Anyon 2
Anyon 2
Anyon 2
Chetan Nayak1,2 , Steven H. Simon3 , Ady Stern4 , Michael Freedman1 , Sankar Das Sarma5 ,
1
Microsoft Station Q, University of California, Santa Barbara, CA 93108
2
Department of Physics and Astronomy, University of California, Los Angeles, CA 90095-1547
3
Alcatel-Lucent, Bell Labs, 600 Mountain Avenue, Murray Hill, New Jersey 07974
4
Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 76100, Israel
5
Department of Physics, University of Maryland, College Park, MD 20742
Topological quantum computation has recently emerged as one of the most exciting approaches to constructing a
fault-tolerant quantum computer. The proposal relies on the existence of topological states of matter whose quasi-
particle excitations are neither bosons nor fermions, but are particles known as Non-Abelian anyons, meaning that
they obey non-Abelian braiding statistics. Quantum information is stored in states with multiple quasiparticles,
which have a topological degeneracy. The unitary gate operations which are necessary for quantum computation
are carried out by braiding quasiparticles, and then measuring the multi-quasiparticle states. The fault-tolerance
of a topological quantum computer arises from the non-local encoding of the states of the quasiparticles, which
makes them immune to errors caused by local perturbations. To date, the only such topological states thought
to have been found in nature are fractional quantum Hall states, most prominently the ν = 5/2 state, although
several other prospective candidates have been proposed in systems as disparate as ultra-cold atoms in optical
lattices and thin film superconductors. In this review article, we describe current research in this field, focusing
on the general theoretical concepts of non-Abelian statistics as it relates to topological quantum computation, on
understanding non-Abelian quantum Hall states, on proposed experiments to detect non-Abelian anyons, and on
proposed architectures for a topological quantum computer. We address both the mathematical underpinnings of
topological quantum computation and the physics of the subject using the ν = 5/2 fractional quantum Hall state
as the archetype of a non-Abelian topological state enabling fault-tolerant quantum computation.
is not only evolving alongside topological quantum computa- ical system ‘condenses’ into a non-Abelian topological phase.
tion but is even informed by it. Therefore, this review must In Section III, we describe the universal low-energy, long-
necessarily be rather sweeping in scope, simply to introduce distance physics of such phases. We also discuss how they can
the concepts of non-Abelian anyons and topological quantum be experimentally detected in the quantum Hall regime, and
computation, their inter-connections, and how they may be re- when they might occur in other physical systems. Our focus
alized in physical systems, particularly in several fractional is on a sequence of universality classes of non-Abelian topo-
quantum Hall states. (For a popular account, see Collins, logical phases, associated with SU(2)k Chern-Simons theory
2006; for a slightly more technical one, see Das Sarma et al., which we describe in section III.A. The first interesting mem-
2006a.) This exposition will take us on a tour extending from ber of this sequence, k = 2, is realized in chiral p-wave
knot theory and topological quantum field theory to conformal superconductors and in the leading theoretical model for the
field theory and the quantum Hall effect to quantum computa- ν = 5/2 fractional quantum Hall state. Section III.B shows
tion and all the way to the physics of gallium arsenide devices. how this universality class can be understood with conven-
The body of this paper is composed of three parts, Sections tional BCS theory. In section III.C, we describe how the topo-
II, III, and IV. Section II is rather general, avoids technical logical properties of the entire sequence of universality classes
details, and aims to introduce concepts at a qualitative level. (of which k = 2 is a special case) can be understood using
Section II should be of interest, and should be accessible, to Witten’s celebrated connection between Chern-Simons theory
all readers. In Section III we describe the theory of topolog- and the Jones polynomial of knot theory. In section III.D, we
ical phases in more detail. In Section IV, we describe how a describe an alternate formalism for understanding the topo-
topological phase can be used as a platform for fault-tolerant logical properties of Chern-Simons theory, namely through
quantum computation. The second and third parts are proba- conformal field theory. The discussion revolves around the
bly of more interest to theorists, experienced researchers, and application of this formalism to fractional quantum Hall states
those who hope to conduct research in this field. and explains how non-Abelian quantum Hall wavefunctions
Section II.A.1 begins by discussing the concept of braiding can be constructed with conformal field theory. Appendix A
statistics in 2 + 1-dimensions. We define the idea of a non- gives a highly-condensed introduction to conformal field the-
Abelian anyon, a particle exhibiting non-Abelian braiding ory. In Section III.E, we discuss the gapless edge excitations
statistics. Section II.A.2 discusses how non-Abelian anyons which necessarily accompany chiral (i.e. parity, P and time-
can arise in a many-particle system. We then review the ba- reversal T -violating) topological phases. These excitations
sic ideas of quantum computation, and the problems of errors are useful for interferometry experiments, as we discuss in
and decoherence in section II.B.1. Those familiar with quan- Section III.F. Finally, in Section III.G, we discuss topological
tum computation may be able to skip much of this section. We phases which do not violate parity and time-reversal symme-
explain in section II.B.2 how non-Abelian statistics naturally tries. These phases emerge in models of electrons, spins, or
leads to the idea of topological quantum computation, and ex- bosons on lattices which could describe transition metal ox-
plain why it is a good approach to error-free quantum compu- ides, Josephson junction arrays, or ultra-cold atoms in optical
tation. In section II.C, we briefly describe the non-Abelian lattices.
quantum Hall systems which are the most likely arena for
observing non-Abelian anyons (and, hence, for producing a In Section IV, we discuss how quasiparticles in topological
topological quantum computer). Section II.C.1 gives a very phases can be used for quantum computation. We first dis-
basic review of quantum Hall physics. Experts in quantum cuss the case of SU(2)2 , which is the leading candidate for
Hall physics may be able to skip much of this section. Section the ν = 5/2 fractional quantum Hall state. We show in Sec-
II.C.2 introduces non-Abelian quantum Hall states. This sec- tion IV.A how qubits and gates can be manipulated in a gated
tion also explains the importance (and summarizes the results) GaAs device supporting this quantum Hall state. We discuss
of numerical work in this field for determining which quantum why quasiparticle braiding alone is not sufficient for universal
Hall states are (or might be) non-Abelian. Section II.C.3 de- quantum computation and how this limitation of the ν = 5/2
scribes some of the proposed interference experiments which state can be circumvented. Section IV.B discusses in detail
may be able to distinguish Abelian from non-Abelian quan- how topological computations can be performed in the sim-
tum Hall states. Section II.C.4 shows how qubits and ele- plest non-Abelian theory that is capable of universal topolog-
mentary gates can be realized in a quantum Hall device. Sec- ical quantum computation, the so-called “Fibonacci-Anyon”
tion II.C.5 discusses some of the engineering issues associ- theory. In IV.C, we show that the SU(2)k theories support
ated with the physical systems where quantum Hall physics universal topological quantum computation for all integers k
is observed. In section II.D we discuss some of the other, except k = 1, 2, 4. In IV.D, we discuss the physical processes
non-quantum-Hall systems where it has been proposed that which will cause errors in a topological quantum computer.
non-Abelian anyons (and hence topological quantum compu-
tation) might occur.
Sections III and IV are still written to be accessible to the Finally, we briefly conclude in section V. We discuss
broadest possible audiences, but they should be expected to questions for the immediate future, primarily centered on the
be somewhat harder going than Section II. Section III intro- ν = 5/2 and ν = 12/5 fractional quantum Hall states. We
duces the theory of topological phases in detail. Topological also discuss a broader set of question relating to non-Abelian
quantum computation can only become a reality if some phys- topological phases and fault-tolerant quantum computation.
3
II. BASIC CONCEPTS Consequently, the notion of a winding of one particle around
another in two dimensions is well-defined. Then, when two
A. Non-Abelian Anyons particles are interchanged twice in a clockwise manner, their
trajectory involves a non-trivial winding, and the system does
1. Non-Abelian Braiding Statistics not necessarily come back to the same state. This topological
difference between two and three dimensions, first realized
Quantum statistics is one of the basic pillars of the quan- by Leinaas and Myrheim, 1977 and by Wilczek, 1982a, leads
tum mechanical view of the world. It is the property which to a profound difference in the possible quantum mechanical
distinguishes fermions from bosons: the wave function that properties, at least as a matter of principle, for quantum sys-
describes a system of many identical particles should satisfy tems when particles are confined to 2 + 1 D (see also Goldin
the proper symmetry under the interchange of any two parti- et al., 1981 and Wu, 1984). (As an aside, we mention that in
cles. In 3 spatial dimension and one time dimension (3 + 1 D) 1 + 1 D, quantum statistics is not well-defined since particle
there are only two possible symmetries — the wave function interchange is impossible without one particle going through
of bosons is symmetric under exchange while that of fermions another, and bosons with hard-core repulsion are equivalent to
is anti-symmetric. One cannot overemphasize, of course, the fermions.)
importance of the symmetry of the wavefunction, which is Suppose that we have two identical particles in two dimen-
the root of the Pauli principle, superfluidity, the metallic state, sions. Then when one particle is exchanged in a counter-
Bose-Einstein condensation, and a long list of other phenom- clockwise manner with the other, the wavefunction can
ena. change by an arbitrary phase,
The limitation to one of two possible types of quantum
symmetry originates from the observation that a process in ψ (r1 , r2 ) → eiθ ψ (r1 , r2 ) (1)
which two particles are adiabatically interchanged twice is The phase need not be merely a ± sign because a second
equivalent to a process in which one of the particles is adi- counter-clockwise exchange need not lead back to the initial
abatically taken around the other. Since, in three dimensions, state but can result in a non-trivial phase:
wrapping one particle all the way around another is topolog-
ically equivalent to a process in which none of the particles ψ (r1 , r2 ) → e2iθ ψ (r1 , r2 ) (2)
move at all, the wave function should be left unchanged by
two such interchanges of particles. The only two possibili- The special cases θ = 0, π correspond to bosons and fermions,
ties are for the wavefunction to change by a ± sign under a respectively. Particles with other values of the ‘statistical an-
single interchange, corresponding to the cases of bosons and gle’ θ are called anyons (Wilczek, 1990). We will often refer
fermions, respectively. to such particles as anyons with statistics θ.
Let us now consider the general case of N particles, where
We can recast this in path integral language. Suppose we
a more complex structure arises. The topological classes of
consider all possible trajectories in 3 + 1 dimensions which
trajectories which take these particles from initial positions
take N particles from initial positions R1 , R2 , . . ., RN at
R1 , R2 , . . ., RN at time ti to final positions R1 , R2 , . . ., RN
time ti to final positions R1 , R2 , . . ., RN at time tf . If the
at time tf are in one-to-one correspondence with the elements
particles are distinguishable, then there are no topologically
of the braid group BN . An element of the braid group can
non-trivial trajectories, i.e. all trajectories can be continu-
be visualized by thinking of trajectories of particles as world-
ously deformed into the trajectory in which the particles do
lines (or strands) in 2+1 dimensional space-time originating
not move at all (straight lines in the time direction). If the
at initial positions and terminating at final positions, as shown
particles are indistinguishable, then the different trajectories
in Figure 1. The time direction will be represented vertically
fall into topological classes corresponding to the elements of
on the page, with the initial time at the bottom and the final
the permutation group SN , with each element of the group
time at the top. An element of the N -particle braid group is
specifying how the initial positions are permuted to obtain the
an equivalence class of such trajectories up to smooth defor-
final positions. To define the quantum evolution of such a sys-
mations. To represent an element of a class, we will draw
tem, we must specify how the permutation group acts on the
the trajectories on paper with the initial and final points or-
states of the system. Fermions and bosons correspond to the
dered along lines at the initial and final times. When drawing
only two one-dimensional irreducible representations of the
the trajectories, we must be careful to distinguish when one
permutation group of N identical particles.1
strand passes over or under another, corresponding to a clock-
Two-dimensional systems are qualitatively different from
wise or counter-clockwise exchange. We also require that
three (and higher dimensions) in this respect. A particle loop
any intermediate time slice must intersect N strands. Strands
that encircles another particle in two dimensions cannot be
cannot ‘double back’, which would amount to particle cre-
deformed to a point without cutting through the other particle.
ation/annihilation at intermediate stages. We do not allow this
because we assume that the particle number is known. (We
will consider particle creation/annihilation later in this paper
1
when we discuss field theories of anyons and, from a mathe-
Higher dimensional representations of the permutation group, known as
‘parastatistics’, can always be decomposed into fermions or bosons with matical perspective, when we discuss the idea of a “category”
an additional quantum number attached to each particle (Doplicher et al., in section IV below.) Then, the multiplication of two ele-
1971, 1974). ments of the braid group is simply the successive execution
4
statement that no local perturbation can have nonzero matrix are so-named for reasons which will become clear in sections
elements within this degenerate space. III.D and III.E), SU(2)2 , and chiral p-superconductors. There
A system with anyonic particles must generally have mul- are slight differences between these three theories, relating to
tiple types of anyons. For instance, in a system with Abelian Abelian phases, but these are unimportant for the present dis-
anyons with statistics θ, a bound state of two such particles cussion. This model has three different types of anyons, which
has statistics 4θ. Even if no such stable bound state exists, we can be variously called 1, σ, ψ or 0, 21 , 1. (Unfortunately, the
may wish to bring two anyons close together while all other notation is a little confusing because the trivial particle is
particles are much further away. Then the two anyons can called ‘1’ in the first model but ‘0‘ in the second, however,
be approximated as a single particle whose quantum num- we will avoid confusion by using bold-faced 1 to denote the
bers are obtained by combining the quantum numbers, in- trivial particle.) The fusion rules for such anyons are
cluding the topological quantum numbers, of the two parti-
cles. As a result, a complete description of the system must σ × σ = 1 + ψ, σ × ψ = σ, ψ × ψ = 1,
also include these ‘higher’ particle species. For instance, if 1 × x = x for x = 1, σ, ψ (7)
there are θ = π/m anyons in system, then there are also
θ = 4π/m, 9π/m, . . . , (m − 1)2 π/m. Since the statistics pa- (Translating these rules into the notation of SU(2)2 , we see
rameter is only well-defined up to 2π, θ = (m − 1)2 π/m = that these fusion rules are very similar to the decomposition
−π/m for m even and π − π/m for m odd. The formation rules for tensor products of irreducible SU(2) representations,
of a different type of anyon by bringing together two anyons but differ in the important respect that 1 is the maximum spin
is called fusion. When a statistics π/m particle is fused with so that 21 × 21 = 0 + 1, as in the SU(2) case, but 21 × 1 = 12 and
a statistics −π/m particle, the result has statistics θ = 0. It 1 × 1 = 0.) Note that there are two different fusion channels
is convenient to call this the ‘trivial’ particle. As far as topo- for two σs. As a result, if there are four σs which fuse together
logical properties are concerned, such a boson is just as good to give 1, there is a two-dimensional space of such states. If
as the absence of any particle, so the ‘trivial’ particle is also we divided the four σs into two pairs, by grouping particles
sometimes simply called the ‘vacuum’. We will often denote 1, 2 and 3, 4, then a basis for the two-dimensional space is
the trivial particle by 1. given by the state in which 1, 3 fuse to 1 or 1, 3 fuse to ψ (2, 4
With Abelian anyons which are made by forming succes- must fuse to the same particle type as 1, 3 do in order that all
sively larger composites of π/m particles, the fusion rule is: four particles fuse to 1). We can call these states Ψ1 and Ψψ ;
n2 π k2 π (n+k)2 π they are a basis for the four-quasiparticle Hilbert space with
m × m = m . (We will use a × b to denote a fused
total topological charge 1. (Similarly, if they all fused to give
with b.) However, for non-Abelian anyons, the situation is
ψ, there would be another two-dimensional degenerate space;
more complicated. As with ordinary quantum numbers, there
one basis is given by the state in which the first pair fuses to 1
might not be a unique way of combining topological quantum
while the second fuses to ψ and the state in which the opposite
numbers (e.g. two spin-1/2 particles could combine to form
occurs.)
either a spin-0 or a spin-1 particle). The different possibili-
Of course, our division of the four σs into two pairs was
ties are called the different fusion channels. This is usually
arbitrary. We could have divided them differently, say, into
denoted by
X the pairs 1, 3 and 2, 4. We would thereby obtain two dif-
φa × φb = Nabc
φc (6) ferent basis states, Ψ̃1 and Ψ̃ψ , in which both pairs fuse to
c 1 or to ψ, respectively. This is just a different basis in the
same two-dimensional space. The matrix parametrizing this
which represents the fact that when a particle of species a
basis change (see also Appendix A) is called the F -matrix:
fuses with one of species b, the result can be a particle of
c Ψ̃a = Fab Ψb , where a, b = 1, ψ. There should really be
species c if Nab 6= 0. For Abelian anyons, the fusion mul-
c c′
6 indices on h F ifi we include indices to specify the 4 parti-
tiplicities Nab = 1 for only one value of c and Nab = 0 for
all c′ 6= c. For particles of type k with statistics θk = πk 2 /m, cle types: Flijk , but we have dropped these other indices
ab
k′′ since i = j = k = l = σ in our case. The F -matrices
i.e. Nkk ′ = δk+k′ ,k′′ . For non-Abelian anyons, there is at
least one a, b such that there are multiple fusion channels c are sometimes called 6j symbols since they are analogous to
with Nab c
6= 0. In the examples which we will be considering the corresponding quantities for SU(2) representations. Recall
in this paper, Nab c
= 0 or 1, but there are theories for which that in SU(2), there are multiple states in which spins j1 , j2 , j3
c
Nab > 1 for some a, b, c. In this case, a and b can fuse to form couple to form a total spin J. For instance, j1 and j2 can add
c in Nab c
> 1 different distinct ways. We will use ā to denote to form j12 , which can then add with j3 to give J. The eigen-
2
the antiparticle of particle species a. When a and ā fuse, they states of (j12 ) form a basis of the different states with fixed
1 j1 , j2 , j3 , and J. Alternatively, j2 and j3 can add to form j23 ,
can always fuse to 1 in precisely one way, i.e. Naā = 1; in
2
the non-Abelian case, they may or may not be able to fuse to which can then add with j1 to give J. The eigenstates of (j23 )
other particle types as well. form a different basis. The 6j symbol gives the basis change
The different fusion channels are one way of accounting for between the two. The F -matrix of a system of anyons plays
the different degenerate multi-particle states. Let us see how the same role when particles of topological charges i, j, k fuse
this works in one simple model of non-Abelian anyons which to total topological charge l. If i and j fuse to a, which then
we discuss in more detail in section III. As we discuss in sec- fuses with k to give topological charge l, the different allowed
tion III, this model is associated with ‘Ising anyons’ (which a define a basis. If j and k fuse to b and then fuse with i to
6
give topological charge l, this defines another basis, and the not tell us when and where they might occur in nature. Elec-
F -matrix is the unitary transformation between the two bases. trons, protons, atoms, and photons, are all either fermions
States with more than 4 quasiparticles can be understood by or bosons even when they are confined to move in a two-
successively fusing additional particles, in a manner described dimensional plane. However, if a system of many electrons (or
in Section III.A. The F -matrix can be applied to any set of 4 bosons, atoms, etc.) confined to a two-dimensional plane has
consecutively fused particles. excitations which are localized disturbances of its quantum-
The different states in this degenerate multi-anyon state mechanical ground state, known as quasiparticles, then these
space transform into each other under braiding. However, two quasiparticles can be anyons. When a system has anyonic
particles cannot change their fusion channel simply by braid- quasiparticle excitations above its ground state, it is in a topo-
ing with each other since their total topological charge can logical phase of matter. (A more precise definition of a topo-
be measured along a far distant loop enclosing the two parti- logical phase of matter will be given in Section III.)
cles. They must braid with a third particle in order to change Let us see how anyons might arise as an emergent prop-
their fusion channel. Consequently, when two particles fuse erty of a many-particle system. For the sake of concreteness,
in a particular channel (rather than a linear superposition of consider the ground state of a 2 + 1 dimensional system of
channels), the effect of taking one particle around the other of electrons, whose coordinates are (r1 , . . . , rn ). We assume
is just multiplication by a phase. This phase resulting from that the ground state is separated from the excited states by
a counter-clockwise exchange of particles of types a and b an energy gap (i.e, it is incompressible), as is the situation in
which fuse to a particle of type c is called Rcab . In the Ising fractional quantum Hall states in 2D electron systems. The
anyon case, as we will derive in section III and Appendix A.1, lowest energy electrically-charged excitations are known as
R1σσ = e−πi/8 , Rψ σσ
= e3πi/8 , R1ψψ = −1, Rσσψ = i. For quasiparticles or quasiholes, depending on the sign of their
an example of how this works, suppose that we create a pair electric charge. (The term “quasiparticle” is also sometimes
of σ quasiparticles out of the vacuum. They will necessarily used in a generic sense to mean both quasiparticle and quasi-
fuse to 1. If we take one around another, the state will change hole as in the previous paragraph). These quasiparticles are
by a phase e−πi/8 . If we take a third σ quasiparticle and take local disturbances to the wavefunction of the electrons corre-
it around one, but not both, of the first two, then the first two sponding to a quantized amount of total charge.
will now fuse to ψ, as we will show in Sec. III. If we now take We now introduce into the system’s Hamiltonian a scalar
one of the first two around the other, the state will change by potential composed of many local “traps”, each sufficient to
a phase e3πi/8 . capture exactly one quasiparticle. These traps may be cre-
In order to fully specify the braiding statistics of a system ated by impurities, by very small gates, or by the potential
of anyons, it is necessary to specify (1) the particle species, created by tips of scanning microscopes. The quasiparticle’s
c
(2) the fusion rules Nab , (3) the F -matrices, and (4) the R- charge screens the potential introduced by the trap and the
matrices. In section IV, we will introduce the other sets of pa- “quasiparticle-tip” combination cannot be observed by local
rameters, namely the topological spins Θa and the S-matrix, measurements from far away. Let us denote the positions of
which, together with the parameters 1-4 above fully charac- these traps to be (R1 , . . . , Rk ), and assume that these posi-
terize the topological properties of a system of anyons. Some tions are well spaced from each other compared to the mi-
readers may be familiar with the incarnation of these mathe- croscopic length scales. A state with quasiparticles at these
matical structures in conformal field theory (CFT), where they positions can be viewed as an excited state of the Hamiltonian
occur for reasons which we explain in section III.D; we briefly of the system without the trap potential or, alternatively, as
review these properties in the CFT context in Appendix A. the ground state in the presence of the trap potential. When
Quasiparticles obeying non-Abelian braiding statistics or, we refer to the ground state(s) of the system, we will often be
simply non-Abelian anyons, were first considered in the con- referring to multi-quasiparticle states in the latter context. The
text of conformal field theory by Moore and Seiberg, 1988, quasiparticles’ coordinates (R1 , . . . , Rk ) are parameters both
1989 and in the context of Chern-Simons theory by Witten, in the Hamiltonian and in the resulting ground state wavefunc-
1989. They were discussed in the context of discrete gauge tion for the electrons.
theories and linked to the representation theory of quantum We are concerned here with the effect of taking these quasi-
groups by Bais, 1980; Bais et al., 1992, 1993a,b. They were particles around each other. We imagine making the quasi-
discussed in a more general context by Fredenhagen et al., particles coordinates R = (R1 , . . . , Rk ) adiabatically time-
1989 and Fröhlich and Gabbiani, 1990. The properties of dependent. In particular, we consider a trajectory in which the
non-Abelian quasiparticles make them appealing for use in a final configuration of quasiparticles is just a permutation of
quantum computer. But before discussing this, we will briefly the initial configuration (i.e. at the end, the positions of the
review how they could occur in nature and then the basic ideas quasiparticles are identical to the intial positions, but some
behind quantum computation. quasiparticles may have interchanged positions with others.)
If the ground state wave function is single-valued with respect
to (R1 , .., Rk ), and if there is only one ground state for any
2. Emergent Anyons given set of Ri ’s, then the final ground state to which the sys-
tem returns to after the winding is identical to the initial one,
The preceding considerations show that exotic braiding up to a phase. Part of this phase is simply the dynamical phase
statistics is a theoretical possibility in 2 + 1-D, but they do which depends on the energy of the quasiparticle state and
7
the
R length of time for the process. In the adiabatic limit, it is Where R(s), s ∈ [0, 2π] is the closed trajectory of the par-
~
dtE(R(t)). There is also a a geometric phase which does ticles and the path-ordering symbol P is defined by the sec-
not depend on how long the process takes. This Berry phase ond equality. Again, the matrix Mab may be the product of
is (Berry, 1984), topological and non-topological parts. In a system in which
I quasiparticles obey non-Abelian braiding statistics, the non-
α = i dR · hψ(R)|∇R~ |ψ(R)i (8) topological part will be Abelian, that is, proportional to the
unit matrix. Only the topological part will be non-Abelian.
The requirements for quasiparticles to follow non-Abelian
where |ψ(R)i is the ground state with the quasiparticles at po-
statistics are then, first, that the N -quasiparticle ground state
sitions R, and where the integral is taken along the trajectory
is degenerate. In general, the degeneracy will not be exact,
R(t). It is manifestly dependent only on the trajectory taken
but it should vanish exponentially as the quasiparticle sepa-
by the particles and not on how long it takes to move along
rations are increased. Second, that adiabatic interchange of
this trajectory.
quasiparticles applies a unitary transformation on the ground
The phase α has a piece that depends on the geometry of
state, whose non-Abelian part is determined only by the topol-
the path traversed (typically proportional to the area enclosed
ogy of the braid, while its non-topological part is Abelian. If
by all of the loops), and a piece θ that depends only on the
the particles are not infinitely far apart, and the degeneracy
topology of the loops created. If θ 6= 0, then the quasipar-
is only approximate, then the adiabatic interchange must be
ticles excitations of the system are anyons. In particular, if
done faster than the inverse of the energy splitting (Thouless
we consider the case where only two quasiparticles are inter-
and Gefen, 1991) between states in the nearly-degenerate sub-
changed clockwise (without wrapping around any other quasi-
space (but, of course, still much slower than the energy gap
particles), θ is the statistical angle of the quasiparticles.
between this subspace and the excited states). Third, the only
There were two key conditions to our above discussion of
way to make unitary operations on the degenerate ground state
the Berry phase. The single valuedness of the wave function
space, so long as the particles are kept far apart, is by braid-
is a technical issue. The non-degeneracy of the ground state,
ing. The simplest (albeit uninteresting) example of degenerate
however, is an important physical condition. In fact, most of
ground states may arise if each of the quasiparticles carried a
this paper deals with the situation in which this condition does
spin 1/2 with a vanishing g–factor. If that were the case, the
not hold. We will generally be considering systems in which,
system would satisfy the first requirement. Spin orbit cou-
once the positions (R1 , .., Rk ) of the quasiparticles are fixed,
pling may conceivably lead to the second requirement being
there remain multiple degenerate ground states (i.e. ground
satisfied. Satisfying the third one, however, is much harder,
states in the presence of a potential which captures quasipar-
and requires the subtle structure that we describe below.
ticles at positions (R1 , .., Rk )), which are distinguished by a
The degeneracy of N -quasiparticle ground states is condi-
set of internal quantum numbers. For reasons that will be-
tioned on the quasiparticles being well separated from one an-
come clear later, we will refer to these quantum numbers as
other. When quasiparticles are allowed to approach one an-
“topological”.
other too closely, the degeneracy is lifted. In other words,
When the ground state is degenerate, the effect of a closed
when non-Abelian anyonic quasiparticles are close together,
trajectory of the Ri ’s is not necessarily just a phase factor.
their different fusion channels are split in energy. This depen-
The system starts and ends in ground states, but the initial and
dence is analogous to the way the energy of a system of spins
final ground states may be different members of this degen-
depends on their internal quantum numbers when the spins are
erate space. The constraint imposed by adiabaticity in this
close together and their coupling becomes significant. The
case is that the adiabatic evolution of the state of the system is
splitting between different fusion channels is a means for a
confined to the subspace of ground states. Thus, it may be ex-
measurement of the internal quantum state, a measurement
pressed as a unitary transformation within this subspace. The
that is of importance in the context of quantum computation.
inner product in (8) must be generalized to a matrix of such
inner products:
~ R |ψb (R)i
mab = hψa (R)|∇ (9) B. Topological Quantum Computation
where |ψa (R)i, a = 1, 2, . . . , g are the g degenerate ground 1. Basics of Quantum Computation
states. Since these matrices at different points R~ do not com-
mute, we must path-order the integral in order to compute the As the components of computers become smaller and
transformation rule for the state, ψa → Mab ψb where smaller, we are approaching the limit in which quantum ef-
I fects become important. One might ask whether this is a prob-
lem or an opportunity. The founders of the field of quantum
Mab = P exp i dR · m
computation (Manin, 1980, Feynman, 1982, 1986, Deutsch,
X∞ Z 2π Z s1
Z sn−1 h 1985, and most dramatically, Shor, 1994) answered in favor of
= in ds1 ds2 . . . dsn Ṙ(s1 )·maa1 (R(s1 )) . . . the latter. They showed that a computer which operates coher-
n=0 0 0 0 ently on quantum states has potentially much greater power
i
than a classical computer (Nielsen and Chuang, 2000).
Ṙ(sn ) · man b (R(sn )) (10)
The problem which Feynman had in mind for a quantum
8
computer was the simulation of a quantum system (Feynman, tem to traverse many trajectories in parallel, and determine
1982). He showed that certain many-body quantum Hamil- its state by their coherent sum. In some sense this coherent
tonians could be simulated exponentially faster on a quantum sum amounts to a massive quantum parallelism. It should
computer than they could be on a classical computer. This not, however, be confused with classical parallel computing,
is an extremely important potential application of a quantum where many computers are run in parallel, and no coherent
computer since it would enable us to understand the properties sum takes place.
of complex materials, e.g. solve high-temperature supercon- The biggest obstacle to building a practical quantum com-
ductivity. Digital simulations of large scale quantum many- puter is posed by errors, which would invariably happen dur-
body Hamiltonians are essentially hopeless on classical com- ing any computation, quantum or classical. For any compu-
puters because of the exponentially-large size of the Hilbert tation to be successful one must devise practical schemes for
space. A quantum computer, using the physical resource of an error correction which can be effectively implemented (and
exponentially-large Hilbert space, may also enable progress which must be sufficiently fault-tolerant). Errors are typically
in the solution of lattice gauge theory and quantum chromo- corrected in classical computers through redundancies, i.e., by
dynamics, thus shedding light on strongly-interacting nuclear keeping multiple copies of information and checking against
forces. these copies.
In 1994 Peter Shor found an application of a quantum com- With a quantum computer, however, the situation is more
puter which generated widespread interest not just inside but complex. If we measure a quantum state during an interme-
also outside of the physics community (Shor, 1994). He in- diate stage of a calculation to see if an error has occurred, we
vented an algorithm by which a quantum computer could collapse the wave function and thus destroy quantum super-
find the prime factors of an m digit number in a length of positions and ruin the calculation. Furthermore, errors need
time ∼ m2 log m log log m. This is much faster than the not be merely a discrete flip of |0i to |1i, but can be continu-
fastest known algorithm for a classical computer, which takes ous: the state a|0i + b|1i may drift, due to an error, to the state
∼ exp(m1/3 ) time. Since many encryption schemes depend → a|0i + beiθ |1i with arbitrary θ.
on the difficulty of finding the solution to problems similar to Remarkably, in spite of these difficulties, error correction is
finding the prime factors of a large number, there is an obvi- possible for quantum computers (Calderbank and Shor, 1996;
ous application of a quantum computer which is of great basic Gottesman, 1998; Preskill, 2004; Shor, 1995; Steane, 1996a).
and applied interest. One can represent information redundantly so that errors can
The computation model set forth by these pioneers of quan- be identified without measuring the information. For instance,
tum computing (and refined in DiVincenzo, 2000), is based if we use three spins to represent each qubit, |0i → |000i,
on three steps: initialization, unitary evolution and measure- |1i → |111i, and the spin-flip rate is low, then we can iden-
ment. We assume that we have a system at our disposal with tify errors by checking whether all three spins are the same
Hilbert space H. We further assume that we can initialize (here, we represent an up spin by 0 and a down spin by 1).
the system in some known state |ψ0 i. We unitarily evolve Suppose that our spins are in in the state α|000i + β|111i. If
the system until it is in some final state U (t)|ψ0 i. This evo- the first spin has flipped erroneously, then our spins are in the
lution will occur according to some Hamiltonian H(t) such state α|100i + β|011i. We can detect this error by checking
that dU/dt = iH(t) U (t)/~. We require that we have enough whether the first spin is the same as the other two; this does
control over this Hamiltonian so that U (t) can be made to be not require us to measure the state of the qubit. (“We measure
any unitary transformation that we desire. Finally, we need to the errors, rather than the information.” (Preskill, 2004)) If
measure the state of the system at the end of this evolution. the first spin is different from the other two, then we just need
Such a process is called quantum computation (Nielsen and to flip it. We repeat this process with the second and third
Chuang, 2000). The Hamiltonian H(t) is the software pro- spins. So long as we can be sure that two spins have not erro-
gram to be run. The initial state is the input to the calculation, neously flipped (i.e. so long as the basic spin-flip rate is low),
and the final measurement is the output. this procedure will correct spin-flip errors. A more elaborate
The need for versatility, i.e., for one computer to effi- encoding is necessary in order to correct phase errors, but the
ciently solve many different problems, requires the construc- key observation is that a phase error in the σz basis is a bit flip
tion of the computer out of smaller pieces that can be manipu- error in the σx basis.
lated and reconfigured individually. Typically the fundamen- However, the error correction process may itself be a little
tal piece is taken to be a quantum two state system known as noisy. More errors could then occur during error correction,
a “qubit” which is the quantum analog of a bit. (Of course, and the whole procedure will fail unless the basic error rate is
one could equally well take general “dits”, for which the fun- very small. Estimates of the threshold error rate above which
damental unit is some d-state system with d not too large). error correction is impossible depend on the particular error
While a classical bit, i.e., a classical two-state system, can be correction scheme, but fall in the range 10−4 − 10−6 (see,
either “zero” or “one” at any given time, a qubit can be in one e.g. Aharonov and Ben-Or, 1997; Knill et al., 1998). This
of the infinitely many superpositions a|0i+b|1i. For n qubits, means that we must be able to perform 104 − 106 operations
the state becomes a vector in a 2n –dimensional Hilbert space, perfectly before an error occurs. This is an extremely strin-
in which the different qubits are generally entangled with one gent constraint and it is presently unclear if local qubit-based
another. quantum computation can ever be made fault-tolerant through
The quantum phenomenon of superposition allows a sys- quantum error correction protocols.
9
Random errors are caused by the interaction between the the problems faced in the two categories might be quite differ-
quantum computer and the environment. As a result of this ent. For quantum information processing, unitary errors, such
interaction, the quantum computer, which is initially in a pure as rotating a qubit by 90.01 degrees instead of 90, are an issue
superposition state, becomes entangled with its environment. of how precisely one can manipulate the system. On the other
This can cause errors as follows. Suppose that the quantum hand, when a qubit is simply storing information, one is likely
computer is in the state |0i and the environment is in the to be more concerned about errors caused by interactions with
state |E0 i so that their combined state is |0i|E0 i. The in- the environment. This is instead an issue of how well isolated
teraction between the computer and the environment could one can make the system. As we will see below, a topologi-
cause this state to evolve to α|0i|E0 i + β|1i|E1 i, where |E1 i cal quantum computer is protected from problems in both of
is another state of the environment (not necessarily orthog- these categories.
onal to |E0 i). The computer undergoes a transition to the
state |1i with probability |β|2 . Furthermore, the computer
and the environment are now entangled, so the reduced den- 2. Fault-Tolerance from Non-Abelian Anyons
sity matrix for the computer alone describes a mixed state,
e.g. ρ = diag(|α|2 , |β|2 ) if hE0 |E1 i = 0. Since we can-
Topological quantum computation is a scheme for using a
not measure the state of the environment accurately, informa-
system whose excitations satisfy non-Abelian braiding statis-
tion is lost, as reflected in the evolution of the density matrix
tics to perform quantum computation in a way that is natu-
of the computer from a pure state to a mixed one. In other
rally immune to errors. The Hilbert space H used for quantum
words, the environment has caused decoherence. Decoher-
computation is the subspace of the total Hilbert space of the
ence can destroy quantum information even if the state of the
system comprised of the degenerate ground states with a fixed
computer does not undergo a transition. Although whether
number of quasiparticles at fixed positions. Operations within
or not a transition occurs is basis-dependent (a bit flip in the
this subspace are carried out by braiding quasiparticles. As
σz basis is a phase flip in the σx basis), it is a useful dis-
we discussed above, the subspace of degenerate ground states
tinction because many systems have a preferred basis, for in-
is separated from the rest of the spectrum by an energy gap.
stance the ground state |0i and excited state |1i of an ion in a
Hence, if the temperature is much lower than the gap and the
trap. Suppose the state |0i evolves as above, but with α = 1,
system is weakly perturbed using frequencies much smaller
β = 0 so that no transition occurs, while the state |1i|E0 i
than the gap, the system evolves only within the ground state
evolves to |1i|E1′ i with hE1′ |E1 i = 0. Then an initial pure
subspace. Furthermore, that evolution is severely constrained,
state (a|0i + b|1i) |E0 i evolves to a mixed state with density
since it is essentially the case (with exceptions which we will
matrix ρ = diag(|a|2 , |b|2 ). The correlations in which our
discuss) that the only way the system can undergo a non-
quantum information resides is now transferred to correlation
trivial unitary evolution - that is, an evolution that takes it
between the quantum computer and the environment. The
from one ground state to another - is by having its quasipar-
quantum state of a system invariably loses coherence in this
ticles braided. The reason for this exceptional stability is that
way over a characteristic time scale Tcoh . It was universally
any local perturbation (such as the electron-phonon interac-
assumed until the advent of quantum error correction (Shor,
tion and the hyperfine electron-nuclear interaction, two ma-
1995; Steane, 1996a) that quantum computation is intrinsi-
jor causes for decoherence in non-topological solid state spin-
cally impossible since decoherence-induced quantum errors
based quantum computers (Witzel and Das Sarma, 2006)) has
simply cannot be corrected in any real physical system. How-
no nontrivial matrix elements within the ground state sub-
ever, when error-correcting codes are used, the entanglement
space. Thus, the system is rather immune from decoherence
is transferred from the quantum computer to ancillary qubits
(Kitaev, 2003). Unitary errors are also unlikely since the uni-
so that the quantum information remains pure while the en-
tary transformations associated with braiding quasiparticles
tropy is in the ancillary qubits.
are sensitive only to the topology of the quasiparticle trajecto-
Of course, even if the coupling to the environment were ries, and not to their geometry or dynamics.
completely eliminated, so that there were no random errors, A model in which non-Abelian quasiparticles are utilized
there could still be systematic errors. These are unitary errors for quantum computation starts with the construction of
which occur while we process quantum information. For in- qubits. In sharp contrast to most realizations of a quantum
stance, we may wish to rotate a qubit by 90 degrees but might computer, a qubit here is a non-local entity, being comprised
inadvertently rotate it by 90.01 degrees. of several well-separated quasiparticles, with the two states
From a practical standpoint, it is often useful to divide er- of the qubit being two different values for the internal quan-
rors into two categories: (i) errors that occur when a qubit is tum numbers of this set of quasiparticles. In the simplest non-
being processed (i.e., when computations are being performed Abelian quantum Hall state, which has Landau-level filling
on that qubit) and (ii) errors that occur when a qubit is simply factor ν = 5/2, two quasiparticles can be put together to
storing quantum information and is not being processed (i.e., form a qubit (see Sections II.C.4 and IV.A). Unfortunately,
when it is acting as a quantum memory). From a fundamen- as we will discuss below in Sections IV.A and IV.C, this sys-
tal standpoint, this is a bit of a false dichotomy, since one can tem turns out to be incapable of universal topological quan-
think of quantum information storage (or quantum memory) tum computation using only braiding operations; some un-
as being a computer that applies the identity operation over protected operations are necessary in order to perform univer-
and over to the qubit (i.e., leaves it unchanged). Nonetheless, sal quantum computation. The simplest system that is capa-
10
ble of universal topological quantum computation is discussed parent problem, however, is evaded by the nature of the com-
in Section IV.B, and utilizes three quasiparticles to form one putations, which correspond to closed world lines that have
qubit. no loose ends: when the computation involves creation and
As mentioned above, to perform a quantum computation, annihilation of a quasiparticle quasi-hole pair, the world-line
one must be able to initialize the state of qubits at the begin- is a closed curve in space-time. If the measurement occurs
ning, perform arbitrary controlled unitary operations on the by bringing two particles together to measure their quantum
state, and then measure the state of qubits at the end. We now charge, it does not matter where precisely they are brought
address each of these in turn. together. Alternatively, when the measurement involves an
Initialization may be performed by preparing the quasipar- interference experiment, the interfering particle must close a
ticles in a specific way. For example, if a quasiparticle-anti- loop. In other words, a computation corresponds to a set of
quasiparticle pair is created by “pulling” it apart from the vac- links rather than open braids, and the initialization and mea-
uum (e.g. pair creation from the vacuum by an electric field), surement techniques necessarily involve bringing quasiparti-
the pair will begin in an initial state with the pair necessarily cles together in some way, closing up the trajectories and mak-
having conjugate quantum numbers (i.e., the “total” quantum ing the full process from initialization to measurement com-
number of the pair remains the same as that of the vacuum). pletely topological.
This gives us a known initial state to start with. It is also Due to its special characteristics, then, topological quan-
possible to use measurement and unitary evolution (both to tum computation intrinsically guarantees fault-tolerance, at
be discussed below) as an initialization scheme — if one can the level of “hardware”, without “software”-based error cor-
measure the quantum numbers of some quasiparticles, one can rection schemes that are so essential for non-topological quan-
then perform a controlled unitary operation to put them into tum computers. This immunity to errors results from the sta-
any desired initial state. bility of the ground state subspace with respect to external lo-
Once the system is initialized, controlled unitary opera- cal perturbations. In non-topological quantum computers, the
tions are then performed by physically dragging quasiparti- qubits are local, and the operations on them are local, lead-
cles around one another in some specified way. When quasi- ing to a sensitivity to errors induced by local perturbations.
particles belonging to different qubits braid, the state of the In a topological quantum computer the qubits are non-local,
qubits changes. Since the resulting unitary evolution depends and the operations — quasiparticle braiding — are non-local,
only on the topology of the braid that is formed and not on leading to an immunity to local perturbations.
the details of how it is done, it is insensitive to wiggles in the Such immunity to local perturbation gives topolgical quan-
path, resulting, e.g., from the quasiparticles being scattered by tum memories exceptional protection from errors due to the
phonons or photons. Determining which braid corresponds to interaction with the environment. However, it is crucial to
which computation is a complicated but eminently solvable note that topological quantum computers are also exception-
task, which will be discussed in more depth in section IV.B.3. ally immune to unitary errors due to imprecise gate operation.
Once the unitary evolution is completed, there are two ways Unlike other types of quantum computers, the operations that
to measure the state of the qubits. The first relies on the fact can be performed on a topological quantum computer (braids)
that the degeneracy of multi-quasiparticle states is split when naturally take a discrete set of values. As discussed above,
quasiparticles are brought close together (within some micro- when one makes a 90 degree rotation of a spin-based qubit, for
scopic length scale). When two quasiparticles are brought example, it is possible that one will mistakenly rotate by 90.01
close together, for instance, a measurement of this energy (or degrees thus introducing a small error. In contrast, braids are
a measurement of the force between two quasiparticles) mea- discrete: either a particle is taken around another, or it is not.
sures the the topological charge of the pair. A second way to There is no way to make a small error by having slight im-
measure the topological charge of a group of quasiparticles is precision in the way the quasiparticles are moved. (Taking
by carrying out an Aharanov-Bohm type interference experi- a particle only part of the way around another particle rather
ment. We take a “beam” of test quasiparticles, send it through than all of the way does not introduce errors so long as the
a beamsplitter, send one partial wave to the right of the group topological class of the link formed by the particle trajectories
to be measured and another partial wave to the left of the – as described above – is unchanged.)
group and then re-interfere the two waves (see Figure 2 and Given the exceptional stability of the ground states, and
the surrounding discussion). Since the two different beams their insensitivity to local perturbations that do not involve
make different braids around the test group, they will experi- excitations to excited states, one may ask then which physical
ence different unitary evolution depending on the topological processes do cause errors in such a topological quantum com-
quantum numbers of the test group. Thus, the re-interference puter. Due to the topological stability of the unitary transfor-
of these two beams will reflect the topological quantum num- mations associated with braids, the only error processes that
ber of the group of quasiparticles enclosed. we must be concerned about are processes that might cause
This concludes a rough description of the way a topologi- us to form the wrong link, and hence the wrong computa-
cal quantum computation is to be performed. While the uni- tion. Certainly, one must keep careful track of the positions of
tary transformation associated with a braid depends only on all of the quasiparticles in the system during the computation
the topology of the braid, one may be concerned that errors and assure that one makes the correct braid to do the correct
could occur if one does not return the quasiparticles to pre- computation. This includes not just the “intended” quasipar-
cisely the correct position at the end of the braiding. This ap- ticles which we need to manipulate for our quantum compu-
11
tation, but also any “unintended” quasiparticle which might system where non-Abelian anyons can be found, manipulated
be lurking in our system without our knowledge. Two pos- (e.g. braided), and conveniently read out. Several theoret-
sible sources of these unintended quasiparticles are thermally ical models and proposals for systems having these proper-
excited quasiparticle-quasihole pairs, and randomly localized ties have been introduced in recent years (Fendley and Frad-
quasiparticles trapped by disorder (e.g. impurities, surface kin, 2005; Freedman et al., 2005a; Kitaev, 2006; Levin and
roughness, etc.). In a typical thermal fluctuation, for example, Wen, 2005b), and in section II.D below we will mention some
a quasiparticle-quasihole pair is thermally created from the of these possibilities briefly. Despite the theoretical work in
vacuum, braids with existing intended quasiparticles, and then these directions, the only real physical system where there is
gets annihilated. Typically, such a pair has opposite electrical even indirect experimental evidence that non-Abelian anyons
charges, so its constituents will be attracted back to each other exist are quantum Hall systems in two-dimensional (2D) elec-
and annihilate. However, entropy or temperature may lead the tron gases (2DEGs) in high magnetic fields. Consequently,
quasiparticle and quasihole to split fully apart and wander rel- we will devote a considerable part of our discussion to pu-
atively freely through part of the system before coming back tative non-Abelian quantum Hall systems which are also of
together and annihilating. This type of process may change great interest in their own right.
the state of the qubits encoded in the intended quasiparticles,
and hence disrupt the computation. Fortunately, as we will
see in Section IV.B below there is a whole class of such pro- 1. Rapid Review of Quantum Hall Physics
cesses that do not in fact cause error. This class includes all
of the most likely such thermal processes to occur: includ- A comprehensive review of the quantum Hall effect is well
ing when a pair is created, encircles a single already existing beyond the scope of this article and can be found in the lit-
quasiparticle and then re-annihilates, or when a pair is created erature (Das Sarma and Pinczuk, 1997; Prange and Girvin,
and one of the pair annihilates an already existing quasipar- 1990). This effect, realized for two dimensional electronic
ticle. For errors to be caused, the excited pair must braid at systems in a strong magnetic field, is characterized by a gap
least two intended quasiparticles. Nonetheless, the possibil- between the ground state and the excited states (incompress-
ity of thermally-excited quasiparticles wandering through the ibility); a vanishing longitudinal resistivity ρxx = 0, which
system creating unintended braids and thereby causing error implies a dissipationless flow of current; and the quantization
is a serious one. For this reason, topological quantum com- of the Hall resistivity precisely to values of ρxy = ν1 eh2 , with
putation must be performed at temperatures well below the ν being an integer (the integer quantum Hall effect), or a frac-
energy gap for quasiparticle-quasihole creation so that these tion (the fractional quantum Hall effect). These values of the
errors will be exponentially suppressed. two resistivities imply a vanishing longitudinal conductivity
2
Similarly, localized quasiparticles that are induced by dis- σxx = 0 and a quantized Hall conductivity σxy = ν eh .
order (e.g. randomly-distributed impurities, surface rough- To understand the quantized Hall effect, we begin by ignor-
ness, etc.) are another serious obstacle to overcome, since ing electron-electron Coulomb interactions, then the energy
they enlarge the dimension of the subspace of degenerate eigenstates of the single-electron Hamiltonian in a magnetic
1
2
ground states in a way that is hard to control. In particular, field, H0 = 2m pi − ec A(xi ) break up into an equally-
these unaccounted-for quasiparticles may couple by tunneling spaced set of degenerate levels called Landau levels. In sym-
to their intended counterparts, thereby introducing dynamics metric gauge, A(x) = 12 B × x, a basis of single particle
to what is supposed to be a topology-controlled system, and wavefunctions in the lowest Landau level (LLL) is given by
possibly ruining the quantum computation. We further note ϕm (z) = z m exp(−|z|2 /(4ℓ0 2 )), where z = x + iy. If the
that, in quantum Hall systems (as we will discuss in the next electrons are confined to a disk of area A pierced by magnetic
section), slight deviations in density or magentic field will flux B · A, then there are NΦ = BA/Φ0 = BAe/hc states
also create unintented quasiparticles that must be carefully in the lowest Landau level (and in each higher Landau level),
avoided. where B is the magnetic field; h, c, and e are, respectively,
Finally, we also note that while non-Abelian quasiparticles Planck’s constant, the speed of light, and the electron charge;
are natural candidates for the realization of topological qubits, and Φ0 = hc/e is the flux quantum. In the absence of dis-
not every system where quasiparticles satisfy non-Abelian order, these single-particle states are all precisely degenerate.
statistics is suitable for quantum computation. For this suit- When the chemical potential lies between the ν th and (ν + 1)th
ability it is essential that the set of unitary transformations in- Landau levels, the Hall conductance takes the quantized value
2
duced by braiding quasiparticles is rich enough to allow for all σxy = ν eh while σxx = 0. The two-dimensional electron
operations needed for computation. The necessary and suffi- density, n, is related to ν via the formula n = νeB/(hc). In
cient conditions for universal topological quantum computa- the presence of a periodic potential and/or disorder (e.g. im-
tion are discussed in Section IV.C. purities), the Landau levels broaden into bands. However, ex-
cept at the center of a band, all states are localized when disor-
der is present (see Das Sarma and Pinczuk, 1997; Prange and
C. Non-Abelian Quantum Hall States Girvin, 1990 and refs. therein). When the chemical potential
lies in the region of localized states between the centers of the
A necessary condition for topological quantum computa- ν th and (ν + 1)th Landau bands, the Hall conductance again
2
tion using non-Abelian anyons is the existence of a physical takes the quantized value σxy = ν eh while σxx = 0. The
12
density will be near but not necessarily equal to νeB/(hc). with ν not of the form ν = 1/k, Haldane and Halperin (Hal-
This is known as the Integer quantum Hall effect (since ν is dane, 1983; Halperin, 1984; Prange and Girvin, 1990) used a
an integer). hierarchical construction in which quasiparticles of a princi-
The neglect of Coulomb interactions is justified when an ple ν = 1/k state can then themselves condense into a quan-
integer number of Landau levels is filled, so long as the en- tized state. In this way, quantized Hall states can be con-
ergy splitting between Landau levels, ~ωc = ~eB mc is much structed for any odd-denominator fraction ν – but only for
e2 odd-denominator fractions. These states all have quasiparti-
larger than the scale of the Coulomb energy, ℓ0 , where ℓ0 =
p cles with fractional charge and Abelian fractional statistics.
hc/eB is the magnetic length. When the electron density
Later, it was noticed by Jain (Heinonen, 1998; Jain, 1989)
is such that a Landau level is only partially filled, Coulomb
that the most prominent fractional quantum Hall states are of
interactions may be important.
the form ν = p/(2p + 1), which can be explained by not-
In the absence of disorder, a partially-filled Landau level
ing that a system of electrons in a high magnetic field can be
has a very highly degenerate set of multi-particle states. This
approximated by a system of auxiliary fermions, called ‘com-
degeneracy is broken by electron-electron interactions. For
posite fermions’ , in a lower magnetic field. If the the elec-
instance, when the number of electrons is N = NΦ /3, i.e.
trons are at ν = p/(2p + 1), then the lower magnetic field
ν = 1/3, the ground state is non-degenerate and there is a
seen by the ‘composite fermions’ is such that they fill an in-
gap to all excitations. When the electrons interact through
teger number of Landau levels ν ′ = p. (See Halperin et al.,
Coulomb repulsion, the Laughlin state
Y 1993; López and Fradkin, 1991 for a field-theoretic imple-
3 2 2
(zi − zj ) e− i |zi | /4ℓ0
P
Ψ= (11) mentations.) Since the latter state has a gap, one can hope
i>j that the approximation is valid. The composite fermion pic-
ture of fractional quantum Hall states has proven to be qual-
is an approximation to the ground state (and is the exact
itatively and semi-quantitatively correct in the LLL (Murthy
ground state for a repulsive ultra-short-ranged model interac- and Shankar, 2003).
tion, see for instance the article by Haldane in Prange and
Girvin, 1990). Such ground states survive even in the pres- Systems with filling fraction ν > 1, can be mapped to
ence of disorder if it is sufficiently weak compared to the gap ν ′ ≤ 1 by keeping the fractional part of ν and using an ap-
to excited states. More delicate states with smaller excita- propriately modified Coulomb interaction to account for the
tion gaps are, therefore, only seen in extremely clean devices, difference between cyclotron orbits in the LLL and those in
as described in subsection II.C.5. However, some disorder is higher Landau levels (Prange and Girvin, 1990). This in-
necessary to pin the charged quasiparticle excitations which volves the assumption that the inter-Landau level coupling is
are created if the density or magnetic field are slightly varied. negligibly small. We note that this may not be a particularly
When these excitations are localized, they do not contribute to good assumption for higher Landau levels, where the compos-
the Hall conductance and a plateau is observed. ite fermion picture less successful.
Quasiparticle excitations above fractional quantum Hall Our confidence in the picture described above for the ν =
ground states, such as the ν = 1/3 Laughlin state (11), are 1/k Laughlin states and the hierarchy of odd-denominator
emergent anyons in the sense described in section II.A.2. An states which descend from them derives largely from nu-
explicit calculation of the Berry phase, along the lines of Eq. merical studies. Experimentally, most of what is known
8 shows that quasiparticle excitations above the ν = 1/k about quantum Hall states comes from transport experiments
Laughlin states have charge e/k and statistical angle θ = π/k — measurements of the conductance (or resistance) tensor.
(Arovas et al., 1984). The charge is obtained from the non- While such measurements make it reasonably clear when a
topological part of the Berry phase which is proportional to quantum Hall plateau exists at a given filling fraction, the na-
the flux enclosed by a particle’s trajectory times the quasipar- ture of the plateau (i.e., the details of the low-energy theory)
ticle charge. This is in agreement with a general argument is extremely hard to discern. Because of this difficulty, nu-
that such quasiparticles must have fractional charge (Laugh- merical studies of small systems (exact diagonalizations and
lin, 1983). The result for the statistics of the quasiparticles Monte Carlo) have played a very prominent role in provid-
follows from the topological part of the Berry phase; it is in ing further insight. Indeed, even Laughlin’s original work
agreement with strong theoretical arguments which suggest (Laughlin, 1983) on the ν = 1/3 state relied heavily on ac-
that fractionally charged excitations are necessarily Abelian companying numerical work. The approach taken was the fol-
anyons (see Wilczek, 1990 and refs. therein). Definitive ex- lowing. One assumed that the splitting between Landau levels
perimental evidence for the existence of fractionally charged is the largest energy in the problem. The Hamiltonian is pro-
excitations at ν = 1/3 has been accumulating in the last few jected into the lowest Landau level, where, for a finite num-
years (De Picciotto et al., 1997; Goldman and Su, 1995; Sam- ber of electrons and a fixed magnetic flux, the Hilbert space
inadayar et al., 1997). The observation of fractional statistics is finite-dimensional. Typically, the system is given periodic
is much more subtle. First steps in that direction have been boundary conditions (i.e. is on a torus) or else is placed on
recently reported (Camino et al., 2005) but are still debated a sphere; occasionally, one works on the disk, e.g. to study
(Godfrey et al., 2007; Rosenow and Halperin, 2007). edge excitations. The Hamiltonian is then a finite-sized ma-
The Laughlin states, with ν = 1/k, are the best understood trix which can be diagonalized by a computer so long as the
fractional quantum Hall states, both theoretically and exper- number of electrons is not too large. Originally, Laughlin ex-
imentally. To explain more complicated observed fractions, amined only 3 electrons, but modern computers can handle
13
sometimes as many as 18 electrons. The resulting ground state 2. Possible Non-Abelian States
wavefunction can be compared to a proposed trial wavefunc-
tion. Throughout the history of the field, this approach has The observation of a quantum Hall state with an even de-
proven to be extremely powerful in identifying the nature of nominator filling fraction (Willett et al., 1987), the ν = 5/2
experimentally-observed quantum Hall states when the sys- state, was the first indication that not all fractional quantum
tem in question is deep within a quantum Hall phase, so that Hall states fit the above hierarchy (or equivalently compos-
the associated correlation length is short and the basic physics ite fermion) picture. Independently, it was recognized Fu-
is already apparent in small systems. bini, 1991; Fubini and Lutken, 1991; Moore and Read, 1991
that conformal field theory gives a way to write a variety of
There are several serious challenges in using such numeri- trial wavefunctions for quantum Hall states, as we describe
cal work to interpret experiments. First of all, there is always in Section III.D below. Using this approach, the so-called
the challenge of extrapolating finite-size results to the thermo- Moore-Read Pfaffian wavefunction was constructed (Moore
dynamic limit. Secondly, simple overlaps between a proposed and Read, 1991):
trial state and an exact ground state may not be sufficiently
informative. For example, it is possible that an exact ground 1 Q m − i |zi |2 /4ℓ0 2
P
ΨPf = Pf i<j (zi − zj ) e (12)
state will be adiabatically connected to a particular trial state, zi − zj
i.e., the two wavefunctions represent the same phase of mat-
ter, but the overlaps may not be very high. For this reason, it is The Pfaffian is the square root of the determinant of an anti-
necessary to also examine quantum numbers and symmetries symmetric matrix or, equivalently, the antisymmetrized sum
of the ground state, as well as the response of the ground state over pairs:
to various perturbations, particularly the response to changes
1
1 1
in boundary conditions and in the flux. Pf =A ... (13)
zj − zk z1 − z2 z3 − z4
Another difficulty is the choice of Hamiltonian to diago- For m even, this is an even-denominator quantum Hall state
nalize. One may think that the Hamiltonian for a quantum in the lowest Landau level. Moore and Read, 1991 suggested
Hall system is just that of 2D electrons in a magnetic field that its quasiparticle excitations would exhibit non-Abelian
interacting via Coulomb forces. However, the small but fi- statistics (Moore and Read, 1991). This wavefunction is the
nite width (perpendicular to the plane of the system) of the exact ground state of a 3-body repulsive interaction; as we
quantum well slightly alters the effective interaction between discuss below, it is also an approximate ground state for more
electrons. Similarly, screening (from any nearby conductors, realistic interactions. This wavefunction is a representative of
or from inter-Landau-level virtual excitations), in-plane mag- a universality class which has remarkable properties which we
netic fields, and even various types of disorder may alter the discuss in detail in this paper. In particular, the quasiparticle
Hamiltonian in subtle ways. To make matters worse, one may excitations above this state realize the second scenario dis-
not even know all the physical parameters (dimensions, dop- cussed in Eqs. 9, 10 in section II.A.2. There are 2n−1 states
ing levels, detailed chemical composition, etc.) of any par- with 2n quasiholes at fixed positions, thereby establishing the
ticular experimental system very accurately. Finally, Landau- degeneracy of multi-quasiparticle states which is required for
level mixing is not small because the energy splitting between non-Abelian statistics (Nayak and Wilczek, 1996). Further-
Landau levels is not much larger than the other energies in the more, these quasihole wavefunctions can also be related to
problem. Thus, it is not even clear that it is correct to truncate conformal field theory (as we discuss in section III.D), from
the Hilbert space to the finite-dimensional Hilbert space of a which it can be deduced that the 2n−1 -dimensional vector
single Landau level. space of states can be understood as the spinor representation
of SO(2n); braiding particles i and j has the action of a π/2
In the case of very robust states, such as the ν = 1/3 state, rotation in the i − j plane in R2n (Nayak and Wilczek, 1996).
these subtle effects are unimportant; the ground state is es- In short, these quasiparticles are essentially Ising anyons (with
sentially the same irrespective of these small deviations from the difference being an additional Abelian component to their
the idealized Hamiltonian. However, in the case of weaker statistics). Although these properties were uncovered using
states, such as those observed between ν = 2 and ν = 4 specific wavefunctions which are eigenstates of the 3-body
(some of which we will discuss below), it appears that very interaction for which the Pfaffian wavefunction is the exact
small changes in the Hamiltonian can indeed greatly affect the ground state, they are representative of an entire universality
resulting ground state. Therefore, a very valuable approach class. The effective field theory for this universality class is
has been to guess a likely Hamiltonian, and search a space SU (2) Chern-Simons theory at level k = 2 together with an
of “nearby” Hamiltonians, slightly varying the parameters of additional Abelian Chern-Simons term (Fradkin et al., 2001,
the Hamiltonian, to map out the phase diagram of the sys- 1998). Chern-Simons theory is the archetypal topological
tem. These phase diagrams suggest the exciting technologi- quantum field theory (TQFT), and we discuss it extensively
cal possibility that detailed numerics will allow us to engineer in section III. As we describe, Chern-Simons theory is re-
samples with just the right small perturbations so as display lated to the Jones polynomial of knot theory (Witten, 1989);
certain quantum Hall states more clearly (Manfra et al., 2007; consequently, the current through an interferometer in such a
Peterson and Das Sarma, 2007). non-Abelian quantum Hall state would give a direct measure
14
of the Jones polynomial for the link produced by the quasipar- of a large class of non-Abelian topological states. We will
ticle trajectories (Fradkin et al., 1998)! see in section III.C that this circle of ideas enables us to use
One interesting feature of the Pfaffian wavefunction is that the theory of knots to understand experiments on non-Abelian
it is the quantum Hall analog of a p + ip superconductor: the anyons.
antisymmetrized product over pairs is the real-space form of In the paragraphs below, we will discuss numerical results
the BCS wavefunction (Greiter et al., 1992). Read and Green, for ν = 5/2, 12/5, and other candidates in greater detail.
2000 showed that the same topological properties mentioned (a) 5/2 State: The ν = 5/2 fractional quantum Hall state
above are realized by a p + ip-wave superconductor, thereby is a useful case history for how numerics can elucidate ex-
cementing the identification between such a paired state and periments. This incompressible state is easily destroyed by
the Moore-Read state. Ivanov, 2001 computed the braiding the application of an in-plane magnetic field (Eisenstein et al.,
matrices by this approach (see also Stern et al., 2004; Stone 1990). At first it was assumed that this implied that the 5/2
and Chung, 2006). Consequently, we will often be able to dis- state is spin-unpolarized or partially polarized since the in-
cuss p + ip-wave superconductors and superfluids in parallel plane magnetic field presumably couples only to the electron
with the ν = 5/2 quantum Hall state, although the experimen- spin. Careful finite-size numerical work changed this percep-
tal probes are significantly different. tion, leading to our current belief that the 5/2 FQH state is
As we discuss below, all of these theoretical developments actually in the universality class of the spin-polarized Moore-
garnered greater interest when numerical work (Morf, 1998; Read Pfaffian state.
Rezayi and Haldane, 2000) showed that the ground state of In rather pivotal work (Morf, 1998), it was shown that spin-
systems of up to 18 electrons in the N = 1 Landau level at polarized states at ν = 5/2 have lower energy than spin-
filling fraction 1/2 is in the universality class of the Moore- unpolarized states. Furthermore, it was shown that varying
Read state. These results revived the conjecture that the lowest the Hamiltonian slightly caused a phase transition between
Landau level (N = 0) of both spins is filled and inert and the a gapped phase that has high overlap with the Moore-Read
electrons in the N = 1 Landau level form the analog of the wavefunction and a compressible phase. The proposal put
Pfaffian state (Greiter et al., 1992). Consequently, it is the forth was that the most important effect of the in-plane field
leading candidate for the experimentally-observed ν = 5/2 was not on the electron spins, but rather was to slightly alter
state. the shape of the electron wavefunction perpendicular to the
Read and Rezayi, 1999 constructed a series of non-Abelian sample which, in turn, slightly alters the effective electron-
quantum Hall states at filling fraction ν = N + k/(M k + 2) electron interaction, pushing the system over a phase bound-
with M odd, which generalize the Moore-Read state in a way ary and destroying the gapped state. Further experimental
which we discuss in section III. These states are referred to as work showed that the effect of the in-plane magnetic field is
the Read-Rezayi Zk parafermion states for reasons discussed to drive the system across a phase transition from a gapped
in section III.D. Recently, a quantum Hall state was observed quantum Hall phase into an anisotropic compressible phase
experimentally with ν = 12/5 (Xia et al., 2004). It is sus- (Lilly et al., 1999a; Pan et al., 1999a). Further numerical work
pected (see below) that the ν = 12/5 state may be (the par- (Rezayi and Haldane, 2000) then mapped out a full phase di-
ticle hole conjugate of) the Z3 Read-Rezayi state, although it agram showing the transition between gapped and compress-
is also possible that 12/5 belongs to the conventional Abelian ible phases and showing further that the experimental systems
hierarchy as the 2/5 state does. Such an option is not possible lie exceedingly close to the phase boundary. The correspon-
at ν = 5/2 as a result of the even denominator. dence between numerics and experiment has been made more
In summary, it is well-established that if the observed quantitative by comparisons between the energy gap obtained
ν = 5/2 state is in the same universality class as the Moore- from numerics and the one measured in experiments (Morf
Read Pfaffian state, then its quasiparticle excitations are non- and d’Ambrumenil, 2003; Morf et al., 2002). Very recently,
Abelian anyons. Similarly, if the ν = 12/5 state is in the this case has been further strengthened by the application of
universality class of the Z3 Read-Rezayi state, its quasiparti- the density-matrix renormalization group method (DMRG) to
cles are non-Abelian anyons. There is no direct experimental this problem (Feiguin et al., 2007b).
evidence that the ν = 5/2 is in this particular universality One issue worth considering is possible competitors to the
class, but there is evidence from numerics, as we further dis- Moore-Read Pfaffian state. Experiments have already told us
cuss below. There is even less evidence in the case of the that there is a fractional quantum Hall state at ν = 5/2. There-
ν = 12/5 state. In subsections II.C.3 and II.C.4, we will fore, our job is to determine which of the possible states is re-
discuss proposed experiments which could directly verify the alized there. Serious alternatives to the Moore-Read Pfaffian
non-Abelian character of the ν = 5/2 state and will briefly state fall into two categories. On the one hand, there is the
mention their extension to the ν = 12/5 case. Both of these possibility that the ground state at ν = 5/2 is not fully spin-
states, as well as others (e.g. Ardonne and Schoutens, 1999; polarized. If it were completely unpolarized, the so-called
Simon et al., 2007c), were constructed on the basis of very (3, 3, 1) state (Das Sarma and Pinczuk, 1997; Halperin, 1983)
deep connections between conformal field theory, knot theory, would be a possibility. However, Morf’s numerics (Morf,
and low-dimensional topology (Witten, 1989). Using meth- 1998) and a recent variational Monte Carlo study (Dimov
ods from these different branches of theoretical physics and et al., 2007) indicate that an unpolarized state is higher in en-
mathematics, we will explain the structure of the non-Abelian ergy than a fully-polarized state. This can be understood as
statistics of the ν = 5/2 and 12/5 states within the context a consequence of a tendency towards spontaneous ferromag-
15
netism; however, a partially-polarized alternative (which may has an overlap of 97% for 14 electrons (Rezayi and Haldane,
be either Abelian or non-Abelian) to the Pfaffian is not ruled 2000).
out (Dimov et al., 2007). Secondly, even if the ground state To summarize, the overlap is somewhat smaller in the 5/2
at ν = 5/2 is fully spin-polarized, the Pfaffian is not the only case than in the 1/3 case when particle-hole symmetry is not
possibility. It was very recently noticed that the Pfaffian state accounted for, but only slightly smaller when it is. This is
is not symmetric under a particle-hole transformation of a sin- an indication that Landau-level mixing – which will favor ei-
gle Landau level (which, in this case, is the N = 1 Landau ther the Pfaffian or the anti-Pfaffian – is an important effect at
level, with the N = 0 Landau level filled and assumed inert), ν = 5/2, unlike at ν = 1/3. Moreover, Landau-level mix-
even though this is an exact symmetry of the Hamiltonian in ing is likely to be large because the 5/2 FQH state is typically
the limit that the energy splitting between Landau levels is realized at relatively low magnetic fields, making the Landau
infinity. Therefore, there is a distinct state, dubbed the anti- level separation energy relatively small.
Pfaffian (Lee et al., 2007b; Levin et al., 2007), which is an Given that potentially large effects have been neglected,
equally good state in this limit. Quasiparticles in this state it is not too surprising that the gap obtained by extrapo-
are also essentially Ising anyons, but they differ from Pfaffian lating numerical results for finite-size systems (Morf and
quasiparticles by Abelian statistical phases. In experiments, d’Ambrumenil, 2003; Morf et al., 2002) is substantially larger
Landau-level mixing is not small, so one or the other state is than the experimentally-measured activation gap. Also, the
lower in energy. On a finite torus, the symmetric combina- corresponding excitation gap obtained from numerics for the
tion of the Pfaffian and the anti-Pfaffian will be lower in en- ν = 1/3 state is much larger than the measured activation
ergy, but as the thermodynamic limit is approached, the anti- gap. The discrepancy between the theoretical excitation gap
symmetric combination will become equal in energy. This is a and the measured activation gap is a generic problem of all
possible factor which complicates the extrapolation of numer- FQH states, and may be related to poorly understood disorder
ics to the thermodynamic limit. On a finite sphere, particle- effects and Landau-level mixing.
hole symmetry is not exact; it relates a system with 2N − 3 Finally, it is important to mention that several very recent
flux quanta with a system with 2N + 1 flux quanta. Thus, the (2006-07) numerical works in the literature have raised some
anti-Pfaffian would not be apparent unless one looked at a dif- questions about the identification of the observed 5/2 FQH
ferent value of the flux. To summarize, the only known alter- state with the Moore-Read Pfaffian (Toke and Jain, 2006; Toke
natives to the Pfaffian state – partially-polarized states and the et al., 2007; Wojs and Quinn, 2006). Considering the ab-
anti-Pfaffian – have not really been tested by numerics, either sence of a viable alternative (apart from the anti-Pfaffian and
because the spin-polarization was assumed to be 0% or 100% partially-polarized states, which were not considered by these
(Morf, 1998) or because Landau-level mixing was neglected. authors) it seems unlikely that these doubts will continue to
With this caveat in mind, it is instructive to compare the evi- persist, as more thorough numerical work indicates (Moller
dence placing the ν = 5/2 FQH state in the Moore-Read Pfaf- and Simon, 2007; Peterson and Das Sarma, 2007; Rezayi,
fian universality class with the evidence placing the ν = 1/3 2007).
FQH state in the corresponding Laughlin universality class. (b) 12/5 State: While our current understanding of the 5/2
In the latter case, there have been several spectacular experi- state is relatively good, the situation for the experimentally
ments (De Picciotto et al., 1997; Goldman and Su, 1995; Sam- observed 12/5 state is more murky, although the possibilities
inadayar et al., 1997) which have observed quasiparticles with are even more exciting, at least from the perspective of topo-
electrical charge e/3, in agreement with the prediction of the logical quantum computation. One (relatively dull) possibility
Laughlin universality class. In the case of the ν = 5/2 FQH is that the 12/5 state is essentially the same as the observed
state, we do not yet have the corresponding measurements of ν = 2/5 state, which is Abelian. However, Read and Rezayi,
the quasiparticle charge, which should be e/4. However, the in their initial work on non-Abelian generalizations of the
observation of charge e/3, while consistent with the Laughlin Moore-Read state (Read and Rezayi, 1999) proposed that the
universality class, does not uniquely fix the observed state in 12/5 state might be (the particle-hole conjugate of) their Z3
this class (see, for example, Simon et al., 2007c; Wójs, 2001. parafermion (or SU (2) level 3) state. This is quite an exciting
Thus, much of our confidence derives from the amazing (99% possibility because, unlike the non-Abelian Moore-Read state
or better) overlap between the ground state obtained from ex- at 5/2, the Z3 parafermion state would have braiding statistics
act diagonalization for a finite size 2D system with up to 14 that allow universal topological quantum computation.
electrons and the Laughlin wavefunction. In the case of the The initial numerics by Read and Rezayi (Read and Rezayi,
ν = 5/2 FQH state, the corresponding overlap (for 18 elec- 1999) indicated that the 12/5 state is very close to a phase
trons on the sphere) between the ν = 5/2 ground state and transition between the Abelian hierarchy state and the non-
the Moore-Read Pfaffian state is reasonably impressive (∼ Abelian parafermion state. More recent work by the same
80%). This can be improved by modifying the wavefunction authors (Rezayi and Read, 2006) has mapped out a detailed
at short distances without leaving the Pfaffian phase (Moller phase diagram showing precisely for what range of parameters
and Simon, 2007). However, on the torus, as we mentioned a system should be in the non-Abelian phase. It was found that
above, the symmetric combination of the Pfaffian and the anti- the non-Abelian phase is not very “far” from the results that
Pfaffian is a better candidate wavefunction in a finite-size sys- would be expected from most real experimental systems. This
tem than the Pfaffian itself (or the anti-Pfaffian). Indeed, the again suggests that (if the system is not already in the non-
symmetric combination of the Pfaffian and the anti-Pfaffian Abelian phase), we may be able to engineer slight changes in
16
neutrality, a variation of the area of the cell varies the flux it of these particles gives the trivial particle since they were cre-
encloses and keeps its bulk Landau filling unaltered. In con- ated from the vacuum. Now consider what happens when a
trast, a variation of the magnetic field changes the filling frac- current-carrying quasiparticle tunnels at one of the two point
tion in the bulk, and consequently introduces quasiparticles in contacts. If it tunnels at the second one, it braids around the N
the bulk. Since the statistics of the quasiparticles is fractional, quasiparticles in the cell (but not the N + 1th , which is outside
they contribute to the phase φ. The back-scattering probabil- the cell). This changes the fusion channel of the N + 1 local-
ity is then determined not only by the two constrictions and ized quasiparticles. In the language introduced in subsection
the area of the cell they define, but also by the number of II.A.1, each e/4 quasiparticle is a σ particle. An odd number
localized quasiparticles that the cell encloses. By varying the N of them can only fuse to σ; fused now with the N + 1th ,
voltage applied to an anti-dot in the cell (the grey circle in Fig. they can either give 1 or ψ. Current-carrying quasiparticles,
2), we can independently vary the number of quasiparticles in when they braid with the N in the cell, toggle the system be-
the cell. Again, however, as the area of the cell is varied, the tween these two possibilities. Since the state of the localized
back-scattered current oscillates. quasiparticles has been changed, such a process cannot inter-
For non-Abelian quantum Hall states, the situation is more fere with a process in which the current-carrying quasiparticle
interesting (Bonderson et al., 2006a,b; Chung and Stone, tunnels at the first junction and does not encircle any of the lo-
2006; Das Sarma et al., 2005; Fradkin et al., 1998; Stern and calized quasiparticles. Therefore, the localized quasiparticles
Halperin, 2006). Consider the case of the Moore-Read Pfaf- ‘measure’ which trajectory the current-carrying quasiparticles
fian state. For clarity, we assume that there are localized e/4 take(Bonderson et al., 2007; Overbosch and Bais, 2001). If N
quasiparticles only within the cell (either at the anti-dot or is even, then we can create (N + 2)/2 pairs and take two of
elsewhere in the cell). If the current in Fig. (2) comes from the the resulting quasiparticles outside of the cell. If the N quasi-
left, the portion of the current that is back-reflected from the particles in the cell all fuse to the trivial particle, then this is
left constriction does not encircle any of these quasiparticles, not necessary, we can just create N/2 pairs. However, if they
and thus does not interact with them. The part of the current fuse to a neutral fermion ψ, then we will need a pair outside
that is back-scattered from the right constriction, on the other the cell which also fuses to ψ so that the total fuses to 1, as
hand, does encircle the cell, and therefore applies a unitary it must for pair creation from the vacuum. A current-carrying
transformation on the subspace of degenerate ground states. quasiparticle picks up a phase depending on whether the N
The final state of the ground state subspace that is coupled to quasiparticles in the cell fuse to 1 or ψ.
the left back–scattered wave, |ξ0 i, is then different from the The Fabry-Perot interferometer depicted in Fig. 2 allows
state coupled to the right partial wave, Û |ξ0 i. Here Û is the also for the interference of waves that are back-reflected sev-
unitary transformation that results from the encircling of the eral times. For an integer filling factor, in the limit of strong
cell by the wave scattered from the right constriction. The in- back-scattering at the constrictions, the sinusoidal dependence
terference term in the four-terminal longitudinal conductance, of the Hall bar’s conductance on the area of the cell gives
the final term in Eq. 14, is then multiplied by the matrix ele- way to a resonance-like dependence: the conductance is zero
ment hξ0 |Û|ξ0 i: unless a Coulomb peak develops. For the ν = 5/2 state,
n
o again, the parity of the number of localized quasiparticles mat-
GL ∝ |t1 |2 + |t1 |2 + 2Re t∗1 t2 eiφ ξ0 Û ξ0 (15) ters: when it is odd, the Coulomb blockade peaks are equally
spaced. When it is even, the spacing between the peaks alter-
In section III, we explain how hξ0 |Û |ξ0 i can be calculated by nate between two values (Stern and Halperin, 2006).
several different methods. Here we just give a brief descrip- The Moore-Read Pfaffian state, which is possibly realized
tion of the result. at ν = 5/2, is the simplest of the non-Abelian states. The
For the Moore-Read Pfaffian state, which is believed to be other states are more complex, but also richer. The geom-
realized at ν = 5/2, the expectation value hξ0 |Û |ξ0 i depends etry of the Fabry-Perot interferometer may be analyzed for
first and foremost on the parity of the number of e/4 quasi- these states as well. In general, for all non-Abelian states the
particles localized in the cell. When that number is odd, the conductance of the Hall bar depends on the internal state of
resulting expectation value is zero. When that number is even, the quasiparticles localized between the constrictions – i.e.
the expectation value is non-zero and may assume one of two the quasiparticle to which they fuse. However, only for the
possible values, that differ by a minus sign. As a consequence, Moore-Read Pfaffian state is the effect quite so dramatic. For
when the number of localized quasiparticles is odd, no inter- example, for the the Z3 parafermion state which may be re-
ference pattern is seen, and the back-scattered current does alized at ν = 12/5, when the number of localized quasipar-
not oscillate with small variations of the area of the cell. When ticles is larger than three, the fusion channel of the quasipar-
that number is even, the back-scattered current oscillates as a ticles determines whether the interference is fully visible or
function of the area of the cell. suppressed
√ by a factor of −ϕ−2 (with ϕ being the golden ra-
A way to understand this striking result is to observe that tio ( 5 + 1)/2) (Bonderson et al., 2006b; Chung and Stone,
the localized quasiparticles in the cell can be viewed as being 2006). The number of quasiparticles, on the other hand, af-
created in pairs from the vacuum. Let us suppose that we want fects only the phase of the interference pattern. Similar to
to have N quasiparticles in the cell. If N is odd, then we can the case of ν = 5/2 here too the position of Coulomb block-
create (N +1)/2 pairs and take one of the resulting quasiparti- ade peaks on the two parameter plane of area and magnetic
cles outside of the cell, where it is localized. Fusing all N + 1 field reflects the non-Abelian nature of the quasiparticles (Ilan
18
111111111111111111111111111111111111
000000000000000000000000000000000000 from one edge to the other, provided that its trajectory passes
000000000000000000000000000000000000
111111111111111111111111111111111111 in between the two localized quasiparticles. This is a sim-
000000000000000000000000000000000000
111111111111111111111111111111111111
000000000000000000000000000000000000
111111111111111111111111111111111111 ple example of how braiding causes non-trivial transforma-
000000000000000000000000000000000000
111111111111111111111111111111111111
000000000000000000000000000000000000
111111111111111111111111111111111111 tions of multi-quasiparticle states of non-Abelian quasiparti-
000000000000000000000000000000000000
111111111111111111111111111111111111
t1 1 tS 2 t2 cles, which we discuss in more detail in section III. If we
000000000000000000000000000000000000
111111111111111111111111111111111111
000000000000000000000000000000000000
111111111111111111111111111111111111 measure the four-terminal longitudinal conductance GL be-
000000000000000000000000000000000000
111111111111111111111111111111111111
000000000000000000000000000000000000
111111111111111111111111111111111111 fore and after applying this NOT gate, we will observe differ-
000000000000000000000000000000000000
111111111111111111111111111111111111 ent values according to (16).
000000000000000000000000000000000000
111111111111111111111111111111111111 For this operation to be a NOT gate, it is important that
just a single quasiparticle (or any odd number) tunnel from
FIG. 3 If a third constriction is added between the other two, the cell one edge to the other across the middle constriction. In or-
is broken into two halves. We suppose that there is one quasiparticle der to regulate the number of quasiparticles which pass across
(or any odd number) in each half. These two quasiparticles (labeled the constriction, it may be useful to have a small anti-dot in
1 and 2) form a qubit which can be read by measuring the conduc- the middle of the constriction with a large charging energy so
tance of the interferometer if there is no backscattering at the middle that only a single quasiparticle can pass through at a time. If
constriction. When a single quasiparticle tunnels from one edge to we do not have good control over how many quasiparticles
the other at the middle constriction, a σx or NOT gate is applied to
the qubit.
tunnel, then it will be essentially random whether an even or
odd number of quasiparticles tunnel across; half of the time,
a NOT gate will be applied and the backscattering probability
(hence the conductance) will change while the other half of
et al., 2007).
the time, the backscattering probability is unchanged. If the
constriction is pinched down to such an extreme that the 5/2
state is disrupted between the quasiparticles, then when it is
4. A Fractional Quantum Hall Quantum Computer
restored, there will be an equal probability for the qubit to be
in either state.
We now describe how the constricted Hall bar may be uti-
This qubit is topologically protected because its state can
lized as a quantum bit (Das Sarma et al., 2005). To that end,
only be affected by a charge e/4 quasiparticle braiding with it.
an even number of e/4 quasiparticles should be trapped in the
If a charge e/4 quasiparticle winds around one of the antidots,
cell between the constrictions, and a new, tunable, constric-
it effects a NOT gate on the qubit. The probability for such
tion should be added between the other two so that the cell is
an event can be very small because the density of thermally-
broken into two cells with an odd number of quasiparticles in
excited charge e/4 quasiparticles is exponentially suppressed
each (See Fig. (3)). One way to tune the number of quasipar-
at low temperatures, nqp ∼ e−∆/(2T ) . The simplest estimate
ticles in each half is to have two antidots in the Hall bar. By
of the error rate Γ (in units of the gap) is then of activated
tuning the voltage on the antidots, we can change the number
form:
of quasiholes on each. Let us assume that we thereby fix the
number of quasiparticles in each half of the cell to be odd. Γ/∆ ∼ (T /∆) e−∆/(2T ) (17)
For concreteness, let us take this odd number to be one (i.e.
let us assume that we are in the idealized situation in which The most favorable experimental situation (Xia et al., 2004)
there are no quasiparticles in the bulk, and one quasihole on considered in (Das Sarma et al., 2005) has ∆ ≈ 500 mK
each antidot). These two quasiholes then form a two-level and T ∼ 5 mK, producing an astronomically low error rate
system, i.e. a qubit. This two-level system can be understood ∼ 10−15 . This should be taken as an overly optimistic es-
in several ways, which we discuss in detail in section III. In timate. A more definitive answer is surely more compli-
brief, the two states correspond to whether the two σs fuse to cated since there are multiple gaps which can be relevant in
1 or ψ or, in the language of chiral p-wave superconductivity, a disordered system. Furthermore, at very low temperatures,
the presence or absence of a neutral (‘Majorana’) fermion; or, we would expect quasiparticle transport to be dominated by
equivalently, as the fusion of two quasiparticles carrying the variable-range hopping of localized quasiparticles rather than
spin-1/2 representation of an SU(2) gauge symmetry in the thermal activation. Indeed, the crossover to this behavior may
spin-0 or spin-1 channels. already be apparent (Pan et al., 1999b), in which case, the
The interference between the t1 and t2 processes depends error suppression will be considerably weaker at the lowest
on the state of the two-level system, so the qubit can be read by temperatures. Although the error rate, which is determined
a measurement of the four-terminal longitudinal conductance by the probability for a quasiparticle to wind around the anti-
dot, is not the same as the longitudinal resistance, which is
GL ∝ |t1 |2 + |t2 |2 ± 2Re t∗1 t2 eiφ (16) the probability for it to go from one edge of the system to the
other, the two are controlled by similar physical processes. A
where the ± comes from the dependence of hξ0 |Û |ξ0 i on the more sophisticated estimate would require a detailed analy-
state of the qubit, as we discuss in section III. sis of the quasiparticle transport properties which contribute
The purpose of the middle constriction is to allow us to ma- to the error rate. In addition, this error estimate assumes that
nipulate the qubit. The state may be flipped, i.e. a σx or NOT all of the trapped (unintended) quasiparticles are kept very far
gate can be applied, by the passage of a single quasiparticle from the quasiparticles which we use for our qubit so that they
19
cannot exchange topological quantum numbers with our qubit sured excitation gaps. In particular, an empirical relation,
via tunneling. We comment on the issues involved in more ∆ = ∆0 − Γ, where ∆ is the measured activation gap and
detailed error estimates in section IV.D. ∆0 the ideal excitation gap with Γ being the level broadening
The device envisioned above can be generalized to one with arising from impurity and disorder scattering, has often been
many anti-dots and, therefore, many qubits. More compli- discussed in the literature (see, e.g. Du et al., 1993). Writing
cated gates, such as a CNOT gate can be applied by braiding the mobility µ = eτ /m, with τ the zero field Drude scattering
quasiparticles. It is not clear how to braid quasiparticles lo- time, we can write (an approximation of) the level broadening
calized in the bulk – perhaps by transferring them from one as Γ = ~/(2τ ), indicating Γ ∼ µ−1 in this simple picture,
anti-dot to another in a kind of “bucket brigade”. This is an and therefore increasing the mobility should steadily enhance
important problem for any realization of topological quantum the operational excitation gap, as is found experimentally. It
computing. However, as we will discuss in section IV, even if has recently been pointed out (Morf et al., 2002) that by re-
this were solved, there would still be the problem that braid- ducing Γ, an FQH gap of 2-3 K may be achievable in the 5/2
ing alone is not sufficient for universal quantum computation FQH state. Much less is currently known about the 12/5 state,
in the ν = 5/2 state (assuming that it is the Moore-Read Pfaf- but recent numerics (Rezayi and Read, 2006) suggest that the
fian state). One must either use some unprotected operations maximal gap in typical samples will be quite a bit lower than
(just two, in fact) or else use the ν = 12/5, if it turns out to be for 5/2.
the Z3 parafermion non-Abelian state. It is also possible to consider designing samples that would
inherently have particularly large gaps. First of all, the in-
teraction energy (which sets the overall scale of the gap) is
5. Physical Systems and Materials Considerations roughly of the 1/r Coulomb form, so it scales as the inverse of
the interparticle spacing. Doubling the density should there-
As seen in the device described in the previous subsection, fore increase the gaps by roughly 40%. Although there are
topological protection in non-Abelian fractional quantum Hall efforts underway to increase the density of samples (Willett
states hinges on the energy gap (∆) separating the many-body et al., 2007), there are practical limitations to how high a den-
degenerate ground states from the low-lying excited states. sity one can obtain since, if one tries to over-fill a quantum
This excitation gap also leads to the incompressibility of the well with electrons, the electrons will no longer remain strictly
quantum Hall state and the quantization of the Hall resistance. two dimensional (i.e., they will start filling higher subbands,
Generally speaking, the larger the size of this excitation gap or they will not remain in the well at all). Secondly, as dis-
compared to the temperature, the more robust the topologi- cussed in section II.C.2 above, since the non-Abelian states
cal protection, since thermal excitation of stray quasiparticles, appear generally to be very sensitive to the precise parameters
which goes as exp(−∆/(2T )), would potentially lead to er- of the Hamiltonian, another possible route to increased exci-
rors. tation gap would be to design the precise form of the inter-
It must be emphasized that the relevant T here is the tem- electron interaction (which can be modified by well width,
perature of the electrons (or more precisely, the quasiparti- screening layers, and particularly spin-orbit coupling (Man-
cles) and not that of the GaAs-AlGaAs lattice surrounding the fra et al., 2007)) so that the Hamiltonian is at a point in the
2D electron layer. Although the surrounding bath temperature phase diagram with maximal gap. With all approaches for re-
could be lowered to 1 mK or below by using adiabatic de- designing samples, however, it is crucial to keep the disorder
magnetization in dilution refrigerators, the 2D electrons them- level low, which is an exceedingly difficult challenge.
selves thermally decouple from the bath at low temperatures Note that a large excitation gap (and correspondingly low
and it is very difficult to cool the 2D electrons below T ≈ temperature) suppresses thermally excited quasiparticles but
20 mK. It will be a great boost to hopes for topological quan- does not preclude stray localized quasiparticles which could
tum computation using non-Abelian fractional quantum Hall be present even at T = 0. As long as their positions are known
states if the electron temperature can be lowered to 1 mK or and fixed, and as long as they are few enough in number to
even below, and serious efforts are currently underway in sev- be sufficiently well separated, these quasiparticles would not
eral laboratories with this goal. present a problem, as one could avoid moving other quasipar-
Unfortunately, the excitation gaps for the expected non- ticles near their positions and one could then tailor algorithms
Abelian fractional quantum Hall states are typically very to account for their presence. If the density of stray local-
small (compared, for example, with the ν = 1/3 fractional ized quasiparticles is sufficiently high, however, this would
quantum Hall state). The early measured gap for the 5/2 state no longer be possible. Fortunately, these stray particles can
was around ∆ ∼ 25 mK (in 1987) (Willett et al., 1987), but be minimized in the same way as one of the above discussed
steady improvement in materials quality, as measured by the solutions to keeping the energy gap large – improve the mo-
sample mobility, has considerably enhanced this gap. In the bility of the 2D electron sample on which the measurements
highest mobility samples currently (2007) available, ∆ ≈ 600 (i.e. the computation operations) are being carried out. Im-
mK (Choi et al., 2007). Indeed, there appears to be a close provement in the mobility leads to both the enhancement of
connection between the excitation gap ∆ and the mobility the excitation gap and the suppression of unwanted quasipar-
(or the sample quality). Although the details of this connec- ticle localization by disorder.
tion are not well-understood, it is empirically well-established We should emphasize, however, how extremely high qual-
that enhancing the 2D mobility invariably leads to larger mea- ity the current samples already are. Current “good” sample
20
mobilities are in the range of 10 − 30 × 106 cm2 /(Volt-sec). emergent symmetry at low energy, which is not present in the
To give the reader an idea of how impressive this is, we note microscopic Hamiltonian of the system. Consequently, given
that under such conditions, at low temperatures, the mean free a Hamiltonian, it is very difficult to determine if its ground
path for an electron may be a macroscopic length of a tenth of state is in a topological phase. It is certainly no easier than
a millimeter or more. (Compare this to, say, copper at room showing that any other low-energy emergent phenomenon oc-
temperature, which has a mean free path of tens of nanometers curs in a particular model. Except for rare exactly solvable
or less). models (e.g. Kitaev, 2006, Levin and Wen, 2005b which we
Nonetheless, further MBE technique and design improve- describe in section III.G), topological ground states are in-
ment may be needed to push low-temperature 2D electron ferred on the basis of approximations and inspired guesswork.
mobilities to 100 × 106 cm2 /(Volt-sec) or above for topo- On the other hand, if topological states exist at all, they will
logical quantum computation to be feasible. At lower temper- be robust (i.e. their topological nature should be fairly insen-
atures, T < 100 mK, the phonon scattering is very strongly sitive to local perturbations, e.g. electron-phonon interaction
suppressed (Kawamura and Das Sarma, 1992; Stormer et al., or charge fluctuations between traps). For this reason, we be-
1990), and therefore, there is essentially no intrinsic limit to lieve that if it can be shown that some model Hamiltonian has
how high the 2D electron mobility can be since the extrin- a topological ground state, then a real material which is de-
sic scattering associated with impurities and disorder can, in scribed approximately by that model is likely to have a topo-
principle, be eliminated through materials improvement. In logical ground state as well.
fact, steady materials improvement in modulation-doped 2D One theoretical model which is known to have a non-
GaAs-AlGaAs heterostructures grown by the MBE technique Abelian topological ground state is a p + ip wave supercon-
has enhanced the 2D electron mobility from 104 cm2 /(Volt- ductor (i.e., a superconductor where the order parameter is of
sec) in the early 1980’s to 30 × 106 cm2 /(Volt-sec) in 2004, a px + ipy symmetry). As we describe in section III.B, vortices
three orders of magnitude improvement in materials quality in in a superconductor of p + ip pairing symmetry exhibit non-
roughly twenty years. Indeed, the vitality of the entire field of Abelian braiding statistics. This is really just a reincarnation
quantum Hall physics is a result of these amazing advances. of the physics of the Pfaffian state (believed to be realized at
Another factor of 2-3 improvement in the mobility seems pos- the ν = 5/2 quantum Hall plateau) in zero magnetic field.
sible (L. Pfeiffer, private communication), and will likely be Chiral p-wave superconductivity/superfluidity is currently the
needed for the successful experimental observation of non- most transparent route to non-Abelian anyons. As we dis-
Abelian anyonic statistics and topological quantum computa- cuss below, there are multiple physical systems which may
tion. host such a reincarnation. The Kitaev honeycomb model (see
also section III.G and ) (Kitaev, 2006) is a seemingly differ-
ent model which gives rise to the same physics. In it, spins
D. Other Proposed Non-Abelian Systems
interact anisotropically in such a way that their Hilbert space
can be mapped onto that of a system of Majorana fermions.
In various parameter ranges, the ground state is in either an
This review devotes a great deal of attention to the non-
Abelian topological phase, or a non-Abelian one in the same
Abelian anyonic properties of certain fractional quantum Hall
universality class as a p + ip superconductor.
states (e.g. ν = 5/2, 12/5, etc. states) in two-dimensional
Chiral p-wave superconductors, like quantum Hall states,
semiconductor structures, mainly because theoretical and ex-
break parity and time-reversal symmetries, although they do
perimental studies of such (possibly) non-Abelian fractional
so spontaneously, rather than as a result of a large magnetic
quantized Hall states is a mature subject, dating back to 1986,
field. However, it is also possible to have a topological phase
with many concrete results and ideas, including a recent pro-
which does not break these symmetries. Soluble theoretical
posal (Das Sarma et al., 2005) for the construction of qubits
models of spins on a lattice have been constructed which have
and a NOT gate for topological quantum computation (de-
P, T -invariant topological ground states. A very simple model
scribed above in subsection II.C.4 and, in greater detail in
of this type with an Abelian topological ground state, called
section IV). But there are several other systems which are po-
the ‘toric code’, was proposed in Kitaev, 2003. Even though it
tential candidates for topological quantum computation, and
is not sufficient for topological quantum computation because
we briefly discuss these systems in this subsection. Indeed,
it is Abelian, it is instructive to consider this model because
the earliest proposals for fault-tolerant quantum computation
non-Abelian models can be viewed as more complex versions
with anyons were based on spin systems, not the quantum Hall
of this model. It describes s = 1/2 spins on a lattice interact-
effect (Kitaev, 2003).
ing through the following Hamiltonian (Kitaev, 2003):
First, we emphasize that the most crucial necessary condi-
X X
tion for carrying out topological quantum computation is the H = −J1 Ai − J2 Fp (18)
existence of appropriate ‘topological matter’, i.e. a physical i p
system in a topological phase. Such a phase of matter has
suitable ground state properties and quasiparticle excitations This model can be defined on an arbitrary lattice. The
manifesting non-Abelian statistics. Unfortunately, the neces- spins are
Q assumed to be on the links of the lattice.
α
sary and sufficient conditions for the existence of topological Ai ≡ α∈N (i) σz , where N (i) is the Q
set of spins on links
α
ground states are not known even in theoretical models. We α which touch the vertex i, and Fp ≡ α∈p σx , where p is
note that the topological symmetry of the ground state is an a plaquette and α ∈ p are the spins on the links comprising
21
the plaquette. This model is exactly soluble because the Ai s metry phases. In the quantum dimer model (Moessner and
and Fp s all commute with each other. For any J1 , J2 > 0, the Sondhi, 2001; Rokhsar and Kivelson, 1988), for instance, an
ground state |0i is given by Ai |0i = Fp |0i = |0i for all i, p. Abelian topological phase must compete with various crys-
Quasiparticle excitations are sites i at which Ai |0i = −|0i talline phases which occupy most of the phase diagram. This
or plaquettes p at which Fp |0i = −|0i. A pair of excited is presumably one obstacle to finding topological phases in
can be created at i and i′ by acting on the ground state
sites Q more realistic models, i.e. models which would give an ap-
with α∈C σxα , where the product is over the links in a chain proximate description of some concrete physical system.
C on the lattice connecting i and i′ . Similarly, a pair of ex- There are several physical systems – apart from fractional
cited plaquettes Q can be created by acting on the ground state quantum Hall states – which might be promising hunting
with connected α∈C̃ σzα where the product is over the links grounds for topological phases, including transition metal ox-
crossed by a chain C˜ on the dual lattice connecting the centers ides and ultra-cold atoms in optical traps. The transition metal
of plaquettes p and p′ . Both types of excitations are bosons, oxides have the advantage that we already know that they give
but when an excited site is taken around an excited plaquette, rise to striking collective phenomena such as high-Tc super-
the wavefunction acquires a minus sign. Thus, these two types conductivity, colossal magnetoresistance, stripes, and thermo-
of bosons are relative semions. electricity. Unfortunately, their physics is very difficult to un-
The toric code model is not very realistic, but it is closely ravel both theoretically and experimentally for this very rea-
related to some more realistic models such as the quantum son: there are often many different competing phenomena in
dimer model (Chayes et al., 1989; Klein, 1982; Moessner and these materials. This is reflected in the models which describe
Sondhi, 2001; Nayak and Shtengel, 2001; Rokhsar and Kivel- transition metal oxides. They tend to have many closely com-
son, 1988). The degrees of freedom in this model are dimers peting phases, so that different approximate treatments find
on the links of a lattice, which represent a spin singlet bond rather different phase diagrams. There is a second advantage
between the two spins on either end of a link. The quantum to the transition metal oxides, namely that many sophisticated
dimer model was proposed as an effective model for frus- experimental techniques have been developed to study them,
trated antiferromagnets, in which the spins do not order, but including transport, thermodynamic measurements, photoe-
instead form singlet bonds which resonate among the links mission, neutron scattering, X-ray scattering, and NMR. Un-
of the lattice – the resonating valence bond (RVB) state (An- fortunately, however, these methods are tailored for detecting
derson, 1973, 1987; Baskaran et al., 1987; Kivelson et al., broken-symmetry states or for giving a detailed understanding
1987) which, in modern language, we would describe as a of metallic behavior, not for uncovering a topological phase.
specific realization of a simple Abelian topological state (Ba- Nevertheless, this is such a rich family of materials that it
lents et al., 1999, 2000; Moessner and Sondhi, 2001; Senthil would be surprising if there weren’t a topological phase hid-
and Fisher, 2000, 2001a). While the quantum dimer model on ing there. (Whether we find it is another matter.) There is one
the square lattice does not have a topological phase for any particular material in this family, Sr2 RuO4 , for which there
range of parameter values (the RVB state is only the ground is striking evidence that it is a chiral p-wave superconductor
state at a critical point), the model on a triangular lattice does at low temperatures, Tc ≈ 1.5 K (Kidwingira et al., 2006;
have a topological phase (Moessner and Sondhi, 2001). Xia et al., 2006). Half-quantum vortices in a thin film of such
Levin and Wen, 2005a,b constructed a model which is, in a superconductor would exhibit non-Abelian braiding statis-
a sense, a non-Abelian generalization of Kitaev’s toric code tics (since Sr2 RuO4 is not spin-polarized, one must use half
model. It is an exactly soluble model of spins on the links quantum vortices, not ordinary vortices). However, half quan-
(two on each link) of the honeycomb lattice with three-spin in- tum vortices are usually not the lowest energy vortices in a
teractions at each vertex and twelve-spin interactions around chiral p-wave superconductor, and a direct experimental ob-
each plaquette, which we describe in section III.G. This servation of the half vortices themselves would be a substan-
model realizes a non-Abelian phase which supports Fibonacci tial milestone on the way to topological quantum computation
anyons, which permits universal topological quantum compu- (Das Sarma et al., 2006b).
tation (and generalizes straightforwardly to other non-Abelian The current status of research is as follows. Three-
topological phases). Other models have been constructed dimensional single-crystals and thin films of Sr2 RuO4 have
(Fendley and Fradkin, 2005; Freedman et al., 2005a) which been fabricated and studied. The nature of the super-
are not exactly soluble but have only two-body interactions conductivity of these samples has been studied by many ex-
and can be argued to support topological phases in some pa- perimental probes, with the goal of identifying the symme-
rameter regime. However, there is still a considerable gulf be- try of the Cooper-pair. There are many indications that sup-
tween models which are soluble or quasi-soluble and models port the identification of the Sr2 RuO4 as a px + ipy super-
which might be considered realistic for some material. conductor. First, experiments that probe the spins of the
Models such as the Kitaev and Levin-Wen models are deep Cooper pair strongly indicate triplet pairing (Mackenzie and
within topological phases; there are no other competing states Maeno, 2003). Such experiments probe the spin susceptibil-
nearby in their phase diagram. However, simple models such ity through measurements of the NMR Knight shift and of
as the Heisenberg model or extensions of the Hubbard model neutron scattering. For singlet spin pairing the susceptibility
are not of this form. The implication is that such mod- vanishes at zero temperature, since the spins keep a zero po-
els are not deep within a topological phase, and topological larization state in order to form Cooper pairs. In contrast, the
phases must compete with other phases, such as broken sym- susceptibility remains finite for triplet pairing, and this is in-
22
deed the observed behavior. Second, several experiments that and should also result in quantum hall physics(Mueller, 2004;
probe time reversal symmetry have indicated that it is broken, Popp et al., 2004; Sørensen et al., 2005).
as expected from a p± ip super-conductor. These experiments The second route to generating topological phases in cold
include muon spin relaxation (Mackenzie and Maeno, 2003) atoms is the idea of using a gas of ultra-cold fermions with
and the polar Kerr effect(Xia et al., 2006). In contrast, mag- a p-wave Feschbach resonance, which could form a spin-
netic imaging experiments designed to probe the edge cur- polarized chiral p-wave superfluid (Gurarie et al., 2005). Pre-
rents that are associated with a super-conductor that breaks liminary studies of such p-wave systems have been made ex-
time reversal symmetry did not find the expected signal (Kirt- perimentally (Gaebler et al., 2007) and unfortunately, it ap-
ley et al., 2007). The absence of this signal may be attributed pears that the decay time of the Feshbach bound states may
to the existence of domains of p + ip interleaved with those be so short that thermalization is impossible. Indeed, recent
of p − ip. Altogether, then, Sr2 RuO4 is likely to be a three di- theoretical work (Levinsen et al., 2007) suggests that this may
mensional p+ip super-conductor, that may open the way for a be a generic problem and additional tricks may be necessary
realization of a two-dimensional super-conductor that breaks if a p-wave superfluid is to be produced in this way.
time reversal symmetry. We note that both the ν = 1 rotating boson system and the
The other very promising direction to look for topological chiral p-wave superfluid would be quite closely related to the
phases, ultra-cold atoms in traps, also has several advantages. putative non-Abelian quantum Hall state at ν = 5/2 (as is
The Hamiltonian can often be tuned by, for instance, tuning Sr2 RuO4 ). However, there is an important difference between
the lasers which define an optical lattice or by tuning through a p-wave superfluid of cold fermions and the ν = 5/2 state.
a Feshbach resonance. For instance, there is a specific scheme Two-dimensional superconductors, as well as superfluids in
for realizing the Hubbard model (Jaksch and Zoller, 2005) any dimension, have a gapless Goldstone mode. Therefore,
in this way. At present there are relatively few experimen- there is the danger that the motion of vortices may cause the
tal probes of these systems, as compared with transition metal excitation of low-energy modes. Superfluids of cold atoms
oxides or even semiconductor devices. However, to look on may, however, be good test grounds for the detection of lo-
the bright side, some of the available probes give information calized Majorana modes associated with localized vortices, as
that cannot be measured in electronic systems. Furthermore, those are expected to have a clear signature in the absorption
new probes for cold atoms systems are being developed at a spectrum of RF radiation (Tewari et al., 2007b), in the form of
remarkable rate. a discrete absorption peak whose density and weight are de-
There are two different schemes for generating topologi- termined by the density of the vortices (Grosfeld et al., 2007).
cal phases in ultra-cold atomic gases that seem particularly One can also realize, using suitable laser configurations, Ki-
promising at the current time. The first is the approach of taev’s honeycomb lattice model (Eq. 55) with cold atoms on
using fast rotating dilute bose gases (Wilkin et al., 1998) to an optical lattice (Duan et al., 2003). It has recently been
make quantum Hall systems of bosons (Cooper et al., 2001). shown how to braid anyons in such a model (Zhang et al.,
Here, the rotation plays the role of an effective magnetic field, 2006).
and the filling fraction is given by the ratio of the number A major difficulty in finding a topological phase in either
of bosons to the number of vortices caused by rotation. Ex- a transition metal oxide or an ultra-cold atomic system is that
perimental techniques (Abo-Shaeer et al., 2001; Bretin et al., topological phases are hard to detect directly. If the phase
2004; Schweikhard et al., 2004) have been developed that can breaks parity and time-reversal symmetries, either sponta-
give very large rotation rates and filling fractions can be gener- neously or as a result of an external magnetic field, then there
ated which are as low as ν = 500 (Schweikhard et al., 2004). is usually an experimental handle through transport, as in the
While this is sufficiently low that all of the bosons are in a sin- fractional quantum Hall states or chiral p-wave superconduc-
gle landau level (since there is no Pauli exclusion, nu¿ 1 can tors. If the state does not break parity and time-reversal, how-
still be a lowest Landau level state), it is still predicted to be ever, there is no ‘smoking gun’ experiment, short of creating
several orders of magnitude too high to see interesting topo- quasiparticles, braiding them, and measuring the outcome.
logical states. Theoretically, the interesting topological states Any detailed discussion of the physics of these ‘alterna-
occur for ν < 10 (Cooper et al., 2001). In particular, evidence tive’ topological systems is well beyond the scope of the cur-
is very strong that ν = 1, should it be achieved, would be rent review. We refer the readers to the existing recent liter-
the bosonic analogue of the Moore-Read state, and (slightly ature on these systems for details. In section III (especially
less strong) ν = 3/2 and ν = 2 would be the Read-Rezayi III.G), however, we discuss some of the soluble models which
states, if the inter-boson interaction is appropriately adjusted support topological phases because many of their mathemat-
(Cooper and Rezayi, 2007; Rezayi et al., 2005). In order to ac- ical features elucidate the underlying structure of topological
cess this regime, either rotation rates will need to be increased phases.
substantially, or densities will have to be decreased substan-
tially. While the latter sounds easier, it then results in all of
the interaction scales being correspondingly lower, and hence III. TOPOLOGICAL PHASES OF MATTER AND
implies that temperature would have to be lower also, which NON-ABELIAN ANYONS
again becomes a challenge. Several other works have pro-
posed using atomic lattice systems where manipulation of pa- Topological quantum computation is predicated on the exis-
rameters of the Hamiltonian induces effective magnetic fields tence in nature of topological phases of matter. In this section,
23
we will discuss the physics of topological phases from several A. Topological Phases of Matter
different perspectives, using a variety of theoretical tools. The
reader who is interested primarily in the application of topo- In Section II of this paper, we have used ‘topological phase’
logical phases to quantum computation can skim this section as essentially being synonymous with any system whose
briefly and still understand section IV. However, a reader with quasiparticle excitations are anyons. However, a precise def-
a background in condensed matter physics and quantum field inition is the following. A system is in a topological phase
theory may find it enlightening to read a more detailed ac- if, at low temperatures and energies, and long wavelengths,
count of the theory of topological phases and the emergence all observable properties (e.g. correlation functions) are in-
of anyons from such phases, with explicit derivations of some variant under smooth deformations (diffeomorphisms) of the
of the results mentioned in section II and used in section IV. spacetime manifold in which the system lives. Equivalently,
These readers may find topological phases interesting in and all observable properties are independent of the choice of
of themselves, apart from possible applications. spacetime coordinates, which need not be inertial or rectilin-
ear. (This is the ‘passive’ sense of a diffeomorphism, while
the first statement uses the active sense of a transformation.)
By “at low temperatures and energies, and long wavelengths,”
we mean that diffeomorphism invariance is only violated by
terms which vanish as ∼ max e−∆/T , e−|x|/ξ for some non-
zero energy gap ∆ and finite correlation length ξ. Thus, topo-
Topological phases, the states of matter which support
logical phases have, in general, an energy gap separating the
anyons, occur in many-particle physical systems. Therefore,
ground state(s) from the lowest excited states. Note that an
we will be using field theory techniques to study these states.
excitation gap, while necessary, is not sufficient to ensure that
A canonical, but by no means unique, example of a field the-
a system is in a topological phase.
ory for a topological phase is Chern-Simons theory. We will
frequently use this theory to illustrate the general points which The invariance of all correlation functions under dif-
we wish to make about topological phases. In section V, we feomorphisms means that the only local operator which
will make a few comments about the problem of classifying has non-vanishing correlation functions is the identity.
topological phases, and how this example, Chern-Simons the- For instance, under an arbitrary change of space-time
ory, fits in the general classification. In subsection III.A, we coordinates x → x′ = f (x), the correlations of a scalar oper-
give a more precise definition of a topological phase and con- ator φ(x) must satisfy h0i |φ(x1 )φ(x2 ) . . . φ(xn )|0j i =
nect this definition with the existence of anyons. We also in- h0i |φ(x′1 )φ(x′2 ) . . . φ(x′n )|0j i, which implies that
troduce Chern-Simons theory, which we will discuss through- h0i |φ(x1 )φ(x2 ) . . . φ(xn )|0j i = 0 unless φ(x) ≡ c for
out section III as an example of the general structure which some constant c. Here, |0i i, |0j i are ground states of the
we discuss in subsection III.A. In subsection III.B, we will system (which may or may not be different). This property
discuss a topological phase which is superficially rather dif- is important because any local perturbation, such as the
ferent but, in fact, will prove to be a special case of Chern- environment, couples to a local operator. Hence, these local
Simons theory. This phase can be analyzed in detail using the perturbations are proportional to the identity. Consequently,
formalism of BCS theory. In subsection III.C, we further ana- they cannot have non-trivial matrix elements between differ-
lyze Chern-Simons theory, giving a more detailed account of ent ground states. The only way in which they can affect the
its topological properties, especially the braiding of anyons. system is by exciting the system to high-energies, at which
We describe Witten’s work (Witten, 1989) connecting Chern- diffeomorphism invariance is violated. At low-temperatures,
Simons theory with the knot and link invariants of Jones and the probability for this is exponentially suppressed.
Kauffman (Jones, 1985; Kauffman, 1987). We show how the The preceding definition of a topological phase may be
latter can be used to derive the properties of anyons in these stated more compactly by simply saying that a system is in
topological phases. In section III.D, we describe a comple- a topological phase if its low-energy effective field theory is
mentary approach by which Chern-Simons theory can be un- a topological quantum field theory (TQFT), i.e. a field the-
derstood: through its connection to conformal field theory. We ory whose correlation functions are invariant under diffeomor-
explain how this approach can be particularly fruitful in con- phisms. Remarkably, topological invariance does not imply
nection with fractional quantum Hall states. In III.E, we dis- trivial low-energy physics.
cuss the gapless excitations which must be present at the edge
of any chiral topological phase. Their physics is intimately
connected with the topological properties of the bulk and, at 1. Chern-Simons Theory
the same time, is directly probed by transport experiments
in quantum Hall devices. In III.F, we apply the knowledge Consider the simplest example of a TQFT, Abelian Chern-
which we have gained about the properties of Chern-Simons Simons theory, which is relevant to the Laughlin states at fill-
theory to the interferometry experiments which we discussed ing fractions of the form ν = 1/k, with k an odd integer. Al-
in II.C.3. Finally, in III.G we discuss a related but different though there are many ways to understand the Laughlin states,
class of topological phases which can arise in lattice models it is useful for us to take the viewpoint of a low-energy ef-
and may be relevant to transition metal oxides or ‘artificial’ fective theory. Since quantum Hall systems are gapped, we
solids such as ultra-cold atoms in optical lattices. should be able to describe the system by a field theory with
24
very few degrees of freedom. To this end, we consider the −1/2 times the Aharonov-Bohm phase. This is cancelled by a
action factor of two coming from the fact that each particle sees the
Z other’s flux.) Thus the contribution to a path integral eiSCS
k
SCS = d2 r dt ǫµνρ aµ ∂ν aρ (19) just gives an Aharonov-Bohm phase associated with moving
4π a charge around the Chern-Simons flux attached to the other
where k is an integer and ǫ is the antisymmetric tensor. Here, charges. The phases generated in this way give the quasipar-
a is a U(1) gauge field and indices µ, ν, ρ take the values 0 (for ticles of this Chern-Simons theory θ = π/k Abelian braiding
time-direction), 1,2 (space-directions). This action represents statistics.3
the low-energy degrees of freedom of the system, which are Therefore, an Abelian Chern-Simons term implements
purely topological. Abelian anyonic statistics. In fact, it does nothing else. An
The Chern-Simons gauge field a in (19) is an emergent de- Abelian gauge field in 2 + 1 dimensions has only one trans-
gree of freedom which encodes the low-energy physics of a verse component; the other two components can be eliminated
quantum Hall system. Although in this particular case, it is by fixing the gauge. This degree of freedom is fixed by the
simply-related to the electronic charge density, we will also be Chern-Simons constraint (22). Therefore, a Chern-Simons
considering systems in which emergent Chern-Simons gauge gauge field has no local degrees of freedom and no dynam-
fields cannot be related in a simple way to the underlying elec- ics.
tronic degrees of freedom. We now turn to non-Abelian Chern-Simons theory. This
In the presence of an external electromagnetic field and TQFT describes non-Abelian anyons. It is analogous to the
quasiparticles, the action takes the form: Abelian Chern-Simons described above, but different meth-
ods are needed for its solution, as we describe in this section.
Z
2 1 µνρ qp The action can be written on an arbitrary manifold M in the
S = SCS − d r dt ǫ Aµ ∂ν aρ + jµ aµ (20) form
2π
Z
where jµqp is the quasiparticle current, j0qp = ρqp is the quasi- k 2
SCS [a] = tr a ∧ da + a ∧ a ∧ a (24)
particle density, jqp = (j1qp , j2qp ) is the quasiparticle spatial 4π M
3
Z
current, and Aµ is the external electromagnetic field. We k 2
will assume that the quasiparticles are not dynamical, but in- = ǫµνρ aµa ∂ν aaρ + fa b c aµa aνb acρ
4π M 3
stead move along some fixed classically-prescribed trajecto-
ries which determine jµqp . The electrical current is: In this expression, the gauge field now takes values in the Lie
algebra of the group G. fa b c are the structure constants of
1 µνρ the Lie algebra which are simply ǫa b c for the case of SU(2).
jµ = ∂L/∂Aµ = ǫ ∂ν aρ (21) a
2π For the case of SU(2), we thus have a gauge field aµ , where
Since the action is quadratic, it is completely solvable, and the underlined indices run from 1 to 3. A matter field trans-
one can integrate out the field aµ to obtain the response of forming in the spin-j representation of the SU(2) gauge group
a
the current to the external electromagnetic field. The result of will couple to the combination aµ xa , where xa are the three
such a calculation is precisely the quantized Hall conductivity generator matrices of su(2) in the spin-j representation. For
σxx = 0 and σxy = k1 e2 /h. gauge group G and coupling constant k (called the ‘level’),
The equation of motion obtained by varying a0 is the we will denote such a theory by Gk . In this paper, we will be
Chern-Simons constraint: primarily concerned with SU(2)k Chern-Simons theory.
To see that Chern-Simons theory is a TQFT, first note that
k 1 the Chern-Simons action (24) is invariant under all diffeomor-
∇ × a = j0qp + B (22)
2π 2π phisms of M to itself, f : M → M. The differential form
notation in (24) makes this manifest, but it can be checked
According to this equation, each quasiparticle has Chern- in coordinate form for xµ → f µ (x). Diffeomorphism in-
Simons flux 2π/k attached to it (the magnetic field is assumed variance stems from the absence of the metric tensor in the
fixed). Consequently, it has electrical charge 1/k, accord- Chern-Simons action. Written out in component form, as in
ing to (21). As a result of the Chern-Simons flux, another (24), indices are, instead, contracted with ǫµνλ .
quasiparticle moving in this Chern-Simons field picks up an
Before analyzing the physics of this action (24), we will
Aharonov-Bohm phase. The action associated with taking
make two observations. First, as a result of the presence of
one quasiparticle around another is, according to Eq. 20, of
the form
Z Z
1
k dr dt j · a = kQ dr · a (23) 3 The Chern-Simons effective action for a hierarchical state is equivalent to
2 C
the action for the composite fermion state at the same filling fraction (Blok
where Q is the charge of the quasiparticle and the final integral and Wen, 1990; Read, 1990; Wen and Zee, 1992). It is a simple gener-
alization of Eq. 19 which contains several internal gauge fields an µ (with
is just the Chern-Simons flux enclosed in the path. (The factor n = 1, 2, ...), corresponding (in essence) to the action for the different
of 1/2 on the left-hand side is due to the action of the Chern- species of particles (either the different levels of the hierarchy, or the dif-
Simons term itself which, according to the constraint (22) is ferent composite fermion Landau levels).
25
ǫµνλ , the action changes sign under parity or time-reversal and a Yang-Mills term, then the Hamiltonian would not van-
transformations. In this paper, we will concentrate, for the ish, and the theory would have both ground states and excited
most part, on topological phases which are chiral, i.e. which states with a finite gap. Since the Yang-Mills term is sublead-
break parity and time-reversal symmetries. These are the ing compared to the Chern-Simons term (i.e. irrelevant in a
phases which can appear in the fractional quantum Hall ef- renormalization group (RG) sense), we can forget about it at
fect, where the large magnetic field breaks P , T . However, energies smaller than the gap and consider the Chern-Simons
we shall also discuss non-chiral topological phases in sec- term alone.
tion III.G, especially in connection which topological phases Therefore, when Chern-Simons theory is viewed as an ef-
emerging from lattice models. fective field theory, it can only be valid at energies much
Secondly, the Chern-Simons action is not quite fully invari- smaller than the energy gap. As a result, it is unclear, at the
ant under gauge transformations aµ → gaµ g −1 + g∂µ g −1 , moment, whether Chern-Simons theory has anything to say
where g : M → G is any function on the manifold taking about the properties of quasiparticles – which are excitations
values in the group G. On a closed manifold, it is only in- above the gap – or, indeed, whether those properties are part
variant under “small” gauge tranformations. Suppose that the of the universal low-energy physics of the system (i.e. are
manifold M is the 3-sphere, S 3 . Then, gauge transformations controlled by the infrared RG fixed point). Nevertheless, as
are maps S 3 → G, which can be classified topologically ac- we will see momentarily, it does and they are.
cording to it homotopy π3 (G). For any simple compact group Although the Hamiltonian vanishes, the theory is still not
G, π3 (G) = Z, so gauge transformations can be classified trivial because one must solve the constraint which follows by
according to their “winding number”. Under a gauge trans- varying a0 . For the sake of concreteness, we will specialize to
formation with winding m, the case G =SU(2). Then the constraint reads:
a b c
SCS [a] → SCS [a] + 2πkm (25) ǫij ∂i aj + f a b c a1 a2 = 0 (27)
(Deser et al., 1982). While the action is invariant under where i, j = 1, 2. The left-hand side of this equation is the
a
“small” gauge transformations, which are continuously con- field strength of the gauge field ai , where a = 1, 2, 3 is an
nected to the identity and have m = 0, it is not invariant su(2) index. Since the field strength must vanish, we can al-
a
under “large” gauge transformations (m 6= 0). However, it ways perform a gauge transformation so that ai = 0 locally.
is sufficient for exp(iS) to be gauge invariant, which will be Therefore this theory has no local degrees of freedom. How-
the case so long as we require that the level k be an integer. ever, for some field configurations satisfying the constraint,
The requirement that the level k be an integer is an example there may be a global topological obstruction which prevents
of the highly rigid structure of TQFTs. A small perturbation us from making the gauge field zero everywhere. Clearly, this
of the microscopic Hamiltonian cannot continuously change can only happen if Σ is topologically non-trivial.
the value of k in the effective low energy theory; only a per- The simplest non-trivial manifold is the annulus, which
turbation which is large enough to change k by an integer can is topologically equivalent to the sphere with two punctures.
do this. Following Elitzur et al., 1989 (see also (Wen and Zee, 1998)
The failure of gauge invariance under large gauge tranfor- for a similar construction on the torus), let us take coordinates
mations is also reflected in the properties of Chern-Simons (r, φ) on the annulus, with r1 < r < r2 , and let t be time.
theory on a surface with boundary, where the Chern-Simons Then we can write aµ = g∂µ g −1 , where
action is gauge invariant only up to a surface term. Conse- φ
quently, there must be gapless degrees of freedom at the edge g(r, φ, t) = eiω(r,φ,t) ei k λ(t) (28)
of the system whose dynamics is dictated by the requirement where ω(r, φ, t) and λ(t) take values in the Lie algebra su(2)
of gauge invariance of the combined bulk and edge (Wen, and ω(r, φ, t) is a single-valued function of φ. The functions ω
1992), as we discuss in section III.E. and λ are the dynamical variables of Chern-Simons theory on
To unravel the physics of Chern-Simons theory, it is use- the annulus. Substituting (28) into the Chern-Simons action,
ful to specialize to the case in which the spacetime manifold we see that it now takes the form:
M can be decomposed into a product of a spatial surface and Z
time, M = Σ × R. On such a manifold, Chern-Simons theory 1
S= dt tr (λ∂t Ω) (29)
is a theory of the ground states of a topologically-ordered sys- 2π
tem on Σ. There are no excited states in Chern-Simons theory R 2π
because the Hamiltonian vanishes. This is seen most simply where Ω(r, t) = 0 dφ (ω(r1 , φ, t)−ω(r2 , φ, t)). Therefore,
in a0 = 0 gauge, where the momentum canonically conjugate Ω is canonically conjugate to λ. By a gauge transformation,
k
to a1 is − 4π a2 , and the momentum canonically conjugate to we can always rotate λ and Ω so that they are along the 3
k
a2 is 4π a1 so that direction in su(2), i.e. λ = λ3 T 3 , Ω = Ω3 T 3 . Since it is
defined through the exponential in (28), Ω3 takes values in
k [0, 2π]. Therefore, its canonical conjugate, λ3 , is quantized
H= tr (a2 ∂0 a1 − a1 ∂0 a2 ) − L = 0 (26) to be an integer. From the definition of λ in (28), we see that
4π
λ3 ≡ λ3 +2k. However, by a gauge transformation given by a
Note that this is a special feature of an action with a Chern- rotation around the 1-axis, we can transform λ → −λ. Hence,
Simons term alone. If the action had both a Chern-Simons the independent allowed values of λ are 0, 1, . . . , k.
26
On the two-punctured sphere, if one puncture is of type a, One extra transformation in the mapping class group, com-
the other puncture must be of type ā. (If the topological charge pared to the braid group, is a 2π rotation of a puncture/particle
at one puncture is measured along a loop around the puncture relative to the rest of the system (a Dehn twist). If we consider
– e.g. by a Wilson loop, see subsection III.C – then the loop particles with a finite extent, rather than point particles, then
can be deformed so that it goes around the other puncture, but we must consider the possibility of such rotations. For in-
in the opposite direction. Therefore, the two punctures neces- stance, if the particles are small dipoles, then we can represent
sarily have conjugate topological charges.) For SU(2), a = ā, their world lines as ribbons. A Dehn twist then corresponds to
so both punctures have the same topological charge. There- a twist of the ribbon. Thickening a world line into a ribbon is
fore, the restriction to only k + 1 different possible allowed called a framing. A given world line has multiple choices of
boundary conditions λ for the two-punctured sphere implies framing, corresponding to how many times the ribbon twists.
that there are k + 1 different quasiparticle types in SU(2)k A framing is actually essential in Chern-Simons theory be-
Chern-Simons theory. As we will describe in later subsec- cause flux is attached to charge through the constraint (22)
tions, these allowed quasiparticle types can be identified with or (27). By putting the flux and charge at opposite edges of
the j = 0, 21 , . . . , k2 representations of the SU(2)2 Kac-Moody the ribbon, which is a short-distance regularization of the the-
algebra. ory, we can associate a well-defined phase to a particle tra-
jectory. Otherwise, we wouldn’t know how many times the
charge went around the flux.
Any transformation acting on a single particle can only re-
2. TQFTs and Quasiparticle Properties sult in a phase; the corresponding phase is called the twist
parameter Θa . Often, one writes Θa ≡ e2πiha , where ha
We will continue with our analysis of Chern-Simons theory is called the spin of the particle.5 (One must, however,
in sections III.C and III.D. Here, we will make some more be careful to distinguish this from the actual spin of the
general observations abut TQFTs and the topological prop- particle, which determines its transformation properties un-
erties of quasiparticles. We turn to the n-punctured sphere, der the three-dimensional rotation group and must be half-
Σ = S 2 \P1 ∪ P2 ∪ . . . ∪ Pn , i.e. the sphere S 2 with the points integral.) However, ha is well-defined even if the system is
P1 , P2 . . . Pn deleted, which is equivalent to n − 1 quasipar- not rotationally-invariant, so it is usually called the topolog-
ticles in the plane (the nth puncture becomes the point at ∞). ical spin of the particle. For Abelian anyons, it is just the
This will allow us to study the topological properties of quasi- statistics parameter, θ = 2πiha .
particle excitations purely from ground state properties. To The ground state properties on arbitrary surfaces, including
see how braiding emerges in this approach, it is useful to the n-punctured sphere and the torus, can be built up from
note that diffeomorphisms should have a unitary representa- more primitive vector spaces in the following way. An ar-
tion on the ground state Hilbert space (i.e. they should com- bitrary closed surface can be divided into a collection of 3-
mute with the Hamiltonian). Diffeomorphisms which can be punctured spheres which are glued together at their bound-
smoothly deformed to the identity should have a trivial ac- aries. This is called a ‘pants decomposition’ because of the
tion on the Hilbert space of the theory since there are no lo- topological equivalence of a 3-punctured sphere to a pair of
cal degrees of freedom. However, ‘large’ diffeomorphisms pants. Therefore, the 3-punctured sphere plays a fundamen-
could have a non-trivial unitary representation on the theory’s tal role in the description of a topological phase. Its Hilbert
c
Hilbert space. If we take the quotient of the diffeomorphism space is denoted by Vab , if a, b, and c are the particle types at
group by the set of diffeomorphisms which can be smoothly the three punctures. If the a and b punctures are fused, a two-
deformed to the identity, then we obtain the mapping class punctured sphere will result. From the above analysis, it has a
group. On the n-punctured sphere, the braid group Bn−1 is a one-dimensional vector space if both punctures have topolog-
subgroup of the mapping class group.4 Therefore, if we study ical charge c and a zero-dimensional vector space otherwise.
Chern-Simons theory on the n-punctured sphere as we did for The dimension of the Hilbert space of the 3-punctured sphere
c c
the 2-punctured sphere above, and determine how the map- is given by the fusion multiplicity Nab = dim (Vab
P ) which ap-
c
ping class group acts, we can learn all of the desired informa- pears in the fusion rule, φa × φb = c Nab φc . The Hilbert
tion about quasiparticle braiding. We do this by two different space on a surface obtained by gluing together 3-punctured
methods in subsections III.C and III.D. spheres is obtained by tensoring together the V ’s and sum-
4 The mapping class group is non-trivial solely as a result of the punctures. In 5 If a is its own anti-particle, so that two as can fuse to 1, then Raa ∗
1 = ±Θa ,
particular, any diffeomorphism which moves one or more of the punctures where the minus sign is acquired for some particle types a which are not
around other punctures cannot be deformed to the identity; conversely, if quite their own antiparticles but only up to some transformation which
two diffeomorphisms move the same punctures along trajectories which squares to −1. This is analogous to the fact that the fundamental repre-
can be deformed into each other, then the diffeomorphisms themselves can sentation of SU(2) is not real but is pseudoreal. Consequently, a spin-1/2
also be deformed into each other. These classes of diffeomorphisms corre- particle ψµ and antiparticle ψµ† can form a singlet, ψµ† ψµ , but two spin-
spond to the braid group which is, in fact, a normal subgroup. If we take 1/2 particles can as well, ψµ ψν i(σy )µν , where σy is the antisymmet-
the quotient of the mapping class group by the Dehn twists of n − 1 of the ric Pauli matrix. When some quantities are computed, an extra factor of
punctures – all except the point at infinity – we would be left with the braid (iσy )2 = −1 results. This ± sign is called the Froebenius-Schur indica-
group Bn−1 . tor. (See, for instance, Bantay, 1997.)
27
ming over the particle types at the punctures where gluing oc- temperatures, one must also consider transitions to excited
curs. For instance, the Hilbert space on the 4-punctured sphere states, but the contributions of these will be ∼ e−∆/T . Fur-
e c e
is given by the direct sum Vabd = ⊕c Vab Vcd ; the Hilbert space thermore if we were to add a time dependent (source) term to
a a
on the torus is VT 2 = ⊕a V1a Va1 . (If one of the particle types the action, these properties will remain preserved so long as
is the vacuum, then the corresponding puncture can simply the frequency of this term remains small compared with the
be removed; the 3-punctured sphere is then actually only 2- gap.
punctured. Gluing two of them together end to end gives a Aside from the n-punctured spheres, the torus is the most
torus. This is one way of seeing that the degeneracy on the important manifold for considering topological phases. Al-
torus is the number of particle types.) though not directly relevant to experiments, the torus is very
The Hilbert space of the n-punctured sphere with topolog- important for numerical simulations since periodic boundary
ical charge a at each puncture can be constructed by sewing conditions are often the simplest choice. As noted above, the
together a chain of (n − 2) 3-punctured spheres. The resulting ground state degeneracy on the torus is equal to the number
1 b1 b2 a of quasiparticle species. Suppose one can numerically solve a
Hilbert space is: Va...a = ⊕bi Vaa Vab1 . . . Vab N −3
. A simple
graphical notation for a set of basis states of this Hilbert space Hamiltonian on the torus. If it has a gap between its ground
is given by a fusion chain (similar to the fusion tree discussed state(s) and the lowest energy excited states, then its ground
in appendix A): state degeneracy is an important topological property of the
state – namely the number of of quasiparticle species. A sim-
ple physical understanding of this degeneracy can be obtained
a a a a a a in the following way. Suppose that we have a system of elec-
a b1 b2 b3 b4 . . . bn−4 bn−3 a trons in a topological phase. If we consider the system on the
torus, then the electrons must have periodic boundary con-
The first two as on the far left fuse to b1 . The next a ditions around either generator of the torus (i.e. around ei-
fuses with b1 to give b2 . The next a fuses with b2 to give ther handle), but the quasiparticles need not. In the Abelian
b3 , and so on. The different basis vectors in this Hilbert ν = 1/m fractional quantum Hall state, for instance, it is pos-
space correspond to the different possible allowed bi s. The sible for a quasiparticle to pick up a phase e2πin/m in going
b1 b2 a around the meridian of the torus, where n can take any of the
dimension of this Hilbert space is Naa Nab1 . . . NabN −3
=
b b a values n = 0, 1, . . . , m−1; electrons would still have periodic
(Na )a1 (Na )b21 . . . (Na )bN −3 . On the right-hand-side of this
c boundary conditions since they are made up of m quasiparti-
equation, we have suggested that the fusion multiplicity Nab cles. Indeed, all m of these possibilities occur, so the ground
c
can be viewed as a matrix (Na )b associated with quasiparticle
state is m-fold degenerate.
species a. Let us denote the largest eigenvalue of the ma- Let us make this a little more precise. We introduce oper-
trix Na by da . Then the Hilbert space of M quasiparticles of
ators T1 and T2 which create a quasiparticle-quasihole pair,
type a has dimension ∼ dM−2 a for large M . For this reason, take the quasiparticle around the meridian or longitude, re-
da is called the quantum dimension of an a quasiparticle. It
spectively, of the torus and annihilate them again. Then T1
is the asymptotic degeneracy per particle of a collection of a and T2 must satisfy:
quasiparticle. For Abelian particles, da = 1 since the multi-
particle Hilbert space is one-dimensional (for fixed particle T2−1 T1−1 T2 T1 = e2πi/m (30)
positions). Non-Abelian particles have da > 1. Note that
da is not, in general, an integer, which is symptomatic of the because T1−1 T1 amounts to a contractible quasiparticle-
non-locality of the Hilbert space: it is not the tensor product quasihole loop, as does T2−1 T2 ; by alternating these pro-
of da -dimensional Hilbert spaces associated locally with each cesses, we cause these loops to be linked. The quasiparticle
particle. trajectories in spacetime (which can be visualized as a thick-
This non-locality is responsible for the stability of this de- ened torus) are equivalent to a simple link between two circles
generate ground state Hilbert space. Not only the Yang-Mills (the Hopf link): the first quasiparticle-quasihole pair is pulled
term, but all possible gauge-invariant terms which we can add apart along the meridian (T1 ); but before they can be brought
to the action (24) are irrelevant. This means that adding such a back together (T1−1 ), the second pair is pulled apart along the
term to the action might split the ∼ dM−2 a -dimensional space longitude (T2 ). After the first pair is brought back together
of degenerate states in a finite-size system, but the splitting and annihilated (T1−1 ), the second one is, too (T2−1 ). In other
must vanish as the system size and the particle separations go words, the phase on the right-hand-side of Eq. 30 is simply the
to infinity. In fact, we can make an even stronger statement phase obtained when one quasiparticle winds around another.
than that. All ground state matrix elements of gauge-invariant This algebra can be represented on a vector space of minimum
a
local operators such as the field strength squared, Fµν F µνa , dimension m. Let us call the states in this vector space |ni,
vanish identically because of the Chern-Simons constraint. n = 0, 1, . . . , m − 1. Then
Therefore, the degeneracy is not lifted at all in perturbation
theory. It can only be lifted by non-perturbative effects (e.g. T1 |ni = e2πin/m |ni
instantons/quantum tunneling), which could cause a splitting T2 |ni = |(n + 1) mod mi (31)
∼ e−gL where g is inversely proportional to the coefficient
of the Yang-Mills term. Therefore, the multi-quasiparticle These m states correspond to n = 0, 1, . . . , m − 1 quanta of
states are degenerate to within exponential accuracy. At finite- flux threaded through the torus. If we were to cut along a
28
meridian and open the torus into an annulus, then these states a is:
would have flux n threaded through the hole in the annulus S1a
and charge n/m at the inner boundary of the annulus (and a da = (34)
S11
compensating charge at the outer boundary). We can instead
switch to a basis in which T2 is diagonal by a discrete Fourier The mathematical structure encapsulating these braiding
transform. If we write |ñi = √1m m−1
P 2πinñ/m
|ni, then and fusion rules is a modular tensor category (Bakalov and
n=0 e
Kirillov, 2001; Kassel, 1995; Kitaev, 2006; Turaev, 1994;
|ñi is an eigenstgate of T2 with eigenvalue e2πiñ/m . In this Walker, 1991). A category is composed of objects and mor-
basis, T1 is an off-diagonal operator which changes the bound- phisms, which are maps between the objects which preserve
ary conditions of quasiparticles around the longitude of the their defining structure. The idea is that one can learn more
torus. In non-Abelian states, a more complicated version of about the objects by understanding the morphisms between
the same thing occurs, as we discuss for the case of Ising them. In our case, the objects are particles with labels (which
anyons at the end of section III.B. The different boundary con- specify their species) as well as fixed configurations of sev-
ditions around the meridian correspond to the different possi- eral particles. The morphisms are particle trajectories, which
ble quasiparticle types which could thread the torus (or, equiv- map a set of labeled partices at some initial time to a set of
alently, could be present at the inner boundary of the annulus labeled particles at some final time. A tensor category has
if the torus were cut open along a meridian). One can switch a tensor product structure for multiplying objects; here, this
to a basis in which the boundary conditions around the longi- is simply the fact that one can take two well-separated (and
tude are fixed. The desired basis change is analogous to the historically well-separated) collections of particles and con-
discrete Fourier transform given above and is given by the ‘S- sider their union to be a new ‘tensor-product’ collection. Since
matrix’ or ‘modular S-matrix’ of the theory. Switching the we consider particles in two dimensions, the trajectories are
longitude and meridian is one of the generators of the map- essentially the elements of the braid group, but they include
ping class group of the torus; the S-matrix expresses how it the additional possibility of twisting. (Allowing twists in the
acts on the ground state Hilbert space. The elements of the strands of a braid yields a braided ribbon category.) We will
S-matrix are closely related to quasiparticle braiding. By fol- further allow the trajectories to include the fusion of two parti-
lowing a similar construction to the one with T1 , T2 above, cles (so that we now have a fusion category). Morphisms can,
one can see that Sab is equal to the amplitude for creating aā therefore, be defined by specifying Θa , Vab c
, R, and F .
and bb̄ pairs, braiding a and b, and annihilating again in pairs. Why is it necessary to invoke category theory simply to
This is why, in an Abelian state, the elements of the S-matrix specify the topological properties of non-Abelian anyons?
are all phases (up to an overall normalization which ensures Could the braid group not be the highest level of abstraction
′
unitarity), e.g. Snn′ = √1m e2πinn /m in the example above. that we need? The answer is that for a fixed number of par-
In a non-Abelian state, the different entries in the matrix can ticles n, the braid group Bn completely specifies their topo-
have different magnitudes, so the basis change is a little more logical properties (perhaps with the addition of twists Θa to
complicated than a Fourier transform. Entries can even vanish account for the finite size of the particles). However, we need
in the non-Abelian case since, after a and b have been braided, representations of Bn for all values of n which are compatible
a and ā may no longer fuse to 1. with each other and with fusion (of which pair creation and an-
In the case of Ising anyons on the torus (SU(2)2 ), there nihilation is simply the special case of fusion to the vacuum).
are three ground states. One basis is |1m i, |σm i, |ψm i, So we really need a more complex – and much more tightly
corresponding to the different allowed topological charges constrained – structure. This is provided by the concept of a
which would be measured at the inner boundary of the re- modular tensor category. The F and R matrices play particu-
sulting annulus if the torus were cut open along its meridian. larly important roles. The F matrix can essentially be viewed
An equally good basis is given by eigenstates of topological an associativity relation for fusion: we could first fuse i with
charge around the longitude: |1l i, |σl i, |ψl i. As we will see in j, and then fuse the result with k; or we could fuse i with
at the end of the next section, the basis change between them the result of fusing j with k. The consistency of this property
is given by leads to a constraint on the F -matrices called the pentagon
equation. (An explicit example of the pentagon equation is
1 √1 1 worked out in Section IV.B.) Consistency between F and R
2 2 2
S= √1 0 − √12
(32) leads to a constraint called the hexagon equation. Modularity
2
1 is the condition that the S-matrix be invertible. These self-
2 − √12 1
2 consistency conditions are sufficiently strong that a solution
to them completely defines a topological phase.6
The S-matrix not only contains information about braiding, An equivalent alternative to studying punctured surfaces is
but also about fusion, according to Verlinde’s formula (Ver- to add non-dynamical charges which are coupled to the Chern-
linde, 1988) (for a proof, see Moore and Seiberg, 1988, 1989): Simons gauge field. Then the right-hand-side of the constraint
X Sax Sbx Sc̄x
c
Nab = (33)
x
S1x
6 Modulo details regarding the central charge c at the edge. e2πic/8 can be
Consequently, the quantum dimension of a particle of species obtained from the topological spins, but not c itself.
29
(27) is modified and a non-trivial gauge field configuration very different in the weak-coupling limit. However, fermionic
is again obtained which is essentially equivalent to that ob- quasiparticles and vortices are really just different types of
tained around a puncture. In the following subsections, we quasiparticle excitations in a superconductor – i.e. different
will discuss the Hilbert spaces of SU(2)k Chern-Simons the- types of localized disturbances above the ground state. There-
ory, either on the n-punctured sphere or in the presence of fore, we will often refer to them both as simply quasiparticles
non-dynamical sources. These discussions will enable us to and use the terms Bogoliubov-de Gennes or fermionic when
compute the braiding and fusion matrices. The non-trivial referring to the former. In a p + ip superconductor, the quasi-
quasiparticle of SU(2)1 is actually Abelian so we do not dis- particles which exhibit non-Abelian statistics are flux hc/2e
cuss this ‘trivial’ case. The next case, SU(2)2 , is non-Abelian vortices.
and may be relevant to the ν = 5/2 fractional quantum Hall
state. It can be understood in several different equivalent
ways, which express its underlying free Majorana fermion 1. Vortices and Fermion Zero Modes
structure. Quantum computation with Majorana fermions is
described in Section IV.A. In the next section, we explain this Let us suppose that we have a system of fully spin-polarized
structure from the perspective of a superconductor with p + ip electrons in a superconducting state of px + ipy pairing sym-
pairing symmetry. Although this description is very elegant, metry. The mean field Hamiltonian for such a superconductor
it cannot be generalized to higher k. Therefore, in the two is,
sections after that, we describe two different approaches to Z
solving SU(2)k Chern-Simons theory for general k. We reca- H = dr ψ † (r)h0 ψ(r) (35)
pitulate the case of SU(2)2 in these other languages and also Z
describe the case of SU(2)3 . The latter has quasiparticles in 1
+ dr dr′ D∗ (r, r′ )ψ(r′ )ψ(r) + D(r, r′ )ψ † (r)ψ † (r′ )
its spectrum which are Fibonacci anyons, a particularly beau- 2
tiful non-Abelian anyonic structure which allows for univer- 1
sal topological quantum computation. It may also underlie the with single-particle term h0 = − 2m ∇2 − µ and complex p-
observed ν = 12/5 fractonal quantum Hall state. More details wave pairing function
of the Fibonacci theory are given in Sections IV.B.
r + r′
D(r, r′ ) = ∆ (i∂x′ − ∂y′ )δ(r − r′ ). (36)
2
The ground state of (36) is the celebrated BCS wave function, u = v ∗ therefore has energy zero. We will soon be consid-
written here in an un-normalized form, ering situations in which there are multiple zero energy solu-
tions (ui , u∗i ), i = 1, 2, . . .. If we denote the corresponding
Y vk † †
P vk † †
|g.s.i = 1+
c c
ck c−k |vaci = e k uk k −k |vaci operators by γi (see eq. 47 below), then they satisfy:
uk
k
(40) γi† = γi (44)
where
! Eq. (44) is the definition of a Majorana fermion operator.
|uk |2 1 µ Let us now consider the BdG equations in the presence of
= 1∓ p (41)
|vk |2 2 µ2 + |∆k|2 vortices when the bulk of the superconductor is in the µ > 0
phase. As usual, a vortex is characterized by a point of van-
are the BCS coherence factors. The wave function (40) de- ishing ∆, and a 2π-winding of the phase of ∆ around that
scribes a coherent state of an undetermined number of Cooper point. In principle we should, then, solve the BdG equations
pairs, each in an internal state of angular momentum ℓ = −1. in the presence of such a non-uniform ∆. However, we can,
Its projection onto a fixed even number of particles N is car- instead, solve them in the presence of a non-uniform µ, which
ried out by expanding the exponent in (40) to the (N/2)th or- is much simpler. All that we really need is to make the core of
der. When written in first quantized language, this wave func- the superconductor topologically distinct from the bulk – i.e.
tion describes a properly anti-symmetrized wave function of a puncture in the superconductivity. Making µ < 0 in the core
N/2 Cooper-pairs, each in an internal state is just as good as taking ∆ to zero, as far as topological prop-
X vk erties are concerned. Therefore, we associate the core of the
g(r) = eikr (42) vortex with a region of µ < 0, whereas the bulk is at µ > 0.
uk Thus, there is a µ = 0 line encircling the vortex core. This
k
line is an internal edge of the system. We will consider the
In first quantized form the multiparticle BCS wavefunction dynamics of edge excitations in more detail in section III.E,
is then of the form of the Pfaffian of an antisymmetric matrix but here we will be content to show that a zero energy mode
whose i − j element is g(ri − rj ), an antisymmetrized product is among them.
of pair wavefunctions The simplest situation to consider is that of azymuthal sym-
metry, with the polar coordinates denoted by r and θ. Imag-
ΨBCS = Pf [g(ri − rj )] (43)
ine the vortex core to be at the origin, so that ∆(r, θ) =
= A [g(r1 − r2 )g(r3 − r4 ) . . . g(rN −1 − rN )] |∆(r)|eiθ+iΩ . Here Ω is the phase of the order parameter
along the θ = 0 line, a phase which will play an important
with A being an antisymmetrization operator. role later in our discussion. Assume that the µ = 0 line is the
The function g(r) depends crucially on the sign of µ, since circle r = r0 , and write
the small k behavior of vk /uk depends on that sign. When
µ > 0, we have g(r) = 1/(x + iy) in the long distance limit µ(r) = ∆ h(r), (45)
(Read and Green, 2000). If we assume this form holds for all
distances, the Pfaffian wave function obtained is identical to with h(r) large and positive for large r, and h(r) < 0 for
the Moore-Read form discussed below in connection with the r < r0 ; therefore, the electron density will vanish for r ≪ r0 .
Ising model and the ν = 5/2 quantum Hall state in section Such a potential defines an edge at r = r0 . There are low-
III.D (see Eqs. ??). The slow decay of g(r) implies a weak energy eigenstates of the BdG Hamiltonian which are spa-
Cooper pairing. (But it does not imply that the state is gapless. tially localized near r = 0 and are exponentially decaying
One can verify that electron Green functions all decay expo- for r → ∞:
nentially for any non-zero µ.) When µ < 0 the function g(r) −iθ/2
decays much more rapidly with r, generically in an exponen- edge iℓθ − 0r h(r ′ )dr ′ e
R
tial way, such that the Cooper pairs are strongly bound. Fur- φE (r, θ) = e e , (46)
eiθ/2
thermore, there is a topological distinction between the µ > 0
and µ < 0 phases. The distinction, which is discussed in de- The spinor on the right-hand-side points in a direction in pseu-
tail in (Read and Green, 2000), implies that, despite the fact dospin space which is tangent to the r = r0 circle at θ. This
that both states are superconducting, the µ > 0 and µ < 0 wavefunction describes a chiral wave propagating around the
states must be separated by a phase transition. (In the analo- edge, with angular momentum ℓ and energy E = ∆ℓ/r0 .
gous quantum Hall state, both states are characterized by the Since the flux is an odd multiple of hc/2e, the Bogoliubov
same Hall conductivity but are separated by a phase transi- quasiparticle (46) must be anti-periodic as it goes around the
tion, and are distinguished by their thermal Hall conductivi- vortex. However, the spinor on the right-hand-side of (46) is
ties(Read and Green, 2000)) Indeed, from (39) we see that the also anti-periodic. Therefore, the angular momentum ℓ must
gap vanishes for a uniform p + ip superconductor with µ = 0. be an integer, ℓ ∈ Z. Consequently, a flux hc/2e vortex has
The low-energy BdG eigenstates at this second-order phase an ℓ = 0 solution, with energy E = 0. (Conversely, if the flux
transition point form a Dirac cone. through the vortex were an even multiple of hc/2e, ℓ would be
For every solution (u, v) of the BdG equations with energy a half-integer, ℓ ∈ Z + 21 , and there would be no zero-mode.)
E, there is a solution (v ∗ , u∗ ) of energy −E. A solution with The operator corresponding to this zero mode, which we will
31
call γ, can be written in the form: refer to the state for which ψ † ψ = 1 as a “filled fermion”,
Z h i and to the ψ † ψ = 0 state as an empty fermion. Note however
1 i i
that the eigenvalue of ψ † ψ has no bearing on the occupation
γ= √ dr F (r) e− 2 Ω ψ(r) + F ∗ (r) e 2 Ω ψ † (r) (47)
2 of single-particle states.)
Rr ′ ′ Of course, the pairing of vortices to form Dirac fermions is
Here, F (r) = e− 0 h(r )dr e−iθ/2 . Since each γ is an equal arbitrary. A given pairing defines a basis, but one can trans-
superposition of electron and hole, it is overall a chargeless, form to a basis associated with another pairing. Consider four
neutral fermion operator vortices with corresponding zero modes γ1 , γ2 , γ3 , γ4 . The
When there are several well separated vortices at posi- F -matrix transforms states from the basis in which iγ1 γ2 and
tions Ri , the gap function near the ith vortex takes the form iγ3 γ4 are diagonal to the basis in which iγ1 γ4 and iγ2 γ3 are
∆(r) = P|∆(r)| exp (iθi + iΩi ), with θi = arg (r − Ri ) and diagonal. Since iγ1 γ4 acts as σx on an iγ1 γ2 eigenstate, the
Ωi = j6=i arg((Rj − Ri )). There is then one zero energy F -matrix is just the basis change from the σz basis to the σx
solution per vortex. Each zero energy solution γi is localized basis:
near the core of its vortex at Ri , but the phase Ωi that replaces
Ω in (47) depends on the position of all vortices. Moreover, σσσ 1 1 1
[Fσ ] = √ (50)
the dependence of the Majorana operators γi on the positions 2 1 −1
Ri is not single valued.
While for any E 6= 0 the operators Γ†E , ΓE are conventional We will refer to this type of non-Abelian anyons by the name
fermionic creation and annihilation operators, the γi ’s are not. ‘Ising anyons’; they are the model introduced in Section
II.A.1. The reason for the name will be explained in Section
In particular, for E 6= 0 we have (Γ†E )2 = Γ2E = 0, but
III.E.
the zero energy operators follow (with a convenient choice of
In a compact geometry, there must be an even number of
normalization) γi2 = 1. The two types of fermion operator
vortices (since a vortex carries half a flux quantum, and the
share the property of mutual anti-commutation, i.e., the γ’s
number of flux quanta penetrating a compact surface must be
satisfy {γi , γj } = 2δij .
integer). In a non-compact geometry, if the number of vortices
is odd, the edge has a zero energy state of its own, as we show
2. Topological Properties of p + ip Superconductors
in Section III.E.
Now, let us examine what happens to the Majorana opera-
The existence of the γi ’s implies a degeneracy of the ground tors and to the ground states as vortices move. The positions
state. The counting of the number of degenerate ground states of the vortices are parameters in the Hamiltonian (36). When
should be done with care. A pair of conventional fermionic they vary adiabatically in time, the operators γi vary adiabati-
creation and annihilation operators span a two dimensional cally in time. In principle, there are two sources for this vari-
Hilbert space, since their square vanishes. This is not true ation - the explicit dependence of γi on the positions and the
for a Majorana operator. Thus, to count the degeneracy of Berry phase associated with the motion. The choice of phases
the ground state when 2N0 vortices are present, we construct taken at (47) is such that the Berry phase vanishes, and the
“conventional” complex (Dirac) fermionic creation and anni- entire time dependence is explicit. The non-single-valuedness
hilation operators, of the phases in (47) implies then that a change of 2π in Ω,
which takes place when one vortex encircles another, does not
ψi = (γi + iγN0 +i )/2 (48) leave the state unchanged.
As vortices adiabatically traverse trajectories that start and
ψi† = (γi − iγN0 +i )/2 (49) end in the same set of positions (Ivanov, 2001; Stern et al.,
2 2004), there is a unitary transformation U within the subspace
These operators satisfy ψi2 = ψi† = 0 and thus span a of ground states that takes the initial state |ψ(t = 0)i to the
two-dimensional subspace of degenerate ground states asso- final one |ψ(t = T )i,
ciated with these operators. Over all, then, the system has
2N0 degenerate ground states. If the fermion number is fixed |ψ(t = T )i = U |ψ(t = 0)i. (51)
to be even or odd, then the degeneracy is 2N0 −1√. Therefore,
Correspondingly, the time evolution of the operators γi is
the quantum dimension of a vortex is dvort = 2 or, in√the
notation introduced in Sec. II.A.1 for Ising anyons, dσ = 2. γi (t = T ) = U γi (t = 0)U † . (52)
For any two vortices i and j, we can associate a two state
system. If we work in the basis of iγi γj eigenstates, then iγi γj By reading the time evolution of γi from their explicit form
acts as σz with eigenvalues ±1, while γi and γj act as σx and (47) we can determine U up to a phase. Indeed, one expects
σy . (However, it is important to keep in mind that Majorana this Abelian phase to depend not only on the topology of the
fermions γk , γl anti-commute with γi , γj , unlike operators trajectory but also on its geometry, especially in the analogous
associated with different spins, which commute.) The two quantum Hall case, where there is an Aharonov-Bohm phase
eigenvalues iγi γj = ∓1 are the two fusion channels of two accumulated as a result of the charge carried by the quasipar-
fermions. If we form the Dirac fermion ψ = (γi +iγj )/2, then ticle.
the two iγi γj eigenstates have ψ † ψ = 0, 1. Therefore, we will When vortex i encircles vortex i + 1, the unitary transfor-
call these fusion channels 1 and ψ. (One is then tempted to mation is simple: both γi and γi+1 are multiplied by −1, with
32
all other operators unchanged. This is a consequence of the change that probability. Rather, they affect phases in the su-
fact that when the order parameter changes by a phase fac- perpositions. Using this point of view it is then possible to
tor 2π, fermionic operators change by a phase π. Exchange show that two ingredients are essential for the non-Abelian
trajectories, in which some of the vortices trade places, are statistics of the vortices. The first is the quantum entangle-
more complicated, since the phase changes of Ωk associated ment of the occupation of states near the cores of distant vor-
with a particular trajectory do not only depend on the winding tices. The second ingredient is familiar from (Abelian) frac-
numbers, but also on the details of the trajectory and on the tional statistics: the geometric phase accumulated by a vortex
precise definition of the cut of the function arg(r) where its traversing a closed loop.
value jumps by 2π. Therefore, we conclude that, for p−wave superconductors,
The simplest example is the interchange of two vortices. the existence of zero-energy intra-vortex modes leads, first, to
Inevitably, one of the vortices crosses the branch cut line a multitude of ground states, and, second, to a particle-hole
of the other vortex. We can place the branch cuts so that symmetric occupation of the vortex cores in all ground states.
a counterclockwise exchange of vortices 1 and 2 transforms When represented in occupation-number basis, a ground state
c1 → c2 and c2 → −c1 while a clockwise exchange trans- is a superposition which has equal probability for the vortex
forms c1 → −c2 and c2 → c1 (Ivanov, 2001). core being empty or occupied by one fermion. When a vortex
This may be summarized by writing the representation ma- traverses a trajectory that encircles another vortex, the phase
trices for the braid group generators (Ivanov, 2001; Nayak and it accumulates depends again on the number of fluid particles
Wilczek, 1996): it encircles. Since a fluid particle is, in this case, a Cooper
π
ρ(σi ) = eiθ e− 4 γi γi+1 (53) pair, the occupation of a vortex core by a fermion, half a pair,
leads to an accumulation of a phase of π relative to the case
where θ is the Abelian part of the transformation. The two when the core is empty. And since the ground state is a su-
eigenvalues iγi γi+1 = ∓1 are the two fusion channels 1 and perposition with equal weights for the two possibilities, the
ψ of a pair of vortices. From (53), we see that the R-matrices relative phase of π introduced by the encircling might in this
σσ
satisfy Rψ = i R1σσ (i.e., the phase of taking two σ particles case transform the system from one ground state to another.
around each other differ by i depending on whether they fuse
to ψ or 1). It is difficult to obtain the Abelian part of the phase Now consider the ground state degeneracy of a p + ip su-
using the methods of this section, but we will derive it by other perconductor on the torus. Let us define, following Oshikawa
methods in Sections III.C and III.D. The non-Abelian part of et al., 2007 (see also Chung and Stone, 2007), the opera-
(53), i.e. the second factor on the right-hand-side, is the same tors A1 , A2 which create a pair of Bogoliubov-de Gennes
as a π/2 rotation in the spinor representation of SO(2n) (see quasiparticles, take one around the meridian or longitude of
Nayak and Wilczek, 1996 for details). The fact that braiding the torus, respectively, and annihilate them again. We then
only enacts π/2 rotations is the reason why this type of non- define B1 , B2 as operators which create a vortex-antivortex
Abelian anyon does not enable universal topological quantum pair, take the vortex around the meridian or longitude of the
computation, as we discuss further in section IV. torus, respectively, and annihilate them. B1 increases the flux
According to (53), if a system starts in a ground state |gsα i through the hole encircled by the longitude of the torus by
and vortex j winds around vortex j + 1, the system’s final one half of a flux quantum while B2 does the same for the
state is γj γj+1 |gsα i. Writing this out in terms of the original other hole. These operators satisfy the commutation relations
electron operators, we have [A1 , A2 ] = 0 and A1 B2 = −B2 A1 , A2 B1 = −B1 A2 . We
i can construct a multiplet of ground states as follows. Since
i i i
cj e 2 Ωj + c†j e− 2 Ωj cj+1 e 2 Ωj+1 + c†j+1 e− 2 Ωj+1 |gsα i , A1 and A2 commute and square to 1, we can label states by
(54) their A1 and A2 eigenvalues ±1. Let |1, 1i be the state with
(†) both eigenvalues equal to 1, i.e. A1 |1, 1i = A2 |1, 1i = |1, 1i.
where cj annihilates a particle in the state F (r − Rj ) and
Then B1 |1, 1i = |1, −1i and B2 |1, 1i = | − 1, 1i. Suppose
(†)
cj+1 creates a particle in the state (F (r − Rj+1 )) localized we now try to apply B2 to B1 |1, 1i = |1, −1i. This will cre-
very close to the cores of the jth and (j + 1)th vortex, respec- ate a vortex-antivortex pair; the Majorana zero modes, γa , γb
tively. Eq. (54) seemingly implies that the motion of the jth associated with the vortex and anti-vortex will be in the state
vortex around the (j + 1)th vortex affects the occupations of |0i defined by (γa + iγb ) |0i = 0. When the vortex is taken
states very close to the cores of the two vortices. This is in around the longitude of the torus, its Majorana mode will be
contrast, however, to the derivation leading to Eq. (54), which multiplied by −1: γa → −γa . Now, the vortex-antivortex pair
explicitly assumes that vortices are kept far enough from one will no longer be in the state |0i, but will instead be in the state
another so that tunneling between vortex cores may be disre- |1i defined by (γa − iγb ) |1i = 0. Consequently, the vortex-
garded. antivortex pair can no longer annihilate to the vacuum. When
This seeming contradiction is analyzed in detail in Stern they fuse, a fermion is left over. Therefore, B2 B1 |1, 1i does
et al., 2004, where it is shown that the unitary transformation not give a new ground state (and, by a similar argument, nei-
(54) does not affect the occupation of the core states of the ther does B1 B2 |1, 1i). Consequently, a p + ip superconductor
j, j + 1 vortices, because all ground states are composed of has ‘only’ three ground states on the torus. A√basis in which
superpositions in which the core states have a probability of B1 is diagonal is given by: (|1, 1i ± |1, −1i)/ 2, with eigen-
one-half to be occupied and one-half to be empty. The uni- value ±1, and | − 1, 1i, with eigenvalue 0 (since there is zero
tary transformation within the ground state subspace does not amplitude for B1 | − 1, 1i to be in the ground state subspace).
33
They can be identified with the states |1m i, |ψm i, and |σm i for k > 2 does not have a free fermion or boson description.7
in Ising anyon language. √Meanwhile, B2 is diagonal in the Therefore, in the next two subsections, we discuss these field
basis (|1, 1i ± | − 1, 1i)/ 2, |1, −1i. By changing from one theories using more general methods.
basis to the other, we find the S-matrix given in the previous Even though its Hamiltonian vanishes and it has no local
subsection follows. degrees of freedom, solving Chern-Simons theory is still a
The essential feature of chiral p-wave superconductors is non-trivial matter. The reason is that it is difficult in a non-
that they have Majorana fermion excitations which have zero Abelian gauge theory to disentangle the physical topological
energy modes at vortices (and gapless excitations at the edge degrees of freedom from the unphysical local gauge degrees
of the system, see section III.E). The Majorana character is a of freedom. There are essentially two approaches. Each has
result of the superconductivity, which mixes particle and hole its advantages, and we will describe them both. One is to
states; the zero modes and gapless edge excitations result from work entirely with gauge-invariant quantities and derive rules
the chirality. Majorana fermions arise in a completely differ- governing them; this is the route which we pursue in this sub-
ent way in the Kitaev honeycomb lattice model (Kitaev, 2006): section. The second is to pick a gauge and simply calculate
X X X within this gauge, which we do in the next subsection (III.D).
H = −Jx σjx σjx − Jy σjy σjy − Jz σjx σjz Consider SU(2)k non-Abelian Chern-Simons theory:
x−links y−links z−links
Z
(55) k 2
where the z-links are the vertical links on the honeycomb lat- SCS [a] = tr a ∧ da + a ∧ a ∧ a (57)
4π M 3
tice, and the x and y links are at angles ±π/3 from the ver-
tical. The spins can be represented by Majorana fermions We modify the action by the addition of sources, j µa , accord-
bx , by , bz , and c, according to σjx = ibxj cj , σjx = ibyj cj , ing to L → L + tr (j · a). We take the sources to be a set
σjx = ibzj cj so long as the constraint bxj byj bzj cj = 1 is satis- of particles on prescribed classical trajectories. The ith parti-
fied. Then, the Hamiltonian is quartic in Majorana fermion cle carries the spin ji representation of SU(2). As we saw in
operators, but the operators bxj bxk , byj byk , bzj bzk commute with subsection III.A, there are only k + 1 allowed representations;
the Hamiltonian. Therefore, we can take their eigenvalues as later in this subsection, we will see that if we give a particle
parameters ujk = bα α
j bk , with α = x, y, or z appropriate to a higher spin representation than j = k/2, then the amplitude
the jk link. These parameters can be varied to minimize the will vanish identically. Therefore, ji must be in allowed set of
Hamiltonian, which just describes Majorana fermions hop- k + 1 possibilities: 0, 12 , . . . , k2 . The functional integral in the
ping on the honeycomb lattice: presence of these sources can be written in terms of Wilson
loops, Wγi ,ji [a], which are defined as follows. The holonomy
iX Uγ,j [a] is an SU (2) matrix associated with a curve γ. It is
H= tjk cj ck (56)
4 defined as the path-ordered exponential integral of the gauge
jk
field along the path γ:
where tjk = 2Jα ujk for nearest neighbor j, k and zero other-
wise. For different values of the Jα s, the tjk ’s take different c c
H
Uγ,j [a] ≡ Pei γ a T ·dl
values. The topological properties of the corresponding cj ∞ Z 2π Z s1 Z
X sn−1 h
bands are encapsulated by their Chern number (Kitaev, 2006).
= in ds1 ds2 . . . dsn γ̇(s1 )·aa1 (γ(s1 )) T a1 . . .
For a certain range of Jα s, a P, T -violating perturbation gives 0 0 0
n=0
the Majorana fermions a gap in such a way as to support zero i
modes on vortex-like excitations (plaquettes on which one of γ̇(sn ) · aan (γ(sn )) T an (58)
the ujk s is reversed in sign). These excitations are identical in
topological character to the vortices of a p+ip superconductor where P is the path-ordering symbol. The Lie algebra genera-
discussed above. tors T a are taken in the spin j representation. ~γ (s), s ∈ [0, 2π]
is a parametrization of γ; the holonomy is clearly independent
of the parametrization. The Wilson loop is the trace of the
C. Chern-Simons Effective Field Theories, the Jones holonomy:
Polynomial, and Non-Abelian Topological Phases
Wγ,j [a] = tr (Uγ,j [a]) (59)
1. Chern-Simons Theory and Link Invariants
Let us consider the simplest case, in which the source is a
In the previous subsection, we have seen an extremely sim- quasiparticle-quasihole pair of type j which is created out of
ple and transparent formulation of the quasiparticle braiding the ground state, propagated for a period of time, and then
properties of a particular non-Abelian topological state which,
as we will see later in this section, is equivalent to SU(2)2
Chern-Simons theory. It describes the multi-quasiparticle
Hilbert space and the action of braiding operations in terms of 7 It is an open question whether there is an alternative description of an
free fermions. Most non-Abelian topological states are not so SU(2)k topological phase with k > 2 in terms of fermions or bosons which
simple, however. In particular, SU(2)k Chern-Simons theory is similar to the p + ip chiral superconductor formulation of SU(2)2 .
34
(a) χ (b) χ
annihilated, returning the system to the ground state. The am-
plitude for such a process is given by:
Z γ+
γ+
h0|0iγ,j = Da eiSCS [a] Wγ,j [a] (60)
t=0 1 2 3 4
Here, γ is the spacetime loop formed by the trajectory of the
quasiparticle-quasihole pair. The Wilson loop was introduced
γ− γ−
as an order parameter for confinement in a gauge theory be-
cause this amplitude roughly measures the force between the
quasiparticle and the quasihole. If they were to interact with ψ = ρ((σ 2)
2
) χ χ
a confining force V (r) ∼ r, then the logarithm of this ampli-
tude would be proportional to the the area of the loop; if they
were to have a short-ranged interaction, it would be propor-
tional to the perimeter of the loop. However, Chern-Simons (c) χ (d) χ
theory is independent of a metric, so the amplitude cannot de-
pend on any length scales. It must simply be a constant. For
j = 1/2, we will call this constant d. As the notation implies,
it is, in fact, the quantum dimension of a j = 1/2 particle. As
we will see below, d can be determined in terms of the level
k, and the quantum dimensions of higher spin particles can be
expressed in terms of d.
We can also consider the amplitude for two pairs of quasi-
particles to be created out of the ground state, propagated for
−1
some time, and then annihilated, returning the system to the ρ(σ2 ) χ ρ(σ2 ) χ
ground state:
FIG. 4 The functional integrals which give (a) hχ|ρ σ22 |χi (b)
` ´
Z ` −1 ´
hχ|χi, (c) hχ|ρ(σ2 ) |χi, (d) hχ|ρ σ2 |χi.
h0|0iγ1 ,j1 ;γ2 ,j2 = Da eiSCS [a] Wγ,j [a] Wγ ′ ,j ′ [a] (61)
This amplitude can take different values depending on how braided with quasiparticle 3, so 1 and 2 may no longer fuse to
γ and γ ′ are linked as in Fig. 4a vs 4b. If the curves are the vacuum. In just a moment, we will see an example of a
unlinked the integral must give d2 , but when they are linked different four-quasiparticle state.
the value can be nontrivial. In a similar way, we can formulate We now interpret the t = 0 to t = ∞ history as the conju-
the amplitudes for an arbitrary number of sources. gate of a t = −∞ to t = 0 history. In other words, it gives us
It is useful to think about the history in figure 4a as a two a four quasiparticle bra rather than a four quasiparticle ket:
step process: from t = −∞ to t = 0 and from t = 0 to Z
t = ∞. (The two pairs are created at some time t < 0 and ∗
χ [A] = Da(x, t) Wγ+ ,j [a] Wγ+′ ,j ′ [a] ×
annihilated at some time t > 0.) At t = 0− , the system is in a(x,0)=A(x)
a four-quasiparticle state. (Quasiparticles and quasiholes are R∞
dt
R
d2 x LCS
topologically equivalent if G =SU(2), so we will use ‘quasi- e 0 (63)
particle’ to refer to both.) Let us call this state ψ: In the state |χi, quasiparticles 1 and 2 fuse to form the trivial
Z quasiparticle, as do quasiparticles 3 and 4. Then we can in-
ψ[A] = Da(x, t) Wγ− ,j [a] Wγ−′ ,j ′ [a] × terpret the functional integral from t = −∞ to t = ∞ as the
a(x,0)=A(x) matrix element between the bra and the ket:
R0
dt
R
d2 x LCS
Z
e −∞ (62)
hχ|ψi = Da eiSCS [a] Wγ1 ,j1 [a] Wγ2 ,j2 [a] (64)
′
where γ− and γ− are the arcs given by γ(t) and γ ′ (t) for
t < 0. A(x) is the value of the gauge field on the t = 0 spatial Now, observe that |ψi is obtained from |χi by taking quasi-
slice; the wavefunctional ψ[A] assigns an amplitude to every particle 2 around quasiparticle 3,i.e. by exchanging quasipar-
spatial gauge field configuration. For G=SU(2) and k > 1, ticles 2 and 3 twice, |ψi = ρ σ22 |χi. Hence,
there are actually two different four-quasiparticle states: if
Z
particles 1 and 2 fuse to the identity field j = 0, then par- hχ|ρ σ2 |χi = Da eiSCS [a] Wγ1 ,j1 [a] Wγ2 ,j2 [a] (65)
2
Consider, now, the state ρ(σ2 ) |χi, in which particles 2 and VL (q) is the Jones polynomial associated with the link L =
3 are exchanged just once. It is depicted in figure 4c. Simi- γ1 ∪ . . . ∪ γn , evaluated at q = −eπi/(k+2) using the skein re-
larly, the state ρ σ2−1 |ψi is depicted in figure 4d. From the lation (73). Note that we assume here that all of the quasiparti-
figure, we see that cles transform under the j = 12 representation of SU (2). The
other quasiparticle types can be obtained through the fusion
hχ|ρ(σ2 ) |χi = d (66) of several j = 1/2 quasiparticles, as we will discuss below in
hχ|ρ σ2−1 |χi = d (67) Section III.C.2.
assume that these arcs continue in some arbitrary way and (a) (b)
form closed loops. By applying the Kauffman bracket recur- t
sion relation in figure 5, we see that these arcs can be replaced
by two arcs which do not cross. In Section II.C.3, we will use Π1
t=0 1 3 4 1 2 3 4
2
= q1/2 + q−1/2
= q +
tain earlier spatial slices. If there are no quasiparticles, then quences of this difference are that Θσ = e−πi/8 while Θ1/2 =
there are no Wilson lines terminating at Σ. However, the e−3πi/8 ; R1σσ = e−πi/8 while R02
1 1
,2
= −e−3πi/8 ; Rψ σσ
=
functional integral can have Wilson loops in the body of the 1 1
h 1 1 1i
3πi/8 2,2 πi/8 σσσ 2,2,2
solid torus as in Figure 7a. These correspond to processes e while R1 = e ; [Fσ ]ab = − F 1 . The
2 ab
in the past, t < 0, in which a quasiparticle-quashole pair rest of the F -matrices are the same, as are the fusion multiplic-
was created, taken around the meridian of the torus and an- ities Nabc
and the S-matrix. In other words, the basic structure
nihilated. The Wilson loop can be in any of the k + 1 al- of the non-Abelian statistics is the same in the two theories,
lowed representations j = 0, 12 , . . . , k2 ; in this way, we obtain but there are some minor differences in the U(1) phases which
k + 1 ground state kets on the torus (we will see momen- result from braiding. Both theories have threefold ground
tarily that they are all linearly independent). Wilson loops state degeneracy on the torus; the Moore-Read Pfaffian state
around the meridian are contractible (Figure 7b), so they can has ground state degeneracy 6 because of an extra U(1) factor
be simply evaluated by taking their Kauffman bracket; they corresponding to the electrical charge degrees of freedom.
multiply the state by dj . Evidently, these Wilson loop oper- Of course, in the k = 2 case we have already obtained all of
ators are diagonal in this basis. Bras can be obtained by in- these results by the method of the previous subsection. How-
tegrating the Chern-Simons Lagrangian over the 3-manifold ever, this approach has two advantages: (1) once Witten’s re-
M′ = Σ × [0, ∞) = S 3 \S 1 × D2 , i.e. the exterior of the sult (74) and Kauffman’s recursion relation (77) are accepted,
torus. Wilson loops in the exterior torus are now contractible braiding matrix elements can be obtained by straightforward
if they are parallel to a longitude but non-trivial if they are high school algebra; (2) the method applies to all levels k,
around the meridian, as in in Figure 7c. Again, we obtain unlike free Majorana fermion methods which apply only to
k + 1 ground state bras in this way. The matrix elements be- the k = 2 case. There is an added bonus, which is that this
tween these bras and kets (appropriately normalized such that formalism is closely related to the techniques used to analyze
the matrix product of a bra with its conjugate ket is unity) lattice models of topological phases, which we discuss in a
are the entries in the S-matrix, which is precisely the basis later subsection.
change between the longitudinal and meridinal bases. A ma-
trix element can be computed by evaluating the corresponding
picture. The ab entry in the S-matrix is given by evaluating D. Chern-Simons Theory, Conformal Field Theory, and
the Kauffman bracket of the picture in Figure 7d (and divid- Fractional Quantum Hall States
ing by the normalization of the states). This figure makes the
relationship between the S-matrix and braiding clear. 1. The Relation between Chern-Simons Theory and
Conformal Field Theory
(a) (b)
Now, we consider Chern-Simons theory in a particular
j gauge, namely holomorphic gauge (to be defined below). The
j ground state wavefunction(s) of Chern-Simons theory can be
obtained by performing the functional integral from the dis-
tant past, t = −∞, to time t = 0 as in the previous subsec-
tion:
Z
j j R0 R 2
(c) (d)
( ψ[A(x)] = Da(x, t) e −∞ dt Σ d x LCS (80)
a(x,0)=A(x)
There are no other cubic terms in the action once az̄ has been (Elitzur et al., 1989; Witten, 1989), we have mapped the prob-
eliminated (as is the case in any such gauge in which one of lem of computing the ground state wavefunction (in 2 + 0-
the components of the gauge field vanishes). The constraint dimensions) of Chern-Simons theory, which is a topological
imposed by the δ-function can be solved by taking theory with a gap, to the problem of computing a correla-
tion function in the chiral WZW model (in 1 + 1-dimensions),
aai = ∂i U U −1 (82) which is a critical theory. This is a bit peculiar since one the-
ory is gapped while the other is gapless. However, the gap-
where U is a single-valued function taking values in the Lie less degrees of freedom of the WZW model for the t = 0
group. Substituting this into the right-hand-side of (81), we spatial slice are pure gauge degrees of freedom for the cor-
find that the action which appears in the exponent in the func- responding Chern-Simons theory. (In the very similar situa-
tional integral takes the form tion of a surface Σ with boundary, however, the correspond-
Z ing conformally-invariant 1 + 1-D theory describes the actual
k
S = ǫij tr ∂i U U −1 ∂z̄ ∂j U U −1 dynamical excitations of the edge of the system, as we discuss
4π ZD2 ×S 1 in section III.E.) Only the topological properties of the chiral
k
WZW conformal blocks are physically meaningful for us.
= ǫij tr ∂i U U −1 ∂z̄ ∂j U U −1 +
4π D2 ×S 1 More complicated topological states with multiple Chern-
tr ∂i U U −1 ∂j U ∂z̄ U −1 Simons fields and, possibly, Higgs fields (Fradkin et al., 2001,
Z 1999, 1998) correspond in a similar way to other chiral ratio-
k nal conformal field theories which are obtained by tensoring
= ǫij ∂j tr ∂i U −1 ∂z̄ U +
4π D2 ×S 1 or cosetting WZW models. (RCFTs are those CFTs which
−1 −1
have a finite number of primary fields – see appendix A for
tr ∂i U U ∂j U ∂z̄ U the definition of a primary field – under some extended chiral
Z
k algebra which envelopes the Virasoro algebra; a Kac-Moody
= tr ∂z U −1 ∂z̄ U + algebra in the WZW case; and, possibly, other symmetry gen-
Z4π T 2 erators.) Consequently, it is possible to use the powerful al-
k
ǫµνλ tr ∂µ U U −1 ∂ν U U −1 ∂λ U U −1 (83) gebraic techniques of rational conformal field theory to com-
12π D2 ×S 1
pute the ground state wavefunctions of a large class of topo-
a logical states of matter. The quasiparticles of the topological
The Jacobian which comes from the δ-function δ(fij ) is can-
celled by that associated with the change of integration vari- state correspond to the primary fields of the chiral RCFT. (It
able from Da to DU . In the final line, the first term has is a matter of convenience whether one computes correlation
been integrated by parts while the second term, although it functions with a primary field or one of its descendants since
appears to be an integral over the 3D manifold, only depends their topological properties are the same. This is a freedom
on the boundary values of U (Wess and Zumino, 1971; Wit- which can be exploited, as we describe below.)
ten, 1983). This is the Wess-Zumino-Witten (WZW) action. The conformal blocks of an RCFT have one property which
What we learn from (83) then, is that, in a particular gauge, the is particularly useful for us, namely they are holomorphic
ground state wavefunction of 2 + 1-D Chern-Simons theory functions of the coordinates. This makes them excellent can-
can be viewed as the partition function of a 2 + 0-dimensional didate wavefunctions for quantum Hall states. We identify pri-
WZW model. mary fields with the quasiparticles of the quantum Hall state,
For positive integer k, the WZW model is a 2D confor- and compute the corresponding conformal block. However,
mal field theory which, in the SU(2) case, has Virasoro cen- there is one important issue which must be resolved: a quan-
3k
tral charge c = c̄ = k+2 . (For a brief review of some of tum Hall wavefunction is normally viewed as a wavefunction
the basics of conformal field theory, see appendix A and ref- for electrons (the quasiparticle positions, by contrast, are usu-
erences therein.) However, in computing properties of the ally viewed merely as some collective coordinates specifying
Chern-Simons theory from which we have derived it, we will a given excited state). Where are the electrons in our RCFT?
couple only to az = ∂z U · U −1 ; i.e. only to the holomor- Electrons have trivial braiding properties. When one electron
phic or right-moving sector of the theory. Thus, it is the chiral is taken around another, the wavefunction is unchanged, ex-
WZW model which controls the ground state wavefunction(s) cept for a phase change which is an odd integral multiple of
of Chern-Simons theory. 2π. More importantly, when any quasiparticle is taken around
If we were to follow the same strategy to calculate the an electron, the wavefunction is unchanged apart from a phase
Chern-Simons ground state wavefunction with Wilson lines change which is an integral multiple of 2π. Therefore, the
or punctures present, then we would end up with a correlation electron must be a descendant of the identity. In other words,
function of operators in the chiral WZW model transforming the RCFT must contain a fermionic operator by which we can
under the corresponding representations of SU(2). (Strictly extend the chiral algebra. This new symmetry generator is es-
speaking, it is not a correlation function, but a conformal sentially the electron creation operator – which is, therefore, a
block, which is a chiral building block for a correlation func- descendant of the identity under its own action. Not all RCFTs
tion. While correlation functions are single-valued, conformal have such an operator in their spectrum, so this is a strong con-
blocks have the non-trivial monodromy properties which we straint on RCFTs which can describe quantum Hall states. If
need, as is discussed in appendix A.) Therefore, following we are interested, instead, in a quantum Hall state of bosons,
39
as could occur with ultra-cold bosonic atoms in a rotating op- 2D Electrons in B field:
tical trap (Cooper et al., 2001), then the RCFT must contain a Observation of FQHE Numerics
bosonic field by which we can extend the chiral algebra.
An RCFT correlation function of Ne electron operators
therefore corresponds to the Chern-Simons ground state Trial Wave Functions
wavefunction with Ne topologically-trivial Wilson lines.
From a purely topological perspective, such a wavefunction
is just as good as a wavefunction with no Wilson lines, so Edge Theory
the Wilson lines would seem superfluous. However, if the de- WZW Model CFT
scendant field which represents the electron operator is cho-
sen cleverly, then the wavefunction with Ne Wilson lines may
be a ‘good’ trial wavefunction for electrons in the quantum
Hall regime. Indeed, in some cases, one finds that these trial
wavefunctions are the exact quantum Hall ground states of
simple model Hamiltonians (Ardonne and Schoutens, 1999;
Blok and Wen, 1992; Greiter et al., 1991; Moore and Read, Low Energy Theory
1991; Read and Rezayi, 1999; Simon et al., 2007a; Wen and
Wu, 1994). In the study of the quantum Hall effect, however, FIG. 8 How one arrives at a low-energy theory of the quantum Hall
a wavefunction is ‘good’ if it is energetically favorable for a effect. At the top, one begins with the experimental observation of
realistic Hamiltonian, which is beyond the scope of the un- the quantized Hall effect. At the bottom, we know the low energy
derlying Chern-Simons theory, which itself only knows about theory should be of Chern-Simons form. One would like to be able
braiding properties. It is unexpected good luck that the trial to “integrate out” high energy degrees of freedom directly to obtain
wavefunctions obtained from Chern-Simons theory are often the low energy theory, as shown by the dotted line, but must instead
found to be ‘good’ from this energetic perspective, which is take a more circuitous route, as described in the text.
a reflection of how highly constrained quantum Hall wave-
functions are, and how central these braiding properties are to
their physics. We emphasize, however, that a wavefunction Rezayi, 1999; Simon et al., 2007c). One may try to justify
obtained in this way will not be the exact ground state wave- it ex post facto by solving for the properties of quasiholes
function for electrons with Coulomb interactions. In some of a system with some unrealistic (e.g. involving 3-body or
cases it might not even have particularly high overlap with the higher interactions) but soluble Hamiltonian. The degeneracy
ground state wavefunction, or have good energetics. The one can be established by counting (Nayak and Wilczek, 1996;
thing which it does capture is the topological structure of a Read, 2006; Read and Rezayi, 1996). The braiding matrices
particular universality class. can be obtained by numerically computing the Berry integrals
for the given wavefunctions (Tserkovnyak and Simon, 2003)
or by using their connection to conformal field theory to de-
2. Quantum Hall Wavefunctions from Conformal Field Theory duce them (Gurarie and Nayak, 1997; Moore and Read, 1991;
Nayak and Wilczek, 1996; Slingerland and Bais, 2001). One
Ideally, the logic which would lead us to a particular RCFT can then deduce the Chern-Simons effective field theory of
would be as follows, as displayed in Fig. 8. One begins the state either from the quasiparticle properties or from the
with the experimental observation of the quantized Hall ef- associated conformal field theory with which both it and the
fect at some filling fraction ν (shown at the top). We certainly wavefunctions are connected.
know that the Hamiltonian for the system is simply that of We now show how such wavefunctions can be constructed
2D electrons in a magnetic field, and at the bottom, we know through some examples. In appendix A, we review some of
the form of the low energy theory should be of Chern-Simons the rudiments of conformal field theory.
form. One would like to be able to “integrate out” high en- (a) Wavefunctions from CFTs: Our goal is to construct
ergy degrees of freedom directly to obtain the low-energy the- a LLL FQH wavefunction Ψ(z1 , . . . , zN ) which describes an
ory. Given the low-energy Chern-Simons effective field the- electron fluid in a circular droplet centered at the origin. Ψ
ory, one can pass to the associated RCFT, as described above. must be a homogeneous antisymmetric analytic function of
With the RCFT in hand, one can construct wavefunctions, as the zi s, independent of the z̄i s apart from the Gaussian fac-
we will describe below. Indeed, such a procedure has been tor, which we will frequently ignore (see Sec. II.C.1). If we
explicitly achieved for Abelian quantum Hall states (López consider the FQHE of bosons, we would need Ψ to instead be
and Fradkin, 1991; Zhang et al., 1989). In some special non- symmetric. The filling fraction ν of a FQH wavefunction Ψ
Abelian cases, progress in this direction has been made (Wen, is given by ν = N/NΦ where N is the number of electrons
1991b, 1999). and NΦ is the number of flux quanta penetrating the droplet
For most non-Abelian theories, however, the situation is not (Prange and Girvin, 1990). In the LLL, NΦ is given by the
so simple. The RCFT is usually obtained through inspired highest power of z occurring in Ψ.
guesswork (Ardonne and Schoutens, 1999; Blok and Wen, We will also frequently need the fact that in an incompress-
1992; Cappelli et al., 2001; Moore and Read, 1991; Read and ible state of filling fraction ν, multiplying a wavefuncton by a
40
Q
factor i (zi − w)m pushes a charge νm away from the point guaranteed that ψqh and ψe are local with respect to each
w. This can be understood (Laughlin, 1983) as insertion of m other, i.e. taking one field around can only produce a phase
flux quanta at the point w, which, via Faraday’s law creates an which is a multiple of 2π. Consequently, the wavefunction
azimuthal electric field, which, then, via the Hall conductivity Ψ remains analytic in the electron coordinates zi even after
transfers charge νm away from the point w. the fields ψqh (w1 ) . . . ψqh (wM ) have been inserted into the
Our strategy will be to choose a particular chiral RCFT, correlation function.
pick an “electron” field ψe in this theory (which, by the rea- One important feature of the conformal block on the right-
soning given above, must be a fermionic generator of the ex- hand-side of (85) is that ψqh (wa ) and ψe (zi ) are on roughly
tended chiral algebra of the theory), and write a ground state the same footing – they are both fields in some conformal field
trial wavefunction Ψgs for N electrons as theory (or, equivalently, they are both fixed sources coupled
to the Chern-Simons gauge field). However, when intepreted
Ψgs = h ψe (z1 ) . . . ψe (zN ) i (84) as a wavefunction on the left-hand-side of (85), the electron
coordinates zi are the variables for which the wavefunction
The field ψe must be fermionic since the quantum Hall wave- gives a probability amplitude while the quasihole coordinates
function on the left-hand-side must be suitable for electrons. wa are merely some parameters in this wavefunction. If we
Not all RCFTs have such a field in their spectrum, so this re- wished to normalize the wavefunction differently, we could
quirement constrains our choice. This requirement also en- multiply by an arbitrary function of the wa s. However, the
sures that we will obtain a wavefunction which has no branch particular normalization which is given by the right-hand-side
cuts; in particular, there will only be one conformal block on of (85) is particularly convenient, as we will see momentar-
the right-hand-side of (84). We must do a little more work in ily. Note that since the quasihole positions wj are merely pa-
choosing ψe so that there are no poles either on the right-hand- rameters in the wavefunction, the wavefunction need not be
side of (84). As discussed above, the correlation function on analytic in these coordinates.
the right-hand-side of (84) is a ground state wavefunction of
(b) Quasiparticle Braiding: The branch cuts in quasihole
Chern-Simons theory with Ne trivial topological charges at
positions wa are symptoms of the fact that there may be a
fixed positions z1 , z2 , . . . , zNe .
vector space of conformal blocks corresponding to the right-
Of course, there isn’t a unique choice of RCFT, even at
hand-side of (85). In such a case, even when the quasihole
a given filling fraction. Therefore, there are different frac-
positions are fixed, there are several possible linearly inde-
tional quantum Hall states which can be constructed in this
pendent wavefunctions. These multiple degenerate states are
way. Which fractional quantum Hall state is actually observed
necessary for non-Abelian statistics, and they will generically
at a particular ν is determined by comparing the energies of
mix when the quasiholes are dragged around each other.
the various possible competing ground states. Having a good
However, there is still a logical gap in the above reason-
wavefunction is, by itself, no guarantee that this wavefunction
ing. The wavefunctions produced by an RCFT have the cor-
actually describes the physical system. Only a calculation of
rect braiding properties for the corresponding Chern-Simons
its energy gives real evidence that it is better than other possi-
ground state wavefunction built into them through their ex-
ble states.
plicit monodromy properties. As a result of the branch cuts
The reason for introducing this complex machinery sim-
in the conformal blocks as a function of the wa s, when one
ply to construct a wavefunction becomes clearer when
quasihole is taken around another, the wavefunction Ψα trans-
we consider quasihole wavefunctions, which are Chern-
forms into M αβ Ψβ , where the index α = 1, 2, . . . , g runs
Simons ground state wavefunctions with Ne trivial topolog-
over the g different degenerate n-quasihole states. However,
ical charges and Nqh non-trivial topological charges. In gen-
when viewed as quantum Hall wavefunctions, their quasipar-
eral, there are many possible quasihole operators, correspond-
ticle braiding properties are a combination of their explicit
ing to the different primary fields of the theory, so we must
monodromy and the Berry matrix which is obtained from:
really consider Nqh1 , Nqh2 , . . . Nqhm numbers of quasiholes
I D E
if there are m primary fields. Each different primary field
eiγαβ = P exp ~ Ψα ∇w~ Ψβ
corresponds to a different topologically-distinct type of “de- dw (86)
fect” in the ground state. (As in the case of electrons, we are
free to choose a descendant field in place of the corresponding where Ψα , α = 1, 2, . . . , g are the g different degenerate n-
primary field since the two have identical topological proper- quasihole states and P is the path ordering symbol. In this
ties although the wavefunction generated by a descendant will equation, the zi s are integrated over in order to compute the
be different from that generated by its primary.) Let us sup- inner product, but the wa s are held fixed, except for the one
pose that we focus attention on a particular type of quasipar- which is taken around some loop.
ticle which, in most cases, will be the quasiparticle of mini- Strictly speaking, the effect of braiding is to transform a
mal electrical charge. Then we can write a wavefunction with state according to Ψα → eiγαβ M βγ Ψγ . By changing the nor-
quasiholes at positions w1 , . . . , wM as malization of the wavefunction, we can alter eiγαβ and M βγ .
Only the product of the two matrices on the right-hand-side
Ψ(w1 . . . wM) = hψqh (w1 ) . . . ψqh (wM ) ψe (z1 ) . . . ψe (zN )i of this equation is gauge invariant and physically meaning-
(85) ful. When we presume that the braiding properties of this
where ψqh is the corresponding primary field. Since ψqh is a wavefunction are given by those of the corresponding CFT
primary field and ψe is a descendant of the identity, we are and Chern-Simons theory, we take it to be equal to M βγ and
41
ignore eiγαβ . This can only be correct if γαβ vanishes up to as the ν = 1/m Laughlin wavefunction. The astute reader
a geometric phase proportional to the area for a wavefunction will notice that the correlator in Eq. 88 actually violates the
given by a CFT conformal block. In the case of the Laughlin neutrality condition discussed in Appendix A and so it should
states, it can be verified that this is indeed correct by repeat- actually have zero value. One fix for this problem is to insert
ing the the Arovas, Schrieffer, Wilczek calculation (Arovas into the correlator (by
√
hand) a neutralizing vertex operator at
et al., 1984) with the Laughlin state normalized according to −iN φ(z=∞) m
infinity e which then makes Eq. 88 valid (up
the quasihole position dependence given by the corresponding to a contant factor). Another approach is to insert an oper-
CFT (see below) (Blok and Wen, 1992). This calculation rests ator that smears the neutralizing background over the entire
upon the plasma analogy originally introduced by Laughlin in system (Moore and Read, 1991). This approach also conve-
his seminal work (Laughlin, 1983). For other, more complex niently results in the neglected Gaussian factors reappearing!
states, it is more difficult to compute the Berry matrix. A ver- We will ignore these neutralizing factors for simplicity. From
sion of a plasma analogy for the MR Pfaffian state was con- now on, we will drop the Gaussian factors from quantum Hall
structed in Gurarie and Nayak, 1997; one could thereby ver- wavefunctions, with the understanding that they result from
ify the vanishing of the Berry matrix for a two-quasihole state including a smeared neutralizing background.
and, with some further assumptions, for four and higher mul- The quasihole operator must be√a primary field; the primary
tiquasihole states. A direct evaluation of the integral in (86) field of minimum charge is eiφ/ m . Using Eq. A7, Eq. 85
by the Monte-Carlo method (Tserkovnyak and Simon, 2003) yields
established that it vanishes for MR Pfaffian quasiholes. The
M N M
effect of Landau level mixing on statistics has also been stud- Ψ(w1 , . . . , wM) = Π (wi − wj )1/m Π Π (zi − wj )Ψgs
ied (Simon, 2007). Although there has not been a complete i<j i=1 j=1
proof that the CFT-Chern-Simons braiding rules are identical Q (89)
to those of the wavefunction, when it is interpreted as an elec- As mentioned above, the factor j (zj − w) “pushes” charge
tron wavefunction (i.e. there has not been a complete proof away from the position w leaving a hole of charge precisely
that (86) vanishes when the wavefunction is a CFT confor- Q = +e/m. The first term on the right of Eq. 89 results from
mal block), there is compelling evidence for the MR Pfaffian the fusion of quasihole operators with each other, and explic-
state, and it is almost certainly true for many other states as itly shows the fractional statistics of the quasiholes. Adia-
well. We will, therefore, take it as a given that we can simply batically taking two quasiholes around each other results in a
read off the braiding properties of the wavefunctions which fractional phase of 2π/m. As promised above, this statistical
we construct below. term appears automatically in the wavefunction given by this
(c) The Laughlin State: We now consider wavefunctions
CFT!
generated by perhaps the simplest CFT, the chiral boson.√We (d) Moore-Read Pfaffian State:
In the Ising CFT (see Appendix A), we might try to use
suppose that the chiral√boson has compactification radius m,
so that φ ≡ φ + 2π m. The U(1) Kac-Moody algebra and ψe (z) = ψ(z) as the electron field (Moore and Read, 1991).
enveloping Virasoro algebra can be extended by the symmetry The ψ fields can fuse together in pairs to give the identity
√
generator eiφ m . Since the dimension of this operator is m/2, (since ψ × ψ = 1) so long as there are an even number of
it is fermionic for m odd and bosonic for even m. The primary fields. However, when we take two ψ fields close to each
√
fields of this extended chiral algebra are of the form einφ/ m , other, the OPE tells us that
with n = 0, 1, . . . , m − 1. They are all of the fields√
which lim ψ(zi )ψ(zj ) ∼ 1/(zi − zj ) (90)
are not descendants and are local with respect to eiφ m (and zi →zj
to the Kac-Moody and Virasoro generators), as may be seen which diverges as zi → zj and is therefore unacceptable as a
from the operator product expansion (OPE) (see Appendix A): wavefunction. To remedy this problem, we tensor the Ising
√ √ √ CFT √ with the chiral boson CFT. There is now an operator
eiφ(z) m inφ(0)/ m
e ∼ z n ei(n+m)φ(0)/ m
+ ... (87)
ψ eiφ m by which we can extend the chiral algebra. (If m
When z is taken around the origin, the right-hand-side is un- is even, this symmetry generator is fermionic; if m is odd, it
changed. It is convenient to normalize√the U (1) current as is bosonic.) As before, we will take this symmetry genera-
tor to be our electron √field. The corresponding primary fields
j = √1m ∂φ; then the primary field einφ/ m has charge n/m. √ √
√ are of the form einφ/ m , σ ei(2n+1)φ/2 m , and ψ einφ/ m ,
We take ψe = eiφ m as our electron field (which has charge where n = 0, 1, . . . , m − 1. Again, these are determined by
1) and consider the resulting ground state wavefunction ac- the requirement of locality with respect to the generators of
cording to Eq. 84. Using Eq. A7 we find the chiral algebra, i.e. that they are single-valued when taken
Q around a symmetry generator, in particular the elecron field
Ψgs = hψe (z1 ) . . . ψe (zN )i = i<j (zi − zj )m (88) √
iφ m
ψe . For instance,
It is now clear why we have chosen this CFT: to have Ψgs √ √
given by correlators of a vertex operator of the form eiαφ an- ψ(z) eiφ(z) m
· σ(0) ei(2n+1)φ(0)/2 m
∼
√
alytic (no branch cuts or poles) we must have α2 = m a non- z −1/2 σ(0) z n+1/2 ei(2(n+m)+1)φ(0)/ m
+ ...
negative integer, and m must be odd to obtain an antisymmet- n
√
i(2(n+m)+1)φ(0)/ m
ric wavefunction (or even for symmetric). We recognize Ψgs = z σ(0) e (91)
42
and similarly for the other primary fields. and the fermion orbital shared by the quasiholes is unoccu-
Using our new symmetry generator as the electron field, we pied. When an odd electron is added, it ‘occupies’ this or-
obtain the ground state wavefunction according to Eq. 84: bital, although the fermion orbital is neutral and the electron
Q is charged (we can think of the electrons’ charge as being
Ψgs = hψ(z1 ) . . . ψ(zN )i i<j (zi − zj )m screened by the superfluid).
When there are many quasiholes, they may fuse together in
1 Q m
= Pf i<j (zi − zj ) (92) many different ways. Thus, even when the quasihole positions
zi − zj are fixed there are many degenerate ground states, each cor-
(See, e.g. Di Francesco et al., 1997 for the calculation of this responding to a different conformal block (see appendix A).
correlation function.) Again, m odd gives an antisymmetric This degeneracy is precisely what is required for non-Abelian
wavefunction and m even gives a symmetric wavefunction. statistics. Braiding the quasiholes around each other produces
For m = 2 (and even N ), Eq. 92 gives precisely the Moore- a rotation within this degenerate space.
Read Pfaffian wavefunction (Eq. 43 with g = 1/z and two Fusing 2m σ fields results in 2m−1 conformal blocks, as
Jastrow factors attached). may be seen by examining the Bratteli diagram of Fig. 22 in
To determine the filling fraction of our newly constructed appendix A. When two quasiholes come together, they may
wavefunction, we need only look at the exponent of the Jas- either fuse to form 1 or ψ. As above, if they come together to
trow factor in Eq. 92. Recall that the filling fraction is de- form 1 then taking the two quasiholes around each other gives
termined by the highest power of any z (See III.D.1 above). a phase of −2π/8 + 2π/4m. On the other hand, if they fuse
There are m(N − 1) factors of z1 in the Jastrow factor. The to form ψ then taking them around each other gives a phase of
Pfaffian has a factor of z1 in the denominator, so the highest 2π3/8 + 2π/4m.
power of z1 is m(N − 1) − 1. However, in the thermody- These conclusions can be illustrated explicitly in the cases
namic limit, the number of factors scales as mN . Thus the of two and four quasiholes. For the case of two quasiholes, the
filling fraction is ν = 1/m. correlation function (93) can be evaluated to give (Moore and
We now consider quasihole operators. As in the Laughlin Read, 1991; Nayak and Wilczek, 1996) (for an even number
√ of electrons):
case we might consider the primary fields ψqh = einφ/ m .
Similar arguments as in the Laughlin case show that the n = 1 Y
case generates precisely the Laughlin quasihole of charge Ψ(w1 , w2 ) = (zj − zk )2 ×
Q = +e/m. However we have other options for our quasi-
j<k
hole which
√
have smaller electrical charge. The primary field (zj − w1 ) (zk − w2 ) + zj ↔ zk
Pf . (94)
σ eiφ/2 m has charge Q = +e/2m. We then obtain the wave- zj − zk
function according to Eq. 85
where w12 = w1 − w2 . For simplicity, we specialize to
Ψ(w1 , . . . , wM) = hσ(w1 ) . . . σ(wm )ψ(z1 ) . . . ψ(zN )i × the case m = 2; in general, there would be a prefactor
1 1
M
1/2m
N M
1/2
N
m (w12 ) 4m − 8 . When the two quasiholes at w1 and w2 are
Π (wi − wj ) Π Π (zi − wj ) Π (zi − zj ) (93)
i<j i=1 j=1 i<j brought together at the point w, a single flux quantum Laugh-
lin quasiparticle results, since two σs can only fuse to the iden-
Using the fusion rules of the σ fields (See Eq. 7, as well as tity in this case, as expected from the above arguments:
Fig. 22 and Table II in Appendix A), we see that it is impossi-
ble to obtain 1 from an odd number of σ fields. We conclude Y Y
1
2
that quasiholes ψqh can only occur in pairs. Let us then con- Ψqh (w) = (zj − zk ) (zi − w) Pf .
zj − zk
sider the simplest case of two quasiholes. If there is an even j<k i
number of electrons, the ψ fields fuse in pairs to form 1, and (95)
the remaining two quasiholes must fuse to form 1 also. As
discussed in Eq. A3 the OPE of the two σ fields will then The situation becomes more interesting when we consider
−1/8 states with 4 quasiholes. The ground state is 2-fold degener-
have a factor of (w1 − w2 )√ . In addition, the fusion of the
two vertex operators e iφ/2 m
results in the first term in the ate (see appendix A). If there is an even number of electrons
second line of Eq. 93, (w1 − w2 )1/(4m) . Thus the phase ac- (which fuse to form the identity), we are then concerned with
cumulated by taking the two quasiholes around each other is the hσσσσi correlator. As discussed in appendix A, two or-
−2π/8 + 2π/4m. thogonal conformal blocks can be specified by whether 1 and
On the other hand, with an odd number of electrons in 2 fuse to form either 1 or ψ. The corresponding wavefunctions
the system, the ψ’s fuse in pairs, but leave one unpaired ψ. obtained by evaluating these conformal blocks are (Nayak and
The two σ’s must then fuse to form a ψ which can then fuse Wilczek, 1996):
with the unpaired ψ to give the identity. (See Eq. A3). In 1
this case, the OPE of the two σ fields will give a factor of (1,ψ) (w13 w24 ) 4 √
Ψ = √ 1/2 Ψ(13)(24) ± x Ψ(14)(23) (96)
(w1 − w2 )3/8 . Thus the phase accumulated by taking the two (1 ± x)
quasiholes around each other is 6π/8 + 2π/4m.
In the language of section III.B above, when there is an where x = w14 w23 /w13 w24 . (Note that we have taken
even number of electrons in the system, all of these are paired a slightly different anharmonic ratio x than in Nayak and
43
Wilczek, 1996 in order to make (96) more compact than Eqs. where M must be odd for electrons, S means the symmetriza-
(7.17), (7.18) of Nayak and Wilczek, 1996.) In this expres- tion over all permutations, and
sion,
Y χr,s = (z3r+1 − z3s+1 )(z3r+1 − z3s+2 ) ×
Ψ(13)(24) = (zj − zk )2 ×
(z3r+2 − z3s+2 )(z3r+2 − z3s+3 ) . . . ×
j<k
(z3r+3 − z3s+3 )(z3r+3 − z3s+1 ) (100)
(zj − w1 )(zj − w3 )(zk − w2 )(zk − w4 ) + (j ↔ k)
Pf
zj − zk With the electron operator in hand, we can determine the
(97) primary fields of the theory. The primary field of minimum
and electrical charge is ψqh = σ1 eiφ/3α To see that this field is
Y local with respect to ψe (i.e., there should be no branch cuts
Ψ(14)(23) = (zj − zk )2 × for the electron coordinates zi ), observe that σ1 (w)ψ1 (z) ∼
j<k (z − w)−1/3 ψ and eiφ/3α (w)eiαφ (z) ∼ (z − w)1/3 . Con-
structing the full wavefunction (as in Eq. 85 and analogous to
(zj − w1 )(zj − w4 )(zk − w2 )(zk − w3 ) + (j ↔ k)
Pf Eq. 93) the fusion of of eiφ/3α (from ψqh ) with eiαφ (from
zj − zk Q
(98) ψe ) again generates a factor of i (zi − w)1/3 . We conclude
that the elementary quasihole has charge Q = +eν/3.
Suppose, now, that the system is in the state Ψ(1) . Braid-
ing 1 around 2 or 3 around 4 simply gives a phase (which is
R1σσ multiplied by a contribution from the Abelian part of the 1 ψ1 ψ2
theory). However, if we take w2 around w3 , then after the (a) @
R @
@ @ @
R @
R
braiding, the system will be in the state Ψ(ψ) as a result of the σ2 σ1 ǫ
branch cuts in (96). Now, 1 and 2 will instead fuse together @
R @
@ R @
@ R
@ ...
σ1 ǫ σ2 σ1
to form ψ, as expected from the general argument in Eq. A6. @
R @
@ R @
@ R @
@ R
@
Thus, the braiding yields a rotation in the degenerate space. 1 ψ1 ψ2 1
The resulting prediction for the behavior under braiding for
the Moore-Read Pfaffian state is in agreement with the results (b) τ - 1 - τ -1
...
obtained in sections III.B and III.C above. @R @
@ R
@
1- τ - 1 - τ - 1 - τ
(e) Z3 Read-Rezayi State (Briefly): We can follow a com-
pletely analogous procedure with a CFT which is the tensor FIG. 9 (a) Bratteli diagram for fusion of multiple σ1 fields in the Z3
product of the Z3 parafermion CFT with a chiral boson. As Parafermion CFT. (b) Bratteli diagram for Fibonacci anyons.
before, the electron operator is a product of a chiral vertex
operator from the bosonic theory with an operator from the The general braiding behavior for the Z3 parafermions has
parafermion theory. The simplest choice is ψe = ψ1 eiαφ . been worked out in Slingerland and Bais, 2001. It is trivial,
We would like this field to be fermionic so that it can be however, to work out the dimension of the degenerate space
an electron creation operator by which we can extend the by examining the Bratteli diagram Fig. 9a (See the appendix
chiral algebra (i.e., so that the electron wavefunction has no for explanation of this diagram). For example, if the number
branch cuts or singularities). (See appendix A for the nota- of electrons is a multiple of 3 then they fuse together to form
tion for parafermion fields.) The fusion rules for ψ1 in the Z3 the identity. Then, for example, with 6 quasiholes one has 5
parafermion CFT are: ψ1 × ψ1 ∼ ψ2 but ψ1 × ψ1 × ψ1 ∼ 1 so paths of length 5 ending at 1 (hence a 5 dimensional degener-
that the correlator in Eq. 84 is only nonzero if N is divisible by ate space). However, if, for example, the number of electrons
3. From the OPE, we obtain ψ1 (z1 )ψ1 (z2 ) ∼ (z1 −z2 )−2/3 ψ2 is 1 mod 3, then the electrons fuse in threes to form 1 but there
so in order is one ψ1 left over. Thus, the quasiholes must fuse together to
p to have the wavefunction analytic, we must choose form ψ2 which can fuse with the leftover ψ1 to form 1. In this
α = m + 2/3 with m ≥ 0 an integer (m odd results
case, for example, with 4 quasiholes there is a 2 dimensional
in an antisymmetric wavefunction and even results in sym-
space. It is easy to see that (if the number of electrons is divis-
metric). The filling fraction in the thermodynamic limit is
ible by 3) the number of blocks with n quasiparticles is given
determined entirely by the vertex operator eiαφ , resulting in
by the n − 1st Fibonacci number, notated Fib(n-1) defined by
ν = 1/α2 = 1/(m + 2/3).
Fib(1) = Fib(2) = 1 and Fib(n) = Fib(n − 1) + Fib(n − 2)
The ground state wavefunction for N = 3n electrons takes
for n > 2.
the form:
Y m
Ψgs (z1 , . . . , z3n ) = (zi − zj ) ×
i<j E. Edge Excitations
Y When a system in a chiral topological phase has a bound-
S χr,s (z3r+1 , . . . , z3r+k , z3k+1 , . . . z3s+3 ) ary (as it must in any experiment), there must be gapless ex-
0≤r<s<n
citations at the boundary (Halperin, 1982; Wen, 1992). To
(99) see this, consider the Chern-Simons action on a manifold M
44
State CFT ν ψe ψqh boundary condition. This boundary condition is not unique,
√ √
Laughlin Boson 1
e iφ m iφ/
e m however. The topological order of the bulk state does not
m √ √
iφ m+1 determine the boundary condition. It is determined by the
Moore-Read Ising 1
ψe σeiφ/(2 m+1)
m+1 √ √ physical properties of the edge. Consider, for instance, the
Z3 RR Z3 Paraf. 1
ψ1 eiφ m+2/3
σ1 eiφ/(3 m+2/3) a a
m+2/3 alternative boundary condition a0 + va1 |∂M = 0 for some
constant v with dimensions of velocity. With this bound-
TABLE I Summary of CFT-wavefunction correspondences dis- ary condition, the quadratic term in the Lagrangian will now
cussed here. In all cases m ≥ 0. Odd (even) m represents a Fermi be tr (∂0 + v∂1 ) U −1 ∂1 U and the edge theory is the chiral
(Bose) wavefunction. WZW model with non-zero velocity.
It is beyond the scope of this paper to discuss the chiral
with boundary ∂M (Elitzur et al., 1989; Witten, 1989), Eq. WZW model in any detail (for more details, see Gepner and
24. The change in the action under a gauge transformation, Qiu, 1987; Gepner and Witten, 1986; Knizhnik and Zamolod-
aµ → gaµ g −1 + g∂µ g −1 , is: chikov, 1984 ). However, there are a few key properties which
Z we will list now. The chiral WZW model is a conformal field
k
SCS [a] → SCS [a] + tr(g −1 dg ∧ a) (101) theory. Therefore, although there is a gap to all excitations in
4π ∂M the bulk, there are gapless excitations at the edge of the sys-
In order for the action to be invariant, we fix the boundary tem. The spectrum of the WZW model is organized into rep-
condition so that the second term on the r.h.s. vanishes. For resentations of the Virasoro algebra and is further organized
a into representations of the Gk Kac-Moody algebra. For the
instance, we could take boundary condition a0 |∂M = 0,
where x0 , x1 are coordinates on the boundary of M and x2 sake of concreteness, let us consider the case of SU(2)k . The
is the coordinate perpendicular to the boundary of M. Then SU(2)k WZW model contains primary fields φj , transforming
the action is invariant under all transformations which respect in the j = 0, 1/2, 1, . . . , k/2 representations. These corre-
this boundary condition, i.e. which satisfy ∂0 g = 0 on the spond precisely to the allowed quasiparticle species: when the
boundary. We separate these into gauge and global symme- total topological charge of all of the quasiparticles in the bulk
tries. Functions g : M → G satisfying g|∂M = 1 are the is j, the edge must be in the sector created by acting with the
gauge symmetries of the theory. (They necessarily satisfy spin j primary field on the vacuum.
∂0 g = 0 since x0 is a coordinate along the boundary.) Mean- The Gk case is a generalization of the U(1)m case, where
while, functions f : M → G which are independent of x0 are g = eiφ and the WZW model reduces to a free chiral bosonic
really global symmetries of the theory. The representations theory:
of this global symmetry form the spectrum of edge excita- Z
m
tions of the theory. (The distinction between gauge and global S= d2 x (∂t + v∂x ) φ ∂x φ (103)
4π
transformations is that a gauge transformation can leave the
t = 0 state unchanged while changing the state of the sys- (In Sec. III.A, we used k for the coefficient of an Abelian
tem at a later time t. Since it is, therefore, not possible for a Chern-Simons term; here, we use m to avoid confusion with
given initial condition to uniquely define the state of the sys- the corresponding coupling of the SU(2) Chern-Simons term
tem at a later time, all physically-observable quantities must in situations in which both gauge fields are present.) The pri-
be invariant under the gauge transformation. By contrast, a mary fields are einφ , with n = 0, 1, ..., m − 1. (The field eimφ
global symmetry, even if it acts differently at different spatial is either fermionic or bosonic for m odd or even, respecitvely,
points, cannot leave the t = 0 state unchanged while chang- so it is not a primary field, but is, rather, included as a genera-
ing the state of the system at a later time t. A global sym- tor of an extended algebra.) A quantum Hall state will always
metry does not prevent the dynamics from uniquely defining have such a term in its edge effective field theory; the U(1)
the state of the system at a later time for a given initial con- is the symmetry responsible for charge conservation and the
dition. Therefore, physically-observable quantities need not gapless chiral excitations (103) carry the quantized Hall cur-
be invariant under global transformations. Instead, the spec- rent.
trum of the theory can be divided into representations of the Therefore, we see that chiral topological phases, such as
symmetry.) fractional quantum Hall states, must have gapless chiral edge
With this boundary condition, the natural gauge choice for excitations. Furthermore, the conformal field theory which
a
the bulk is a0 = 0. We can then transform the Chern-Simons models the low-energy properties of the edge is the same con-
functional integral into the chiral WZW functional integral formal field theory which generates ground state wavefunc-
following the steps in Eqs. 81-83 (Elitzur et al., 1989): tions of the corresponding Chern-Simons action. This is clear
Z from the fact that the two derivations (Eqs. 83 and 102)
k are virtually identical. The underlying reason is that Chern-
S= tr ∂0 U −1 ∂1 U +
4π ∂M Simons theory is a topological field theory. When it is solved
Z on a manifold with boundary, it is unimportant whether the
k
ǫµνλ tr ∂µ U U −1 ∂ν U U −1 ∂λ U U −1 (102) manifold is a fixed-time spatial slice or the world-sheet of the
12π M
edge of the system. In either case, Chern-Simons theory re-
Note the off-diagonal form of the quadratic term (analogous duces to the same conformal field theory (which is an example
to the z − z̄ form in Eq. 83), which follows from our choice of of ‘holography’). One important difference, however, is that,
45
in the latter case, a physical boundary condition is imposed point, the Majorana fermions become massless. Therefore,
and there are real gapless degrees of freedom. (In the former the edge excitations are the right-moving chiral part of the
case, the CFT associated with a wavefunction for a fixed-time critical Ising model. (This is why the vortices of a p + ip su-
spatial slice may have apparent gapless degrees of freedom perconductor are call Ising anyons.) However, the edge exci-
which are an artifact of a gauge choice, as discussed in Sec- tations have non-trivial topological structure for the same rea-
tion III.D.) son that correlation functions of the spin field are non-trivial
The WZW models do not, in general, have free field rep- in the Ising model: while the fermions are free, the Ising spin
resentations. One well-known exception is the equivalence field is non-local in terms of the fermions, so its correlations
between the SU(N)1 × U(1)N chiral WZW model and N free are non-trivial. The Ising spin field σ(z) inserts a branch cut
chiral Dirac fermions. A somewhat less well-known excep- running from z = vn x + it to infinity for the fermion ψ. This
tion is the SU(2)2 chiral WZW model, which has a represen- is precisely what happens when a flux hc/2e vortex is created
tation in terms of 3 free chiral Majorana fermions. Before dis- in a p + ip superconductor.
cussing this representation, we first consider the edge excita- The primary fields of the free Majorana fermion are 1, σ,
tions of a p + ip superconductor, which supports Ising anyons and ψ with respective scaling dimensions 0, 1/16, and 1/2,
which, in turn, differ from SU(2)2 only by a U(1) factor. as discussed in Section III.D. When there is an odd number
Let us solve the Bogoliubov-de Gennes Hamiltonian (38) of flux hc/2e vortices in the bulk, the edge is in the σ(0)|0i
with a spatially-varying chemical potential, just as we did in sector. When there is an even number, the edge is in either
Section III.B. However, instead of a circular vortex, we con- the |0i or ψ(0)|0i sectors, depending on whether there is an
sider an edge at y = 0: even or odd number of fermions in the system. So long as
quasiparticles don’t go from the edge to the bulk or vice versa,
µ(y) = ∆ h(y), (104) however, the system remains in one of these sectors and all
excitations are simply free fermion excitations built on top of
with h(y) large and positive for large y, and h(y) < 0 for the ground state in the relevant sector.
y < 0; therefore, the electron density will vanish for y large However, when a quasiparticle tunnels from the edge to the
and positive. Such a potential defines an edge at y = 0. There bulk (or through the bulk), the edge goes from one sector to
are low-energy eigenstates of the BdG Hamiltonian which are another – i.e. it is acted on by a primary field. Hence, in the
spatially localized near y = 0: presence of a constriction at which vortices of fermions can
Ry ′ ′ tunnel from one edge to another, the edge Lagrangian of a
φedge
E (x) ≈ eikx e− 0 h(y )dy φ0 , (105) p + ip superconductor is (Fendley et al., 2007a):
with φ0 = 11 an eigenstate of σ x . This wavefunction de- Z
scribes a chiral wave propagating in the x−direction localized S= dτ dx (Lfermion (ψa ) + Lfermion (ψb ))
on the edge, with wave vector k = E/∆. A more complete Z Z
solution of the superconducting Hamiltonian in this situation + dτ λψ iψa ψb + dτ λσ σa σb (108)
would involve self-consistently solving the BdG equations, so
that both the density and the gap ∆(y) would vanish for large where a, b denote the two edges. (We have dropped all irrel-
positive y. The velocity of the chiral edge mode would then
evant terms, e.g descendant fields.) In other words, although
depend on how sharply h(y) varies. However, the solutions
the edge theory is a free theory in the absence of coupling to
given above with fixed constant ∆ are sufficient to show the
the bulk or to another edge through the bulk, it is perturbed by
existence of the edge mode.
primary fields when quasiparticles can tunnel to or from the
If we define an edge fermion operator ψ(x):
edge through the bulk. The topological structure of the bulk
Ry ′ ′ X constrains the edge through the spectrum of primary fields.
ψ(x) = e− 0 h(y )dy [ψk eikx φ0 + ψ−k e−ikx φ0 ]. As in the discussion of Section III.D, the edge of the Moore-
k>0
Read Pfaffian quantum Hall state is a chiral Majorana fermion
together with a free chiral boson φ for the charge sector of the
The
P fermion operators, ψk , satisfy ψ−k = ψk† , so ψ(x) = theory. As in the case of a p + ip superconductor, the primary
k ψk e
ikx
is a real Majorana field, ψ(x) = ψ † (x). The edge fields of this theory determine how the edge is perturbed by
Hamitonian is: the tunneling of quasiparticles between two edges through the
X Z bulk (Fendley et al., 2006, 2007a):
Ĥedge = vn k ψk† ψk = dx ψ(x)(−ivn ∂x )ψ(x),
Z Z
k>0
(106) S= dτ dx (Ledge (ψa , φa ) + Ledge (ψb , φb ))
where the edge velocity v = ∆. The Lagrangian density takes √
the form: + λ1/2 cos((φa (0) − φb (0))/ 2) + λψ,0 iψa ψb
√ i
Lfermion = iψ(x)(∂t + vn ∂x )ψ(x) (107) + λ1/4 σa (0)σb (0) cos((φa (0) − φb (0))/2 2) (109)
The 2D Ising model can be mapped onto the problem of The most relevant coupling is λ1/4 , so the tunneling of charge
(non-chiral) Majorana fermions on a lattice. At the critical e/4 quasiparticles dominates the transport of charge from
46
one edge to the other at the point contact. (The tunneling which is not quite the SU(2)2 WZW model.
of charge e/2 quasiparticles makes a subleading contribution
while the tunneling of neutral fermions contributes only to
thermal transport.) At low enough temperatures, this relevant F. Interferometry with Anyons
tunneling process causes the point contact to be pinched off
(Fendley et al., 2006, 2007a), but at temperatures that are not In Section II of this review we described an interference
too low, we can treat the tunneling of e/4 quasiparticles per- experiment that is designed to demonstrate the non-Abelian
turbatively and neglect other the other tunneling operators. Of statistics of quasiparticles in the ν = 5/2 state. We start this
course, the structure of the edge may be more complex than section by returning to this experiment, and using it as an exer-
the minimal structure dictated by the bulk which we have an- cise for the application of the calculational methods reviewed
alyzed here. This depends on the details of the confining po- above. We then generalize our analysis to arbitrary SU(2)k
tential defining the system boundary, but at low enough tem- non-Abelian states and also describe other experiments that
peratures, the picture described here should still apply. In- share the same goal.
teresting information about the non-Abelian character of the In the experiment that we described in Section II, a Fabry-
Moore-Read Pfaffian state can be obtained from the tempera- Perot interference device is made of a Hall bar perturbed by
ture dependence of the tunneling conductance (Fendley et al., two constrictions (see Fig. 2). The back-scattered current is
2006, 2007a) and from current noise (Bena and Nayak, 2006). measured as a function of the area of the cell enclosed by the
Finally, we return to SU(2)2 . The SU(2)2 WZW model is two constrictions and of the magnetic field. We assume that
a triplet of chiral Majorana fermions, ψ1 , ψ2 , ψ3 – i.e. three the system is at ν = 5/2 and consider interference experi-
identical copies of the chiral Ising model. This triplet is the ments which can determine if the electrons are in the Moore-
spin-1 primary field (with scaling dimension 1/2). The spin- Read Pfaffian quantum Hall state.
1/2 primary field is roughly ∼ σ1 σ2 σ3 with dimension 3/16 Generally speaking, the amplitude for back–scattering is a
(a more precise expression involves the sum of products such sum over trajectories that wind the cell ℓ times, with ℓ =
as σ1 σ2 µ3 , where µ is the Ising disorder operator dual to σ). 0, 1, 2... an integer. The partial wave that winds the cell ℓ
This is one of the primary differences between the Ising model times, winds the n quasiparticles localized inside the cell ℓ
and SU(2)2 : σ is a dimension 1/16 field, while the spin-1/2 times. From the analysis in Section III.B, if the electrons are
primary field of SU(2)2 has dimension 3/16. Another way to in the Pfaffian state, the unitary transformation that the tun-
understand the difference between the two models is that the neling quasiparticle applies on the wave function of the zero
SU(2)2 WZW model has two extra Majorana fermions. The energy modes is
pair of Majorana fermions can equally well be viewed as a
Dirac fermion or, through bosonization, as a free chiral bo- " n
#ℓ
ℓ Y
son, which has U(1) symmetry. Thus, the Ising model is often Ûn = eiαn γan γi (110)
written as SU(2)2 /U(1) to signify that the the U(1) chiral bo- i=1
son has been removed. (This notion can be made precise with
the notion of a coset conformal field theory (Di Francesco where the γi ’s are the Majorana modes of the localized bulk
et al., 1997) or by adding a U(1) gauge field to the 2D action quasiparticles, γa is the Majorana mode of the quasiparticle
and coupling it to a U(1) subgroup of the SU(2) WZW field g that flows around the cell, and αn is an Abelian phase that
(Gawȩdzki and Kupiainen, 1988; Karabali et al., 1989). The will be calculated below.
gauge field has no Maxwell term, so it serves only to elimi- The difference between the even and odd values of n, that
nate some of the degrees of freedom, namely the U(1) piece.) we described in Section II of the review, is evident from Eq.
As we discussed in subsection III.C, these differences are also (110) when we we look at the lowest order, ℓ = 1. For even n,
manifested in the bulk, where they lead to some differences Ûn is independent of γa . Thus, each tunneling quasiparticle
in the Abelian phases which result from braiding but do not applies the same unitary transformation on the ground state.
change the basic non-Abelian structure of the state. The flowing current then measures the operator Ûn (more pre-
On the other hand, the edge of the Moore-Read Pfaffian cisely, it measures the interference term, which is an hermitian
quantum Hall state is a chiral Majorana fermion together with operator. From that term the value of Ûn may be extracted).
a free chiral boson φ which carries the charged degrees of In contrast, when n is odd the operator Ûn depends on γa .
freedom. So we restore the chiral boson which we eliminated Thus, a different unitary operation is applied by every incom-
in passing from SU(2)2 to the Ising model, with one important ing quasiparticle. Moreover, the different unitary operators
difference. The compactification radius R (i.e., the theory is do not commute, and share no eigenvectors. Thus, their ex-
invariant under φ → φ + 2πR) of the charged boson need not pectation values average to zero, and no interference is to be
be the same as that of the boson which was removed by coset- observed. This analysis holds in fact for all odd values of ℓ.
ting. For the special case of bosons at ν = 1, the boson is, in The phase αn is composed of two parts. First, the quasi-
fact, at the right radius. Therefore, the charge boson can be particle accumulates an Aharonov-Bohm phase of 2πe∗ Φ/hc,
fermionized so that there is a triplet of Majorana fermions. In where e∗ = e/4 is the quasiparticle charge for ν = 5/2 and Φ
this case, the edge theory is the SU(2)2 WZW model (Fradkin is the flux enclosed. And second, the tunneling quasiparticle
et al., 1998). In the case of electrons at ν = 2 + 1/2, the chiral accumulates a phase as a consequence of its interaction with
boson is not at this radius, so the edge theory is U(1)2 ×Ising, the n localized quasiparticles. When a charge e/4 object goes
47
as depicted in figure 11. This difference allows us to read out by both the bulk and the edge. However, the edge theory also
the value of a topologically-protected qubit (Das Sarma et al., enables one to determine the temperature and voltage depen-
2005). dences of I0 , I1 , ... in (111), (116) (Ardonne and Kim, 2007;
What happens if the qubit is in a superposition of j = 0 and Bishara and Nayak, 2007; Fidkowski, 2007). As is discussed
j = 1? The interference measurement causes the tunneling in these papers, at finite temperature, interference will not be
quasiparticles to become entangled with the bulk quasiparti- visible if the two point contacts are furtherapart than the
ther-
cle (Bonderson et al., 2007; Freedman et al., 2006; Overbosch −1 1/8 1/8
mal length scale Lφ , where Lφ = kB T vc + vn , if the
and Bais, 2001). When the integrated current is large enough
charged and neutral mode velocities are vc , vn . Another im-
that many quasiparticles have tunneled and equilibrated at the
portant feature is that the interference term (when it is non-
current leads, the j = 0 and j = 1 possibilities will have deco-
vanishing) is oscillatory in the source-drain voltage while the
hered. The measurement will see one of the two possibilities
I0 term has a simple power law dependence.
with corresponding probabilities.
The assumption that the edge and the bulk are well sepa-
rated is crucial to that above calculations of interference, but
0 1 1 in practice this may not be the case. When there is bulk-
= = (−1) edge tunneling one might imagine that a quasiparticle mov-
ing along the edge may tunnel into the bulk for a moment and
thereby evade encircling some of the localized quasiparticles
(a) (b) thus smearing out any interference pattern. The first theoreti-
cal steps to analysing this situation have been taken in (Over-
FIG. 11 We can obtain the result of taking a j = 1/2 quasiparticle bosch and Wen, 2007; Rosenow et al., 2007) where tunnling
around a qubit from the two diagrams in this figure. In (a) the qubit is to a single impurity is considered. Surprisingly it is found
in the state 0, while in (b) it is in state 1. These figures are similar to that the interfernece pattern is full strength both in the strong
the left-hand-side of Fig. 10, but with the loop on the right replaced
tunneling limit as well as in the weak tunneling limit.
by a loop with (a) j = 0 or (b) j = 1.
While the experiment we described for the ν = 5/2 state
does not require a precise determination of n, as it is only
The experiment that we analyzed above for ν = 5/2 may
its parity that determines the amplitude of the interference
be analyzed also for other non-Abelian states. The com-
pattern, it does require that the number n does not fluctu-
putation using knot invariants can be immediately√adapted
ate within the duration of the experiment. Generally, fluc-
to other SU(2)k states by simply replacing d = 2 with
tuations in n would be suppressed by low temperature, large
d = 2 cos π/(k + 2). We should calculate the value of the
charging energy and diminished tunnel coupling between the
Hopf link as in figures 10 and 11, with one of the loops corre-
bulk and the edge. However, when their suppression is not
sponding to the tunneling quasiparticle and the other loop cor-
strong enough, and n fluctuates over a range much larger than
responding to the total topological charge of the bulk quasi-
1 within the time of the measurement, two signatures of the
particles. The result can be written in the more general form
non-Abelian statistics of the quasiparticles would still survive,
(Bonderson et al., 2006b):
at least as long as the characteristic time scale of these fluctu-
Ibs (a) = I0 + I1 |Mab | cos(β + θab ) (118) ations is much longer than the time between back–scattering
events. First, any change in n would translate to a change in
where Mab is defined in terms of the S-matrix: the back–scattered current, or the two-terminal conductance
Sab S11 of the device. Hence, fluctuations in n would introduce cur-
Mab = (119)
S1a S1b rent noise of the telegraph type, with a unique frequency de-
pendence (Grosfeld et al., 2006). Second, fluctuations in n
and Mab = |Mab | eiθab . The expression (118) gives the cur- would suppress all terms in Eq. (111) other than those where
rent to due to a quasiparticles if the quasiparticles in the bulk m = 4k with k an integer. Thus, the back–scattered current
fuse to b. If the contribution of j = 1/2 quasiparticles domi- will have a periodicity of one flux quatum Φ0 , and the visi-
nates, as in the ν = 5/2 case, then we should set a = 21 in this bility of the flux oscillations, for weak back–scattering, would
expression. For the level k = 3 case, taking a = 12 , |Mab | = 1 be II04 ∝ I03 .
for b = 0, 32 while |Mab | = φ−2 for b = 21 , 1, where φ is the A similar relation holds also for another type of interfer-
√
golden mean, φ = (1 + 5)/2. (In Z3 parafermion language, ence experiment, in which the interferometer is of the Mach-
b = 0, 32 correspond to the fields 1, ψ1,2 while b = 21 , 1 corre- Zehnder type, rather than the Fabry-Perot type. (A Mach-
spond to the fields σ1,2 , ε.) Zehnder interferometer has already been constructed in the
Finally, we can analyze the operation of an interferometer integer quantum Hall regime (Ji et al., 2003)). If we are
using the edge theory (109). The preceding discussion esen- to describe the Mach-Zehnder interferometer in a language
tially assumed that the current is carried by non-interacting close to that we used for the Fabry-Perot one, we would note
anyonic quasiparticles. However, the edge is gapless and, the following important differences: first, no multiple back–
in general, does not even have well-defined quasiparticles. scattering events are allowed. And second, since the area en-
Therefore, a computation using the edge theory is more com- closed by the interfering partial waves now encompasses the
plete. The expected results are recovered since they are deter- inner edge, the quantum state of the encircled area changes
mined by the topological structure of the state, which is shared with each tunneling quasiparticle. Thus, it is not surprising
49
that the outcome of an interference experiment in a Mach- are the natural degrees of freedom in a topological phase be-
Zehnder geometry will be very close to that of a Fabry-Perot cause the topological charge of a particle or collection of par-
experiment with strong fluctuations in n. The telegraph noise ticles can only be determined, in general, by taking a test par-
in the Fabry-Perot case(Grosfeld et al., 2006) becomes shot ticle around the particle or collection in question. Therefore,
noise in the Mach-Zehnder case. Remarkably(Feldman et al., the most direct way in which a system can settle into a topo-
2006) the effective charge extracted from that noise carries a logical phase is if the microscopic degrees of freedom orga-
signature of the non-Abelian statistics: as the flux is varied, nize themselves so that the low-energy degrees of freedom are
the charge changes from e/4 to about 3e. loops or, as we will see below, string nets (in which we allow
Other than interference experiments, there are several pro- vertices into which three lines can run). As we will describe
posals for experiments that probe certain aspects of the more fully below, the Hilbert space of a non-chiral topologi-
physics of non-Abelian states. The degeneracy of the ground cal phase can be described very roughly as a ‘Fock space for
state in the presence of vortices may be probed(Grosfeld and loops’ (Freedman et al., 2004). Wilson loop operators are es-
Stern, 2006) by the consequences of its removal: when the fill- sentially creation/annihilation operators for loops. The Hilbert
ing factor is ν = 5/2 + ǫ with ǫ ≪ 1, quasiparticles are intro- space is spanned by basis states which can be built up by act-
duced into the bulk of the system, with a density proportional ing with Wilson loop operators on the state with no loops, i.e.
to ǫ. For a clean enough sample, and a low enough density, |γ1 ∪ . . . ∪ γn i = W [γn ] . . . W [γ1 ]|∅i is (vaguely) analgous
the quasiparticles form a lattice. In that lattice, the Majorana to |k1 , . . . , k2 i ≡ a†kn . . . a†k1 |0i. An important difference is
zero modes of the different quasiparticles couple by tunneling, that the states in the topological theory must satisfy some ex-
and the degeneracy of the ground states is removed. The sub- tra constraints in order to correctly represent the algebra of
space of multiply-degenerate ground states is then replaced by the operators W [γ]. If we write an arbitrary state |Ψi in the
a band of excitations. The neutrality of the Majorana modes is basis given above, Ψ[γ1 ∪ . . . ∪ γn ] = hΨ|γ1 ∪ . . . ∪ γn i,
removed too, and the excitations carry a charge that is propor- then the ground state(s) of the theory are linearly independent
tional to their energy. This charge makes these modes weakly Ψ[γ1 ∪ . . . ∪ γn ] satisfying some constraints.
coupled to an externally applied electric field, and provides a In fact, we have already seen an example of this in section
unique mechanism for a dissipation of energy, with a charac- II.D: Kitaev’s toric code model (18). We now represent the
teristic dependence on the wave vector and frequency of the solution in a way which makes the emergence of loops clear.
electric field. Since the tunnel coupling between neighboring We color every link of the lattice on which the spin points up.
quasiparticles depends exponentially on their separation, this Then, the first term in (18) requires that there be an even num-
mechanism will be exponentially sensitive to the distance of ber of colored links emerging from each site on the lattice. In
the filling factor from 5/2 (Grosfeld and Stern, 2006). other words, the colored links form loops which never termi-
nate. On the square lattice, loops can cross, but they cannot
cross on the honeycomb lattice; for this reason, we will often
G. Lattice Models with P, T -Invariant Topological Phases find it more convenient to work on the honeycomb lattice. The
second term in the Hamiltonian requires that the ground state
Our discussion of topological phases has revolved around satisfy three further properties: the amplitude for two config-
fractional quantum Hall states because these are the only ones urations is the same if one configuration can be transformed
known to occur in nature (although two dimensional 3 He-A into another simply by (1) deforming some loop without cut-
(Leggett, 1975; Volovik, 1994) and Sr2 RuO4 may join this list ting it, (2) removing a loop which runs around a single plaque-
(Kidwingira et al., 2006; Xia et al., 2006)). However, there is tte of the lattice, or (3) cutting open two loops which approach
nothing inherent in the definition of a topological phase which each other within a lattice spacing and rejoining them into a
consigns it to the regime of high magnetic fields and low tem- single loop (or vice-versa), which is called surgery. A vertex
peratures. Indeed, highly idealized models of frustrated mag- at which the first term in the Hamiltonian is not satisfied is an
nets also show such phases, as we have discussed in section excitation, as is a plaquette at which the second term is not
II.D. Of course, it is an open question whether these models satisfied. The first type of excitation acquires a −1 when it is
have anything to do with any real electronic materials or their taken around the second.
analogs with cold atoms in optical lattices, i.e. whether the The toric code is associated with the low-energy physics
idealized models can be adiabatically connected to more real- of the deconfined phase of Z2 gauge theory (Fradkin and
istic models. In this section, we do not attempt to answer this Shenker, 1975; Kogut, 1979); see also Senthil and Fisher,
question but focus, instead, on understanding how these mod- 2000 for an application to strongly-correlated electron sys-
els of topological phases can be solved. As we will see, their tems). This low-energy physics can be described by an
solubility lies in their incorporation of the basic topological Abelian BF-theory (Hansson et al., 2004):
structure of the corresponding phases. Z
1
One way in which a topological phase can emerge from S = eµ ǫµνλ ∂ν aλ
π
some microscopic model of interacting electrons, spins, or = SCS a + 21 e − SCS a − 12 e (120)
cold atoms is if the low-lying degrees of freedom of the mi-
croscopic model can be mapped to the degrees of freedom of eµ is usually denoted bµ and ǫµνλ ∂ν aλ = 21 ǫµνλ fνλ , hence
the topological phase in question. As we have seen in section the name. Note that this theory is non-chiral. Under a com-
III.C, these degrees of freedom are Wilson loops (59). Loops bined parity and time-reversal transformation, eµ must change
50
sign, and the action is invariant. This is important since it en- nient regularization of loops in the continuum, essentially be-
ables the fluctuating loops described above to represent the cause when d is large, the lattice fills up with loops which then
Wilson loops of the gauge field aµ . In a chiral theory, it is have no freedom to fluctuate (Freedman et al., 2004). Instead,
not clear how to do this since a1 and a2 do not commute with trivalent graphs on the lattice prove to be a better way of pro-
each other. They cannot both be diagonalized; we must arbi- ceeding (and, in the case of SU(3)k and other gauge groups,
trarily choose one direction in which Wilson loops are diago- trivalent graphs are essential (Kuperberg, 1996; Turaev and
nal operators. It is not clear how this will emerge from some Viro, 1992)). The most convenient lattice is the honeycomb
microscopic model, where we would expect that loops would lattice, since each vertex is trivalent. A trivalent graph is sim-
not have a preferred direction, as we saw above in the toric ply a subset of the links of the honeycomb lattice such that no
code. Therefore, we focus on non-chiral phases, in particular, vertex has only a single link from the subset emanating from
the SU(2)k analog of (120) (Cattaneo et al., 1995): it. Zero, two, or three links can emanate from a vertex, cor-
responding to vertices which are not visited by the trivalent
k k
S = SCS Z(a +e) − SCS (a − e) graph, vertices through which a curve passes, and vertices at
k 1 which three curves meet. We will penalize energetically ver-
= tr e ∧ f + e ∧ e ∧ e (121)
4π 3 tices from which a single colored link emanates. The ground
state will not contain such vertices, which will be quasipar-
We will call this theory doubled SU(2)k Chern-Simons theory
ticle excitations. Therefore, the ground state Ψ[Γ] assigns a
(Freedman et al., 2004). complex amplitude to a trivalent graph Γ.
We would like a microscopic lattice model whose low-
energy Hilbert space is composed of wavefunctions Ψ[γ1 ∪ Such a structure arises in a manner analogous to the loop
. . . ∪ γn ] which assign a complex amplitude to a given con- structure of the toric code: if we had spins on the links of
figuration of loops. The model must differ from the toric code the honeycomb lattice, then an appropriate choice of interac-
in the constraints which it imposes on these wavefunctions. tion at each vertex will require that colored links (on which
The corresponding constraints for (121) are essentially the the spin points up) form a trivalent graph. We note that links
rules for Wilson loops which we discussed in subsection III.C can be given a further labeling, although we will not dis-
(Freedman et al., 2004). For instance, ground state wavefunc- cuss this more complicated situation in any detail. Each col-
tions should not give the same the amplitude for two config- ored link can be assigned a j in the set 12 , 1, . . . , k2 . Uncol-
urations if one configuration can be transformed into another ored links are assigned j = 0. Rather than spin-1/2 spins
simply by removing a loop which runs around a single plaque- on each link, we should take spin-k/2 on each link, with
tte of the lattice. Instead, the amplitude for the former config- Sz = −k/2 corresponding to j = 0, Sz = −k/2 + 1
uration should be larger by a factor of d = 2 cos π/(k + 2), corresponding to j = 1/2, etc. (or perhaps, we may want
which is the value of a single unknotted Wilson loop. Mean- to consider models with rather different microscopic degrees
while, the appropriate surgery relation is not the joining of of freedom). In this case, we would further require that the
two nearby loops into a single one, but instead is the condi- links around each vertex should satisfy the branching rules
tion that when k + 1 lines come close together, the amplitudes of SU(2)k : |j1 − j2 | ≥ j3 ≤ min j1 + j2 , k2 − j1 − j2 . The
for configurations in which they are cut open and rejoined in case which we have described in the previous paragraph, with-
different ways satisfy some linear relation. This relation is es- out the additional j label could be applied to the level k = 1
sentially the requirement that the j = (k + 1)/2 Jones-Wenzl case, with colored links carrying j = 1/2 or to level k = 3,
projector should vanish within any loop configuration, as we with colored links carrying j = 1, as we will discuss further
might expect since a Wilson loop carrying the corresponding below. A trivalent graph represents a loop configuration in the
SU(2) representation should vanish. manner depicted in Figure 12a. One nice feature is that the
The basic operators in the theory are Wilson loops, W [γ], Jones-Wenzl projections are enforced on every link from the
a
of the gauge field aµ in (121) in the fundamental (j = 1/2) start, so no corresponding surgery constraint is needed.
representation of SU(2). A Wilson loop in a higher j repre- If we would like a lattice model to be in the doubled SU(2)3
sentation can be constructed by simply taking 2j copies of a universality class, which has quasiparticle excitations which
j = 1/2 Wilson loop and using the appropriate Jones-Wenzl are Fibonacci anyons, then its Hamiltonian should impose the
projector to eliminate the other representations which result following: all low-energy states should have vanishing am-
in the fusion of 2j copies of j = 1/2. If the wavefunction plitude on configurations which are are not trivalent graphs,
satisfies the constraint mentioned above, then it will vanish as defined above; and the amplitude for a configuration with
identically if acted on by a j > k/2 Wilson loop. a contractible loop should be larger than the amplitude for a
These conditions are of a topological nature, so they are configuration without this loop by a factor of d = 2 cos π5 =
√
most natural in the continuum. In constructing a lattice model φ = (1 + 5)/2 for a closed, contractible loop. These condi-
from which they emerge, we have a certain amount of free- tions can be imposed by terms in the Hamiltonian which are
dom in deciding how these conditions are realized at the lattice more complicated versions of the vertex and plaquette terms
scale. Depending on our choice of short-distance regulariza- of (18). It is furthermore necessary for the ground state wave-
tion, the model may be more of less easily solved. In some function(s) to assign the same amplitude to any two trivalent
cases, an inconvenient choice of short-distance regularization graphs which can be continuously deformed into each other.
may actually drive the system out of the desired topological However, as mentioned above, surgery is not necessary. The
phase. Loops on the lattice prove not to be the most conve- Hamiltonian takes the form (Levin and Wen, 2005b) (see also
51
is a single-qubit π/8 phase gate and a two-qubit measure- B. Fibonacci Anyons: a Simple Example which is
ment. One way to implement these extra gates is to use some Universal for Quantum Computation
non-topological operations (Bravyi, 2006). First, consider the
single-qubit phase gate. Suppose quasiparticles 1, 2, 3, 4 com- One of the simplest models of non-Abelian statistics is
prise the qubit. The states |0i and |1i correspond to 1 and 2 known as the Fibonacci anyon model, or “Golden theory”
fusing to 1 or ψ (3 and 4 must fuse to the same as 1 and 2, (Bonesteel et al., 2005; Freedman et al., 2002a; Hormozi et al.,
since the total topological charge is required to be 1). If we 2007; Preskill, 2004). In this model, there are only two fields,
bring quasiparticles 1 and 2 close together then their splitting the identity (1) as well as single nontrivial field usually called
will become appreciable. We expect it to depend on the sepa- τ which represents the non-Abelian quasiparticle. (Note there
ration r as ∆E(r) ∼ e−r∆/c , where r is the distance between is no field representing the underlying electron in this simpli-
the quasiparticles and c is some constant with dimensions of fied theory). There is a single nontrivial fusion rule in this
velocity. If we wait a time Tp before pulling the quasiparti- model
cles apart again, then we will apply the phase gate (Freedman
τ ×τ =1+τ (126)
et al., 2006) UP = diag(1, ei∆E(r) Tp ). If the time T and dis-
tance r are chosen so that ∆E(r) Tp = π/4, then up to an which results in the Bratteli diagram given in Fig. 9b. This
overall phase, we would apply the phase gate: model is particularly simple in that any cluster of quasiparti-
! cles can fuse only to 1 or τ .
e−πi/8 0 The j = 0 and j = 1 quasiparticles in SU(2)3 satisfy the fu-
Uπ/8 = (125)
0 eπi/8 sion rules of Fibonacci anyons. Therefore, if we simply omit
the j = 1/2 and j = 3/2 quasiparticles from SU(2)3 , we will
We note that, in principle, by measuring the energy when the have FIbonacci anyons. This is perfectly consistent since half-
two quasiparticles are brought together, the state of the qubit integral j will never arise from the fusions of integral js; the
can be measured. model with only integer spins can be called SO(3)2 or, some-
The other gate which we need for universal quantum com- times, ‘the even part of SU(2)3 ’. As a result of the connection
putation is the non-destructive measurement of the total topo- to SU(2)3 , sometimes 1 is called q-spin “0” and τ is called
logical charge of any four quasiparticles. This can be done q-spin “1” (see (Hormozi et al., 2007)). Z3 parafermions are
with an interference measurement. Suppose we have two equivalent to a coset theory SU(2)3 /U(1). This can be real-
qubits which are associated with quasiparticles 1, 2, 3, 4 and ized with an SU(2)3 WZW model in which the U(1) subgroup
quasiparticles 5, 6, 7, 8 and we measure the total topologi- is coupled to a gauge field (Gawȩdzki and Kupiainen, 1988;
cal charge of 3, 4, 5, 6. The interference measurement is of Karabali et al., 1989). Consequently, Z3 parafermions have
the type described in subsection II.C.3: edge currents tunnel essentially the same fusion rules as SU(2)3 ; there are some
across the bulk at two points on either side of the set of four phase differences between the two theories which show up in
quasiparticles. Depending on whether the four quasiparticles the R and F -matrices. In the Z3 parafermion theory, the field
have total topological charge 1 or ψ, the two possible trajec- ǫ which results from fusing σ1 with ψ1 satisfies the Fibonacci
tories interfere with a phase ±1. We can thereby measure the fusion rule Eq. 126, i.e., ǫ × ǫ = 1 + ǫ.
total parity of two qubits. (For more details, see Freedman As with the Z3 parafermion model described above, the di-
et al., 2006.) mension of the Hilbert space with n quasiparticles (i.e., the
Neither of these gates can be applied exactly, which means number of paths through the Bratteli diagram 9b terminating
that we are surrendering some of the protection which we have at 1) is given by the Fibonacci number Fib(n − 1), hence the
worked so hard to obtain and need some software error cor- name Fibonacci anyons. And similarly the number terminat-
rection. However, it is not necessary for the π/8 phase gate or ing at τ is Fib(n). Therefore, the quantum dimension√ of the
the two qubit measurement to be extremely accurate in order τ particle is the golden mean, dτ = φ ≡ (1 + 5)/2 (from
for error correction to work. The former needs to be accurate which the theory receives the name “golden” theory). The Fi-
to within 14% and the latter to within 38% (Bravyi, 2006). bonacci model is the simplest known non-Abelian model that
Thus, the requisite quantum error correction protocols are not is capable of universal quantum computation (Freedman et al.,
particularly stringent. 2002a). (In the next section, the proof will be described for
An alternate solution, at least in principle, involves chang- SU(2)3 , but the Fibonacci theory, which is its even part, is also
ing the topology of the manifold on which the quasiparticles universal.) It is thus useful to study this model in some detail.
live (Bravyi and Kitaev, 2001). This can be realized in a de- Many of the principles that are described here will generalize
vice by performing interference measurements in the presence to other non-Abelian models. We note that a detailed discus-
of moving quasiparticles (Freedman et al., 2006). sion of computing with the Fibonacci model is also given in
However, a more elegant approach is to work with a non- Hormozi et al., 2007.
Abelian topological state which supports universal topological (a) Structure of the Hilbert Space: An important feature
quantum computation through quasiparticle braiding alone. In of non-Abelian systems is the detailed structure of the Hilbert
the next subsection, we give an example of such a state and space. A given state in the space will be described by a “fusion
how quantum computation can be performed with it. In sub- path”, or “fusion tree” (See appendix A). For example, using
section IV.C, we sketch the proof that a large class of such the fusion rule (126), or examining the Bratteli diagram we
states is universal. see that when two τ particles are present, they may fuse into
54
τ τ τ
|0i = |((•, •)1 , •)τ i = τ =
A, we can write a change of basis using the F -matrix as
1 1
τ P
|(•, (•, •)i )k i = j [Fkτ τ τ ]ij |((•, •)j , •)k i (128)
τ τ τ
|1i = |((•, •)τ , •)τ i = τ τ = where i, j, k take the values of the fields 1 or τ . (This is just a
τ
τ rewriting of a special case of Fig. 23). Clearly from Eq. 127,
τ τ τ F1τ τ τ is trivially unity. However, the two-by-two matrix Fττ τ τ
|N i = |((•, •)τ , •)1 i = = is nontrivial
τ 1 τ
! p !
1 F F φ −1
φ −1
[Fττ τ τ ] = 11 1τ
= p −1 (129)
FIG. 14 The three possible states of three Fibonacci particles, shown Fτ 1 Fτ τ φ −φ−1
in several common notations. The “quantum number” of an individ-
ual particle is τ . In the parenthesis and ellipse notation (middle), Using this F matrix, one can translate between bases that de-
each particle is shown as a black dot, and each pair of parenthesis or scribe arbitrary fusion orders of many particles.
ellipse around a group of particles is labeled at the lower right with
For the Fibonacci theory (Preskill, 2004), it turns out to be
the total quantum number associated with the fusion of that group.
Analogously in the fusion tree notation (right) we group particles as easy to calculate the F -matrix using a consistency condition
described by the branching of the tree, and each line is labeled with known as the pentagon equation (Fuchs, 1992; Goméz et al.,
the quantum number corresponding to the fusion of all the particles 1996; Moore and Seiberg, 1988, 1989). This condition simply
in the branches above it. For example on the top line the two parti- says that one should be able to make changes of basis for four
cles on the left fuse to form 1 which then fuses with the remaining particles in several possible ways and get the same result in
particle on the right to form τ . As discussed below in section IV.B.c, the end. As an example, let us consider
three Fibonacci particles can be used to represent a qubit. The three
possible states are labeled (far left) as the logical |0i, |1i and |N i |(•, (•, (•, •)1 )τ )1 i = |((•, •)1 , (•, •)1 )1 i
(noncomputational) of the qubit.
= |((•, •)1 , •)τ , •)1 i (130)
each other, we can never change |(•, •)1 i to |(•, •)τ i. To qubit Hilbert space. However, finding such braids is now a
remedy this problem, it is convenient to use three quasipar- much more formidable task. The full Hilbert space for six Fi-
ticles to represent a qubit as suggested by Freedman et al., bonacci particles (constituting two qubits) is now 13 dimen-
2002a (many other schemes for encoding qubits are also pos- sional, and searching for a desired result in such a high di-
sible (Freedman et al., 2002a; Hormozi et al., 2007)). Thus, mensional space is extremely hard even for a powerful classi-
we represent the two states of the qubit as the |0i and |1i states cal computer. Therefore, the problem needs to be tackled by
shown in Fig. 14. The additional state |N i is a “noncompu- divide-and-conquer approaches, building up two-qubit gates
tational” state. In other words, we arrange so that at the be- out of simple braids on three particles (Bonesteel et al., 2005;
ginning and end of our computations, there is no amplitude in Hormozi et al., 2007). A simple example of such a construc-
this state. Any amplitude that ends up in this state is known as tion is sketched in Fig. 16. First, in Fig. 16.a, we consider
“leakage error”. We note, however, that the braiding matrices braids on three strands that moves (“weaves” (Simon et al.,
ρ(σ1 ) and ρ(σ2 ) are block diagonal and therefore never mix 2006)) only a single particle (the blue particle in the figure)
the noncomputational state |N i with the computational space through two stationary particles (the green particles). We
|0i and |1i (This is just another way to say that the overall search for such a braid whose action on the Hilbert space is
quantum number of the three particles must remain unchanged equivalent to exchanging the two green particles twice. Since
under any amount of braiding). Therefore, braiding the three this is now just a three particle problem, finding such a braid,
particles gives us a way to do single qubit operations with no to arbitrary accuracy, is computationally tractable. Next, for
leakage. the two qubit problem, we label one qubit the control (blue
In section IV.C, we will describe a proof that the set of in Fig. 16.b) and another qubit the target (green). We take a
braids has a “dense image” in the set of unitary operations for pair of particles from the control qubit (the control pair) and
the Fibonacci theory. This means that there exists a braid that weave them as a group through two of the particles in the tar-
corresponds to a unitary operation arbitrarily close to any de- get, using the same braid we just found for the three particle
sired operation. The closer one wants to approximate the de- problem. Now, if the quantum number of the control pair is 1
sired unitary operation, the longer the braid typically needs to (i.e, control qubit is in state |0i) then any amount of braiding
be, although only logarithmically so (i.e, the necessary braid of this pair will necessarily give just an Abelian phase (since
length grows only as the log of the allowed error distance moving 1 around is like moving nothing around). However,
to the target operation). The problem of actually finding the if the quantum number of the control pair is τ (i.e, the con-
braids that correspond to desired unitary operations, while ap- trol qubit is in state |1i) then we can think of this pair as be-
parently complicated, turns out to be straightforward (Bones- ing equivalent to a single τ particle, and we will cause the
teel et al., 2005; Hormozi et al., 2007). One simple approach same nontrivial rotation as in Fig.16.a above (Crucially, this
is to implement a brute force search on a (classical) computer is designed to not allow any leakage error!). Thus, we have
to examine all possible braids (on three strands) up to some constructed a “controlled rotation” gate, where the state of the
certain length, looking for a braid that happen to generate a target qubit is changed only if the control qubit is in state |1i,
unitary operation very close to some desired result. While this where the rotation that occurs is equivalent to exchanging two
approach works very well for searching short braids (Bones- particles of the target qubit as shown in Fig. 16.b. The result-
teel et al., 2005; Hormozi et al., 2007), the job of searching ing two-qubit controlled gate, along with single qubit rota-
all braids grows exponentially in the length of the braid, mak- tions, makes a universal set for quantum computation (Brem-
ing this scheme unfeasible if one requires high accuracy long ner et al., 2002). More conventional two-qubit gates, such as
braids. Fortunately, there is an iterative algorithm by Solovay the controlled NOT gates (CNOT), have also been designed
and Kitaev (see Nielsen and Chuang, 2000) which allows one using braids (Bonesteel et al., 2005; Hormozi et al., 2007).
to put together many short braids to efficiently construct a long (d) Other theories: The Fibonacci theory is a particularly
braid arbitrarily close to any desired target unitary operation. interesting theory to study, not only because of its simplicity,
While this algorithm does not generally find the shortest braid but also because of its close relationship (see the discussion at
for performing some operation (within some allowed error), the beginning of section IV.B) with the Z3 parafermion theory
it does find a braid which is only polylogarithmically long in — a theory thought to actually describe (Rezayi and Read,
the allowed error distance to the desired operation. Further- 2006) the observed quantum Hall state at ν = 12/5 (Xia
more, the (classical) algorithm for finding such a braid is only et al., 2004). It is not hard to show that a given braid will
algebraically hard in the length of the braid. perform the same quantum computation in either theory (Hor-
Having solved the single qubit problem, let us now imag- mozi et al., 2007) (up to an irrelevant overall Abelian phase).
ine we have multiple qubits, each encoded with three parti- Therefore, the Fibonacci theory and the associated braiding
cles. To perform universal quantum computation, in addi- may be physically relevant for fractional quantum Hall topo-
tion to being able to perform single qubit operations, we must logical quantum computation in high-mobility 2D semicon-
also be able to perform two-qubit entangling gates (Bremner ductor structures.
et al., 2002; Nielsen and Chuang, 2000). Such two-qubit gates However, there are many other non-Abelian theories, which
will necessarily involve braiding together (physically “entan- are not related to Fibonacci anyons. Nonetheless, for arbitrary
gling”!) the particles from two different qubits. The result non-Abelian theories, many of the themes we have discussed
of Freedman et al., 2002a generally guarantees that braids ex- in this section continue to apply. In all cases, the Hilbert space
ist corresponding to any desired unitary operation on a two- can be understood via fusion rules and an F -matrix; rotations
57
Lie group G is of type BCD and braid generators have three Definition IV.1. Let G be a compact Lie group and V a faith-
eigenvalues, corresponding to the BMW algebra and the two ful, irreducible, unitary representation. The pair (G, V ) has
variable Kauffman polynomial. For JW-representations of the the two eigenvalue property (TEVP) if there exists a conju-
exceptional group at level k, the number of eigenvalues of gacy class [g] of G such that:
braid generators can be composite integers (such as 4 for G2 )
1. [g] generates a dense set in G
and this has so far blocked attempts to prove density for these
JW-representations. 2. For any g ∈ [g], g acts on V with exactly two distinct
In order to perform quantum computation with anyons, eigenvalues whose ratio is not −1.
there are many details needed to align the topological pic-
ture with the usual quantum-circuit model from computer sci- Let H be the closed image of some Jones representation
ence. First, qubits must be located in the state space Vn . Since ρ : Bn → U (Vn ). We would like to use figure 18 to assert that
Vn has no natural tensor factoring (it can have prime dimen- the fundamental representation of U (Vn ) restricted to H, call
sion) this alignment (Freedman et al., 2002a) is necessarily a it θ, has the TEVP. All braid generators σi are conjugate and,
bit inefficient9 ; some directions in Vn are discarded from the in nontrivial cases, the eigenvalue ratio is −q 6= −1. However,
computational space and so we must always guard against un- we do not yet know if the restriction is irreducible. This prob-
intended “leakage” into the discarded directions. A possible lem is solved by a series of technical lemmas in (Freedman
research project is how to adapt computation to “Fibonacci” et al., 2002a). Using TEVP, it is shown first that the further
space (see subsection IV.B) rather than attempting to find bi- restriction to the identity component H0 is isotipic and then
nary structure within Vn . A somewhat forced binary structure irreducible. This implies that H0 is reductive, so its derived
was explained in subsection IV.B in connection with encod- group [H0 , H0 ] is semi-simple and, it is argued, still satisfies
ing qubits into SU(2)3 , as it was done for level 2 in subsection the TEVP. A final (and harmless) variation on H is to pass to
IV.A. (A puzzle for readers: Suppose we write integers out as the universal cover H ′ := [H^ 0 , H0 ]. The pulled back repre-
“Fibonacci numerals”: 0 cannot follow 0, but 0 or 1 can fol- sentation θ′ still has the TEVP and we are finally in a situation,
low 1. How do you do addition and multiplication?) However, namely irreducible representations of semi-simple Lie groups
we will not dwell on these issues but instead go directly to the of bounded dimension, where we can hope to apply the clas-
essential mathematical point: How, in practice, does one tell sification of such representations (McKay and Patera, 1981)
which braid group representations are dense and which are to show that our mysterious H ′ is none other than SU (Vn ).
not, i.e. which ones are sufficient for universal topological If this is so, then it will follow that the preceding shenanigans
quantum computation and which ones need to be augmented H → H0 → [H0 , H0 ] → [H^ 0 , H0 ] did nothing (beyond the
by additional non-topological gate operations? first arrow, which may have eliminated some components of
We begin by noting that the fundamental skein relation of H on which the determinant is a nontrivial root of unity).
Jones’ theory is: In general, milking the answer (to the question of which
Jones representations are projectively dense) out of the clas-
sification requires some tricky combinatorics and rank-level
(Freedman et al., 2002b) duality. Here we will be content
q -1 q
1
(q 2
-1
q 2) with doing the easiest nontrivial case. Consider six Fibonacci
= anyons τ with total charge = 1. The associated V6 ∼ = C5 ∼=
2 qubits ⊕ non-computational C as shown:
τ τ τ
τ τ
FIG. 18 Jones skein relation. (See (73)
= +
(see (73) and the associated relation for the Kauffman bracket
dim = 5 4
(77)) This is a quadratic relation in each braid generator σi and C + C
by inspection any representation of σi will have only two dis-
3 1
tinct eigenvalues q 2 and −q 2 . It turns out to be exceedingly FIG. 19 The charge on the dotted circle can be 1 or τ providing the
qubit.
rare to have a representation of a compact Lie group H where
H is densely generated by elements σi with this eigenvalue
restriction. This facilitates the identification of the compact
In coordinates, ρ takes the braid generators (projectively) to
closure H = image(ρ) among the various compact subgroups these operators:
of U (Vn ).
−1
q
2πi
q = e− 5
9
σ1 7−→ −1 ,
Actually, current schemes use approximately half the theoretical number
of qubits. One finds αlog2 (dimV
√ n
) computational qubits in Vn , for α = q
1+ 5
(log2 τ 3 )−1 ≈ 0.48, φ = 2
. q
59
metic as done on the abacus, the quantum computer will com- tion of the form
pute in superposition. We have some knowledge about what
this will allow us to do. Select mathematical problems (fac- hJµ (q)Jν (−q)i = C ǫµνλ qλ (136)
toring, finding units in number fields, searching) have efficient
where C is a topological invariant. If the topological phase
solutions in the quantum model. Many others may succumb
to quantum heuristics (e.g. adiabatic computation (Farhi et al., does not break P and T or if there is no conserved current
in the low-energy effective field theory, then there will not be
2000)) but we will not know until we can play with real quan-
tum computers. Some physical problems, such as maximizing such a dramatic signature. However, even in the quantum Hall
context, in which we have a leg up thanks to the Hall conduc-
Tc within a class of superconductors, should be advanced by
quantum computers, even though, viewed as math problems, tance, it is still a subtle matter to determine which topological
phase the system is in.
they lie even outside class NP (i.e. they are very hard). A con-
jectural view of relative computational complexity is shown As we have described, we used theoretical input to focus
our attention on the ν = 5/2 and ν = 12/5 states. With-
in Fig. 21.
out such input, the available phase space is simply too large
and the signatures of a topological phase are too subtle. One
benefit of having a particular theoretical model of a topologi-
cal phase is that experiments can be done to verify other (i.e.
poly−time non-topological) aspects of the model. By corroborating the
classically poly−time model in this way, we can gain indirect evidence about the
NP nature of the topological phase. In the case of the ν = 5/2
quantum
state, the Pfaffian model wavefunction (Greiter et al., 1992;
Moore and Read, 1991) for this state is fully spin-polarized.
Therefore, measuring the spin polarization at ν = 5/2 would
confirm this aspect of the model, thereby strengthening our
travelling salesman
maximizing T c belief in the the model as a whole – including its topologi-
multiplying
problem matrices primality cal features (see Tracy et al., 2007 for such a measurement at
factoring testing ν = 1/2). In the case of Sr2 RuO4 , the p + ip BCS model
predicts a non-zero Kerr rotation (Xia et al., 2006). This is
FIG. 21 A conjectural view of relative computational complexity not a topological invariant, but when it is non-zero and the su-
perconducting order parameter is known to be a spin-triplet,
we can infer a non-zero spin quantum Hall effect (which is a
But, before we can enter this quantum computing paradise, topological invariant but is much more difficult to measure).
there are fundamental issues of physics to be tackled. The first Thus, non-topological measurements can teach us a great deal
problem is to find a non-Abelian topological phase in nature. when we have a particular model in mind.
The same resistance to local perturbation that makes topologi- In frustrated magnets, one often cuts down on the com-
cal phases astonishing (and, we hope, useful) also makes them plex many-dimensional parameter space in the following way:
somewhat covert. An optimist might hope that they are abun- one focusses on systems in which there is no conventional
dant and that we are merely untutored and have trouble notic- long-range order. Although it is possible for a system to be
ing them. At present, our search is guided primarily by a pro- in a topological phase and simultaneously show conventional
cess of elimination: we have focussed our attention on those long-range order (quantum Hall ferromagnets are an exam-
systems in which the alternatives don’t occur – either quantum ple), the absence of conventional long-range order is often
Hall states for which there is no presumptive Abelian candi- used as circumstantial evidence that the ground state is ‘ex-
date or frustrated magnets which don’t order into a conven- otic’ (Coldea et al., 2003; Shimizu et al., 2003). This is a rea-
tional broken-symmetry state. What we need to do is observe sonable place to start, but in the absence of a theoretical model
some topological property of the system, e.g. create quasi- predicting a specific topological state, it is unclear whether the
particle excitations above the ground state, braid them, and ground state is expected to be topological or merely ‘exotic’
observe how the state of the system changes as a result. In in some other way (see below for a further discussion of this
order to do this, we need to be able to (1) create a specified point).
number of quasiparticles at known positions, (2) move them While theoretical models and indirect probes can help to
in a controlled way, and (3) observe their state. All of these identify strong candidates, only the direct measurement of a
are difficult, but not impossible. topological property can demonstrate that a system is in a
It is instructive to see how these difficulties are manifested topological phase. If, as in the quantum Hall effect, a sys-
in the case of quantum Hall states and other possible topologi- tem has been shown to be in a topological phase through
cal states. The existence of a topological phase in the quantum the measurement of one property (e.g. the Hall conduc-
Hall regime is signaled by the quantization of the Hall con- tance), then there is still the problem of identifying which
ductance. This is a special feature of those chiral topological topological phase. This requires the complete determination
phases in which there is a conserved current Jµ (e.g. an elec- of all of its topological properties (in principle, the quasi-
trical charge current or spin current). Topological invariance particle species, their topological spins, fusion rules, R- and
and P, T -violation permit a non-vanishing correlation func- F -matrices). Finding non-trivial quasiparticles is the first
62
step. In the quantum Hall regime, quasiparticles carry elec- this is an unimportant distinction. This experiment was per-
trical charge (generally fractional). Through capacitive mea- formed on an underdoped cuprate superconductor by Wynn
surements of quasiparticle electric charges (Goldman and Su, et al., 2001. The result was negative, implying that there isn’t
1995) or from shot noise measurements (De Picciotto et al., a topological phase in the low-doping part of the phase di-
1997; Saminadayar et al., 1997), one can measure the mini- agram of that material, but the experimental technique may
mal electric charges and infer the allowed quasiparticle elec- still prove to be a valuable way to test some other candidate
tric charges. The observation of charge e/4 quasiparticles by material in the future. It would be interesting and useful to de-
either of these methods would be an important step in char- sign analogous experiments which could exploit the possible
acterizing the ν = 5/2 state. Detecting charged quasipar- proximity of topological phases to other long-range ordered
ticles capacitatively or through noise measurements necessi- states besides superconductors.
tates gated samples: anti-dots and/or point contacts. In the Even if non-trivial quasiparticles have been found, there is
case of delicate states such as ν = 5/2, this is a challenge; we still the problem of determining their braiding properties. In
don’t want the gates to reduce the quality of the device and the quantum Hall case, we have described in Secs. II.C.3,
excessively degrade the robustness of the states. Even if this III.F how this can be done using quasiparticle tunneling and
proves not to be surmountable, it only solves the problem of interferometry experiments. This requires even more intri-
measuring charged quasiparticles; it does not directly help us cate gating. However, even these difficult experiments are the
with non-trivial neutral quasiparticles (such as those which we most concrete that we have, and they work only because these
believe exist at ν = 5/2). states are chiral and have gapless edge excitations – and, there-
Again, a particular theoretical model of the state can be ex- fore, have non-trivial DC transport properties – and because
tremely helpful. In the case of the toric code, an excited pla- charged anyons contribute directly to these transport proper-
quette or Z2 vortex (see Secs. II.D III.G) is a neutral spinless ties. Neutral quasiparticles are an even bigger challenge. Per-
excitation and, therefore, difficult to probe. However, when haps they can be probed through thermal transport or even, if
such a phase arises in models of superconductor-Mott insula- they carry spin, through spin transport.
tor transitions, Z2 vortices can be isolated by going back and As we have seen in Sec. II.C.3, abelian and non-Abelian in-
forth through a direct second-order phase transition between terference effects are qualitatively different. Indeed, the latter
a topological phase and a superconducting phase (Senthil and may actually be easier to observe in practice. It is striking that
Fisher, 2001a). Consider a superconductor in an annular ge- quasiparticle interferometry, which sounds like an application
ometry with a single half-flux quantum vortex through the of topological phases, is being studied as a basic probe of the
hole in the annulus. Now suppose that some parameter can state. The naive logical order is reversed: to see if a system
be tuned so that the system undergoes a second-order phase is in a topological phase, we are (ironically) saying “shape
transition into an insulating state which is a topological phase the system into a simple computer and if it computes as ex-
of the toric code or Z2 variety. Then the single vortex ground pected, then it must have been in the suspected phase.” This
state of the superconductor will evolve into a state with a Z2 is a charming inversion, but it should not close the door on the
vortex in the hole of the annulus. The magnetic flux will es- subject of probes. It is, however, important to pause and note
cape, but the Z2 vortex will remain. (Eventually, it will either that we now know the operational principles and methodology
quantum tunnel out of the system or, at finite temperature, be for carrying out quasiparticle braiding in a concrete physical
thermally excited out of the system. It is important to per- system. It is, therefore, possible that non-Abelian anyons will
form the experiment on shorter time scales.) If the system is be observed in the quantum Hall regime in the near future.
then taken back into the superconducting state, the Z2 vor- This is truly remarkable. It would not close the book on non-
tex will evolve back into a superconducting vortex; the flux Abelian anyons, but open a new chapter and encourage us to
must be regenerated, although its direction is arbitrary. Al- look for non-Abelian anyons elsewhere even while trying to
though Senthil and Fisher considered the case of a Z2 topolog- build a quantum computer with a quantum Hall state.
ical phase, other topological phases with direct second-order One important feature of non-Abelian anyons is that they
phase transitions into superconducting states will have a sim- generally have multiple fusion channels. These different fu-
ilar signature. On the other hand, in a non-topological phase, sion channels can be distinguished interferometrically, as dis-
there will be nothing left in the insulating phase after the flux cussed in Secs. II.C.3, III.F. This is not the only possibility.
has escaped. Therefore, when the system is taken back into In ultra-cold neutral atom systems, they can be optically de-
the superconducting phase, a vortex will not reappear. The ef- tected (Grosfeld et al., 2007; Tewari et al., 2007b) in the case
fect described above is not a feature of the topological phase of states with Ising anyons. Perhaps, in a solid, it will be pos-
alone, but depends on the existence of a second-order quan- sible to measure the force between two anyons. Since the two
tum phase transition between this topological state and a su- fusion channels will have different energies when the anyons
perconducting state. However, in the happy circumstance that are close together, there will be different forces between them
such a transition does exist between two such phases of some depending on how the anyons fuse. If an atomic force micro-
material, this experiment can definitively identify a topolog- scope can ‘grab’ an anyon in order measure this force, perhaps
ically non-trivial neutral excitation. In practice, the system it can also be used to drag one around and perform a braid.
is not tuned through a quantum phase transition but instead Thus, we see that new ideas would be extremely helpful
through a finite-temperature one; however, so long as the tem- in the search for non-Abelian topological phases. It may be
perature is much smaller than the energy gap for a Z2 vortex, the case that each physical system, e.g. FQHE, cold atoms,
63
Sr2 RuO4 films, etc. . . , may be suited to its own types of mea- ful in the way that the closely-related problem of classifying
surements, such as the ones described above and in Secs. rational conformal field theories has proved useful in under-
II.C.3, III.F, but general considerations, such as topological standing classical and quantum critical points.
entropy (Kitaev and Preskill, 2006; Levin and Wen, 2006), We refer here, as we have throughout this article, to topo-
may inform and unify these investigations. Another diffi- logical phases as we have defined them in Sec. III (and which
culty is that, as mentioned above, we are currently searching we briefly recapitulate below). There are many other possi-
for non-Abelian topological phases in those systems in which ble ‘exotic’ phases which share some characteristics of topo-
there is an absence of alternatives. It would be far better to logical phases, such as the emergence of gauge fields in their
have positive a priori reasons to look at particular systems. low-energy theories (Wen, 2004), but do not satisfy all of the
This state of affairs points to the dire need for general prin- criteria. These do not appear to be useful for quantum com-
ciples, perhaps of a mathematical nature, which will tell us putation.
when a system is likely to have a topological phase. Equiv- Finally, the three-dimensional frontier must be mentioned.
alently, can we define the necessary conditions for the ex- Most theory (and experiment) pertains to 2D or quasi-2D sys-
istence of a topological phase with non-Abelian quasiparti- tems. In 3+1-dimensions, even the underlying mathematical
cle statistics? For contrast, consider the case of magnetism. structure of TQFTs is quite open. Little is known beyond
Although there is a great deal which we don’t know about finite group gauge theories. For example, we do not know
magnetism, we do know that we need solids containing ions if quantum information can (in the thermodynamic limit) be
with partially filled d or f shells. Depending on the effec- permanently stored at finite temperature in any 3-dimensional
tive Coulomb interaction within these orbitals and their filling system. (By Dennis et al., 2002, this is possible in 4+1-
fractions, we understand how various mechanisms such as ex- dimensions, not possible in 2+1-dimensions, and is an open
change and superexchange can lead to effective spin-spin in- question in 3+1-dimensions.) The case of 2+1-dimensions
teractions which, in turn, can lead to ferromagnetism, antifer- has been the playground of anyons for 30 years. Will loop-
romagnetism, spin-density-waves, etc.. We need a compara- like “particles” in 3+1-dimensions be as rich a story 30 years
ble understanding of topological phases. One direction, which from now?
we have described in Sec. III.G, is to analyze models in which Perhaps it is fitting to end this review with a succinct state-
the interactions encode some combinatorial relations, such as ment of the definition of a topological phase: the ground state
those associated with string nets or loop gases (Fendley, 2007; in the presence of multiple quasiparticles or in a non-trivial
Fendley and Fradkin, 2005; Fidkowski et al., 2006; Freedman topology has a stable degeneracy which is immune to weak
et al., 2005a; Levin and Wen, 2005b). However, we only have (but finite) local perturbations. Note that the existence of
a few examples of microscopic interactions which give rise an excitation gap is not needed as a part of this definition
to these intermediate scale structures. We sorely need more although, as should be obvious by this point, the stability
general guidelines which would enable us to look at a given of the ground state degeneracy to local perturbations almost
Hamiltonian and determine if it is likely to have a non-Abelian always necessitates the existence of an excitation gap. We
topological phase; a more detailed analysis or experimental make three comments about this definition before conclud-
study could then be carried out. This is a particularly im- ing: (1) incompressible FQH states satisfy our definition and
portant direction for future research because, although nature they are, so far, the only experimentally-established topolog-
has given us the quantum Hall regime as a promising hunting ical phases. (2) The existence of a topological phase does
ground for topological phases, the energy scales are very low. not, by itself, enable topological quantum computation – one
A topological phase in a transition metal oxide might have a needs quasiparticles with non-Abelian braiding statistics, and
much larger gap and, therefore, be much more robust. for universal topological quantum computation, these quasi-
An important problem on the mathematical side is a com- particles’ topological properties must belong to a class which
plete classification of topological phases. In this review, we includes SU(2)k , with k = 3, 5, 6, 7, 8, 9, . . ., as we have dis-
have focussed on a few examples of topological phases: those cussed extensively in this article. (3) Possible non-Abelian
associated with SU(2)k Chern-Simons theory, especially the quantum Hall states, such as ν = 5/2 and 12/5 are the first
k = 2, 3 cases. These are part of a more general class as- among several possible candidates, including Sr2 RuO4 , which
sociated with an arbitrary semi-simple Lie group G at level has recently been shown to be a chiral p-wave superconductor
k. Another class is associated with discrete groups, such as (Kidwingira et al., 2006; Xia et al., 2006), and p-wave paired
phases whose effective field theories are lattice gauge theo- cold atom superfluids.
ries with discrete gauge group. New topological phases can Note added in proof: A measurement of the charge of a
be obtained from both of these by coset constructions and/or quasi-particle in a ν = 5/2 fractional quantum Hall state has
tensoring together different effective field theories. However, been recently reported by Dolev et al. in arXiv:0802.0930 (to
a complete classification is not known. With a complete clas- appear in Nature). In that measurement, current tunnels across
sification in hand, if we were to observe a topological phase a constriction between two opposite edge states of a Hall bar,
in nature, we could identify it by comparing it against the list and the quasi-particle charge is extracted from the current shot
of topological phases. Since we have observed relatively few noise. Dolev et al. have found the charge to be consistent
topological phases in nature, we have not needed a complete with e/4, and inconsistent with e/2. A quasi-particle charge
classification. If, however, many more are lurking, waiting of e/4 is consistent with paired states at ν = 5/2, including
to be observed, then a complete classification could be use- both the Moore-Read state, the anti-Pfaffian state, and also
64
Abelian paired states. Thus, the observation of charge e/4 We write the OPE for two arbitrary fields φi and φj as
quasiparticles is necessary but not sufficient to show that the P k
ν = 5/2 state is non-Abelian. lim φi (z)φj (w) = k Cij (z − w)∆k −∆i −∆j φk (w) (A1)
z→w
Note added in proof:
k
Dolev et al. (arXiv:0802.0930; Nature, in press) have where the structure constants Cij are only nonzero as indi-
recently measured the low-frequency current noise (‘shot cated by the fusion table. (For our purposes, we can assume
noise’) at a point contact in the ν = 5/2 state. They find the that all fields φk are primary fields. So called “descendant”
noise to be consistent with charge-e/4 quasiparticles, and in- fields, which are certain types of “raising operators” applied
consistent with e/2. A quasi-particle charge of e/4 is consis- to the primary fields, can also occur on the right hand side,
tent with paired states at ν = 5/2, including both the Moore- with the dimension of the descendant being greater than that
Read (Pfaffian) state, the anti-Pfaffian state, and also Abelian of its primary by an integer. Since we will be concerned only
paired states. with leading singularities in the OPE, we will ignore descen-
In another recent experiment, Radu et al. dants. For all the CFTs that we consider the coefficient of
(arXiv:0803.3530) measured the dependence on voltage the primary on the right hand side will not vanish, although
and temperature of the tunneling current at a point contact this can happen.) Note that the OPE works inside a correlator.
in the ν = 5/2 state. They find that the current is well fit For example, in the Z3 parafermion CFT (see Table II), since
by the form I = T α F (e∗ V /kB T ) where e∗ = e/4, and the σ1 × ψ1 = ǫ, for arbitrary fields φi we have
exponent α and scaling function F (x) are at least consistent lim h φ1 (z1 ) . . . φM (zM ) σ1 (z)ψ1 (w) i (A2)
with the anti-Pfaffian state, although it is premature to rule z→w
out other states. ∼ (z − w)2/5−1/15−2/3 h φ1 (z1 ) . . . φM (zM )ǫ(w) i
In a recent preprint (arXiv:0803.0737), Peterson et al. have
performed finite-system exact diagonalization studies which In addition to the OPE, there is also an important “neutral-
find the correct ground state degeneracy on the torus at ν = ity” condition: a correlator is zero unless all of the fields can
5/2 and also observe the expected degeneracy between Pfaf- fuse together to form the identity field 1. For example, in the
fian and anti-Pfaffian states. The key new ingredient in their Z3 parafermion field theory hψ2 ψ1 i = 6 0 since ψ2 × ψ1 = 1,
calculation is the inclusion of the effects of the finite-thickness but hψ1 ψ1 i = 0 since ψ1 × ψ1 = ψ2 6= 1.
of the 2D layer which also appears to enhance the overlap be-
Ising CFT: (c = 1/2)
tween the non-Abelian states and the exact numerical finite- Chiral Bose Vertex: (c = 1)
system wavefunction at ν = 5/2. ∆ × ψ σ
∆ × eiαφ
The first two papers provide the first direct experimental ψ 1/2 ψ 1
eiαφ α2 /2 eiβφ ei(α+β)φ
evidence in support of the 5/2 state being non-Abelian while σ 1/16 σ σ 1+ψ
the third paper strengthens the case from numerics.
Z3 Parafermion CFT: (c = 4/5)
∆ × ψ1 ψ2 σ1 σ2 ǫ
ψ1 2/3 ψ1 ψ2
Acknowledgments
ψ2 2/3 ψ2 1 ψ1
The authors are grateful for support from Microsoft Sta- σ1 1/15 σ1 ǫ σ2 σ2 + ψ1
tion Q, the National Science Foundation under grant DMR- σ2 1/15 σ2 σ1 ǫ 1 + ǫ σ1 + ψ2
0411800, the Army Research Office under grant W911NF- ǫ 2/5 ǫ σ2 σ1 σ1 + ψ2 σ2 + ψ1 1 + ǫ
04-1-0236, the Israel Science Foundation, the U.S.-Israel Bi-
national Science Foundation, and Alcatel-Lucent Bell Labs.
TABLE II Conformal data for three CFTs. Given is the list of pri-
mary fields in the CFT with their conformal dimension ∆, as well as
the fusion table. In addition, every CFT has an identity field 1 with
APPENDIX A: Conformal Field Theory (CFT) for dimension ∆ = 0 which fuses trivially with any field (1 × φi = φi
Pedestrians for any φi ). Note that fusion tables are symmetric so only the lower
part is given. In the Ising CFT the field ψ is frequently notated as
ǫ. This fusion table indicates the nonzero elements of the fusion ma-
We consider chiral CFTs in 2 dimensions. “Chiral” means c
trix Nab . For example in the Z3 CFT, since σ1 × σ2 = 1 + ǫ,
that all of our fields will be functions of z = x + iy only
Nσ11 σ2 = Nσǫ1 σ2 = 1 and Nσc1 σ2 = 0 for all c not equal to 1 or ǫ.
and not functions of z̄. (For a good introduction to CFT see
(Belavin et al., 1984; Di Francesco et al., 1997)). (b) Conformal Blocks: Let us look at what happens when
(a) OPE: To describe a CFT we give its “conformal data”, a fusion has more than one possible result. For example, in
including a set of primary fields, each with a conformal di- the Ising CFT, σ × σ = 1 + ψ. Using the OPE, we have
mension ∆, a table of fusion rules of these fields and a central 1
charge c (which we will not need here, but is fundamental to lim σ(w1 )σ(w2 ) ∼ +(w1 −w2 )3/8 ψ (A3)
w1 →w2 (w1 − w2 )1/8
defining each CFT). Data for three CFTs are given in Table II.
k
The operator product expansion (OPE) describes what hap- where we have neglected the constants Cij . If we consider
pens to two fields when their positions approach each other. hσσi, the neutrality condition picks out only the first term in
65
Theory of computing (ACM Press, New York, NY, USA), pp. changes, Proc. Roy. Soc. Lond. A392, 45.
176–188, ISBN 0-89791-888-6. Bishara, W., and C. Nayak, 2007, Non-Abelian anyon superconduc-
Anderson, P. W., 1973, Resonating valence bonds: a new kind of tivity, phys. Rev. Lett. (in press); cond-mat/0703097.
insulator?, Mat. Res. Bull. 8, 153. Blanchet, C., N. Habegger, G. Masbaum, and P. Vogel, 1995, Topo-
Anderson, P. W., 1987, The resonating valence bond state in la2 cuo4 logical quantum field theories derived from the kauffman bracket,
and superconductivity, Science 235, 1196. Topology 34, 883.
Aneziris, C., A. P. Balachandran, M. Bourdeau, S. Jo, T. R. Ra- Blok, B., and X. G. Wen, 1990, Effective theories of the fractional
madas, and R. D. Sorkin, 1989, Statistics and general relativity, quantum Hall effect: Hierarchy construction, Phys. Rev. B 42(13),
Mod. Phys. Lett. A4, 331. 8145.
Ardonne, E., 2002, Parafermion statistics and the application to non- Blok, B., and X.-G. Wen, 1992, Many-body systems with non-
Abelian quantum Hall states, J. Phys. A 35, 447. Abelian statistics, Nuc. Phys. B 374, 615.
Ardonne, E., P. Bouwknegt, and K. Schoutens, 2001, Non-Abelian Boebinger, G., H. Stomer, D. Tsui, A. Chang, J. Hwang, A. Cho,
quantum Hall states - exclusion statistics, k-matrices and duality, C. Tu, and G. Weimann, 1987, Activation energies and localiza-
J. of Statistical Phys. 102, 421. tion in the fractional quantum Hall effect, Phys. Rev. B 36, 7919.
Ardonne, E., and E.-A. Kim, 2007, Hearing non-Abelian statis- Bonderson, P., A. Kitaev, and K. Shtengel, 2006a, Detecting non-
tics from a Moore-Read double point contact interferometer, Abelian statistics in the ν = 5/2 fractional quantum Hall state,
arXiv:0705.2902. Phys. Rev. Lett. 96(1), 016803.
Ardonne, E., and K. Schoutens, 1999, New class of non-Abelian Bonderson, P., K. Shtengel, and J. K. Slingerland, 2006b, Prob-
spin-singlet quantum Hall states, Phys. Rev. Lett. 82(25), 5096. ing non-Abelian statistics with quasiparticle interferometry, Phys.
Ardonne, E., and K. Schoutens, 2007, Wavefunctions for topological Rev. Lett. 97(1), 016401.
quantum registers, Ann. Phys. (N.Y.) 322, 221. Bonderson, P., K. Shtengel, and J. K. Slingerland, 2007, Decoher-
Arovas, D., J. R. Schrieffer, and F. Wilczek, 1984, Fractional statis- ence of anyonic charge in interferometry measurements, Phys.
tics and the quantum Hall effect, Phys. Rev. Lett. 53(7), 722. Rev. Lett. 98(7), 070401.
Bais, F. A., 1980, Flux metamorphosis, Nucl. Phys. B170, 32. Bonesteel, N. E., 1989, Valence bonds and the Lieb-Schultz-Mattis
Bais, F. A., P. van Driel, and M. de Wild Propitius, 1992, Quantum theorem, Phys. Rev. B 40(14), 8954.
symmetries in discrete gauge theories, Phys. Lett. B280, 63. Bonesteel, N. E., 2000, Chiral spin liquids and quantum error-
Bais, F. A., P. van Driel, and M. de Wild Propitius, 1993a, Anyons correcting codes, Phys. Rev. A 62(6), 062310.
in discrete gauge theories with Chern-Simons terms, Nucl. Phys. Bonesteel, N. E., L. Hormozi, G. Zikos, and S. H. Simon, 2005,
B393, 547. Braid topologies for quantum computation, Phys. Rev. Lett.
Bais, F. A., A. Morozov, and M. de Wild Propitius, 1993b, Charge 95(14), 140503.
screening in the Higgs phase of Chern-Simons electrodynamics, Bonesteel, N. E., I. A. McDonald, and C. Nayak, 1996, Gauge fields
Phys. Rev. Lett. 71, 2383. and pairing in double-layer composite fermion metals, Phys. Rev.
Bakalov, B., and A. Kirillov, 2001, Lectures on Tensor Categories Lett. 77(14), 3009.
and Modular Functors, volume 21 of University Lecture Series Bonesteel, N. E., and K. Yang, 2007, Infinite-randomness fixed
(American Mathematical Society). points for chains of non-abelian quasiparticles, Physical Review
Balatsky, A., and E. Fradkin, 1991, Singlet quantum Hall effect and Letters 99(14), 140405/1.
Chern-Simons theories, Phys. Rev. B 43(13), 10622. Bouwknegt, P., and K. Schoutens, 1999, Exclusion statistics in con-
Balents, L., M. P. A. Fisher, and S. M. Girvin, 2002, Fractionaliza- formal field theory - generalized fermions and spinons for level-1
tion in an easy-axis kagome antiferromagnet, Phys. Rev. B 65(9), wzw theories, Nucl. Phys. B 547, 501.
224412. Bravyi, S., 2006, Universal quantum computation with the ν = 5/2
Balents, L., M. P. A. Fisher, and C. Nayak, 1998, Nodal liquid theory fractional quantum Hall state, Phys. Rev. A 73(4), 042313.
of the pseudo-gap phase of high-Tc superconductors, Int. J. Mod. Bravyi, S., and A. Kitaev, 2001, Quantum invariants of 3-manifolds
Phys. B 12(10), 1033. and quantum computation, unpublished.
Balents, L., M. P. A. Fisher, and C. Nayak, 1999, Dual order param- Bremner, M. J., C. M. Dawson, J. L. Dodd, A. Gilchrist, A. W. Har-
eter for the nodal liquid, Phys. Rev. B 60, 1654. row, D. Mortimer, M. A. Nielsen, and T. J. Osborne, 2002, Prac-
Balents, L., M. P. A. Fisher, and C. Nayak, 2000, Dual vortex theory tical scheme for quantum computation with any two-qubit entan-
of strongly interacting electrons: A non-fermi liquid with a twist, gling gate, Phys. Rev. Lett. 89(24), 247902.
Phys. Rev. B 61(9), 6307. Brennen, G. K., and J. K. Pachos, 2007, Why should anyone care
Bantay, P., 1997, The frobenius-schur indicator in conformal field about computing with anyons?, arXiv:0704.2241.
theory, Phys.Lett. B 394, 87. Bretin, V., S. Stock, Y. Seurin, and J. Dalibard, 2004, Fast rotation of
Baskaran, G., Z. Zou, and P. W. Anderson, 1987, The resonating a bose-einstein condensate, Phys. Rev. Lett. 92(5), 050403/1.
valence bond state and high-Tc superconductivity – a mean field de C. Chamon, C., D. E. Freed, S. A. Kivelson, S. L. Sondhi, and
theory, Solid State Commun. 63(11), 973. X. G. Wen, 1997, Two point-contact interferometer for quantum
Belavin, A. A., A. M. Polyakov, and A. B. Zamolodchikov, 1984, In- Hall systems, Phys. Rev. B 55(4), 2331.
finite conformal symmetry in two-dimensional quantum field the- Calderbank, A. R., and P. W. Shor, 1996, Good quantum error-
ory, Nucl. Phys. B241, 333. correcting codes exist, Phys. Rev. A 54, 1098.
Bena, C., and C. Nayak, 2006, Effects of non-Abelian statistics on Camino, F. E., W. Zhou, and V. J. Goldman, 2005, Realization of
two-terminal shot noise in a quantum Hall liquid in the Pfaffian a laughlin quasiparticle interferometer: Observation of fractional
state, Phys. Rev. B 73(15), 155335. statistics, Phys. Rev. B 72(7), 075342.
Bergholtz, E. J., J. Kailasvuori, E. Wikberg, T. H. Hansson, and Cappelli, A., L. S. Georgiev, and I. T. Todorov, 1999, A unified
A. Karlhede, 2006, Pfaffian quantum Hall state made simple: conformal field theory description of paired quantum Hall states,
Multiple vacua and domain walls on a thin torus, Phys. Rev. B Comm. Math. Phys. 205, 657.
74(8), 081308. Cappelli, A., L. S. Georgiev, and I. T. Todorov, 2001, Parafermion
Berry, M. V., 1984, Quantal phase factors accompanying adiabatic Hall states from coset projections of Abelian conformal theories,
67
on the triangular lattice, Phys. Rev. B 66, 214513. Fubini, S., 1991, Vertex operators and quantum hall effect, Mod.
Fetter, A., C. Hanna, and R. Laughlin, 1989, Random-phase approx- Phys. Lett. A6, 347.
imation in the fractional-statistics gas, Phys. Rev. B 39, 9679. Fubini, S., and C. A. Lutken, 1991, Vertex operators in the fractional
Feynman, R. P., 1982, Simulating physics with computers, Interna- quantum Hall effect, Mod. Phys. Lett. A6, 487.
tional Journal of Theor. Physics 21, 467. Fuchs, J., 1992, Affine Lie Algebras and Quantum Groups (Cam-
Feynman, R. P., 1986, Quantum mechanical computers, Foundations bridge University Press).
of Physics 16(6), 507. Funar, L., 1999, On TQFT representations, Pac. J. Math. 188, 251.
Fidkowski, L., 2007, Double point contact in the k=3 read-rezayi Gaebler, J. P., J. T. Stewart, J. L. Bohn, and D. S. Jin, 2007, p-wave
state, arXiv:0704.3291. feshbach molecules, Phys. Rev. Lett. 98(20), 200403/1.
Fidkowski, L., M. Freedman, C. Nayak, K. Walker, and Z. Wang, Gawȩdzki, K., and A. Kupiainen, 1988, G/H conformal field theory
2006, From string nets to nonabelions, cond-mat/0610583. from gauged WZW model, Phys. Lett. B 215, 119.
Fradkin, E., M. Huerta, and G. Zemba, 2001, Effective Chern- Georgiev, L. S., 2006, Topologically protected gates for quantum
Simons theories of Pfaffian and parafermionic quantum Hall computation with non-Abelian anyons in the Pfaffian quantum
states, and orbifold conformal field theories, Nucl. Phys. B 601, Hall state, Phys. Rev. B 74(23), 235112.
591. Georgiev, L. S., and M. R. Geller, 2006, Aharonov-bohm effect in the
Fradkin, E., and L. P. Kadanoff, 1980, Disorder variables and para- non-Abelian quantum Hall fluid, Phys. Rev. B 73(20), 205310.
fermions in two-dimensional statistical mechanics, Nucl. Phys. B Gepner, D., and Z.-a. Qiu, 1987, Modular invariant partition func-
170(1), 1. tions for parafermionic field theories, Nucl. Phys. B285, 423.
Fradkin, E., C. Nayak, and K. Schoutens, 1999, Landau-Ginzburg Gepner, D., and E. Witten, 1986, String theory on group manifolds,
theories for non-Abelian quantum Hall states, Nucl. Phys. B 546, Nucl. Phys. B278, 493.
711. Girvin, S. M., and A. H. MacDonald, 1987, Off-diagonal long-range
Fradkin, E., C. Nayak, A. Tsvelik, and F. Wilczek, 1998, A Chern- order, oblique confinement and the fractional quantum Hall effect,
Simons effective field theory for the Pfaffian quantum Hall state, Phys. Rev. Lett. 58(12), 1252.
Nucl. Phys. B 516(3), 704. Godfrey, M. D., P. Jiang, W. Kang, S. H. Simon, K. W. Baldwin,
Fradkin, E., and S. H. Shenker, 1975, Phase diagrams of lattice gauge L. N. Pfeiffer, and K. W. West, 2007, Aharonov-bohm-like oscil-
theories with Higgs fields, Phys. Rev. D 19(12), 3682. lations in quantum Hall corrals, arXiv:0708.2448.
Fredenhagen, K., K. H. Rehren, and B. Schroer, 1989, Superselection Goerbig, M. O., P. Lederer, and C. M. Smith, 2004, Second genera-
sectors with braid group statistics and exchange algebras, Com- tion of composite fermions and the self-similarity of the fractional
mun. Math. Phys. 125, 201. quantum Hall effect, Int J. Mod. Phys. B 18, 3549.
Freedman, M., A. Kitaev, M. Larsen, and Z. Wang, 2003a, Topolog- Goldin, G. A., R. Menikoff, and D. H. Sharp, 1981, Representations
ical quantum computation, Bull. Amer. Math. Soc. 40, 31. of a local current algebra in nonsimply connected space and the
Freedman, M., C. Nayak, and K. Shtengel, 2003b, Non-Abelian aharonovbohm effect, J. Math. Phys. 22(8), 1664.
topological phases in an extended Hubbard model. Goldman, V. J., and B. Su, 1995, Resonant Tunneling in the Quan-
Freedman, M., C. Nayak, and K. Shtengel, 2005a, Extended Hubbard tum Hall Regime: Measurement of Fractional Charge, Science
model with ring exchange: A route to a non-Abelian topological 267(5200), 1010.
phase, Phys. Rev. Lett. 94(6), 066401. Goldstone, J., and F. Wilczek, 1981, Fractional quantum numbers of
Freedman, M., C. Nayak, and K. Shtengel, 2005b, Line of critical solitons, Phys. Rev. Lett. 47, 986.
points in 2 + 1 dimensions: Quantum critical loop gases and non- Goméz, C., M. Ruiz-Altaba, and G. Sierra, 1996,
Abelian gauge theory, Phys. Rev. Lett. 94(14), 147205. Quantum Groups in Two-Dimensional Physics (Cambridge
Freedman, M., C. Nayak, K. Shtengel, K. Walker, and Z. Wang, Monographs on Mathematical Physics, Cambridge University
2004, A class of P, T -invariant topological phases of interacting Press).
electrons, Ann. Phys. (N.Y.) 310, 428. Gottesman, D., 1998, Theory of fault-tolerant quantum computation,
Freedman, M., C. Nayak, and K. Walker, 2006, Towards universal Phys. Rev. A 57(1), 127.
topological quantum computation in the ν = 5/2 fractional quan- Greiter, M., X.-G. Wen, and F. Wilczek, 1991, Paired Hall state at
tum Hall state, Phys. Rev. B 73(24), 245307. half filling, Phys. Rev. Lett. 66(24), 3205.
Freedman, M. H., 2001, Quantum computation and the localization Greiter, M., X. G. Wen, and F. Wilczek, 1992, Paired Hall states,
of modular functors, Found. Comput. Math. 1, 183. Nucl. Phys. B 374(3), 567.
Freedman, M. H., 2003, A magnetic model with a possible Chern- Grosfeld, E., N. R. Cooper, A. Stern, and R. Ilan, 2007, Predicted sig-
Simons phase, Commun. Math. Phys. 234, 129. natures of p-wave superfluid phases and Majorana zero modes of
Freedman, M. H., M. J. Larsen, and Z. Wang, 2002a, A modular fermionic atoms in rf absorption, Phys. Rev. B 76(10), 104516/1.
functor which is universal for quantum computation, Commun. Grosfeld, E., S. H. Simon, and A. Stern, 2006, Switching noise as a
Math. Phys. 227, 605. probe of statistics in the fractional quantum Hall effect, Phys. Rev.
Freedman, M. H., M. J. Larsen, and Z. Wang, 2002b, The two- Lett. 96(22), 226803.
eigenvalue problem and density of Jones representation of braid Grosfeld, E., and A. Stern, 2006, Electronic transport in an array of
groups, Commun. Math. Phys. 228, 177. quasiparticles in the ν = 5/2 non-Abelian quantum Hall state,
Fröhlich, J., and F. Gabbiani, 1990, Braid statistics in local quantum Phys. Rev. B 73(20), 201303.
theory, Rev. Math. Phys. 2, 251. Gurarie, V., and C. Nayak, 1997, A plasma analogy and Berry ma-
Fröhlich, J., and P. A. Marchetti, 1988, Quantum field theory of trices for non-Abelian quantum Hall states, Nucl. Phys. B 506(3),
anyons, Lett. Math. Phys. 16, 347. 685.
Fröhlich, J., and P. A. Marchetti, 1989, Quantum field theories of Gurarie, V., L. Radzihovsky, and A. V. Andreev, 2005, Quantum
vortices and anyons, Commun. Math. Phys. 121, 177. phase transitions across a p-wave feshbach resonance, Phys. Rev.
Frohlich, J., and P. A. Marchetti, 1991, Spin-statistics theorem and Lett. 94(23), 230403.
scattering in planar quantum field theories with braid statistics, Haldane, F. D. M., 1983, Fractional quantization of the Hall effect: A
Nucl. Phys. B356, 533. hierarchy of incompressible quantum fluid states, Phys. Rev. Lett.
69
compressible quantum fluid with fractionally charged excitations, fraction 5/2, Nature Phys. 3, 561.
Phys. Rev. Lett. 50(18), 1395. Milovanović, M., and N. Read, 1996, Edge excitations of paired frac-
Laughlin, R. B., 1988a, The relationship between high-temperature tional quantum Hall states, Phys. Rev. B 53(20), 13559.
superconductivity and the fractional quantum Hall effect, Science Misguich, G., C. Lhuillier, B. Bernu, and C. Waldtmann, 1999, Spin-
242, 525. liquid phase of the multiple-spin exchange hamiltonian on the tri-
Laughlin, R. B., 1988b, Superconducting ground state of nonin- angular lattice, Phys. Rev. B 60(2), 1064.
teracting particles obeying fractional statistics, Phys. Rev. Lett. Mochon, C., 2003, Anyons from nonsolvable finite groups are suf-
60(25), 2677. ficient for universal quantum computation, Phys. Rev. A 67(2),
Lee, D.-H., G.-M. Zhang, and T. Xiang, 2007a, Edge solitons of 022315.
topological insulators and fractionalized quasiparticles in two di- Mochon, C., 2004, Anyon computers with smaller groups, Phys. Rev.
mensions, arXiv:0705.3499. A 69(3), 032306.
Lee, S.-S., S. Ryu, C. Nayak, and M. P. Fisher, 2007b, Particle-hole Moessner, R., and S. L. Sondhi, 2001, An RVB phase in the triangu-
symmetry and the ν = 5/2 quantum Hall state, arXiv:0707.0478, lar lattice quantum dimer model, Phys. Rev. Lett. 86, 1881.
to appear in Phys. Rev. Lett. Moessner, R., S. L. Sondhi, and P. Chandra, 1999, Two-dimensional
Lee, T., and P. Oh, 1994, Non-Abelian Chern-Simons quantum me- periodic frustrated ising models in a transverse field, Phys. Rev.
chanics, Phys. Rev. Lett. 72(8), 1141. Lett. 84(19), 4457.
Leggett, A. J., 1975, A theoretical description of the new phases of Moller, G., and S. H. Simon, 2007, Paired composite fermion wave-
liquid 3 He, Rev. Mod. Phys. 47(2), 331. functions, arXiv:0708.2680.
Leinaas, J. M., and J. Myrheim, 1977, On the theory of identical Moore, G., and N. Read, 1991, Nonabelions in the fractional quan-
particles, Nuovo Cimento 37B, 1. tum Hall effect, Nucl. Phys. B 360(2-3), 362.
Levin, M., B. I. Halperin, and B. Rosenow, 2007, Particle-hole sym- Moore, G., and N. Seiberg, 1988, Polynomial equations for rational
metry and the Pfaffian state, arXiv:0707.0483, to appear in Phys. conformal field theories, Phys. Lett.. B 212, 451.
Rev. Lett. Moore, G., and N. Seiberg, 1989, Classical and quantum conformal
Levin, M. A., and X.-G. Wen, 2005a, Colloquium: Photons and elec- field theory, Commun. Math. Phys. 123, 177.
trons as emergent phenomena, Rev. of Mod. Phys. 77(3), 871. Morf, R., and N. d’Ambrumenil, 2003, Disorder in fractional quan-
Levin, M. A., and X.-G. Wen, 2005b, String-net condensation: A tum Hall states and the gap at ν = 5/2, Phys. Rev. B 68,
physical mechanism for topological phases, Phys. Rev. B 71(4), 113309/1.
045110. Morf, R. H., 1998, Transition from quantum Hall to compressible
Levin, M. A., and X.-G. Wen, 2006, Detecting topological order in a states in the second Landau level: new light on the ν = 5/2
ground state wave function, Phys. Rev. Lett. 96(11), 110405. enigma, Phys. Rev. Lett. 80(7), 1505.
Levinsen, J., N. Cooper, and V. Gurarie, 2007, Strongly-resonant p- Morf, R. H., and N. d’Ambrumenil, 1995, Stability and effective
wave superfluids, arXiv:0705.4460. masses of composite fermions in the first and second Landau level,
Lilly, M. P., K. B. Cooper, J. P. Eisenstein, L. N. Pfeiffer, and K. W. Phys. Rev. Lett. 74(25), 5116.
West, 1999a, Anisotropic states of two-dimensional electron sys- Morf, R. H., N. d’Ambrumenil, and S. Das Sarma, 2002, Excitation
tems in high Landau levels: Effect of an in-plane magnetic field, gaps in fractional quantum Hall states: An exact diagonalization
Phys. Rev. Lett. 83(4), 824. study, Phys. Rev. B 66(7), 075408.
Lilly, M. P., K. B. Cooper, J. P. Eisenstein, L. N. Pfeiffer, and K. W. Motrunich, O. I., 2003, Bosonic model with z3 fractionalization,
West, 1999b, Evidence for an anisotropic state of two-dimensional Phys. Rev. B 67, 115108.
electrons in high Landau levels, Phys. Rev. Lett. 82(2), 394. Motrunich, O. I., and T. Senthil, 2002, Exotic order in simple models
Lo, H.-K., and J. Preskill, 1993, Non-abelian vortices and non- of bosonic systems, Phys. Rev. Lett. 89, 277004.
abelian statistics, Phys. Rev. D 48(10), 4821. Mudry, C., and E. Fradkin, 1994a, Mechanism of spin and charge
López, A., and E. Fradkin, 1991, Fractional quantum Hall effect and separation in one-dimensional quantum antiferromagnets, Phys.
Chern-Simons gauge theories, Phys. Rev. B 44(10), 5246. Rev. B 50, 11409.
López, A., and E. Fradkin, 2004, Fermion Chern-Simons theory of Mudry, C., and E. Fradkin, 1994b, Separation of spin and charge
hierarchical fractional quantum Hall states, Phys. Rev. B 69(15), quantum numbers in strongly correlated systems, Phys. Rev. B
155322/1. 49, 5200.
Mackenzie, A. P., and Y. Maeno, 2003, The superconductivity of Mueller, E. J., 2004, Artificial electromagnetism for neutral atoms:
sr2 ruo4 and the physics of spin-triplet pairing, Rev. Mod. Phys. Escher staircase and laughlin liquids, Phys. Rev. A 70(4), 041603.
75(2), 657. Murthy, G., and R. Shankar, 2003, Hamiltonian theories of the frac-
Manfra, M. J., R. de Picciotto, Z. Jiang, S. H. Simon, L. N. Pfeiffer, tional quantum hall effect, Rev. Mod. Phys. 75(4), 1101.
K. W. West, and A. M. Sergent, 2007, Impact of spin-orbit cou- Nayak, C., and K. Shtengel, 2001, Microscopic models of two-
pling on quantum Hall nematic phases, Phys. Rev. Lett. 98(20), dimensional magnets with fractionalized excitations, Phys. Rev.
206804. B 64, 064422.
Manin, Y. I., 1980, Computable and Uncomputable, volume 21 Nayak, C., and F. Wilczek, 1996, 2n-quasihole states realize 2n−1 -
(Sovetskoye Radio). dimensional spinor braiding statistics in paired quantum Hall
Marzuoli, A., and M. Rasetti, 2005, Spin network setting of topolog- states, Nucl. Phys. B 479(3), 529.
ical quantum computation, Int. J. of Quant. Info. 3, 65. Nielsen, M. A., and I. L. Chuang, 2000, Quantum Computation and
McKay, W., and J. Patera, 1981, Tables of dimensions, indices, and Quantum Information (Cambridge University Press, Cambridge).
branching rules for representations of simple Lie algebras,, vol- Niu, Q., D. J. Thouless, and Y.-S. Wu, 1985, Quantized Hall conduc-
ume 69 of Lecture Notes in Pure and Applied Mathematics (M. tance as a topological invariant, Phys. Rev. B 31(6), 3372.
Dekker, New York). Ortalano, M. W., S. He, and S. Das Sarma, 1997, Realistic calcu-
Miller, J. B., I. P. Radu, D. M. Zumbuhl, E. M. Levenson-Falk, M. A. lations of correlated incompressible electronic states in GaAs-
Kastner, C. M. Marcus, L. N. Pfeiffer, and K. W. West, 2007, Alx Ga1−x As heterostructures and quantum wells, Phys. Rev. B
Fractional quantum Hall effect in a quantum point contact at filling 55(12), 7702.
71
Oshikawa, M., Y. B. Kim, K. Shtengel, C. Nayak, and S. Tewari, Rev. Lett. 95(16), 160404.
2007, Topological degeneracy of non-Abelian states for dummies, Rokhsar, D. S., 1990, Solitons in chiral-spin liquids, Phys. Rev. Lett.
Ann. Phys. (N.Y.) 322, 1477. 65(12), 1506.
Overbosch, B. J., and F. A. Bais, 2001, Inequivalent classes of in- Rokhsar, D. S., and S. A. Kivelson, 1988, Superconductivity and the
terference experiments with non-Abelian anyons, Phys. Rev. A quantum hard-core dimer gas, Phys. Rev. Lett. 61(20), 2376.
64(6), 062107. Rosenow, B., B. Halperin, S. Simon, and A. Stern, 2007, Bulk-edge
Overbosch, B. J., and X.-G. Wen, 2007, Dynamical and scaling prop- coupling in the non-abelian ν = 5/2 quantum Hall interferometer.
erties of ν = 5/2 interferometer. Rosenow, B., and B. I. Halperin, 2007, Influence of interactions on
Pachos, J. K., 2005, Quantum computation with Abelian anyons on flux and back-gate period of quantum Hall interferometers, Phys.
the honeycomb lattice, cond-mat/0511273. Rev. Lett. 98(10), 106801.
Pachos, J. K., 2006, The wavefunction of an anyon, cond- Saminadayar, L., D. C. Glattli, Y. Jin, and B. Etienne, 1997, Obser-
mat/0605068. vation of the e/3 fractionally charged laughlin quasiparticle, Phys.
Pan, W., R. R. Du, H. L. Stormer, D. C. Tsui, L. N. Pfeiffer, K. W. Rev. Lett. 79, 2526.
Baldwin, and K. W. West, 1999a, Strongly anisotropic electronic Scarola, V. W., J. K. Jain, and E. H. Rezayi, 2002, Possible pairing-
transport at Landau level filling factor ν = 9/2 and ν = 5/2 induced even-denominator fractional quantum Hall effect in the
under a tilted magnetic field, Phys. Rev. Lett. 83(4), 820. lowest Landau level, Phys. Rev. Lett. 88(21), 216804.
Pan, W., H. L. Stormer, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, Schwarz, A. S., 1978, The partition function of degenerate quadratic
and K. W. West, 2001, Experimental evidence for a spin-polarized functional and ray-singer invariants, Lett. Math. Phys. 2, 247.
ground state in the ν = 5/2 fractional quantum Hall effect, Sol. Schweikhard, V., I. Coddington, P. Engels, V. P. Mogendorff, and
State Comm. 119, 641. E. A. Cornell, 2004, Rapidly rotating bose-einstein condensates in
Pan, W., J.-S. Xia, V. Shvarts, D. E. Adams, H. L. Stormer, D. C. and near the lowest landau level, Physical Review Letters 92(4),
Tsui, L. N. Pfieffer, K. W. Baldwin, and K. W. West, 1999b, 040404.
Exact quantization of the even-denominator fractional quantum Seidel, A., and D.-H. Lee, 2006, Abelian and non-Abelian Hall liq-
Hall state at ν = 5/2 Landau level filling factor, Phys. Rev. Lett. uids and charge-density wave: Quantum number fractionalization
83(17), 3530. in one and two dimensions, Phys. Rev. Lett. 97(5), 056804.
Peterson, M. R., and S. Das Sarma, 2007, unpublished. Semenoff, G. W., and P. Sodano, 2006a, Stretching the electron as
Popp, M., B. Paredes, and J. I. Cirac, 2004, Adiabatic path to frac- far as it will go, Electron. J. Theor. Phys. 10, 157.
tional quantum hall states of a few bosonic atoms, Phys. Rev. A Semenoff, G. W., and P. Sodano, 2006b, Teleportation by a majorana
70(5), 053612. medium, cond-mat/0601261.
Prange, R., and S. M. Girvin (eds.), 1990, The Quantum Hall effect Senthil, T., and M. P. A. Fisher, 2000, z2 gauge theory of electron
(Springer-Verlag, New York). fractionalization in strongly correlated systems, Phys. Rev. B 62,
Preskill, J., 2004, Lecture notes for physics 7850.
219:quantum computation, available at Senthil, T., and M. P. A. Fisher, 2001a, Fractionalization in the
http://www.theory.caltech.edu/∼preskill/ph219/ph219 2004.html. cuprates: Detecting the topological order, Phys. Rev. Lett. 86, 292.
Read, N., 1989, Order parameter and Ginzburg-Landau theory for Senthil, T., and M. P. A. Fisher, 2001b, Fractionalization, topological
the fractional quantum Hall effect, Phys. Rev. Lett. 62(1), 86. order, and cuprate superconductivity, Phys. Rev. B 63, 134521.
Read, N., 1990, Excitation structure of the hierarchy scheme in the Senthil, T., and O. Motrunich, 2002, Microscopic models for frac-
fractional quantum Hall effect, Phys. Rev. Lett. 65(12), 1502. tionalized phases in strongly correlated systems, Phys. Rev. B 66,
Read, N., 2003, Non-Abelian braid statistics versus projective per- 205104.
mutation statistics, J. Math Phys 44, 558. Shimizu, Y., K. Miyagawa, K. Kanoda, M. Maesato, and G. Saito,
Read, N., 2006, Wavefunctions and counting formulas for quasiholes 2003, Spin liquid state in an organic Mott insulator with a trian-
of clustered quantum Hall states on a sphere, Phys. Rev. B 73(24), gular lattice, Phys. Rev. Lett. 91(10), 107001.
245334. Shor, P. W., 1994, in Proc. 35nd Annual Symposium on Foundations
Read, N., and B. Chakraborty, 1989, Statistics of the excitations of of Computer Science (IEEE Computer Society Press, Los Alami-
the resonating-valence-bond state, Phys. Rev. B 40(10), 7133. tos, CA), pp. 124–134, shafi Goldwasser, ed.
Read, N., and D. Green, 2000, Paired states of fermions in two Shor, P. W., 1995, Scheme for reducing decoherence in quantum
dimensions with breaking of parity and time-reversal symme- computer memory, Phys. Rev. A 52(4), R2493.
tries and the fractional quantum Hall effect, Phys. Rev. B 61(15), Simon, S. H., 2000, Proposal for a quantum hall pump, Phys. Rev. B
10267. 61(24), R16327.
Read, N., and E. Rezayi, 1996, Quasiholes and fermionic zero modes Simon, S. H., 2007, Effect of landau level mixing on braiding statis-
of paired fractional quantum Hall states: the mechanism for non- tics, arXiv:0710.1088.
Abelian statistics, Phys. Rev. B 54(23), 16864. Simon, S. H., N. E. Bonesteel, M. H. Freedman, N. Petrovic, and
Read, N., and E. Rezayi, 1999, Beyond paired quantum Hall states: L. Hormozi, 2006, Topological quantum computing with only one
Parafermions and incompressible states in the first excited Landau mobile quasiparticle, Phys. Rev. Lett. 96(7), 070503.
level, Phys. Rev. B 59(12), 8084. Simon, S. H., E. H. Rezayi, and N. R. Cooper, 2007a, General-
Rezayi, E. H., 2007, unpublished. ized quantum Hall projection hamiltonians, Phys. Rev. B 75(7),
Rezayi, E. H., and F. D. M. Haldane, 2000, Incompressible paired 075318.
Hall state, stripe order, and the composite fermion liquid phase in Simon, S. H., E. H. Rezayi, and N. R. Cooper, 2007b, Pseudopo-
half-filled Landau levels, Phys. Rev. Lett. 84(20), 4685. tentials for multiparticle interactions in the quantum Hall regime,
Rezayi, E. H., and N. Read, 2006, Non-Abelian quantized Hall states Phys. Rev. B 75(19), 195306.
of electrons at filling factors 12/5 and 13/5 in the first excited Lan- Simon, S. H., E. H. Rezayi, N. R. Cooper, and I. Berdnikov, 2007c,
dau level, cond-mat/0608346. Construction of a paired wave function for spinless electrons at
Rezayi, E. H., N. Read, and N. R. Cooper, 2005, Incompressible filling fraction ν = 2/5, Phys. Rev. B 75(7), 075317/1.
liquid state of rapidly rotating bosons at filling factor 3/2, Phys. Slingerland, J. K., and F. A. Bais, 2001, Quantum groups and non-
72
Abelian braiding in quantum Hall systems, Nuc. Phys. B 612, 229. manifolds and quantum 6j-symbols, Topology 31(4), 865902.
Sørensen, A. S., E. Demler, and M. D. Lukin, 2005, Fractional quan- Verlinde, E., 1988, Fusion rules and modular transformations in 2d
tum hall states of atoms in optical lattices, Physical Review Letters conformal field theory, Nuc. Phys. B 300, 360.
94(8), 086803. Verlinde, E., 1990, in Proceedings of the International Colloquium
Steane, A., 1996a, Multiple particle interference and quantum error on Modern Quantum Field Theory (World Scientific, Singapore).
correction, Proc. Roy. Soc. Lond.A 452, 2551. Volovik, G. E., 1994, The Universe in a Helium Droplet (Oxford
Steane, A. M., 1996b, Error correcting codes in quantum theory, University Press).
Phys. Rev. Lett. 77(5), 793. Walker, K., 1991, On witten’s 3-manifold invariants, available at
Steane, A. M., 1996c, Simple quantum error-correcting codes, Phys. http://canyon23.net/math/1991TQFTNotes.pdf.
Rev. A 54(6), 4741. Wan, X., K. Yang, and E. H. Rezayi, 2006, Edge excitations and non-
Stern, A., and B. I. Halperin, 2006, Proposed experiments to probe Abelian statistics in the Moore-Read state: A numerical study in
the non-Abelian ν = 5/2 quantum Hall state, Phys. Rev. Lett. the presence of coulomb interaction and edge confinement, Phys.
96(1), 016802. Rev. Lett. 97(25), 256804.
Stern, A., F. von Oppen, and E. Mariani, 2004, Geometric phases and Wen, X. G., 1990, Topological orders in rigid states, Int. J. Mod.
quantum entanglement as building blocks for non-Abelian quasi- Phys. B 4(2), 239.
particle statistics, Phys. Rev. B 70(20), 205338. Wen, X. G., 1991a, Mean-field theory of spin-liquid states with finite
Stone, M., and S.-B. Chung, 2006, Fusion rules and vortices in px + energy gap and topological orders, Phys. Rev. B 44(6), 2664.
ipy superconductors, Phys. Rev. B 73(1), 014505. Wen, X. G., 1991b, Non-Abelian statistics in the fractional quantum
Stormer, H. L., L. N. Pfeiffer, K. W. Baldwin, and K. W. West, 1990, Hall states, Phys. Rev. Lett. 66(6), 802.
Observation of a Bloch-Grüneisen regime in two-dimensional Wen, X. G., 1992, Theory of the edge states in fractional quantum
electron transport, Phys. Rev. B 41(2), 1278. Hall effects, Intl. J. Mod. Phys. B 6(10), 1711.
Su, W. P., and J. R. Schrieffer, 1981, Fractionally charged excitations Wen, X.-G., 1999, Projective construction of non-Abelian quantum
in charge-density-wave systems with commensurability 3, Phys. Hall liquids, Phys. Rev. B 60(12), 8827.
Rev. Lett. 46(11), 738. Wen, X.-G., 2004, Quantum Field Theory of Many-body Systems
Su, W. P., J. R. Schrieffer, and A. J. Heeger, 1980, Soliton excitations (Oxford University Press).
in polyacetylene, Phys. Rev. B 22(4), 2099. Wen, X. G., and Q. Niu, 1990, Ground-state degeneracy of the frac-
Sutherland, B., 1988, Systems with resonating-valence-bond ground tional quantum Hall states in the presence of a random potential
states: Correlations and excitations, Phys. Rev. B 37, 3786. and on high-genus riemann surfaces, Phys. Rev. B 41(13), 9377.
Tao, R., and Y.-S. Wu, 1984, Gauge invariance and fractional quan- Wen, X.-G., and Y.-S. Wu, 1994, Chiral operator product algebra
tum Hall effect, Phys. Rev. B 30(2), 1097. hidden in certain fractional quantum Hall wave functions, Nucl.
Tewari, S., S. Das Sarma, and D. H. Lee, 2007a, An index theorem Phys. B 419, 455.
for the majorana zero modes in chiral p-wave superconductors, Wen, X. G., and A. Zee, 1992, Classification of Abelian quantum
Phys. Rev. Lett. (in press); arXiv:cond-mat/0609556. Hall states and matrix formulation of topological fluids, Phys. Rev.
Tewari, S., S. Das Sarma, C. Nayak, C. Zhang, and P. Zoller, 2007b, B 46(4), 2290.
Quantum computation using vortices and Majorana zero modes Wen, X.-G., and A. Zee, 1998, Topological degeneracy of quantum
of a px + ipy superfluid of fermionic cold atoms, Phys. Rev. Lett. Hall fluids, Phys. Rev. B 58(23), 15717.
98(1), 010506. Wess, J., and B. Zumino, 1971, Consequences of anomalous ward
Tewari, S., C. Zhang, S. Das Sarma, C. Nayak, and D.-H. Lee, 2007c, identities, Phys. Lett. B 37, 95.
Non-local quantum phenomena associated with the majorana Wilczek, F., 1982a, Magnetic flux, angular momentum, and statistics,
fermion bound states in a px + ipy superconductor, arXiv:cond- Phys. Rev. Lett. 48(17), 1144.
mat/0703717. Wilczek, F., 1982b, Quantum mechanics of fractional-spin particles,
Thouless, D. J., and Y. Gefen, 1991, Fractional quantum Hall ef- Phys. Rev. Lett. 49(14), 957.
fect and multiple Aharonov-Bohm periods, Phys. Rev. Lett. 66(6), Wilczek, F., 1990, Fractional Statistics and Anyon Superconductivity
806. (World Scientific, Singapore).
Toke, C., and J. K. Jain, 2006, Understanding the 5/2 fractional quan- Wilkin, N. K., J. M. F. Gunn, and R. A. Smith, 1998, Do attractive
tum Hall effect without the Pfaffian wave function, Phys. Rev. bosons condense?, Phys. Rev. Lett. 80(11), 2265.
Lett. 96, 246805/14. Willett, R., J. P. Eisenstein, H. L. Stormer, D. C. Tsui, A. C. Gos-
Toke, C., N. Regnault, and J. K. Jain, 2007, Nature of excitations sard, and J. H. English, 1987, Observation of an even-denominator
of the 5/2 fractional quantum Hall effect, Phys. Rev. Lett. 98, quantum number in the fractional quantum Hall effect, Phys. Rev.
036806/1. Lett. 59(15), 1776.
Tracy, L. A., J. P. Eisenstein, L. N. Pfeiffer, and K. W. West, 2007, Willett, R., H. Stormer, D. Tsui, A. Gossard, and J. English, 1988,
Spin transition in the half-filled Landau level, Phys. Rev. Lett. Quantitative experimental test for the theoretical gap energies in
98(8), 086801. the fractional quantum Hall effect, Phys. Rev. B 37, 8476.
Trebst, S., P. Werner, M. Troyer, K. Shtengel, and C. Nayak, 2007, Willett, R. L., M. J. Manfra, L. N. Pfeiffer, and K. W. West, 2007,
Breakdown of a topological phase: Quantum phase transition in a Confinement of fractional quantum Hall states in narrow conduct-
loop gas model with tension, Phys. Rev. Lett. 98(7), 070602. ing channels, Appl. Phys. Lett. 91.
Tserkovnyak, Y., and S. H. Simon, 2003, Monte carlo evaluation of Witten, E., 1983, Global aspects of current algebra, Nucl. Phys.
non-Abelian statistics, Phys. Rev. Lett. 90(1), 016802. B223, 422.
Tsui, D. C., H. L. Stormer, and A. C. Gossard, 1982, Two- Witten, E., 1989, Quantum field theory and the Jones polynomial,
dimensional magnetotransport in the extreme quantum limit, Comm. Math. Phys. 121, 351.
Phys. Rev. Lett. 48(22), 1559. Witzel, W. M., and S. Das Sarma, 2006, Quantum theory for electron
Turaev, V. G., 1994, Quantum Invariants of Knots and 3-Manifolds spin decoherence induced by nuclear spin dynamics in semicon-
(Walter de Gruyter, Berlin, New York). ductor quantum computer architectures: Spectral diffusion of lo-
Turaev, V. G., and O. Y. Viro, 1992, State sum invariants of 3- calized electron spins in the nuclear solid-state environment, Phys.
73
Rev. B 74(3), 035322. ulnik, 2006, High resolution polar kerr effect measurements of
Wójs, A., 2001, Electron correlations in partially filled lowest and Sr2 RuO4 : Evidence for broken time-reversal symmetry in the su-
excited Landau levels, Phys. Rev. B 63(12), 125312. perconducting state, Phys. Rev. Lett. 97(16), 167002.
Wojs, A., and J. J. Quinn, 2002, Electron correlations in a partially Xia, J. S., W. Pan, C. L. Vicente, E. D. Adams, N. S. Sullivan,
filled first excited Landau level, Physica E 12, 63. H. L. Stormer, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, and
Wojs, A., and J. J. Quinn, 2006, Landau level mixing in the ν = 5/2 K. W. West, 2004, Electron correlation in the second Landau level:
fractional quantum Hall state, Phys. Rev. B 74(23), 235319. A competition between many nearly degenerate quantum phases,
Wojs, A., D. Wodzinski, and J. J. Quinn, 2006, Second generation of Phys. Rev. Lett. 93(17), 176809.
Moore-Read quasiholes in a composite-fermion liquid, Phys. Rev. Zhang, C., V. Scarola, S. Tewari, and S. Das Sarma, 2006, Anyonic
B 74(3), 035315. braiding in optical lattices, quant-ph/0609101.
Wojs, A., K.-S. Yi, and J. J. Quinn, 2004, Fractional quantum Zhang, C., S. Tewari, and S. Das Sarma, 2007, Bell’s inequality and
Hall states of clustered composite fermions, Phys. Rev. B 69(20), universal quantum gates in a cold atom chiral fermionic p-wave
205322. superfluid, arXiv:0705.4647.
Wu, Y.-S., 1984, General theory for quantum statistics in two dimen- Zhang, F., and S. Das Sarma, 1986, Excitation gap in the fractional
sions, Phys. Rev. Lett. 52(24), 2103. quantum Hall effect: Finite layer thickness corrections, Phys.Rev.
Wynn, J. C., D. A. Bonn, B. W. Gardner, Y.-J. Lin, R. Liang, B 33, 2903.
W. N. Hardy, J. R. Kirtley, and K. A. Moler, 2001, Limits on Zhang, S. C., T. H. Hansson, and S. Kivelson, 1989, Effective-field-
spin-charge separation from h/2e fluxoids in very underdoped theory model for the fractional quantum Hall effect, Phys. Rev.
YBa2 Cu3 O6+x , Phys. Rev. Lett. 87, 197002/1. Lett. 62(1), 82.
Xia, J., Y. Maeno, P. T. Beyersdorf, M. M. Fejer, and A. Kapit-