a

r
X
i
v
:
c
o
n
d
-
m
a
t
/
9
4
0
4
0
1
1
v
1


6

A
p
r

1
9
9
4
Network Models of Quantum Percolation
and
Their Field-Theory Representations
Dung-Hai Lee

IBM Research Division, T. J. Watson Research Center, Yorktown Heights, NY 10598
Abstract
We obtain the eld-theory representations of several network models that are
relevant to 2D transport in high magnetic elds. Among them, the simplest one,
which is relevant to the plateau transition in the quantum Hall eect, is equivalent
to a particular representation of an antiferromagnetic SU(2N) (N 0) spin chain.
Since the later can be mapped onto a = 0, U(2N)/U(N) U(N) sigma model, and
since recent numerical analyses of the corresponding network give a delocalization
transition with 2.3, we conclude that the same exponent is applicable to the
sigma model.
PACS numbers: 73.50.Jt, 05.30.-d, 74.20.-z
1
In strong magnetic elds, a two-dimensional electron gas exhibits the quantum Hall
eect[1] (in this paper we will restrict ourselves to the integer eect) over a wide range of
sample disorder.[2] The hallmark of the quantum Hall eect is that in the neighborhood of
a series of magic lling factors f = S
xy
,
xy
exhibits quantized plateaus at values S
xy
e
2
/h
while
xx
0.[1] Between these lling factors,
xy
interpolates between the plateaus, and

xx
peaks in the middle of the transitions. By analyzing the temperature dependence of the
widths of the transitions, Wei et al.[3] have shown convincingly that, if extrapolated to zero
temperature, the plateau transition is a continuous phase transition. The single diverging
length scale at this transition is the electron localization length | B B
c
|

. Although
Wei et al.s results do not give directly, a recent experiment using narrow Hall bars suggests
2.3.[4] Moreover, considerable numerical work [5] has been done for the plateau transition
of non-interacting electrons. The result for is consistent with the experimental ndings.[6]
It is known that the eective theory for zero-temperature, long-wavelength electronic
transport in disordered media is a non-linear sigma model with symplectic symmetry.[7] In
2D this model has a nite localization length (hence gives an insulating state) for arbitrarily
weak disorders. This result holds when a weak magnetic eld is applied, where the symmetry
of the sigma model is reduced to unitary. These results are in apparent contradiction with
the quantum Hall eect. In trying to reconcile them, Levine, Libby and Pruisken[8] proposed
that the proper eld theory for electronic transport in strong magnetic elds is a N 0,
U(2N)/U(N) U(N) sigma model with a topological term. Subsequently, starting from a
microscopic basis, Pruisken[9] derived the previously postulated sigma model
L =
1
8

xx
Tr[(

Q)
2
]
1
8

xy

Tr[Q

Q], (1)
where = x, y,
xy
=
yx
= 1, and

are dimensionless conductivities in units of e

2
/h. In
Eq.(1) Q = u
+
u, where u is a 2Nx2N unitary matrix, and is a diagonal matrix with
a
a
= 1
2
for a N and
a
a
= 1 for N < a 2N. Pruisken et al. argued that with the addition of
the second (topological) term, Eq.(1) possesses stable xed points at (
xx
,
xy
) = (0, n) (cor-
responding to the quantized plateaus) and critical points at (constant, n+
1
2
) (corresponding
to the transition points between the plateaus). Using the sigma model as the paradigm,
Pruisken conjectured the scaling behavior of the plateau transition.[10] This conjecture was
later veried experimentally.[3] Although Pruiskens arguments are compelling, serious cal-
culations of the critical properties are absent. It is still an open question as to whether
Eq.(1) can produce the experimentally or numerically observed exponent.[11]
Quite independently, in an insightful paper, Chalker and Coddington [12] proposed a
network model to describe the quantum percolation of the semi-classical (equal-potential)
orbitals of non-interacting electrons in a strong magnetic eld and a smooth random poten-
tial. Both that work and a recent extension by Lee, Wang and Kivelson[13] show that the
network model has a plateau transition with = 2.40.2. In brief, the network model[12, 13]
consists of a checker-board of low-potential- energy plaquettes (marked by 1 in Fig.1a) that
are occupied by electrons. At the boundary of each plaquette, there is an edge state at the
Fermi energy, in which an electron undergoes

E

B drift in the direction indicated by the
arrows. In this model, quantum tunneling takes place at the centers of the open squares
only. The eect of changing the Fermi energy is reected in the modication of the tun-
neling matrix elements. The fact that, in reality, the semi-classical orbitals have dierent
spatial extent is reected in the random Aharonov-Bohm phases the electrons accumulate
when reaching the tunneling points. In numerical calculations[12, 13], the localization length
is determined in cylindrical systems. Using nite-size scaling the thermodynamic was
deduced, from which the authors of Ref.[13] concluded that = 2.4 0.2. Since to calculate
(either numerical or analytically) using Eq.(1) is notoriously dicult, one purpose of the
3
present work is to establish a connection between the network model and the sigma model,
so that the exponent obtained for the former can be used for the latter.
Due to a recent experiment on the spin-unresolved plateau transition[14], questions con-
cerning the nature of plateau transitions in the presence of a strong inter-Landau level
mixing have been raised. In this paper we generalize the network model,[15] and obtain
the corresponding eld-theory representation to describe this situation as well. Finally, to
demonstrate the importance of the random phase on the critical properties of the delocal-
ization transition, we consider a network model in its absence. We show that the transition
in this model belongs to a dierent universality class.
A representation of the network model is to view it as a one dimensional ( x) stacking of
alternating upper-right ( y) and lower-left ( y) moving edge states shown in Fig.1b. Electrons
that tunnel at the centers of the open squares can be localized or extended depending on
the assignment of the (random) tunneling matrix elements. To capture the linear dispersion
and the chiral nature of the edge states[16], we use the following Hamiltonian to model the
eigenstates of the network near the Fermi energy (E = 0):

H =

x
(1)
x/a

dyv
+
(x, y)

y
i
(x, y)

y
[t

(x, y)e
i(x,y)

+
(x+a, y)(x, y)+h.c]. (2)
Here is the electron annihilation operator, the s are the random Aharonov-Bohm phases
for electrons at the Fermi energy, and the t

s are random (positive) tunneling matrix el-

ements. In the second term, the y-sum is extended over the discrete set of y coordinates
at which tunneling takes place. Here, without changing the long-distance physics, we have
assumed that the value of the drift velocity v is uniform. Since the eects of changing E
can be absorbed into a dependence of and t

on E, we can concentrate on E = 0 without

losing generality. In order to study the critical behavior of the conductivities, we consider
4
the following transport action[9]
S =

dyiS
p

p
(x, y)
p
(x, y) H

dy[iS
p

p
(x, y)
p
(x, y) (1)
x/a

p
(x, y)

y
i

p
(x, y)]
+

y
[t(x, y)e
i(x,y)

p
(x + a, y)
p
(x, y) + h.c.]. (3)
In the above, repeated indices p = are summed over,
p
and

p
are Grassmann elds,
S
p
sign(p), 1/(v) is the linear density of states, is a positive innitesimal, and
t t

. In going from the rst to the second lines in Eq.(3) we have rescaled the fermion
elds so as to absorb the edge velocity v.
By redening
p

p
(i
p
) and

p
i

p
(

p
) for even (odd) x/as[17] the action
becomes
S =

dy[(1)
x/a
S
p

p
(x, y)
p
(x, y) +

p
(x, y)
y

p
(x, y)]
+

y
[t(x, y)e
i(x,y)

p
(x + a, y)
p
(x, y) + h.c.]. (4)
Next, we replicate the action and integrate out the random s and ts to obtain
S =

dy[(1)
x/a
S
p

p
(x, y)
p
(x, y) +

p
(x, y)
y

p
(x, y)]
+

dy

F
x
[

p
(x + a, y)
p

(x + a, y)

(x, y)
p
(x, y)]. (5)
We note that because is random, after integrating it out only the charge neutral combi-
nations of s and
+
s are present in Eq.(5). Moreover, the symmetry of the replicated
action guarantees that only the SU(2N) invariant combinations appear at = 0. As usual,
the repeated replica indices , = 1...N are summed over, and

F
x
(Z) =
1
a
< t
2
(x) > Z +....
We note that

F has an explicit x-dependence. It reects the fact that the electrons tunnel
across the occupied/unoccupied regions for the even-odd/odd-even columns. As a result, the
corresponding < t
2
> are dierent.
5
If we view y as the imaginary time , Eq.(5) is the coherent-state path-integral action of
a 1D quantum eld theory described by the following Hamiltonian

H =

x
[(1)
x/a
S
p

+
p
(x)
p
(x)] +

x
F
x
[
+
p
(x + a)
p

(x + a)
+
p

(x)
p
(x)]. (6)
The relation between F in Eq.(6) and

F in Eq.(5) is that upon normal ordering F

F. The
relation to the SU(2N) spin chain becomes explicit after we rewrite Eq.(6) in terms of the
SU(2N) generators

S
b
a

+
a

b

ab
1
2N

+
c

c
, where a (p, ) and takes on 2N values.
The nal spin Hamiltonian is

H =

x
(1)
x/a
Tr[

S(x)] +

x
F
x
(Tr[

S(x + a)

S(x)]), (7)
where is the diagonal c-number matrix dened earlier, and Tr[

S]

a,b

b
a

S
a
b
, Tr[

S

S

a,b

S
b
a

a
b
. Since

H commutes with the site occupation number n(x) =

a

+
a
(x)
a
(x),
Hilbert spaces corresponding to dierent {n(x)} decouple. The ground state of Eq.(7) lies in
the Hilbert space where n(x) = N for all x. In this Hilbert space, a particular representation
for SU(2N) is realized. This representation is characterized by a Young tableau with a single
column of length N.
Analogous to the SU(2) quantum spin chain, which in the large-spin limit is equivalent
to the U(2)/U(1)xU(1)=O(3) sigma model, one can show that at = 0 the SU(2N) spin
chain of Eq.(7) is equivalent to the following U(2N)/U(N) U(N) sigma model in the large
representation limit[18, 19]
L =
M
16
(

1 R
2
)Tr(

Q)
2
+
M
16
(1 R)

Tr[Q

Q], (8)
where = , x, and R [F

x+a
(Z
0
) F

x
(Z
0
)]/[F

x+a
(Z
0
) +F

x
(Z
0
)] (here Z
0
NM
2
/4 and
F

(Z) dF/dZ). For the network model M = 1, and Eq.(8) is massless at R = 0. For R = 0
the system remains massive. Thus we identify the transition from R < 0 to R > 0 as the
6
plateau transition. At the critical point (R = 0) the x-translational symmetry is restored.
If we assume that a) the mapping from Eq.(7) to Eq.(8) is valid down to M = 1, and b) at
the critical point the system is described by a translationally-invariant SU(2N) spin chain
with M=1, then by comparing Eq.(1) and Eq.(8) we conclude that
xx
=
xy
= 1/2 at the
critical point.[13, 21] In any case, if the density of states stays nite at the transition,[20]
then the dimension of Q remains zero. As a result, the

in Eq.(1) are dimensionless, and

hence are generically universal at the transition.[21]
Modications can be made so that the network model represents other physically relevant
situations. For example, to study the eects of neighboring Landau-level mixing on the
plateau transition we introduce two edge states with the same circulation in each plaquette
of Fig.1.[15] The Hamiltonian including the inter-channel mixing is given by

H =

x
(1)
x/a

dyv

(x, y)

y
i

(x, y)

y
[t

(x, y)e
i(x,y)

(x + a, y)

(x, y) + h.c]

y
[t

12
(x, y)e
i
12
(x,y)

+
1
(x + a, y)
2
(x, y) + h.c]. (9)
Here = 1, 2 is the channel index and is implicitly summed over. Straightforward general-
ization of the steps between Eq.(2) and Eq.(7) now gives

H =

x
(1)
x/a

Tr[

(x)] +

x
F
x,
(Tr[

(x + a)

(x)]) +

x
F
x,12
(Tr[

S
1
(x)

S
2
(x)])
(10)
In the above F
x,
(Z) =

2

a
< t

(x) > Z + ..., and are antiferromagnetic interactions. On

the contrary, F
x,12
(Z) =

2
a
< t

2
12
(x) > Z + ..., and is ferromagnetic. Therefore inter-
Landau mixing causes a ferromagnetic coupling between two otherwise decoupled SU(2N)
spin chains.
In the absence of the inter-chain coupling there are two transitions as we tune R
1
and
R
2
keeping R
1
> R
2
. (R

[F

x+a,
(Z
0
) F

x,
(Z
0
)]/[F

x+a,
(Z
0
) + F

x,
(Z
0
)]) All three
7
massive phases break the translational symmetry and correspond to the (even-odd, even-
odd), (even-odd, odd-even), and (odd-even, odd-even) spin-Peierls phases. In the presence
of a strong ferromagnetic inter-chain coupling we expect vertical spin pairs to form the
M = 2 representation of SU(2N). At long wave-length, the problem is equivalent to a single
antiferromagnetic spin chain in the M = 2 representation. In this case there are also two
transitions as we tune R
1
and R
2
. Among the three massive phases, two break translation
symmetry and they correspond to the (even-odd) and (odd-even) spin-Peierls phases. The
phase which remains translationally invariant is the SU(2N) analog of the Haldane phase.[22]
Thus the phase structure and the universality class of the phase transition are preserved in the
presence of a strong inter-Landau level mixing. Although we have not done calculations for
intermediate inter-chain couplings, based on the knowledge about the SU(2) spin chains[23]
we expect the same behavior as in the zero and strong coupling limits.
In addition to the quantum spin chains, the network model contains another interesting
eld theory. In specic, if we set the random phases = 0, hold t(x+a, y) +t(x, y) xed and
let t(x+a, y)t(x, y) be random, the network model is equivalent to a 2N-component Gross-
Neveu model[24] in the limit N 0. Since the latter model has a dierent localization length
exponent, we conclude that the random phase is essential in determining the universality
class of the delocalization transition. To obtain the eld theory we group each adjacent
upper-right and lower-left moving edge states into a doublet. For each doublet we dene

+
and

as the annihilation operators for the upper-right and lower-left moving electrons
respectively. Moreover, we construct a Dirac spinor such that =

. It is simple to
show that in the continuum limit Eq.(2) reduces to
H =

dy
+
(x, y){
z
p
y
+[(t(x+1, y)t(x, y)]
x
+a[t(x+a, y)+t(x, y)]
y
p
x
}(x, y). (11)
Here p

=

i
( = x, y), and

are the Pauli matrices. In obtaining the above we have

8
ipped the signs of and
+
for every other doublets. After going to the spinor basis that
diagonalize
y
, we let

x

dx
2a
,
t(x,y)+t(x+a,y)
2
y y and x/2a x. As the result, we obtain
H =

dxdy
+
(x, y)[

+ m(x, y)
z
](x, y), (12)
where m(x, y) = 2[t(x + a, y) t(x, y)]/[t(x + a, y) + t(x, y)]. We recognize that Eq.(12) is
the Hamiltonian for 2D Dirac electrons with random masses. To study the conductivities
we study the following E = 0 transport action
S =

d
2
x

p
[

+ miS
p

z
]
p
. (13)
Here we have let x

, and redened

so that

z
. Finally, we replicate and
integrating out m = m < m > to obtain:
S =

d
2
x

p
[

+ < m > iS
p

z
]
p

g
2
(

p
)(

), (14)
where g =< (m)
2
>. In the limit 0, Eq.(14) describes the 2N-component Gross-Neveu
model. We have checked numerically[25] that for small < m
2
> the phase transition of
the network model has = 1, consistent with the result of the perturbative analysis of
Eq.(14).[26]
Acknowledgments: We thank E. Fradkin, S. Kivelson, N. Read, Z. Wang and X-G Wen
for useful discussions. We also thank J. Chalker for communicating his numerical results to
us prior to their publication. Part of the work was initiated at the Aspen Center for Physics.
[ * New address: Department of physics, University of California at Berkeley, Berkeley, CA
94720]
9
References
[1] See e.g. The Quantum Hall Eect, 2nd edition, edited by R.E. Prange and S.M.
Girvin, Springer-Verlag (1990).
[2] H.W. Jiang, C.E. Johnson, K.L. Wang, and S.T. Hannahs, Phys. Rev. Lett. 71, 1439
(1993).
[3] H.P. Wei, D.C. Tsui, M. Paalanen, and A.M.M. Pruisken, Phys. Rev. Lett. 61, 1294
(1988).
[4] S. Koch, R. Haug, K. v. Klitzing, and K. Ploog, Phys. Rev. Lett. 67, 883 (1991).
[5] B. Huckestein and B. Kramer, Phys. Rev. Lett. 64, 1437 (1990); B. Mieck, Europhysics
Lett. 13, 453 (1990); Y. Huo and R.N. Bhatt, Phys. Rev. Lett. 68, 1375 (1992); D. Liu
and S. Das Sarma, Mod. Phys. Lett. B 7, 449 (1993).
[6] One can take this agreement as an experimental indication that the Coulomb interaction
is irrelevant at the plateau transition. Nonetheless, the rigorous theoretical justication
for such irrelevance is still lacking.
[7] F.Wegner, Z. Physik B 36,209 (1979);K. Efetov, A. Larkin and D. Khemelnitskii, JETP
52, 568 (1980).
[8] H. Levine, S.B. Libby and A.M.M. Pruisken, Phys. Rev. Lett. 51, 1915 (1983).
[9] A.M.M. Pruisken, Nucl. Phys. B 235 FS[11], 277 (1984), and chapter 5 of Ref.[1].
[10] A.M.M. Pruisken, Phys. Rev. Lett. 61, 1297 (1988).
[11] I. Aeck, Nucl. Phys. B 265, 409 (1986).
10
[12] J.T. Chalker and P.D. Coddington, J. Phys. C 21, 2665 (1988).
[13] D-H Lee, Z. Wang and S.A. Kivelson, Phys. Rev. Lett. 70, 4130 (1993).
[14] H.P Wei, S.W. Hwang, D.C. Tsui, and A.M.M. Pruisken Surf. Sci., 229 34 (1990).
[15] For the numerical results on these new networks see D-H Lee, Z. Wang, and X-G Wen,
to be published; J. Chalker to be published.
[16] B.I. Halperin, Phys. Rev. B 25, 2185 (1982).
[17] Since and

are independent Grassmann elds, they can be transformed independently.
[18] The large representations are characterized by Young tableaus with M columns (M >>
1) of length N.
[19] I. Aeck, Nucl. Phys. B 257 FS[14], 409 (1986); D-H Lee to be published.
[20] In the transfer matrix calculation, one only computes the connected part of the spin-
spin correlation function. Hence it bares no implication on whether the spin operator
itself has nonzero expectaion value (or the density of state is nite) at the transition.
This question is currently under investigation. For relevant works on this issue see, e.g.
E. Fradkin, Phys. Rev. B 33, 3257 (1986); and ibid 3263 (1986).
[21] S.A. Kivelson, D-H Lee and S-C Zhang, Phys. Rev. B 46, 2223 (1992). Y.Huo, R.E.
Hetzel and R.N. Bhatt, Phys. Rev. Lett. 70, 481 (1993).
[22] F.D.M. Haldane, Phys. Lett. 93A, 464 (1983); Phys. Rev. Lett. 50, 1153 (1983); J.
Appl. Phys. 57, 3359 (1985); I. Aeck, Nucl. Phys. B257, 397 (1985).
[23] J. Solyom and J. Timonen, Phys. Rev. B 34, 487 (1986).
11
[24] D.J. Gross and A. Neveu, Phys. Rev. D 10, 3235 (1974)
[25] Z. Wang and D-H Lee, to be published.
[26] See e.g. R. Shankar, BCSPIN lectures, Kathmandu, 1991.
12
FIGURE CAPTIONS
Figure 1. a) The network model. Here the arrows indicate the direction of the edge
velocity, and the open squares enclose the tunneling points. b) A representation of the
network model as alternating y and y-moving fermions that tunnel at the centers of
the squares.
13