An Introduction to Quantum Order, String-net Condensation,

and Emergence of Light and Fermions

Xiao-Gang Wen

Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139

(Dated: Jan. 2004)
We review some recent work on new states of matter. Those states contain a new kind of order
quantum order because they cannot be described symmetry breaking. Some quantum orders are
shown to be closely related to string-net condensations. Those quantum orders lead to an emergence
of gauge bosons and fermions from pure bosonic models.
I. INTRODUCTION
A. Origin of light/fermion and new orders
The existences of light and fermions are two big mys-
teries in nature. The mysteries are so deep that the
questions like, What are light and fermions?, Where
do light and fermions come from?, Why do light and
fermions exist?, are regarded by many people as philo-
sophical or even religious questions.
To appreciate the physical signicance those questions
let us ask three simpler questions: What are phonons?,
Where do phonons come from?, Why do phonons ex-
ist?. We know these three questions to be scientic
questions and we know their answers. Phonons are vi-
brations of a crystal. Phonons come from a spontaneous
translation symmetry breaking. Phonon exists because
the translation-symmetry-breaking phase actually exists
in nature. It is quite interesting to see that our under-
standing of a gapless excitation phonon is rooted in
our understanding of the phases of matter as symmetry
breaking states [1, 2].
However, our picture for massless photons and nearly
massless fermions[70] (such as electrons and quarks) is
quite dierent from our picture of gapless phonons. We
regard photons and fermions as elementary particles
the building block of our universe.
But why should we regard photons and fermions as
elementary particles? Why dont we regard photons and
fermions as emergent quasiparticles like phonons? We
can view this question from several dierent points of
view.
First point of view: Before late 1970s, we felt that we
understood, at least in principle, all the physics about
phases and phase transitions based on Landaus sym-
metry breaking theory [3, 4]. In such a theory, if we
start with a bosonic model, the only way to get gapless
excitations is via spontaneous breaking of a continuous

A more comprehensive description of topological/quantum orders

can be found in Quantum eld theory of many-body systems from
the origin of sound to an origin of light and fermions, Xiao-Gang
Wen, Oxford Univ. Press, 2004.

URL: http://dao.mit.edu/~wen
symmetry [1, 2], which will lead to gapless scalar bosonic
excitations. It seems that there is no way to obtain gap-
less gauge bosons and fermions from symmetry breaking.
This may be the reason why people think our vacuum
(with massless gauge bosons and nearly-gapless fermions)
is very dierent from bosonic many-body systems (which
were believed to contain only gapless scalar bosonic col-
lective excitations, such as phonons). It seems there does
not exist any order that give rise to massless photons and
nearly-massless fermions. This may be the reason why
we regard photons and fermions as elementary particles
and introduce them by hand into our theory of nature.
Second point of view: On the other hand, the resem-
blance between the photons and the phonons makes it
odd to regard photons as elementary. To appreciate this
point, let us imagine another universe which contains a
massless excitation with three components. This mass-
less excitation behaves in every way like the phonon in a
crystal. We will not hesitate to declare that the vacuum
in that universe is actually a crystal even when no one
can see the particles that form the crystal. Our convic-
tion of the existence of the crystal does not come from
seeing the lattice structure, but from seeing the low en-
ergy collective modes of the crystal.
Third point of view: Now back to our universe. Are
the massless photons and nearly massless fermions also
collective modes of certain order in our vacuum. Not
knowing what order can give rise to photons and fermions
may not imply the photons and fermions to be elemen-
tary. More likely, it means that our understanding of
order is incomplete. The very existence of light and
fermions may indicate that our vacuum contain a new
kind of order. The new order will produce light and
fermions, and protect its masslessness.
Fourth point of view: If we had a material which is
described by bosons (such as a spin system) and if we
found that the low energy excitations in the material are
gauge bosons and fermions, we would not hesitate to de-
clare that the material contains a new kind of order be-
yond the symmetry breaking description. But so far, we
have not nd any material that contain emergent gauge
bosons and emergent fermions. So we do not know if
new order beyond the symmetry breaking exists or not.
Other other hand, we may regard our vacuum as a special
material. From this point of view, the light and the elec-
trons in the vacuum provided an experimental evidence
2
Quantum system Classical system
Gapped
NambuGoldstone mode
"Particle" condensation
Orders
Fermi liquids
Fermi surface topology
Gapless Gauge bosons/Fermions
Projective symmetry group
Conformal algebra, ??
Topological field theory
Nonsymmetry breaking orders Symmetry breaking orders
Topological orders
Quantum orders
Symmetry group
Stringnet condensation
FIG. 1: A classication of dierent orders in matter (and in
vacuum).
of the existence of new order.
B. Topological order and quantum order
Historically, our understanding of new order beyond
symmetry breaking starts at an unexpected place frac-
tional quantum Hall (FQH) systems. The FQH states
discovered in 1982 [5, 6] opened a new chapter of con-
densed matter physics. What is really new in FQH states
is that FQH systems contain many dierent phases at
zero temperature which have the same symmetry. Thus
those phases cannot be distinguished by symmetries and
cannot be described by Landaus symmetry breaking the-
ory.
Since FQH states cannot be described by Landaus
symmetry breaking theory, it was proposed that FQH
states contain a new kind of order topological order
[7]. Topological order is new because it cannot be de-
scribed by symmetry breaking, long range correlation, or
local order parameters. None of the usual tools that we
used to characterize a phase applies to topological order.
Despite this, topological order is not an empty concept
since it can be characterized by a new set of tools, such
as the number of degenerate ground states [8, 9], the
non-Abelian Berrys phase under modular transforma-
tions [10], quasiparticle statistics [11], and edge states
[12, 13]. Just like Ginzburg-Landau theory is the eec-
tive theory of symmetry breaking order, the topological
eld theory [14] is the eective theory of topological order
[7].
It was shown that the ground state degeneracy of a
topologically ordered state is robust against any pertur-
bations [9]. Thus the ground state degeneracy is a uni-
versal property that can be used to characterize a phase.
The existence of topologically degenerate ground states
proves the existence of topological order. The topolog-
ically degenerate ground states were found to useful in
fault tolerant quantum computing [15].
The concept of topological order only applies to state
with nite energy gap. It was recently generalized to
quantum order [16] to describe new kind of orders in
gapless quantum states. There are two general but vague
understanding quantum orders.
In the rst understanding, we assume that the order
in a quantum state is encoded in the many-body ground
state wave function. We believe that the symmetry of
the ground state wave function cannot characterize all
the possible orders in the many-body state. The extra
structure in the ground state can be viewed as a pattern
of quantum entanglement in the many-body state. From
this point of view, we may say that quantum orders are
patterns quantum entanglement in quantum many-body
states.
The second way to understand quantum order is to
see how it ts into a general classication scheme of
orders (see Fig. 1). First, dierent orders can be di-
vided into two classes: symmetry breaking orders and
non-symmetry breaking orders. The symmetry breaking
orders can be described by a local order parameter and
can be said to contain a condensation of point-like ob-
jects. The amplitude of condensation corresponds to the
order parameter. All the symmetry breaking orders can
be understood in terms of Landaus symmetry breaking
theory. The non-symmetry breaking orders cannot be de-
scribed by symmetry breaking, nor by the related local
order parameters and long range correlations. Thus they
are a new kind of orders. If a quantum system (a state
at zero temperature) contains a non-symmetry breaking
order, then the system is said to contain a non-trivial
quantum order. We see that a quantum order is simply
a non-symmetry breaking order in a quantum system.
Quantum orders can be further divided into many sub-
classes. If a quantum state is gapped, then the corre-
sponding quantum order will be called topological order.
The second class of quantum orders appear in Fermi liq-
uids (or free fermion systems). The dierent quantum
orders in Fermi liquids are classied by the Fermi surface
topology [17]. We will discuss this class of quantum order
briey in section III. The third class of quantum orders
arises from a condensation of nets of strings (string-nets)
[1824]. We will discuss it in sections IV and VI. This
class of quantum orders shares some similarities with the
symmetry breaking orders of particle condensation.
We know that dierent symmetry breaking orders can
be classied by symmetry groups. Using group theory,
we can classify all the 230 crystal orders in three di-
mensions. The phase transitions between dierent sym-
metry breaking orders are described by critical point
with algebraic correlations. The symmetry also produces
and protects gapless collective excitations the Nambu-
Goldstone bosons above the symmetry breaking ground
state. Similarly, dierent string-net condensations (and
the corresponding quantum orders) can be classied by
a mathematical object called projective symmetry group
[16] (see subsection IVD). Using the projective symme-
try group, we can classify over 100 dierent 2D spin liq-
uids that all have the same symmetry. The phase transi-
tions between dierent quantum orders are also described
by critical points. Those phase transitions do not change
any symmetry and cannot be described by order param-
eters associated with broken symmetries [2529]. Just
3
FIG. 2: Our vacuum may be a state lled with string-nets.
The uctuations of the string give rise to gauge bosons. The
ends of the strings correspond to electrons, quarks, etc .
like the symmetry group, the projective symmetry group
can also produce and protect gapless excitations. How-
ever, unlike the symmetry group, the projective symme-
try group produces and protects gapless gauge bosons
and fermions [16, 30, 31]. Because of this, we can say
that light and massless fermions can have a unied ori-
gin. They can emerge from string-net condensations.
C. String-net picture of light and fermions
We used to believe that to have light and fermions in
our theory, we have to introduce by hand a fundamental
U(1) gauge eld and anti-commuting fermion elds, since
at that time we did not know any collective modes that
behave like gauge bosons and fermions. However, due to
the advances of the last 20 years, we now know how to
construct local bosonic systems that have emergent un-
conned gauge bosons and/or fermions [15, 19, 30, 32
38]. In particular, one can construct ugly bosonic spin
models on a cubic lattice whose low energy eective the-
ory is the beautiful QED and QCD with emergent pho-
tons, electrons, quarks, and gluons [39].
This raises an issue: do light and fermions in nature
come from a fundamental U(1) gauge eld and anti-
commuting elds as in the U(1)SU(2)SU(3) standard
model or do they come from a particular quantum order
in our vacuum? Is Coulombs law a fundamental law of
nature or just an emergent phenomenon? Clearly it is
more natural to assume light and fermions, as well as
the Coulombs law, come from a quantum order in our
vacuum. From the connections between string-net con-
densation, quantum order, and massless gauge/fermion
excitations, it is very tempting to propose the follow-
ing possible answers to the three fundamental questions
about light and fermions:
What are light and fermions?
Light is the uctuation of condensed strings (of arbitrary
sizes) [21, 23, 37]. Fermions are ends of condensed strings
[19].
Where do light and fermions come from?
Light and fermions come from the collective motions of
nets of strings (or string-net) that ll our vacuum (see
Fig. 2).
Why do light and fermions exist?
P
T
Vapor
Ice
Water
FIG. 3: The phase diagram of water.
Light and fermions exist because our vacuum happen to
have a property called string-net condensation.
Had our vacuum chose to have particle condensation,
there would be only Nambu-Goldstone bosons at low en-
ergies. Such a universe would be very boring. String
condensation and the resulting light and fermions pro-
vide a much more interesting universe, at least interest-
ing enough to support intelligent life to study the origin
of light and fermions.
Our understanding of quantum/topological orders are
base on many researchs in three main areas: (1) the study
of topological phases in condensed matter systems such
as FQH systems [9, 4042], quantum dimer models [32,
4346], quantum spin models [10, 16, 3336, 4749], or
even superconducting states [50, 51], (2) the study of
lattice gauge theory [2123, 52], and (3) the study of
quantum computing by anyons [15, 53, 54]. In this paper,
we will use some simple models to introduce the main
points of topological/quantum order.
II. STATE OF MATTER AND CONCEPT OF
ORDER
To start our journey to search new state of matter
with emergent gauge bosons and fermions, we like to
rst discuss the concept of order and review the symme-
try breaking description of order. With low temperature
technology developed around 1900, physicists discovered
many new states of matter (such as superconductors and
superuids). Those dierent states have dierent inter-
nal structures, which are called dierent kinds of orders.
The precise denition of order involves phase transition.
Two states of many-body systems have the same order
if we can smoothly change one state into the other (by
smoothly changing the Hamiltonian) without encounter
a phase transition (ie without encounter a singularity in
the free energy). If there is no way to change one state
into the other without a phase transition, than the two
states will have dierent orders. We note that our def-
inition of order is a denition of equivalent class. Two
states that can be connected without a phase transition
are dened to be equivalent. The equivalent class dened
this way is called the universality class. Two states with
dierent orders can be also be said as two states belong
to dierent universality classes. According to our deni-
tion, water and ice have dierent orders while water and
vapor have the same order. (See Fig. 3)
4
(a)
A
(b)
A A
B B
(c)
F F F

FIG. 4: In the presence of symmetry, switching
of the minima can be continuous and causes a second-order
phase transition. (a) The ground state is symmetric under
the transformation . (c) The ground state breaks the
symmetry.
After discovering so many dierent kinds of orders, a
general theory is needed to gain a deeper understanding
of states of matter. In particular, we like to understand
what make two orders really dierent so that we can-
not change one order into the other without encounter
a phase transition. It is a deep insight to connect the
singularity in free energy to a symmetry breaking pic-
ture in Fig. 4. Based on the relation between orders and
symmetries, Landau developed a general theory of orders
and phase transitions [3, 55]. According to Landaus the-
ory, the states in the same phase always have the same
symmetry and the states in dierent phases always have
dierent symmetries. So symmetry is a universal prop-
erty that characterized dierent phases. Landaus theory
is very successful. Using Landaus theory and the related
group theory for symmetries, we can classify all the 230
dierent kinds of crystals that can exist in three dimen-
sions. By determining how symmetry changes across a
continuous phase transition, we can obtain the critical
properties of the phase transition. The symmetry break-
ing also provides the origin of many gapless excitations,
such as phonons, spin waves, etc , which determine the
low energy properties of many systems [1, 2]. A lot of
the properties of those excitations, including their gap-
lessness, are directly determined by the symmetry.
III. QUANTUM ORDERS AND QUANTUM
TRANSITIONS IN FREE FERMION SYSTEMS
However, not all orders are described by symmetry. In
fact, free fermion systems are the simplest systems with
non-trivial quantum order. In this section, we will study
the quantum order in a free fermion system to gain some
intuitive understanding of quantum order.
To nd quantum order, or to even dene quantum or-
der, we must nd universal properties. The universal
properties are the properties which do not change under
any perturbations of the Hamiltonian that do not aect
the symmetry. Once we nd those universal properties,
we can use them to group many-body wave functions
into universal classes such that the wave functions in
each class have the same universal properties. Hopefully
those universal classes correspond to quantum phases in
a phase diagram. To really show that those universal
classes do correspond to quantum phases, we must show
(a) (b) (d) (c)
FIG. 5: Two sets of oriented Fermi surfaces in (a) and (b)
represent two dierent quantum orders. The two possible
transition points between the two quantum order (a) and (b)
are described by the Fermi surfaces (c) and (d).
that as we deform the Hamiltonian to drive the ground
state from one universality class to another, the ground
state energy always has a singularity at the transition
point. (For zero temperature quantum transition, the
ground state energy play the role of free energy for nite
temperature transition. A singularity in the ground state
energy signal a quantum phase transition.)
We know that symmetry is a universal property. The
order determined by such a universal property is our old
friend the symmetry-breaking order. So to show the
existence of new quantum order, we must nd universal
properties that are dierent from symmetry.
Let us consider free fermion system with only the
translation symmetry and the U(1) symmetry from the
fermion number conservation. The Hamiltonian has a
form
H =

ij
_
c

i
t
ij
c
j
+h.c.
_
(1)
with t

ij
= t
ji
. The ground state is obtained by lling
every negative energy level with one fermion. In general,
the system contains several pieces of Fermi surfaces.
We note that any small change of t
ij
do not change
the topology of the Fermi surfaces as long as the change
do not break the translation symmetry and do not vio-
late the fermion number conservation. So the the Fermi
surface topology is a universal properties.
To show that the Fermi surface topology really denes
quantum phases, we need to show that any change of
the Fermi surface topology will lead to a singularity in
the ground state energy. The Fermi surface topology can
change in two ways as we continuously changing t
ij
: (a)
a Fermi surface shrinks to zero (Fig. 5d) and (b) two
Fermi surfaces join (Fig. 5c).
When a Fermi surface is about to disappear in a D-
dimensional system, the ground state energy density has
a form

E
=
_
d
D
k
(2)
D
(k M k )(k M k +) +...
where the ... represents non-singular contribution and
the symmetric matrix M is positive (or negative) def-
inite. The integral in the above equation simply rep-
resents the total energy of the lled states enclosed by
the small Fermi surface. The small Fermi surface is
about to shrink to zero as pass zero. We nd that
5
the ground state energy density has a singularity at
= 0:
E
= c
(2+D)/2
() + ..., where (x > 0) = 1,
(x < 0) = 0. When two Fermi surfaces are about
to join, the singularity is still determined by the above
equation, but now M has both negative and positive
eigenvalues. The ground state energy density has a sin-
gularity
E
= c
(2+D)/2
() + ... when D is odd and

E
= c
(2+D)/2
log || +... when D is even.
We nd that the ground state energy density has a
singularity at = 0 which is exactly the same place
where the topology of the Fermi surfaces has a change
[17]. Therefore the topology of the Fermi surface is a
universal property that dene a order. We note that the
states with dierent Fermi surface topologies all have the
same symmetry. Thus the quantum phase transition that
change the the topology of the Fermi surface does not
change any symmetry. Therefore the order dened by
the Fermi surface topology is a new kind of order that
cannot be characterized by symmetries. Such an order is
an example of quantum order.
IV. QUANTUM ORDER IN BOSON/SPIN
LIQUIDS
A. Quantum order and new universal properties
After realizing the existence of the quantum order in
free fermion systems, we may expect quantum order to
be a general phenomena. In this section we would like
to study the existence of quantum order in interacting
boson or spin systems. Instead of looking for univer-
sal properties, we would like to rst look for boson/spin
states that contain emergent gauge bosons and fermions.
The emergence of gauge bosons and fermions indicate the
appearance of new quantum order. Then we will study
the universal properties which give a more systematic
description of quantum orders.
We like to point out that a spin system is a special
case of boson system since we can regard a site with a
down spin as an empty site and a site with a up spin as
a occupied site for bosons. In this section we will inter-
changeably use both the boson and the spin languages to
describe the same system.
B. Projective construction
In the introduction, we argue that the existence of
light and electron implies that our vacuum contains a
non-trivial quantum order. However, we do not know
to which system does the quantum order belong. Now
we look for quantum order in spin systems. So we know
our system. But we do not know what to look for, since
we have no clue what does a quantum order look like
at microscopic level. So instead of directly searching for
quantum order, let us look for something slightly more
familiar: a spin liquid state that does not break any sym-
metry.
1. A mean-eld theory of spin liquids
To be concrete, let us consider a spin-1/2 system on a
square lattice
H =

ij
J
ij
S
i
S
j
. (2)
In the conventional mean-eld theory, we use the ground
|
m
i
mean
of a free spin Hamiltonian
H
mean
=

i
m
i
S
i
to approximate the ground state of the interacting Hamil-
tonian H. The mean-eld ground state described by
m
i
= m
i
, |
m
i
mean
, is obtained by minimizing the av-
erage energy
m
i
mean
|H|
m
i
mean
. However, no matter
how we choose the mean-eld ansatz m
i
, the mean-eld
ground state always break spin rotation symmetry and
there is no way to obtain a spin liquid.
We have to use another approach to obtain a spin
liquid [56, 57]. We start with a free fermion mean-
eld Hamiltonian that contains two fermion elds
i
=
_

1i

2i
_
:
H
mean
=

i
u
ij

i
(3)
where u
ij
are two by two complex matrices dened on
the links ij that describe the hopping of the fermions.
However, the mean-eld ground state of H
mean
, |
u
ij
mean
,
does not correspond to a spin state. But, we can obtain
a spin state from the fermion state by projecting the
fermion state |
u
ij
mean
into the subspace where every site
has even numbers of fermion:
|
u
ij
spin
= P|
u
ij
mean

This is because there are only two states, on each site,

that have even number of fermion. One is the empty
site |0 which can be viewed as a spin-down state, and
the other is the state with two fermions

1i

2i
|0 which
corresponds to the spin-up state.
Although it is not obvious, one can show [16] that i
u
ij
satisfy
Tr(u
ij
) = imaginary, Tr(u
ij

l
) = real, l = 1, 2, 3
where
1,2,3
are the Pauli matrices, then
u
ij
spin
describes
a spin rotation invariant state. Since the spin state is ob-
tained through the projection P, we will call the above
construction projective construction. It is a special case
of the slave-boson construction [56, 57] at zero doping.
6
We see that at least we can use the projective construc-
tion to construct a spin liquid state which does not break
the spin rotation symmetry.
Through the projective construction, we introduced a
label u
ij
that labels a class of spin wave functions. (u
ij
does not label all possible spin wave functions.) So we do
not have to directly deal with the many-body functions
of of spin liquids which are very hard to visualize. We
only need to deal with u
ij
to understand the properties
of the spin liquids.
2. The variational ground state
We may view u
ij
as variational parameters. An
approximate many-body ground state wave function
|
u
ij
spin
can be obtained by minimizing the average en-
ergy
u
ij
spin
|H|
u
ij
spin
where H is the spin Hamiltonian.
Aside from its many variational parameters, there is
no reason to expect the projective construction to give a
good approximation of the ground state for a generic spin
Hamiltonian. So there is no reason to trust any results
obtained from projective construction. As we will see be-
low that the projective construction leads to many unbe-
lievable results, such as the emergence of gauge bosons,
fermions, or even anyons from purely bosonic systems.
So those amazing results may just be the artifacts of a
unreliable approach.
However, for certain type of Hamiltonians, the projec-
tive construction leads to a good approximation of the
ground state. In certain large N limits [33, 56], the uc-
tuations around the mean-eld ansatz are weak, and the
projective construction gives a good description of the
ground state and the excitations. We can also construct
special Hamiltonians where the projective construction
leads to an exact ground state (and all the exact ex-
cited states. See section V). For those Hamiltonians, the
projective construction does provide a good description
of spin liquid states which cannot be provided by other
conventional method. In the following, we will only con-
sider those friendly Hamiltonians and trust the results of
the projective construction.
I would like to mention that in practice, most of the
unbelievable predictions from the projective construction
turn out to be correct. For example, in the research of
high T
c
superconductors, both the d-wave superconduct-
ing state and the pseudo-gap metallic state was predicted
by the projective construction prior to experimental ob-
servation [56, 58]. This may be the rst time in the his-
tory of condensed matter physics that a truly new state
of matter the pseudo-gap metallic state is predicted
before the experimental observation [59].
3. Low energy excitations
In the conventional mean-eld theory for spin or-
dered state, after we obtain the mean-eld ground state
|
m
i
mean
that minimize the average energy, we can create
collective excitations above the ground state through the
uctuation of the mean-eld ansatz m
i
= m
i
+ m
i
.
Those collective excitations correspond to the spin wave
excitations.
In the projective construction, we can create collec-
tive excitations in the exactly the same way. The collec-
tive excitations above the mean-eld ground state
u
ij
spin
correspond to the uctuations of the mean-eld ansatz
u
ij
= u
ij
+ u
ij
. The physical spin wave function for
such type of excitations is obtained via the projection of
the deformed fermion state |
u
ij
+u
ij
mean
:
|
u
ij
spin
= P|
u
ij
+u
ij
mean

The ground state obtained from the projective con-
struction also contains a second type of excitations. This
type of excitations corresponds to fermion pair excita-
tions. We start with the fermion ground state with a
pair of particle-hole excitations

ai

bj
|
u
ij
mean
. After
the projection, we obtain the corresponding physical spin
state
|
(a,i,b,j)
spin
=

ai

bj
|
u
ij
mean

that describes a pair of fermions.

Clearly the fermions excitations interact with the col-
lective modes u
ij
. The eective Lagrangian that de-
scribes the two types of excitations has a formL(, u
ij
).
It appears that the spin liquid state obtained through the
projective construction always contain fermionic excita-
tions described by . The emergent fermions will imply
that the spin liquid state is a new state of matter and
contain non-trivial quantum order.
However, the thing is not that easy. It turns out the
collective uctuations represent gauge uctuations and
the fermions carry gauge charges. Those uctuations can
mediate an conning interactions between the fermions.
As a result, the spin liquid state may not contain any
fermionic excitations and may not represent new state of
matter.
To see that the uctuations of u
ij
represent gauge uc-
tuations, we note that the mean-eld Hamiltonian H
mean
is invariant under the following SU(2) gauge transforma-
tion

i
W
i

i
, W
i
SU(2)
u
ij
W
i
u
ij
W

j
(4)
So the two fermion ground states of the H
mean
corre-
sponding to two ansatz u
ij
and u

ij
are related by an
SU(2) gauge transformation if u

ij
= W
i
u
ij
W

j
. Since
the even-fermion states |0 and

1i

2i
|0 are invariant
under SU(2) gauge transformation, the projected state
|
u
ij
spin
= P|
u
ij
mean
is invariant under the SU(2) gauge
transformation:
|
u
ij
spin
= |
W
i
u
ij
W

j
spin

7
FIG. 6: The fermion dispersion for the ansatz Eq. (6).
As a result, u
ij
is not a one-to-one label of the physi-
cal spin state, but a many-to-one label. The mean-eld
energy E(u
ij
) =
u
ij
spin
|H|
u
ij
spin
, as a function of real
physical spin state |
u
ij
spin
, is invariant under the SU(2)
gauge transformation E(u
ij
) = E(W
i
u
ij
W

j
). Similarly,
the eective Lagrangian L(, u
ij
) is also invariant under
the SU(2) gauge transformation:
L(, u
ij
) = L(W
i
, W
i
u
ij
W

j
)
The SU(2) gauge invariance of the eective Lagrangian
strongly aect the dynamics of u
ij
uctuations. It makes
the u
ij
uctuations to behave like SU(2) gauge uctua-
tions. If we write u
ij
= ie
ia
l
ij

l
, then a
l
ij
play the role
of the SU(2) gauge potential on the lattice.
C. Deconned phase and new state of matter
We have mentioned that to obtain a new state of mat-
ter from the projective construction, the gauge uctua-
tions u
ij
must not mediate a conning interaction. One
may say that the SU(2) gauge uctuations always me-
diate a conning interaction in 1+2D, so the projective
construction can never produce a spin liquid with emer-
gent fermions. Well again thing is not that simple. It
turns out that whether the gauge uctuations conne
the fermions or not depend on the form of the mean-eld
ansatz u
ij
that minimize the average energy H.
To understand how the ansatz u
ij
aect the dynamics
of the gauge uctuations, it is convenient to introduce
the loop variable
P(C
i
) = u
ij
u
jk
... u
li
(5)
If we write P(C
i
) as P(C
i
) = e
i
l
(C
i
)
l
, then
l

l
is
the SU(2) ux through the loop C
i
: i j k ..
l i with base point i. The SU(2) ux correspond to
the gauge eld strength in the continuum limit.
1. SU(2) spin liquid
If for a certain spin Hamiltonian H, u
ij
has a form
u
i,i+x
= i()
i
y
,
u
i,i+y
= i, (6)
the SU(2) ux through any loops is trivial. In this
case the uctuations of u
ij
behave like the usual SU(2)
gauge uctuations. The fermion dispersion is deter-
mined by mean-eld Hamiltonian H
mean
in Eq. (3):
E
k
= 2
_
sin
2
k
x
+ sin
2
k
y
. At low energies (E
k
0)
the fermions have a linear dispersion E |k| (see Fig.
6) and can be described by massless Dirac fermions in
the continuum limit. So in the continuum limit, the low
energy uctuations are described by
L =
N

=1

ia
l

l
)

+
1
2g
Trf
l

f
l,
(7)
where a
l

and f
l

, l = 1, 2, 3 the vector potential and the

eld strength of the SU(2) gauge eld.
It is interesting to see that for the spin liquid described
by the ansatz Eq. (6), the low energy uctuations is de-
scribed by 1+2D QCD! The low energy physical proper-
ties of 1+2D QCD is complicated.
But the above construction can be easily generalize to
higher dimensions. The low energy properties of 1+4D
QCD is much easier to determine and it contains de-
conned phase. As a result, the 1+4D spin liquid con-
tains emergent massless fermions and emergent massless
SU(2) gauge bosons. Such a state represent a new state
of matter and contains a non-trivial quantum order. We
get what we are looking for (if we live in four dimensional
space). Because of the low energy SU(2) gauge uctua-
tions, we will call the ansatz Eq. (6) SU(2) ansatz and
the correspond spin liquid SU(2) spin liquid.
2. U(1) spin liquid
If the ansatz that minimizes the average energy has a
form
u
i,i+x
=
3
i()
i
,
u
i,i+y
=
3
+i()
i
, (8)
where ()
i
()
i
x
+i
y
, then the low energy u
ij
uctua-
tions are actually described by U(1) gauge uctuations.
This is because the above ansatz contains non-trial SU(2)
ux: P(C
i
) e
i
3

3
. Unlike the ux of U(1) gauge eld,
the ux of SU(2) gauge eld is not invariant under the
SU(2) gauge transformations. Instead, the SU(2) ux
transforms like a Higgs eld that carries non-zero SU(2)
charge. The non-zero SU(2) ux has a similar eect as
the condensation of a Higgs eld, which can give gauge
bosons a mass term via the Anderson-Higgs mechanism
8
[60, 61]. For our case, the SU(2) ux in the
3
direction
give the a
1

and a
2

components of the gauge eld a mass,

but a
3

remains massless. So the SU(2) ux break the

SU(2) gauge structure down to a U(1) gauge structure
[34]. This is why the spin liquid described by the ansatz
Eq. (8) contains only massless U(1) gauge uctuations.
The low energy fermions are still described by massless
Dirac fermions. So the low energy eective theory of the
spin liquid is a 1+2D QED
L =
N

=1

ia
3

3
)

+
1
2g
f
3

f
3,
(9)
where a
3

and f
3

are the vector potential and the eld

strength of the U(1) gauge eld. Again, the above con-
struction can be easily generalize to higher dimensions.
Now we do not need to go to 1+4 dimensions. In 1+3
dimensions, the spin liquid [30] already contains emer-
gent massless fermions and emergent massless U(1) gauge
bosons since 1+3D QED is not conning. Such a 1+3D
spin liquid represents another new state of matter and
will be called U(1) spin liquid.
The close resemblance of the low energy eective the-
ory of the spin liquids and the QED/QCD in the standard
model makes one really wonder: is this how the QED and
QCD emerge to be the eective theory that describe our
vacuum [39]?
Even in 1+2D, the U(1) spin liquid was shown to be
a stable phase [16, 62, 63] based a combined analysis of
instanton [64] and projective symmetry [16] (see section
IVD). The existence of the U(1) spin liquid is a striking
phenomenon since the gapless excitations interact down
to zero energy, and yet remain to be gapless. The in-
teraction is so strong that that are no free fermionic or
bosonic quasiparticles ar low energies. Since the U(1)
gauge bosons and the fermions are not well dened at any
energy, the U(1) spin liquid was more correctly called the
algebraic spin liquid [16, 62].
Since there is no spontaneous broken symmetry to
protect the above interacting gapless excitations, there
should be a principle that prevents the gapless excita-
tions from opening an energy gap and makes the alge-
braic spin liquids stable. Ref. [16] proposed that quan-
tum order is such a principle. To support this idea, it
was shown that just like the symmetry group of sym-
metry breaking order protects gapless Nambu-Goldstone
modes, the projective symmetry group (see section IVD)
of quantum order protects the interacting gapless excita-
tions in the algebraic spin liquid. This result implies
that the stabilities of algebraic spin liquids are protected
by their projective symmetry groups. The existence of
gapless excitations without symmetry breaking is a truly
remarkable feature of quantum ordered states.
3. Z
2
spin liquid
For the ansatz [34]
u
i,i+x
=u
i,i+y
=
3
,
u
i,i+x+y
=
1
+
2
,
u
i,ix+y
=
1

2
, (10)
the SU(2) ux
l
(C
i
)
l
for dierent loops points in
dierent directions in the (
1
,
2
,
3
) space. The non-
collinear SU(2) ux break the SU(2) gauge structure
down to a Z
2
gauge structure. Since the Z
2
gauge uc-
tuations only mediate a short ranged interaction, the
fermions are not conned even in 1+2D. The spin liq-
uid obtained from the ansatz Eq. (10) is a new state of
matter that has emergent fermions and Z
2
gauge theory.
Non-trivial quantum order can appear in two dimensional
space. The spin liquid will be called Z
2
spin liquid. Such
a spin liquid corresponds to the short-ranged Resonating
Valence Bound state proposed in Ref. [43, 65].
4. Summary
The projective construction is a powerful way con-
struct states that represent new state of matter. Those
states have emergent fermions and gauge bosons, and
thus contain a new kind of order that cannot be described
by symmetry. The new order is called quantum orders.
Certainly, not all states obtained via the projection con-
struction contain non-trivial quantum orders. But many
of them do. The projective construction not only can
produces states with emergent QED and QCD, it can
also produce states with fractional statistics (including
non-Abelian statistics) [36, 66, 67].
D. Quantum order and projective symmetry group
We know that dierent symmetry-breaking orders can
be systematically characterize by dierent symmetry
group. The group theory description allows us to classify
230 dierent 3D crystals. Knowing the existence of new
quantum order in spin liquids, we would like to ask what
mathematical object that we can use to systematically
describe dierent quantum orders? In this subsection, we
are going to introduce a mathematical object projective
symmetry group and show that the projective symmetry
group can (partially) characterize dierent quantum or-
ders.
1. The diculty of seeing quantum order
In subsection IVA, we argue that to describe or to even
dene quantum order, we must nd universal properties
9
that are dierent from the symmetry (such as the topol-
ogy of the Fermi surfaces discussed in the section III).
However, it is very dicult to nd new universal prop-
erties of generic many-body wave functions. Let us con-
sider the free fermion systems that we discussed before to
gain some intuitive understanding of the diculty. We
know that a free fermion ground state is described by an
anti-symmetric wave function of N variables. The anti-
symmetric function has a form of Slater determinant:
(x
1
, ..., x
N
) = det(M) where the matrix elements of
M is given by M
mn
=
n
(x
m
) and
n
are single-fermion
wave functions. The rst step to nd quantum orders
in free fermion systems is to nd a reasonable way to
group the Slater-determinant wave functions into classes.
This is very dicult to do if we only know the real space
many-body function (x
1
, ..., x
N
). However, if we use
Fourier transformation to transform the real-space wave
function to momentum-space wave function, then we can
group dierent wave functions into classes according to
their Fermi surface topologies. This leads to our under-
standing of quantum orders in free fermions systems (see
section III). The Fermi surface topology is the quan-
tum number that allows us characterize dierent quan-
tum phases of free fermions. Here we would like to stress
that without the Fourier transformation, it is very di-
cult to see Fermi surface topologies from the real space
many-body function (x
1
, ..., x
N
).
For the boson/spin systems, what is missing here is the
corresponding Fourier transformation. Just like the
topology of Fermi surface, it is very dicult to see uni-
versal properties (if any) directly from the real space wave
function. At moment there are two ways to understand
the quantum order in boson/spin systems. The rst one
is through the projective symmetry group which will be
discussed below. The second one is through string-net
condensation which will be discuss in section VI. Both
the projective symmetry group and the string-net con-
densation play the role of the Fourier transformation in
the free fermion system. They allow us the extract the
universal properties from the very complicated many-
body wave functions.
2. Symmetry of the spin liquids
To motivate the projective symmetry group, let us rst
consider the symmetry of the spin liquid states obtained
from the SU(2), U(1) and Z
2
ansatz Eq. (6), Eq. (8),
and Eq. (10). At rst sight, those spin liquids appear
not to have all the symmetries. For example, the U(1)
ansatz Eq. (8) is not invariant under translation in the
x-direction.
However, those ansatz do describe spin states that
have all the symmetries of square lattice, namely the two
translation symmetries T
x
: (i
x
, i
y
) (i
x
+ 1, i
y
) and
T
y
: (i
x
, i
y
) (i
x
, i
y
+ 1), and three parity symmetries,
P
x
: (i
x
, i
y
) (i
x
, i
y
), P
y
: (i
x
, i
y
) (i
x
, i
y
), and
P
xy
: (i
x
, i
y
) (i
y
, i
x
). This is because the ansatz u
ij
is a many-to-one label of the physical spin state. The
non-invariance of the ansatz does not imply the non-
invariance of the corresponding physical spin state after
the projection. We only require the mean-eld ansatz to
be invariant up to a SU(2) gauge transformation in order
for the projected physical spin state to have a symmetry.
For example, a T
x
translation transformation changes the
U(1) ansatz Eq. (8) to
U
i,i+x
=
3
+i()
i
,
U
i,i+y
=
3
i()
i
,
The translated ansatz can be transformed into the orig-
inal ansatz via a SU(2) gauge transformation W
i
=
()
i
i
1
. Therefore, after the projection, the ansatz
Eq. (8) describes a T
x
translation symmetric spin state.
Using the similar consideration, one can show that
the SU(2), U(1), and Z
2
ansatz are invariant under
translation T
x,y
and parity P
x,y,xy
symmetry transfor-
mations followed by corresponding SU(2) gauge trans-
formations G
T
x
,T
y
and G
P
x
,P
y
,P
xy
respectively. Thus the
three ansatz all describe symmetric spin liquids. In the
following, we list the corresponding gauge transforma-
tions G
T
x
,T
y
and G
P
x
,P
y
,P
xy
for the three ansatz:
for the SU(2) ansatz Eq. (6):
G
T
x
(i) =()
i
x
G
T
y
(i) =
0
, G
P
xy
(i) =()
i
x
i
y

0
,
()
i
x
G
P
x
(i) =()
i
y
G
P
y
(i) =
0
, G
0
(i) =e
i
l

l
(11)
for the U(1) ansatz Eq. (8):
G
T
x
(i) =G
T
y
(i) = i()
i

1
, G
P
xy
(i) =i()
i

1
,
G
P
x
(i) =G
P
y
(i) =
0
, G
0
(i) =e
i
3
(12)
for the Z
2
ansatz Eq. (10):
G
T
x
(i) =G
T
y
(i) = i
0
, G
P
xy
(i) =
0
,
G
P
x
(i) =G
P
y
(i) = ()
i

1
, G
0
(i) =
0
(13)
In the above we also list the pure gauge transfor-
mation G
0
(i) that leave the ansatz invariant: u
ij
=
G
0
(i) u
ij
G

0
(j).
3. Denition of PSG
The SU(2), U(1) and Z
2
ansatz after the projection,
give rise to three spin liquid states. The three states
have the exactly the same symmetry. The question here
is whether the three spin liquids belong to the same phase
or not. According to Landaus symmetry breaking the-
ory, two states with the same symmetry belong to the
same phase. However, we now know that Landaus sym-
metry breaking theory does not describe all the phases.
It is possible that the three spin liquids contain dierent
orders that cannot be characterized by symmetries. The
10
issue here is to nd a new set of quantum numbers that
characterize the new orders.
To nd a new set of universal quantum numbers that
distinguish the three spin liquids, we note that although
the three spin liquids have the same symmetry, their
ansatz are invariant under the symmetry translations fol-
lowed by dierent gauge transformations (see Eq. (11),
Eq. (12), and Eq. (13)). So the invariant group of three
ansatz are dierent. We can use the invariant group of
the three ansatz to characterize the new order in the spin
liquid. In a sense, the invariant group dene a new order
quantum order.
The invariant group of an ansatz is formed by
all the combined symmetry transformations and the
gauge transformations that leave the ansatz in-
variant. Those combined transformations from a
group. Such a group is called the Projective
Symmetry Group (PSG). The combined transforma-
tions (G
T
x
T
x
, G
T
y
T
y
, G
P
x
P
x
, G
P
y
P
y
, G
P
xy
P
xy
) and G
0
in
Eq. (11), Eq. (12), and Eq. (13) generate the three PSGs
for the three ansatz Eq. (6), Eq. (8), and Eq. (10).
4. Properties of PSG
To understand the properties of the PSG, we would
like to point out that a PSG contains a special subgroup,
which will be called the invariant gauge group (IGG). An
IGG is formed by pure gauge transformations that leave
the ansatz unchanged
IGG {G
0
| u
ij
= G
0
(i)u
ij
G

0
(j)}
For the ansatz Eq. (6), Eq. (8), and Eq. (10), the IGGs
are SU(2), U(1), and Z
2
respectively. We note that
SU(2), U(1), and Z
2
happen to be the gauge groups that
describe the low energy gauge uctuations in the three
spin liquids This relation is not an accident. In general
the gauge group of the low energy gauge uctuations for
a spin liquid described by an ansatz u
ij
is given by the
IGG of the ansatz [16]. This result generalizes the anal-
ysis of the low energy gauge group based on the SU(2)
ux.
If an ansatz is invariant under the translation T
x
fol-
lowed by a gauge transformation G
x
, then it is also invari-
ant under the translation T
x
followed by another gauge
transformation G
0
G
x
, as long as G
0
IGG. So the
gauge transformation associated with a symmetry trans-
formation is not unique. The number of the choices of
the gauge transformations is the number of the elements
in IGG. We see that as sets, PSG = SGIGG where SG
is the symmetry group. But as groups, PSG is not the
direct product of SG and IGG. It is a twisted product.
Using the more rigorous mathematical notation, we have
SG = PSG/IGG
We may also say that the PSG is a projective extension
of SG by IGG.
The SU(2), U(1) and Z
2
ansatz all have the same sym-
metry and hence the same symmetry group SG. They
have dierent PSGs since the same SG is extended by
dierent IGGs. Here we would like to remark that even
for a given pair of SG and IGG, there are many dier-
ent ways to extent the SG by the IGG, leading to many
dierent PSGs. For example there are over 100 ways
to extend the symmetry group of a square lattice by a
Z
2
IGG. This implies that there are over 100 dierent
Z
2
spin liquids on a square lattice and those spin liquids
all have the exactly the same symmetry! Finding dier-
ent ways of extending a symmetry group SG is a pure
mathematical problem. Such a calculation will lead to
a (partial) classication of the quantum orders (and the
spin liquids).
5. PSG is a universal property which protects gapless
excitations
From the above discussion, we see that a PSG con-
tains two parts. The rst part is SG which describe the
symmetry of the spin liquid. The second part is IGG
which describe the gauge symmetry of the spin liquid.
A generic elements in the PSG is a combination of the
symmetry transformation and the gauge transformation.
We know that symmetry and gauge symmetry are
universal properties, ie perturbative uctuations cannot
break the symmetry, nor can they break the gauge sym-
metry. So both SG and IGG are universal properties.
This strongly suggests that the PSG is also a universal
property.
To directly show a PSG to be a universal property, we
note that the fermion mean-eld Hamiltonian H
mean
in
Eq. (3) is invariant under the lattice symmetry and the
SU(2) gauge transformations Eq. (4). But the mean-
eld ansatz u
ij
is not invariant under the separate lat-
tice symmetry and SU(2) gauge transformations. So
the mean-eld state break the separate lattice symmetry
and SU(2) gauge symmetry down to a smaller sym-
metry. The symmetry group of this smaller symmetry
is the PSG. So, the PSG is the symmetry of the mean-
eld theory with u
ij
ansatz. As a result, the PSG is the
symmetry for the eective Lagrangian L
u
ij
(, u
ij
) that
describes the uctuations around the mean-eld ansatz.
If the mean-eld uctuations do not have any infrared
divergence, then those uctuations will be perturbative
in nature and cannot change the symmetry the PSG.
What do we mean by perturbative uctuations can-
not change the PSG? We know that a mean-eld ground
state is characterized by u
ij
. If we include perturba-
tive uctuations to improve our calculation of the mean-
eld energy
u
ij
spin
|H|
u
ij
spin
, then we expect the u
ij
that
minimize the improved mean-eld energy to receive per-
turbative corrections u
ij
. The statement perturbative
uctuations cannot change the PSG? means that u
ij
and u
ij
+ u
ij
have the same PSG.
As the perturbative uctuations (by denition) do not
11
change the phase, u
ij
and u
ij
+ u
ij
describe the same
phase. In other words, we can group u
ij
into classes
(which are called universality classes) such that the u
ij
in each class are connected by the perturbative uctua-
tions. Each universality class describes one phase. We
see that, if the above argument is true, then the ansatz
in a universality class all share the same PSG. In other
words, the universality classes (or the phases) are classi-
ed by the PSGs. Thus the PSG is a universal property.
We can use the PSG to describe the quantum order in
the spin liquid, as long as the low energy eective theory
L
u
ij
(, u
ij
) does not have any infrared divergence.
In the standard renormalization group analysis of the
stability of a phase or a critical point, one needs to in-
clude all the counter terms that have the right symme-
tries into the eective Lagrangian, since those terms can
be generated by perturbative uctuations. Then we ex-
amine if those allowed counter terms are relevant pertur-
bations or not. In our problem, u
ij
discussed above cor-
respond to thoe counter terms. The eective Lagrangian
with the counter term is given by L
u
ij
+ u
ij
(, u
ij
).
The new feature here is that it is incorrect to use the
symmetry group alone to determine the allowed counter
terms u
ij
. We should use PSG to determine the al-
lowed counter terms in our analysis of the stability of
phases and critical points. That is the counter terms
u
ij
should not change the PSG of u
ij
.
The Z
2
spin liquid Eq. (10) (and other 100 plus Z
2
spin liquids) contains no diverging uctuations. So the
PSG description of the quantum order is valid for this
case. For the SU(2) spin liquid Eq. (6) and the U(1) spin
liquid Eq. (8), their low energy eective theory Eq. (7)
and Eq. (9) contain log divergence. These are marginal
cases where the PSG description of the quantum order
still apply. In a renormalization group analysis of the
stability of the U(1) spin liquid Eq. (8), one can show
that, in a large N limit, the counter terms allowed by
the U(1) PSG Eq. (12) are all irrelevant [62, 63], even if
we include the instanton eect [64]. Thus the (large N)
U(1) spin liquid is a stable quantum phase. One can also
show that non of the allowed counter terms can give the
gapless fermions and gapless gauge bosons an energy gap
[16, 31]. Thus the gapless excitations in the U(1) spin
liquid are protected by by the U(1) PSG despite those
gapless excitations interact down to zero energy.
E. An intuitive understand of quantum order and
the emergent gauge bosons and fermions
The projective construction produces a correlated
many-body ground state |
spin
= P|
u
ij
mean
. We may
view the complicated correlation in the ground state as
a pattern of quantum entanglement. The quantum order
and the associated PSG is a characterization of such a
pattern of entanglement. The gauge uctuation above
the many-body ground state can be viewed as a uctua-
tion of the entanglement. The fermion excitations can be
viewed as topological defects in the entanglement. From
this point of view, the theory of quantum order can be
regarded as a theory of many-body quantum entangle-
ment.
V. AN EXACT SOLUBLE MODEL FROM
PROJECTIVE CONSTRUCTION
Usually, the projective construction does not give us
exact results. In this section, we are going to construct an
exactly soluble model on 2D square lattice [15, 68]. The
model has a property that the projective construction
give us exact ground states and all other exact excited
states.
A. An exact soluble model for the -fermions
First, we would like to construct an exact soluble model
for the -fermions. It is convenient to write the exact
soluble Hamiltonian in terms of four Majorana fermions
2
1,i
=
x
i
+i
x
i
, 2
2,i
=
y
i
+i
y
i
(14)
The Majorana fermions satisfy the algebra {
a,i
,
b,j
} =
2
ab

ij
. where a, b = x, x, y, y. The exact soluble fermion
Hamiltonian is given by
H =

i
g

F
i
, (15)

F
i
=

U
i,i+x

U
i+x,i+x+y

U
i+x+y,i+y

U
i+y,i
,

U
i,i+ x
=
x
i

x
i+ x
,

U
i,i+ y
=
y
i

y
i+ y
,

U
ij
=

U
ji
.
To see why the above interacting fermion model is ex-
actly soluble, we note that

U
ij
commute with each other
and H commute with all the

U
ij
. So we can nd the
eigenvalues and eigenstates of H by nding the common
eigenstates of

U
ij
:

U
ij
|{s
ij
} = s
ij
|{s
ij
}
Since (

U
ij
)
2
= 1 and

U
ij
=

U
ji
, s
ij
satises s
ij
= i
and s
ij
= s
ji
. Since H is a function of

U
ij
s, |{s
ij
} is
also an energy eigenstate of Eq. (15) with energy
E =

i
gF
i
,
F
i
=s
i,i+x
s
i+x,i+x+y
s
i+x+y,i+y
s
i+y,i
. (16)
To see if |{s
ij
}s represent all the exact eigenstates
of H, we need count the states. Let us assume the 2D
square lattice to have N
site
lattice sites and a periodic
boundary condition in both directions. In this case the
lattice has 2N
site
links. Since there are total of 2
2N
site
dierent choices of s
ij
(two choices for each link), the
states |{s
ij
} exhaust all the 4
N
site
states in the (
1
,
2
)
12
Hilbert space. Thus the common eigenstates of

U
ij
is not
degenerate and the above approach allows us to obtain
all the eigenstates and eigenvalues of the H.
We note that the eigenstate |{s
ij
} is the ground state
of the following free fermion Hamiltonian
H
mean
=

ij
s
ij

U
ij
(17)
by choosing dierent s
ij
= i, the ground state of the
above mean-eld Hamiltonian give rise to all the eigen-
states of the interacting fermion Hamiltonian Eq. (15).
B. An exact soluble spin-1/2 model
We note that the Hamiltonian H can only change the
fermion number on each site, n
i
=

1,i

1,i
+

2,i

2,i
,
by an even number. Thus the H acts within a subspace
which has an even number of fermions on each site. We
will call such a subspace physical Hilbert space. The
physical Hilbert space has only two states per site cor-
responding to a spin-up and a spin-down state. When
restricted within the physical space, H actually describes
a spin-1/2 system. To obtain the corresponding spin-1/2
Hamiltonian, we note that

x
i
= i
y
i

x
i
,
y
i
= i
x
i

y
i
,
z
i
= i
x
i

x
i
(18)
act within the physical Hilbert space and satisfy the al-
gebra of Pauli matrices. Thus we can identify
l
i
as the
spin operator. Using the fact that
()
n
i
=
x
i

y
i

x
i

y
i
= 1
within the physical Hilbert space, we can show that the
fermion Hamiltonian Eq. (15) becomes (see Fig. 7)
H
spin
=

i
g

F
i
,

F
i
=
x
i

y
i+ x

x
i+ x+ y

y
i+ y
(19)
within the physical Hilbert space.
C. The projective construction leads to exact
results
All the states in the physical Hilbert space (ie all
the states in the spin-1/2 model) can be obtained from
the |{s
ij
} states by projecting into the physical Hilbert
space: P|{s
ij
}. The projection operator is given by
P =

i
1 + ()
n
i
2
Since [P, H] = 0, the projected state P|{s
ij
}, if non-
zero, remain to be an eigenstate of H (or H
spin
) and
remain to have the same eigenvalue. We see that the
ground state of the mean-eld Hamiltonian Eq. (17), af-
ter the projection, give rise to all the exact eigenstates of
the spin Hamiltonian Eq. (19). The project construction
is exact for Eq. (19)!
D. The Z
2
gauge structure
The physical states (with even numbers of fermions per
site) are invariant under local Z
2
transformations gener-
ated by

G =

i
G
n
i
i
where G
i
is an arbitrary function with only two values
1. The Z
2
transformation change
ai
to

ai
= G
i

ai
and s
ij
to s
ij
= G
i
s
ij
G
j
, or more precisely
|{ s
ij
} =

G|{s
ij
}
Using P

G = P, we nd that |{s
ij
} and |{ s
ij
} give
rise to the same physical state after projection (if their
projection is not zero):
P|{ s
ij
} = P|{s
ij
} (20)
Thus, s
ij
is a many-to-one label of the physical spin state.
The above results indicate that we can view is
ij
as a
Z
2
gauge potential and the local Z
2
transformation is a
Z
2
gauge transformation. Eq. (20) implies that gauge
equivalent gauge potential described the same physical
state. The uctuations of s
ij
is described by a Z
2
gauge
theory, which is the low energy eective theory of the
spin system Eq. (19).
VI. CLOSED-STRING CONDENSATION
The ground state of the exactly soluble model contains
a special property closed-string condensation. In this
section, we will see that the closed-string condensation is
intimately related to the emergence of the gauge struc-
ture and the fermions. We have shown that the ground
state of the exactly soluble model can be constructed via
the projective construction. This indicates that the pro-
jective construction is probably just a trick to construct
string condensed states.
A. String operators and closed-string condensation
First let us dene a string C as a curve that connect
the midpoints of neighboring links (see Fig. 7). The
string operator has the following form
W(C) =

l
n
i
n
(21)
where i
n
are sites on the string. l
n
= z if the string does
not turn at site i
n
. l
n
= x or y if the string makes a
turn at site i
n
. l
n
= x if the turn forms a upper-right or
lower-left corner. l
n
= y if the turn forms a lower-right
or upper-left corner. (See Fig. 7.)
By an explicit calculation, one can show that the closed
string operator dened above commute with the spin
13
F
i

y
FIG. 7: The string is formed by a curve connecting the mid-
points of the neighboring links. The string operator is form
by the product of
x,y,z
i
for sites on the string. The operator

F
i
is also presented.
Hamiltonian Eq. (19). So the ground state |ground of
the spin Hamiltonian is also an eigenstate of the closed-
string operator. Since the eigenvalues of the closed string
operator are 1. We have ground|W(C
closed
)|ground =
1. The ground state has a closed-string condensation.
We note that in a symmetry breaking state, the op-
erator representing the order parameter has a non-zero
expectation value

O
i
= . If we dene a string oper-
ator as the product of

O
i

along a loop, the average of

the string operator will be non-zero. However, we do not
regard the non-zero average of such a string operator to
represent a string condensation. The real closed-string
condensation must be unbreakable, in the sense that
we cannot break a closed string operator into several seg-
ments and nd condensation in each segment. The string
operator

i

O
i

does not satisfy this property. However,

the string operator dened in Eq. (21) is indeed unbreak-
able. This is because an open-string operator does not
commute with the spin Hamiltonian and does not con-
dense ground|W(C
open
)|ground = 0. It is such a un-
breakable closed-string condensation indicates a new or-
der in the ground state.
In terms of the Majorana fermions, the closed-string
can be written as
W(C
closed
) =

ij
i

U
ij
where ij are the nearest neighbor links that form the
closed string C
closed
. For a spin state obtained from the
ansatz s
ij
(via the projection, P|{s
ij
), we nd that
{s
ij
}|PW(C
closed
)P|{s
ij
} =

ij
is
ij
(note that for closed strings [W(C
closed
), P] = 0). There-
fore the closed string operator is nothing but the Wegner-
Wilson loop operator [22, 69] for the corresponding Z
2
gauge theory. We see an intimate relation between the
closed-string condensation and the emergence of a gauge
structure.
B. Open string operators and Z
2
charges
If the uctuations of s
ij
represent Z
2
gauge uctua-
tions, what are the Z
2
charges? In a gauge theory, we
know that a Wegner-Wilson operator for an open string
is not gauge invariant and is not a physical operator (ie is
not an operator that acts within the physical Hilbert
space). The open-string operator dened in Eq. (21)
act with the physical Hilbert space and is a gauge in-
variant physical operator. Thus the Wegner-Wilson op-
erator for an open string does not correspond to our
open string operator. However, in the gauge theory, two
charge operators connected by the Wegner-Wilson oper-
ator,

x
1

x
2
e
i
R
x
2
x
1
dxa
, is gauge invariant. It is such a
operator that correspond to our open string operator.
In the gage theory, the Wegner-Wilson loop operator
create a loop of electric ux. The closed-string operator
dened in Eq. (21) has the same physical meaning. In
the gage theory, the operator

x
1

x
2
e
i
R
x
2
x
1
dxa
creates
two charges connected by a line of the electric ux. Our
open string operator dened in Eq. (21) does the same
thing: it creates two Z
2
charges at its two ends and a
electric ux line connecting the two charges.
Due to the closed-string condensation, the string con-
necting the two Z
2
charges is unobservable and costs no
energy. Thus the open string does not create an extended
line-like object, it creates two point-like objects at its
ends. Those point-like objects are the Z
2
charges. De-
spite the point-like appearance, the Z
2
charges are in-
trinsically non-local. There is no way to create a lone Z
2
charge. Because of this, the Z
2
charge (and other gauge
charges) usually carry fractional quantum numbers.
C. Statistics of the Z
2
charges
What is the statistics of the Z
2
charges? Usually,
bosons are dened as particles described by commuting
operators and fermions as particles described by anti-
commuting operators. But this denition is too formal
and is hard to apply to our case. The nd a new way
to calculate statistics, we need to gain a more physi-
cal understanding of the dierence between bosons and
fermions.
Let us consider the following many-body hopping sys-
tem. The Hilbert space is formed by a zero-particle state
|0, one-particle states |i
1
, two-particle states |i
1
, i
2
,
etc , where i
n
labels the sites in a lattice. As an iden-
tical particle system, the state |i
1
, i
2
, ... does not de-
pend on the order of the indexes i
1
, i
2
, .... For example
|i
1
, i
2
= |i
2
, i
1
. There are no doubly-occupied sites and
we assume |i
1
, i
2
, ... = 0 if i
m
= i
n
.
A hopping operator

t
ij
is dened as follows. When

t
ij
acts on state |i
1
, i
2
, ..., if there is a particle at site j but
no particle at site i, then

t
ij
moves the particle at site j to
site i and multiplies a complex amplitude t(i, j; i
1
, i
2
, ...)
to the resulting state. Note that the amplitude may de-
14
Exchange No exchange
1
3
4
5
1
2
3
4
5
2
j
k l
i j
k l
i
FIG. 8: (a) The rst way to arrange the ve hops swaps the
two particles. (b) The second way to arrange the same ve
hops does not swap the two particles.
pend on the locations of all the other particles. The
Hamiltonian of our system is given by
H
hop
=

ij

t
ij
where the sum

ij
is over a certain set of pairs ij,
such as nearest-neighbor pairs. In order for the above
Hamiltonian to represent a local system we require that
[

t
ij
,

t
kl
] = 0
if i, j, k, and l are all dierent.
What does the hopping Hamiltonian H
hop
describe? A
hard-core boson system or a fermion system? Whether
a many-body hopping system is a boson system or a
fermion system (or even some other statistical systems)
has nothing to do with the Hilbert space. The fact that
the many-body states are labeled by symmetric indexes
(eg |i
1
, i
2
= |i
2
, i
1
) does not imply that the many-body
system is a boson system. The statistics are determined
by the Hamiltonian H
hop
.
Clearly, when the hopping amplitude t(i, j; i
1
, i
2
, ...)
only depends on i and j, t(i, j; i
1
, i
2
, ...) = t(i, j), the
many-body hopping Hamiltonian will describe a hard-
core boson system. The issue is under what condition
the many-body hopping Hamiltonian describes a fermion
system.
This problem was solved in Ref. [19]. It was found that
the many-body hopping Hamiltonian describes a fermion
system if the hopping operators satisfy

t
lk

t
il

t
lj
=

t
lj

t
il

t
lk
(22)
for any three hopping operators

t
lj
,

t
il
, and

t
lk
with i, j,
k, and l all being dierent. (Note that the algebra has a
structure

t
1

t
2

t
3
=

t
3

t
2

t
1
.)
To understand this result, consider the state |i, j, ....
with two particles at i, j, and possibly other particles
further away. We apply a set of ve hopping operators
{

t
jl
,

t
lk
,

t
il
,

t
lj
,

t
ki
} to the state |i, j, .... but with dier-
ent order (set Fig. 8)

t
jl

t
lk

t
il

t
lj

t
ki
|i, j, .... =C
1
|i, j, ....

t
jl

t
lj

t
il

t
lk

t
ki
|i, j, .... =C
2
|i, j, ....
k
l
j
k
i
B
A
i
j
l

y
2 1
3
FIG. 9: The Z
2
charge live on the links. The hopping of the
Z
2
charges is induced by the open string operator.
where we have assumed that there are no particles at
sites k and l. We note that after ve hops we get back
to the original state |i, j, .... with additional phases C
1,2
.
However, from Fig. 8, we see that the rst way to arrange
the ve hops (Fig. 8a) swaps the two particles at i and
j, while the second way (Fig. 8b) does not swap the two
particles. Since the two hopping schemes use the same
set of ve hops, the dierence between C
1
and C
2
is due
to exchanging the two particles. Thus we require C
1
=
C
2
in order for the many-body hopping Hamiltonian
to describe a fermion system. Noting that the rst and
the last hops are the same in the two hopping schemes,
we nd that C
1
= C
2
if the hopping operators satisfy
Eq. (22). Eq. (22) serves as an alternative denition of
Fermi statistics if we do not want to use anti-commuting
algebra.
For our spin model Eq. (19), the open strings end at
the midpoint of the links. So the Z
2
charges live on the
links. To apply the above result to the Z
2
charges, we
note that an open string operator connecting midpoints i
and j (see Fig. 9) play the role of a hopping operator

t
ij
.
Near the site 1 in Fig. 9, the hoping operators between
the midpoints i, j, k, and l are given by

t
lj
=
y
1
,

t
il
=
z
1
,

t
lk
=
x
1
.
The fermion hopping algebra Eq. (22) becomes
y
1

z
1

x
1
=

x
1

z
1

y
1
which is satised. Near the site 2 and 3 in Fig.
9, the hoping operators between the midpoints i, j, k,
and l are given by

t
lj
=
y
2
,

t
il
=
z
2
,

t
lk
=
x
3
The fermion hopping algebra becomes
y
2

z
2

x
3
=

x
3

z
2

y
2
which is again satised. We see that the hop-
ping operators of the Z
2
charges satisfy a fermion hop-
ping algebra. So the Z
2
charges are fermions. Fermions
can emerge in a pure bosonic model as ends of condensed
strings.
VII. SUMMARY
Symmetry breaking have dominated our understand-
ing of phase and phase transition for over 50 years. We
now know that symmetry breaking cannot describe all
the possible orders that matter can have. The study of
15
the quantum order [16] and the associated condensation
of string-nets [24, 39] and other extended objects suggest
that a new world beyond symmetry breaking exists. It
appears that the new world of quantum order is much
richer than the world of symmetry breaking order. The
most striking picture from the new world is that gauge
bosons, fermions, and string-net condensation are just
the dierent sides of the same coin. Or in other words,
the string-net condensation provides a way to unify gauge
interactions and Fermi statistics! So far, we have only
seen some small fragmented pieces of the new world. The
exciting time is still ahead of us. Comparing with our
understanding of symmetry breaking order, we need to
understand the following aspects of quantum order and
the associated string-net condensation:
(1) We know that the group theory is the mathematical
frame work behind the symmetry breaking order. What
is the mathematical frame work behind string condensa-
tion? PSG only provides a partial answer to this ques-
tion. A recent work [24] suggests that tensor category
theory may play the same role in string-net condensed
states as group theory in symmetry breaking states.
(2) We know that crystal orders can be measured through
X-ray diraction. How to measure dierent quantum or-
ders associated with dierent string-net condensations?
(3) We know that many material contain non-trivial sym-
metry breaking orders. What material has string-net
condensation and emergent gauge bosons and fermions?
(A believer can always say that we actually live inside
one such material. However, one needs more to convince
a non-believer.)
The more we explore, the more are we fascinated by
the endless richness that the nature reveals to us.
This research is supported by NSF Grant No. DMR
0123156 and by NSF-MRSEC Grant No. DMR02
13282.
[1] Y. Nambu, Phys. Rev. Lett. 4, 380 (1960).
[2] J. Goldstone, Nuovo Cimento 19, 154 (1961).
[3] L. D. Landau and E. M. Lifschitz, Statistical Physics -
Course of Theoretical Physics Vol 5 (Pergamon, London,
1958).
[4] L. D. Landau, Phys. Z. Sowjetunion 11, 26 (1937).
[5] D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys. Rev.
Lett. 48, 1559 (1982).
[6] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983).
[7] X.-G. Wen, Advances in Physics 44, 405 (1995).
[8] F. D. M. Haldane and E. H. Rezayi, Phys. Rev. B 31,
2529 (1985).
[9] X.-G. Wen and Q. Niu, Phys. Rev. B 41, 9377 (1990).
[10] X.-G. Wen, Int. J. Mod. Phys. B 4, 239 (1990).
[11] D. Arovas, J. R. Schrieer, and F. Wilczek, Phys. Rev.
Lett. 53, 722 (1984).
[12] B. I. Halperin, Phys. Rev. B 25, 2185 (1982).
[13] X.-G. Wen, Int. J. Mod. Phys. B 6, 1711 (1992).
[14] E. Witten, Comm. Math. Phys. 121, 351 (1989).
[15] A. Y. Kitaev, Ann. Phys. (N.Y.) 303, 2 (2003).
[16] X.-G. Wen, Phys. Rev. B 65, 165113 (2002).
[17] I. M. Lifshitz, Sov. Phys. JETP 11, 1130 (1960).
[18] X.-G. Wen, Phys. Rev. B 68, 115413 (2003).
[19] M. Levin and X.-G. Wen, Phys. Rev. B 67, 245316
(2003).
[20] M. Freedman, C. Nayak, K. Shtengel, K. Walker, and
Z. Wang, cond-mat p. 0307511 (2003).
[21] T. Banks, R. Myerson, and J. B. Kogut, Nucl. Phys. B
129, 493 (1977).
[22] F. Wegner, J. Math. Phys. 12, 2259 (1971).
[23] J. Kogut and L. Susskind, Phys. Rev. D 11, 395 (1975).
[24] M. Levin and X.-G. Wen, cond-mat/0404617 (2004).
[25] X.-G. Wen and Y.-S. Wu, Phys. Rev. Lett. 70, 1501
(1993).
[26] W. Chen, M. P. A. Fisher, and Y.-S. Wu, Phys. Rev. B
48, 13749 (1993).
[27] T. Senthil, J. B. Marston, and M. P. A. Fisher, Phys.
Rev. B 60, 4245 (1999).
[28] N. Read and D. Green, Phys. Rev. B 61, 10267 (2000).
[29] X.-G. Wen, Phys. Rev. Lett. 84, 3950 (2000).
[30] X.-G. Wen, Phys. Rev. Lett. 88, 11602 (2002).
[31] X.-G. Wen and A. Zee, Phys. Rev. B 66, 235110 (2002).
[32] R. Moessner and S. L. Sondhi, Phys. Rev. Lett. 86, 1881
(2001).
[33] N. Read and S. Sachdev, Phys. Rev. Lett. 66, 1773
(1991).
[34] X.-G. Wen, Phys. Rev. B 44, 2664 (1991).
[35] V. Kalmeyer and R. B. Laughlin, Phys. Rev. Lett. 59,
2095 (1987).
[36] X.-G. Wen, F. Wilczek, and A. Zee, Phys. Rev. B 39,
11413 (1989).
[37] D. Foerster, H. B. Nielsen, and M. Ninomiya, Phys. Lett.
B 94, 135 (1980).
[38] O. I. Motrunich and T. Senthil, Phys. Rev. Lett. 89,
277004 (2002).
[39] X.-G. Wen, Phys. Rev. D 68, 065003 (2003).
[40] B. Blok and X.-G. Wen, Phys. Rev. B 42, 8145 (1990).
[41] N. Read, Phys. Rev. Lett. 65, 1502 (1990).
[42] J. Fr ohlich and T. Kerler, Nucl. Phys. B 354, 369 (1991).
[43] D. S. Rokhsar and S. A. Kivelson, Phys. Rev. Lett. 61,
2376 (1988).
[44] N. Read and B. Chakraborty, Phys. Rev. B 40, 7133
(1989).
[45] E. Ardonne, P. Fendley, and E. Fradkin, Annals Phys.
310, 493 (2004).
[46] R. Moessner and S. L. Sondhi, Phys. Rev. B 68, 184512
(2003).
[47] T. Senthil and M. P. A. Fisher, Phys. Rev. B 62, 7850
(2000).
[48] S. Sachdev and K. Park, Annals of Physics (N.Y.) 298,
58 (2002).
[49] L. Balents, M. P. A. Fisher, and S. M. Girvin, Phys. Rev.
B 65, 224412 (2002).
[50] X.-G. Wen, Int. J. Mod. Phys. B 5, 1641 (1991).
[51] T. H. Hansson, V. Oganesyan, and S. L. Sondhi, cond-
mat/0404327 (2004).
[52] J. B. Kogut, Rev. Mod. Phys. 51, 659 (1979).
[53] L. B. Ioe, M. V. Feigelman, A. Ioselevich, D. Ivanov,
M. Troyer, and G. Blatter, Nature 415, 503 (2002).
[54] M. Freedman, M. Larsen, and Z. Wang, Commun. Math.
16
Phys. 227, 605 (2002).
[55] V. L. Ginzburg and L. D. Landau, Zh. Ekaper. Teoret.
Fiz. 20, 1064 (1950).
[56] I. Aeck and J. B. Marston, Phys. Rev. B 37, 3774
(1988).
[57] G. Baskaran, Z. Zou, and P. W. Anderson, Solid State
Comm. 63, 973 (1987).
[58] G. Kotliar and J. Liu, Phys. Rev. B 38, 5142 (1988).
[59] D. S. Marshall, D. S. Dessau, A. G. Loeser, C.-H. Park,
A. Y. Matsuura, J. N. Eckstein, I. Bozovic, P. Fournier,
A. Kapitulnik, W. E. Spicer, et al., Phys. Rev. Lett. 76,
4841 (1996).
[60] P. W. Anderson, Phys. Rev. 130, 439 (1963).
[61] P. W. Higgs, Phys. Rev. Lett. 12, 132 (1964).
[62] W. Rantner and X.-G. Wen, Phys. Rev. B 66, 144501
(2002).
[63] M. Hermele, T. Senthil, M. P. A. Fisher, P. A. Lee, N. Na-
gaosa, and X.-G. Wen, cond-mat/0404751 (2004).
[64] L. Ioe and A. Larkin, Phys. Rev. B 39, 8988 (1989).
[65] S. A. Kivelson, D. S. Rokhsar, and J. P. Sethna, Phys.
Rev. B 35, 8865 (1987).
[66] X.-G. Wen, Phys. Rev. B 60, 8827 (1999).
[67] J. K. Jain, Phys. Rev. B 41, 7653 (1991).
[68] X.-G. Wen, Phys. Rev. Lett. 90, 016803 (2003).
[69] K. G. Wilson, Phys. Rev. D 10, 2445 (1974).
[70] All the observed particles can be treated as massless when
compared with Planck mass.