1: \documentclass[aps,prl,twocolumn,groupedaddress,showpacs]{revtex4}
2: %\renewcommand{\topmargin}{-0.05in}
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4: \usepackage{graphicx}
5: \usepackage{wasysym}
6: %TCIDATA{OutputFilter=Latex.dll}
7: %TCIDATA{Version=5.00.0.2552}
8: %TCIDATA{<META NAME="SaveForMode" CONTENT="1">}
9: %TCIDATA{LastRevised=Saturday, February 04, 2006 14:03:17}
10: %TCIDATA{<META NAME="GraphicsSave" CONTENT="32">}
11: \begin{document}
12: \title{Quantum Hall Effect of Dirac Fermions in Graphene: Disorder
13: Effect and Phase Diagram}
14: \author{D. N. Sheng$^{1}$, L. Sheng$^{2}$ and Z. Y. Weng$^3$}
15: \address{
16: $^1$Department of Physics and Astronomy, California State
17: University, Northridge, California 91330\\
18: $^2$Department of Physics and Texas Center for Superconductivity,
19: University of Houston, Houston, Texas 77204\\
20: $^3$Center for Advanced Study, Tsinghua University, Beijing
21: 100084, China}
22: \begin{abstract}
23: We numerically study the interplay of band structure, topological
24: invariant and disorder effect in two-dimensional electron system
25: of graphene in a magnetic field. Two \emph{distinct} quantum Hall
26: effect (QHE) regimes exist in the energy band with the
27: unconventional \textquotedblleft half-integer\textquotedblright\
28: QHE appearing near the band center, consistent with the
29: experimental observation. The latter is more robust against
30: disorder scattering than the conventional QHE states near the band
31: edges. The phase diagram for the unconventional QHE is obtained
32: where the destruction of the Hall plateaus at strong disorder is
33: through the float-up of extended levels toward band center and
34: higher plateaus always disappear first. We further predict a new
35: insulating phase between $\nu =\pm 2$ QHE states at the band
36: center, which may explain the experimentally observed resistance
37: discontinuity near zero gate voltage.
38: \end{abstract}
39: \pacs{73.43.Cd; 73.40.Hm; 72.10.-d; 72.15.Rn}
40: \maketitle
41: With the advances in micromechanical extraction and fabrication
42: techniques, high mobility single atomic layer of graphite, called
43: graphene, became available recently~\cite{G0,G1,G2,G3}, which has
44: attracted much experimental and theoretical interest. This new
45: material has many extraordinary properties such as submicron-scale
46: ballistic transport at room temperature, ability to sustain high
47: electric currents, and ease of tuning carrier density for both
48: electrons and holes. It is generally believed that graphene may
49: have potential applications in electronic
50: devices~\cite{G0,G1,G2,G3}. Graphene is of great fundamental
51: interest as well because of its special band structure~\cite{S0}. Undoped
52: graphene has one $\pi$ electron per lattice site forming a
53: two-dimensional (2D) extended electronic structure. The only
54: states at the Fermi energy ($E_{F}=0$) are at two corners of the
55: Brillouin zone, where the conduction and valence bands touch. The
56: low energy excitations have a linear dispersion relation similar
57: to that of the massless Dirac equation, where the electron Fermi
58: velocity plays the role of an effective ``speed of light''. So the
59: electron system of graphene can be viewed as a condensed-matter
60: realization of relativistic Dirac fermions, and the band touching
61: points are often referred to as Dirac points.
62:
63: Remarkably, novel quantum Hall effect (QHE) with Hall plateaus
64: obeying the unconventional quantization rule
65: $\sigma_{xy}=(k+1/2)g_s \frac {e^2} h$ has recently been observed
66: experimentally~\cite{Hall0,Hall1} in graphene films in a strong
67: magnetic field. Here, $k$ is an integer and $g_s=4$ stands for the
68: spin and sublattice-related degeneracy. In units of
69: $g_{s}\frac{e^{2}}{h}$, the \textquotedblleft
70: half-integer\textquotedblright\ quantization of $\sigma _{xy}$ was
71: conjectured to be due to the nontrivial Berry phase of the Dirac
72: fermions in a magnetic field~\cite{Hall0,Hall1,phystoday}.
