Work supported in part by US Department of Energy contract DE-AC02-76SF00515.

Algebraic spin liquid in an exactly solvable spin model

Hong Yao, Shou-Cheng Zhang, and Steven A. Kivelson
Department of Physics, Stanford University, Stanford, California 94305
(Dated: November 8, 2009)
We have proposed an exactly solvable quantum spin-3/2 model on a square lattice. Its ground
state is a quantum spin liquid with a half integer spin per unit cell. The fermionic excitations are
gapless with a linear dispersion, while the topological vison excitations are gapped. Moreover,
the massless Dirac fermions are stable. Thus, this model is, to the best of our knowledge, the rst
exactly solvable model of half-integer spins whose ground state is an algebraic spin liquid.
The term spin liquid is widely used to give sharp
meaning to the more general intuitive notion of a Mott
insulator; a spin liquid is an insulating state that cannot
be adiabatically connected to a band insulator, i.e. to
an insulating Slater determinant state. In a system that
preserves time reversal symmetry, any insulating state
with an odd number of electrons (or a half-integer spin)
per unit cell is a spin liquid. Interesting proposals [1, 2]
have been made concerning the relevance of such states to
the theory of high temperature superconductivity in the
cuprates and other materials. Indeed, various spin-liquid
phases have been proposed, which are distinguished by
the character of any gapless spinons and the exchange
statistics of the topological vison excitations.
Since they are new and exotic quantum phases of
matter, it is desirable to construct solvable models with
short range interactions with stable spin-liquid ground-
state phases. A breakthrough occurred when Moessner
and Sondhi [3] demonstrated the existence of a gapped
spin liquid ground state in the quantum dimer model [4],
analogous to the short range version of the RVB state
[5, 6]. An exactly solvable spin-1/2 model in a gapped Z
2
spin liquid phase was later constructed by Wen [7]. How-
ever, much of the recent interest, spurred in part by the
possible observation of such a state in -(ET)
2
Cu
2
(CN)
3
[8, 9, 10] and Zn(Cu)
3
(OH)
6
Cl
2
[11], has focussed on
spin-liquids with gapless spin excitations, so-called al-
gebraic spin liquids.
The exactly solvable Kitaev model on the honeycomb
lattice [12] can exhibit gapless excitations. However, be-
cause the honeycomb lattice has two sites per unit cell,
this model has an integer spin, hence an even number of
electrons per unit cell. In the present paper we construct
an exactly solvable model, in much the same spirit as the
Kitaev model, whose ground state is a spin liquid with
an odd number electrons per unit cell and stable gapless
fermionic excitations. To the best of our knowledge, this
is the rst exactly solvable model with this sort of spin
liquid ground state -algebraic spin liquid.
The Kitaev model has a spin-1/2 on each site of a
trivalent lattice, where the coordination number is dic-
tated by the existence of three Pauli matrices. In order to
study a model on a square lattice, we instead consider a
model with a spin-3/2 on each lattice site. The resulting
larger Hilbert space, with 4 spin polarizations per site,
permits us to express the model in terms of the 4 4 an-
ticommuting Gamma matrices,
a
(a = 1, , 5) which
form Cliord algebra,
a
,
b
= 2
ab
. Specically, the
5 Gamma matrices can be represented [13] by symmetric
bilinear combinations of the components of a spin 3/2
operator, S

, as

1
=
1

3
S
y
, S
z
,
2
=
1

3
S
z
, S
x
,
3
=
1

3
S
x
, S
y
,

4
=
1

3
[(S
x
)
2
(S
y
)
2
],
5
= (S
z
)
2

5
4
. (1)
Model Hamiltonian: We dene our model on a square
lattice, with a spin-3/2 on each site, and corresponding
matrices dened as in Eq. (1). In terms of these,
H =

