Basics of Quantum Spin Liquids

Thomas Johnson
May 10, 2018

Abstract
A quantum spin liquid is a state of matter in which the spins are
highly entangled and don’t order, even at T = 0. They are predicted
to exhibit fractional particle excitations and emergent gauge fields.
In this essay, I introduce the basics of quantum spin liquids with an
emphasis on their universal features, and cover some experiments done
on ZnCu3 (OH)6 Cl2 .

1
1 Introduction
A spin liquid is a state of matter in which the spins are correlated, yet
fluctuate strongly, even at low temperatures [1]. For example, consider an
Ising antiferromagnet on a hexagonal lattice (see figure 1). Only two of the
spins can be antiparallel, leading to six degenerate ground states; thermal
fluctuations can drive transitions between these states. This is an example
of a classical spin liquid, i.e., a spin ice. Spin ices have been realized ex-
perimentally (e.g., Dy2 Ti2 O7 ) and exhibit magnetic monopoles. At very low
temperatures, however, other terms may become relevant in the Hamiltonian
(e.g., a stray field) which cause the spins to either order or form a spin glass.
It’s also possible to have quantum fluctuations,
which are fluctuations due to the commutation re-
lations of spin operators (i.e., we can no longer rep-
resent the spins as vectors). These fluctuations can
be strong enough to prevent ordering even at T=0
[1]. If, in addition to a lack of order, the spins are
also highly entangled, then we have a quantum spin
liquid (QSL) [2]. These states are expected to have
fractional, non-local, quasiparticle excitations and
emergent gauge fields. Since these states never or-
der, they have no broken symmetries and are not
described by Landau’s theory of phase transitions.

Figure 1: AF Ising spins

on a triangle. Figure is 1.1 History
from [1]. In 1973, Anderson came up with the idea of QSLs
[7]. When two spins interact antiferromagnetically,
they can pair into a singlet state, forming a valence
bond [1]. If all of the spins in a system form valence
bonds, the ground state is given by the product of
the valence bonds, i.e., a valence bond solid (VBS) (see figure 2). VBS states
are only short-range entangled and can order (e.g, by aligning along an axis),
so they are not QSLs. A superposition of VBS states, which Pauling named
a resonating valence bond (RVB), could form a QSL, however, and Anderson
proposed this to be the ground state of the antiferromagnetic Heisenberg
model on a hexagonal lattice [7]. This turned out to be incorrect, as the
spins order at 120o [3].

2
Figure 2: A valence bond solid (VBS) state. The ovals are valence bonds (paired
spins). Figure is from [4].

In 1986, Bednorz and Mueller discovered the high-temperature copper

oxide superconductors, and in 1987 Anderson proposed that La2 CuO4 had
an RVB ground state [8]. This sparked interest in QSLs, and over the next few
years QSLs were connected to topology [2]. In 1989, Wen created our current
understanding of topological order, when he realized that many different spin
liquids have the same symmetry, and that a new order was needed beyond the
Landau paradigm [5, 6]. While some QSLs have topological order, many do
not [2]. In 2000, Senthil and Fisher discovered QSL quasiparticles could have
fractional quantum numbers (i.e, fractionalization). In 2003 Kitaev created
his toric code model, and in 2006 created his honeycomb model: these models
are exactly solvable and have QSL ground states. This has led to great effort
to realize QSL states experimentally, an effort which continues to this day.
While there are several QSL candidates, there is no definitive evidence that
they are QSLs [2].
In the following sections I shall attempt to elucidate on the terms I’ve
introduced (e.g., topological order), as well as explain the phenomenology of
QSLs without focusing on any particular model. I will focus on U(1) gauge
fields, as that is electromagnetism on a lattice and therefore familiar. This
term paper is based mostly on the 2010 review by Balents [1] and the 2017
review by Savary and Balents [2], which is relatively accessible and recom-
mended to anyone interested in this topic. Additionally, ref. [11] is a book
(written in 2011) that has several chapters on QSLs, as well as chapters on
experimental techniques such as neutron scattering, NMR, and µSR.