73: Interestingly, calculations~\cite{T1,T2,T3,T0} by using an analogy
74: to the $2+1$ dimensional Quantum Electro Dynamics have predicted a
75: ``half-integer'' quantized QHE for graphene. Disorder effect has
76: been studied by using self-consistent Born approximation within
77: the continuous model~\cite{T0}. However, so far the interplay of
78: the band structure, particle-hole symmetry, disorder effect, and
79: topological property of the energy band in the unconventional QHE
80: has not yet been investigated. It is thus highly desirable to
81: perform exact numerical calculations by taking into account the
82: full band structure and random disorder, in order to reveal the
83: fundamental nature of the new QHE phases and related quantum phase
84: transitions in graphene.
85:
86: In this Letter, we study the QHE in graphene using a tight-binding
87: model. The experimentally observed unconventional Hall plateaus
88: are reproduced around the band center $E_{F}=0$, while the
89: conventional integer QHE plateaus appear near the band edges. The
90: unusual distribution of the topological invariant quantity-Chern
91: number in the energy band and the conservation of total Chern
92: number (which is also the geometric Berry phase~\cite{thouless1})
93: for the particle and hole bands account for the unconventional
94: \textquotedblleft half-integer\textquotedblright\ QHE. The latter
95: is found to be much more stable than the conventional ones near
96: the band edges against disorder scattering. We map out the whole
97: phase diagram for the QHE and demonstrate that the QHE plateaus
98: are destroyed at strong disorder (or weak magnetic field) through
99: the float-up of extended levels toward the band center. We further
100: identify a new insulating phase between $\nu=\pm 2$ QHE states,
101: which may explain the experimentally observed discontinuity in
102: resistance\cite{Hall0,Hall1} near zero gate voltage.
103:
104: We consider a rectangular sample of 2D graphene sheet consisting
105: of carbon atoms on a honeycomb lattice~\cite{S0,lisheng}, which
106: has totally $L_y$ zigzag chains with $L_{x}$ atomic sites on each
107: chain~\cite{lisheng}. The size of the sample will be denoted as
108: $L_{x}\times L_{y}$. In the presence of an applied magnetic field
109: perpendicular to the graphene plane, the lattice model can be
110: written in the tight-binding form~\cite{S0,sheng}:
111: \begin{eqnarray}
112: H&=&-t\sum\limits_{\langle
113: ij\rangle\sigma}e^{ia_{ij}}c_{i\sigma}^\dagger
114: c_{j\sigma}+\sum\limits_{i}w_ic_{i\sigma}^\dagger c_{i\sigma}\ ,
115: \label{HAMIL}
116: \end{eqnarray}
117: where $c_{i\sigma}^+ $ ($c_{i\sigma}$) creates (annihilates) a
118: $\pi$ electron of spin $\sigma$ on lattice site $i$
119: with $t$ as the nearest-neighbor hopping integral,
120: and $w_i$ is random disorder potential uniformly distributed in the interval
121: $w_i\in [-W/2, W/2]$. The magnetic flux per hexagon
122: $\phi=\sum_{{\small {\mbox{\hexagon}}}}a_{ij}=\frac{2\pi}{M}$ is
123: proportional to the strength of the applied magnetic field $B$.
124: This tight-binding model is valid for describing the full energy
125: band and realizes the energy dispersion relation of the Dirac
126: fermions near the band center~\cite {S0}.
127: \begin{figure}[tbp]
128: %\includegraphics[width=2.0in]{figure1.eps}
129: \includegraphics[width=3.2in]{figure1.eps}
130: \caption{(a) Calculated Hall conductance and electron density
131: of states in the full energy band for magnetic flux strength $%
132: \protect\phi=\frac{2\protect\pi}{48}$ or $M=48$, and (b) the
133: unconventional Hall conductance for three different strengths of
134: magnetic flux near the band center. Here the disorder strength is
135: set to $W=0$ and the sample size is taken to be $96\times 48$.}
136: \vskip-0.25cm
137: \end{figure}
138:
139: The Hall conductance $\sigma_{xy}$ can be calculated by using the
140: Kubo formula through exact diagonalization of the system
141: Hamiltonian~\cite{sheng}. In Fig.\ 1a, the Hall conductance
142: $\sigma_{xy}$ and electron density of states are plotted as
143: functions of electron Fermi energy $E_{F}$ for a clean sample
144: $W=0$ with magnetic flux $\phi=\frac{2\pi}{48}$, which illustrates
145: the overall picture of the QHE in the full energy band. According
146: to the behavior of $\sigma_{xy}$, the energy band is naturally
147: divided into three different regimes. Around the band center,
148: $\sigma_{xy}=\nu \frac{e^2}{h}$ is indeed quantized according to
149: the unconventional quantization rule $\nu=(k+1/2)g_s$ with a
150: degeneracy factor $g_{s}=4$ for each Landau level (LL) due to two spin
151: directions and two Dirac points. These Hall plateaus explain the
152: experimentally observed unconventional QHE~\cite{Hall0,Hall1} and
153: agree with the results from the theory based upon the continuous
154: model~\cite{T1,T2,T3,T0}. In Fig.\ 1b, the Hall conductance in
155: this unconventional region for three different strengths of
156: magnetic flux is shown. With decreasing magnetic flux from
157: $\phi=\frac{2\pi}{24}$ to $\phi=\frac{2\pi}{96}$, more quantized
158: Hall plateaus emerge following the same quantization rule. We see
159: that the widths of the plateaus are roughly proportional to
160: $\sqrt{\phi}\propto \sqrt{B}$, in agreement with the continuous
161: theory~\cite{T1,T2}. The unconventional QHE in the present band
162: model can be understood in terms of the topological invariant
163: Chern number~\cite{sheng,bhatt}. Inside each LL, there are
164: extended states characterized by a nonzero Chern integer. The
165: total Hall conductance in units of $\frac{e^2}{h}$ is exactly the
166: sum of the Chern numbers of all the occupied extended
167: states~\cite{sheng,bhatt}. The additional degeneracy $g_s=4$
168: around the band center gives rise to a total Chern number $C=4$
169: for each LL. Thus when each additional LL is occupied, the Hall
170: conductance increments by $g_{s}\frac {e^2} {h}$. At the
171: particle-hole symmetric point $E_{F}=0$, corresponding to the
172: half-filling of the central LL, $\sigma_{xy}=0$ and the total
173: Chern number of all the occupied states (hole band) must sum up to
174: zero. Now one can count $\sigma_{xy}$ from this point, and find
175: that the central LL effectively contributes $\pm
176: (\frac{g_s}{2})\frac{e^2}{h}$ to $\sigma_{xy}$, when $E_{F}$ is
177: shifted away from the central LL by adding particles or holes.
178: This leads to the experimental ``half-integer'' quantization of
179: $\sigma_{xy}$ in units of $g_s \frac {e^2}{h}$. As will be shown
180: below, the total zero Chern number for the particle or hole band
181: resulting from the particle-hole symmetry in the pure system
182: remains to be true in the presence of disorder.
183:
184: Near the band edges, each LL carries a total Chern number $C=2$,
185: and thus the Hall conductance is quantized as
186: $\sigma_{xy}=kg_s\frac {e^2} h$ with $k$ an integer and $g_{s}=2$
187: for spin degeneracy only, which is as same as that in the
188: conventional QHE systems. Remarkably, around $E_{F}=\pm t$, there
189: are two new critical regions, which separate the unconventional
190: and conventional QHE states. The extended states in each critical
191: region carry a large negative total Chern number, e.g., $\sum
192: C=-72$ for $\phi=\frac {2\pi} {48}$, as shown in Fig.\ 1a. The
193: existence of the negative Chern number regimes around $E_{F}=\pm
194: t$ is crucial for understanding the behavior of the Hall
195: conductance in the whole energy band. When the Fermi energy
196: $E_{F}$ is increased from the band bottom toward band center
197: continuously, following a whole sequence of the conventional Hall
198: plateaus, the negative Chern numbers cause a dramatic reduction
199: and a sign inversion of $\sigma_{xy}$, so that the unconventional
200: low Hall plateaus with $\nu=-6,$ $-2,$ $2\cdots$ can reoccur near
201: the band center. This is in contrast to the QHE on a square
202: lattice, where stable Hall plateaus can only be observed near the
203: band edges~\cite{sheng}.
204: \begin{figure}[tbp]
205: %\includegraphics[width=2.0in]{figure1.eps}
206: \par
207: \includegraphics[width=3.2in]{figure2.eps}
208: \caption{Unconventional Hall conductance as a function of electron
209: Fermi energy near the band center for four different disorder
210: strengths each averaged over $200$ disorder configurations.
211: Inset: conventional Hall conductance near the lower
212: band edge. Here $M=96$ and the sample size is $96\times 48$.}
213: \vskip-0.3cm
214: \end{figure}
215:
216: We have shown that the unconventionally quantized QHE observed in
217: the experiments can be reproduced in the lattice model, and is due
218: to the unusual topological property of the energy band. We next
219: turn to the effect of random disorder on the QHE. In Fig.\ 2, the
220: Hall conductance around the band center is shown as a function of
221: $E_{F}$ for magnetic flux $\phi=\frac{2\pi}{96}$ and four
222: different values of disorder strength $W$. We see that with
223: increasing $W$, higher Hall plateaus (with larger $|\nu|$) are
224: destroyed first. At $W=0.5t$, the plateaus with $\nu=\pm 10,\pm 6$
225: and $\pm 2$ remain well quantized, while at $W=2.0t$ all the
226: plateaus except for the $\nu=\pm 2$ ones are destroyed. These last
227: two plateaus will eventually disappear at $W \sim 2.5t$. For
228: comparison, the QHE near the lower band edge is shown in the
229: inset, where all plateaus are found to be destroyed at a much
230: weaker disorder $W=1.0t$. This clearly indicates that, under the
231: same conditions, the unconventional QHE is much more stable than
232: the conventional one. This is attributed to the Dirac-fermion-like
233: linear dispersion relation around the band center, where the
234: widths of the LL gaps are proportional to $\sqrt B$ instead of
235: $B$. We also notice that, $\sigma_{xy}$ always vanishes at
236: $E_{F}=0$ for all $W$, due to the fact that the whole particle or
237: hole band carries zero total Chern number as a topological invariant
238: in the disordered system.
239: \begin{figure}[tbp]
240: %\includegraphics[width=2.0in]{figure1.eps}
241: \par
242: \includegraphics[width=3.2in]{figure3.eps}
243: \caption{(a) Phase diagram for the unconventional QHE regime in
244: graphene at $M=48$, which is symmetric about $E_F=0$. (b) to (d):
245: Normalized localization lengths calculated for three bar widths
246: $L_{y}=48$, $96$ and $144$, as the
247: phase boundary is crossed by the paths indicated by the arrows
248: $B$, $C$ and $D$ in (a), respectively.}
249: \vskip-0.25cm
250: \end{figure}
251:
252: We further study the quantum phase transition of the graphene
253: electron system and establish the phase diagram for the QHE. This
254: can be done relatively conveniently by calculation of the
255: finite-size localization length $\lambda$ on an essentially
256: infinitely long bar of width $L_y$ (length $L_{x}\geq 10^{6}$) by
257: using the well-established recursive Green's function
258: approach~\cite{RECUR}. We present the calculated phase diagram in
259: Fig.\ 3a, for a relatively large magnetic flux
260: $\phi=\frac{2\pi}{48}$ for clarity, while the topology of the
261: phase diagram is essentially universal, independent of $\phi$. In
262: the $W-E_{F}$ plane, different QHE plateaus with $\sigma_{xy}=\nu
263: \frac {e^2} h$ are separated by extended states, where $\lambda$
264: grows linearly with increasing bar width $L_y$. With the increase
265: of $W$, each plateau can be destroyed through a transition $\nu
266: \rightarrow 0$ to the insulating phase and higher plateaus
267: disappear first. In Fig.\ 3b-3d, we show examples of calculated
268: localization length to explain how the phase boundaries in the
269: phase diagram are determined. In Fig.\ 3b, the normalized
270: localization length $\lambda/L_y$ for $E_{F}=-0.3t$ and three
271: sample widths $L_y=48$, $96$ and $144$ is plotted as a function of
272: $W$, which corresponds to a $\nu=-2\rightarrow 0$ transition as
273: indicated by the arrow $B$ in the phase diagram. The sample length
274: $L_{x}$ ranges from $10^{6}$ up to $5\times 10^{6}$, so that the
275: relative error due to statistical fluctuations in $\lambda$
276: reduces to about $2\%$. We see clearly that $\lambda/L_y$ is
277: peaked at $W=W_c\simeq 2.7t$, an indication of an extended
278: critical point separating the $\nu=-2$ plateau from the outside
279: insulating phase. Finite-size scaling~\cite{RECUR} confirms that
280: the localization length at the thermodynamic limit becomes
281: divergent at $W_c$. This is consistent with a new ``float-up''
282: picture~\cite{sheng}, where some negative Chern number states are
283: coming from lower energy, and moving toward the band center with
284: increasing $W$, which sweep across $E_{F}=-0.3t$ at $W=W_c$,
285: causing the collapse of the $\nu=-2$ plateau.
286:
287: Figure 3c shows the normalized localization length as a function
288: of $W$ at $E_{F}=-0.55t$, corresponding to the path indicated by
289: the arrow $C$ in the phase diagram Fig.\ 3a. We see that a peak of
290: localization length occurs at $W=W_c\simeq 1.6t$. Similarly to
291: Fig.\ 3b, the peak indicates the destruction of the $\nu=-6$ QHE
292: state and its transition into the insulating phase. However, we
293: note that here the localization length $\lambda$ is relatively
294: large, being much greater than the largest $L_{y}$ that is
295: reachable in our calculations. So $\lambda/L_{y}$ does not
296: decrease visibly with increasing $L_{y}$, and one cannot rule out
297: a possibility that the higher plateau to insulator transition
298: happens in a critical region with a small finite width $\Delta
299: W_c$ instead of at a critical point $W_c$, where electron states
300: become delocalized. All the phase boundaries separating the QHE
301: phases from the insulating phase at strong $W$, indicated by the
302: solid line with open circles in Fig. 3a, are determined in the
303: same way.
304:
305: To determine the phase boundary between different QHE states, the
306: localization length is calculated as a function of electron Fermi
307: energy $E_F$ for fixed $W$. As shown in Fig.\ 3d, corresponding to the
308: path indicated by the arrow $D$ in Fig. 3a, a peak in
309: $\lambda/L_y$ occurs at $E_{F}=E_c\simeq -0.46t$, which indicates
310: a critical point separating $\nu=-6$ and $-2$ plateaus. All the
311: phase boundaries indicated by dotted lines with cross symbols are
312: determined in the same manner.
313:
314: \begin{figure}[tbp]
315: %\includegraphics[width=2.0in]{figure1.eps}
316: \par
317: \includegraphics[width=3.2in]{figure4.eps}
318: \caption{The normalized localization length
319: $\protect\lambda/L_{y}$ for three bar widths $L_{y}=48$, $96$ and
320: $144$, when the phase boundary is crossed with varying disorder
321: strength $W$ at $E_{F}=-0.01t$, as indicated by the arrow $E$ in
322: Fig.\ 3a. Inset shows the localization length
323: $\protect%
324: \xi$ at the thermodynamic limit determined through one-parameter
325: scaling.}
326: \vskip-0.25cm
327: \end{figure}
328: An important feature of the phase diagram is that the $\nu=-2$ and
329: $2$ plateaus around the band center are no longer connected to
330: each other in the presence of disorder, separated by a new
331: insulating phase in between. Corresponding to such a phase
332: transition, along the path indicated by the arrow $E$ in Fig.\ 3a,
333: we show $\lambda/L_y$ at $E_{F}=-0.01t$ as a function of $W$ in
334: Fig.\ 4. Clearly a largely reduced critical disorder strength
335: $W_c\simeq 1.0t$ is observed in Fig.\ 4. Following the standard
336: finite-size scaling analysis, we find that all the data in Fig.\ 4
337: can be well fitted by a one-parameter scaling
338: relation~\cite{RECUR} $\lambda/L_y=f(L_y/\xi)$ for $L_y=48, 96 $
339: and $144$. The fitting parameter $\xi(W)$ is the localization
340: length at the thermodynamic limit. Its value is plotted in the
341: inset of Fig.\ 4 as a function of $W$, which becomes divergent at
342: $W_c\simeq 1.0t$. Interestingly, the splitting indicates a
343: singularity in the resistance near $E_{F}=0$, as an insulating
344: phase is characterized by a divergent resistance $\rho_{xx}$,
345: while a plateau state has zero or finite $\rho_{xx}$ (being finite
346: only at the critical point) at low temperature limit. This feature may
347: explain the divergent trend in $\rho_{xx}$ and discontinuity in
348: $\rho_{xy}$ observed near zero gate voltage in the
349: experiments~\cite{Hall0,Hall1}.
350:
351: We have also confirmed the above phase boundaries by calculating
352: the Thouless number~\cite{thouless}, which is proportional to the
353: longitudinal conductance. In particular, we observed that the
354: Thouless number as a function of $E_{F}$ at fixed $W$ shows two
355: peaks near the band center, while a dip occurs at $E_{F}=0$, which
356: is consistent with the splitting of the $\nu=-2$ to 2 transition
357: with a new insulating phase emerging near $E_{F}=0$.
358:
359: In summary, we have numerically investigated the QHE in 2D
360: graphene based upon a lattice model. The experimentally discovered
361: unconventional quantization of QHE is reproduced near the band
362: center, which is understood in terms of the novel distribution of
363: the topological Chern integers in the energy band. The phase
364: diagram indicates a new float-up picture, in which the extended
365: levels move toward band center with increasing disorder strength,
366: causing higher plateaus to disappear first. The unconventional QHE
367: plateaus around the band center are found to be much more stable
368: than the conventional ones near the band edges. A new insulating
369: phase is predicted to emerge at the band center, between two
370: $\nu=\pm 2$ QHE states, which is consistent with the
371: experimentally observed resistance discontinuity near zero gate
372: voltage.
373:
374: \textbf{Acknowledgment:} This work is supported by ACS-PRF
375: 41752-AC10, Research Corporation Fund CC5643, the NSF
376: grant/DMR-0307170 (DNS), a grant from the Robert A. Welch
377: Foundation under the grant no. E-1146 (LS), and the NSFC grants
378: 10374058 and 90403016 (ZYW).
379: \begin{references}
380: \bibitem{G0} K. S. Novoselov $et$ $al.$, Science {\bf 306}, 666
381: (2004).
382: \bibitem{G1} C. Berger $et$ $al.$, J. Phys. Chem. B {\bf 108}, 19912
383: (2004).
384: \bibitem{G2} Y. Zhang, J. P. Small, W. V. Pontius and P.
385: Kim, Appl. Phys. Lett. {\bf 86}, 073104 (2005); Y. Zhang, J. P.
386: Small, E. S. Amori and P. Kim,
387: Phys. Rev. Lett. {\bf 94}, 176803 (2005).
388: \bibitem{G3} J. S. Bunch $et$ $al.$, Nano Lett. {\bf 5}, 287
389: (2005).
390: \bibitem{S0} F. D. M. Haldane, Phys. Rev. Lett. {\bf 61}, 2015
391: (1988).
392: \bibitem{Hall0} K. S. Novoselov $et$ $al.$, Nature {\bf 438}, 197 (2005).
393: \bibitem{Hall1} Y. Zhang, Y.-W. Tan, H. L. Stormer, and Philip Kim,
394: Nature {\bf 438}, 201 (2005).
395: \bibitem{phystoday} M. Wilson, Physics Today {\bf 59}, 21 (2006).
396: \bibitem{T1} V. P. Gusynin and S. G. Sharapov
397: Phys. Rev. Lett. 95, 146801 (2005).
398: \bibitem{T2} N. M. R. Peres, F. Guinea and A. H. C. Neto,
399: cond-mat/0506709.
400: \bibitem{T3} E. McCann and V. I. Fal'ko, cond-mat/0510237.
401: \bibitem{T0} Y. Zheng and T. Ando, Phys. Rev. B {\bf 65}, 245420
402: (2002).
403: \bibitem{thouless1} D. J. Thouless, \emph{et. al.}, Phys. Rev.
404: Lett. \textbf{49}, 405 (1982).
405: \bibitem{lisheng} L. Sheng, D. N. Sheng, C. S. Ting and F. D. M.
406: Haldane, Phys. Rev. Lett. {\bf 95}, 136602 (2005).
407: \bibitem{sheng} D. N. Sheng and Z. Y. Weng, Phys. Rev. Lett. {\bf 78},
408: 318 (1997); D. N. Sheng, Z. Y. Weng and X. G. Wen, Phys. Rev. B
409: {\bf 64}, 165317 (2001).
410: \bibitem{bhatt}
411: Y. Huo and R. N. Bhatt, Phys. Rev. Lett. ${\bf 68}$, 1375 (1992).
412: \bibitem{RECUR} A. MacKinnon and B. Kramer, Phys. Rev. Lett. {\bf
413: 47}, 1546 (1981); Z. Phys. B {\bf 53}, 1 (1983).
414: \bibitem{thouless} J. T. Edwards and D. J. Thouless, J. Phys. C {\bf
415: 5}, 807 (1972).
416: \end{references}
417: \end{document}
418:
419: