1: % User-ID: cond-mat/0210428

2: % Password: teprf

3: 

4: 

5: \documentclass[twocolumn,showpacs,prl]{revtex4}

6: %\documentclass[showpacs,prl]{revtex4}

7: 

8: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

9: 

10: \newcommand{\BE}{\begin{equation}}

11: \newcommand{\EE}{\end{equation}}

12: \newcommand{\BA}{\begin{eqnarray}}

13: \newcommand{\EA}{\end{eqnarray}}

14: \textheight 9.75in

15: %\textwidth 6.7in \oddsidemargin -0.25in

16: %\topmargin -0.5in

17: 

18: \usepackage{graphicx}% Include figure files

19: \usepackage{dcolumn}% Align table columns on decimal point

20: \usepackage{bm}% bold math

21: 

22: \begin{document}

23: 

24: \title{Are waves all localized in two dimensional random media?}

25: 

26: \author{Zhen Ye}\affiliation{Department of Physics, National Central University,

27: Chungli, Taiwan 32054}

28: 

29: %\date{February 2, 2002}

30: \date{October 16, 2002}

31: 

32: \begin{abstract}

33: 

34: It has been the dominant view for over two decades that all waves

35: are localized in two dimensions for any given amount of disorder.

36: Here, questions are raised about this assertion. It is shown that

37: there is a lack of the convincing support of the claim. Rather,

38: the recent evidence tends to indicate that waves are not

39: necessarily always localized in two dimensional random systems.

40: 

41: \end{abstract}

42: 

43: \pacs{43.25.Fx, 71.55.Jv} \maketitle

44: 

45: The concept of localization was originally introduced by

46: Anderson\cite{Anderson58} for electrons in a crystal. In the case

47: of a perfectly periodic lattice, except in the gaps all the

48: electronic states are extended and are represented by Bloch

49: states. When a sufficient amount of disorders is added to the

50: lattice, for example in the form of random potentials, the

51: electrons may become spatially localized due to the multiple

52: scattering by the disorders. In such a case, the eigenstates are

53: exponentially confined in the space. By a scaling

54: analysis\cite{gang4}, Abraham {\it et al}. suggested that there

55: can be no metallic state or metal-insulator transition in two

56: dimensions in zero magnetic field. In other words, all electrons

57: are always localized in two dimensions (2D), as reviewed in

58: \cite{Lee,EA}.

59: 

60: The fact that the electronic localization is due to the wave

61: nature of electrons has led to the conjecture that the

62: localization phenomenon also exists for classical waves in random

63: media. And all predictions for the electronic localization are

64: believed to hold for classical waves.

65: %Now the general perception

66: %of localization is: It refers to situations in which due to

67: %multiple scattering by disorders waves are confined in space, and

68: %remain trapped until dissipated; meanwhile the energy density

69: %decays exponentially from the transmitting point.

70: Following the scaling analysis of the electronic localization, it

71: was widely accepted that all waves are localized in 2D random

72: media. This has been the prevailing view for the past twenty years

73: (e.~g. Ref.~\cite{Condat,McCall,Sigalas}). Hereafter I will refer

74: to this view as `2D Conjecture'

75: 

76: In this Letter, I propose that the popular view on 2D localization

77: may not be valid. While the `2D Conjecture' has been challenged

78: and shown to be likely incorrect for electronic systems (Reviewed

79: in \cite{EA}), here I will focus on classical waves. For the

80: purpose, I will first review the origin of the `2D Conjecture' and

81: the current theory on 2D wave localization. Then point out the

82: ambiguities in the theory and discuss the evidence that is in

83: conflict with the `2D Conjecture'. We take the following steps.

84: (1) Re-inspect the scaling analysis, checking for ambiguities. (2)

85: Examine the predictions from the current theory, checking for its

86: validity. (3) Examine the previous experimental and numerical

87: results that claim to support the `2D Conjecture', checking for

88: their appropriateness. (4) Find the self-conflicting points in the

89: current theory, and discuss an apparent mechanism which is in

90: conflict with the theory. The idea is that if there are conflicts,

91: the 2D conjecture should be at least skeptical. The confidence is:

92: Wave scattering in many 2D systems is exactly calculable following

93: Twersky\cite{Victor}; thus previous predictions can be put under a

94: close scrutiny. We note that the electronic system is more

95: complicated because effects such as the Coulomb interaction makes

96: data interpretation difficult. In this sense, the classical wave

97: systems are advantageous in studying localization effects.

98: 

99: \input{epsf}

100: \begin{figure}[hbt]

101: \begin{center}

102: \epsfxsize=2.25in \epsffile{fig1.eps} \caption{ \label{fig1}\small

103: The scaling function $\beta$ versus  $\ln R$ from

104: Eq.~(\ref{eq:4})}

105: \end{center}

106: \end{figure}

107: 

108: {\bf The scaling analysis} According to \cite{gang4}, an

109: hypercubic geometry is used for the scaling analysis. In the

110: metallic state, the resistance follows the Ohmic behavior \BE R

111: \sim L^{2-d},\label{eq:1}\EE where $d$ is the dimension. For a

112: localized state, i.~e. large $R$, the resistance grows

113: exponentially \BE R\sim e^{L/L_1},\label{eq:2}\EE where $L_1$ is

114: the localization length which may differ for different dimensions.

115: A scaling function is defined as $\beta = \frac{\partial \ln

116: R}{\partial \ln L}.$ Taking Eqs.~(\ref{eq:1}) and (\ref{eq:2})

117: into this, we obtain the asymptotic behavior \BE \beta \sim

118: \left\{

119: \begin{array}{ll} \ln R, & \mbox{as} \ R\rightarrow \infty \

120: (\mbox{Localized})

121: \\ 2-d, & \mbox{as} \ R \rightarrow 0 \ (\mbox{Ohmic})

122: \end{array}\right. \label{eq:4}

123: \EE From the asymptotic behavior in Eq.~(\ref{eq:4}), one can

124: sketch the universal curves in $d=1,2,3$ dimensions. The central

125: assumptions in \cite{gang4} are (1) $\beta$ is continuous; (2)

126: $\beta$ is a function of $R$ and depends on other parameters such

127: as disorders and length scale only through $R$; and (3) once wave

128: is localized, the increasing sample size would always mean more

129: localization.

130: 

131: The behavior of $\beta$ is plotted in Fig.~\ref{fig1}. It is clear

132: that in the 3D case, the curve crosses the horizontal axis,

133: yielding an unstable fixed point ($B$). Above this point, the

134: waves become more and more localized as the sample size increases.

135: Below the critical point, the system tends to follow the Ohmic

136: behavior as the sample size is enlarged. This fixed point

137: separates the localized and non-localized states. For the two

138: dimensional case, in the Ohmic regime $\beta$ approaches zero as

139: $\ln(R) \rightarrow 0$. But the perturbation calculation including

140: the wave interference effect shows that $\beta$ is always greater

141: than zero. Therefore for both one and two dimensions, the curves

142: do not cross the horizontal axis, and there is thus no fixed

143: point. As the sample size increases, all states move towards the

144: localization regime. {\it This has been the main reason that led

145: previously to the conclusion that all waves are localized in one

146: and two dimensions.}

147: 

148: {\bf The current localization theory} Now I briefly review the

149: existing theory for localization. As wave propagates in random

150: media, it experiences multiple scattering, and as a result, the

151: wave loses its phase, leading to the gradual decreases of the

152: coherence of the wave in the absence of absorption. Meanwhile,

153: diffusive wave is built up as more and more scattering takes

154: place. The procedure to obtain the localization state can be

155: briefly summarized as follows.

156: 

157: The quantity, $D^{(B)}$ which is a measure of diffusion of

158: classical waves is called the classical Boltzman diffusion

159: constant and it may be derived under the coherent potential

160: approximation \cite{lax}, and is given as

161: \begin{equation}

162: D^{(B)} \sim \frac{v_t l}{d}

163: \end{equation}

164: where $v_t$ is the transport velocity, $l$ is the mean free path

165: and $d$ is the dimensionality.

166: 

167: As waves scattered along any two reversed paths in the backward

168: direction interfere constructively, leading to the enhanced

169: backscattering effect, which will add corrections to the diffusion

170: coefficient. In the field theory approach, such an enhanced

171: backscattering effect is represented by a set of maximally crossed

172: ladder diagrams\cite{DV}. In the two dimension case, the

173: evaluation of these diagrams leads to an integration for which two

174: cut-off limits have to be introduced to avoid the divergence. The

175: correction to the diffusion constant for two dimensional systems

176: is thus found as

177: \begin{equation}

178: \delta D \sim  - {\rm ln}(L_M/l_m) \label{eq:correction}

179: \end{equation}

180: where $L_M$ and $l_m$ are the two cut-off limits. It is then

181: interpreted in the previous theory that the cut-off limit $l_m$ is

182: a measure of the minimum scaling for the waves and is thought to

183: be related to (for example) the mean free path, whereas $L_M$ is a

184: measure of the effective size of the sample. It is rather

185: important to note that the correction in Eq.~(\ref{eq:correction})

