Anyons and The Quantum Hall Effect - A Pedagogical Review: Ady Stern
Anyons and The Quantum Hall Effect - A Pedagogical Review: Ady Stern
Anyons and The Quantum Hall Effect - A Pedagogical Review: Ady Stern
com
Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 76100, Israel
Abstract
The dichotomy between fermions and bosons is at the root of many physical phenomena, from
metallic conduction of electricity to super-fluidity, and from the periodic table to coherent propaga-
tion of light. The dichotomy originates from the symmetry of the quantum mechanical wave function
to the interchange of two identical particles. In systems that are confined to two spatial dimensions
particles that are neither fermions nor bosons, coined ‘‘anyons’’, may exist. The fractional quantum
Hall effect offers an experimental system where this possibility is realized. In this paper we present the
concept of anyons, we explain why the observation of the fractional quantum Hall effect almost
forces the notion of anyons upon us, and we review several possible ways for a direct observation
of the physics of anyons. Furthermore, we devote a large part of the paper to non-abelian anyons,
motivating their existence from the point of view of trial wave functions, giving a simple exposition
of their relation to conformal field theories, and reviewing several proposals for their direct
observation.
Ó 2007 Elsevier Inc. All rights reserved.
1. Introduction
There are two basic principles on which the entire formidable world of non-relativistic
quantum mechanics resides. First, the world is described in terms of wave functions that
*
Fax: +972 89344477.
E-mail address: Adiel.Stern@Weizmann.ac.il
0003-4916/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.aop.2005.03.001
A. Stern / Annals of Physics 323 (2008) 204–249 205
satisfy Schroedinger’s wave equation. And second, these wave functions should satisfy cer-
tain symmetry properties with respect to the exchange of identical particles. For fermions
the wave function should be anti-symmetric, for bosons it should be symmetric. It is
impossible to overrate the importance of these symmetries in determining the properties
of quantum systems made of many identical particles. Bosons form superfluids, fermions
form Fermi liquids. The former may carry current without dissipating energy, the latter
dissipate energy. The periodic table of elements, and with it chemistry and biology, looks
the way it does because electrons are fermions. And radiation may propagate in a coherent
way since photons are bosons.
Given this set of reminders, it should be clear that finding particles that are neither
fermions nor bosons is an exciting development. Finding them as a theoretical construct
is exciting enough, as realized by Leinaas and Myrheim [1] and by Wilczek [2]. Having
them in the laboratory, open for experimental investigation with an Amperemeter and a
Voltmeter is plain wonderful. Luckily, two dimensional electronic systems subjected to
a strong magnetic field provide both the theoretical and experimental playground for such
an investigation, all through the Quantum Hall effect [3–6].
This paper is aimed at reviewing the physics of Anyons, particles whose statistics is nei-
ther fermionic not bosonic, and the way it is manifested in the quantum Hall effect. We
will start with introducing the basic characters of this play—the Quantum Hall effect,
the Aharonov–Bohm effect [7] and (more briefly) the Berry phase [8]. We will then show
why the mere experimental observation of the quantum Hall effect, coupled with very
basic principles of physics, forces us to accept the existence of excitations that effectively
are Anyons, and how it raises the distinction between abelian and non-abelian anyons.
Following that, we will explore what the experimental consequences of anyonic quantum
statistics may be, with a strong emphasis on interference phenomena. After this subject is
covered, we will raise one level in complexity, and introduce non-abelian anyons [9], look-
ing first at the general concept, then at the simplest example, that of the m ¼ 5=2 quantum
Hall state, and finally at the more complicated example of the Read–Rezayi states [10].
Some subjects will be left out. In particular, the relation of anyons to topological quan-
tum computation [11], to topological quantum field theories and to group theory are all
covered at great length in a recent review article whose list of authors has some overlap
with the corresponding list of the present paper [12]. These subjects will not be repeated
here. Left out are also realizations of anyons in systems out of the realm of the quantum
Hall effect, the relation of anyons to exclusion statistics (a generalized Pauli principle for
anyons [13–17]) and studies of statistical physics of anyons in abstract models.
The paper attempts to be pedagogical, and should be accessible to graduate students,
both theorists and experimentalists.
e VH
vB ¼ ð1Þ
c w
where v is the electron’s velocity, e is the electron charge and c is the speed of light. Since
the current is I ¼ nevw, with n being the electron’s density, we get
VH B
¼ ð2Þ
I nec
This ratio of the Hall voltage to the current is known as the Hall resistance, denoted by
Ryx ¼ Rxy or RH . The simple-minded line of thoughts goes even further: once the force
that results from the Hall voltage cancels the force that results from the magnetic field,
there will be no other effect of the magnetic field. Thus, the longitudinal voltage, the volt-
age drop parallel to the current, will be independent of the magnetic field. The ratio of this
voltage to the current is the longitudinal resistance Rxx .
Classical arguments are not the whole story. There are two other things to add: real-life
experiments and quantum mechanics. Both contradict the classical arguments in a very
fundamental way. Before we get to the quantum Hall effect, we should pay homage to
an even earlier contradiction—the sign of the Hall voltage. Indeed, while classical physics
relates this sign to the direction of the Lorentz force that acts on the negatively charged
electrons, experiments show a different sign in different materials. Quantum mechanics,
in one of its greatest triumphs in solid state physics, explains this varying sign in terms
of the theory of bands, and the notion of holes.
Several decades after band theory introduced the holes, the quantum Hall effect was
discovered [3–6]. Look at Fig. 1: when the Hall effect is measured in high mobility two
dimensional electronic systems at low temperatures, the Hall resistance is not linear in
magnetic field, in contrast to what Eq. (2) suggests. Rather, it shows steps. The magnetic
field at which a step starts or ends varies between samples, but the value of RH at the steps
is universal. At the steps,
h 1
RH ¼ ð3Þ
e2 m
Here h=e2 is the quantum of resistance, equals to about 25.813 kX. The dimensionless
number m has, in the observed steps, either integer values m ¼ 1; 2; 3 . . . going up to several
tens, or ‘‘simple’’ fractions p=q. Most of the simple fractions are of odd denominators (one
of the exceptions to that rule will be later discussed), usually smaller than about 15. Of
those the most prominent fractions are of the series m ¼ p=ð2p þ 1Þ, with p an integer. Gen-
erally speaking, the lower is the temperature and the cleaner is the sample, the more steps
are observed. Integer values of m go under the name Integer Quantum Hall effect (IQHE)
while fractional values go under the name Fractional Quantum Hall effect (FQHE).
Let us pause to emphasize—these steps are astonishing. First, they are amazingly pre-
cise. Eq. (3) is satisfied to a level of one part in 109 ! Second, their value is independent of
any properties of the material being measured. Rather, they are determined by the ratio of
two of the four universal constants, the charge of the electron and Planck’s constant.
At the same magnetic field in which Rxy is on a step, the longitudinal resistance Rxx van-
ishes. Current flows without local dissipation of energy. Ohm’s law may then be put in
local terms, as a ratio between the gradient of the electro-chemical potential E (loosely
speaking, the electric field) and the current density j as,
A. Stern / Annals of Physics 323 (2008) 204–249 207
4
T ~ 35 mK
n=1.0×1011 cm-2 2/7
3 μ=10×106 cm2/Vs 1/3
Rxy (h/e2)
2/5
2
2/3
1 1
2
0 1/3
2/7
3/4 1/4
1 2/5
1/2
1 2/3
Rxx (kΩ)
3/2
5/2
6/23
10 10 6/25
19 21
0
0 5 10 15
MAGNETIC FIELD [T]
Fig. 1. The quantum Hall effect. When the Hall resistance is measured as a function of magnetic field plateaus at
quantized values are observed. In regions of the magnetic field where the Hall resistance is in a plateau, the
longitudinal resistance vanishes (Sample grown by L.N. Pfeiffer (Lucent-Alcatel) and measured by W. Pan
(Sandia)).
h
E¼ ^z j ð4Þ
e2 m
where both vectors lie in the x y plane. As the electric field and the current are perpen-
dicular to one another, the vanishing longitudinal resistivity (commonly denoted by qxx )
implies also a vanishing longitudinal conductivity rxx , and the quantized Hall resistivity
qxy ¼ h=e2 m implies a quantized Hall conductivity rxy ¼ e2 m=h.
These universal values are the zero temperature limit of the experimental observations.
The deviations at finite temperatures are a rich issue by itself, which we touch only briefly:
at least within a certain range of temperatures, the deviation of rxx ; rxy Tfrom their zero tem-
0
perature value follows an activation law, i.e., is proportional to e T , with T being the
temperature and T 0 being a temperature scale that depends on many details. This temper-
ature dependence, which is familiar also from other contexts, such as low temperature con-
ductance of intrinsic semi-conductors or low temperature dependence of the heat capacity
of super-conductors, indicates the existence of an energy gap between the ground state and
208 A. Stern / Annals of Physics 323 (2008) 204–249
the first excited states. The deviations from the activation law, which are observed at low
temperatures, probably indicate that the gap is not a ‘‘true’’ gap (a finite range of energies
at which there are no electronic states) but rather a mobility gap (a finite range of energies
at which there are no extended electronic states).
Faced with these dramatic experimental observations, one naturally asks ‘‘why does the
effect happen?’’. We will address this question very partially later. At the moment, how-
ever, we are more interested in a different question, namely, ‘‘given that the effect is
observed, what does it teach us?’’ As we will see, the very observation of a fractional quan-
tized Hall conductivity, a vanishing longitudinal conductivity and a mobility gap between
the ground state and the first excited states, when combined with general principles of
physics, will force us to accept the notion of Anyons.
But for going along that route, we first need to review the Aharonov–Bohm effect [7].
Let us start by digressing a bit, and reminding ourselves a few basic facts on the physics
of electrons in a magnetic field. High school physics tells us that an electron in a magnetic
field B and an electric field E is subjected to a force
e
F ¼ eE þ v B ð5Þ
c
Undergraduate classical physics tells us that this force may be derived out of an action.
In particular, if we choose a gauge in which the scalar potential vanishes, this force results
from one term in the action,
Z
e
dtvðtÞ Aðx; tÞ ð6Þ
c
Here A is the vector potential whose various derivatives constitute the electric and mag-
netic fields.
Quantum mechanics tells us that an action is not only a device to generate an equation
of motion. Rather, when divided by h it gives the phase of the contribution of the trajec-
tory xðtÞ to the propagator of the particle. If we look at the case in which the vector poten-
tial is independent of time, we see that the term (6) becomes geometric, i.e., independent of
the velocity the trajectory is traversed by,
Z Z
e e
dtvðtÞ AðxÞ ¼ dl AðxÞ ð7Þ
c c
where the integral is taken along the trajectory. Gauge invariant quantities always involve
closed trajectories (loops). For those, Stokes theorem tells us that the integral (7) becomes
2pU=U0 , where U is the flux enclosed in the trajectory, and U0 hc=e is the flux quantum.
This is the Aharonov–Bohm phase.
A particularly important case is the case of the vector potential created by a solenoid that
cannot be penetrated. In that case, the trajectories of an electron may be topologically clas-
sified by nw , the number of times they wind the solenoid. The phase that corresponds to a tra-
jectory is then 2pnw Us =U0 , where Us is the flux enclosed by the solenoid. For all trajectories,
then, this phase is periodic in the flux, with the period being the flux quantum U0 .
There are two useful experimental set-ups to think about in the context of the Aharo-
nov–Bohm effect (see Fig. 2). The first is the famous double slit interference experiment.
A. Stern / Annals of Physics 323 (2008) 204–249 209
a b
Fig. 2. The Aharonov–Bohm effect. Two realizations of the Aharonov–Bohm effect. In the first, shown in (a), the
magnetic flux induces a shift in the interference pattern of a double slit experiment. An analog effect is to be seen
in the quantum Hall interferometers discussed in Sections 7 and 12. In the second, shown in (b), the magnetic flux
affects the thermodynamic properties of the annulus around it. This realization is very useful in the analysis of the
fractional charge and fractional statistics of quasi-particles in the quantum Hall effect, see Section 4.
When a solenoid is put in between the two slits, it shifts the interference patter due to the
phase shift it induces. Note that this shift in the interference pattern is very surprising from
a classical point of view, as the electron, which cannot penetrate the solenoid, does not feel
any Lorenz force when it moves.
The second set-up is that of a ring, or an annulus, with a magnetic flux U going through
the hole. The electron is now confined to the annulus and again does not experience any
Lorenz force due to the magnetic field in the solenoid. However, due to the Aharonov–
Bohm effect its spectrum does depend on the magnetic flux, with a period of U0 . And since
this is true for the spectrum, it is true for all thermodynamic properties, that may all be
expressed as derivatives of the partition function.
The dependence of various thermodynamical quantities on the magnetic flux through the
hole, despite the absence of any force exerted on the electron by the magnetic field in the
solenoid, may be understood in the following way: suppose that we look at a ring where ini-
tially, at time t ¼ 1, no flux penetrates the hole. Then a flux U is turned on adiabatically in
time. Classically, the process of turning the flux on involves the application of an electric
field on the electron, and hence its acceleration. While there is a freedom of the position
at which the electric field would operate, the condition of adiabaticity implies that the time
period at which the electric field operates is much longer than the time it takes an electron to
encircle the ring, and thus the electron cannot avoid the effect of the field. As long as the elec-
tron is not prevented by a trapping potential from encircling the ring, it will experience a
force due to the electric field, and will absorb (kinetic) momentum and angular momentum.
Due to the adiabaticity, the electron stays in an eigen-state throughout the process in which
the flux is turned on. Thus, the eigen-state evolves in such a way that it absorbs kinetic
momentum and angular momentum. As long as this momentum is not fully transferred
to static degrees of freedom (impurities etc.), the eigenstate will have the electron encircling
the ring in a non-zero velocity. This is the source of the persistent current, or thermodynamic
orbital magnetization, observed in normal mesoscopic rings [18].
There is a subtle point that needs to be understood in this argument: consider first an
electron in a ballistic circular ring, subjected to a magnetic flux that is adiabatically turned
e U
on. As we said, classically the electron is accelerated by the field. Its velocity is mc 2pL
, with
210 A. Stern / Annals of Physics 323 (2008) 204–249
L being the circumference of the ring and m the electron mass. Quantum mechanically, if
the electron was initially in the ground state it stays in an eigenstate, due to adiabaticity. It
does not necessarily stay in the ground state, however. As the flux is being turned on, there
are points of time in which the gap in the spectrum closes, and two eigenstates, one whose
energy increases with flux and one whose energy decreases with flux, become degenerate in
energy. Even when these states are degenerate, there is no matrix element that allows for a
transition between them, and thus the electron does not go through such a transition. The
reason for that is that the two states have different (canonical) angular momenta, and the
rotational symmetry of the problem dictates a conservation of angular momentum. Any
deviation from these perfect conditions, however, for example by the ring being imper-
fectly circular or imperfectly ballistic, introduces transitions between the two states,
removes the degeneracy, and leads to the electron staying, in the adiabatic limit, in the
lowest energy state. This subtle point will be of importance later.