3
2 Theory
We begin with the parton approach to QSLs, often called the slave-boson
or slave-fermion approach. Anderson used this method in 1987 [8]; it has
developed since then and is reviewed in refs. [2, 4, 5]. In this approach, we
break the spins into parts (partons) which we hope become quasiparticles
upon substitution in the Hamiltonian [2]. This isn’t necessarily true, but it
works exactly in Kitaev’s honeycomb model. These partons are chosen to be
Schwinger bosons or Abrikosov fermions. Let us use the fermion representa-
tion, which is simpler because it avoids Bose Einstein condensation. On our
lattice, the spins reside on the bonds between lattice points. We can represent
a spin S~ operator as
X1
~=
S fα† ~σαβ fβ (1)
αβ
2
with constraint α |fα |2 = 1. Assuming they become quasiparticles, these
P
Abrikosov fermions are called spinons and are spin 1/2 with no electric
charge. Excitations of conventional phases have integer spin, so spinons are
fractional quasiparticles.
Upon substitution of equation (1) in a Hamiltonian, which for a Heisen-
berg model depends on terms like S ~1 · S~2 , we end up with four-fermion terms.
To create an effective quadratic Hamiltonian, theorists use mean field theory.
This involves many possible groupings of operators; the decoupling scheme is
chosen beforehand based on symmetry and energy [2]. Following Savary and
Balents[2], we shall just assume the effective Hamiltonian is the Bogoliubov
Hamiltonian X αβ † † †
[tij fiα fjβ + ∆αβ ij fiα fjβ + h.c.]. (2)
ij

2.1 Gauge fields

This Hamiltonian has a Z2 symmetry under f → −f . This symmetry is un-
physical and occurs as a result of equation (1): the formal Abrikosov fermions
have changed, but the physical operator S ~ has not [2]. By using this equa-
tion, we have expanded our Hilbert space; to create a one-to-one mapping
between this enlarged Hilbert space and the physical Hilbert space, the gauge
symmetry must be built into the Hamiltonian by introducing a lattice gauge
field [5]. The type of gauge symmetry (e.g., Z2 , U(1), or SU(2)) depends on
t and ∆.

4
 Suppose ∆ = 0. Then equation (2) is invariant under f → f eiχ (still
unphysical). Introducing a U(1) lattice gauge field, our Hamiltonian becomes
X αβ †
[tij eiAij fiα fjβ + h.c.] + Hg (3)
ij

where Hg is a Hamiltonian associated with the lattice gauge field [2].

A U(1) lattice gauge theory is just electromagnetism on a lattice [2]. For
a cubic lattice [2],
X X V X 2
Hg = −K cos(∇ × A) + K 0 (∇ · E)2a + E (4)
p a
2 <ab> ab

where we define X
(∇ · E)a = Eab
b∈nn(a)
X
Bp = (∇ × A)p = Aa,a+1 .
a∈p

Here A and E are our ‘field’ variables which satisfy [Aab ,Eab ] = i, Aab =-Aba ,
and Eab =-Eba . The letters a and b refer to lattice points (the field variables
are defined on bonds), nn means nearest neighbor, and p refers to plaquette,
which is a unit square on the lattice.
Returning to equation (3), the Hamiltonian and its physical states are
gauge-invariant under Aij → Aij + χi − χj , f → f eiχ , and f † → f † e−iχ [2].
The generator of this symmetry is Q, where
†
Qi = (∇ · E)i − fiα fiα = −1. (5)
Since equation (3) is gauge-invariant, Q is conserved, and when E = 0, our
Abrisokov constraint (1) gives -1. According to Savary and Balents [2], this
Hamiltonian “corresponds to the mean-field approximation in the limit of
large K, in which case fluctuations of Aij are suppressed.” Excitations can
cause A to fluctuate, however, and in this case the U(1) state is stable in 3D,
but not in 2D [2].
For large K we can approximate A = 0, and the Hamiltonian can be
diagonalized [2]. This depends on the specific form of t, but in simple cases
we expect the spinons to be gapped out. In these cases we obtain a pure U(1)
gauge theory described by Hg with equation (5) replaced by Gauss’s law
Qi = (∇ · E)i .

5
Up to a constant and with renormalized couplings, the Hamiltonian is ap-
proximately [2] 1
X X V X 2
Hg = K Bp2 + K 0 Q2a + E . (6)
p a
2 <ab> ab

The Hamiltonian approximation is valid when V  K, and is called the

(pure) Coulomb phase [2].
Before moving on, there are some points I should address. This parton
method is a mean field theory, and thus qualitative. It’s possible to use vari-
ational methods to guess the ground state using partons, but I don’t cover
that: it is reviewed in refs. [2, 4]. It’s also possible for the spinons to be
gapless, but that case is more complicated and not fully solved [2].

2.2 Quasiparticles
Consider the Coulomb phase. Since Q is conserved, we can choose our eigen-
states to be eigenstates of Q. Let Q = 0. In the continuum limit, the low-
energy, long-wavelength physics is given by the electromagnetic energy
Z  1 
H = d3 x |E|2 + |B|2 .
2 2µ
By analogy to electromagnetism, there are emergent gapless photons.
When Q is non-zero, there are emergent gapped (K’) electric charges
(not literal charges). There are also magnetic charges, or monopoles, which
emerge as topological point defects in 3D [2, 9]. Both of these quasiparti-
cles are bosons, have a 1/r Coulomb potential, and are non-local [2]. By
non-local, I mean that local operators cannot create just one quasiparticle.
Rather, local operators can only create charge-neutral configurations, which
means that local operators create particles in pairs. To create a single quasi-
particle, a semi-infinite string operator, which is some semi-infinite product
of connected (i.e., on neighboring bonds) local operators, is required [2, 9].
The quasiparticles appear at the ends of this ‘string.’ The electric and mag-
netic quasiparticles can also pair, or fuse, to form a third quasiparticle, called
a dyon, which is a fermion [2].
1
Savary and Balents [2], page 11: “ A naive analysis consists of taking very small
but non-zero
P U, which we expect to introduce small fluctuations of Bp .” Here U =
exp(i a χa Qa ). I don’t understand how U is small, since it’s unitary, or how a gauge
transformation can make B non-zero.

6
2.3 Stability
The Coulomb phase in 3D is stable to weak perturbations, even those which
break the gauge symmetry [2]. To paraphrase Savary and Balents, when there
are perturbations which break the gauge symmetry yet the QSL phase sur-
vives, we say there is an emergent gauge symmetry [2]. Strong perturbations
can lead to confinement. As a reminder, isolated quasiparticles are created
by semi-infinite string operators. Confinement is when there is an energy cost
to such a string such that the energy to create an isolated quasiparticle, say
the electric quasiparticle, is infinite. When this happens the quasiparticles
can only be created in pairs and are confined to their antiparticle (oppo-
sitely charged particle). The magnetic quasiparticles are also removed, as
their energy is set to zero: they can be viewed as a condensate [2]. There is a
first-order phase transition from the unconfined phase to the confined phase,
which can be viewed as Bose condensation of either the elecric or magnetic
quasiparticles.

2.4 Topological phases

Topological phases are defined by topological invariants; for example, Ki-
taev’s toric code model has a topological ground state degeneracy [2]. Ac-
cording to Wen [6], “topological order is defined to describe gapped quantum
liquids that cannot be deformed into a product state without gap-closing
phase transitions .” Here quantum liquids refers to more than just quantum
spin liquids; for example, fractional quantum Hall states have topological
order. Topological order is due to long-range entanglement, and therefore
different from topological insulators, defects, etc. [6].
The pure Coulomb phase is not a topological phase, as all topological
phases only have gapped excitations. It does have non-local excitations, how-
ever, which indicates it is a highly entangled phase [2]. Examples of topolog-
ical QSLs include Kitaev’s toric code model, which is a Z2 gauge phase [2],
and the frustrated Coulomb phase on a pyrochlore lattice [9].

2.5 Entanglement entropy

Savary and Balents argue that the fundamental attribute of QSLs is not the
absence of symmetry breaking but the presence of a high degree of entan-
glement [2]. But what does it mean to be ‘highly’ entangled? The degree of

7
entanglement can be quantified by the von Neumann entanglement entropy.
Suppose we have a 2D system split, in real space, into parts A and B. The
von Neumann entropy is [2]

S(A) = S(B) = −TrA (ρA logρA ) ∼ s0 L − γ. (7)

Here ρA is the density operator for part A, L is the length of the boundary
between the two parts, and γ is the topological entanglement entropy; s0
and γ are constants. For a pure state, the entanglement entropy is zero. For
topologically ordered states, γ is quantized and (at least for the toric code) is
positive [2]. Since the entropy cannot be negative, it’s impossible to have s0 =
0, and thus a non-zero topological entanglement entropy indicates long-range
entanglement.
In the thermodynamic limit, however, the L term dominates. This term
is called the area law, since in 3D it is an area, not a length. Savary and
Balents [2] acknowledge that in the thermodynamic limit, “quantum-spin-
liquid states, including topologically ordered phases, are typically no more
entangled than any other ground states.” In this sense, the word ‘highly’
seems meaningless, and I think that the definition Savary and Balents propose
is more of a working definition which emphasizes the type of physics they are
interested in. They say their definition is vague because entanglement itself
is still an active research subject.
For temperatures greater than zero, the entanglement entropy is extensive
and no longer a good metric [2]. As a result, most QSLs are only well-defined
as T=0, and undergo a smooth crossover to a paramagnetic phase for T
> 0. An exception is the 3D toric code, which undergoes a second order
phase transition unassociated with any local order parameter. For T > 0,
the physics can be described in terms of a dilute gas of quasiparticles.

3 Experiment
While there are several QSL candidates (see figure 3), none have been proven
to be QSLs [2]. This is difficult to do, as QSLs have few experimental sig-
natures and conventional probes focus on local excitations. Thus, it is nec-
essary to study candidate materials with many different probes. Most QSL
candidates exhibit no symmetry breaking, which suggests there may be new
physics at play.

8
Figure 3: Some QSL candidates. The chart is from Balents 2010 [1] and is slightly
out of date. Na4 Ir3 O8 has AF order, but may still exhibit residual QSL physics
[2].

When looking for new QSLs, there are several features people look for
[2]. First, frustration is necessary to avoid ordering. This frustration could be
geometric, or it could be due to something else such as next-nearest-neighbor
interactions. Second, people look for spin 1/2 systems, since for Heisenberg
models the commutator of two spin operators is of order 1/S, which means
quantum fluctuations are stronger for smaller spins [2]. But this may not be
true for frustrated systems, and theorists are constructing higher spin models
[2, 4]. Third, people look for proximity to a Mott transition, as this leads to
more quantum fluctuations [2].

3.1 Herbertsmithite
I shall focus on ZnCu3 (OH)6 Cl2 , named herbertsmithite, as this is one of the
most well-known and well-characterized QSL candidates. A 2010 experimen-
tal review on powder samples is given in ref. [10]; single crystals were grown
in 2011, and brief reviews as of 2017 are given in refs. [2, 4]. This material has
Cu ions in a kagomé lattice: the lattice was named after a type of Japanese
basket weave (see figure 4) [3]. Spins on this lattice have a larger geometric
degeneracy compared with spins on hexagonal lattices, and thus the kagomé
lattice has a greater potential of realizing a QSL [4].
Herbertsmithite has been shown not to order down to low temperatures
by NMR, µSR, neutron scattering, and magnetic susceptibility measurements

9
[2]. First I shall discuss µSR, which stands for muon spin rotation. Basically,
a polarized beam of positively charged muons is produced by proton-proton
collisions, then implanted into a sample. The muons then precesses about the
local magnetic field before decaying and emitting a positron. This positron
is emitted in the direction of the muon spin and can be detected. From this
we can measure the muon precession and learn about the local magnetism
of a material. Zero field µSR measurements were done by Mendels et al. in
2007, and shows no ordering down to 50 mK for herbertsmithite [10].2
NMR is another probe which allows us to learn
about local magnetism [2]. The nuclei feel the mag-
netic field of the nearby electronic spins (i.e., hyper-
fine interaction), which splits the energy levels of the
nuclui. The nuclei can be driven between the levels
using an external AC magnetic field. The resonance
frequency shifts with temperature, as the field due
to the hyperfine interaction is proportional to the
local electronic susceptibility: this is known as the
Knight shift. Thus NMR measures the local mag-
netic susceptibility.
The 17 O spectra was measured in 2008 by Olariu
et al, and the 35 Cl spectra measured that same year
by Imai et al [10]. The 17 O NMR spectra is more
sensitive to the kagomé plans, and this shows no
ordering down to 0.45 K. It levels off at low temper-
atures, which indicates herbertsmithite is gapless.
It also has a peak which suggests there is there is
Cu-Zn disorder: some Zn is in the kagomé planes,
and the displaced Cu between planes. This can also
be seen in figure 5, as the bulk magnetic suscep-
tibility doesn’t match the local susceptibility given
by NMR; the bulk susceptibility is interpreted as
Curie-Weiss behaviour of the Cu defects, which be- Figure 4: Figure is from
have as quasi-free spins. [3].
Neutrons can scatter off both atomic nuclei
and magnetic moments, giving information on both
structure and bulk magnetic order. For QSLs, elastic scattering can reveal
2
The cited review says 20 mK, which doesn’t match the actual paper.

10
low-energy excitations; this is because neutrons don’t have perfect energy
resolution [2]. These low-energy excitaitons (e.g., photons of the Coulumb
phase) are expected to be localized in momentum space. Inelastic neutron
scattering is used to study the excitations of materials; for QSLs, a smooth
continuous signal is usually expected. It’s possible for QSL quasiparticles for
pair and form a bound state, which could give a sharp peak.
For herbertsmithite, de Vries et al. did structure refinements in 2008
which show 7-10% mixing of Zn and Cu [10]. Inelastic neutron scattering
on a single crystal showed smooth spectral weight and no peaks, indicating
herbertsmithite is gapless [2]. But the inelastic signal does not agree with
any QSL model, which have features at specific momentum points.
The last probe I’ll discuss is the specific heat,
which reveals phase transitions and, when inte-
grated, can be used to determine the entropy. Spe-
cific heat measurements can be used to distinguish
gapped and gapless states: the magnetic specific
heat is exponentially suppressed for gapped states,
and obeys power laws for gapless states [2]. Different
gapless QSLs are predicted to have different scalings
[2].
The specific heat for herbertsmithite was mea-
sured measured by Helton et al. in 2007, and de
Vries et al. in 2008 [4]. The low temperature data
shows the specific heat depends on field, and has a
field-dependent peak associated with Cu-Zn disor-
der. The slope is roughly linear at low temperatures
in zero field, which is associated with gapless behav- Figure 5: NMR spectra
ior. overplotted with bulk
In summary, various probes show that herbert- susceptibility (solid
smithite doesn’t order down to low temperatures line). Figure is from
and suggest gapless behavior, but the results are not [10].
fully consistent with any model. They also show Cu-
Zn disorder, which complicates the analysis. Savary
and Balents mention a recent NMR experiment on a single crystal which
shows a gap, but that result is unpublished and needs to be reconciled with
previous experiments [2].

11
4 Conclusion
Quantum spin liquids are a state of matter which avoids symmetry breaking
and are highly entangled. These states can be approached phenomenologi-
cally through an effective Hamiltonian coupled with a gauge field, which leads
to emergent non-local quasiparticles with fractional quantum numbers. There
are also exactly solvable models, notably Kiteav’s toric code and honeycomb
model, which have QSL ground states. Gapped QSLs exhibit topological or-
der.
The progress on QSL physics has been largely theoretical, but there is
currently a great experimental search for these states. There are some QSL
candidates, but none have been proven to be QSLs. Since QSLs have few ex-
perimental signatures, their identification is difficult, and, as noted by Savary
and Balents, will likely require careful modeling [2].

References
[1] Balents L. Spin liquids in frustrated magnets. Nature. 464, 199-208 (2010).

[2] Savary L and Balents L. Quantum spin liquids: a review. Rep. Prog. Phys. 80
016502 (2017).

[3] Lee P. An End to the Drought of Quantum Spin Liquids. Science 321, 1306-
1307 (2008).

[4] Zhou Y, Kanoda K, and Ng T. Quantum spin liquid states. Rev. Mod. Phys.
89, 025003 (2017).

[5] Wen X. G. Quantum orders and symmetric spin liquids. Phys. Rev. B.. 65,
165113 (2002).

[6] Wen X. G. Colloquium: Zoo of quantum-topological phases of matter. Rev.

Mod. Phys. 89, 041004 (2017).

[7] Anderson P. W. Resonating valence bonds: a new kind of insulator? Mat. Res.
Bull. 8, 153-160 (1973).

[8] Anderson P. W. The Resonating Valence Bond State in La2 CuO4 and Super-
conductivity. Science. 235, 1196-1198 (1987).

12
[9] Hermele M, Fisher M, and Balents L. Pyrochlore photons: The U(1) spin liquid
in a S=1/2 three-dimensional frustrated magnet. Phys. Rev. B. 69, 064404
(2004).

[10] Mendels P and Bert F. Quantum Kagome Antiferromagnet ZnCu3 (OH)6 Cl2 .
J. Phys. Soc. Jpn. 79, 011001 (2010).

[11] Lacroix C, Mendels P, and Mila F (Eds.). Introduction to Frustrated Mag-

netism. Springer. (2011).

13