186: is not only negative but diverges as $L_M$ approach infinity. This

187: is obviously unphysical, since the corrected diffusion constant

188: cannot be negative. To avoid the problem, it was suggested that

189: $L_M$ is related to the localization range, or simply the

190: localization length, in such way that when $L_M$ is equal to the

191: localization length denoted by $\xi$ say, the corrected diffusion

192: coefficient becomes zero: \BE D_R(\xi) = D^{(B)} + \delta D(\xi) =

193: 0. \label{eq:d} \EE The localization length $\xi$ is subsequently

194: solved for from this equation. It is obvious that this equation

195: {\it always} allows a solution. {\it Therefore a localization

196: length can always be found in two dimensions.} {\it Such a

197: backscattering induced absence-of-diffusion mechanism is the core

198: of the current theory of localization in two dimensions, and is

199: considered a strong support of the `2D Conjecture' from the

200: scaling analysis}\cite{gang4}.

201: 

202: {\bf On the scaling analysis} While it is simple and

203: straightforward, the above scaling analysis is not without

204: ambiguities in investigating the localization effect. The reasons

205: follow. Whether a system has non-localized or only localized

206: states is an intrinsic property of the system, and should not rely

207: on neither the boundary nor the source. As long as the analysis

208: cannot exclude the possibility that the boundary or the source is

209: playing a role, the consequence from the analysis is deemed to be

210: questionable. In order to isolate the localization or

211: non-localization effect, therefore, a genuine analysis should not

212: be plagued by boundary effects not only in the localization region

213: but also in the non-localization region. Of course, if the system

214: has indeed only localized states, the boundary is not an issue, as

215: the dependence on the boundary is exponentially vanishing.

216: However, the care must be taken for the non-localized regime. It

217: is not difficult to see that the above scaling theory may work for

218: situations when both probing contacts, used to measure the

219: resistance or conductance from which the localization is inferred,

220: are located outside the system. In this case, the Ohmic behavior

221: given by Eq.~(\ref{eq:1}) is valid under the condition that the

222: current flows uniformly in one direction. This is possible only

223: with properly scaled sources and with the presence of confining

224: boundaries, obviously in conflict with the proclamation that

225: whether it is a localization or non-localized state is the

226: intrinsic property of the system and should not rely on a boundary

227: nor a source. Thus the above analysis is more appropriate for

228: studying transport phenomena. It is our opinion that the reduction

229: in the conductance does not necessarily mean that all waves are

230: actually localized. In other words, it is necessary to

231: differentiate the situation that the electrons are prohibited from

232: transmission through a random medium from the situation that the

233: system has only localized states.

234: 

235: Not to be neglected, the recent numerical results on acoustic

236: waves\cite{PRE2002} do support the above comments on the scaling

237: theory. They point out that waves can be blocked from propagation

238: {\it through} a medium by disorders, but such an inhibition is not

239: necessarily an indication that waves can be localized in the

240: medium when the transmitting source is put {\it inside} the

241: medium. The same conclusion is also obtained from the simulations

242: on propagation of electromagnetic waves (EM) in 2D random

243: dielectric media\cite{Bikash2}. More explicitly, it is shown that

244: waves certainly cannot propagate through a random medium when it

245: has only localized states. But the observation that waves cannot

246: propagate through does not necessarily imply that the system has

247: only localized states.

248: 

249: To this end, it is proper to mention that a recent scaling

250: analysis shows that the difference between 2D and 3D disordered

251: systems is insignificant\cite{scaling}, i.~e. like 3D the

252: localization-delocalization transition is also possible in 2D,

253: which could explain the recently observed `unusual' metallic

254: behavior in 2D electronic systems (See references cited in

255: \cite{EA}). This scaling analysis has been pointed out to be

256: absolutely in line with recent findings obtained through direct

257: calculation of the conductance with the use of the Kubo formula

258: (Y. Tarasov, {\it private communication}).

259: 

260: {\bf On the validity of the current theory} Among many, there are

261: two basic ways to check the validity of the aforementioned theory.

262: 

263: First, the key parameter obtained from the current theory is the

264: localization length. This quantity can also be obtained exactly by

265: numerical computation using the scheme detailed in\cite{Victor}.

266: Then we can make a comparison of the two results. We consider the

267: model of acoustic scattering in water with air-cylinders detailed

268: in \cite{Emile}. The comparison is shown in Fig.~\ref{fig2}. The

269: comparison clearly shows that there is a significant difference

270: between the numerically exact results and the results obtained

271: from the theory. Further comparisons indicate that the difference

272: between the two results is significant not only quantitatively but

273: qualitatively\cite{Bikash3}.

274: 

275: \input{epsf}

276: \begin{figure}[hbt]

277: \begin{center}

278: \epsfxsize=2.25in \epsffile{fig2.eps} \caption{Localization length

279: ($\xi$) is shown as a function of frequency ($ka$) for

280: $\beta$=0.001. The dashed curve with circles represent the exact

281: values obtained numerically while the solid curve is obtained from

282: theory. Here $k$ is the wavenumber, $a$ is the radius of the

283: cylinder, and $\beta$ is the fraction of area occupied by the

284: cylinders per unit area.} \label{fig2}

285: \end{center}

286: \end{figure}

287: 

288: Second, one can evaluate the localization length from the theory.

289: Using an exactly solvable model, then inspect the spatial

290: distribution of the wave energy density and check whether

291: localization occurs when the system is larger than the obtained

292: localization length. Again we consider the model of acoustic

293: localization in air-filled cylinders in water. From

294: Fig.~\ref{fig2}, the theory predicts that the shortest

295: localization length is around $ka=0.005$ which is the nature

296: frequency of an air-cylinder in water\cite{Emile}. The numerical

297: results show, however, that no localization occurs around this

298: frequency. Instead, localization appears at somewhat higher

299: frequency ranges with the same system size, referring to Fig.~3 of

300: \cite{Emile}.

301: 

302: {\bf On the previous experimental and numerical evidence} It was

303: widely believed that the theory of 2D localization has been tested

304: experimentally to be successful. We find that the claimed success

305: is mainly based on two types of experiments. One is the indirect

306: method which measures the effects of the enhanced backscattering

307: (e.~g. Ref.~\cite{2Dexp}). In a rigorous simulation, it has been

308: shown that the enhanced backscattering is not related to the

309: localization\cite{aad}. Waves are not necessarily localized when a

310: strong enhanced backscattering exits, and sometimes waves can be

311: localized even when the enhanced backscattering is weak. Asides

312: from few exceptions\cite{McCall}, the other type of experiments is

313: based on observations of the exponential decay of waves as they

314: propagate {\it through} disordered media, as stated in

315: Ref.~\cite{Nat}. According to the above and following discussions,

316: this type of experiments is {\it not} sufficient to discern

317: whether the medium really only has localized states. Unwanted

318: effects of non-localization origin can also contribute to the

319: wanted exponential decay, making data interpretation ambiguous. In

320: fact, the authors in Ref.~\cite{Sigalas} pointed out that there is

321: no conclusive experimental evidence for localization of EM waves

322: in 2D. We mention that there was a report of the observation of

323: microwave localization in two dimensions when a transmitting

324: source is inside disordered media\cite{McCall}. However, the

325: diffusion based theory has not been verified against this

326: experimental result.

327: 

328: The same situation can be said about the numerical simulations.

329: Take Ref.~\cite{Sigalas} as the example. The authors considered

330: the EM propagation in a random array of dielectric cylinders. The

331: localization length is computed from the reduction in the

332: transmission {\it across} the random sample. Following their

333: method, we first compute the transmission versus the sample size

334: at various frequencies with the source and the receiver located at

335: the opposite sides of the sample. We do observe exponential

336: decays, and the decay rates depend on the frequency. According to

337: \cite{Sigalas}, the localization lengths can be estimated from

338: this decay rates. Then as long as the sample size is bigger than

339: this length, we would also expect to observe an exponential decay

340: in the transmission when source is put inside the medium. But we

341: found that the exponential decay disappears for some frequencies.

342: As an example, the results for two frequencies are shown in

343: Fig.~\ref{fig3}. In line with \cite{Sigalas}, the following

344: parameters are used in the computation. The dielectric constants

345: of the cylinders and the medium are 10 and 1 respectively. The

346: fraction of area occupied by the cylinders per unit area, is 0.28.

347: The radius $a$ of the cylinders is 0.38 cm. The lattice constant

348: $d$ of the corresponding square lattice array is calculated as

349: 1.28 cm. All lengths are scaled by the lattice constant $d$. Here

350: we see that the exponential decay at 14.70GHz, shown when the

351: transmission is across the sample, disappears when the source is

352: moved into the medium. The results suggest that waves are not

353: localized at this frequency. One may still argue that the

354: non-localization is due to the fact that the localization length

355: is long compared to the sample size in the `Inside' case. Even if

356: this were the case, the exponential decay shown for the `Outside'

357: case could not be due to the localization effect. Reiterating, it

358: is not sufficient to extract the localization effect by merely

359: computing the transmission reduction across the sample. Therefore

360: the claim about the 2D localization like in \cite{Sigalas} is not

361: appropriate. At 11.75GHz, we observe that the exponential decay

362: with nearly the same slop holds for both cases - the slight

363: difference is due to the finite width in the `Outside'

364: case\cite{PRE2002}, and reveals the genuine localization behavior.

365: 

366: \input epsf.tex

367: \begin{center}

368: \begin{figure}[hbt]

369: \epsfxsize=2.5in\epsffile{fig3.eps} \caption{The logarithmic

370: average transmission T and its fluctuation versus the sample size

371: for two frequencies. The estimated slops for the transmission are

372: indicated in the figure. The `Outside' and `Inside' cases refer

373: respectively to when the transmitting source is located outside

374: and inside the medium. More details on the notations and the model

375: are in \cite{Bikash2}} \label{fig3}

376: \end{figure}

377: \end{center}

378: 

379: {\bf The physical picture of localization} It seems that a general

380: picture of localization may be obtained. For quantum mechanic or

381: acoustic waves (the same argument also holds for EM

382: systames\cite{PRER}), the current can be written as $ \vec{J} \sim

383: \mbox{Re}[\psi(-i)\nabla\psi],$ where $\psi$ stands for the wave

384: function for quantum mechanical systems and for the pressure in

385: acoustic systems. Writing the field as $ \psi = |\psi|

386: e^{i\theta},$ the current becomes $\vec{J} \sim

387: |\psi|^2\nabla\theta.$ It is clear that when $\theta$ is constant

388: at least by domains while $|\psi| \neq 0$, the flow stops, i.~e.

389: $\vec{J}=0$, and the wave or the energy is localized in space,

390: i.~e. $|\psi|^2 \neq 0$. Obviously the constant phase $\theta$

391: indicates the appearance of a coherence in the system. This

392: coherent-phase picture has been demonstrated successfully not only

393: for two dimensional media\cite{Emile}, but for one and three

394: dimensions as well\cite{Luan,Hsu}.

395: 

396: The current diffusion-based theory does not support the above

397: picture. The physical picture of the theory is: Waves will undergo

398: a diffusion process when the system is smaller than the

399: localization size. As the system increases, the diffusion

400: gradually diminishes and finally comes to a complete stop when the

401: size exceeds the localization length. If this picture were valid,

402: then one would expect a significant change in the spatial

403: distribution of the energy density, from that of diffusion to the

404: exponentially confined envelop. This is not evident in the theory

405: and is not supported by numerical results. In contrast, the

406: numerical results in Fig.~\ref{fig3} show that the exponential

407: decay starts even when the sample size is smaller than the

408: localization length; the distribution of the characters of a

409: diffusion process does not appear.

410: 

411: In summary, concerns have been raised about the previous claim

412: that all waves are localized in 2D. In fact, non-localized states

413: have already been reported by numerical computation in, for

414: instance, Ref.~\cite{Emile}. Even if this could be argued to be

415: due to the finite sample size limited by computing facilities,

416: there are still many other reasons for being doubtful about the

417: `2D Conjecture'. Some important reasons are corroborated here.

418: 

419: \begin{thebibliography}{99}

420: 

421: \bibitem{Anderson58} P. W. Anderson, Phys. Rev. {\bf 109}, 1492 (1958).

422: 

423: \bibitem{gang4} E. Abrahams, P. W. Anderson, D. C. Licciardello, and

424: T. V. Ramakrishnan, Phys. Rev. Lett. {\bf 42}, 673 (1979).

425: 

426: \bibitem{Lee} P. A. Lee and Ramakrishnan, Rev. Mod. Phys. {\bf

427: 57}, 287 (1985).