i
_
J
x

1
i

2
i+ x
+J
y

3
i

4
i+ y
_
(2)
+

i
_
J

15
i

25
i+ x
+J

35
i

45
i+ y
_
J
5

5
i
,
where
ab
[
a
,
b
]/(2i) and i labels the lattice site
at r
i
= (x
i
, y
i
). We call this model the Gamma ma-
trix model (GMM). Suppose that the square lattice has
N = L
x
L
y
sites, where L
x
and L
y
are the lattices linear
sizes and are assumed, for simplicity, to be even in this
paper. Moreover, we consider periodic boundary condi-
tions. Obviously, the GMM can be written explicitly as a
spin-3/2 model. The GMM model respects translational
symmetry and time reversal symmetry (TRS). It does not
have global spin SU(2) or even U(1) rotational symmetry,
but is invariant under 180

rotations about the z axis in

spin space, i.e. it has global Ising symmetry. The lack of
any continuous symmetry implies that the fermionic ex-
citations (discussed below) do not have well-dened spin
quantum number, so their relation to spinons is un-
clear. A feature of the model which makes it solvable is
an innite set of conserved uxes: [

W
i
, H] = 0 for any
i and [

W
i
,

W
j
] = 0, where

W
i

13
i

23
i+ x

14
i+ y

24
i+ x+ y
is
the plaquette operator at site i.
Note that, in contrast to the spin-1/2 Kitaev model on
the honeycomb lattice, the GMM has an odd number (3)
of electrons per unit cell. The present model in the limit
J
x
= J
y
= J

x
= J

y
and J
5
= 0 is similar to a model pro-
posed by Wen in Ref. [14], where, however, the Gamma
SLAC-PUB-13933
SIMES, SLAC National Accelerator Center, 2575 Sand Hill Road, Menlo Park, CA 94309
2
matrices were constructed from two spin-1/2 operators
on each site, and therefore behave dierently under time
reversal than in the present realization. Moreover, the
present model generically does not possess the special
symmetries of Wens model that are responsible for some
of the behaviors of that model.
Fermionic Representation: Spin-3/2 operators can be
expressed as bilinear forms involving three avors of
fermion operators, S
z
= a

a + 2b

b 3/2 and S
+
=

3f

a +

3af +2a

b, subject to the constraint that the

physical states are only those with odd fermion parity,
(1)

N
= 1, where

N = f

f +a

a+b

b. Rather than us-

ing this representation in terms of three Dirac fermions,
we will directly represent the Gamma matrices in term
of a related set of 6 Majorana fermions:

i
= ic

i
d
i
,
5
i
= ic

i
d

i
, = 1, 2, 3, 4,
5
i
= id
i
d

i
(3)
where c

i
, d
i
, and d

i
are Majorana fermions on site i.
Six Majorana fermions form an eight dimensional Hilbert
space, which is an enlarged one from the physical Hilbert
space of an spin-3/2. In terms of spin-3/2 operators,

1
i

2
i

3
i

4
i

5
i
= 1 for all i. Consequently, all allowed
physical states [) in term of Majorana fermions must
satisfy the following constraint, for all i,
D
i
[)
_
ic
1
i
c
2
i
c
3
i
c
4
i
d
i
d

[) = [). (4)
In terms of Majorana fermions, the Hamiltonian in the
enlarged Hilbert space can be written as
H =

i
_
J
x
u
i,x
id
i
d
i+ x
+J
y
u
i,y
id
i
d
i+ y
+J

x
u
i,x
id

i
d

i+ x
+J

y
u
i,y
id

i
d

i+ y
J
5
id
i
d

i
_
, (5)
where u
i,x
ic
1
i
c
2
i+ x
and u
i,y
ic
3
i
c
4
i+ y
. It is obvious
that u
i,
are conserved quantities with eigenvalues u
i,
=
1, = x, y. The enlarged Hilbert space can be divided
into sectors u. In each sector, the Hamiltonian Eq. (5)
describes free Majorana fermions:
H(u) =

i
_
J
x
u
i,x
id
i
d
i+ x
+J
y
u
i,y
id
i
d
i+ y
+J

x
u
i,x
id

i
d

i+ x
+J

y
u
i,y
id

i
d

i+ y
J
5
id
i
d

i
_
, (6)
where u
i,
are emergent Z
2
gauge elds. The Z
2
gauge
transformations are given by d
i

i
d
i
, d

i

i
d

i
, and
u
i,

i
u
i,

i+
, where
i
= 1. In the enlarged
Hilbert space, the eigenstates [) = [)
c
[)
d,d
can
be written as a direct product of a part that involves
the c

fermions and a part that involves the d and d

fermions, respectively. Here [)

c
is dened by u
i,
[)
c
=
u
i,
[)
c
and [)
d,d
are eigenstates of Eq. (6) with the
corresponding u
i,
s.
The spectrum of H(u) depends only on gauge invari-
ant quantities - the ux on local plaquettes, exp(i
i
)
u
i,x
u
i+ x,y
u
i,y
u
i+ y,x
, and two global uxes exp(i
x
)

i(yi=1)
u
i,x
and exp(i
y
)

i(xi=1)
u
i,y
, where
i
and

x,y
= 0, . [Note that the previously dened W
i
=
exp(i
i
).] It is obvious that

i
= 0 (mod 2), so
there are N 1 independent local uxes. Including the
two global uxes, the number of independent uxes is
N + 1. Since there are 2N Z
2
gauge elds, the num-
ber of dierent gauge eld choices corresponding to each
ux sector is 2
N1
. In other words, in the enlarged
Hilbert space, each state is 2
N1
-fold degenerate. Note
that in the thermodynamic limit, the energy is indepen-
dent of the two global uxes, which gives rise to a fourfold
topological degeneracy of the physical ground states.
Projection operators: Most of the states in the en-
larged Hilbert space are not physical states. To obtain
a physical eigenstate, we must nd a linear combination
of the degenerate eigenstates which is simultaneously an
eigenstate of every D
i
with eigenvalue 1. This is realized
through the projection operator P:
[) = P[)

i
_
(1 +D
i
)/2
_
[). (7)
Clearly, Eq. (7) implies D
i
[) = [) for any i. Explicitly,
P is given by
P =
_
1 +

i
D
i
+

i1<i2
D
i1
D
i2
+ +

i
D
i
_
/2
N
. (8)
D
i
acting on a direct product state [) is equivalent
to a gauge transformation on site i. A subtlety here
is that there are 2
N
operators in the sum in Eq. (8),
but only 2
N1
inequivalent gauge transformations. In
fact, D

i
D
i
implements a gauge transformation on
every site, thus leaving all gauge elds unchanged. It fol-
lows that P = P

(1 + D), where P

includes all inequiv-

alent transformations. Since [D, H] = 0 and D
2
= 1,
D[) = [). Moreover, D =

i
_
u
i,x
u
i,y

i
_
id
i
d

.
In terms of the number

N
f
of the Dirac fermions,
f
j
i
j
(d
j
+id

j
)/2,

N
f
=

i
f

i
f
i
(9)
and uxes,

N

, we obtain D = (1)

N

(1)

N
f
.

N
f
is con-
served modulo 2 by the Hamiltonian.

N

is dened by
dividing the plaquettes into two sublattices and counting
the number of -uxes through one or the other sublat-
tice [24].
Thus, depending on the fermion and ux parity, P ei-
ther annihilates a given direct product state, [) , or
maps it to the equal weight linear superposition of all
gauge transformations acting on [). For instance, when
there is a -ux through each plaquette, which, as dis-
cussed below, is the ground state sector, D = (1)

N
f
,
i.e. all physical states must have an even number of
fermions. Conversely, where 1 -ux is added to each
3
sublattice, only states with

N
f
= odd survive projec-
tion. Conserving the parity of fermion number reects
the fact that physical fermionic excitations are created
by non-local (string) operators [14, 23].
-ux state and gapped visons: In each ux sector
, the lowest energy of the Hamiltonian is denoted
by E
0
(). The ground state energy of the model is
achieved by minimizing E
0
() with respect to .
Formally, by integrating out the fermions, an eective
action for the Z
2
gauge elds can be derived. However,
in general, it is nontrivial to obtain an explicit form of
the eective action.
For J
5
= 0, fortunately, there is a theorem due to Lieb
[15] which implies that the energy minimizing ux sector
of a half-lled band of electrons hopping on a planar,
bipartite lattice is per square plaquette. We dene
this uniform -ux, the ground state sector, as being
vortex-free. Z
2
vortex excitations, visons, are dened
to be plaquettes with
i
= 0. Due to the constraint

i
= 0 (mod 2), only an even number of visons
are allowed. It is special for this model that visons do
not have dynamics even though they interact with one
another. (Numerical calculations of the two vison energy
reveals that the interaction between visons is repulsive.)
The minimum energy cost of creating two visons in the
ground state is always nite and is dened as the vison
gap energy
v
(
_
[J
x
J
y
[ +
_
[J

x
J

y
[). Since the visons
are gapped, the uniform -ux is still the ground state
sector as long as J
5
is much smaller than
v
. In the rest
of paper, we restrict our attention to cases (e.g. small
J
5
) in which the ground state lie in the -ux sector.
Spin correlation: The state with uniform -uxes pre-
serves the translational symmetry and TRS of the model.
To prove the ground state of the model is a spin liquid,
we need to show that there is no magnetic order and
that the spin correlations are short-ranged. Spin-3/2 op-
erators can be expressed as bilinear forms of Majorana
fermions. For instance, S
z
i
= ic
3
i
c
4
i
+
1
2
ic
1
i
c
2
i
. The action
of S
z
i
on the ground state [) creates visons in the four
surrounding plaquettes around site i [16]. The eect of
S
x
i
and S
y
i
on the ground state is similar. Because visons
are non-dynamical in the present model, [S

i
S

j
[) is
identically zero unless sites i and j are nearest neighbors,
i.e. the spin correlations are unphysically short-ranged.
Presumably, if additional small, local terms are added to
the Hamiltonian, perturbative corrections to this corre-
lator would lead to a nite correlation length.
Gapless fermions: To obtain the excitation spectrum
in the -ux (ground state) sector, we x the gauge
by choosing u
i,x
= 1, u
i,y
= (1)
i
. The corresponding
Hamiltonian is given by
H
0
=

i
_
t
x
f

i
f
i+ x
+t
y
(1)
i
f

i
f
i+ y
J
5
f

i
f
i

x
(1)
i
f

i
f

i+ x

y
f

i
f

i+ y
+H.c.
_
, (10)
which describes a p-wave superconductor of spinless
fermions [17]. Here t

+ J

and

.
(Note that in Ref. [14], the pairing terms are absent
due to projective symmetries.) In terms of the Bloch
states, f
k
=

i
exp(ik r
i
)f
i
/

N, Eq. (10) in momen-

tum space is given by
H
0
=

k
H
k

k
,

k
= (f

k
, f

k+Q
, f
k
, f
kQ
), (11)
where Q = (, ) and the summation is over only half
of the Brillouin zone since k is equivalent to k + Q. In
general, the analytical form of the eigenvalues of the 44
matrix H
k
are complicated. Due to time reversal sym-
metry, however, it is straightforward to derive the quasi-
particle excitation spectrum as follows:
E
,k
= 2
_
J
2
5
+ 2g
+,k
2
_
g
2
,k
+J
2
5
g
k
, (12)
where g
,k
= (J
2
x
J

x
2
) cos
2
k
x
+ (J
2
y
J

y
2
) sin
2
k
y
and
g
k
= (J
x
+J

x
)
2
cos
2
k
x
+ (J
y
+J

y
)
2
sin
2
k
y
.
When J
5
= 0, E
+,k
= 4(J
2
x
cos
2
k
x
+J
2
y
sin
2
k
y
)
1/2
and
E
,k
= 4(J

x
2
cos
2
k
x
+J

y
2
sin
2
k
y
)
1/2
are the energies of
the d and d

Majorana fermions respectively since they

decouple from each other. Both spectra are gapless at
nodes K = (/2, 0), around which the spectrum is
linear in momentum; the excitations are massless Dirac
fermions. However, the massless fermions are unstable in
the sense that additional small, local (further neighbor
hopping and pairing) terms can gap the nodes [23].
When 0 < J
5

v
, (so the ground state lies in the
-ux sector,) it is clear that E
+,k
is always gapped. The
conditions for E
,k
to have gapless excitations are:
J
x
J

x
cos
2
k
x
+J
y
J

y
sin
2
k
y
= J
2
5
/4, (13)
(J
x
J

y
J
y
J

x
) cos k
x
sin k
y
= 0. (14)
For simplicity, we consider the case in which J
x
, J
y

J
5
> 0 so that J
5

v
is satised for arbitrary J

x
and
J

y
. From the two conditions we obtain the phase diagram
shown in Fig. 1 as a function of J

x
and J

y
: (i) When
J

x
> J
2
5
/(4J
x
), J

y
> J
2
5
/(4J
y
), and J

x
/J

y
,= J
x
/J
y
, the
fermion spectrum has eight Dirac nodes at (/2
x
, 0)
and (/2,
y
) with

= arcsin(J
5
/
_
4J

); (ii)
When J

x
> J
2
5
/(4J
x
) and J

y
< J
2
5
/(4J
x
), there are four
Dirac nodes at (/2
x
, 0); (iii) When J

x
< J
2
5
/(4J
x
)
and J

y
> J
2
5
/(4J
x
), there are also four Dirac nodes
but at (/2,
y
); (iv) When J

x
< J
2
5
/(4J
x
) and
J

y
< J
2
5
/(4J
y
), all fermionic excitations are gapped.
For 0 < J
5

v
, the Dirac nodes, if they exist, are
topologically stable in the sense of Wigner-Von Neumann
theorem. Any weak translational and time reversal in-
variant perturbation only shifts the positions of the nodes
[23]. Consequently, the present phase with Dirac nodes
is characteristic of a stable quantum phase of matter, i.e.
an algebraic spin liquid [18, 19, 20].
4
x
J'
Eight nodes
E
i
g
h
t

n
o
d
e
s
F
o
u
r

n
o
d
e
s
Four nodes
Gapped
Pesudo-FS
y
J'
FIG. 1: The quantum phase diagram of the Gamma matrix
model as a function of J

x
and J

y
in the case J5 Jx, Jy,
where the ground state lies in the uniform -ux sector. The
Dirac nodes in the phase diagram are topologically stable. At
the critical (red) line J

x
/J

y
= Jx/Jy, fermions form a Fermi
surface (FS). The other two critical (blue) lines are dened as
J

x
= J
2
5
/(4Jx) or J

y
= J
2
5
/(4Jy).
It is worth noting that along the critical line J

x
/J

y
=
J
x
/J
y
and J

x
> J
2
5
/(4J
x
), corresponding to the red line
in Fig. 1, the discrete nodes broaden into a line of nodes.
In short, in this special case the present model realizes
a spinon Fermi surface. Since spin-3/2 operators can be
written as a bosonic bilinear in term of two Schwinger
bosons, S

= b

ss
b
s
/2, s =, with the constraint
b

+b

= 3, the present spin model can be written as

a bosonic model, albeit one with four-body interactions.
It follows as a corollary that by tuning a single coupling
constant to a critical value, a purely bosonic model can
exhibit an emergent Fermi surface.
Upon approach to the critical line J

x
= J
2
5
/(4J
x
)
from the eight node phase, both of the two Dirac cones
at (/2 +
x
, 0) approach (, 0) leading, at critical-
ity, to a single node with the unusual dispersion E
k

_
Aq
4
x
+Bq
2
y
) for small [q[ = [k (, 0)[, where A and
B are constants depending on J

, J

, and J
5
. Another
two nodes at (/2
x
, 0) approach (0, 0) leading to
the same unusual dispersion. Similar physics is obtained
at the other critical line J

y
= J
2
5
/(4J
y
).
Large-J
5
limit: The ground-states for 1/J
5
= 0 form
the low energy manifold S
z
i
= 3/2 for each i. Den-
ing an eective spin-1/2 , such that S
z
i
= 3/2 cor-
responds to
z
i
= 1, and employing degenerate per-
turbation theory in J

and J

, the eective Hamilto-

nian can be shown be H
e
= J
e

x
i

y
i+ x

x
i+ x+ y

y
i+ y
,
J
e
= (J
x
J

x
)
2
(J
y
J

y
)
2
/(16J
3
5
), which is exactly
Wens plaquette model [7]. In terms of uxes, H
e
=

i
J
e
exp[i
i
]. Consequently, the ground state is still
in the -ux sector. We expect that the gapped phase
in small J
5
limit can be adiabatically connected to the
large J
5
gapped phase.
Discussion: It is straightforward to generalize the
GMM to other lattices in 2D and also to higher dimen-
sions. For the 2D triangular lattice, a quantum spin-7/2
(or spin-
1
2
-
1
2
-
1
2
[14]) model can be dened via the seven
Gamma matrices. Due to the non-bipartiteness of the
lattice, the model spontaneously breaks TRS. Thus, it is
expected that a chiral spin liquid [22] could be realized
in this model on a triangular lattice [23].
Acknowledgments: We would like to thank Eduardo
Fradkin, Zhengcheng Gu, Taylor Hughes, Alexei Ki-
taev, Robert B. Laughlin, Joseph Maciejko, Joel Moore,
Chetan Nayak, Xiaoliang Qi, and Shivaji Sondhi for
helpful discussions. We are grateful to Xiao-Gang Wen
for pointing out a dierent interpretation of the present
Gamma matrix model in Ref. [14]. S.A.K. and H.Y. are
supported in part by DOE grant DEFG03-01ER45925.
S.C.Z. is supported by NSF DMR-0342832 and DOE
grant DE-AC03-76SF00515. H.Y. thanks the support by
a Stanford Graduate Fellowship.
[1] P.W. Anderson, Science 235, 1196 (1987).
[2] For a review, see P. A. Lee, N. Nagaosa, X.-G. Wen, Rev.
Mod. Phys. 78, 17 (2006).
[3] R. Moessner and S. Sondhi, Phys. Rev. Lett. 86, 1881
(2001).
[4] D. S. Rokhsar and S. A. Kivelson, Phys. Rev. Lett. 61,
2376 (1988).
[5] P. W. Anderson, Mater. Res. Bull. 8, 153 (1973).
[6] S. A. Kivelson, D. S. Rokhsar, and J. P. Sethna, Phys.
Rev. B 35, 8865 (1987).
[7] X. G. Wen, Phys. Rev. Lett. 90, 016803 (2003).
[8] Y. Shimizu et al, Phys. Rev. Lett. 91, 107001 (2003).
[9] S. Yamashita et al, Nat. Phys. 4, 459 (2008).
[10] Y. Qi, C. Xu, and S. Sachdev, Arxiv: 0809.0694.
[11] J. S. Helton et al, Phys. Rev. Lett. 98, 107204 (2007).
[12] A. Kitaev, Annals of Physics 321, 2 (2006).
[13] S. Murakami, N. Nagaosa, and S. C. Zhang, Phys. Rev.
B 69, 235206 (2004).
[14] X.-G. Wen, Phys. Rev. D, 68, 065003 (2003).
[15] E. H. Lieb, Phys. Rev. Lett. 73, 2158 (1994).
[16] G. Baskaran, S. Mandal, and R. Shankar, Phys. Rev.
Lett. 98, 247201 (2007).
[17] H.-D.Chen and J.-P.Hu, Phys. Rev. B 76, 193101 (2007).
[18] W. Rantner and X.-G. Wen, Phys. Rev. Lett. 86, 3871
(2001).
[19] M. Hermele et al, Phys. Rev. B 70, 214437 (2004).
[20] M. Hermele, T. Senthil, and M. P. A. Fisher, Phys. Rev.
B 72, 104404 (2005).
[21] Y. Shimizu et al, Phys. Rev. B 73, 140407(R) (2006).
[22] H. Yao and S. A. Kivelson, Phys. Rev. Lett. 99, 247203
(2007)
[23] H. Yao et al, unpublished.
[24] The total number of -uxes must be even, so the choice
of which sublattice is immaterial.