4. Fractionally charged excitations: combining the Aharonov–Bohm effect and the quantum
Hall effect
With very little detailed modelling, we are now able to conclude, following Laughlin’s
original arguments, that a two dimensional system that shows the fractional quantum Hall
effect must have quasi-particles that carry a fraction of an electron charge [19]. Following
that, we will add in a few basic observations about the dynamics of vortices in fluids, and
conclude that the fractional charge of the quasi-particles in the fractional quantum Hall
effect implies also that the quasi-particles are anyons. While the anyonic nature of the
quasi-particles was first realized by Halperin [20], based on the sequence of observed frac-
tional quantum Hall states, the derivation we will follow here is closer to that of Arovas,
Schrieffer and Wilczek [21].
The ingredients that will go into the heuristic derivation of the fractional charge are the
Aharonov–Bohm effect, the fractional quantum Hall effect and the adiabatic theorem.
Rather than looking at the ring we discussed before, let us consider a wide annulus, where
the electrons on the annulus are in the (say) m ¼ 1=3 fractional quantum Hall state (see
Fig. 3). Again, we start with no flux threading the hole at the center of the annulus,
Fig. 3. The gedanken experiment to create an eigenstate of a fractional charge, later realized to be an anyon. The
electrons on the annulus are in a Laughlin fractional quantum Hall state of filling fraction m ¼ 1=m.
The introduction of a flux quantum into the hole pushes a fraction of an electron charge from the interior to the
exterior, leaving the system in a eigenstate.
A. Stern / Annals of Physics 323 (2008) 204–249 211
and turn the flux on adiabatically. Again, an electric field is exerted on the electrons, and
for simplicity we assume this electric field to be azimuthally symmetric. Let us analyze
what happens when the flux is increased.
First, with the electrons being in a fractional quantum Hall state, the azimuthal electric
field leads to a purely radial current flow, according to Eq. (4). At a distance r from the
1 oU ^
origin the electric field is E ¼ 2prc ot
h, and the resulting radial current density is
e2 m
jr ¼ h E. Integrating the current density that goes through a circle of radius r to get the
total current, we find that the current does not depend on r. Thus, it leaves behind a charge
lump at the interior edge of the annulus. Integrating the current over time, starting when
there was no flux at the ring and ending at a time t when the flux is UðtÞ, we find that the
amount of charge in that lump is
e2 m
QðtÞ ¼ UðtÞ ð8Þ
hc
Second, in the quantum Hall effect there is an energy gap between the ground state and
the first excited state. Thus, we can apply the adiabatic theorem, and conclude that the sys-
tem, that started in a ground state when there was no flux, remains in an eigenstate
throughout the process.
Third, what happens when the flux is U0 ? Then, Eq. (8) gives a charge of em. The system is at
an eigenstate. The spectrum at U ¼ U0 is the same as the spectrum at U ¼ 0 and P the eigen-
states at U ¼ U0 are the same as those at U ¼ 0, up to the phase factor exp i i /i (where
/i is the azimuthal coordinate of the ith electron), which does not affect the charge density.
Thus, we found that the annulus with U ¼ 0 has an eigenstate in which a lump of charge of em
is localized on its interior edge. The assignments of energies to eigenstates may have inter-
changed, as in the case of a ballistic circular ring, so this is not necessarily the ground state.
Clearly, an eigenstate in which a charge em has been transferred from the interior of the annu-
lus to its exterior is physically different from the ground state the system started in. We call
this state a quasi-particle when it carries a negative charge and a quasi-hole when it carries a
positive charge. In much of the following the sign of the charge will not be of our concern, and
we will use the terms quasi-hole and quasi-particle interchangeably.
As in the case of a ring, if the ground state at U ¼ 0 evolves into an excited state at
U ¼ U0 then for some intermediate value of U two states must have been degenerate in
energy, with no transition between them taking place as the flux is varied. The absence
of such a transition may be due to the absence of any matrix element connecting them
(due to symmetries of the problem), or due to that matrix element being so small such that
the rate at which UðtÞ would need to be changed becomes so slow to be impractical.
Indeed, for the fractional quantum Hall annulus the latter is the case—the matrix element
between the states is exponentially small in the size of the annulus (in the units of the mag-
netic length), and thus the two states are effectively uncoupled.
That this is the case may be understood in the following manner: the applied electric
field acting in the azimuthal direction pushes a current to flow radially. If the system ends
at U ¼ U0 in the initial state of U ¼ 0, then the charge that was driven from the interior to
the exterior must have tunnelled back. This process of tunneling requires a breaking of
rotational invariance, e.g., by impurities, but being a tunneling process, its amplitude is
exponentially small [22].
In the thought experiment we described above the quasi-particle/quasi-hole were exci-
tations above the ground state. It is, however, easy to imagine the energetics to change in
212 A. Stern / Annals of Physics 323 (2008) 204–249
such a way that a state with such ‘‘excitations’’ becomes lower in energy than a state with-
out them. That happens when electrostatic considerations force the density to deviate from
a ‘‘magic’’ m ¼ 1=m filling fraction. For example, imagine that rather than turning a flux
quantum at the center of the annulus we would turn on a potential that would repel elec-
trons away. If the potential is very small, it would have no effect, since the quantum Hall
fluid is incompressible, due to its energy gap. But if the potential is strong enough it would
lead to a ground state that includes one or more quasi-holes. Fractional charges of quasi-
holes and quasi-particles have been observed in various types of measurements, such as
resonant tunnelling through antidots [23], shot noise [24,25] and local compressibility [26].
The argument we presented so far for the existence of fractionally charged quasi-parti-
cles and quasi-holes has been very general, being based on the existence of the fractional
quantum Hall effect as a gapped ground state, and on general principles such as the adi-
abatic theorem and the Aharonov–Bohm effect. Historically, these quasi-particles were
first understood through the use of trial wave functions. As Laughlin discovered [19], if
the ground state for a m ¼ 1=3 fractional quantum Hall state is assumed to be made solely
of lowest Landau level wave functions, the following is a very good variational wave func-
tion for it (using complex coordinates zi ¼ xi þ iy i for the two Cartesian coordinates of the
ith particle):
Y 3
Y 2
wg:s: ðfzi g; fzi gÞ ¼ ðzi zj Þ exp jzi j =4l2h ð9Þ
i<j i
The virtues of this wave function will be explained later on (see Section 13). At the
moment we just say that it minimizes the kinetic energy by placing all the electrons in
the lowest Landau level, and minimizes the potential energy by efficiently keeping elec-
trons away from one another. To generate a quasi-hole at the origin, Laughlin proposed
the wave function
! P
Y Y i /i
zi wg:s: ¼ jzi j e i wg:s: ð10Þ
i i
Q
Indeed, the operator i zi doesP exactly what we expect the adiabatic introduction of a
flux tubeQ to do: the phase exp i i /i gives each electron an angular momentum h, while the
product i jzi j moves each electron radially outwards, thus generating the radial current
and the lump of charge at the interior. Over all, the operator does not introduce transitions
between different Landau levels, and thus keep the wave function purely within the lowest
Landau level.
The angular momentum carried by the quasi-hole (10) is there not only for the Laughlin
wave function of the quasi-hole. In fact, it is introduced also in the thought experiment
that we described: as the flux is turned on, a radial current flows, with no introduction
of angular
P momentum. However, when the wavefunction for U ¼ U0 is multiplied by
exp i i /i in order to get the wave function for the quasi-hole at U ¼ 0, an angular
momentum of h is given to every electron. The quasi-hole then carries a quantum of
vorticity.
To summarize, the process we just described generates an eigenstate that carries a frac-
tional charge em and a single quantum of vorticity, both localized at the interior edge of the
annulus. By shrinking the size of the hole we can localize this charge to a size of the mag-
netic length (below that size the relation (4) does not hold). The state being an eigenstate
A. Stern / Annals of Physics 323 (2008) 204–249 213
implies that both the lump of charge and the vorticity do not decay in time. They are there
to stay.
The vorticity carried by the quasi-holes in the fractional quantum Hall effect will turn
out to be crucial for their quantum statistics. Thus, let us remind ourselves of a few basic
facts regarding the motion of vortices in two dimensions. In a Galilean invariant system, a
vortex always moves at the speed of the fluid. In the absence of such symmetry, however,
there may be a relative velocity between the vortex and the fluid. Under these conditions,
Magnus force operates on the vortex. With this force being relevant for what comes next,
we now review its basic features, and draw its similarity to the Lorenz force acting on
charged particles in a magnetic field. We start by considering a vortex in a neutral super-
fluid, say superfluid Helium.
The origin of the Magnus force lies in the consideration of the vortex as a collective
degree of freedom. Having a vortex centered at the point R implies that each fluid particle
has an angular momentum l^z relative to the point R. When the vortex is static, there is a
velocity field of the fluid, of vðrÞ ¼ ml ^zðrRÞ
jrRj2
. When the vortex moves the velocity field of the
fluid is the sum of this motion and the velocity of the vortex’ center relative to the fluid, R. _
_
If the vortex’ velocity R is in the ^x direction, the velocity difference along the ^y -axis leads to
a pressure difference, and hence to a force in the ^y direction. The force is
F Mag ¼ 2pnl^z ^v ð11Þ
where n is the number density of the fluid. This force is proportional to the product nlv
since the pressure difference between the two sides of the vortex is proportional to
_ 2 ðv RÞ
nððv þ RÞ _ 2 Þ.
Like the Lorenz force, the Magnus force is proportional and perpendicular to the veloc-
ity. Comparing Eq. (11) to Eq. (5) we see that the role played by the product eB=c in Lor-
enz force is played by the product 2pnl in Magnus force. As in the Lorenz force, the
transition from classical to quantum mechanics of vortices is best understood when inter-
ference effects are considered. The classical Lorenz force gives rise to the quantum
mechanical Aharonov–Bohm phase accumulated by the wave function when an electron
is moving, which then leads to the definition of the flux quantum. Similarly, the classical
Magnus force gives rise in quantum mechanics to a phase that is accumulated
R by the wave
function when a vortex is moving. The Aharonov–Bohm phase is hec ds B. The analo-
gous phase should then be
Z
l
2p ds n ð12Þ
h
For a vortex with a single quantum of vorticity (and all those we are interested in are of this
type) l ¼
h and
R thus the phase accumulated by a vortex traversing a closed loop is 2p times
the integral ds n which is nothing but 2p times the number of fluid particles encircled by the
vortex as it goes around the loop. The analog of the flux quantum is a single fluid particle.
We limited ourselves to a neutral fluid when we took the velocity field of the vortex to
decay as 1=jr Rj, but, as we now argue, both the classical force and the quantum
mechanical phase we obtained hold more generally. When the superfluid is charged, as
214 A. Stern / Annals of Physics 323 (2008) 204–249
with /ðrÞ arctanðyxÞ being the angle conjugate to the angular momentum, and with the
real function GðrÞ vanishing at r ¼ 0. This operator creates a circularly symmetric vortex,
but that simplification is not crucial. When it is applied to a featureless state, which is
216 A. Stern / Annals of Physics 323 (2008) 204–249
independent of R, the geometric vector potential (14) is, with qðrÞ being the fluid density at
the point R,
Z
^z ðr RÞ
AB ðRÞ ¼ drqðrÞ 2
ð16Þ
jr Rj
This is precisely what our semi-heuristic analysis has led us to believe: the vortex, whose
coordinate is R, sees the fluid particles as an electron sees flux tubes. When it moves rel-
ative to the fluid, it experiences a force perpendicular and proportional to its velocity rel-
ative to the fluid, and the magnitude of that force is exactly that given by the Magnus
force. Eqs. (14) and (16) also explain why the magnitude of the force is the same for neu-
tral and charged vortices: it is determined by derivatives of the wave function with respect
to R, and the wave functions for vortices in neutral and super-conducting vortices share
the same crucial property: the canonical angular momentum of h given to each fluid par-
ticle relative to the vortex’ center. The magnetic field created by the currents that form the
vortex in a super-conductor affects the kinetic angular momentum mv r of the Cooper-
pairs around the center of the vortex, but does not affect their canonical angular momen-
tum p r. Finally, the phase accumulated by the quantum mechanical wave function
when the vortex traverses a closed loop is 2p times the expectation value of the number
of fluid particles enclosed in the loop [21].
So we finally got to the quasi-holes in the fractional quantum Hall effect. As we saw,
these quasi-holes, at the Laughlin fractions of m ¼ 1=m, carry a fraction 1=m of the electron
charge and one quantum of vorticity. What can we say about their dynamics, using the
two other types of vortices as references?
The phase factors in the operator (15) create the vorticity carried by the quasi-hole, and
with the proper choice of GðrÞ this operator will create also the correct depletion of charge.
While there are many choices that would satisfy this condition, the most Q natural one, as
suggested by Laughlin, is GðrÞ ¼ r, making the quasi-hole operator i zi [19]. Under this
choice, the operator that creates the quasi-hole does not introduce mixing of Landau lev-
els. When it is applied to the Laughlin wave function (9), the resulting quasi-hole state is
still purely within the lowest Landau level.
Similar to the case of a vortex in a charged super-conductor, the velocity field of the
vortex created by the quasi-hole creation operator does not decay inversely with the dis-
tance from the center of the quasi-hole. Within the Laughlin picture, the charge density
and azimuthal current associated with the quasi-hole both fall off exponentially at dis-
tances larger than the magnetic length. A more elaborate analysis [30] realizes that the azi-
muthal currents must be proportional to 1=r2 (with r being the distance from the quasi-
hole), since the charge of the quasi-hole creates a radial electric field that decays as
1=r2 , and that field generates an azimuthal Hall current. Furthermore, the charge density
is found to decay as 1=r3 . Independent of these details, however, as the quasi-hole moves
relative to the quantum Hall fluid, it experiences a force proportional and perpendicular to
its velocity, given by 2pnlv ^z, resulting from the vector potential (16). This is the same
Magnus force experienced by a vortex in a neutral super-fluid, and the same Lorenz force
that the vortex in a charged super-conductor experienced due to its interaction with the
A. Stern / Annals of Physics 323 (2008) 204–249 217
current. Here this force has a rather simple description: noting that the angular momen-
tum is h and that nU0 =B ¼ m we see that the force acting on the quasi-hole is emc v B,
i.e., it is the Lorenz force acting on a charge em as it moves in a magnetic field B. Viewed
as such, this is not too surprising—if we accept the quasi-hole as a collective degree of free-
dom that carries a fractional charge, it should be subjected to the Lorenz force that cor-
responds to that charge.
But the story gets more interesting when we think of a quasi-hole as a quantum
mechanical degree of freedom, i.e., when we consider the phase it accumulates due to
the vector potential (16). The quantum Hall fluid is incompressible. Thus, when the den-
sity is such that the local filling factor deviates from 1=m, the deviation is accommodated
in integer number of quasi-particles or quasi-holes. Given the charge em of each quasi-hole,
when a quasi-hole goes around a closed loop, every quasi-hole it encircles contributes a phase
of 2pm to its wave function. And since a winding of one quasi-hole around another is just two
interchanges, we find that an interchange of the position of two quasi-holes multiplies the
wave function by a phase of pm, which makes the quasi-holes anyons [21].