428: 

429: \bibitem{EA} E. Abrahams, S. Kravchenko, and M. P. Sarachik, Rev.

430: Mod. Phys. {\bf 73}, 251 (2001).

431: 

432: \bibitem{Condat} C. A. Condat and T. R. Kirkpatrick, Phys. Rev. B {\bf 36}, 6782

433: (1987); C. A. Condat, J. Acoust. Soc. Am. {\bf 83}, 441 (1988);

434: 

435: %\bibitem{local} S. John, H. Sompolinski, and M. J. Stephen, Phys. Rev.  B {\bf

436: %27}, 5592 (1983).

437: 

438: \bibitem{McCall} R. Dalichaouch, J. P. Armstrong, S. Schultz, P. M. Platzman

439: and S. L. McCall, Nature {\bf 354}, 53 (1991); S. L. McCall, et

440: al., Phys. Rev. Lett. {\bf 67}, 2017 (1991).

441: 

442: %\bibitem{opt} A. R. McGun, P. Sheng, and A. A. Maradudin, Opt.

443: %Commun. {\bf 91}, 175 (1992).

444: 

445: \bibitem{Sigalas} M. M. Sigalas, C. M. Soukoulis, C.-T. Chan, and

446: D. Turner, Phys. Rev. B {\bf 53}, 8340 (1996).

447: 

448: %\bibitem{AAA} A. A. Asatryan, et al. Phys. Rev. B {\bf 57}, 13535

449: %(1998).

450: 

451: \bibitem{Victor} V. Twersky, J. Math. Phys. (N.Y.) {\bf 3}, 700 (1962).

452: 

453: \bibitem{lax} M. Lax, Rev. Mod. Phys. {\bf 23}, 287 (1951).

454: 

455: \bibitem{DV}  D. Vollhardt and P. W\"olfle, Phys. Rev. B {\bf 22},

456:               4666 (1980).

457: 

458: \bibitem{PRE2002} Y.-Y. Chen and Z. Ye, Phys. Rev. E {\bf 65},

459: 056612 (2002).

460: 

461: \bibitem{Bikash2} B. C. Gupta, Y.-Y. Chen, and Z. Ye,

462: cond-mat/0210171.

463: 

464: \bibitem{scaling} Z. Ye, cond-mat/0106481; Z. Ye, Chin.

465: J. Phys. {\bf 39}, L207 (2001)

466: 

467: %\bibitem{Tarasov} Yu. V. Tarasov, private communication.

468: %Cond-mat/0207480.

469: 

470: %\bibitem{PRL2001} Y.-Y. Chen and Z. Ye, Phys. Rev. Lett. {\bf 87}, 184301 (2001).

471: 

472: \bibitem{Emile} E. Hoskinson and Z. Ye, Phys. Rev. Lett. {\bf 83},

473: 2734 (1999).

474: 

475: \bibitem{Bikash3} B. C. Gutpa and Z. Ye, cond-mat/0205161.

476: 

477: \bibitem{2Dexp} G. J. Dolan and D. D. Osheroff, Phys. Rev. Lett.

478: {\bf 43}, 721 (1979); D. J. Bishop, C. Dynes, and D. C. Tsui,

479: Phys. Rev. B {\bf 26}, 773 (1982); M. van Albada, A. Lagendijk,

480: Phys. Rev. Lett. 55, 2692 (1985); P. E. Wolf, G. Maret, Phys. Rev.

481: Lett. 55, 2696 (1985).

482: 

483: \bibitem{aad} Z. Ye and A. Alvarez, Phys. Rev. Lett. {\bf 80},

484: 3503 (1998).

485: 

486: %\bibitem{Sridhar} P. Pradhan and S. Sridhar, Phys. Rev. Lett. {\bf 85}, 2360 (2000).

487: 

488: \bibitem{Nat} A. A. Chabanov, M. Stoytchev, and A. Z. Genack,

489:                 Nature {\bf 404}, 850 (2000).

490: 

491: \bibitem{PRER} Z. Ye, S. Li, and X. Sun, Phys. Rev.

492: E {\bf 66}, 045602(R) (2002).

493: 

494: \bibitem{Luan} P. G. Luan and Z. Ye, Phys. Rev. E {\bf 63}, 066611

495:                (2001).

496: 

497: \bibitem{Hsu} Z. Ye and H. Hsu, Appl. Phys. Lett. {\bf 79}, 1724

498: (2001).

499: 

500: \end{thebibliography}

501: \end{document}

502: