If we
replace τ x by τ y in Eq. (2.4), then we obtain an interacting relativistically invariant
theory for the equal velocity case, which was studied in Ref. Vojta et al. (2000c,b,
2008) as a description of the transition from a dx2 −y2 to a dx2 −y2 +idxy superconductor.
A renormalization group analysis of the above action has been presented previously
(Vojta et al., 2000c,b, 2008) in an expansion in (3 − d). No fixed point describing
the onset of nematic order was found, with the couplings u0 and λ0 flowing away
to infinity. This runaway flow was closely connected to the non-Lorentz-invariant
structure of SΨφ and the resulting flow of the velocities away from each other. In
contrast, for the transition to a dx2 −y2 + idxy superconductor, a stable fixed point
was found at which the velocities flowed to equal values at long scales (Vojta et al.,
2000c,b, 2008). Similar relativistically-invariant fixed points are also found in other
cases involving the onset of spin or charge density wave orders at wavevectors which
nest the separation between nodal points (Balents et al., 1998; Vojta et al., 2000a).
2.2.1 Large Nf expansion

We will analyze S in the context of the 1/Nf expansion. Formally, this involves
integrating out the Ψ1,2 fermions, and obtaining the resulting non-local effective action
for φ. Near the potential quantum critical point, the non-local terms so generated
are more important in the infrared than the terms Sφ0 in Eq. (2.3), apart from the
“mass” term r. So we can drop the remaining terms in Sφ0 ; also for convenience we
rescale φ → φ/λ0 and r → Nf rλ20 , and obtain the local field theory
 
Z Nf
Nf r 2 X
S = SΨ + d2 xdτ  φ +φ Ψ†1a τ x Ψ1a + Ψ†2a τ x Ψ2a  . (2.5)
2 a=1
This local field theory will form the basis of all our RG analysis. Note that this field
theory has only 3 parameters: r, vF , and v∆ . So any RG equations can be expressed
only in terms of these parameters, as will be presented in the following section.
The large Nf expansion proceeds by integrating out the Ψ fermions, yielding an
effective action Sφ for the nematic order φ. It is important to note that Sφ is a
complicated non-local functional of φ, which however depends upon only r, vF , and
v∆ . Also, in writing out explicit forms for Sφ it is essential to keep subtle issues
on orders of limits in mind. In particular, for the large Nf phase diagram, we need
the effective potential for φ at zero k and ω. Thus we need an expansion of the
effective potential in the regime |φ| |k|, |ω| — the structure of this was discussed in
Ref. Kim et al. (2008), and yielded a second-order transition in the limit of large Nf .
In contrast, for our RG analysis here, we will work in the quantum critical region,
where hφi = 0, and so we need the functional for |φ| |k|, |ω|. In this case, we can
expand Sφ in powers of φ to yield the formal result
Z
Sφ 1
= (r + Γ2 (K)) |φ(K)|2
Nf 2 K
4 Z
!
1Y X
+ δ Ki Γ4 (K1 , K2 , K3 , K4 )φ(K1 )φ(K2 )φ(K3 )φ(K4 ) + . . .
4 i=1 Ki i
(2.6)
Here the Ki ≡ (ki , ωi ) are 3-momenta. The functions Γi are all given by one fermion
loop Feynman graphs with i insertions of the external φ vertices, as shown in Fig. 2.2;
we denote the external momenta of these vertices, Ki , clockwise around the fermion
loop. The Feynman loop integrals are quite tedious to evaluate, especially for large
Figure 2.2: Feynman graph expansion for the effective action Sφ . The full lines are
fermion propagators, while the wavy lines are φ insertions.
i, and so below we only present explicit expressions for the needed low order terms.
However, all the Γi are universal functions of only the momenta and vF and v∆
Our analysis will require the explicit form of Γ2 (K). This can be written as
Γ2 (K) = Π2 (kx , ky , ω) + Π2 (ky , kx , ω) (2.7)
with the 2 terms representing the contributions of the Ψ1 and Ψ2 fermions respectively.
The one fermion loop diagram yields
Z 2 Z
dp dΩ h x
Π2 (kx , ky , ω) = Tr τ (−i(Ω + ω) + vF (px + kx )τ z + v∆ (py + ky )τ x )−1
4π 2 2π
i
× τ x (−iΩ + vF px τ z + v∆ py τ x )−1
1 (ω 2 + vF2 kx2 )
= (2.8)
16vF v∆ (ω 2 + vF2 kx2 + v∆
2 2 1/2
ky )
Here, we have subtracted out a constant which shifts the position of the critical point.
For Γ4 we will only need the cases where 2 of the four external 3-momenta vanish.
The first is
Γ4 (K, −K, 0, 0) = Π4a (kx , ky , ω) + Π4a (ky , kx , ω) (2.9)
where
d2 p
Z Z
dΩ h x
Π4a (kx , ky , ω) = Tr τ (−iΩ + vF px τ z + v∆ py τ x )−1
4π 2 2π
n o3 i
× τ x (−i(Ω + ω) + vF (px + kx )τ z + v∆ (py + ky )τ x )−1
1 ∂ 2 Π2 (kx , ky , ω)
= 2
2v∆ ∂ky2
1 (ω 2 + vF2 kx2 )(ω 2 + vF2 kx2 − 2v∆
2 2
ky )
= − 2 2 2 2 2 5/2
. (2.10)
32vF v∆ (ω + vF kx + v∆ ky )
The other case with 2 vanishing 3-momenta is
Γ4 (K, 0, −K, 0) = Π4b (kx , ky , ω) + Π4b (ky , kx , ω) (2.11)
where
d2 p
Z Z
dΩ hn x o2
Π4b (kx , ky , ω) = Tr τ (−iΩ + vF px τ z + v∆ py τ x )−1
4π 2 2π
n o2 i
× τ x (−i(Ω + ω) + vF (px + kx )τ z + v∆ (py + ky )τ x )−1
1 2 2
(ω 2 + vF2 kx2 )(ω 2 + vF2 kx2 − 2v∆ ky )
= 2 2 2 2 2 5/2
. (2.12)
16vF v∆ (ω + vF kx + v∆ ky )
From the effective action Sφ in (2.6), we can obtain the 1/Nf corrections to various
observable quantities.
For the nematic susceptibility, χφ , we have
2Γ4 (K, −K, 0, 0) + Γ4 (K, 0, −K, 0)
Z
−1 1
χφ = r + . (2.13)
Nf K Γ2 (K) + r
Using the values in Eq. (2.10) and (2.12), we observe that the 1/Nf correction is
identically zero. This can be traced to the arguments in Ref. Kim et al. (2008) (based
upon gauge invariance and the fact that the coupling in Eq. (2.4) is to a globally
conserved fermion current) that the effective potential for the field φ is unrenormalized
by the low energy fermion action. Indeed, this argument implies that all higher order
terms in 1/Nf also vanish, and that χ−1
φ = r exactly in the present continuum field
theory. So we may conclude that the susceptibility exponent, γ, for the nematic order
parameter has the exact value
γ = 1. (2.14)
(a)
(b)
Figure 2.3: Order 1/Nf contributions to the (a) self energy Σ1 of Ψ1 fermions and
(b) the vertex Ξ.
Note that there are 1/Nf corrections to the momentum and frequency dependence of
the nematic susceptibility, and these will appear in our RG analysis below.
Turning to the 1/Nf expansion for the Ψ1 Green’s function, we write this as
G−1 z x
Ψ1 (k, ω) = −iω + vF kx τ + v∆ ky τ − Σ1 (k, ω), (2.15)
the self-energy is given by the Feynman graph in Fig. 2.3:

Z 2 Z
1 dp dΩ [i(Ω + ω) − vF (px + kx )τ z + v∆ (py + ky )τ x ]
Σ1 (k, ω) =
Nf 4π 2 2π (Ω + ω)2 + vF2 (px + kx )2 + v∆ 2
(py + ky )2
1
× . (2.16)
Γ2 (p, Ω) + r
Finally, we will also need the 1/Nf correction to the boson-fermion vertex (the
Yukawa couping) in Eq. (2.4). At zero external momenta and frequencies, this vertex
is renormalized by Fig. 2.3 to
Z 2 Z
1 dp dΩ h x
Ξ = τ + x
τ (−iΩ + vF px τ z + v∆ py τ x )−1
Nf 4π 2 2π
1 i
× τ x (−iΩ + vF px τ z + v∆ py τ x )−1 τ x (2.17)
Γ2 (p, Ω) + r
The RG equations will be obtained in the next section from Eqs. (2.13), (2.16)
and (2.17).
2.3 Renormalization group analysis

We begin our discussion of the RG by describing the general structure which ap-
plies to all orders in the 1/Nf expansion. The RG will be based upon a consideration
of the renormalization of the local field theory S in Eq. (2.5). We are interested in
the behavior of this field theory under the rescaling transformation
k = k 0 e−`
ω = ω 0 e−` . (2.18)
Note that we have not introduced a dynamic critical exponent z for the rescaling
of the frequency. This is because we will allow both velocities vF and v∆ to flow,
and this flow will effectively account for any anomalous dynamic scaling. Under this
spacetime rescaling, we also have a rescaling of the fermion and boson fields
Z `
0 0 0 1
Ψ1,2 (k, ω) = Ψ1,2 (k , ω ) exp d`(4 − ηf )
2 0
Z `
0 0 0 1
φ(k, ω) = φ (k , ω ) exp d`(5 − ηb ) . (2.19)
2 0
Here we have allowed the anomalous dimensions ηb,f to be scale-dependent, as that
will be the case below. To implement these field rescalings, we have to determine how
the field scales are defined. For the fermions, as is conventional, we set the scale of
Ψ1,2 so that the co-effecients of the time derivative terms in SΨ remain unity. For the
nematic order parameter, φ, there is no “kinetic energy” term in Eq. (2.5), and so we
cannot use the conventional method. Instead, recall that the “Yukawa” coupling λ0
was absorbed into the overall scale of φ in Eq. (2.5). Therefore it is natural to set the
scale of φ so that the boson-fermion vertex in Eq. (2.5) remains fixed at unity. This
is also consonant with the fact that the boson “kinetic energy” comes entirely from
the fermion loops.
The RG equations now follow from the low frequency forms of the fermion self
energy Σ1 and the vertex Ξ. These will have a dependence upon an ultraviolet cutoff,
Λ. We will discuss below an explicit method of applying this cutoff. For now, we
note that for the RG we only need the logarithmic derivative of the self energy and
the vertex. Thus let us write

d
Λ Σ1 (k, ω) = C1 (−iω) + C2 vF kx τ z + C3 v∆ ky τ x
dΛ
d
Λ Ξ = C4 τ x . (2.20)
dΛ
On the right hand side, we assume (as in the usual field-theoretic RG) that the limit
Λ → ∞ can be safely taken. Then a simple dimensional analysis shows that the C1−4
are all universal dimensionless functions of the velocity ratio v∆ /vF . In principle, the
above method can be generalized to any needed order in the 1/Nf expansion by taking
the appropriate logarithmic cutoff derivative of the self energy Σ1 and the vertex Ξ,
as reviewed in Ref. Brézin et al. (1976).
With the results in Eq. (2.20), we can set the scaling dimensions of the fields.
Considering the renormalization of the −iω term in SΨ , we obtain
ηf = −C1 . (2.21)
Imposing the unit value of the Yukawa coupling in Eq. (2.5) we obtain
ηb = 1 + 2C4 − 2ηf = 1 + 2C4 + 2C1 (2.22)
Note the large value of ηb in the large Nf theory. This is a consequence of the fact that
the φ kinetic energy is tied to the non-analytic fermion loop contribution. We can now
also use the non-renormalization of the φ2 term in Eq. (2.5), as noted in Eq. (2.14),
to determine the flow equation for the “mass” r. This defines the correlation length
exponent ν by the RG equation
dr 1
= r (2.23)
d` ν
for small r, and we have the non-mean-field value
1
ν= . (2.24)
2 − ηb
The results in Eq. (2.20) also yield the flow equations for the velocities. By
considering the renormalization of the k-dependent terms in SΨ we determine that
the velocity vF flows according
dvF
= (−ηf − C2 )vF = (C1 − C2 )vF , (2.25)
d`
while the velocity v∆ obeys

dv∆
= (−ηf − C3 )v∆ = (C1 − C3 )v∆ . (2.26)
d`
Clearly, the ratios of the velocity obeys
d(v∆ /vF )
= (C2 − C3 )(v∆ /vF ). (2.27)
d`
All that remains now is to determine the C1−4 as a function of v∆ /vF . These
functions also depend upon the “mass” r, but we will henceforth restrict ourselves
to the vicinity of the critical point by setting r = 0. We reiterate that, in principle,
our method allows us to obtain these functions to any needed order in the 1/Nf
expansion. However, we will only obtain explicit results below to first order 1/Nf .
The computation of C1−4 requires the imposition of a ultraviolet momentum cutoff.
We implement this (Vojta et al., 2000c,b, 2008) by multiplying both the Fermi and
Bose propagators by a smooth cutoff function K(k 2 /Λ2 ). Here K(y) is an arbitrary
function with K(0) = 1 and which falls off rapidly with y at y ∼ 1 e.g. K(y) = e−y .
We will show below that the RG equations are independent of the particular choice
of K(y).
Let us now discuss the evaluation of C1−3 . Schematically, we can write for the
fermion self energy
d3 P (p + k)2 p2
Z
Σ1 (K) = F (P + K)G(P )K K (2.28)
8π 3 Λ2 Λ2
where K ≡ (k, ω) and P ≡ (p, Ω) are 3-momenta, and F and G are functions which
can be obtained by comparing this expression to Eq. (2.16). Also notice that (at r = 0)
F and G are both homogenous functions of 3-momenta of degree -1. Expanding to
first order in Kµ (µ is a spacetime index) we have
Z 3 2 2
2pµ 0 p2

d P ∂F (P ) 2 p p
Σ1 (K) ≈ Kµ G(P )K + F (P )G(P )K K
8π 3 ∂Pµ Λ2 Λ2 Λ2 Λ2
(2.29)
where pµ is the three vector (0, px , py ) (while Pµ = (Ω, px , py )). Now taking the Λ
derivative
Z 3
4p2 ∂F (P )
2 2
d dP pµ p p
Λ Σ1 (K) ≈ Kµ 3
− 2 − 4F (P ) 2 G(P )K 2
K0
dΛ 8π Λ ∂Pµ Λ Λ Λ2
4p2 pµ p2 p2 p2

00 02
− F (P )G(P ) K K + K (2.30)
Λ4 Λ2 Λ2 Λ2
Now we convert to cylindrical co-ordinates in spacetime by writing
Pµ = yΛ(vF x, cos θ, sin θ) and integrate over y, x, and θ. Also, let us define P̂µ =
(vF x, cos θ, sin θ) and p̂µ = (0, cos θ, sin θ). We use the homogeneity properties of the
functions F and G to obtain
Z 2π "( )
vF Kµ ∞
Z Z ∞
d ∂F (P̂ )
Λ Σ1 (K) ≈ dx dθ −4 − 4p̂µ F (P̂ ) G(P̂ ) ydyK(y 2 )K0 (y 2 )
dΛ 8π 3 −∞ 0 ∂P µ 0
Z ∞
3
2
00 2 02 2

− 4p̂µ F (P̂ )G(P̂ ) y dy K y K y + K y (2.31)
0
Now the integrals over y can be evaluated exactly by integration by parts, and the
final result is independent of the precise form of K(y):
vF Kµ ∞
Z Z 2π
d ∂F (P̂ )
Λ Σ1 (K) = 3
dx dθ G(P̂ ) (2.32)
dΛ 8π −∞ 0 ∂Pµ
In a similar manner, we can write the result for the vertex Ξ in Eq. (2.17) in the
schematic form
d3 P
Z 2
x 3 p
Ξ=τ + H(P )K , (2.33)
8π 3 Λ2
where H(P ) is a homogenous function of P of degree -3. Proceeding as above we
obtain the analog of Eq. (2.32)
Z ∞ Z 2π
d vF
Λ Ξ= 3 dx dθH(P̂ ). (2.34)
dΛ 8π −∞ 0
We can now use the above expressions to obtain explicit results for the constants
C1−4 at order 1/Nf . For C1−3 we combine Eqs. (2.16), (2.20), (2.28), and (2.32) to
obtain
2(v∆ /vF ) ∞
Z 2π
(x2 − cos2 θ − (v∆ /vF )2 sin2 θ)
Z
C1 = dx dθ G(x, θ)
π 3 Nf −∞ 0 (x2 + cos2 θ + (v∆ /vF )2 sin2 θ)2
2(v∆ /vF ) ∞
Z 2π
(−x2 + cos2 θ − (v∆ /vF )2 sin2 θ)
Z
C2 = dx dθ G(x, θ)
2(v∆ /vF ) ∞
Z 2π
(x2 + cos2 θ − (v∆ /vF )2 sin2 θ)
Z
C3 = dx dθ G(x, θ) (2.35)
where
x2 + sin2 θ x2 + cos2 θ
G −1 (x, θ) = p + p (2.36)
x2 + (v∆ /vF )2 cos2 θ + sin2 θ x2 + cos2 θ + (v∆ /vF )2 sin2 θ
is the φ propagator inverse. All the integrals in Eqs. (2.35) are well-defined and
convergent for all non-zero v∆ /vF , and have to be evaluated numerically. It is also
evident that the results are functions only of v∆ /vF . The results of a numerical
evaluation are shown in Fig. 2.4.
0.1
0.5 1.0 1. 5 2.0

- 0.1
- 0.2
- 0.3
- 0.4
Figure 2.4: Plots of the functions C1−3 in Eq. (2.35).
Similarly, we can combine Eqs. (2.17), (2.20), (2.33), and (2.34) to deduce that
C4 = −C3 , (2.37)
at leading order in the 1/Nf expansion.

With the values in Eq. (2.35), we are now in a position to numerically solve the
RG flow for the velocity ratio v∆ /vF in Eq. (2.27). We plot the right-hand-side of
this flow equation in Fig. 2.5. Note that the only zero of the flow equation is at
v∆ /vF = 0, and that this fixed point is attractive in the infrared. So for all starting
values of the parameter v∆ /vF , the flow is towards v∆ /vF → 0 for large `. At large
enough scales, it is always possible to approximate the constants C1−3 in Eqs. (2.35)
0.5 1.0 1.5 2.0
- 0.2
- 0.4
-0.6
- 0.8
Figure 2.5: Right-hand-side of the RG flow equation Eq. (2.27) for v∆ /vF .
by their limiting values for small v∆ /vF . Evaluating the integrals in Eq. (2.35) in this
limit, we obtain
(v∆ /vF )
+ O (v∆ /vF )3

C1 = −0.4627
Nf
(v∆ /vF )
+ O (v∆ /vF )3

C2 = −0.3479
Nf

8 (v∆ /vF )
+ O (v∆ /vF )3 .

C3 = 2
ln(vF /v∆ ) − 0.9601 (2.38)
π Nf
The logarithmic dependence on v∆ /vF arises from a singularity in the integrand at
x = 0 and θ = ±π/2: this corresponds to an integral across the underlying Fermi
surface over ω and kx where the inverse fermion propagator ∼ −iω+vF kx τ z . Inserting
these values into the flow equations (2.25), (2.26), and (2.27) we obtain an explicit
form of the RG flow for small v∆ /vF :
2
dv∆ 8 v∆
= − 2
ln (vF /v∆ ) − 0.4974 (2.39)
d` π Nf vF
dvF v∆
= −0.1148 (2.40)
d` Nf
(v∆ /vF )2

d(v∆ /vF ) 8
= − ln (vF /v ∆ ) − 0.6122 . (2.41)
d` π2 Nf
The right-hand-side of Eq. (2.41) has a zero for v∆ /vF ≈ 2; this is spurious as this
equation is valid only for v∆ /vF 1. The full expression for the right-hand-side,
valid for arbitrary v∆ /vF , is plotted in Fig. 2.5, and this makes it clear that the only
zero of the beta function is at v∆ /vF = 0.
The results of a numerical integration of Eqs. (2.25), (2.26) and (2.27) are shown
in Figs. 2.6, 2.7, and 2.8. We start the RG integration at ` = 0 corresponding to a
1.0
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1.0
Figure 2.6: Integration of Eqs. (2.25), (2.26) and (2.27) from = 0 to ` for Nf = 2,
starting from the velocities vF0 and v∆
0 0
. Results above are for v∆ /vF0 = 1.
temperature T = T0 at the nematic critical point, at which point the velocities have
the values vF0 and v∆
0
. Integrating to the scale ` yields the velocities vF (`) and v∆ (`)
at a temperature T = T0 e−` . The results depend upon the initial value of the velocity
0
ratio, v∆ /vF0 , and are shown in the figures for v∆
0
/vF0 = 1, 0.1, 0.05. These results can
also be converted into dependence on the deviation from the nematic critical point,
r, by integrating Eq. (2.23).
In the asymptotic low temperature regime (` → ∞), we have v∆ /vF → 0, and so
we can integrate the asymptotic expression in Eq. (2.41). This yields
v∆ π 2 Nf 1 1
= (2.42)
vF 8 ` ln[0.3809 `/Nf ]
The T dependence follows by using ` = ln(T0 /T ). Using the result for v∆ /vF in
10
1.0
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1.0
0
Figure 2.7: As in Fig. 2.6 but for v∆ /vF0 = 0.1.
Eq. (2.42) and integrating Eq. (2.40), we find that vF only has a finite renormalization
from its bare value as ` → ∞. So the decrease in v∆ /vF as ` → ∞ (i.e. as T → 0)
comes primarily from the decrease in v∆ . This is also clear from the plots in Figs. 2.6-
2.8.
We can also apply these results to obtain the T = 0 evolution of the velocities
as a function of the tuning parameter, r, across the nematic ordering transition. In
this case we use Eq. (2.23), along with the fact that ν → 1 as v∆ /vF → 0. Then
we deduce that ` = ln(1/|r|) to leading logarithm accuracy, and so obtain the r
dependence of the velocities from Figs. 2.6-2.8. When both T and r are non-zero, we
can use ` = ln(max(|r|, T )) to leading logarithmic accuracy. Note that this predicts
a minimum in v∆ /vF as r is tuned across the nematic ordering transition, with the
minimum value of order 1/ ln(1/T ).
20
1.0
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1.0
0
Figure 2.8: As in Fig. 2.6 but for v∆ /vF0 = 0.05.
2.4 Theory for small v∆/vF

The RG analysis in Section 2.3 has shown that the β functions at order 1/Nf are
such that the flow is towards the strong anisotropy limit, v∆ /vF → 0 at large scales.
Here we will show that this conclusion holds at all orders in the 1/Nf expansion.
Indeed, for small v∆ /vF we will show that v∆ /vF can itself be as the control parameter
for the computation, and Nf is not required to be large. In other words, the RG flow
equations in Eqs. (2.39-2.41) are asymptotically exact, even for the physical value of
Nf = 2. All corrections to Eq. (2.39-2.41) are higher order in v∆ /vF .
Our conclusions follow from an examination of the structure of the action Sφ
for the nematic order in Eq. (2.6) in the limit of v∆ /vF → 0. As we will shortly
verify, the natural scale for the fluctuations of φ are at frequency and momenta with
ω ∼ vF kx ∼ vF ky . Under these conditions, we can just take the v∆ → 0 limit of the
expressions in Eqs. (2.8), (2.10), and (2.12). This limit exists, and to quadratic order
in φ we have the following effective action as v∆ → 0 (at criticality, r = 0):

Z 2 Z q
Nf dk dω 2
q
Sφ = |φ(k, ω)| ω 2 + vF2 kx2 + ω 2 + vF2 ky2 + . . . (2.43)
16vF v∆ 4π 2 2π
Rescaling momenta k → k/vF , we notice that the prefactor in Eq. (2.43) is propor-
tional to the dimensionless number Nf vF /v∆ . Thus each propagator of φ will appear
with the small prefactor of v∆ /(Nf vF ). Further, as claimed above, each propogator
has the typical scales ω ∼ vF kx ∼ vF ky .
These considerations are easily extended to terms to all orders in φ in Eq.(2.6).
The full expression for Sφ involves the sum of the logarithms of the determinants of
the Dirac operators of Ψ1,2 fermions in a background φ field. The expansion of these
determinants in powers of φ involves only terms with a single fermion loop. For each
Ψ1 loop we rescale the fermion momentum px → px /vF and py → py /v∆ . Similarly for
each Ψ2 loop we rescale the fermion momentum px → px /v∆ and py → py /vF . In both
cases, the integral over the fermion loop momentum yields a prefactor of 1/(vF v∆ ).
For the φ vertices emerging from the Ψ1 loops, we have momenta ω ∼ vF kx ∼ vF ky , as
noted above; when this external momenta enters the fermion loop, we have kx ∼ px ,
the typical Ψ1 x-momentum. However, ky py ∼ 1/v∆ , the typical Ψ1 y-momentum.
Consequently, we may safely set ky = 0 within the fermion loop, and there is no
further dependence upon v∆ in the loop integral. Similarly, in the Ψ2 loops, we
may set kx = 0 within the fermion loop. To summarize, the typical φ momenta are
kx ∼ 1/vF , ky ∼ 1/vF , the typical Ψ1 momenta are px ∼ 1/vF , py ∼ 1/v∆ , and typical
Ψ2 momenta are px ∼ 1/v∆ , py ∼ 1/vF . Further, the resulting effective action for φ
has the form
Nf
Sφ = S φ(k, ω); vF kx , vF ky , ω + terms higher order in v∆ /vF . (2.44)
vF v∆
Here S is a functional of φ which is independent of v∆ , whose expansion in φ has co-
efficients which depend only upon vF kx , vF ky , and ω. It can be verified that Eq. (2.43)
and the quartic terms in Eqs. (2.10) and (2.12) are indeed of the form in Eq. (2.44).
From Eq. (2.44), it is evident (after rescaling momenta k → k/vF and momentum
space fields φ → vF2 φ) that the natural expansion parameter controlling φ fluctuations
is v∆ /(N vF ). This is the main result of this section.
It remains to understand the ln(vF /v∆ ) factor in Eq. (2.41). If we compute ob-
servable φ correlations under the reduced effective action S, the resulting expansion
in powers of v∆ /(N vF ) leads to Feynman integrals which are not necessarily infrared
or ultraviolet finite. Because the action S is free of any dimensionful coupling con-
stant, the Feynman graph divergences can at most be powers of logarithms. These
divergences are cutoff by using the full fermion propagators, including the terms pro-
portional to v∆ times the momenta of the boson propagators. This leads to factors
of ln(vF /v∆ ), as was the case in obtaining Eq. (2.41) from Eq. (2.35).
2.5 Conclusions
This paper has described the RG properties of the field theory (Vojta et al.,
2000c,b, 2008) in Eq. (2.5) for nematic ordering. It was previously noted (Kim
et al., 2008) that the effective potential for the nematic order, φ, remained unrenor-
malized upon integrating out the nodal fermions Ψ1,2 . This happens because the
φ couples to a conserved fermion current, and a spacetime-independent φ can be
‘gauged away’ applying a gauge transformation to the fermions. One consequence
of this non-renormalization is that the susceptibility exponent, γ, takes the simple
value in Eq. (2.14). However, non-trivial renormalizations of the field scale of both φ
and Ψ1,2 are still possible, along with renormalizations of the velocities, and we have
shown here that this leads to an interesting RG flow structure, with non-mean-field
exponents.
It is interesting to note here the similarities to the RG flow of supersymmetric field
theories (Strassler, 2003). There the effective potential also remains unrenormalized
(albeit for very different reasons), but the ‘wavefunction renormalizations’ lead to
many non-trivial RG fixed points.
Our main result was that the transition is described by a fixed point in which the
fermion velocities at the nodal points have a ratio which approaches a fixed point with
v∆ /vF → 0 logarithmically slowly. However, it is not valid to set the pairing-induced
velocity v∆ = 0 in the computation, and so deal with a metallic Fermi surface. For
2 2 1/2
the case where the fermion dispersion is ∼ (vF2 kx2 + v∆ ky ) , the typical fermion
momenta contributing to the critical theory scale as kx ∼ 1/vF and ky ∼ 1/v∆ , and
so the full form of the Bogoliubov quasi-particle dispersion is important. The flow of
the velocities is described by Eqs. (2.39-2.41) for v∆ /vF 1, and these equations are
believed to be asymptotically exact.
Unlike the fermions, the fluctuations of the nematic order, φ, are isotropic in
momenta. These are described asymptotically exactly by Eq. (2.43). Note that the
propagator is very different from free field, and has a large anomalous dimension, as
was found in the Nf = ∞ theory (Kim et al., 2008). Experimental detection of this
unusual spectrum in e.g. inelastic X-ray scattering would be most interesting.
We thank M. P. A. Fisher, E. Fradkin, E.-A. Kim, S. Kivelson, and M. Lawler for
useful discussions.
Chapter 3
Quantum criticality of the kagome

antiferromagnet
with Dzyaloshinskii-Moriya
interactions
3.1 Introduction
1
The nearest neighbor spin S = 2
antiferromagnet on the kagome lattice has been
the focus of extensive theoretical and experimental study because it is a prime can-
didate for realizing a ground state without antiferromagnetic order.
On the experimental side, much attention has focused on the S = 1/2 compound
herbertsmithite ZnCu3 (OH)6 Cl2 . It has J ≈ 170 K and no observed ordering or
structural distortion. (Helton et al., 2007; Ofer et al., 2006; Mendels et al., 2007;
Imai et al., 2008) However, there is an appreciable amount of substitutional disorder
between the Zn and Cu sites (believed to be of the order 5-10 %) which affects the low
T behavior. (Olariu et al., 2008; Gregor and Motrunich, 2008; Rozenberg and Chitra,
2008; Lee et al., 2007) More importantly, there is an upturn in the susceptibility at
T = 75 K which has been ascribed to the DM interactions (Rigol and Singh, 2007;
35
Chapter 3: Quantum criticality of the kagome antiferromagnet
Zorko et al., 2008; Ofer and Keren, 2009).

On the theoretical side, the most recent evidence (Nikolic and Senthil, 2003; Singh
and Huse, 2007; Jiang et al., 2008; Evenbly and Vidal, 2010; Poilblanc et al., 2010;
Schwandt et al., 2010) on the the nearest-neighbor antiferromagnet points consistently
to a ground state with a spin gap of 0.05J and valence bond solid (VBS) order. The
pattern of the VBS order is quite complex with a large unit cell, but was anticipated by
Marston and Zeng (Marston and Zeng, 1991) by an application of the VBS selection
mechanism described in the 1/N expansion of the SU(N ) antiferromagnet. (Read
and Sachdev, 1989a)
The influence of the DM interactions has also been studied theoretically (Cépas
et al., 2008; Tovar et al., 2009; Hermele et al., 2008). Starting with an “algebraic
spin liquid” ground state, Hermele et al. (Hermele et al., 2008) argued that the DM
coupling, D, was a relevant perturbation, implying that an infinitesimal D would
induce long-range magnetic order. In a recent exact diagonalization study, Cepas et
al. (Cépas et al., 2008) reach a different conclusion: they claim that there is a non-
zero critical DM coupling Dc beyond which magnetic order is induced. They estimate
Dc /J ≈ 0.1, quite close to the value measured (Zorko et al., 2008) for ZnCu3 (OH)6 Cl2
which has D/J ≈ 0.08. This proximity led Cepas et al. to suggest that the quantum
criticality of the DM-induced transition to magnetic order controls the observable
properties of this kagome antiferromagent.
The purpose of this paper is to propose a theory for the quantum critical point
discovered by Cepas et al. (Cépas et al., 2008). We will compute various observables
of this theory, allowing a potential comparison with numerics and experiments.
Given the evidence for VBS order in the model without DM interactions (Nikolic
and Senthil, 2003; Singh and Huse, 2007; Jiang et al., 2008; Evenbly and Vidal,
2010), it would appear we need a theory for the transition from the VBS state to
the magnetically ordered state. However, the VBS ordering is very weak, and can
reasonably be viewed as a small perturbation on some underlying spin-liquid ground
state. Schwandt, Mambrini, and Poilblanc (Poilblanc et al., 2010; Schwandt et al.,
2010) have recently presented evidence that the kagomé antiferromagnet is proximate
to a Z2 spin liquid state, and that vison condensation in this state leads to weak VBS
ordering. Their dimer representation leads naturally to Z2 spin liquid states in the
same class as that originally described (Sachdev, 1992; Wang and Vishwanath, 2006)
by the Schwinger boson method (Arovas and Auerbach, 1988; Auerbach and Arovas,
1988; Sachdev, 2010). We will therefore neglect the complexities associated with the
VBS ordering and work with the parent Z2 spin liquid state. This is equivalent to
ignoring the physics of the vison sector, and assumes the magnetic ordering transition
can be described in a theory of the spinons alone (Xu and Sachdev, 2009). The
main result of this paper will be a theory of the quantum phase transition from the
Schwinger boson Z2 spin liquid to the magnetically ordered state as induced by the
DM interactions.
We will begin in Section 3.2 with a description of the mean-field theory of the
Z2 spin liquid and its transition to the magnetically ordered state in the presence of
DM interactions. This will be carried using the Sp(N ) Schwinger boson formulation
(Sachdev, 1992, 2010), for which the mean-field theory becomes exact in the large N
limit. We will turn to fluctuation corrections and the nature of the quantum critical
point in Section 3.3. Here we will show that the critical theory is the familiar three
dimensional XY model. However, its connection to experimental observables is subtle:
in particular, the XY field itself is not directly observable.
While this paper was in preparation, a description of the Schwinger boson mean
field theory in the presence of DM interactions also appeared in Ref. Messio et al.
(2010); they consider mean-field solutions with larger unit cells than we do, but did
not analyze the critical field theory. Where they overlap, our results are in agree-
ment with theirs. We also note the recent experimental observations of Helton et al.
(Helton et al., 2010), who present evidence for quantum criticality in ZnCu3 (OH)6 Cl2 .
Dz =D
Figure 3.1: Staggered DM interaction from triangle to triangle in z-direction as in

Ref. Elhajal et al. (2002). The arrowheads indicate D to come out of the plane,
whereas the tails denote D to go into the plane. Note that the triangles are summed
clock- and anticlockwise, respectively, indicated by the arrows on the bonds.
3.2 Mean field theory

The model we consider is a standard Heisenberg Hamiltonian supplemented by an
additional DM interaction. It assumes the form
1X
H= [Jij Si · Sj + Dij · (Si × Sj )] . (3.1)
2 i,j
Si in this notation denotes the spin operator at site i, Jij is assumed to be uniform and
of the nearest neighbor type, and Dij = Dij ez is taken along the z-axis and staggered
from triangle to triangle (Elhajal et al., 2002), see Fig. 3.1. This additional term
explicitly breaks the spin-rotation symmetry by favoring configurations lying in the x-
y plane. Furthermore this term increases the tendency of classical spin ordering. It has
been shown in earlier works (Arovas and Auerbach, 1988; Auerbach and Arovas, 1988;
Sachdev, 2010) that Schwinger bosons are ideally suited to describe phase transitions
between paramagnetic and magnetically ordered phases in spin models. Following
Ref. Sachdev (1992) we introduce a Sp(N ) generalization of the spin operators, which
formally allows to consider a controlled large N limit particularly suited for the study
of frustrated spin systems such as kagome or traingular antiferromagnets. In the
Sp(1) case, which is isomorphic to SU(2), one can represent the spin variables as
τσσ0
Si = b∗iσ b 0 (3.2)
2 iσ
with ~τ being the Pauli matrices and with
!
bi↑ X
biσ = and b∗iσ biσ = 2S = nb . (3.3)
∗
bi↓ σ
In the case of the large-N generalization the Schwinger bosons acquire another index
counting the copies of the system (we drop this index in the following discussions, but
display N whenever it is essential). The large-N generalization formally justifies the
mean-field, with the saddle point becoming exact in the limit N → ∞. In order to
properly reformulate the problem at hand we introduce two decoupling parameters
X
Qij = σσ0 biσ bjσ0
σσ 0
X
x
Pij = τσσ 0 biσ bjσ 0 (3.4)
σσ 0
where σσ0 is the antisymmetric tensor and τ x is just the standard Pauli matrix. We
see from the above expressions that Qij = −Qji whereas Pij = Pji . This implies that
the bond variables Pij do not have a direction.
The constraint Eq. (3.3) is implemented via a Lagrangian multiplier in a standard
way. The Hamiltonian of the system formulated in the fields defined in Eq. (3.4)
consequently reads
H 1X iX
= − Jij Q∗ij Qij − Dij (Pij∗ Qij − Q∗ij Pij )
N 2 i,j 4 i,j
X
+ λi (b∗iσ biσ − κ) (3.5)
i
where κ = nb /N . We furthermore introduce Nu as the number of unit cells in the

systems and Ns as the number of sites within the unit cell. We can write the mean-
field Hamiltonian per flavor and unit cell as

HM F 1 X ∗
= Ψ (k)H(qij , pij , λ, k)Ψ(k)
N Nu Nu k
JX iD X ∗
+ |qij |2 + (pij qij − qij∗ pij )
2 0
4 0
(ij) (ij)
− Ns λ(1 + κ) (3.6)
P
where (ij)0 denotes the sum over bonds belonging to the unit cell,
Ψ∗ (k) = (b∗1↑ (k), ..., b∗Ns ↑ (k), b1↓ (−k), ..., bNs ↓ (−k)) , (3.7)
and the matrix

!
λI C∗ (k)
H= (3.8)
C(k) λI
with the matrices I (identity) and C(k) being Ns × Ns matrices; the explicit form
of these matrices is given in Appendix 3.A. As mentioned before, one of the major
assets of the Schwinger boson approach is that it can describe magnetically disordered
gapped spin liquid phases as well as magnetically ordered states. On a formal level
in the large N approach this is achieved by introducing the following notation for the
Schwinger bosons
√
biσ = ( N xi , bm̃
i ) where m̃ = 2, ..., N . (3.9)
The first component is thus a classical field. If x 6= 0 it signals condensation which

causes long range order to appear. Following Ref. Arovas and Auerbach (1988);
Auerbach and Arovas (1988); Sachdev (1992) we can integrate out the Schwinger
bosons and the zero-temperature mean field energy assumes the following form
EM F 1 X X
= ωµ (k) − Ns λ(1 + κ) + λ |xiσ |2
N Nu Nu k,µ=1,..,N i0 σ
s
J X
|qij |2 − qij∗ σσ0 xiσ xjσ0 ) + h.c.

+
2 0 (ij)
iD X ∗
+ (pij qij − qij∗ pij )
4 0
(ij)
iD X ∗ x ∗ ∗
− (pij σσ0 xiσ xjσ0 + qij τσσ 0 xiσ xjσ 0 )
4 0 (ij)
iD X
+ (pij σσ0 x∗iσ x∗jσ0 + qij∗ τσσ
x
0 xiσ xjσ 0 ) (3.10)
4 0 (ij)
P
where i0 denotes the sum over all sites within one unit cell. In the following we
solve the self-consistency equations according to
κ = hb∗iσ biσ iM F
qij = hQij iM F ,
pij = hPij iM F . (3.11)
with the Hamiltonian defined in Eq. (3.6).

Our solution of the mean-field equations follows previous work (Sachdev, 1992;
Wang and Vishwanath, 2006), which classified physically different Z2 spin liquid so-
lutions without the DM interactions. We found that these solutions have a natural
generalization in the presence of DM terms, with values of the pij which reflect the
symmetries of the qij . Two stable solutions were found in previous work, with only
two possibly distinct values of qij as illustrated in Fig. 3.2. Including the DM inter-
actions, these solutions extended to
(i ) q1 = q2 real, p1 = p2 pure imaginary: upon increasing κ, the Schwinger bosons
condense at k = 0, with the spins at angles of 120◦ to each other within the unit cell.
This states is therefore called the k = 0 state.
(ii ) q1 = −q2 real, p1 = −p2 pure imaginary: upon increasing κ, the Schwinger bosons
√ √
condense at wavevector k = ±(2π/3, 0) into a state which is called the 3 × 3 an-
tiferromagnet, characterized by an enlarged unit cell.
v
w
Figure 3.2: We take a unit cell of three sites, labeled u, v, w. The arrows indicate the
values of the oriented variables qij . The links with single arrows have qij = q1 , while
those with double arrows have qij = q2 . The Pij are unoriented, and take values p1
and p2 on these links respectively.
Solutions with larger unit cells can be present with additional frustrating interactions
(Wang and Vishwanath, 2006), but we will not consider them here.
3.2.1 Phase diagram

Our phase diagram is shown in Fig. 3.2.1 as a function of κ = nb /N (which
corresponds to the spin size) and the parameter D/J. Our phase diagram is similar
to that obtained recently by Messio et al. (Messio et al., 2010). They also considered
solutions with a larger unit cell which were stable over some portion of the phase
diagram.
We start with a discussion of the classical limit. While for D = 0 the long-
√ √
range ordered state of the 3 × 3 type is generically preferred (Sachdev, 1992),
infinitesimal D favors the so-called k = 0 state. This is reproduced by our mean-field
equations in the large spin limit κ → ∞. For finite values of κ there is a small slab in
√ √
which long-range order of the 3 × 3 type is favored over the k = 0. These states
are separated by a first order transition driven by the ratio D/J.
A similar behavior appears for the corresponding spin liquid states at small κ.
The q1 = −q2 is favored at small D/J, and then undergoes a first order transition to
D!J
1.2
1.0
0.8
LRO
0.6 k=0
Z2 spin liquid
0.4 q1 = q2
0.2
LRO
√ √
3× 3
Z2 spin liquid q1 = −q2
0.2 0.3 0.4 0.5 0.6 0.7 Κ
Figure 3.3: Phase diagram in the N → ∞ limit of the Sp(N ) theory. The x-axis
shows κ = 2S/N and the y-axis D/J. The phases have long-range magnetic order
type (LRO), or gapped Z2 spin liquids. The thick line is a first-order transition, while
the thin lines are second-order transitions: the transition to the k = 0 LRO state is
described in Section 3.3. The limit of D/J → 0 reduces to the results presented in
Ref. Sachdev (1992).
the q1 = q2 state at large D/J.

1
In exact diagonalization studies of the spin S = 2
kagome antiferromagnet with
DM interactions a second order phase transition between a phase with short ranged
k = 0 correlations (Jiang et al., 2008; Laeuchli and Lhuillier, 2009) (obtained in the
pure Heisenberg case) and phase with k = 0 long range order was found (Messio
et al., 2010). Such a transition is also present in our mean-field theory, which therefore
can be used for a study of critical properties in Section 3.3.
3.3 Quantum criticality

We will consider only the transition out of the q1 = q2 spin liquid, because that is
what is seen in the numerical studies (Cépas et al., 2008). The corresponding transi-
tion out of the q1 = −q2 spin liquid can be treated in a similar manner. Throughout
this section we will consider the physical SU(2) antiferromagnet directly, and not take
the large N limit. The method followed below has been reviewed in a more general
context in Ref. Sachdev (2010).
Since we are at k = 0, we can write the effective action for the bosons by making
the small momentum expansion of the matrix in Eq. (3.34). We take 3 sites, u, v, w,
in each unit cell (see Fig. 3.2), and then take the continuum limit of the saddle-point
Lagrangian. We write the boson operators on these sites as buσ = Uσ , bvσ = Vσ etc.,
and set q1 = q2 = q, and p1 = p2 = ip with q and p real. Then, we write the final
Lagrangian in the form
L = LH + LDM (3.12)
representing the contributions of the Heisenberg exchange and the DM coupling,

respectively.
From Eq. (3.34) we obtain the Lagrangian in the absence of a DM term (which
describes the k = 0 solution of Ref. Sachdev (1992)):
∂Uσ ∂Vσ ∂Wσ
LH = Uσ∗ + Vσ∗ + Wσ∗
∂τ ∂τ ∂τ
+ λ |Uσ | + |Vσ | + |Wσ |2
2 2
− Jqσσ0 (Uσ Vσ0 + Vσ Wσ0 + Wσ Uσ0 ) + c.c.

Jq
+ σσ0 (∂1 Uσ ∂1 Vσ0 + ∂2 Vσ ∂2 Wσ0 + ∂3 Wσ ∂3 Uσ0 )
2
+ c.c. (3.13)
where ∂i is the gradient along the direction ei in Eq. (3.33).

We now perform a unitary transformation to new variables Xσ , Yσ , Zσ . These are
chosen to diagonalize only the non-gradient terms in L.

     
Uσ 1 i
∗ 
 
 Vσ  = √ Zσ  ζ  + σσ0 Z
 σ0 

√ iζ 2 
  6  6 
2
Wσ ζ iζ
   
1 −i
Yσ   Yσ∗0  
+ √  ζ 
 + σσ 0 √  −iζ 2 
6  6 
2
ζ −iζ
 
1
Xσ  
+√  1 . (3.14)
3 
1
where ζ ≡ e2πi/3 . The tensor structure above makes it clear that this transformation
is rotationally invariant, and that Xσ , Yσ , Zσ transform as spinors under SU(2) spin
rotations. Inserting Eq. (3.14) into LH we find
∂Xσ ∂Zσ ∂Yσ √
LH = Xσ∗ + Yσ∗ + Zσ∗ + (λ + 3Jq)|Zσ |2
∂τ √ ∂τ ∂τ
+ (λ − 3Jq)|Yσ | + λ|Xσ |2
2
√
Jq 3
|∂x Zσ |2 + |∂y Zσ |2 + . . .

+ (3.15)
2
The ellipses indicate omitted terms involving spatial gradients in the Xσ and Yσ which
we will not keep track of. This is because the fields Yσ and Xσ are massive relative to
Zσ (for q < 0 which is the case in our mean-field solution), and so can be integrated
out. This yields the effective Lagrangian
√
1 Jq 3
LZH 2
|∂x Zσ |2 + |∂y Zσ |2

= √ |∂τ Zσ | +
(λ − 3Jq) 2
√ 2
+ (λ + 3Jq)|Zσ | + . . . (3.16)
Note that the omitted spatial gradient terms in Xσ , Yσ do contribute a correction

to the spatial gradient term in Eq. (3.16), and we have not accounted for this. This
Lagrangian shows that the mean-field theory has a transition to magnetic order at
√
λ = | 3Jq|, which agrees with earlier results (Sachdev, 1992).
The effective Lagrangian LZH is almost the complete solution for the critical theory
in the system without the DM interactions. However, we also need higher order terms
in Eq. (3.16), which will arise from including the fluctuations of the gapped fields Q
and λ. Rather than computing these from the microscopic Lagrangian, it is more
efficient to deduce their structure from symmetry considerations. The representation
in Eq. (3.14), and the connection of the U , V , W to the lattice degrees of freedom,
allow us to deduce the following symmetry transformations of the X, Y , Z:
• Under a global spin rotation by the SU(2) matrix gσσ0 , we have Zσ → gσσ0 Zσ0 ,
and similarly for Y , and Z. When DM interactions are included, the global
symmetry is reduced to U(1) rotations about the z axis, under which
Z↑ → eiθ Z↑ , Z↓ → e−iθ Z↓
Y↑ → eiθ Y↑ , Y↓ → e−iθ Y↓
X↑ → eiθ X↑ , X↓ → e−iθ X↓ . (3.17)
• Under a 120◦ lattice rotation, we have Uσ → Vσ , Vσ → Wσ , Wσ → Uσ . From

(3.14), we see that this symmetry is realized by
Zσ → ζZσ , Yσ → ζYσ , Xσ → Xσ . (3.18)
Note that this is distinct from the SU(2) rotation because det(ζ) 6= 1.
• Under time-reversal, we have Uσ → σσ0 Uσ∗0 , and similarly for Vσ , Wσ . This is

realized in Eq. (3.14) by
Zσ → iZσ , Yσ → −iYσ , Xσ → σσ0 Xσ∗0 . (3.19)
In particular, note that Z↑ does not map to Z↓ under time-reversal.
These symmetry operators make it clear that the only allowed quartic term for the
2
Heisenberg Hamiltonian is ( σ |Zσ |2 ) : this implies that the Z2 spin liquid to anti-
P
ferromagnetic order transition of this model is in the universality class of the O(4)
model (Chubukov et al., 1994b; Chubukova et al., 1994).
Let us now include the DM interactions. From Eq. (3.34), we see that
Dq x
LDM = i τ 0 (Uσ Vσ0 + Vσ Wσ0 + Wσ Uσ0 ) + c.c.. (3.20)
2 σσ
We have dropped a term proportional to Dp which has the same structure as the
terms in Eq. (3.15), and ignored spatial gradients. In terms of the fields in Eq. (3.14),
this takes the simple form
Dq x
LDM = iτσσ0 Xσ Xσ0 + c.c.
2
− Zσ∗ τσσ
z ∗ z
0 Zσ 0 + Yσ τσσ 0 Yσ 0 , (3.21)
and it can be verified that these terms are invariant under Eqs. (3.17,3.18,3.19). As
before, we now integrate out Xσ and Yσ from LH + LDM in Eqs. (3.15) and (3.21).
We obtain a Lagrangian with the same structure as Eq. (3.16), but all couplings
become dependent upon σ; in other words, we have 2 separate XY models for Z↑ and
Z↓ . Performing a careful analysis of allowed higher order terms as restricted by the
symmetry constraints discussed above, and after appropriate rescalings of the spatial,
temporal, and field scales, we obtain the field theory with the Lagrangian
LZ = |∂τ Z↑ |2 + |∇Z↑ |2 + s↑ |Z↑ |2 + u↑ |Z↑ |4

+ |∂τ Z↓ |2 + |∇Z↓ |2 + s↓ |Z↓ |2 + u↓ |Z↓ |4
+ v|Z↑ |2 |Z↓ |2 + w (Z↑ Z↓ )6 + (Z↑∗ Z↓∗ )6 .

(3.22)
Note that s↑ 6= s↓ in general (and similarly for u↑,↓ etc.), and equality is not required
by the time-reversal symmetry in Eq (3.19). Time-reversal symmetry does prohibit
a term ∼ (Z↑ Z↓ )3 which is allowed by the other symmetries. Thus we expect only
one of Z↑ or Z↓ to condense at the quantum critical point: as we will see from the
analysis of observables in Section 3.3.1, this transition does indeed correspond to the
development of spiral magnetic order in the x-y plane. The choice between Z↑ and
Z↓ is controlled by the sign of D.
Eq. (3.22) also contains terms which couple the two XY models to each other. The
lowest allowed term, v, couples the energy densities and does not have any important
effects. More interesting is the w term, which shows that the global symmetry is not
O(2)⊗O(2) but O(2)⊗Z3 . In the magnetically ordered phase with hZ↑ i 6= 0 (say),
this term will induce a small cubic ordering field ∼ Z↓6 in the XY model for Z↓ .
However, the action for Z↓ has a ‘mass’ term s↓ with a positive co-efficient, and this
sixth order term will not immediately induce ordering in Z↓ ; i.e. a magnetic phase
with hZ↑ i =
6 0 and hZ↓ i = 0 has a finite range of stability. Thus close to the transition
we can neglect the Z↓ field entirely, and transition is in the universality class of the
three-dimensional XY model.
The choice above of Z↑ over Z↓ gives the incorrect appearance that we are breaking
the spin reflection symmetry Sz → −Sz of H, suggesting the appearance of a net z
ferromagnetic moment. However, notice that the theory of Z↑ is relativistic, and
so contains both spinons and anti-spinons which carry Sz = +1/2 and Sz = −1/2
respectively. The spinon of Z↓ also carries Sz = −1/2, and this is degenerate with the
anti-spinon of Z↑ in the O(4) invariant theory in Eq. (3.16). It is this latter degeneracy
which is lifted by the DM interactions, which induce a vector spin chirality along the
z direction (Isakov et al., 2005) (as we will see below). We will also see there is no
net ferromagnetic moment, because time-reversal symmetry is preserved.
3.3.1 Observables
To determine the operators corresponding to the ferromagnetic moment, let us
couple a uniform external field h to the lattice Hamiltonian. This adds to the con-
tinuum Lagrangian the term
Lh = −h · τσσ0 (Uσ∗ Uσ0 + Vσ∗ Vσ0 + Wσ∗ Wσ0 ) (3.23)
Inserting the parameterization in Eq. (3.14) this becomes
Lh = −h · τσσ0 (Xσ∗ Xσ0 + Yσ∗ Zσ0 + Zσ∗ Yσ0 ) (3.24)
We now need to integrate out Xσ and Yσ in the Lagrangian LH +LDM +Lh defined by
the sum of Eqs. (3.15), (3.21), and (3.24), and collect the terms linear in h. Without
the DM coupling, we obtain
∂Zσ∗

∗ ∂Zσ 0
∼ h · τσσ0 Zσ − Zσ 0 (3.25)
∂τ ∂τ
Comparing with LZH in Eq. (3.16) we see that this is just the coupling to the conserved
SU(2) charges of the O(4) model: this is the usual term which determines the magnetic
susceptibility of the Heisenberg antiferromagnet (Chubukov et al., 1994b; Chubukova
et al., 1994). Upon including the effects of LDM we find that the essential structure
of Eq. (3.25) does not change: the τ matrices get multiplied by some σ-dependent
factors τσσ0 → fσ τσσ0 fσ0 which do not modify the scaling considerations. No term
with a new structure is generated by the DM coupling. It can now be seen that these
expressions have vanishing expectation values under LZ in Eq. (3.22), and so there is
no net ferromagnetic moment in the absence of an external field.
We now turn to the antiferromagnetic order parameter; for a coplanar antiferro-
magnet, this is described by
Si ∝ N1 cos(Q · ri ) + N2 sin(Q · ri ), (3.26)
where N1,2 are 2 orthogonal vectors representing the spiral order, and Q is wavevector
at which the spin structure factor is peaked. For our model, we can see that
N1 + iN2 = Su + ζSv + ζ 2 Sw . (3.27)
Using Eq. (3.14), and keeping only the lowest order term, we therefore obtain (Chubukov
et al., 1994b; Chubukova et al., 1994)
 
i(Z↑2 − Z↓2 )/2
 
N1 + iN2 =  2 2
 −(Z ↑ + Z ↓ )/2 ;
 (3.28)
−iZ↑ Z↓
in a notation that makes the rotational invariance evident, this relationship is

i
N1 + iN2 = αβ τβσ Zσ Zα . (3.29)
2
Note that a phase with hZ↑ i =
6 0 and hZ↓ i = 0 has spiral order in the x-y plane.
To complete the list of operators which are quadratic in the Zσ , we consider the
vector spin chirality (Isakov et al., 2005). This is defined here by
Su × Sv + Sv × Sw + Sw × Su . (3.30)
Using Eq. (3.14) we find that the leading operator mapping to vector spin chirality
is (dropping an overall factor of |Z↑ |2 + |Z↓ |2 )
Zσ∗ τσσ0 Zσ0 . (3.31)
Note that in the presence of the DM term, the couplings in the effective theory (3.22)
imply that the z component of the vector spin chirality is always non-zero.
3.3.2 Critical properties

Let us assume the transition to magnetic order proceeds via the condensation
of Z↑ . The transition is in the XY universality class, and the dimension of the
antiferromagnetic order parameter is
1+η
dim[N1 ] = dim[N2 ] = dim[Z↑2 ] = . (3.32)
2
The value of the exponent η can be read off from results for the three-dimensional XY
model (Campostrini et al., 2001; Calabrese et al., 2002), and we obtain η ≈ 1.474.
The antiferromagnetic susceptibility will diverge at the critical point as T −(2−η) . We
note the recent work of Ref. Grover and Senthil (2009) in a different context, which
also considered a model with an XY critical point at which the physically measurable
magnetic order was the square of the XY field.
The behavior of the uniform magnetic susceptibility follows from the scaling di-
mension of the operators in Eq. (3.25). For h along the z direction, the magnetization
is the just the conserved U(1) charge of the XY model: so (Chubukov et al., 1994a)
it has scaling dimension 2, and the susceptibility ∼ T . For h along the x or y direc-
tions, we have to integrate out Z↓ , and then the lowest dimension operator coupling
to the square of the field is |Z↑ |2 . This means that the susceptibility only has a weak
singularity at the quantum critical point given by that in h|Z↑ |2 i: at the quantum
critical point, there is a non-analytic term ∼ T 3−1/ν , above an analytic background.
3.4 Conclusion
We have presented a theory for the quantum critical point between a Z2 spin liquid
and an ordered antiferromagnet for the kagomé antiferromagnet in the presence of DM
interactions. The critical theory is just the three dimensional XY model. However,
the XY order parameter carries a Z2 gauge charge, and so it is not directly observable.
In particular, the antiferromagnetic order parameter is the square of the XY order
parameter. Specifically, the theory is given by LZ in Eq. (3.22), and its observables
are described in Section 3.3.1.
It is interesting to compare our results with recent observations of quantum critical
scaling in ZnCu3 (OH)6 Cl2 by Helton et al. (Helton et al., 2010). Their neutron scat-
tering measurements show an antiferromagnetic susceptibility which scales as T −0.66 .
This is actually in reasonable agreement with our theory, which has a susceptibility
∼ T −0.526 . However, they also observe a similar divergence in measurements of the
uniform magnetization, while our theory only predicts a very weak singularity. We
suspect that this difference is due to the present of impurities (Olariu et al., 2008;
Gregor and Motrunich, 2008; Rozenberg and Chitra, 2008; Lee et al., 2007), which
can mix the uniform and staggered components. A complete study of impurities near
the quantum critical point described above is clearly called for.
We thank L. Messio, O. Cépas, C. Lhuillier, and E. Vicari for useful discussions.
3.A Microscopic form of the Hamiltonian

We here give explicit expressions for the Hamiltonian introduced in Sec. 3.2. We
introduce the following set of unit vectors
√ !
1 3
e1 = a ,
2 2
√ !
1 3
e2 = a ,−
2 2
e3 = a(−1, 0) (3.33)
which allows to express ki = k·ei . In the following we concentrate on the states with 3
sites per unit cell. In that case the vector Ψ introduced in Eq. (3.7) has 6 components
and the matrix C consequently is a 3 × 3 matrix with the following entries:
J ∗ ik1 J ∗ −ik1 iD ∗ ik1 iD ∗ ik1 iD ∗ −ik1 iD ∗ −ik1
Cuv = q e + q2 e + pe + q e + pe + q e
2 1 2 4 1 4 1 4 2 4 2
J J iD ∗ −ik3 iD ∗ −ik3 iD ∗ ik3 iD ∗ ik3
Cuw = − q1∗ e−ik3 − q2∗ eik3 − pe + q e − pe + q e
2 2 4 1 4 1 4 2 4 2
Cvw = q e + q2 e + pe + q e + pe + q e
2 1 2 4 1 4 1 4 2 4 2
Cvu = − q1∗ e−ik1 − q2∗ eik1 − pe + q e − pe + q e
2 2 4 1 4 1 4 2 4 2
Cwv = − q1∗ e−ik2 − q2∗ eik2 − pe + q e − pe + q e
2 2 4 1 4 1 4 2 4 2
Cwu = q e + q2 e + pe + q e + pe + q e .
2 1 2 4 1 4 1 4 2 4 2
(3.34)
3.B Dispersion of the lowest excited spinon state

For q1 = q2 and D = 0, the ground state is doubly degenerate. The degeneracy
splits as D moves away from 0. In all cases, the minimum excitation energy occurs
at k = 0.
Fig. 3.4 plots the momentum dependence of the energy of the lowest excited spinon
at D/J = 0.3, κ = 0.2 and D/J = 0.3, κ = 0.4. In the former case, there is a finite
energy gap at k = 0 making it a gapped spin liquid. On the other hand, with a larger
κ value, it shows long-range magnetic order. The energy gap at k = 0 is closed and
condensation occurs at this wave vector.
For q1 = −q2 , there is a unique ground state even for D = 0. Energy minima is
at k = ±(2π/3, 0). Fig. 3.5 shows the dispersion of the lowest lying state of the spin
liquid with D/J = 0.03, κ = 0.55 and a case with long range ordering with a higher
D/J.
Figure 3.4: Left : Momentum dependence of the energy ω(k) of the lowest excited
spinon state of the kagome-lattice quantum antiferromagnet for the q1 = q2 state at
D/J = 0.3, κ = 0.2. The minimum excitation energy is at k = 0 and has a finite
energy gap. Right : Dispersion of the lowest excited spinon state for the q1 = q2 state
at D/J = 0.3, κ = 0.4. The energy gap closes at k = 0 and condensation occurs.
Figure 3.5: Left: disordered q1 = −q2 , D/J = 0.05, κ = 0.3 Right : ordered q1 =
−q2 , D/J = 0.03, κ = 0.55
3.C Condensation
For the q1 = q2 state Eq. (3.14) leads to the following parametrization of the
condensation of Z↓ field at k = 0.
! !
x↑u i
= `
x↓u 1
! !
x↑v iζ 2
= `
x↓v ζ
! !
x↑w iζ
= ` (3.35)
x↓w ζ2
where ` is the size of the condensate. Condensation of Z↑ can be written similarly.

For D > 0, Z↓ field condensation is energetically favored while the opposite is true
for D < 0. The two condensations are degenerate for D = 0.
For the q1 = −q2 state, condensation occurs at k̃ ≡ ~k = (2π/3, 0) or ~k = −k̃.
The two states have identical energies and the condensation is spontaneously chosen.
Similar analysis to Eqns. (3.13) and (3.14) gives the eigenvectors corresponding to the
lowest lying state. For ~k = k̃ this is (i, −i, i, 1, −1, 1) while for ~k = −k̃ the correspond-
ing eigenvector is (−i, i, −i, 1, −1, 1). Therefore condensations can be parametrized
as
! !
xuk↑ i
= `
xu−k↓ 1
! !
xvk↑ −i
= `
xv−k↓ −1
! !
xw
k↑ i
= ` (3.36)
xw
−k↓ 1
for ~k = k̃, and

! !
xuk↑ −i
= `
xu−k↓ 1
! !
xvk↑ i
= `
xv−k↓ −1
! !
xw
k↑ −i
= ` (3.37)
xw
−k↓ 1
for ~k = −k̃.
Chapter 4
Vison states and confinement

transitions of Z2 spin liquids on the
kagome lattice
4.1 Introduction
The recent DMRG study of Yan et al. (Yan et al., 2011) has provided striking
evidence for a spin liquid ground state for the S = 1/2 Heisenberg antiferromagnet on
the kagome lattice. Yan et al. found a gap to all excitations, and it is plausible that
their ground state realizes a Z2 spin liquid. (Read and Sachdev, 1991; Wen, 1991;
Sachdev, 1992; Wang and Vishwanath, 2009; Tay and Motrunich, 2011; Lu et al.,
2011; Iqbal et al., 2011b,a)
The results of Yan et al. also indicate the presence of proximate valence bond solid
(VBS) states in which the space group symmetry of the kagome lattice is broken, and
the fractionalized excitations of the spin liquid are confined into integer spin states.
The confinement quantum phase transition should be accessible in extended models
with further neighbor exchange interactions, and numerical studies of such transitions
can serve as a valuable probe of characteristics of the spin liquid.
This paper shall classify elementary vortex excitations of the Z2 spin liquid, carry-
56
Chapter 4: Vison states and confinement transitions of Z2 spin liquids on the
kagome lattice 57
Figure 4.1: Visualization of two different valence bond solid states as dimer coverings
of the kagome lattice. Each dimer represents a frustrated bond in the ordered phase of
the corresponding Ising model on the dual dice lattice. The left pattern, representing
the VBS 1F phase (see text), is a hardcore dimer covering with a 12-site unit cell that
maximizes the number of perfectly flippable diamonds (highlighted in gray). Note
however, that the choice of diamonds is not unique because the dimer pattern is
symmetric with respect to π/3-rotations around the hexagons marked by a black dot.
The pattern on the right represents the VBS 1A phase (see text), has a 36-site unit
cell and maximizes the number of perfectly flippable hexagons (highlighted in gray).
kagome lattice 58
ing Z2 magnetic flux (Read and Chakraborty, 1989; Read and Sachdev, 1991; Senthil
and Fisher, 2000), often called ‘visons’, which are analogous to the Abrikosov vor-
tices of BCS superconductors (Sachdev and Read, 1991) (after electromagnetism is
replaced by strong coupling to a compact U(1) gauge theory). We will compute their
projective symmetry group (Wen, 2002) (PSG), and their spectrum using an effec-
tive frustrated Ising model. (Jalabert and Sachdev, 1991) Note that the Ising ‘spin’
has nothing to do with the S = 1/2 spin of the underlying antiferromagnet, and
it is instead the creation or annihliation operator of the vortex excitation which is
centered on sites of a lattice dual to that of the antiferromagnet. For the kagome
antiferromagnet, the Ising model resides on the dice lattice, and the simplest effec-
tive Ising model has a degenerate momentum-independent spectrum. (Nikolic and
Senthil, 2003; Sengupta et al., 2006) We shall show how the PSG constraints allow
a systematic analysis of further neighbor interactions in the effective Ising model.
Such extended interactions must generically be present, (Xu and Balents, 2011) and
they lead to well-defined vison states with a finite effective mass. Depending upon
the values of these effective interactions, we find 2 possibilities for the vison states:
the simplest case has them transforming under the 48 element group GL(2, Z3 ), the
group of 2 × 2 matrices with non-zero determinant whose matrix elements belong
to the field Z3 . The more complex case has visons states of the 288 element group
GL(2, Z3 ) × D3 .
Armed with this description of the vison states, will propose quantum field the-
ories for the confinement transitions of Z2 spin liquids on the kagome lattice. These
transitions are associated with condensation of visons, (Senthil and Fisher, 2000; Jal-
abert and Sachdev, 1991; Sachdev and Vojta, 2000; Moessner and Sondhi, 2001b,a;
Xu and Sachdev, 2009; Xu and Balents, 2011; Poilblanc et al., 2010; Schwandt et al.,
2010) and are expressed in terms of a multi-component ‘relativistic’ scalar field; the
field theory with GL(2, Z3 ) symmetry appears in Eq. (4.22). These field theories also
place constraints on the specific patterns of spatial broken symmetry in the confining
VBS states found next to the quantum critical point and we will present phase dia-
grams illustrating these states. Visualizations of two possible VBS states are shown
kagome lattice 59
in Fig. 4.1. The left VBS pattern in Fig. 4.1 appears for the simplest case of the
GL(2, Z3 ) visons, and it is interesting that it is closely related to the “diamond pat-
tern” which is enhanced in the numerical studies of Yan et al. (Yan et al., 2011). The
right pattern in Fig. 4.1 is one of many possible VBS states for the GL(2, Z3 ) × D3
visons, and maps to the “honeycomb” VBS states found in earlier studies (Marston
and Zeng, 1991; Singh and Huse, 2007; Jiang et al., 2008; Evenbly and Vidal, 2010;
Nikolic and Senthil, 2003).
We will begin in Section 4.2 with a review of the basic characteristics of Z2 spin
liquids and of their vison excitations. A key property of a vison is that it picks up
a Aharanov-Bohm phase of π upon encircling every S = 1/2 spin on the sites of the
antiferromagnet; (Jalabert and Sachdev, 1991; Senthil and Fisher, 2000; Sachdev and
Vojta, 2000; Moessner and Sondhi, 2001b), as will be described in Section 4.2.
Section 4.3 contains our main new results. We begin with general effective theories
of vison motion which incorporate the Aharanov-Bohm phase (Xu and Sachdev, 2009;
Xu and Balents, 2011) of π: for the kagome antiferromagnet, these are conveniently
expressed in terms of an effective, fully-frustrated Ising model on the dice lattice. A
key parameter in this Ising model is the sign of a particular next-nearest-neighbor
interaction. A ‘ferromagnetic’ sign leads to the simpler vison PSG, and is discussed
in Section 4.3.1; the more complicated ‘antiferromagnetic’ case is discussed in Sec-
tion 4.3.2. In both cases, we use the vison PSG to present quantum field theories
for the confinement transitions, and discuss renormalization group analyses of these
quantum critical points. The field theories are also used to classify the patterns of
lattice symmetry breaking in the confining valence bond solid states.
In Appendix 4.A the explicit calculation of a visons Berry phase can be found.
The results of a PSG analysis of visons for different lattice geometries are summarized
in Appendix 4.B.
kagome lattice 60
4.2 Z2 spin liquids and visons

We begin by a review of the basic properties of Z2 spin liquids, following the
description in Ref. Sachdev (1992).
~i using Schwinger bosons (Arovas
It is convenient to describe the S = 1/2 spins S
and Auerbach, 1988) biα (α =↑, ↓)
~i = 1 b† ~σαβ biβ ,
S (4.1)
2 iα
where ~σ are the Pauli matrices, and the bosons obey the local constraint
X
b†iα biα = 1 (4.2)
α
on every site i. Our analysis below can be easily extended to gapped Z2 spin liquids
obtained from the Schwinger fermion formulation, (Wen, 1991; Lu et al., 2011; Iqbal
et al., 2011b,a) but we will only consider the Schwinger boson case for brevity.
The Z2 spin liquid is described by an effective boson Hamiltonian
X X
Hb = − Qij εαβ b†iα b†jβ + H.c. + λ b†iα biα , (4.3)
i<j i
where ε is the antisymmetric unit tensor, λ is chosen to satisfy the constraint in

Eq. (4.2) on average, and the Qij = −Qji are a set of variational parameters chosen
to optimize the energy of the spin liquid state. Generally, the Qij are chosen to be
non-zero only on near neighbor links. The ‘Z2 ’ character of the spin liquid requires
that the links with non-zero Qij can form closed loops with an odd number of links,
(Read and Sachdev, 1991; Wen, 1991; Sachdev, 1992). If the state preserves time-
reversal symmetry, all the Qij can be chosen real.
This effective Hamiltonian also motivates a wavefunction of the spin liquid (Read
and Sachdev, 1989b; Tay and Motrunich, 2011)
!
X
|SLi = P exp fij εαβ b†iα b†jβ |0i, (4.4)
i<j
where |0i is the boson vaccum, P is a projection operator which selects only states
which obey Eq. (4.2), and the boson pair wavefunction fij = −fji is determined
kagome lattice 61
Figure 4.2: A vison on the kagome lattice. The center of the vison is marked by the
X. We have sgn(Qvij ) = −sgn(Qij ) only on the links marked by the wavy lines.
by diagonalizing Eq. (4.3) by a Bogoliubov transformation (see Eq. (4.35) in Ap-

pendix 4.A).
The Schwinger boson approach also allows a description of the vison excited states.
(Read and Sachdev, 1991; Sachdev and Read, 1991) We choose the vison state as the
ground state of a Hamiltonian, Hbv , obtained from H by mapping Qij → Qvij (see
Eq. (4.31)); then the vison state |Ψv i has a wavefunction as in Eq. (4.4), but with
fij → fijv (see Eq. (4.34)). Far from the center of the vison, we have |Qvij | = |Qij |,
while closer to the center there are differences in the magnitudes. However, the key
difference is in the signs of the link variables, as illustrated in Fig. 4.2: there is a
‘branch-cut’ emerging from the vison core along which sgn(Qvij ) = −sgn(Qij ). This
branch-cut ensures that the Z2 flux equals -1 on all loops which encircle the vison
core, while other loops do not have non-trivial Z2 flux. Such a configuration of the Qvij
is expected to be a metastable solution of the Schwinger boson mean-field equations,
representing a vison excitation.
The previous solution for the vison state (Sachdev and Read, 1991) was obtained
using a continuum field theoretic representation of the Schwinger boson theory, valid
kagome lattice 62
in the limit of a small energy gap towards spinful excitations. In principle, such a
solution can also be obtained by a complete solution of the Schwinger boson equations
on the lattice, but this requires considerable numerical effort. Here, we illustrate the
solution obtained in the of a large spin gap. In the large spin gap limit, (Tchernyshyov
et al., 2006) we can integrate out the Schwinger bosons, and write the energy as a local
functional of the Qij . This functional is strongly constrained by gauge-invariance: for
time-independent Qij , this functional takes the form
X β

2 4
E[{Qij }] = − α|Qij | + |Qij |
i<j
2
X
+K Qij Q∗jk . . . Q∗`i (4.5)
even loops
Here α, β, and K are coupling constants determined by the parameters in the Hamil-
tonian of the antiferromagnet. We have shown them to be site-independent, because
we have only displayed terms in which all links/loops are equivalent; they can de-
pend upon links/loops for longer range couplings provided the full lattice symmetry
is preserved. We describe the results of a minimization of E[{Qij }] on the simpler
case of the triangular lattice in Fig. 4.3. The magnitudes of Qvij are suppressed close
to the vison, and converge to Qij as we move away from the vison (modulo the sign
change associated with the branch cut), analogous to the Abrikisov vortices. Despite
the branchcut breaking the 3-fold rotation symmetry, the gauge-invariant fluxes of
Qvij preserve the rotation symmetry.
Let us now consider the motion of a single vison. A key ingredient is the Berry
phase a vison accumulates while moving through the background spin liquid. The
gauge-invariant Berry phases are those associated with a periodic motion, and so let
us consider the motion of a vison along a general closed loop C. We illustrate the
simple case where C encloses a single site of the triangular lattice antiferromagnet
in Fig. 4.4. The Berry phase for this periodic motion can be computed in a manner
analogous to that presented in Section III.A of Ref. Read and Sachdev (1990) for a
monopole in a U(1) spin liquid; details appear in Appendix 4.A. The final gauge-
invariant Berry phase turns out to be given by the gauge-transformation required
kagome lattice 63
-0.
10
- 0.
04
00
00
00
00
- 0.
10
-0.
06
-0.0013 -0.0028
-0.
08
63
37
-0.
00
00
00
00
16
00
- 0.
-0.
-0.
48
-0.0022
-0.0112 -0.0185
57
00
-0.
-0.0128 -0.1113
89
02 X
-0.
-0.0354
-0.2406
-0.
00
04
-0.
00
05
Figure 4.3: A vison on the triangular lattice (a similar solution is expected on the
kagome lattice). The center of the vison is marked by the X. The wavy line is the
‘branch-cut’ where we have sgn(Qvij ) = −sgn(Qij ) only on the links crossed by the
line. Plotted is the minimization result of E[{Qij }] with α = 1, β = −2, K = −0.5.
Minimization is done with the cluster embedded in a vison-free lattice with all nearest
neighbor links Qij . The numbers are (Qij − Qvij ) and the thickness of the links
are proportional to (Qvij − Qij )1/2 . K < 0 gives rise to the zero flux state while
K > 0 favors the π flux state where the unit cell is doubled. The zero and π flux
states without the vison have been studied previously. (Sachdev, 1992; Wang and
Vishwanath, 2006)
to map the final state to the initial state. The analysis in Fig. 4.4 shows that the
required gauge transformation is
biα → −biα , for i inside C

biα → biα , for i outside C. (4.6)
By Eq. (4.2), each site has one boson, and so the total Berry phase accumulated by
kagome lattice 64
Figure 4.4: Periodic motion of a vison around a closed loop C on the triangular lattice.
Here C encloses the single site marked by the filled circle. The wavy lines indicate
sgn(Qvij ) = −sgn(Qij ), as in Fig. 4.2. The bottom state is gauge-equivalent to the
top state, after the gauge transformation biα → −biα only for the site i marked by
the filled circle.
kagome lattice 65
|Ψv i is
π × (number of sites enclosed by C) , (4.7)
as recognized in earlier works (Senthil and Fisher, 2000; Jalabert and Sachdev, 1991;
Sachdev and Vojta, 2000; Xu and Sachdev, 2009; Xu and Balents, 2011). It is also
clear that for a spin S antiferromagnet, the Berry phase would be multiplied by a
factor of 2S.
Finally, let us also mention the spinon states, although these will not play a
role in the subsequent analysis of the present paper. These are created by applying
†
the Bogoliubov quasiparticle operator γµα (Appendix 4.A) on the spin-liquid ground
state; in this manner we obtain the spinon state
X
U −1∗

|µαi = µ`
|`αi
`
!
X
|`αi = P b†`α exp fij εαβ b†iα b†jβ |0i, (4.8)
i<j
where Uiµ is a Bogoliubov rotation matrix defined in Appendix 4.A. We can now
also consider a spinon well-separated from a vison, and describe the motion of a
vison along a large contour C which encircles the spinon. First, consider the motion
of a vison around the spinon state |`αi localized on the site `. Proceeding with
the argument as above, the projection onto states which obey Eq. (4.2) now implies
that the Berry phase for such a process is π × ((number of sites enclosed by C) − 1).
The transformation to the spinon state |µαi will not change the result, provided the
‘wavefunction’ (U −1∗ )µ` is localized well within the contour C. Thus relative to the
Berry phase in the case without a spinon, the vison acquires an additional phase of π
upon encircling a spinon i.e. the spinons and visons are relative semions. (Read and
Chakraborty, 1989)
kagome lattice 66
4 8
u
2 6 10
A B C 1 5 9 12
3 7 11
v
Figure 4.5: Left: unit cell of the dice lattice consisting of the 3 independent sites
labeled A, B and C. The A sites are 6-coordinated,
√ whereas the B √ and C sites are 3-
coordinated. The basis vectors u = (3/2, 3/2) and v = (3/2, − 3/2) are indicated
by dashed arrows. Right: gauge choice for the fully frustrated Ising model on the dice
lattice with a 12-site unit cell. Red thick bonds are frustrated, i.e. sgn(Jij ) = −1.
4.3 PSG analysis of the fully frustrated Ising model

on the dice lattice
The analysis of Section 4.2 suggests a simple effective model for the vison fluctu-
ations about the Z2 spin liquid ground state. In the framework of the path-integral
formations of Ref. Sachdev (1992), the visons are saddle-points of the Qij with π-flux,
as shown in Fig. 4.3. Each saddle-point traces a world-line in spacetime, representing
the time evolution of the vison. The Berry phase computation in Section 4.2 shows
that this world-line picks up π-flux each time it encircles a kagome lattice site. We
know further that two such worldlines can annihilate each other, because 2π magnetic
flux is equivalent to zero flux. So if we interpret the wordlines as the trajectory of
a particle, that particle must be its own anti-particle, and has a real field operator.
In this manner, we see that the fluctuations represented as the sum over all vison
worldlines is precisely that of a frustrated Ising model in a transverse field (Jalabert
and Sachdev, 1991; Sachdev and Vojta, 2000; Senthil and Fisher, 2000):
X
H=− Jij φi φj + . . . , (4.9)
i<j
kagome lattice 67
where the product of bonds around each elementary plaquette is negative

Y
sgn(Jij ) = −1 . (4.10)
plaq.
Also, we have not displayed the transverse-field term, because most of our symmetry
considerations are restricted to time-independent, static configurations.
In the following we will use a soft-spin formulation where the φj ’s take real values.
This model is invariant under Z2 gauge transformations φj → σj φj , Jij → σi σj Jij
with σj = ±1. For the case of the Z2 spin liquid on a kagome lattice the dual
Ising model lives on the dice lattice, shown in Fig. 4.5. The dice lattice has three
independent sites per hexagonal unit cell, two of them are three-coordinated and one
is six-coordinated. Flipping a dual Ising spin changes the flux through a plaquette on
the kagome lattice by π, thereby creating or annihilating a vison. The magnetically
disordered phase of the Ising model corresponds to a Z2 spin liquid with deconfined
spinon excitations, whereas the ordered phases describe different valence bond solids
where the visons are condensed and fractional excitations are confined.
We study the confinement transitions of the spin liquid by constructing a Ginzburg-
Landau functional that is consistent with the projective symmetry group (PSG), i.e.
the combination of lattice-symmetry- and Z2 gauge-transformations that leave the
Hamiltonian (4.9) together with (4.10) invariant. In order to determine the PSG
transformations we fix the gauge of nearest neighbor interactions as shown in Fig.
4.5, thereby obtaining a unit cell with twelve sites. The generators of the dice lattice
symmetry group are translations by one of the two basis vectors, e.g. u, reflections
about one axis, e.g. the x-axis, and π/3 rotations about the central 6-coordinated
site. Since our gauge choice is already invariant under rotations, we only need to
determine the gauge-transformations corresponding to translations and reflections to
specify the PSG. These are shown in Fig. 4.6.
The dispersion relations of the soft-spin modes can be obtained directly from the
Hamiltonian (4.9). The corresponding action takes the form
X (i) (j)
S= φq,Ω (Ω2 + m2 )δi,j − Jq(ij) φ−q,−Ω (4.11)
Ω,q
kagome lattice 68
Figure 4.6: Gauge transformations associated with translations by u and reflections

about the x-axis. Shown is the unit cell of the fully frustrated dice model after a
lattice translation by u (left) and after a reflection about the x-axis (right). The
corresponding gauge-transformations consist of spin flips on the sites marked by blue
points, which restore the original gauge pattern as in Fig. 4.5.
(ij)
where the summation over the sublattice indices i, j = 1...12 is implicit, Jq denotes
the Fourier transform of the interaction matrix Jij , and we have included a kinetic
energy with frequency Ω (which descends from the transverse field), and a mass
term, m. As noted earlier (Nikolic and Senthil, 2003; Sengupta et al., 2006), the
frustrated Ising model on the dice lattice with nearest neighbor interactions gives
rise to three flat bands (each being four-fold degenerate in our gauge choice with
a 12-site unit cell), which would result in infinitely many critical modes. In order
to lift this degeneracy we include further interactions beyond nearest neighbors (Xu
and Balents, 2011) that are consistent with the PSG. Different masses on the 3-
and 6-coordinated sites would be allowed by the PSG, but they don’t give rise to a
momentum dependence of the modes and thus don’t change the picture qualitatively.
Out of the five possible additional interactions up to a distance of two times the
nearest neighbor bond length only one is consistent with the PSG. This interaction,
shown in Fig. 4.7, connects different 3-coordinated lattice sites and gives rise to a
non-flat dispersion. The Fourier transform of the interaction matrix Jij takes the
form
kagome lattice 69
Figure 4.7: Additional next-nearest neighbor interactions between 3-coordinated sites

on the dice lattice (shown as dashed blue lines) that are allowed by the PSG. For
illustrative clarity not all additional bonds in the unit cell are shown. All other bonds
can be obtained from the ones that are shown via translations by the basis vectors u
and/or v.
κ∗1 e−i2qu
 
0 1 1 0 0 0 0 0 0 0
1 0 0 1 1 −1 0 0 0 0 −e−i2qv e−i2qv
ei2q(v−u) 0 e−i2qu −e−i2qu
 1 0 0 −1 0 1 0 0

 
 κ1 1 ei2q(u−v) 0 0 0 0 1 0 0 0 0 
 0 1 −1 0 0 0 0 0 1 0 0 κ∗2 
 0 −1 0 0 0 0 0 1 1 0 κ3∗ 0

Jq(ij) = .
 0 0 1 0 0 0 0 −ei2q(v−u) 1 κ∗1 0 0 
 i2qu
1 −ei2q(u−v) 0 −1 ei2q(u−v)

e 0 0 1 0 0 0 
 0 0 0
i2qu
0 1 1 1 0 0 1 1 1 
 0 0 e 0 0 0 κ1 −1 1 0 0 0 
0 −ei2qv 0 0 0 κ3 0 ei2q(v−u) 1 0 0 0
0 ei2qv −ei2qu 0 κ2 0 0 0 1 0 0 0
(4.12)
i2qu i2q(u−v) i2qu i2qv
Here we have abbreviated κ1 ≡ t(1 + e +e ), κ2 ≡ t(1 + e +e ) and
κ3 ≡ t(1 + ei2qv + ei2q(v−u) ), which are the additional terms coming from the next
nearest neighbor interaction that is allowed by the PSG and t denotes the relative
strength of the interaction with respect the the nearest neighbor coupling J = 1.
(ij)
Diagonalizing Jq results in three dispersing bands, each being four-fold degenerate.
Depending on the sign of the next-nearest neighbor interaction t, we get different
transitions to valence bond solids that we are going to discuss in the following.
4.3.1 Ferromagnetic n.n.n. interactions

For ferromagnetic next-nearest neighbor interactions (t > 0) the dispersion mini-
mum of the lowest band is at zero momentum q = 0. The corresponding eigenvalue
kagome lattice 70
(ij) √
of the interaction matrix J0 is given by λ+ = (3t + 24 + 9t2 )/2 and one choice for
the four degenerate eigenvectors is
h i
v (1) = λ−1
− 1 0 λ−1
−
λ+
6
−λ−1
− 0 0 0 0 −λ−1
−
λ+
6
||v||−1 (4.13)
h i
v (2) = λ−1
− 0 1 λ−1
− − λ6+ 0 λ−1
− 0 0 λ−1− 0 − λ6+ ||v||−1 (4.14)
h i
v (3) = λ−1− 0 0 λ −1
− 0 λ−1
− −λ −1
− 1 0 −λ −1
− λ−1
− 0 ||v||−1 (4.15)
h i
v (4) = 0 0 0 0 6 λ+ λ+
6
λ+
6
0 1 6 λ+ λ+
6
λ+
6
||v||−1 (4.16)
√ p
with λ− = (3t− 24 + 9t2 )/2 and ||v|| = (λ2+ + 6)/6. This set of eigenvectors forms
an orthonormal basis for the four critical modes at the transition to the confined
phase. In the magnetically ordered state the magnetization φj (R) at lattice site
R = 2nu + 2mv (n, m ∈ N) and sublattice site j ∈ {1, ..., 12} is given by
X (n)
φj (R) = ψn vj (4.17)
n=1...4
and is independent of the lattice site R for ferromagnetic n.n.n. interactions, i.e. the
ordered VBS phases have a 12-site unit cell. The values of the four mode amplitudes
ψn are obtained by minimizing the Ginzburg-Landau functional, which in turn is
given by all homogeneous polynomials in the mode amplitudes ψn that are invariant
under the PSG transformations. In order construct these polynomials we have to
determine how the mode amplitudes transform under the PSG. This can be done by
looking at the transformation properties of the eigenvectors v (n) . For example, the
transformation properties of the mode amplitudes with respect to translations Tu are
determined via
X (n)
X (m)
T̂u φj = ψn T̂u vj = (Tu )mn ψn vj
n n,m
X (m)
= (T̂u ψ)m vj . (4.18)
m
The resulting PSG transformation matrices for the amplitudes of the four critical
modes with respect to translations (Tu ), reflections (Ix ) and rotations (R6 ) are given
kagome lattice 71
by
 
0 0 −1 0
 
0 0 0 −1
Tu =  , (4.19)
 
1 0 0 0
 
0 1 0 0
 
0 1 0 0
 
1 0 0 0
Ix =  , (4.20)
 
0 0 1 0
 
0 0 0 −1
 
0 0 1 0
 
1 0 0 0
R6 =   . (4.21)
 
0 1 0 0
 
0 0 0 1
We used the GAP program (GAP) to show that these three matrices generate a finite,
48 element subgroup of O(4) which is isomorphic to GL(2, Z3 ). We also determined
that this group is isomorphic to the group ± 21 [O × C2 ] in the classification of Conway
and Smith. (Conway and Smith, 2003)
Next we determined the most general Ginzburg-Landau functional of the ψn . It
turns out that there are three fourth order polynomials that are invariant under this
group, thus our functional for the four mode amplitudes depends on the coupling
constants, r, u, a and b, and takes the form
X
L = ((∇ψn )2 + (∂τ ψn )2 + rψn2 + uψn4 )
n=1...4
X
ψn2 ψm
2
+ b ψ12 (ψ2 ψ3 − ψ2 ψ4 + ψ3 ψ4 )

+a
n<m
2
+ψ2 (ψ1 ψ3 + ψ1 ψ4 − ψ3 ψ4 ) + ψ32 (ψ1 ψ2 − ψ1 ψ4
+ψ2 ψ4 ) − ψ42 (ψ1 ψ2 + ψ1 ψ3 + ψ2 ψ3 ) .

(4.22)
At b = 0 and a = 2u, this is just the well-known φ4 field theory with O(4) symmetry,
whose critical properties are described by the extensively studied Wilson-Fisher fixed
kagome lattice 72
point. The b coupling breaks the O(4) symmetry down to GL(2, Z3 ). Remarkably,
the renormalization group (RG) properties of just such a quartic coupling have been
studied earlier by Toledano et al.; (Toledano et al., 1985) they denoted this symmetry
class as [D3 /C2 ; O/D2 ], following the analysis of Du Val. (Du Val, 1964) Toledano
et al. found that the O(4) fixed point was unstable, but were unable to find a stable
critical fixed point at two-loop order. It would be interesting to extend the analysis of
Eq. (4.22) to higher loop order, and search for a suitable fixed point by the methods
reviewed in Ref. Vicari (2007).
Despite the difficulty in finding a suitable critical fixed point, we can assume
the existence of a second-order quantum phase transition, and use Eq. (4.22) to
determine the structures of possible confining phases. Minimizing this functional for
the magnetically ordered phase r < 0 gives rise to two possible phases, depending
on the values of the two parameters a and b in the GL-functional (4.22). The phase
diagram is shown in Fig. 4.8.
1.5
b
1.0
0.5 VBS 1F
a
-1 1 2 3
-0.5
unstable
-1.0 VBS 2F
-1.5
-2.0
Figure 4.8: Phase diagram of Eqn. (4.22) as a function of the two couplings a and
b (here we have set u = 1). The VBS 1F phase is not reflection symmetric, whereas
the VBS 2F phase is reflection symmetric. In the VBS 2F phase there is a crossover
(indicated by the dashed line) from a phase where one of the ψn ’s is zero (left of the
dashed line) to a phase where three of the ψn ’s are zero (right of the dashed line).
In the VBS 1F phase the GL-functional (4.22) has 16 degenerate minima corre-
kagome lattice 73
sponding to different magnetization patterns. Note, however, that the magnetization

itself is not a gauge invariant quantity. In fact, all degenerate minima give rise to
the same bond patterns, which are related by simple lattice symmetry transforma-
tions. The bond pattern in the VBS 1F phase is symmetric under rotations by π/3,
translations by 2u and 2v but it is not reflection symmetric. The 16 minima thus
correspond to 8 different bond patterns that are related by translations by u and v
as well as by a reflection about the x-axis (the factor of two arises from a global spin
flip symmetry). In Fig. 4.9 we show the bond pattern in the VBS 1F phase, i.e. we
plot the gauge-invariant value of the bond-strength Jij φi φj for every nearest neighbor
bond. In this figure we mapped the bond pattern from the dice lattice back to the
kagome lattice by assigning the value of the bond-strength on the dice lattice to the
bond on the kagome lattice that intersects the dice lattice bond. Deep in the confined
phase the bonds plotted in Fig. 4.9 thus represent the expectation value hσijx i of the
Z2 gauge field on the kagome bonds in the transverse field direction.
One way to relate the symmetries of valence bond solid phases on the kagome
lattice to the ordered phases of the frustrated Ising model is to make use of the
mapping between hardcore dimer models and frustrated Ising models (Moessner
et al., 2001), i.e. by putting a dimer on every bond of the kagome lattice that
intersects a frustrated bond on the dual dice lattice. The dimer covering that is
obtained in this way for the VBS 1F phase is shown in Fig. 4.1. In comparison to
previously obtained dimer coverings of the kagome lattice (Nikolic and Senthil, 2003;
Poilblanc et al., 2010; Schwandt et al., 2010) which maximize the number of perfectly
flippable hexagons on a 36-site unit cell, this dimer covering maximizes the number
of perfect flippable diamonds on a 12-site unit cell. Note that this bond pattern is
similar to the diamond pattern observed by Yan et al. (Yan et al., 2011).
We note that the VBS 1F state has recently been identified as the ground state of
the deformed kagome lattice spin-1/2 antiferromagnet Rb2 Cu3 SnF12 . (Matan et al.,
2010) It is possible that the spin physics being discussed here played a role in the
lattice distortion observed in this experiment. (Yang and Kim, 2009) Also in Zn-
paratacamite, Znx Cu4−x (OH)6 Cl2 , there is a transition (Lee et al., 2007) between a
kagome lattice 74
Figure 4.9: Bond pattern in the VBS 1F phase. Plotted is the gauge invariant bond-
strength Jij φi φj for nearest neighbor bonds on the dice-lattice at a = b = 1, which has
been assigned to each respective kagome bond. Black dashed lines indicate satisfied
bonds (−Jij φi φj < 0), red solid lines are frustrated bonds (−Jij φi φj > 0). The
thickness of the bonds is proportional to the bond-strength.
spin-liquid phase near x = 1 to a distorted kagome lattice near x = 0, and it has

been argued (Lawler et al., 2008) that a ‘pinwheel’ VBS state, which is identical in
symmetry to our VBS 1F state, plays a role in the latter case.
In the VBS 2F phase the GL-functional (4.22) has 8 degenerate minima and the
corresponding bond-patterns have an additional reflection symmetry as compared to
the VBS 1F phase. Moreover, there is a crossover from a phase where one of the ψn ’s
is zero to a phase where three of the ψn ’s are zero, indicated by the dashed line in
Fig. 4.8. The bond patterns are shown in Fig. 4.10.
4.3.2 Antiferromagnetic n.n.n. interactions

If the additional next-nearest neighbor interactions are antiferromagnetic (t <
0) and smaller than a critical value |t| < tc ∼ 1, the dispersion-minimum of the
lowest band lies at the edges of the Brillouin zone, i.e. at q = ±Q1 = 0, ± 32π

√
3
kagome lattice 75
Figure 4.10: Bond patterns in the VBS 2F phase. Plotted is the gauge invariant bond-
strength Jij φi φj for nearest neighbor bonds on the dice lattice, shown on the corre-
sponding kagome bonds. Black dashed lines represent satisfied bonds (−Jij φi φj < 0).
In this phase there are no frustrated bonds at all. The thickness of the bonds is
proportional to the bond-strength.
for our gauge choice with a hexagonal 12-site unit cell. In this case eight modes
become critical at the confinement transition and the resulting unit cell has 36 sites.
(ij) √
The corresponding eigenvalue of the interaction matrix J±Q1 is 6 and the eight
eigenvectors occur in complex conjugate pairs
h i
(1) (1)∗ −1 iπ/3 −e iπ/3
vQ1 = v−Q1 = √1 √1 0 √1 √1 √ 0 0 0 0 e√ √
12 2 12 12 12 12 12
h i
(2) (2)∗ −eiπ/3 −1 −iπ/3
−e√ e−iπ/3
vQ1 = v−Q1 = √1 0 √1 √ √ 0 √1 0 0 0 √
12 2 12 12 12 12 12
h i
(3) (3)∗ −e iπ/3 e−iπ/3 −iπ/3
−e√
vQ1 = v−Q1 = √ 0 0 √1 0 √1 √ √1 0 − √112 0
12 12 12 12 2 12
h i
(4) (4)∗
vQ1 = v−Q1 = 0 0 0 0 √1 √1 √1 0 √1 √1 √1 √1 (4.23)
12 12 12 2 12 12 12
Note that in this case the magnetization at lattice site R = 2nu + 2mv and sublattice
site j is given by
X (n)
φj (R) = eiQ1 ·R ψn vQ1 ,j + c.c. (4.24)
n=1...4
kagome lattice 76
The PSG transformations of the four complex mode amplitudes ψn can be determined
similarly to the ferromagnetic case. Quite generally they are defined by
h X i
(n)
Ôφj (R) = Re eiQ1 ·(ÔR) ψn ÔvQ1 ,j
n=1...4
.
h X i
iQ1 ·R (n)
= Re e (Ôψ)n vQ1 ,j (4.25)
n=1...4
If we define the vector Ψ = ψ1 , .., ψ4 , ψ1∗ , .., ψ4∗ , the PSG transformation matrices

corresponding to translations by u, reflections about the x-axis and rotations by π/3

around the central site, that act on the vector Ψ take the form
 
0 0 −1 0 0 0 0 0
 
 0
 0 0 −1 0 0 0
0
 i2π/3 
e 0 0 0 0 0 0 0
 
ei2π/3 0
 
 0 0 0 0 0 0
Tu =  , (4.26)
 0 0 0 0 0 0 −1 0 
 
 
 0
 0 0 0 0 0 0 −1
 
 0
 0 0 0 e−i2π/3 0 0 0
0 0 0 0 0 e−i2π/3 0 0
 
0 0 0 0 0 1 0 0
 
0 0 0 0 1 0 0 0
 
 
0
 0 0 0 0 0 e−i2π/3 0

−1
 
0 0 0 0 0 0 0
Ix =  , (4.27)
0 1 0 0 0 0 0 0
 
 
1 0 0 0 0 0 0 0
 
 
0
 0 ei2π/3 0 0 0 0 0
0 0 0 −1 0 0 0 0
kagome lattice 77
 
0 0 0 0 0 1 0 0
 
0
 0 0 0 0 0 e−i2π/3 0

 
0 0 0 0 1 0 0 0
 
 
0 0 0 0 0 0 0 1
R6 =  . (4.28)
0 1 0 0 0 0 0 0
 
 
0
 0 ei2π/3 0 0 0 0 0

 
1 0 0 0 0 0 0 0
 
0 0 0 1 0 0 0 0
As in the previous subsection, we used the GAP program to determine that these three
matrices generate a 288 element subgroup of O(8) which is isomorphic to GL(2, Z3 ) ×
D3 The Ginzburg-Landau functional is again given by all homogeneous polynomials
that are invariant under this group. At fourth order there are five such polynomials,
thus the GL-functional depends on the coupling constants r, u, a1 , ..., a4 and is given
kagome lattice 78
by
X h i
rψn2 + uψn4 + ψ22 ψ32 a1 − 2a4 cos(2(θ2 − θ3 ))

L4 =
n=1..4
h √ i
+ψ12 ψ22 a1 + a4 cos(2(θ1 − θ2 )) + 3a4 sin(2(θ1 − θ2 ))
h √ i
h √ i
+ψ12 ψ42 a1 + a4 cos(2(θ1 − θ4 )) − 3a4 sin(2(θ1 − θ4 ))
2 2
h √ i
h √ i
h √ i
+ψ1 ψ22 ψ4 − a2 cos(θ1 − θ4 ) − 2a3 cos(θ1 − 2θ2 + θ4 ) − 3a2 sin(θ1 − θ4 )
h √ i
+ψ1 ψ32 ψ4 a2 cos(θ1 − θ4 ) + 2a3 cos(θ1 − 2θ3 + θ4 ) + 3a2 sin(θ1 − θ4 )
h √ i
+ψ1 ψ2 ψ42 a2 cos(θ1 − θ2 ) + 2a3 cos(θ1 + θ2 − 2θ4 ) − 3a2 sin(θ1 − θ2 )
2
h √ i
+ψ1 ψ2 ψ4 2a3 cos(2θ1 − θ2 − θ4 ) + a2 cos(θ2 − θ4 ) − 3a2 sin(θ2 − θ4 )
h √
+ψ2 ψ32 ψ4 − a2 cos(θ2 − θ4 ) + a3 cos(θ2 − 2θ3 + θ4 ) + 3a2 sin(θ2 − θ4 )
√ i
+ 3a3 sin(θ2 − 2θ3 + θ4 )
2
h √
+ψ1 ψ2 ψ3 − a2 cos(θ1 − θ2 ) + a3 cos(θ1 + θ2 − 2θ3 ) + 3a2 sin(θ1 − θ2 )
√ i
− 3a3 sin(θ1 + θ2 − 2θ3 )
h √ i
+ψ1 ψ3 ψ42 a2 cos(θ1 − θ3 ) + 2a3 cos(θ1 + θ3 − 2θ4 ) − 3a2 sin(θ1 − θ3 )
h √ i
+ψ12 ψ2 ψ3 a3 cos(2θ1 − θ2 − θ3 ) + 2a2 cos(θ2 − θ3 ) + 3a3 sin(2θ1 − θ2 − θ3 )
h √
+ψ22 ψ3 ψ4 − a3 cos(2θ2 − θ3 − θ4 ) + a2 cos(θ3 − θ4 ) + 3a3 sin(2θ2 − θ3 − θ4 )
√ i
− 3a2 sin(θ3 − θ4 )
h √ i
+ψ12 ψ3 ψ4 − 2a3 cos(2θ1 − θ3 − θ4 ) − a2 cos(θ3 − θ4 ) + 3a2 sin(θ3 − θ4 )
h √
+ψ1 ψ22 ψ3 − a2 cos(θ1 − θ3 ) + a3 cos(θ1 − 2θ2 + θ3 ) + 3a2 sin(θ1 − θ3 )
√ i
− 3a3 sin(θ1 − 2θ2 + θ3 )
h √ i
+ψ2 ψ3 ψ42 − 2a2 cos(θ2 − θ3 ) − a3 cos(θ2 + θ3 ) − 3a3 sin(θ2 + θ3 ) . (4.29)
kagome lattice 79
0.4 a4
VBS 1A 0.2
VBS 2A a2
-0.4 -0.2 0.2 0.4
VBS 4A -0.2
VBS 3A
-0.4
Figure 4.11: Phase diagram obtained from the GL-functional (4.30) as a function of
the couplings a2 and a4 . The other parameters are fixed at u = 1, a1 = 1/2, a3 = a2 ,
a5 = 1/20 and a6 = −1/25. The different phases are described in the text.
Here we have expressed the complex mode amplitudes in terms of their absolute
value and phase ψn eiθn . Note that this fourth order Ginzburg-Landau functional has
a remaining continuous U (1) symmetry, since it is invariant under a change of all
phases θn → θn + χ. The “magnetization” (4.24) is not invariant under this U (1)
transformation, however. In order to break this continuous degeneracy we need to
include higher order terms in the Ginzburg-Landau functional. Among the invariant
sixth order polynomials there are five which break the U (1) symmetry. A full analysis
of the invariant GL-functional at sixth order is beyond the scope of this paper, and
we restrict ourselves to the simplest U (1)-breaking sixth order term. The U (1)-
invariant sixth order terms are not expected to qualitatively change our results at the
confinement transition.
In the remainder of this section we are going to consider the following fourth order
GL-functional, including the simplest invariant sixth order U (1)-breaking polynomial,
which takes the form
X
ψn6 a5 + a6 cos[6θn ] .

L = L4 + (4.30)
n=1...4
kagome lattice 80
In total our simplified GL-functional thus has seven coupling constants. Again, a
complete analysis of the phase diagram as a function of these seven couplings is
hardly feasible. A representative slice of the phase diagram is shown in Fig. 4.11,
where we have fixed the values u = 1, a1 = 1/2, a3 = a2 , a5 = 1/20 and a6 = −1/25
and show the different phases as function of the two remaining parameters a2 and
a4 . All phases have a 36-site unit cell and are invariant with respect to translations
by 4u − 2v and 4v − 2u. In the above mentioned parameter regime there are four
different ordered phases. VBS 1A has a π/3 rotational symmetry, but is not reflection
symmetric. A dimer representation of this state is shown in Fig. 4.1, which was
obtained by putting a dimer on every kagome bond that intersects a frustrated bond
on the dice lattice. This dimer covering suggests that our VBS 1A state is identical to
previously found valence bond solid states on the kagome lattice which maximize the
number of perfectly flippable hexagons (Nikolic and Senthil, 2003; Poilblanc et al.,
2010; Schwandt et al., 2010). The VBS 2A phase is symmetric under 2π/3 rotations
and has a reflection symmetry. No rotational symmetry is present in the VBS 3A and
4A phases. The 3A phase has a reflection symmetry, however, which is not present in
the 4A phase. Bond patterns of all four phases are shown in Fig. 4.12.
We performed a one-loop RG calculation for (4.29) and found six additional fixed
points besides the Gaussian one, but all of them turned out to be unstable. Again it
would be useful to revisit this issue using higher loop methods. (Vicari, 2007)
4.4 Conclusions
We have studied confinement transitions of Z2 spin liquids of Heisenberg antifer-
romagnets on the kagome lattice by constructing field theories that are consistent
with the projective symmetry group of the vison excitations. Depending on the sign
of the next-nearest neighbor interaction between the visons, we found that the visons
transformed under the group GL(2, Z3 ) for the simpler case, and under GL(2, Z3 )×D3
for the other case. Our analysis shows that possible VBS phases close to the con-
finement transition are strongly constrained by the vison PSG. We found VBS states
kagome lattice 81
Figure 4.12: Bond patterns in the VBS 1A , 2A , 3A and 4A phase (clockwise, starting
from the upper left). Again we plot the gauge invariant bond-strength Jij φi φj for
nearest neighbor bonds on the dice lattice, mapped to the corresponding kagome
bonds. Black dashed lines represent satisfied bonds (−Jij φi φj < 0), red lines are
frustrated bonds. The thickness of the lines is proportional to the bond strength.
kagome lattice 82
that break the translational symmetry of the kagome lattice with either 12- or 36-site
unit cells, for the two vison PSGs respectively. The two possible VBS states with
12-site unit cells do not break the rotation symmetry of the kagome lattice but one of
them breaks the reflection symmetry; this state is closely connected to the “diamond
pattern” enhancement observed in the recent numerical study of Yan et al. (Yan
et al., 2011) As far as possible VBS states with 36-site unit cells are concerned, our
analysis is not exhaustive. Nevertheless, we found different VBS states with either
full, reduced or no rotation symmetry, as well as states that do or do not break the
reflection symmetry of the kagome lattice. Our results should be useful in more com-
pletely characterizing spin liquids in numerical or experimental studies of the kagome
antiferromagnet.
Analogous analyses for Z2 spin liquids on other lattices have been carried out in
other cases (see Appendix 4.B). In all other cases, the effective theory for confining
transition has an emergent continuous symmetry, and the criticality can be computed
using properties of the Wilson-Fisher fixed point; reduction to the discrete lattice
symmetry appears only upon including higher-order couplings which are formally
“irrelevant” at the critical fixed point. The kagome lattice is therefore the unique
case (so far) in which the reduction to discrete lattice symmetry appears already in
the critical theory: these is the theory in Eqs. (4.22), and its relevant quartic couplings
are invariant only under discrete symmetries. This suggests that the numerical studies
of confinement transitions may be easier on the kagome lattice.
We thank Steve White for sharing the results of Ref. Yan et al. (2011) prior to
publication, and for valuable discussions. We are also grateful to M. Lawler and
C. Xu for discussions.
4.A Berry phase of a vison

This appendix will compute the Berry phase of a vison moving around a S = 1/2
spin of the antiferromagnet, as illustrated in Fig. 4.4. We will follow the method of
Section III.A of Ref. Read and Sachdev (1990), generalized to a Z2 spin liquid as in
kagome lattice 83
Ref. Sachdev (1992).

We consider the time-dependent Schwinger boson Hamiltonian
X X
Hbv (τ ) = − Qvij (τ )εαβ b†iα b†jβ + H.c. + λvi (τ )b†iα biα , (4.31)
i<j i
where the τ dependence of Qvij and λvi is chosen so that the vison executes the motion
shown in Fig. 4.4, while always maintaining the constraint in Eq. (4.2).
We compute the Berry phase by working with the instantaneous ground state of
Hb (τ ). This is facilitated by a diagonalization of the Hamiltonian by performing a
Bogoliubov transformation to a set of canonical Bose operators, γµα , where the index
µ = 1 . . . Ns , where Ns is the number of lattice sites. These are related to the biα by
X †

biα = Uiµ (τ )γµα − Viµ∗ (τ )εαβ γµβ . (4.32)
µ
The Ns ×Ns matrices Uiµ (τ ), Viµ (τ ) perform the Bogoliubov transformation, and obey
the following identities: (Sachdev, 1992)
! ! !
λv −Qv U U
= ω̂
−Qv∗ −λv V V
U †U − V †V = 1
U U † − V ∗V T = 1
V T U + UT V = 0
U V † + V ∗ U T = 0, (4.33)
where ω̂ is a diagonal matrix containing the excitation energies of the Bogoliubov

quasiparticles, and all quantities in Eq. (4.33) have an implicit τ dependence.
We can use the above transformations to write down the instantaneous (unnor-
malized) wavefunction of the vison as the unique state which obeys γµα |Ψv i = 0 for
all µ, α: !
X
|Ψv i = exp fijv εαβ b†iα b†jβ |0i, (4.34)
i<j
where
X
fijv = U −1† V†

iµ µj
. (4.35)
µ
kagome lattice 84
Then the Berry phase accumulated during the τ variation of Hb (τ ) is
i Im hΨv | dτd |Ψv i

† −1 dU −1 dV
= i Im Tr V V U −V . (4.36)
hΨv |Ψv i dτ dτ
We now assume that the Hamiltonian in Eq. (4.31) preserves time-reversal sym-
metry. Then, we can always choose a gauge in which the Qij , Uiµ and Viµ are all
real. Under these conditions, the expression in Eq. (4.36) vanishes identically. It is
clear that this argument generalizes to the case where we project the wavefunction
to boson states which obey the constraint in Eq. (4.2).
We have now shown that no instantaneous Berry phase is accumulated during
the vison motion of Fig. 4.4. Under these conditions, the total gauge-invariant Berry
phase is simply equal to the phase difference between the wavefunctions in the initial
and final states. (Read and Sachdev, 1991) As shown in Fig. 4.4, this phase difference
is π.
4.B Effective Ising models for visons on various

other lattice geometries
In this Appendix we summarize the results of a Ginzburg-Landau analysis of
frustrated transverse field Ising models (TIMs) on various lattice geometries. These
models describe the low energy properties of different frustrated Heisenberg antiferro-
magnets in terms of their vison exciations. Some of these results have been discussed
previously in the literature.
4.B.1 Triangular lattice

The vison excitations of a Heisenberg antiferromagnet on the triangular lattice are
described by a frustrated TIM on the dual honeycomb lattice, which has been studied
previously by Moessner and Sondhi (Moessner and Sondhi, 2001b). A Ginzburg-
Landau analysis reveals four critical modes and the corresponding PSG transfor-
mation matrices generate a 288 element subgroup of O(4) which is isomorphic to
kagome lattice 85
Figure 4.13: Confining phase on the triangular lattice. Plotted is the gauge invariant
bond-strength Jij φi φj for nearest neighbor bonds on the frustrated honeycomb lattice,
shown on the corresponding triangular lattice bonds. Black lines represent satisfied
bonds (−Jij φi φj < 0). There are no frustrated bonds in the confining phase. The
thickness of the bonds is proportional to the bond-strength. The two different patterns
arise due to a crossover when the sign of the O(4)-breaking term in the GL-functional
changes.
(C3 × GL(2, Z3 )) n C2 , (GAP) where Cn denotes the cyclic group of order n. PSG
matrices for a specific gauge choice can be found in Ref. Moessner and Sondhi (2001b).
An O(4)-breaking term appears at sixth order in the GL-functional, the minimization
of which gives rise to a single confined phase with a 24-site unit cell (i.e. a 12-site unit
cell on the triangular lattice) that is symmetric under 2π/3-rotations and reflections.
Bond patterns of this phase are shown in Fig. 4.13. Note that there is a transition
when the sign of the O(4)-breaking term changes.
4.B.2 Honeycomb lattice

For the frustrated honeycomb lattice antiferromagnet the visons are described by
an antiferromagnetic TIM on the dual triangular lattice. There are two critical modes
at the Brillouin zone edges Q = ±(4π/3, 0) and the PSG matrices corresponding to
translations T by any basis vector, rotations R6 and reflections Iy about the y-axis
of the two mode amplitudes are given by
√
T = − 12 + i 3 σz /2 ,

R6 = Iy = σx (4.37)
kagome lattice 86
where σi denote the Pauli matrices. These PSG matrices generate the 6 element
dihedral group D3 , i.e. the symmetry group of the equilateral triangle. The invari-
ant Ginzburg-Landau functional has been discussed previously by Blankenschtein et
al. (Blankschtein et al., 1984b), who showed that an O(2) symmetry breaking term
appears at sixth order. (Xu and Balents, 2011) Minimizing the GL functional gives
rise to only one possible confining phase which breaks the translational symmetry.
For a particular sign of the sixth order term, (Xu and Balents, 2011) the confining
phase has a three-site unit cell (i.e. six sites per unit cell on the honeycomb lattice)
and is symmetric both with respect to rotations and reflections; the corresponding
dimer pattern on the honeycomb lattice has the maximal number of one perfectly
flippable hexagon per six-site unit cell (Moessner and Sondhi, 2001b) and is identical
to the VBS state found in Ref. Read and Sachdev (1990). A plaquette-like phase is
obtained for the other sign of the sixth-order term. (Xu and Balents, 2011) More com-
plex minima structure for the vison dispersion have also been considered in Ref. Xu
and Balents (2011).
4.B.3 Square lattice

The effective vison model for the frustrated square lattice Heisenberg antiferro-
magnet is a frustrated TIM on the dual square lattice. (Jalabert and Sachdev, 1991;
Balents et al., 2005) In an appropriate gauge the two critical modes appear at zero
momentum and the corresponding (gauge-dependent) PSG-matrices for translations
Tx , rotations R4 and reflections Ix take the form
√
Tx = σx + σz )/ 2 , R 4 = σz , Ix = 12 . (4.38)
These matrices generate the 16 element dihedral group D8 . An invariant GL-polynomial

that breaks the O(2) symmetry appears only at eight order, as discussed by Blanken-
schtein et al. (Blankschtein et al., 1984a). Depending on the sign of this eight
order term, two different confining phases are possible. Both phases are reflection
symmetric, break the translational symmetries and have a four-site unit cell. One of
the two phases is invariant under π/2-rotations, whereas the other one has a reduced
kagome lattice 87
rotational symmetry and is only invariant with respect to π-rotations. These are the
familiar ‘plaquette’ and ‘columnar’ VBS states. (Senthil et al., 2004b,a)
More complex vison dispersion structures, with further-neighbor couplings, have
been described recently in Ref. Xu and Balents (2011).
Bibliography
Affleck, I. and Marston, J., 1988. Large-n limit of the Heisenberg-Hubbard model:
Implications for high-Tc superconductors. Physical Review B, 37:3774.
Anderson, P., 1973. Resonating valence bonds: a new kind of insulator? Materials
Research Bulletin, 8(2):153.
Arovas, D. P. and Auerbach, A., 1988. Functional integral theories of low-dimensional

quantum Heisenberg models. Phys. Rev. B, 38(1):316.
Auerbach, A. and Arovas, D. P., 1988. Spin Dynamics in the Square-Lattice Antifer-
romagnet. Phys. Rev. Lett., 61(5):617.
Balents, L., Bartosch, L., Burkov, A., et al., 2005. Putting competing orders in their
place near the Mott transition. Phys. Rev. B, 71(14):144508.
Balents, L., Fisher, M. P. A., and Nayak, C., 1998. Nodal Liquid Theory of the
Pseudo-Gap Phase of High-Tc Superconductors. International Journal of Modern
Physics B, 12:1033.
Blankschtein, D., Ma, M., and Berker, A. N., 1984a. Fully and partially frus-
trated simple-cubic Ising models: Landau-Ginzburg-Wilson theory. Phys. Rev. B,
30(3):1362.
Blankschtein, D., Ma, M., Berker, A. N., et al., 1984b. Orderings of a stacked frus-
trated triangular system in three dimensions. Phys. Rev. B, 29(9):5250.
Brézin, E., Guillou, J. C. L., and Zinn-Justin, J., 1976. In Domb, C. and Green, M. S.,
eds., Phase Transitions and Critical Phenomena, volume 6 (Academic, London).
Calabrese, P., Pelissetto, A., and Vicari, E., 2002. Critical structure factors of bilinear
fields in O(N ) vector models. Phys. Rev. E, 65(4):046115.
Campostrini, M., Hasenbusch, M., Pelissetto, A., et al., 2001. Critical behavior of the
three-dimensional XY universality class. Phys. Rev. B, 63(21):214503.
88
Bibliography 89
Cépas, O., Fong, C. M., Leung, P. W., and Lhuillier, C., 2008. Quantum phase transi-
tion induced by Dzyaloshinskii-Moriya interactions in the kagome antiferromagnet.
Phys. Rev. B, 78(14):140405.
Chubukov, A. V., Sachdev, S., and Ye, J., 1994a. Theory of two-dimensional quantum
Heisenberg antiferromagnets with a nearly critical ground state. Phys. Rev. B,
49(17):11919.
Chubukov, A. V., Senthil, T., and Sachdev, S., 1994b. Universal magnetic proper-
ties of frustrated quantum antiferromagnets in two dimensions. Phys. Rev. Lett.,
72(13):2089.
Chubukova, A. V., Sachdevb, S., and Senthilb, T., 1994. Quantum phase transitions
in frustrated quantum antiferromagnets. Nuclear Physics B, 3:601.
Conway, J. and Smith, D., 2003. On quaternions and octonions: their geometry,
arithmetic, and symmetry (AK Peters / CRC Press).
Du Val, P., 1964. Homographies, quaternions, and rotations (Clarendon Press Ox-
ford).
Elhajal, M., Canals, B., and Lacroix, C., 2002. Symmetry breaking due
to Dzyaloshinsky-Moriya interactions in the kagomé lattice. Phys. Rev. B,
66(1):014422.
Evenbly, G. and Vidal, G., 2010. Frustrated Antiferromagnets with Entanglement

Renormalization: Ground State of the Spin-12 Heisenberg Model on a Kagome
Lattice. Phys. Rev. Lett., 104(18):187203.
GAP, 2008. GAP – Groups, Algorithms, and Programming, Version 4.4.12. The
GAP Group.
Gregor, K. and Motrunich, O. I., 2008. Nonmagnetic impurities in the spin- 1?2
kagome antiferromagnet. Phys. Rev. B, 77(18):184423.
Grover, T. and Senthil, T., 2009. Quantum phase transition from an antiferromagnet
to a spin liquid in a metal. ArXiv:0910.1277v1.
Helton, J. S., Matan, K., Shores, M. P., et al., 2007. Spin Dynamics of the
Spin-1/2 Kagome Lattice Antiferromagnet ZnCu3 (OH)6 Cl2 . Phys. Rev. Lett.,
98(10):107204.
Helton, J. S., Matan, K., Shores, M. P., et al., 2010. Dynamic Scaling in the Sus-
ceptibility of the Spin-12 Kagome Lattice Antiferromagnet Herbertsmithite. Phys.
Rev. Lett., 104(14):147201.
Bibliography 90
Hermele, M., Ran, Y., Lee, P. A., and Wen, X.-G., 2008. Properties of an algebraic
spin liquid on the kagome lattice. Phys. Rev. B, 77(22):224413.
Hinkov, V., Bourges, P., Pailhés, S., et al., 2007. Spin dynamics in the pseudogap
state of a high-temperature superconductor. Nature Physics, 3:780.
Hinkov, V., Haug, D., Fauqué, B., et al., 2008. Electronic Liquid Crystal State in the
High-Temperature Superconductor Y Ba2 Cu3 O6.45 . Science, 319(5863):597.
Hinkov, V., Pailhés, S., Bourges, P., et al., 2004. Two-dimensional geometry of
spin excitations in the high-transition-temperature superconductor Y Ba2 Cu3 O6+x .
Nature, 430:650.
Imai, T., Nytko, E. A., Bartlett, B. M., et al., 2008. Cu63 , Cl35 , and H 1 NMR in the
S = 12 Kagome Lattice ZnCu3 (OH)6 Cl2 . Phys. Rev. Lett., 100(7):077203.
Iqbal, Y., Becca, F., and Poilblanc, D., 2011a. Projected wave function study of Z2
spin liquids on the kagome lattice for spin-1/2 quantum Heisenberg antiferromagnet.
Arxiv preprint arXiv:1105.0341.
Iqbal, Y., Becca, F., and Poilblanc, D., 2011b. Valence-bond crystal in the extended
kagome spin-1/2 quantum Heisenberg antiferromagnet: A variational Monte Carlo
approach. Physical Review B, 83(10):100404.
Isakov, S. V., Senthil, T., and Kim, Y. B., 2005. Ordering in Cs2 CuCl4 : Possibility
of a proximate spin liquid. Phys. Rev. B, 72(17):174417.
Jalabert, R. A. and Sachdev, S., 1991. Spontaneous alignment of frustrated bonds in

an anisotropic, three-dimensional Ising model. Phys. Rev. B, 44(2):686.
Jiang, H. C., Weng, Z. Y., and Sheng, D. N., 2008. Density Matrix Renormaliza-
tion Group Numerical Study of the Kagome Antiferromagnet. Phys. Rev. Lett.,
101(11):117203.
Kim, E.-A., Lawler, M. J., Oreto, P., et al., 2008. Theory of the nodal nematic
quantum phase transition in superconductors. Phys. Rev. B, 77(18):184514.
Kivelson, S. A., Fradkin, E., and Emery, V. J., 1998. Electronic liquid-crystal phases
of a doped Mott insulator. Nature, 393:550.
Kogut, J., 1979. An introduction to lattice gauge theory and spin systems. Reviews
of Modern Physics, 51(4):659.
Laeuchli, A. and Lhuillier, C., 2009. Dynamical Correlations of the Kagome S = 1/2
Heisenberg Quantum Antiferromagnet. ArXiv:0901.1065v1.
Bibliography 91
Lawler, M. J., Fritz, L., Kim, Y. B., and Sachdev, S., 2008. Theory of Néel and
Valence-Bond Solid Phases on the Kagome Lattice of Zn Paratacamite. Phys. Rev.
Lett., 100(18):187201.
Lee, S.-H., Kikuchi, H., Qiu, Y., et al., 2007. Quantum-spin-liquid states in the
two-dimensional kagome antiferromagnets Znx Cu4−x (OD)6 Cl2 . Nature Materials,
6:853.
Lu, Y.-M., Ran, Y., and Lee, P. A., 2011. Z2 spin liquids in the S = 12 Heisen-
berg model on the kagome lattice: A projective symmetry-group study of Schwinger
fermion mean-field states. Phys. Rev. B, 83(22):224413.
Marston, J. and Zeng, C., 1991. Spin-Peierls and spin-liquid phases of Kagomé
quantum antiferromagnets. Journal of Applied Physics, 69(8):5962.
Matan, K., Ono, T., Fukumoto, Y., et al., 2010. Pinwheel valence-bond solid and
triplet excitations in the two-dimensional deformed kagome lattice. Nature Physics,
6:865.
Mendels, P., Bert, F., de Vries, M. A., et al., 2007. Quantum Magnetism in the Parat-
acamite Family: Towards an Ideal Kagomé Lattice. Phys. Rev. Lett., 98(7):077204.
Messio, L., Cépas, O., and Lhuillier, C., 2010. Schwinger-boson approach to the
kagome antiferromagnet with Dzyaloshinskii-Moriya interactions: Phase diagram
and dynamical structure factors. Phys. Rev. B, 81(6):064428.
Moessner, R. and Sondhi, S. L., 2001a. Ising models of quantum frustration. Phys.
Rev. B, 63(22):224401.
Moessner, R. and Sondhi, S. L., 2001b. Resonating Valence Bond Phase in the Tri-
angular Lattice Quantum Dimer Model. Phys. Rev. Lett., 86(9):1881.
Moessner, R., Sondhi, S. L., and Fradkin, E., 2001. Short-ranged resonating valence
bond physics, quantum dimer models, and Ising gauge theories. Phys. Rev. B,
65(2):024504.
Nikolic, P. and Senthil, T., 2003. Physics of low-energy singlet states of the Kagome
lattice quantum Heisenberg antiferromagnet. Phys. Rev. B, 68(21):214415.
Ofer, O. and Keren, A., 2009. Symmetry of the spin Hamiltonian for herbertsmithite:
A spin- 12 kagome lattice. Phys. Rev. B, 79(13):134424.
Ofer, O., Keren, A., Nytko, E. A., et al., 2006. Ground state and excitation prop-
erties of the quantum kagom system ZnCu3 (OH)6 Cl2 investigated by local probes.
ArXiv:cond-mat/0610540v2.
Bibliography 92
Olariu, A., Mendels, P., Bert, F., et al., 2008. O17 NMR Study of the Intrinsic Mag-
netic Susceptibility and Spin Dynamics of the Quantum Kagome Antiferromagnet
ZnCu3(OH)6Cl2. Phys. Rev. Lett., 100(8):087202.
Poilblanc, D., Mambrini, M., and Schwandt, D., 2010. Effective quantum dimer
model for the kagome Heisenberg antiferromagnet: Nearby quantum critical point
and hidden degeneracy. Phys. Rev. B, 81(18):180402.
Read, N. and Chakraborty, B., 1989. Statistics of the excitations of the resonating-
valence-bond state. Phys. Rev. B, 40(10):7133.
Read, N. and Sachdev, S., 1989a. Some features of the phase diagram of the square
lattice SU(N) antiferromagnet. Nuclear Physics B, 316:609.
Read, N. and Sachdev, S., 1989b. Valence-bond and spin-Peierls ground states of
low-dimensional quantum antiferromagnets. Phys. Rev. Lett., 62(14):1694.
Read, N. and Sachdev, S., 1990. Spin-Peierls, valence-bond solid, and Néel ground
states of low-dimensional quantum antiferromagnets. Phys. Rev. B, 42(7):4568.
Read, N. and Sachdev, S., 1991. Large-N expansion for frustrated quantum antifer-
romagnets. Physical review letters, 66(13):1773.
Rigol, M. and Singh, R. R. P., 2007. Magnetic Susceptibility of the Kagome Antifer-
romagnet ZnCu3 (OH)6 Cl2 . Phys. Rev. Lett., 98(20):207204.
Rokhsar, D. and Kivelson, S., 1988. Superconductivity and the quantum hard-core
dimer gas. Physical review letters, 61(20):2376.
Rozenberg, M. J. and Chitra, R., 2008. Disorder effects in the quantum kagome
antiferromagnet ZnCu3(OH)6Cl2. Phys. Rev. B, 78(13):132406.
Sachdev, S., 1992. Kagomé-and triangular-lattice Heisenberg antiferromagnets: Or-

dering from quantum fluctuations and quantum-disordered ground states with un-
confined bosonic spinons. Physical Review B, 45(21):12377.
Sachdev, S., 1999. Quantum Phase Transitions (Cambridge University Press).
Sachdev, S., 2010. Quantum phase transitions of antiferromagnets and the cuprate
superconductors [ Lectures at the Les Houches School on Modern theories of corre-
lated electron systems]. ArXiv:1002.3823v3.
Sachdev, S. and Read, N., 1991. Large N Expansion for Frustrated and Doped Quan-
tum Antiferromagnets. International Journal of Modern Physics B, 5:219. Cond-
mat/0402109.
Bibliography 93
Sachdev, S. and Vojta, M., 2000. Translational symmetry breaking in two-dimensional

antiferromagnets and superconductors. J Phys Soc Jpn, 69(Supplement B):1.
Schwandt, D., Mambrini, M., and Poilblanc, D., 2010. Generalized hard-core
dimer model approach to low-energy Heisenberg frustrated antiferromagnets: Gen-
eral properties and application to the kagome antiferromagnet. Phys. Rev. B,
81(21):214413.
Sengupta, K., Isakov, S. V., and Kim, Y. B., 2006. Superfluid-insulator transitions of
bosons on the kagome lattice at noninteger fillings. Phys. Rev. B, 73(24):245103.
Senthil, T., Balents, L., Sachdev, S., et al., 2004a. Quantum criticality beyond the
Landau-Ginzburg-Wilson paradigm. Physical Review B, 70(14):144407.
Senthil, T. and Fisher, M. P. A., 2000. Z2 gauge theory of electron fractionalization

in strongly correlated systems. Phys. Rev. B, 62(12):7850.
Senthil, T., Vishwanath, A., Balents, L., et al., 2004b. Deconfined quantum critical
points. Science, 303(5663):1490.
Singh, R. R. P. and Huse, D. A., 2007. Ground state of the spin-1/2 kagome-lattice
Heisenberg antiferromagnet. Phys. Rev. B, 76(18):180407.
Strassler, M. J., 2003. An Unorthodox Introduction to Supersymmetric Gauge Theory.

ArXiv:hep-th/0309149v1.
Tay, T. and Motrunich, O., 2011. Variational study of J1-J2 Heisenberg model on
Kagome lattice using projected Schwinger boson wave functions. Arxiv preprint
arXiv:1103.4429.
Tchernyshyov, O., Moessner, R., and Sondhi, S., 2006. Flux expulsion and greedy
bosons: Frustrated magnets at large N. EPL (Europhysics Letters), 73:278.
Toledano, J.-C., Michel, L., Toledano, P., and Brezin, E., 1985. Renormalization-
group study of the fixed points and of their stability for phase transitions with four-
component order parameters. Phys. Rev. B, 31(11):7171.
Tovar, M., Raman, K. S., and Shtengel, K., 2009. Dzyaloshinskii-Moriya interactions
in valence-bond systems. Phys. Rev. B, 79(2):024405.
Vicari, E., 2007. Critical phenomena and renormalization-group flow of multi-

parameter Φ4 field theories. Arxiv preprint arXiv:0709.1014.
Vojta, M., Zhang, Y., and Sachdev, S., 2000a. Competing orders and quantum criti-
cality in doped antiferromagnets. Phys. Rev. B, 62(10):6721.
Bibliography 94
Vojta, M., Zhang, Y., and Sachdev, S., 2000b. Quantum Phase Transitions in d-Wave
Superconductors. Phys. Rev. Lett., 85(23):4940.
Vojta, M., Zhang, Y., and Sachdev, S., 2000c. Renormalization group analysis of
quantum critical points in d-wave superconductors. International Journal of Modern
Physics B, 14:3719.
Vojta, M., Zhang, Y., and Sachdev, S., 2008. Erratum: Quantum Phase Transitions
in d-Wave Superconductors [Phys. Rev. Lett. 85, 4940 (2000)]. Phys. Rev. Lett.,
100(8):089904.
Wang, F. and Vishwanath, A., 2006. Spin-liquid states on the triangular and Kagomé
lattices: A projective-symmetry-group analysis of Schwinger boson states. Phys.
Rev. B, 74(17):174423.
Wang, F. and Vishwanath, A., 2009. Z2 spin-orbital liquid state in the square lattice
Kugel-Khomskii model. Phys. Rev. B, 80(6):064413.
Wen, X., 1991. Mean-field theory of spin-liquid states with finite energy gap and
topological orders. Physical Review B, 44(6):2664.
Wen, X.-G., 2002. Quantum orders and symmetric spin liquids. Phys. Rev. B,
65(16):165113.
Xu, C. and Balents, L., 2011. Quantum Phase Transitions around the Staggered
Valence Bond Solid. Arxiv preprint arXiv:1103.1638.
Xu, C. and Sachdev, S., 2009. Global phase diagrams of frustrated quantum anti-
ferromagnets in two dimensions: Doubled Chern-Simons theory. Phys. Rev. B,
79(6):064405.
Yan, S., Huse, D., and White, S., 2011. Spin-Liquid Ground State of the S= 1/2
Kagome Heisenberg Antiferromagnet. Science, 332(6034):1173.
Yang, B.-J. and Kim, Y. B., 2009. Valence bond solid phases on deformed kagome
lattices: Application to Rb2 Cu3 SnF12 . Phys. Rev. B, 79(22):224417.
Zorko, A., Nellutla, S., van Tol, J., et al., 2008. Dzyaloshinsky-Moriya Anisotropy
in the Spin-1/2 Kagome Compound ZnCu3 (OH)6 Cl2 . Phys. Rev. Lett.,
101(2):026405.

Yejin Huh

Uploaded by

Copyright:

Available Formats

Yejin Huh

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Yejin Huh

Uploaded by

Copyright:

Available Formats

Quantum Phase Transitions

All rights reserved.

Quantum Phase Transitions in d-wave Superconductors

2 Renormalization group theory of nematic ordering

3 Quantum criticality of the kagome antiferromagnet

4 Vison states and confinement transitions of Z2 spin liquids on the

Chapter 2 has been previously published as

“Renormalization group theory of nematic ordering in d -wave supercon-

Chapter 3 has been previously published as

“Quantum criticality of the kagome antiferromagnet with Dzyaloshinskii-

Chapter 4 has been previously published as

“Vison states and confinement transitions of Z2 spin liquids on the kagome

Electronic preprints (shown in typewriter font) are available on the Internet at

1.1 d-wave Superconductivity and Nematic Order-

order wave vector QAF = (π, π).

Figure 1.2: (Hinkov et al., 2008) Spin excitation spectra at T = 5K at 3, 7, and

1.2 Phase Transitions in Kagome Lattice Z2 Spin

1.2.1 Schwinger Boson

with  a fully anti-symmetric tensor.

We can describe the Hamiltonian as

where J = Jij > 0 for nearest neighbors i and j.

up to an additive constant, where J is the Sp(N ) generalization of .

1.2.2 Z2 Spin Liquid

In the Schwinger boson description of spins there is a local U (1) transformation,

brσ → eiφ(r) brσ , (1.10)

1.2.3 Quantum Dimer Model

This quantum dimer model can be derived as a perturbation expansion of the

1.2.4 Ising Gauge Theory

1.2.5 Duality of Ising Gauge Theory and Ising Model in

1.3 Organization of the thesis

In Chapter 3 we investigate the zero-temperature phase diagram of the nearest-

Renormalization group theory of

detwinned crystals of YBa2 Cu3 O6.45 .

2.2 Field theory

The action for the field theory, S, has three components

S = SΨ + Sφ0 + SΨφ . (2.1)

where λ0 is a coupling constant.

2.2.1 Large Nf expansion

Γ2 (K) = Π2 (kx , ky , ω) + Π2 (ky , kx , ω) (2.7)

Γ4 (K, 0, −K, 0) = Π4b (kx , ky , ω) + Π4b (ky , kx , ω) (2.11)

the self-energy is given by the Feynman graph in Fig. 2.3:

2.3 Renormalization group analysis

the vertex. Thus let us write

ηb = 1 + 2C4 − 2ηf = 1 + 2C4 + 2C1 (2.22)

while the velocity v∆ obeys

0.5 1.0 1. 5 2.0

Figure 2.4: Plots of the functions C1−3 in Eq. (2.35).

at leading order in the 1/Nf expansion.

0.5 1.0 1.5 2.0

0.2 0.4 0.6 0.8 1.0

0.2 0.4 0.6 0.8 1.0

0.2 0.4 0.6 0.8 1.0

2.4 Theory for small v∆/vF

in φ we have the following effective action as v∆ → 0 (at criticality, r = 0):

Quantum criticality of the kagome

Zorko et al., 2008; Ofer and Keren, 2009).

Figure 3.1: Staggered DM interaction from triangle to triangle in z-direction as in

3.2 Mean field theory

with a fully anti-symmetric tensor.

up to an additive constant, where J is the Sp(N ) generalization of .

− Jqσσ0 (Uσ Vσ0 + Vσ Wσ0 + Wσ Uσ0 ) + c.c.

• Under time-reversal, we have Uσ → σσ0 Uσ∗0 , and similarly for Vσ , Wσ . This is

Zσ → iZσ , Yσ → −iYσ , Xσ → σσ0 Xσ∗0 . (3.19)