Note again that what gives the quasi-holes their anyonic character is a combination of
two ingredients—the fractional charge and the vorticity they carry. Both are essential for
the quasi-hole to exist, as we described here in detail.
If anyons are defined by the quantum mechanical phase that they accumulate when they
encircle one another, it makes sense to design experimental devices in which quantum
mechanical phases are observables and quasi-particles encircle one another. The observa-
tion of phases accumulated by waves is the essence of interferometry. We thus turn to
interferometers as the devices for the observation of anyons, dealing with the Fabry–Perot
and Mach–Zehnder interferometers.
In the context of the quantum Hall effect, the Fabry–Perot interferometer, discussed in
details first by Chamon et al. [31], is a Hall bar (for concreteness, lying along the x-axis,
between 0 < y < w) perturbed by two constrictions, known as quantum point contacts
(say, at x ¼ d=2). Current flows chirally along the edges, towards the left along the
y ¼ w edge and towards the right along the y ¼ 0 edge. The right moving edge is put at
a chemical potential difference V relative to the left moving one. In the absence of the point
contacts there is no back scattering of current, and thus when the bulk is in a quantum
Hall state the two-terminal conductance (the ratio of the current difference to the voltage
difference between the two edges) is quantized. The point contacts introduce an amplitude
for inter-edge tunnelling, which makes some of the current flowing along the right moving
edge back-scattered. The probability for back-scattering is directly reflected in the voltage
between the edges, i.e., in the two terminal conductance. Fig. 4b folds the Hall bar into an
annulus, for the purpose of comparison with the Mach–Zehnder interferometer, to be dis-
cussed later. In this setting, the point contacts introduce a flow of tunnelling from the
source at the exterior to the drain in the interior.
The presence of two point contacts makes the probability for back-scattering the quan-
tum Hall analog of ‘‘the interference screen’’. If the tunnelling amplitudes introduced by
218 A. Stern / Annals of Physics 323 (2008) 204–249
Fig. 4. The Fabry–Perot (a and b) and Mach–Zehnder (c) interferometers. The second drawing is meant to
emphasize the difference between the two interferometers. The interior edge is a part of the interference loop in the
Mach–Zehnder interferometer, while it is not part of that loop in the Fabry–Perot interferometer. Furthermore,
in the former only single tunnelling events take place, while the latter allows for multiple reflections and the
formation of resonances.
the two point contacts are t1 ; t2 respectively, then the back-scattered current is, to lowest
2
order in these amplitudes, proportional to jt1 þ t2 j . To this lowest order, the Fabry–Perot
interferometer is the QHE analog of the two slit experiment. The relative phase between
the two amplitudes may be varied in three principal ways: by varying the magnetic field,
by varying the electron density and by varying the area of the ‘‘cell’’ defined by the two
edges and two point contacts. The latter is implemented by a side gate that ‘‘moves the
walls’’ of the cell in the region it operates on. Within the crudest approximation, a varia-
tion of the area of the cell does not vary the filling factor in the bulk. Thus, it varies the
number of electrons in the cell (which is the product of density and area), but does not
introduce quasi-particles into the bulk. The variation of the electron density and/or mag-
netic field lead to a more complicated effect. Electrostatic interaction tends to fix the elec-
tronic charge density to be such that it exactly neutralizes the positive background. When
this interaction dominates, a variation of the bulk Landau filling from that of the quantum
Hall state (say, a variation from an integer m or from m ¼ 1=m) leads to the introduction of
localized quasi-holes or quasi-particles. Such a variation may be created either by varying
the magnetic field or by varying the positive background density (usually done by applying
a voltage to a gate).
Given this picture, the Fabry–Perot interferometer gives us a device in which quasi-
holes that are flowing on the edge split into partial waves that interfere, the interference
pattern is measured by measuring the backscattered current as a function of three possible
parameters, and the area enclosed by the interfering trajectories enclose localized quasi-
holes. It then makes sense to expect the dependence of the back-scattered current on
the area of the cell, the magnetic field and the density to reflect the statistics of the
quasi-particles. Let us see how that happens.
We start with the simplest toy model, non-interacting electrons at m ¼ 1, and consider
how the back-scattered current depends on the area of the cell and the magnetic field. By
the Aharonov–Bohm effect, the relative phase between the two interfering waves is 2p
A. Stern / Annals of Physics 323 (2008) 204–249 219
times the number of flux quanta enclosed in the cell between the two interfering waves.
That number is BS=U0 , with S being the area of the cell. As S is varied, then, we expect
a sinusoidal variation of the back-scattered current. The period of this variation is
DS ¼ e=n0 , with n0 being the density of electrons. In other words, the number of electrons
in the cell changes by one for each period of the oscillations.
As B is varied the picture might be slightly more complicated, since the variation of B
might also lead to a variation of S, by changing the occupation of states at the edge. These
edge effects are important when the back-scattering probability is close to one, such that
the cell becomes a quantum dot. Deferring this limit to a later stage, we assume the area of
the cell to be independent of the magnetic field, and thus the dependence of the back-scat-
tering current on the magnetic field to be sinusoidal as well. The period for these oscilla-
tions is DB ¼ U0 =S.
It is important to note that as the magnetic field varies and the bulk filling factor
becomes, say, 1 , a density n of quasi-holes, which for m ¼ 1 are just missing electrons,
appears in the bulk. On average, a variation of the magnetic flux through the cell by one
flux quantum introduces one quasi-hole into the bulk of the cell. The precise values of
magnetic field in which a quasi-hole is introduced into the bulk depends on the detailed
geometry and the disorder potential within the cell. While for m ¼ 1 the introduction of
quasi-holes into the bulk does not affect the back-scattered current, this is not the case
for the fractional quantum Hall states, as we now see.
When the filling factor is the fraction m ¼ 1=m several factors should be taken into
account. First, several objects may be tunnelling across the point contacts, starting from
quasi-particles of charge em and ending with electrons of charge e. For several reasons,
we expect the tunnelling of em-charged quasi-particles to be the strongest: being the object
with the smallest charge, its bare tunnelling matrix elements are likely to be the largest.
And under renormalization at low temperatures and low voltage, it has the fastest growth
rate. We therefore focus on this type of tunnelling. Second, the phase accumulated by a
quasi-particle encircling the cell between the two point contacts is
/ ¼ 2pm½BS=U0 N qh ð17Þ
This phase could be thought in two ways: it is the sum of the Aharonov–Bohm phase
accumulated by a charge em going around a flux BS and the statistical phase of a quasi-
particle going around N qh others. Alternatively, it is the geometric phase we calculated
above, a phase of 2p per every fluid particle encircled. In any case, when the area S is var-
ied, the period of the oscillations of back-scattering is DS ¼ e=n0 , as before. When the
magnetic field is varied, on the other hand, there is a continuous evolution of the Aharo-
nov–Bohm part of the phase, accompanied by occasional jumps of the statistical phase,
taking place whenever the number N qh increases by one. These jumps are the manifestations
of the statistical interaction of the quasi-holes flowing along the edge with those that are
localized in the bulk.
When the amplitude for tunnelling across the two point contacts is not very small, mul-
tiple reflections should be taken into account. Waves that are back-scattered at the second
point-contact wind the cell several times before leaving the cell towards one of the con-
tacts. The electronic Fabry–Perot interferometer develops resonances: rather than having
the backscattered current oscillate sinusoidally as a function of magnetic field and area,
most of the current is backscattered, except in certain lines in the magnetic field-area plane
in which constructive interference of the waves that are transmitted through the cell build
220 A. Stern / Annals of Physics 323 (2008) 204–249
up the probability of the current not to be back-scattered. Generally, the transmitted cur-
rent takes the form:
X
1
I bs ¼ I n cos nð/ þ a0 Þ ð18Þ
n¼0
where the phase / is given in (17) and a0 is a phase that originates from the phases of the
tunnelling amplitudes of the two point contacts. The phase a0 is independent of the mag-
2pN
netic field and the area. Resonances take place when / m qh ¼ 2p‘, with ‘ an integer. The
area between the resonances is then e=n0 .
This state of affairs can be rephrased in a different language: in the limit of strong back-
scattering the cell becomes a quantum dot, in which the number of electrons is quantized.
For most values of the area and the magnetic field, the dot’s energy is minimized with a
unique number of electrons N ðS; BÞ. Current through the dot is then blocked by the energy
cost associated with adding the flowing electron to the dot. Resonances take place at those
value of S and B for which the energies of the dot with N electrons and with N þ 1 elec-
trons are degenerate. One expects these degeneracy points to appear at an area separation
of e=n0 .
The picture we described is based on a sharp distinction between edge quasi-holes,
which are flowing from one contact to another, and bulk quasi-holes, which are localized.
If this distinction is realized experimentally, the Fabry–Perot interferometer takes quasi-
holes to encircle one another in a way that realizes the gedanken experiments rather faith-
fully. Practically, however, it is difficult to avoid the number of quasi-particles enclosed in
the cell, N qh , to fluctuate during the experimental time scale. When such fluctuations take
place on a characteristic time scale s0 in which many quasi-holes are being back-scattered,
they result in two effects: first, for a given value of S and B the back-scattered current
becomes time dependent, and hence noisy. In the limit of weak backsttering the noise
h i
2 2
hI 2bs i hI bs i / hI bs i s0 ð19Þ
x¼0
Second, as seen from (18), certain harmonics, those for which nm is an integer, survive
the averaging over temporal fluctuations of N qh . The time averaged back-scattered current
then becomes U0 –periodic. Furthermore, in the limit of weak back-scattering, the Aharo-
nov–Bohm oscillations of the back-scattered current would originate primarily from the
nm ¼ 1 term in (18). Since
I m / I m1
0 ; ð20Þ
the visibility of the Aharonov–Bohm oscillations, for a fixed voltage and varying back-
scattering strength, would scale like the average current to the m 1 power.
The Mach–Zehnder interferometer is another device that allows for an interference of tra-
jectories of particles that are back-scattered through two point contacts. The two important
differences between the Fabry–Perot and Mach–Zehnder interferometers may be conve-
niently viewed when they are depicted as in Fig. 4. First, while the former allows for multiple
reflections and the formation of resonances, the latter interferes only two waves, each one
going through one tunnelling event. The structure is built in such a way that each quasi-hole
A. Stern / Annals of Physics 323 (2008) 204–249 221
may tunnel between edges at most once before being collected at the contact. Second, while in
the former the interference loop encloses only the cell between the two point contacts, in the
latter the entire internal edge is part of the interference loop.
The inclusion of one of the edges in the interference loop has a profound effect on the
interference pattern that the interferometer shows, as analyzed by Law et al. [32,33] (see
also [34,35]). The reason for that is that N qh now includes the number of quasi-holes on
that edge, and this number changes by one with every tunnelling quasi-hole. Rather than
discussing the back-scattered current directly it is then instructive to analyze the tunnelling
rate of quasi-holes between the two edges. This rate depends on the number N qh :
1 2pN qh
¼ C0 þ C1 cos / þ ð21Þ
s m
Here C0 is the classical term, of tunnelling either at the first or at the second point contact,
and C1 is the interference contribution. The latter has m possible values, determined by
N qh mod m. At any given time, then, the system is in a state in which, say, j quasi-holes have
tunnelled from the exterior edge to the interior one. The transition rates from that state to the
state with j 1 quasi-holes depend on j mod m. A transition from the state j to the state j 1
implies a transfer of 1 quasi-hole from one edge to another, i.e., a current.
The flow of current through the interferometer is a statistical process, and the current
average and noise may be calculated by means of rate equations. At zero temperature tran-
sitions are allowed only in one direction, determined by the difference of chemical poten-
tials between the edges. The probability of the system being in a state where j quasi-holes
have tunnelled by time t, denoted by P j ðtÞ, then satisfies,
dP j ðtÞ P j1 P j
¼ ð22Þ
dt sj1 sj
where 1=sj is the rate for a transition from the state j to the state j þ 1. The initial condi-
tion is P j ðt ¼ 0Þ ¼ dj;0 . The average current is hIihe
dj
dt
i ¼ e
hjðT Þjð0Þi
T
for a very long time T,
2 dj dj
and the zero-frequency noise is S x¼0 ¼ ðe Þ hdt dt ix¼0 . Here e ¼ e=m is the charge of the
and
Pm
dj dj 2 s2
¼ 2m Pmi¼1 i3 ð24Þ
dt dt x¼0 i¼1 si
Substituting the rates Eq. (21) into these expressions, we find that both the average cur-
rent and the current noise depend periodically on the flux, with a period of U0 . The ratio
S=2hIi, the Fano factor, that is frequently interpreted as the effective charge q, is
Pm 2
s
q ¼ e Pmi¼1 i2 ð25Þ
i¼1 si
For m ¼ 1, where quasi-holes are fermions, the effective charge is just the electron
charge. While both the current and the noise depend on the flux, their ratio does not.
In contrast, for the Laughlin fractions, with an odd m > 1, the effective charge generally
222 A. Stern / Annals of Physics 323 (2008) 204–249
depends on flux. This dependence is a direct consequence of the fractional statistics of the
quasi-holes. It originates from the fact that the tunnelling rate of a quasi-hole from one
edge to another depends periodically on the number of quasi-holes that have already tun-
nelled, and that dependence is a direct consequence of the geometric phase that a quasi-
hole accumulates when it encircles another one.
The effective charge in a Mach–Zehnder interferometer, Eq. (25) may be rather easily
understood in two cases. In the case where all the tunnelling rates are the same, the effec-
tive charge is the charge of the quasi-particle, e=m. In the case where the rate to tunnel out
of one state is significantly smaller than all others, the system will spend most of its time
waiting to tunnel out of that state. Once it does so, it very quickly goes through the m steps
until it gets back to that state. In that case, then, the effective charge will be m times larger
than the quasi-particle charge, and that would amount to the charge of the electron. All
other cases yield an effective charge between e=m and e. As we will see later, values of
the charge that are larger than e are indicative of non-abelian anyons.
The Mach–Zehnder interferometer exhibits also a relation between the visibility of the
Aharonov–Bohm oscillations, similar to that of (20), as verified by using Eqs. (23) and
(21). The current depends on the flux through IðUÞ ¼ I 0 þ I 1 cosð2p UU0 þ a0 Þ, with a power
law relation between I 1 and I 0 [32].
On a practical note it is worth mentioning that the increase of the visibility of the oscil-
lations with increasing current, which we find both for a noisy Fabry–Perot interferometer
and for a Mach–Zehnder interferometer, is opposite to the dependence one would expect
from the obvious effect of heating. The latter increases with increased current, and sup-
presses the visibility due to loss of coherence.
A different scheme for observing subtle signs of non-abelian statistics in current noise
was suggested in [36].
So far we dealt with anyons in the m ¼ 1=m states, basing our analysis on little more
than the experimental input of the fractional quantum Hall effect and general physics prin-
ciples. In this section we show how anyons emerge from composite fermion theory, which
is the most commonly used theoretical method to deal with the fractional quantum Hall
effect [4,37–42]. There are some reasons to do that. First, the name composite fermion the-
ory seems to suggest that one can understand the quantum Hall effect without ever wor-
rying about fractional statistics. Here we explain how quasi-particles are anyons also in
this theory. And second, composite fermion theory gives a theoretical tool for the treat-
ment of fractional quantum Hall states not from the m ¼ 1=m type. We will use this tool
to analyze the statistics of quasi-particles in these states [43].
So, in brief, what is composite fermion theory? The concept was first introduced by Jain
in a first-quantized way, suitable for use in numerical work ([40] and see Section 13). In
this Section we present it in a field theoretical way [39,41]. The starting point is the well
accepted Hamiltonian
Z
1
H ¼ d2 r jði$ AÞwðrÞj2 þ H int ð26Þ
2m
where H int is the Coulomb interaction part of the Hamiltonian, and AðrÞ ¼ 2ce B r. In this
Hamiltonian wðrÞ is the electronic annihilation operator. The goal of the theory is to
A. Stern / Annals of Physics 323 (2008) 204–249 223
where qðrÞ ¼ wy ðrÞwðrÞ ¼ wycf ðrÞwcf ðrÞ is the density operator. In terms of this operator the
kinetic part of the Hamiltonian acquires a new vector potential and becomes,
Z
1 2
H ¼ d2 r jði$ A þ aÞwðrÞj þ H int ð28Þ
2m
with $ a ¼ 2U0 qðrÞ. Figuratively, the transformation attaches two flux quanta to each
electron, transforming it to a composite particle that carries a charge e and two flux quan-
ta, and follows fermionic statistics. While this description is correct, it should be used with
caution, as we will shortly see.
On the face of it, the introduction of this vector potential just makes the problem
harder, since on top of their electrostatic interaction, the particles now interact also with
the flux tubes of one another. However, the transformation also opens the way for the
desired approximation, a Hartree mean field approximation. Replacing the dynamical vec-
tor potential aðrÞ by its expectation value maps the problem of electrons at a magnetic field
B to composite fermions at a magnetic field b ¼ B 2U0 n, with n the average density. That
mapping also introduces a correspondence between Landau level filling factors: the elec-
tronic filling factor me translates to a composite fermion filling factor mcf according to
m1 1
e ¼ mcf þ 2. In particular, the prominently observed series of fractional quantum Hall
p
states at filling fractions of me ¼ 2pþ1 maps onto an integer quantum Hall effect for the com-
posite fermions, with mcf ¼ p (Note that the number 2 in the exponent in Eq. (27) may be
replaced by any other even number for the description of states that do not belong to this
series, such as m ¼ 1=5 or m ¼ 2=3). The energy gap then becomes a natural consequence of
the filling of an integer number of composite fermions Landau levels. The Hall resistivity
of the composite fermions is quantized to be h=pe2 . Since a current of the composite ferm-
ions involves also a motion of their flux tubes, which creates a Chern-Simons transverse
electric field, the resistivity of the electrons and that of the composite fermions are related
by qexx ¼ qcf e cf 2 e 2
xx and qxy ¼ qxy þ 2h=e . This relation reproduces the correct qxy ¼ h=e m value.
With the featureless fractional quantum Hall liquid of m ¼ p=ð2p þ 1Þ being described as p
filled Landau levels of composite fermions, what would be the quasi-particles and quasi-
holes? The procedure we described above for creating a quasi-hole in a m ¼ 1=m state, by
the adiabatic turning on of a U0 flux tube, can be applied here as well, and will create a
quasi-hole with a charge of em ¼ ep=ð2p þ 1Þ. It is possible, however, to create a quasi-hole
with a smaller charge. That should come as no surprise: In the m ¼ n integer quantum Hall
effect, in a model where electron-electron interaction is neglected, the charge of the quasi-par-
ticle is clearly the charge of the electron. It is created by adding an electron to an otherwise
empty Landau level, or —for a quasi-hole—by removing an electron from an otherwise full
level. In contrast, the introduction of a flux quantum expels a charge larger by a factor of n.
To identify the charge of the quasi-hole or quasi-particle for the m ¼ p=ð2p þ 1Þ states
we repeat the same process, for composite fermions. Let us imagine adiabatically annihi-
224 A. Stern / Annals of Physics 323 (2008) 204–249
lating a composite fermion at the origin, at the lowest angular momentum state of one of
the filled Landau levels. This process of annihilation is done in two steps. First, the charge
is taken out, a charge of e. Then, a flux tube carrying 2U0 parallel to the external magnetic
field is introduced. Its adiabatic introduction introduces an azimuthal electric field, which
drives away a charge of 2pe=ð2p þ 1Þ, leaving a net charge of magnitude e=ð2p þ 1Þ as
the charge of the quasi-hole.
This process may look confusing at first sight: if the charge of a composite fermion is
the charge of the electron, how come that by annihilating a composite fermion we create a
lump of a fractional charge? The answer is that by adiabatically annihilating a composite
fermion in one Landau level we also affect the wave functions of composite fermions in the
other Landau levels. Let us look at this carefully: the annihilation of the composite fer-
mion takes away a charge q (which we are now set to determine again). Thus, it also varies
the mean-field effective magnetic field b ¼ B 2U0 n in the region from which the charge
was taken. Since a charge is taken out, b grows, and the magnetic flux in the region grows
by 2U0 q, making the charge in each of the filled Landau levels grow by 2q, since a filled
Landau level has a fermion per flux quantum. Altogether, then, one Landau level lost
the composite fermion that was taken out, but got an extra charge 2q because of the mag-
netic flux growing. The other p 1 levels got a charge of 2q each. The total charge then
satisfies
q ¼ 1 2q 2qðp 1Þ ¼ 1 2pq ð29Þ
There has recently been a surge of experimental activity in the field of mesoscopic quan-
tum Hall devices in general, and interferometry in particular (see for example [44–52]).
Interesting and only partially explained phenomena were seen in Mach–Zehnder interfer-
ometers of integer filling factors [51], but those presumably do not involve anyons. In a
series of beautiful experiments of Camino et al., devices of the Fabry–Perot type were fab-
ricated, and were measured in the integer and fractional quantum Hall regime. The results
A. Stern / Annals of Physics 323 (2008) 204–249 225
of these experiments are not yet fully understood, and several interpretations have been
put forward in subsequent theoretical works [36,53–57]. While we will not get into a
detailed discussion of these experiments here, we will describe the main results and com-
ment on several factors that are crucial for their interpretation.
Naturally, the place to start is with the integer quantum Hall effect. The experiment
measured the dependence of the back-scattered current on the magnetic field and the
electronic density, which was varied by means of a back-gate. Both dependencies were
oscillatory. The period of the oscillations with respect to the back-gate voltage was
independent of the integer filling factor f at which the measurement was carried out.
In contrast, the period with respect to the magnetic field was inversely proportional
to f.
These measurements reveal a major difference between the theoretical construct we
introduced above and its experimental realization: unlike in the theoretical construct,
the ‘‘cell’’ of the interferometer is confined by a smooth potential, and therefore may
break into several regions of different phases. One example to that is the center of the
cell being an ‘‘island’’ of a quantum Hall state of one filling factor surrounded by a
quantum Hall state of a different (typically lower) filling factor. Another example is
one in which the center of the cell is a compressible island, surrounded by a quantum
Hall state. Yet another is one in which the bulk both outside and inside the cell is in
one quantized Hall state but the region of the point contacts is in another. In all cases,
the bulk of the cell has a compressible region, either as an edge separating an island of
one quantized state from a bulk of another, or as a compressible island within a quan-
tized region. These compressible regions complicate the experiment in several ways.
First, being confined by insulating quantized Hall regions, their charge is quantized. Sec-
ond, they add indirect paths for tunnelling from one edge to another. And third, their
size is a degree of freedom that may vary as the magnetic field or back-gate voltage
are varied.
Both the flux and the charge periodicity of the back-scattered current in the integer
quantum Hall regime are understood in terms of Coulomb blockade physics of the com-
pressible region within the interferometer’s cell. As the flux within the island is varied by
1=f flux quanta, the occupation of the highest occupied Landau level (the one that is occu-
pied only within the island) changes by one electron, hence giving rise to a Coulomb block-
ade periodicity of 1=f . The periodicity in gate voltage is then naturally independent of the
filling factor.
Further measurements were carried out in the fractional regime, when the filling frac-
tion at the constrictions was m ¼ 1=3 and that at the island between the constrictions was
m ¼ 2=5. Oscillations of the conductance were observed again, with a flux period of five
flux quanta through the estimated area of the m ¼ 2=5 island and a charge period of
two electronic charges. It must be noted that the estimate of the size of the m ¼ 2=5 island
has a certain dependence on modelling, and is not directly measured. In any case, the any-
onic nature of the quasi-particles is probably involved in the determination of the periods
of the oscillations, but the precise way, and the role of the other factors, are not fully
understood yet. In fact, different models of the experimental system yield different period-
icities [54–56], none of which is presently able to fully account for the experimental obser-
vations. The models differ from one another in their identification of the dominant inter-
edge tunnelling route and by the way they account for the relative roles of Coulomb block-
ade physics and interference effects.
226 A. Stern / Annals of Physics 323 (2008) 204–249
So far we reviewed how the two-dimensional fractional quantum Hall effect extends the
notion of quantum statistics and introduces anyons, particles whose interchange phases
are between the zero of bosons and p of fermions. From now on we will present the next
level of extension, of particles satisfying non-abelian statistics, or non-abelian anyons
[9,12,58–65]. The first quantum Hall state suspected of being non-abelian is the m ¼ 5=2
state [9,66]. The study of this state has then led to the introduction of the Read–Rezayi
series, the series of spin polarized m ¼ 2 þ k=ðk þ 2Þ states, for which non-abelian theories
have been proposed [10]), and to other non-abelian states, including states that are spin-
singlets [67–69]. Numerical works indicate that the m ¼ 5=2 and m ¼ 12=5 ground state are
indeed non-abelian for a wide range of interaction parameters [70–74].
We will start this Section with the m ¼ 5=2 state, both because of its relative simplicity
and because of it being the most relevant to present days experiments, and will continue
with the way Conformal Field Theories are used to reason and analyze the more compli-
cated non-abelian states, taking the Read–Rezayi series as an example.
Let us start by expanding on the statement we made regarding the effect of the quasi-
hole’s vorticity on the dynamics of its coordinate. As we said, if the wave function of the
system with a quasi-hole at a coordinate R is jwðRÞi, then the kinetic energy that dictates
the dynamics of R includes a Berry vector potential
AB ¼ ImhwðRÞj$R jwðRÞi ð32Þ
Then, Rwhen the vortex traverses a closed loop the wave function accumulates a Berry
phase of AB dl.
There was an assumption in this statement, and its violation is the source of non-abe-
lian quasi-particles. The assumption was that once the positions of the quasi-holes are
fixed, there is just one ground state to the system. Under this assumption, when the param-
eters in that ground state, the positions Rs, adiabatically traverse a closed loop, the state of
the system must evolve from its initial ground state jwðRÞi to a final state that differs from
the original one by a phase factor only. This is the adiabatic theorem. A calculation then
shows that the phase factor is just the Berry phase mentioned above.
What if the assumption does not hold, and rather than having one ground state per
each configuration of quasi-holes we have several ground states that differ by some inter-
nal quantum numbers? Such a degeneracy may in principle appear accidentally, but here
we are interested in the case where the degeneracy originates from deep properties of the
quantum states.
At any rate, with such a degeneracy, if the initial state is a ground state the adiabatic
theorem guarantees that the final state will be a ground state as well, but it does not guar-
antee that it would be the initial state multiplied by a phase factor. Rather, under these
conditions, the geometric vector potential AðRÞ becomes a matrix,
ABij ¼ Imhwi ðRÞj$R jwj ðRÞi ð33Þ
with the jwi i being the various degenerate ground states. Then, the effect of a motion of the
parameters Rs along a closed loop is not a phase factor, but rather a unitary transforma-
A. Stern / Annals of Physics 323 (2008) 204–249 227
tion that acts within the subspace of degenerate ground states. This unitary transformation
is,
R
P^ ei AB dl ð34Þ
This description was very abstract, and an example would probably help to clarify the
idea. The most prominent example for non-abelian anyons are believed to be the quasi-
holes and quasi-particles of the m ¼ 5=2 fractional quantum Hall state, as realized first
by Moore and Read [9]. This state is also the easiest to analyze, using its mapping onto
a p-wave super-conductor of composite fermions [75–78]. Let us see how this comes about,
following closely the work of Read and Green [75]:
In the m ¼ 5=2 state two Landau levels are full and one is half full. The two full levels
are inert and for our discussion have no effect. The half filled Landau level may be
mapped, by means of the Chern-Simons transformation (27), onto a system of composite
fermions at an average of zero magnetic field, since the external field is cancelled by the
Chern-Simons field. At very low temperatures it then becomes energetically favorable
for the composite fermions, at least in some regime of interaction parameters, to form
Cooper-pairs and have them condense to form a super-conductor. If the electrons are
assumed to be spin polarized (which is definitely the case in the limit of infinite magnetic
field and density, and perhaps also for the experimentally relevant values) so are also the
composite fermions. Then they cannot form s-wave Cooper-pairs, and the simplest pairing
they may undergo is that of p-wave symmetry. In the absence of a spontaneous breaking of
rotational symmetry, the two possible p-wave types of pairing would be px ipy . To give
an intermediate summary of this line of argument, then, the m ¼ 5=2 quantum Hall state is
to be thought of as a px ipy super-conductor of composite fermions.
With that in mind, note the important difference between the effect of a slight increase
of the magnetic field (or a slight decrease of the density) on the m ¼ p=ð2p þ 1Þ and the
m ¼ 5=2 states: while for the former this change introduces vacancies in an otherwise full
Landau level, for the m ¼ 5=2 the residual magnetic field is accommodated by the super-
conductor in the form of vortices. These vortices are the non-abelian quasi-holes and
quasi-particles of the m ¼ 5=2 state, as we now see.
As seen in Eq. (28), the composite fermions are subjected to the vector potential A a.
The electromagnetic part of that vector potential is externally applied. The tiny currents
that flow in a two dimensional electronic system do not produce a significant magnetic
field. In contrast, the Chern-Simons vector potential is dynamical, i.e., created by the sys-
tem itself. As in a super-conductor the vortex currents are screened by one half of a flux
228 A. Stern / Annals of Physics 323 (2008) 204–249
quantum of the gauge field a. Since two flux quanta correspond to one electron charge,
half a flux quantum corresponds to one quarter of a charge. The quasi-hole/quasi-particle
are quarter charged, then.
But this is not the end of the story. The next piece of it is the degeneracy of the ground
state in the presence of these quasi-holes. To see that we need to analyze the px þ ipy -wave
super-conductor in some more detail (the choice of the relative sign is motivated by the
need to obtain a ground state wave function of the lowest Landau level). The most straight
forward way of analyzing a super-conductor is through the BCS mean field Hamiltonian,
Z Z
y 1
H ¼ dr w ðrÞh0 wðrÞ þ dr dr0 D
ðr; r0 Þwðr0 ÞwðrÞ þ Dðr; r0 Þwy ðrÞwy ðr0 Þ ð35Þ
2
with the single-particle term h0 and the complex p-wave pairing function
r þ r0
Dðr; r0 Þ ¼ D ðiox0 oy 0 Þdðr r0 Þ: ð36Þ
2
The dynamics of D is governed by a Landau–Ginzburg-type Hamiltonian. The qua-
dratic Hamiltonian (35) may be diagonalized by solving the corresponding Bogolubov-
de-Gennes equations, and following its diagonalization may be written as
X y
H ¼ E0 þ ECE CE ð37Þ
E>0
R
where CyE dr½uE ðrÞwðrÞ þ vE ðrÞwy ðrÞ is the creation operator formed by the positive en-
ergy solutions of the Bogolubov deGennes (BdG) equations,
i
!
uðrÞ lðrÞ 2
DðrÞ; ox þ ioy uðrÞ
E ¼ i
; ð38Þ
vðrÞ 2
D ðrÞ; ox ioy lðrÞ vðrÞ
and E0 is the ground state energy. As becomes clear from Eq. (37), for the ground state of
the Hamiltonian (35) to be degenerate it is essential that the BdG equations have zero
eigenvalues.
Such zero energy eigenvalues appear when the super-conductor has vortices. Vortices
are introduced into the Bogolubov-de-Gennes equations through D. A vortex at the point
R implies a winding of the phase of D by 2p for every trajectory that encircles R, and a
vanishing of D at the point R itself. Assuming, for simplicity, azimuthal symmetry,
DðrÞ ¼ jDðjr RjÞjeihþiX , where h is the angle of the vector r R relative to the x-axis,
and X is the phase of the order parameter along the h ¼ 0 line.
When a vortex configuration for D is considered, and when the chemical potential l is
assumed larger than zero, the Bogolubov-de-Gennes equations are found to possess a sin-
gle zero energy solution, localized close to the point R, of the form
Z h i
1 i i
c ¼ pffiffiffi dr F ðrÞ e2X wðrÞ þ F
ðrÞ e2X wy ðrÞ ð39Þ
2
Here, F ðrÞ decays for large r.
When there are several well separated vortices at positions Ri , the gap function near the
P vortex takes the form DðrÞ ¼ jDðrÞj expðihi þ iXi Þ, with hi ¼ argðr Ri Þ and Xi ¼
ith
j6¼i argððRj Ri ÞÞ. There is one zero energy solution per vortex, and we correspondingly
add an index to c. Each zero energy solution ci is localized near the core of its vortex at Ri ,
A. Stern / Annals of Physics 323 (2008) 204–249 229
but the phase Xi that replaces X in (39) depends on the position of all vortices. Moreover,
the dependence of ci on the positions Ri is not single valued. If the angle of the vector
Ri Rj with respect to the x-axis is changed by 2p then ci acquires a minus sign. These
minus signs will play an important role later.
The operators ci have a few properties that we need to dwell on. First, as is obvious
from (39), these operators are hermitian, ci ¼ cyi . Second, they mutually anti-commute.
With the proper normalization, fci ; cj g ¼ 2dij . These two properties make the ci ’s Major-
ana fermions. Since c2i 6¼ 0 and ðcyi Þ2 6¼ 0, we cannot talk about the ci ’s as being ‘‘empty’’
or ‘‘full’’ as we are so used to when talking about fermions. We need to be more careful
when counting the dimensionality of the Fock space that is spanned by these fermions. To
that end, we first note that the number of zero energy states must be even: for a compact
geometry, Dirac’s quantization of the monopole enforces the number of vortices going
through the boundary-less surface to be even. For a non-compact geometry, if the bulk
of the system has an odd number of vortices, the edge will have a zero mode. If the bulk
has an even number of vortices, the edge will not have one of its own [75]. Thus, there is
always an even number of ci ’s, with i ¼ 1 . . . 2N . To properly count the dimensionality of
the Hilbert space they span, it is convenient to define complex fermionic operators,
Cj cj icjþN ð40Þ
Cyj cj þ icjþN ð41Þ
with j ¼ 1 . . . N . The N pairs of fermionic operators Cj ; Cyj
satisfy the anticommutation rela-
tions we are used to from ‘‘conventional’’ fermionic operators, and hence define N fermionic
modes. In particular, each of these modes satisfy C2j ¼ ðCyj Þ ¼ 0, and hence may be referred
2
to as ‘‘empty’’ and ‘‘full’’. The dimension of the Hilbert space they span is therefore 2N , and a
basis that spans it can be described in terms of binary numbers of N digits, with 0 denoting an
‘‘empty’’ mode, and 1 denoting a ‘‘full’’ one. Note that the BCS mean field Hamiltonian does
not commute with the particle number operator, but does commute with the parity of that
number. Thus, its eigenstates may be characterized by the parity of the number of particles
they contain. Since the operators Cj and Cyj change that parity (each is a superposition of cre-
ating and annihilating a particle), half of the 2N ground states have an even number of par-
ticles, and the other half have an odd number of particles.
This subspace of ground states is an interesting Hilbert space. The basis we chose is
obviously arbitrary, being based on an arbitrary enumeration of the vortices. A different
enumeration is just a different basis spanning the same subspace. But note how unique
these states are: suppose that we look at a ground state jwi where all fermionic modes
are ‘‘empty’’, Cj jwi ¼ 0 for all j. From Eq. (41) we find that,
cj jwi ¼ icjþN jwi ð42Þ
Noting the structure (39) of the c operators, this relation tells us something interesting
about the ground state jwi. The vortices j and j þ N may be arbitrarily far from one another.
Yet, when we operate on jwi with cj , which is an operator that creates and annihilates a par-
ticle at the neighborhood of the vortex j, we get essentially the same state (up to an eip=2 phase)
that we get when we operate with cjþN , which is an operator that creates and annihilates a par-
ticle at the neighborhood of the vortex j þ N . Clearly, this state involves some long range cor-
relations between the local occupation of single particle states near the cores of the vortices.
How can we move around within this subspace of degenerate ground states? Can we
design some kind of a perturbation to the Hamiltonian that, if tuned properly, will take
230 A. Stern / Annals of Physics 323 (2008) 204–249
us from one ground state to another? That is a very subtle question. Generally speaking,
we could think of several types of perturbations. The first is a perturbation that takes one
of the particles in the super-conductor and moves it from one single-particle state to
another. An example to that would be an interaction of the composite fermion with an
externally applied electro-magnetic field that does not involve a flip of the spin. The second
is a perturbation that scatters a composite fermion (or an electron) out of the super-con-
ductor, for example by a spin-flip. And the third, the most interesting one, involves an adi-
abatic motion of the vortices. Remarkably, a perturbation of the first type would have no
effect on the ground state subspace. Typically, a perturbation of this sort looks like,
Z Z
H pert ¼ dx dx0 F ðx x0 Þwy ðxÞwðx0 Þbðx; x0 Þ ð43Þ
with the range of F ðx x0 Þ being very short and with b being some kind of a bosonic oper-
ator (e.g., creation and annihilation of a photon or a phonon). Most generally, the projec-
tion of H pert onto the ground state subspace is
X
H gs
pert ¼ hij ci cj ð44Þ
ij
When the wy ðxÞ and wðx0 Þ are expressed as a superposition of CE and CyE of Eq. (37),
the proximity of the points x; x0 to one another implies that both will have a significant
overlap with the same Majorana operator ci . As a consequence, in Eq. (44), as long as
the vortices are well separated from one another, hij / dij , making Eq. (44) nothing but
a c-number (remember that c2i ¼ 1 for all is). Thus, as long as it is local, the perturbation
does not induce transitions from one ground state to another and does not remove the
degeneracy of the ground states.
The stability of the degeneracy of the ground state with respect to local perturbations
may also be understood from the point of view of the Bogolubov-deGennes equations.
Since the spectrum of Eq. (38) is symmetric with respect to E ¼ 0, for a perturbation to
affect the degeneracy of the ground states it must mix two E ¼ 0 solutions of the BdG
equations into two solutions of non-zero energiesE. For that to happen, however, the
perturbation must have matrix elements that couple the two zero energy solutions. Since
those are localized within vortex cores, as long as vortices are well separated the removal
of the degeneracy will be exponentially small.
Having the super-conductor shift from one ground state to another by a perturbation
that changes the parity of the number of particles in the super-conductor is easier to do.
For example, consider a perturbation of the type (43) in which the spin state that is created
by wy is opposite to that annihilated by the w. Such a perturbation, when translated to the
BdG operator, would involve only a single BdG operator, and thus would not assume the
form (44). Note, however, that if the spin polarized super-conductor is the ground state of
the system, taking a particle out of it may involve an energy cost, and thus put the entire
system in an excited state.
The most interesting way to evolve the super-conductor from one ground state to
another is through braiding of vortices [79–81]. Look first at the case where vortex i encir-
cles vortex j: independent of the geometry of the trajectory taken by the vortices, when
that happens argðri rj Þ changes by 2p. Hence both Xi and Xj change by p, and both
ci and cj are multiplied by 1. Note, these are changes in operators. The unitary transfor-
A. Stern / Annals of Physics 323 (2008) 204–249 231
mation that operates on the ground state when vortex i encircles vortex j, call it U ij , can be
regarded as transforming the operators ck according to
ck ! U yij ck U ij ð45Þ
The requirement that this transformation multiplies ci and cj by 1 while leaving all
other c’s unaltered is enough to fix U ij up to an abelian phase to be,
U ij ¼ ci cj ¼ exp pci cj =2 ¼ ci cj ð46Þ
The effect of vortex i going around vortex j must be identical to the effect of two inter-
changes of vortices i and j, and thus it is not surprising that (up to an abelian phase again)
the unitary transformation that corresponds to an interchange is the square root of U ij
namely, exp pci cj =4, where the sign is determined by the sense of the interchange. Uni-
tary transformations that correspond to different interchanges do not necessarily commute
(in fact, two such transformations do not commute if they share one vortex), and hence the
vortices satisfy a non-abelian statistics.
So to summarize this section, the combination of a degenerate ground state subspace
and a topological unitary transformation that is applied to the state of the super-con-
ductor whenever vortices braid makes the vortices satisfy a non-abelian statistics.
Remember that the super-conductor we think of is the m ¼ 5=2 state and the vortices
are the quasi-particles and quasi-holes of that state. Would similar phenomena occur
in two dimensional superfluid He-3 of the px ipy pairing, or in two dimensional
super-conductors of that symmetry? We will not get into a detailed discussion of this
question here, but merely comment on two important differences between the quantum
Hall m ¼ 5=2 state and the latter two examples: first, as we commented earlier, two
dimensional super-fluids and super-conductors have a gapless bulk mode in the form
of a charge density wave. And second, their vortices do not carry electrical charge
and therefore cannot be manipulated with the help of a magnetic or electric field.
The Majorana intra-vortex states and the ground state degeneracy they create are both
there and have measurable consequences, but the question of the statistics of the vorti-
ces is more subtle.
11. The geometric phases and quantum entanglement behind the non-abelian statistics
Before we look into the way non-abelian statistics may be observed through interferom-
etry, we will dwell a little bit on the meaning of the unitary transformations that are
applied when vortices interchange, or encircle one another. These transformations were
first worked out using Conformal Field Theory by Nayak and Wilczek [80], and then using
the p-wave super-conductor description described above by Ivanov [79] (see also [81–84]).
Look again on the unitary transformation that results from vortex j encircling vortex j þ 1
(Eq. (46)). Written in full detail, it is
i i
i i
cj e2Xj þ cyj e2Xj cjþ1 e2Xjþ1 þ cyjþ1 e2Xjþ1 ; ð47Þ
ðyÞ ðyÞ
where the operators cj ; cjþ1 annihilate (create) a composite fermion localized very
close to the cores of the jth and ðj þ 1Þth vortex, respectively. Eq. (47) seemingly im-
plies that the motion of the jth vortex around the ðj þ 1Þth vortex affects the occupa-
tions of states very close to the cores of the two vortices. This is in contrast, however,
232 A. Stern / Annals of Physics 323 (2008) 204–249
to the derivation leading to Eq. (47), which explicitly assumes that vortices are kept far
enough from one another so that tunneling between vortex cores may be disregarded.
How can that happen [81]?
In fact, no tunnelling takes place. Rather, as we now explain, two ingredients are essen-
tial for the nonabelian statistics of the vortices. The first is the quantum entanglement of the
occupation of states near the cores of distant vortices. The second ingredient is familiar
from our discussion of abelian anyons: the geometric phase accumulated by a vortex tra-
versing a closed loop.
As we saw, the anyonic statistics in the m ¼ 1=m state is related to the geometric phase
accumulated by a vortex traversing a close trajectory. Roughly speaking, the vortex accu-
mulates a phase of 2p per fluid particle which it encircles. When another vortex with its
fractional charge is introduced to the encircled area, this phase changes by a fraction of
2p, due to the fractional charge carried by the vortex. Upon adapting the argument to
interchanging of vortices, one finds that this fraction of 2p translates into fractional
statistics.
Similarly, the Moore-Read theory of the m ¼ 5=2 state describes it also as a superfluid,
with the quasiparticles being vortices in that superfluid. However, the ‘‘effective bosons’’
forming the superfluid are Cooper pairs of composite fermions. Consequently, the super-
fluid has excitation modes associated with the breaking of Cooper-pairs. In the presence of
vortices, a Cooper-pair may be broken such that one or two of its constituents are local-
ized in the cores of vortices. For pwave superconductors, the existence of zero-energy
intra-vortex modes leads, first, to a multitude of ground states, and, second, to a parti-
cle-hole symmetric occupation of the vortex cores in all ground states. When represented
in occupation-number basis, each of the ground states we deal with is a superposition
which has equal probability for the vortex core being empty or occupied by one composite
fermion. What distinguishes ground states from one another is the relative phases between
the different components of the super-position.
When a vortex traverses a trajectory that encircles another vortex, the phase it accu-
mulates depends again on the number of fluid particles that it encircles. Since a fluid
particle is in this case a Cooper pair, the occupation of a vortex core by a fermion, half
a pair, leads to an accumulation of a phase of p relative to the case when the core is not
occupied. It is this relative phase of p that the encircling introduces between the different
components of the wave function that might transform the system from one ground
state to another.
This line of thoughts leads to the following distinction between three types of fractional
quantum Hall states. The Laughlin m ¼ 1=m fractions are condensates of one type of par-
ticle, the composite bosons formed by the attachment of m flux quanta to each electron
[85]. The Jain states are several condensates living in parallel, each condensate being
formed of a different type of composite boson. The non-abelian states are bosons made
up of several electrons with flux attached to them. The most prominent example of the
third type is the m ¼ 5=2 state, but this state is followed by a whole series, the Read–Rezayi
series. In Section 13 we will use this line of thoughts to describe the Read–Rezayi series.
As we described above for abelian anyons, interferometers seem to be the most direct
probe of the topological interaction between anyons, since it is a geometry where winding
A. Stern / Annals of Physics 323 (2008) 204–249 233
of particles around one another is inherent in the way transport takes place. Moreover, it
has recently been shown that, if realized, a Fabry–Perot interferometer of non-abelian
anyons may be used as a topologically-protected qubit [86], with an astonishingly low
error rate. In this section we revisit the two geometries we discussed above with the goal
of analyzing how they manifest the unique properties of the non-abelian m ¼ 5=2 state.
Again, we consider the Fabry–Perot geometry and analyze the dependence of the back-
scattered current on the area of the cell between the two point contacts and on the mag-
netic field. The magnetic field determines the number of quasi-holes localized within the
island and the area of the cell determines the number of fluid particles encircled by the
interference loop. The effect of the localized quasi-particles in the non-abelian case is more
subtle than in the abelian one, since the edge quasi-hole that encircles them may modify
the quantum state they are in [87–92].
Let us see how this happens: suppose that N qp quasi-particles are localized in the cell in
between the two point contacts of Fig. 4a, and that their quantum state is jwi i. Transport
current flows rightwards on the lower edge, and may be back-scattered to flow leftward on
the upper edge. To lowest order, the back-scattering probability is the interference of two
amplitudes, of back-scattering by the left and by the right quantum point contact. The
partial wave that is back-scattered by the left point contact does not encircle the cell
and thus does not modify the state jwi i. By Eq. (46) the partial wave that is back-scattered
by the right point contact applies a unitary transformation on jwi i, whose form is, up to a
N QN qp
phase, ca qp j¼1 cj , where cj (j ¼ 1 . . . N qp ) are the Majorana modes associated with the
localized quasi-holes and ca is the Majorana mode of the quasi-hole that is being back-
scattered. Clearly, there is a difference between even and odd N qp . When N qp is even,
the interfering waves take the form,
t1 jwi i þ t2 ccell jwi i ð48Þ
Qnis
where ccell j¼1 cj ; and t1 ; t2 are again the amplitudes for tunnelling at the two point
contacts. The interference term is then
2Re½t
1 t2 hwi jccell jwi i ð49Þ
There are two possible eigenvalues to ccell , either 1 or i, depending on whether N qp is
divisible by four. Thus, as the area of the cell is varied, there are two possible interference
patterns, that differ by a p phase shift. If jwi i is an eigenvector of ccell the observed inter-
ference pattern would follow the corresponding eigenvalue. If that is not the case, the
back-scattered current would measure the value of ccell and one of the two interference pat-
tern would emerge. For an even N qp there are 2N qp =2 ground states, characterized by N qp =2
quantum numbers, each taking one of two possible values. The Fabry–Perot interferom-
eter measures just one of these numbers.
As shown by Das Sarma et al. [86], the Fabry–Perot interferometer, if brought to work
for the m ¼ 5=2 state with an even N qp , may be used as a qubit, with the two possible values
of ccell being the two states of the qubit. To flip the qubit from one value of ccell to another,
a quasi-hole should tunnel through the cell, encircling an odd numbered subset of the N qp
localized quasi-holes. This qubit is protected from the sources of dephasing that are so
234 A. Stern / Annals of Physics 323 (2008) 204–249
when N qp is even vm ðN qp Þ ¼ mp
2
according to the eigenvalue of ccell . When N qp ¼ 2j þ 1
(with j an integer) then vm ¼ jp. The calculation of the prefactors I m becomes complicated
for large m, but is not needed for our purpose.
Similar to the abelian case, when the number N qp fluctuates in time the visibility of the
Aharonov–Bohm oscillations develops a power-law relation with the average current, due
to the averaging of the harmonics m in (50) that do not divide by 4. In that case the vis-
ibility increases with the increase of the average current, and the fluctuations in N qp result
in a current noise [93].
Again, as for the abelian case, the main difference between the Fabry–Perot and Mach–
Zehnder interferometers is in the fact that for the former N qp is a parameter whose dynam-
ics is (at least as a matter of principle) independent of the current flowing in the interfer-
ometer, while for the latter N qp changes by one with each tunnelling quasi-hole. Similar to
the abelian case, the time evolution of the system is a stochastic process, but now the state
of the system is not characterized only by the number of quasi-holes that have tunnelled
from the exterior to the interior, but also by the state of the system: when N qp is even there
are two possible states, and a different transition rate corresponds to each one of them.
This state of affairs can be diagrammatically expressed by means of Brattelli diagrams,
see Fig. 5. The x-axis of the diagram is N qp . The y-axis is the state of the cell, i.e., the inter-
ference pattern it shows (this is a rather careless definition of the y-axis. A more elaborate
one will be given in the Section 13). For reasons that have to do with conformal field the-
ory, and will be explained below, the state of odd N qp is commonly called r, while the two
A. Stern / Annals of Physics 323 (2008) 204–249 235
Fig. 5. A Brattelli diagram as a way to describe interference at m ¼ 5=2. The x-axis counts the number of quasi-
particles in the interference loop. The y-axis describes the state of the quasi-particles. For an even N qp there are
two possible interference patterns, mutually shifted by p, and the interferometer may be utilized as a qubit [86].
For N qp there is zero amplitude of interference in lowest order of tunnelling. An ideal Fabry–Perot interferometer
stays on one of the nodes indefinitely. A noisy Fabry–Perot interferometer, where N qp fluctuates, diffuses on the
Brattelli diagram. A Mach–Zehder interferometer propagates on the diagram: each tunnelling quasi-particle
changes N qp by one. The notation 1; w; r is explained in the text.
states of even N qp are called 1 and w (or sometimes 1 and ). When N qp is odd, the system is
in the state w and may go to either 1 or w. The transition rates to these states differ from
one another since the relative phase of the two interfering waves depends on the final state.
When N qp is even, there are two possible states the system may be in (1 and w), and one
state to which it may go. Again, two different scattering rates. It turns out that altogether
there are four transition rates, which depend on the number N qp mod4 also due to the abe-
lian phase accumulated around the interference loop. The detailed calculation of the var-
ious transition rates is easiest to carry out using the techniques that will be explained
below, in Section 13 [94].
The evolution of the state of the system is determined by a set of equations similar to
(22), with results that are similar in spirit, but different in important details: the current is
periodic in the flux through the interferometer, with a period of U0 . The visibility scales as
a power law of the average current, with the power being unique to the m ¼ 5=2 state. And
the effective charge (Fano factor) measured in a shot noise measurement depends on the
flux as well [33].
The range of effective charges (Fano factors) to be seen in various values of the flux is a
signature of the m ¼ 5=2 state. The dependence of the effective charge on the flux is, as we
saw for the abelian case, a consequence of the existence of several scattering rates. Similar
to the abelian case, two cases are rather easy to analyze. When all transition rates are the
same, the effective charge is that of the quasi-hole, namely e=4. When one transition rate is
significantly smaller than all others, again the effective charge will be larger, since the sys-
tem will be stuck for long times in the state from which it is hard to move on, and then
move with a burst of steps until the next time it gets stuck. The structure of the Brattelli
diagram allows the system to ‘‘bypass’’ the state from which the transition rate is small if
that rate happens on a transition out of r, and thus does not limit the number of steps in
236 A. Stern / Annals of Physics 323 (2008) 204–249
the ‘‘burst’’. This structure allows for an effective charge that is larger than the electronic
charge. In fact, in the case of the m ¼ 5=2 state the largest Fano factor should be higher
than three electronic charges.
Another scheme that uses current noise for the observation of signs of non-abelian sta-
tistics was suggested in [95].
When the two quantum point contacts in a Fabry–Perot interferometer (Fig. 4a are
tuned to strong back-scattering, the cell between the point contacts becomes effectively
a quantum dot, where the number of electrons is quantized. Equivalently, in this regime
the conductance of the interferometer may be viewed as determined by multiple windings
of the cell, induced by multiple reflections by the point contacts. Rather than having a
sinusoidal dependence of the back-scattered current on the area of the dot, we then expect
to have most of the current back-scattered by the point contacts, except in those values of
the area where resonant tunnelling takes place and all the current is transmitted through
the dot. Those values correspond to area where the ground state energy of the dot with N
and with N þ 1 electrons is degenerate.
In view of this condition, both for integer quantum Hall states and for abelian frac-
tional quantum Hall states we expect the difference in area between two consecutive
resonances to be 1=n0 , where n0 is the electronic density in the dot. This prediction
is little more than the observation that when the dot has several thousand electrons,
the physics of the degeneracy point between the different nearby values of N is identi-
cal. One needs to increase the area of the dot by 1=n0 in order for it to accommodate
another electron.
The non-abelian m ¼ 5=2 state is different, since in that state electrons are paired. Thus,
we may wonder whether the periodicity of the peaks on the area axis would be the area
needed for accomodating another pair, or that needed for accommodating another elec-
tron, 2=n0 or 1=n0 (here n0 is the density of electrons in the half filled uppermost Landau
level). A similar question arises, of course, for super-conductors, where the peaks are
scanned on the axis of a gate voltage rather than area. For super-conductors, the spacing
between peaks indeed alternate, depending on whether the added electron has an electron
in the dot to pair with, or whether it remains unpaired. The spacings in the two cases differ
by the energy gap of the super-conductor.
In the m ¼ 5=2 case, the answer depends on N qp [88]. If N qp is odd, there is a zero
energy mode on the edge of the dot. This is, in the language of a super-conductor, a
mid-gap state that is available for the tunnelling electron, since it lies at the edge. Thus,
in this case the spacing between transmission peaks will not alternate, and the periodicity
in area will be 1=n0 . In contrast, this zero energy state is absent from the edge when N qp
is even. As long as there is no edge-bulk coupling the tunnelling electron cannot make
use of the zero energy states in the bulk, and the resonant peaks alternate: an odd elec-
tron must reside on the edge, occupying the lowest energy edge mode, while an even elec-
tron forms a Cooper-pair and is absorbed by the bulk. Thus, when N qp is even, the
periodicity in area becomes 2=n0 . Even when the measurement is of a transmission
through a quantum dot, where the number of electrons is quantized, the combined
dependence on area and magnetic field carries a signature of the non-abelian nature of
the quasi-holes.
A. Stern / Annals of Physics 323 (2008) 204–249 237
The m ¼ 5=2 state is the simplest example of a non-abelian quantum Hall state. There
are several ways to approach the others. A possible starting point, which we choose here,
is that of trial wave functions. Despite the obvious difficulty in constructing a good
approximation for an eigenstate wave function of a system of about 108 interacting elec-
trons, trial wave functions led to enormous progress in the study of the fractional quantum
Hall effect, as we will review. Other starting points, originating from the non-abelian
Chern-Simons theories and from quantum groups, are covered extensively in [12] and [96].
Our journey towards non-abelian quantum Hall states on the trial wave function route
starts early, with the construction of the ground state wave function for a system of non-
interacting electrons at m ¼ 1. These electrons would fill up the lowest Landau level, with
all spins polarized, and would form a many-body wave function that is just the Slater
determinant of all single particle lowest Landau level states. In the symmetric gauge the
single particle states are characterized by their angular momentum m. When 2
the ith elec-
tron occupies the mth state its single particle wave function is zmi exp jz4li 2j and the many-
H
body Slater determinant becomes,
wm¼1 ðfzi gÞ ¼Detzmi exp jzi j2 =4l2H ð51Þ
Y Y
¼ ðzi zj Þ exp jzi j2 =4l2H ð52Þ
i<j i
When a is an odd integer, this is a Laughlin wave function, but we will think here of a as
a parameter to be varied. Generally, the value of a determines the filling factor. This may
be understood either by observing that the probability distribution for finding a single
2aðN 1Þ jzj2 =2l2
electron at a distance jzj very far from the origin scales like jzj e H , with N being
the number of electrons, thus making the effective area of the droplet scale like aN and the
2
filling factor scale as 1=a, or by means of the plasma analogy. Writing jwðfzi gÞj
exp H, with
X X 2
H ¼ 2a logðzi zj Þ jzi j =2l2H ; ð54Þ
i<j i
the function H can be regarded as the classical Hamiltonian of a fictitious two dimensional
plasma, and the electronic probability distribution as the partition function corresponding
to that plasma. In this two-dimensional plasma the fictitious particles, whose coordinates
are zi , mutually interact through a a logðzÞ interaction, and interact with a uniform back-
238 A. Stern / Annals of Physics 323 (2008) 204–249
ground density (that in two dimensions creates a quadratic potential). The condition of
charge neutrality then fixes the density to be 1=2pal2H and the filling factor to be 1=a.
If we combine together: (a). the derivation of Eq. (52) as the exact ground state for
non-interacting electrons at m ¼ 1, (b). the observations made above regarding the wave
function (53), and (c) the requirement that for filling factors smaller than one, in the limit
of a very large cyclotron energy, the ground state wave function should be constructed of
states of the lowest Landau level only; And if we add to this list a lot of Laughlin–inspired
hindsight, then it becomes almost natural to use the wave function (53) with a ¼ 3; 5 as
trial wave functions for the m ¼ 1=3; 1=5 ground states. As long as a is an odd number,
the wave function (53) is of fermionic symmetry and describes—as seen above using the
plasma analogy—a droplet of filling factor m ¼ 1=a. Its exceptional success in describing
the actual m ¼ 13 ; 15 states may be ascribed to the clever use it makes in its zeroes. Other than
the exponential factors, this wave function is a polynomial in the zi ’s. The largest power it
has for each zi is determined by the filling factor to be of order of aN . Thus, this is also the
number of zeroes for each zi ’s. Surely, there must be a zero whenever two electrons are in
the same position, due to the Pauli principle of Fermions. That forces the position of
N 1 zeroes per each coordinate, but does not force the position of the other
ða 1ÞðN 1Þ zeroes per coordinate. The Laughlin wave function devotes all its zeroes
to keeping electrons away from one another, and thus minimizes their probability for
being close to one another. As the distance dz between two electrons gets smaller and smal-
2a 2
ler, the probability to find them at that distance decreases as jdzj , faster than the jdzj
forced by the Pauli principle.
What happens, however, when m ¼ 1=a and a is not an odd integer? There are two
approaches that have been taken for constructing wave functions for this case: Jain’s com-
posite fermions approach and Read–Rezayi’s clustering approach. The composite ferm-
ions approach leads to a series of abelian quantum Hall states, while the clustering
approach leads to a series of non-abelian states. For those filling fractions for which both
abelian and non-abelian trial wave functions may be constructed, the details of electron-
electron interaction determine whether one of the two, and which one, is indeed the phase
of lowest energy.
Jain’s approach, which has been the impetus for the development of the composite
fermions field theory described above in Section (8), starts from the following decomposi-
tion of wa for any odd a and m ¼ 1=a:
Y a1
Y 2
Y 2
wa ¼ ðzi zj Þ exp 2pnjzi j ða 1Þ ðzi zj Þ exp 2pnjzi j ð55Þ
i<j i i<j
2
Given that H ¼ 2pna, where n is the electronic density, this breaks the wave function
into two factors. The first one, to the left of thesign, attaches an even number ða 1Þ
of flux quanta to each electron, transforming it into a composite fermion. The filling factor
of the composite fermions is 1 and the second factor describes their occupation of a single
Landau level of a reduced magnetic field. Based on this observation, Jain constructed the
wave function for p filled Landau levels of composite fermions, corresponding to a filling
p
factor of ða1Þpþ1 by replacing the single Landau level wave function, the second line in (55),
by the wave function for p filled Landau levels, and then projecting the resulting wave
function into the lowest Landau level. The resulting wave functions are identical in their
topological properties to those obtained from the field theoretical approach for composite
fermions. Specifically, the quasi-particles and quasi-holes follow abelian statistics.
A. Stern / Annals of Physics 323 (2008) 204–249 239
Read–Rezayi’s clustering approach, which is our focus in this Section, starts by consid-
ering the wave function (53) for a general value of a. This wave function describes a drop-
let of the correct filling factor, but—lamentably—for an even a it does not have the proper
fermionic symmetry, and for a non-integer a it is not even single-valued. As it stands, then,
Eq. (53) is not an acceptable trial wave function for electrons in a general filling factor
m ¼ 1=a.
Can this be fixed? What does it take to fix it? We look for a trial wave function of the
form,
Wðfzi gÞ ¼ wa ðfzi gÞF ðfzi gÞ ð56Þ
where the function F will fix all the harm done by a non-odd value of a, without varying
the filling factor. For the relatively easy case of an even a, we need to find a function
F ðfZ i gÞ that is odd with respect to interchanging any of its two arguments, yet does
not involve powers of zi that are of order N. For the harder case of a non-integer a the
function F should guarantee the single-valuedness and fermionic symmetry of W. To do
b
that, it should scale like ðzi zj Þ when zi ! zj , with a b an odd positive integer.
Yet, again, it should not involve high powers of each of the coordinates. The Read–Rezayi
approach searches for such a function.
Following the rationale of flux attachment as a means to map problems of electrons in a
magnetic field onto composite particles at a different magnetic field, we may view, at least
roughly, the wave function wa (Eq. (53)) as a flux attachment transformation that attaches
a flux quanta to each electron. For a filling fraction m ¼ 1=a that makes the composite par-
ticles experience zero magnetic field on average. Thus, in Eq. (56) W is the electronic wave
function, wa is the flux attachment transformation, and the function F is the wave func-
tion of the composite particles, that carry one electron charge and a flux quanta, and are
subjected to zero average magnetic field. For an odd a the electrons are turned into bosons
at zero average magnetic field, whose wave function is just a constant. This viewpoint of
the Laughlin fractions (as well as the m ¼ 1 completely filled lowest Landau level), com-
monly known as composite boson theory [85], views the Laughlin fractions as Bose con-
densates, characterized—as all condensates—by the dissipationless flow of currents. For
an even a the electrons are transformed into composite fermions. The function F is then
supposed to be the function of spin polarized fermions at zero field, and as we saw above,
they may form a p-wave super-conductor. The p-wave super-conductor is again a conden-
sate. This time the condensate is formed by Cooper-pairs, a cluster of two fermions that,
effectively, group together to form a boson.
For a non-integer a the flux attachment turns electrons into anyons, and the function F
is to be the wave function of anyons at zero magnetic field. For these anyons to form con-
densates, where current flows with no dissipation, they have to form effective bosons,
made of clusters of an integer number k of anyons. If the phase accumulated when one
anyon encircles another is 4p=k, the phase accumulated by an anyon encircling a cluster
of k anyons will be 4p, and that accumulated by two clusters encircling one another will
be a multiple of 4p. The cluster will then be an effective boson. Thus, for forming a con-
densate of this sort, the flux attachment transformation should attach ð1 þ 2k Þ flux quanta
to each electron, which would make the electronic filling factor k=ðk þ 2Þ.
When the Laughlin fractions m ¼ 1=m are described as condensates of bosons, the
Laughlin quasi-particles are the quantized vortices in this condensate. To form a quan-
tized vortex at the point R in a superfluid of bosons we need to award each boson an angu-
240 A. Stern / Annals of Physics 323 (2008) 204–249
lar momentum of h relative to the point R. When the boson is made of two fermions, the
relative angular momentum per fermion is h=2. When the fermion goes around the vortex,
then, it accumulates a phase of p. In analogy, when the boson is made of k anyons, each
anyon gets, on average, an angular momentum of h=k and an anyon that goes around a
vortex accumulates a phase of 2p=k.
The arguments of the last two paragraphs tell us that there should be a description of
the m ¼ k=ðk þ 2Þ fractional quantum Hall state as a condensate of effective bosons made
of clusters of k anyons. Within this description an anyon that goes around another would
accumulate a phase of 4p=k while an anyon that goes around a vortex in the condensate
would accumulate a phase of 2p=k. Our considerations do not tell us, however, the infor-
mation we need about a collection of vortices: what is the phase accumulated when vor-
tices go around one another, what are the degenerate ground states, if any, that many
vortices form when they are distant from one another, and how these degenerate ground
states lead to non-abelian statistics, if at all. Indeed, for the case of m ¼ 5=2, described in
details earlier, we needed the machinery of super-conductivity, particularly the Bogo-
lubov-deGennes equations for a p-wave super-conductor, to give us that information.
The tool that will help us to analyze the clustering of anyons into effective bosons and
the states formed by vortices in the condensate of these bosons will be parafermionic Con-
formal Field Theories. As we now review, these theories provide a tool for constructing the
function F both in the absence and in the presence of quasi-particles in the ground state.
We start with an introduction to the Conformal Field Theories we will use (Our intro-
duction is very short and very incomplete. Useful references for a more comprehensive
study of the subject are [97–104]. Following that, we show how the Ising CFT reproduces
our analysis of the m ¼ 5=2 state, and finally we show how a certain set of parafermionic
CFTs generate the Read–Rezayi states for m ¼ 2 þ k=ðk þ 2Þ. When analyzing conformal
field theories, general considerations of symmetries and internal consistency go a long way
towards the calculation of certain properties of correlation functions. Here these correla-
tion functions will construct the trial wave functions for the ground state and the quasi-
particles. What is needed for these calculations are three inputs. The first is the list of fields
in the theory. Some of these fields we may anticipate: the field of the anyon, which we will
denote by w1 (this field is called a parafermion in the CFT context), the field of the vortex,
which we will denote by r1 and the field of the boson that is formed by k anyons. As we
will see below, at m ¼ 5=2 these are the only three fields, while in the more complicated
CFTs there are more fields. The second input we need is the conformal dimension of each
field. The third is the list of operator product expansion rules. The operator product
expansion rules tell us that when two fields Ua ðz1 Þ and Ub ðz2 Þ appear together in a corre-
lation function, and when the coordinates z1 and z2 approach one another, the following
substitution may be carried out:
X d
Ua ðz1 ÞUb ðz2 Þ ! C abc ðz1 z2 Þ Uc ð57Þ
c
where C abc are constants and d ¼ hc ha hb with the hs being the conformal dimensions
of the respective fields. The fields Ua and Ub are then said to be ‘‘fused’’ to the field(s) Uc .
These rules for fusion are very important for us, in several contexts. First, we will use them
to ‘‘create’’ clusters of anyons. Indeed, we will see that when kw1 ’s are fused together, the
resulting field is the identity. The identity field is an ‘‘invisible’’ one: when it is fused with
any field, d ¼ 0, and the right hand side of (57) does not have the ðz1 z2 Þ factor. Second,
A. Stern / Annals of Physics 323 (2008) 204–249 241
we will use the fusion rules to figure out the phase accumulated when one field encircles
another. This phase is simply 2pd, with the d that correspond to the fusion of the two
fields. And third, we will use the rules to figure out the structure of the ground state sub-
space created when several quasi-particles (vortices) are fused together.
So here is how it will go [9,10]: we will have a description of the electron as a product of
two conformal fields,
pffiffiffiffi
kþ2
ffi
we ðzÞ ¼ ei k /ðzÞ w1 ðzÞ ð58Þ
pffiffiffiffi
kþ2
ffi
In the first piece, ei k /ðzÞ , commonly called a ‘‘vertex operator’’, the field /ðzÞ is a free
chiral boson. The ground state wave function will be a correlator of many electron fields
Wðfzi gÞ ¼ hwe ðz1 Þ . . . we ðzn Þi: ð59Þ
The correlation functions of the bosonic vertex operators produce Eq. (53). The corre-
lation function of the w1 ðzÞ’s will produce the function F . There are two important prop-
erties that will make it happen. Firstly, the fusion of a finite number k of w1 ’s will produce
the identity field. That will guarantee that the power of each zi in the function F will scale
like k, and not like N. Second, the conformal dimension of w1 will guarantee that the entire
correlator will create a fermionic wave function. Generally, the operator product expan-
sion for the bosonic fields is,
0 ab
eia/ðzÞ eib/ðz Þ ! eiðaþbÞ/ðzÞ ðz z0 Þ ð60Þ
Electrons must accumulate a phase of 2p when they encircle one another. They will
accumulate a phase of 2pðk þ 2Þ=k from the bosonic factors, and will therefore need to
accumulate 4p=k from the parafermionic fields. This will make the combined wave func-
tion single-valued and anti-symmetric with respect to interchanging the coordinates of two
electrons.
To generate vortices, we will introduce terms of r1 eib/ into the correlators. The value of
b will be chosen as the minimal value for which the correlators will remain single valued
with respect to the electronic coordinates. As we will see, the fusion rule of two r1 ’s will be
fundamentally different from the fusion rule of two w1 ’s. While two w1 ’s fuse to a single
fusion product, two r1 ’s will have two fusion channels to go into. This multiplicity of pos-
sible outcomes will be the source of the ground state degeneracy and the non-abelian
statistics.
This description looks complicated at first reading, and—to be honest—also at later
readings. It will hopefully get simpler after we see how it works for the m ¼ 5=2 state,
and then how it works for the more complicated Read–Rezayi states.
The simplest parafermionic theory to be mentioned here is known as the Z 2 (or Ising)
theory. It is the CFT method of dealing with the m ¼ 5=2 state. Let us see how it works: the
theory has three fields: w1 , r1 and the identity field. In view of that, we will omit the sub-
script from w and r. The identity field corresponds to the Cooper pair. The w is a fermion.
The r corresponds to a vortex. The conformal dimensions of these fields are 0, 1/2 and 1/
16 respectively. What makes the Z 2 theory simple is the fact that the action for the w is
known to be that of a free chiral Majorana fermion [101]. As explained above, the electron
242 A. Stern / Annals of Physics 323 (2008) 204–249
and quasi-particle fields are composed of the parafermionic fields together with the boson-
ic factors.
There are three rules for operator product expansion. For creating the wave function
with no vortices we need only the following:
In the language of super-conductivity, this says that two composite fermions may form a
1
Cooper-pair whose wave function is ðz z0 Þ . Let us put:
* +
Y
N pffiffi pffiffi
i 2/ðzi Þ iN 2/ðz1 Þ
Wðfzi gÞ ¼ e wðzi Þe ð62Þ
i¼1
with z1 an arbitrary point at infinity, and the the last factor guarantees that the correlator
of the free boson factor does not vanish. The free action for both w and the boson / en-
ables the use of Wick’s theorem, and the rule (61) allows for the explicit calculation of this
correlator. The result is
Y 2
Y jzi2j2 1 1 1
wa ðfzi gÞ ¼ ðzi zj Þ e 4lH A ... ð63Þ
i<j i
z1 z2 z3 z4 zN 1 zN
Thus, r is an object that gives the fermion w a minus sign when the fermion goes
around it. Sounds like a vortex in the super-conductor, and indeed it is. The quasi-hole
would be rðzÞeib/ðzÞ ; and b is fixed in such a way that when the electron goes around a
quasi-hole (or a quasi-particle) it accumulates a phase of 2p such that the wave function
remains single
pffiffiffi valued. For that to happen, Eqs. (60) and (64) tell us that we need
b ¼ 1=2 2, a value that fixes the charge of the quasi-hole/particle to be a quarter
of the electron charge.
When two r’s fuse, the Z 2 theory has,
Two r’s may then be fused either to 1 or to w, which is the CFT way of saying that two
vortices in a px þ ipy super-conductor introduce a two-fold degeneracy of the ground state,
and that the two degenerate states differ by the parity of the total number of fermions.
The only remaining piece of information about operator product expansion that we
need is that the 1 field fuses trivially with the w and the r, leaving them unchanged,
and with no factors of ðz z0 Þ.
Finally for the Z 2 theory, how do we understand the interferometry in the language of
the CFT? Let us look at the Fabry–Perot geometry. The encircling of the bulk quasi-par-
ticles by the interference loop amounts to a trajectory in which an edge quasi-particle
A. Stern / Annals of Physics 323 (2008) 204–249 243
pffiffi pffiffi
rðzÞei/ðzÞ=2 2 goes around the bulk, whose N qp quasi-particles fuse to eiN qp /ðz0Þ=2 2 nðz0 Þ where
nðz0 Þ is rðz0 Þ when N qp is odd, and may be either the identity or wðz0 Þ when N qp is even.
Now, the phase accumulated when an edge quasi-particle goes around the bulk one can
N =8
be obtained from their fusion rules. First, the bosonic fields yield a factor of ðz z0 Þ qp .
Second, by (65) and (64), the fusion of r with n gives no extra phase when the bulk fuses to
the identity, and an extra phase of p when it fuses to w. Thus, when the bulk quasi-par-
ticles fuse to the identity the edge quasi-particles accumulates only the abelian phase. This
phase is just that of a vortex of a super-conductor going around 1/8 of a Cooper-pair,
namely a quarter of a charge. When the bulk quasi-particles fuse to w an extra phase of
p is added to the phase accumulated in the previous case. Furthermore, when the bulk
fuses to r, the fusion of the edge r with the bulk r gives, according to (65) a sum of
two interference patterns of equal magnitude, with a mutual phase shift of p. Obviously,
these two terms cancel one another and no interference is to be seen, exactly as predicted
by the px þ ipy superconductor picture.
Before we turn to discuss the other non-abelian states, for which we do not have but the
CFT description, let us summarize what use we made of CFT for understanding the
m ¼ 5=2 state, since similar steps will be useful for the understanding of the other non-abe-
lian states:
(1) We carried out a Chern-Simons transformation that is not necessarily single valued,
aimed at cancelling the external magnetic field within mean field approximation. This
transformation effectively maps the problem of electrons in a magnetic field onto a
problem of composite particles at zero magnetic field. For m ¼ 5=2 those particles
were fermions. For other filling fractions, they will be anyons.
(2) As the wave function F for the composite particles we looked for a function that will
have the needed power laws when any two particles approach one another, but will
be rather low in the degree of its polynomials. We then introduced the parafermionic
CFT’s, that happen to have exactly the needed properties.
(3) We studied the Z 2 CFT, and used it to identify the field for the electron, the field for
the quasi-hole, the degeneracy of the ground state subspace, and the phases accumu-
lated when one type of particle goes around another.
Note that this line of thought cannot but offer the possibility of a non-abelian phase at a
particular filling fraction and analyze its properties. It cannot judge if, and at what condi-
tions, this phase is energetically favorable to competing phases. This question may be
answered either through experiments or through numerical exact diagonalization studies.
Let us now apply these ideas to other filling fractions, and obtain the Read–Rezayi ser-
ies. Consider a filling fraction of k=ðk þ 2Þ. The cases k ¼ 1; 2 are relatively easy, and were
discussed at length above (True, we discussed the m ¼ 5=2 as the k ¼ 2, and not m ¼ 1=2. It
turns out that details of the electron-electron interaction in the N ¼ 1 Landau level are
more favorable for composite fermion pairing than those of the lowest, N ¼ 0, Landau
level. That is apluasibly the case for the higher values of k [72]). Now we look at higher
integers. The Chern-Simons transformation (53) is now multiply valued, and we resort
again to the CFTs to fix this problem. Thus, we are looking for a CFT whose correlators
244 A. Stern / Annals of Physics 323 (2008) 204–249
2=k
for one of its fields scales like ðz z0 Þ . Luckily, as we will now see, the parafermionic
CFTs do just that.
Similar to the m ¼ 5=2 case, we will combine two CFTs, a free chiral boson /ðzÞ and a
‘‘parafermionic’’ one. The parafermionic CFT will have more than the three fields we had
in the Z 2 case. This should come as no surprise: it takes the fusion of k anyons to get to the
identity, so we need to have k fields wj ðj ¼ 1 . . . kÞ where wj is just j anyons fused together,
and wk is the identity. Since a vortex carries a flux of 1=k, we would also require that k
vortices are allowed to fuse to the identity (but are not forced to do so. Remember that
vortices have several fusion channels). That requires k fields rj which would be j vortices
fused together (the field rk ¼ wk would be the identity field). But even that is not all: let us
call rk1 the anti-vortex. If the vortex and anti-vortex do not necessarily fuse to the iden-
tity, then a collection of j vortices and l anti-vortices does not necessarily fuse to rjl , but
may fuse to another state. That state will accumulate the same phase when encircling a
parafermion, but a different phase when encircling a vortex.
Altogether, a more elaborate analysis along these lines leads to the following picture: it
turns out to be convenient to label the fields by two quantum numbers, Ulm . The integer m
is called ‘‘the topological charge’’. Fusion processes conserve the topological charge mod-
ulo 2k, and it is the topological charge that determines the phase accumulated when a
parafermion encircle the field Ulm . The integer index l satisfies l 2 f0; 1; . . . ; kg, and the
fields Ulm satisfy Ulm ¼ Ulmþ2k ¼ Ukl mk and l þ m 0ðmod2Þ. These fields include all the
fields anticipated in the paragraphs above: w1 U02 is the parafermion, which will play
the role of w in the Z 2 case; The fields wj U02j are collections of j parafermions; The field
r1 U11 is the vortex (called also the spin field); The field U20 is the field that may be
obtained (side by side with the identity) when a vortex and and an anti-vortex fuse
together; And finally Ukk ¼ U00 which is the identity.
The conformal dimension of the field Ulm is given by (for l 6 m 6 l)
lðl þ 2Þ m2
hlm ¼ : ð66Þ
4ðk þ 2Þ 4k
as we need. Second, that when we fuse k parafermions w1 we get the unity. Thus, the state
we will construct this way will be a condensate of clusters, each cluster having k paraferm-
ions.
As we said above, dissipationless transport is a property of a bosonic system at zero
magnetic field. At m ¼ 1=m the flux attachment term (53) turned the electrons into compos-
ite bosons. At m ¼ 1=2 the attachment of flux turned electrons into composite fermions and
A. Stern / Annals of Physics 323 (2008) 204–249 245
pairs of composite fermions acted like bosons. At k ¼ 3 . . . the electrons were turned into
anyons, and the grouping of k anyons creates the bosons that condense.
The identification of the electron creation operator is done as for the m ¼ 5=2 case. It is
pffiffiffiffiffiffi
ip/ffi kþ2 ipuffi pffiffiffiffi
1 ffi
e
w1 e k . Similarly, the quasi-particle creation operator is r1 e k kþ2 . Their charge is kþ2
and they carry an average angular momentum of h=k per particle.
In the absence of quasi-particles this scheme creates a single ground state, which is the
correlation function of many electron operators. For each zi , the polynomial part of the
wave function has a maximal power of about N ðk þ 2Þ=k, and hence about that number
of zeroes (the word ‘‘about’’ stands for the neglect of numbers of order one, and for
the neglect of the question of whether N divides by k). Out of these zeroes for, say,
z1 ; N 1 zeroes coincide with the positions of the other electrons, as required by the Pauli
principle. The other 2=k zeroes are not attached to other electrons. Rather, Q it can be shown
that if the electronic wave function W in Eq. (56) is written as W ¼ i<j ðzi zj Þvðfzi gÞ
(separating out the zeroes that are dictated by the Pauli principle) then the function v van-
ishes quadratically when k þ 1 of the coordinates approach one another [10,102,103] .
Thus, when k of the coordinates are close together around the point z0 , we expect the elec-
2
tronic wave function W to vanish like ðz1 z0 Þ , attaching two zeroes—figuratively speak-
ing—to k electrons.
The fusion rules Eq. (67) introduce the non-abelian side of these quasi-particles, since as
we fuse many r1 ’s together, we get a number of fusion channels that increases exponen-
tially with the number of fused r’s. Each of these channels corresponds to a ground state.
For keeping track of this number it is useful to use again the Brattelli diagram, see Fig. (6).
In the Brattelli diagrams that are useful to us the x-axis counts the number of quasi-par-
ticles (which, when divided modulo 2k, gives the topological charge). The y axis is the
index l. A fusion with r1 increases or decreases l by one, and l is limited between 0 and
k, and hence the structure reflected in the figure: the diagram has k þ 1 ‘‘floors’’ and each
steps towards the right or left takes you one floor up or down. A trajectory that starts at
the origin and ends at some node of a particular number of quasi-particles corresponds to
Fig. 6. A Brattelli diagram for a Read–Rezayi state. Each node on the diagram describes the state that the quasi-
particles fuse to. Each trajectory on the diagram is a ground state.
246 A. Stern / Annals of Physics 323 (2008) 204–249
a ground state with that number of quasi-particles. Each node in the diagram corresponds
to a field Ulm . The node at which a trajectory ends corresponds to the state to which the
quasi-particles fuse in that particular ground state.
The Brattelli diagram makes the counting of the number of ground states relatively sim-
ple—it becomes the counting of trajectories on a graph. For a large number N of quasi-
particles, it scales like
N
p
2 cos ð69Þ
kþ2
To propagate horizontally on the diagram, we need to fuse a state with r1 ’s. To propagate
vertically, we need to fuse with the field U20 .
The information we gave here on the CFT formulation of the Read–Rezayi gives most
of the background needed for generalizing the analysis of interferometry, described above
in detail for the m ¼ 5=2 state, to these states. We will not carry out this generalization
here, but rather refer the reader to [105–109]. As shown in these works, lowest order inter-
ference in the Fabry–Perot interferometer and Couomb blockade peak spacings in a quan-
tum dot reflect the number and state of the localized quasi-particles in the cell of the
interferometer, and the effective charge in a shot noise measurement in a Mach–Zehnder
interferometer may identify the Z 3 nature of the m ¼ 12=5 state.
14. Summary
Much of physics is about setting rules, and then looking for ways to break them. Any-
ons were the breaking of the rule that the wave function of identical particles must follow
fermionic or bosonic symmetry. The breaking of this rule signaled the door to a rich and
fascinating part of physics, in which ohmic contacts and semi-conductors mingle with
topology and CFT’s, and the immensely complicated nature of the first two converges
in certain limits to the pure simplicity of the last two. Much is presently understood about
that part of physics, as we tried to describe. Much more is awaiting for future research, we
believe.
Acknowledgements
I am indebted to Merav Dolev, Eytan Grosfeld, Roni Ilan and Gil Tayar for their help
in preparing this manuscript. This work was partially funded by the US-Israel Binational
Science Foundation, the Minerva Foundation and the Israel Science Foundation.
References
[54] B. Rosenow, B.I. Halperin, Phys. Rev. Lett. 98(10) (2007) 106801, URL http://link.aps.org/abstract/PRL/
v98/e106801.
[55] J.K. Jain, C. Shi, Phys. Rev. Lett. 96 (13) (2006) 136802, pages 4.
[56] V.J. Goldman, Phys. Rev. B 75 (2007) 045334.
[57] G.A. Fiete, Refael Gil, Fisher M.P.A., arXiv.org:0705.1543 (2007).
[58] B. Blok, X.-G. Wen, Nuc. Phys. B 374 (1992) 615.
[59] H.-K. Lo, J. Preskill, Phys. Rev. D 48 (10) (1993) 4821.
[60] X.G. Wen, Phys. Rev. Lett. 66 (6) (1991) 802.
[61] J. Fröhlich, P.A. Marchetti, Commun. Math. Phys. 121 (1989) 177.
[62] J. Fröhlich, F. Gabbiani, Rev. Math. Phys. 2 (1990) 251.
[63] X.-G. Wen, A. Zee, Phys. Rev. B 58 (23) (1998) 15717.
[64] X.-G. Wen, Phys. Rev. B 60 (12) (1999) 8827.
[65] F.A. Bais, Nucl. Phys. B170 (1980) 32.
[66] R. Willett, J.P. Eisenstein, H.L. Stormer, D.C. Tsui, A.C. Gossard, J.H. English, Phys. Rev. Lett. 59 (15)
(1987) 1776.
[67] E. Ardonne, K. Schoutens, Phys. Rev. Lett. 82 (25) (1999) 5096.
[68] E. Ardonne, N. Read, E. Rezayi, K. Schoutens, Nucl. Phys. B 607 (2001) 549.
[69] S.H. Simon, E.H. Rezayi, N. R. Cooper, I. Berdnikov, Phys. Rev. B 75(7) (2007) 075317,
http://link.aps.org/abstract/PRB/v75/e075317.
[70] R.H. Morf, Phys. Rev. Lett. 80 (7) (1998) 1505.
[71] R. Morf, N. d’Ambrumenil, Phys. Rev. B 68 (2003) 113309/1.
[72] E. H. Rezayi, N. Read, 2006, Non-Abelian quantized Hall states of electrons at filling factors 12/5 and 13/5
in the first excited Landau level, cond-mat/0608346.
[73] V.W. Scarola, J.K. Jain, E.H. Rezayi, Phys. Rev. Lett. 88 (21) (2002) 216804.
[74] X. Wan, K. Yang, E.H. Rezayi, Phys. Rev. Lett. 97 (25) (2006) 256804.
[75] N. Read, D. Green, Phys. Rev. B 61 (15) (2006) 10267.
[76] M. Greiter, X.-G. Wen, F. Wilczek, Phys. Rev. Lett. 66 (24) (1991) 3205.
[77] M. Greiter, X.G. Wen, F. Wilczek, Nucl. Phys. B 374 (3) (1992) 567.
[78] N. Read, E. Rezayi, Phys. Rev. B 54 (23) (1996) 16864.
[79] D.A. Ivanov, Phys. Rev. Lett. 86 (2) (2001) 268.
[80] C. Nayak, F. Wilczek, Nucl. Phys. B 479 (3) (1996) 529, cond-mat/9605145.
[81] A. Stern, F. von Oppen, E. Mariani, Phys. Rev. B 70 (20) (2004) 205338.
[82] S.B. Chung, M. Stone, J. Physics A: Math. Theor. 40(19) (2007) 4923, URL http://stacks.iop.org/
1751-8121/40/4923.
[83] M. Stone, S.-B. Chung, 2006, Phys. Rev. B 73(1), 014505, URL http://link.aps.org/abstract/PRB/v73/
e014505.
[84] Y. Tserkovnyak, S.H. Simon, Phys. Rev. Lett. 90 (1) (2003) 016802.
[85] S.C. Zhang, T.H. Hansson, S. Kivelson, Phys. Rev. Lett. 62 (1) (1989) 82.
[86] S. Das Sarma, M. Freedman, C. Nayak, Phys. Rev. Lett. 94 (16) (2005) 166802.
[87] E. Fradkin, C. Nayak, A. Tsvelik, F. Wilczek, Nucl. Phys. B 516 (3) (1998) 704, cond-mat/9711087.
[88] A. Stern, B.I. Halperin, Phys. Rev. Lett. 96 (1) (2006) 016802.
[89] P. Bonderson, A. Kitaev, K. Shtengel, Phys. Rev. Lett. 96 (1) (2006) 016803.
[90] C.-Y. Hou, C. Chamon, Phys. Rev. Lett. 97 (2006) 146802.
[91] B.J. Overbosch, F.A. Bais, Phys. Rev. A 64 (6) (2001) 062107.
[92] L.S. Georgiev, M.R. Geller, Phys. Rev. B 73 (20) (2006) 205310.
[93] E. Grosfeld, S.H. Simon, A. Stern, Phys. Rev. Lett. 96 (22) (2006) 226803.
[94] D.E. Feldman, A. Kitaev, Phys. Rev. Lett. 97 (2006) 186803.
[95] C. Bena, C. Nayak, Phys. Rev. B 73 (2006) 155335.
[96] J.K. Slingerland, F.A. Bais, 2001, Nuc. Phys. B 612, 229, URL http://www.citebase.org/abstract?
id=oai:arXiv.org:cond-mat/01 04035.
[97] A.B. Zamolodchikov, V.A. Fateev, Zh. Eksp. Teor. Fiz. 89 (1985) 380.
[98] D. Gepner, Nucl. Phys. B290 (1987) 10.
[99] D. Gepner, E. Witten, Nucl. Phys. B278 (1986) 493.
[100] D. Gepner, Z.-a. Qiu, Nucl. Phys. B285 (1987) 423.
[101] P. Di Francesco, P. Mathieu, D. Sénéchal, Conformal Field Theory, Springer, New York, 1997.
[102] A. Cappelli, L.S. Georgiev, I.T. Todorov, Comm. Math. Phys. 205 (1999) 657.
A. Stern / Annals of Physics 323 (2008) 204–249 249
[103] A. Cappelli, L.S. Georgiev, I.T. Todorov, Nuc. Phys. B 599 (3) (2001) 499.
[104] E. Verlinde, Nuc. Phys. B 300 (1988) 360.
[105] P. Bonderson, K. Shtengel, J.K. Slingerland, Phys. Rev. Lett. 97 (1) (2006) 016401.
[106] P. Bonderson, K. Shtengel, J.K. Slingerland, Phys. Rev. Lett. 98(7) (2007) 070401, URL http://link.aps.org/
abstract/PRL/v98/e070401.
[107] K.T. Law, 2007, Probing non-abelian statistics in nu = 12/5 quantum hall state, URL arXiv.org:0707.3995.
[108] R. Ilan, E. Grosfeld, A. Stern, 2007, Coulomb blockade as a probe for non-Abelian statistics in Read–
Rezayi states, arXiv:0705.2187.
[109] L. Fidkowski, 2007, Double point contact in the k=3 read-rezayi state, arXiv:0704.3291.
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: