cond-mat0005339/bf4.tex
1: \documentstyle[aps,prl,multicol,epsfig]{revtex}
2: 
3: \newcommand {\be}{\begin{equation}}
4: \newcommand {\ee}{\end{equation}}
5: 
6: \begin{document}
7: 
8: \title{Spreading and localization of wavepackets in 
9: disordered wires in a magnetic field}
10: \author{Matthias~Weiss, Tsampikos~Kottos, and Theo~Geisel\\
11: Max-Planck-Institut f\"ur Str\"omungsforschung und Fakult\"at Physik
12: der Universit\"at G\"ottingen,\\ Bunsenstra\ss e 10, D-37073 G\"ottingen, Germany}
13: \maketitle
14: 
15: \begin{abstract}
16: We study the diffusive and localization properties of wavepackets in disordered 
17: wires in a magnetic field. In contrast to a recent supersymmetry approach our
18: numerical results show that the decay rate of the steady state changes {\em smoothly} 
19: at the crossover from preserved to broken time-reversal symmetry. Scaling and 
20: fluctuation properties are also analyzed and a formula, which was derived analytically
21: only in the pure symmetry cases is shown to describe also the steady state 
22: wavefunction at the crossover regime. Finally, we present a scaling for the variance 
23: of the packet which shows again a smooth transition due to the magnetic field.
24: \end{abstract}
25: \pacs{PACS numbers: 71.55.Jv ; 05.44.+b }
26: 
27: \begin{multicols}{2}
28: 
29: In recent years, considerable progress has been made in understanding the structure
30: of eigenstates of quasi-one-dimensional (quasi-1D) disordered systems (see~\cite{FM94}
31: and references therein). Most of the investigations have been concerned with the
32: cases of completely preserved and totally broken time-reversal symmetry. The transition
33: between these limits, however, has been studied much less and little is known about 
34: it. The only results for the transition were based on a heuristic approach by 
35: Bouchaud~\cite{B91}, a semiclassical analysis by Imry and Lerner~\cite{LI95} and 
36: numerical studies based on transfer matrices by Pichard et al.~\cite{PSSD90}. The main 
37: prediction of the above studies is the doubling of the localization length in the 
38: limit of strong magnetic fields while a smooth transition towards this asymptotic 
39: limit was assumed. Similar results were obtained for quantum chaotic models~\cite{SB92}. 
40: Recently, the doubling of the localization length was 
41: observed in sub-micron thin wires of doped GaAs~\cite{KGB98}, where for increasing 
42: magnetic field strengths a continously decreasing activation energy was reported, 
43: which saturated indeed at half of its field free value. Motivated by this experiment, 
44: the first attempt to study such a transition using the supersymmetry technique was 
45: undertaken in~\cite{KE99,KE00}. These authors predict that the decay of wavefunctions 
46: in disordered wires in weak magnetic fields, is characterized by two localization 
47: lengths; the far tails decay with the length $l_{\rm tails}$ characteristic for 
48: completely broken time-reversal symmetry, while at shorter distances the decay 
49: length is $l_{\rm orig} =l_{\rm tails}/2$. 
50: 
51: The predictions of~\cite{KE99,KE00} imply two different temperature regimes in the 
52: hopping conductivity separated by a boundary which depends on the magnetic field 
53: strength. A direct numerical test of the above prediction by means of the transmission 
54: approach was carried out recently in~\cite{SB99}. The results are contradicting 
55: the two-scale behaviour of the wavefunction in the crossover regime. The following 
56: possibilities were proposed in order to explain this discrepancy~\cite{KE99,KE00,SB99}:
57: (a) the Borland conjecture (which connects the asymptotic decay length of the
58: transmittance to the asymptotic decay length of the wavefunction) breaks down
59: in the crossover regime between the preserved and broken time reversal symmetry,
60: and (b) the two-scale localization phenomenon is due to anomalously localized 
61: states that are irrelevant for a typical wire. The latter case appears more likely
62: since in~\cite{KE99,KE00} the average wavefunction was calculated whereas, in~\cite{SB99} 
63: under the assumption of the Borland conjecture the authors calculated the average 
64: of the logarithm of the wavefunction. We are going to show that none of 
65: the above reasons can resolve the contradictions!
66: 
67: In this Letter the problem of the crossover behaviour, from preserved to broken 
68: time reversal symmetry, is addressed from a totally different perspective. In 
69: particular we study the evolution of initially $\delta$-like wavepackets in a 
70: quasi-1D geometry under the influence of a magnetic field. We then directly 
71: calculate the logarithmic and the arithmetic average of the steady state 
72: wavefunction. We show that the extracted decay rates change smoothly as a function 
73: of the magnetic field for both observables. Moreover, we study the whole steady 
74: state distribution and its fluctuations as a function of the magnetic field. 
75: From our numerical data we extract an analytical formula for the asymptotic 
76: wavefunction in the crossover regime. Finally, we present arguments according 
77: to which the diffusive constant scales smoothly with respect to the magnetic 
78: field. This imposes a smooth scaling behaviour of the variance of the packet. 
79: Our predictions are confirmed by extensive numerical calculations.
80: 
81: The mathematical model we consider is the time-dependent Schr\"odinger 
82: equation on a 1D lattice,  
83: %
84: \be\label{eqmo}
85: i\,{\frac{dc_n(t)}{{dt}}}=\,\sum_{m=n-b}^{n+b}H_{nm}c_m\quad , 
86: \ee
87: %
88: where $c_n(t)$ is the probability amplitude for an electron to be at site~$n$
89: and $H_{nm}=H_{nm}^0 + i\alpha A_{nm}$ is a complex Hermitian Band Random 
90: Matrix (BRM), which is decomposed into a real symmetric matrix~$H^0$ and a real 
91: antisymmetric matrix~$A$ with imaginary weight $i\alpha$. The entries of the
92: two matrices are independent Gaussian random numbers with variance $\sigma^2=1
93: +\delta_{nm}$ (where $\delta_{nm}$ is the Kronecker symbol) if $|n-m|\le b$ 
94: and zero otherwise. The parameter~$b$ defines the hopping range between 
95: neighbouring sites, or, in the quasi-1D interpretation, the number of transverse 
96: channels along a thin wire. To relate the parameter~$\alpha$ to the magnetic 
97: field~$B$, we note that the perturbation of the levels is proportional to 
98: the magnetic flux $\Phi =k B b l_{\infty} $ through an area limited by the 
99: localization length. Here $k$ is a dimensionless constant of order unity 
100: that depends on the specific geometry of the disordered wire~\cite{KE00,E97}. 
101: Hence we expect $\alpha \sim \Phi/\Phi_0$, where $\Phi_0=h/e$ is the 
102: elementary flux quantum. 
103: 
104: We have integrated Eq.~(\ref{eqmo}) numerically using a Cayley scheme in 
105: order to preserve the norm~\cite{PRT00}. Moreover a self-expanding algorithm
106: was implemented to eliminate finite-size effects. Whenever the probability
107: of finding the particle at the edges of the lattice exceeded $10^{-15}$,
108: $20b$ new sites were added to each end. Since all eigenstates of the 
109: Hamiltonian~$H_{nm}$ are known to be exponentially localized with a localization 
110: length $l_{\infty}(E)\sim b^2$~\cite{loc}, the evolution of the wavepacket 
111: is expected to exhibit a relaxation to a steady-state distribution in the 
112: limit $t\rightarrow \infty$. On the basis of numerical calculations it was 
113: shown in~\cite{IKPRT96} that for the time reversal symmetry the asymptotic 
114: profile $f_s(n)\equiv|c_n(t\to \infty)|^2$ is given by the following expression:
115: %
116: \be\label{Gogolin}
117: f_s(x)=\frac{\pi ^2}{16 l_{\infty}}\int_0^\infty \frac{\eta {\rm sh}(\pi
118: \eta )(1+\eta ^2)^2}{(1+{\rm ch}(\pi \eta ))^2}e^{
119:  -{1+\eta^2 \over 4 l_{\infty}} |x|} d\eta  \;,
120: \ee
121: %
122: where~$l_{\infty}$ is the averaged (over energy) localization length. 
123: Equation~(\ref{Gogolin}) was later proved analytically, for quasi-1D systems 
124: with preserved and broken time reversal symmetry~\cite{Z97}. It is interesting to 
125: note that Eq.~(\ref{Gogolin}) was derived first for continous 1D models 
126: with white noise potential~\cite{G76} with~$l_{\infty}$ being the mean free path. 
127: Remarkably, in spite of the relevant difference between the 1D and quasi-1D case 
128: (the mean free path is of the order of~$b$ in the latter and thus much smaller 
129: than the localization length $l_\infty \sim b^2$), the asymptotic shape remains 
130: the same in both cases. From~(\ref{Gogolin}), one finds that close to the origin,
131: %
132: \be
133: f_s(x)\sim \exp (-|x|/l_{\infty});\,\,\, |x| \leq l_{\infty}, 
134: \ee
135: %
136: while the asymptotic decay is described by 
137: %
138: \be\label{tail}
139: f_s(x)\sim |x|^{-3/2}\exp(-|x|/4l_{\infty});\,\,\, |x| \gg 4l_{\infty}, 
140: \ee
141: %
142: revealing that the decay rate $s(x) = d \log f_s/dx$ changes by a factor~$4$.
143: 
144: \begin{figure}
145: \begin{center}
146: \epsfig{figure=bfigp1.eps,width=8cm}\hfill
147: \end{center}
148: \noindent
149: {\footnotesize {\bf FIG.~1.}
150: Arithmetic average of the asymptotic wavefunction~$f_s(n)$. The bandwidth is $b=15$,
151: and the snapshot is taken at $t=10^4$. The data correspond to various
152: crossover parameters: (a) $\alpha=0$; (b) $\alpha=0.02$; (c) $\alpha=0.085$; 
153: and (d) $\alpha=0.7$.}
154: \end{figure}
155: 
156: In Fig.~1 we report some asymptotic profiles~$f_s(x,\alpha)$ for~$b=15$ and various 
157: values of the time-reversal symmetry breaking parameter~$\alpha$. These data strongly 
158: suggest that the decay of~$f_s(n)$ in the vicinity of the origin is definitely faster 
159: than in the tails. As no analytical results are available for the crossover
160: regime, it is very tempting to compare our numerical results with the theoretical 
161: dependence~(\ref{Gogolin}) derived for quasi-1D systems with pure symmetries. 
162: The profiles reported in Fig.~2a are the result of an arithmetic average over 
163: many realizations of the asymptotic profile for $b=5,10,15$ and several~$\alpha$. 
164: They are plotted with the scaling assumption
165: %
166: \be\label{prof}
167: f_s(n,t\rightarrow \infty )=l_{\infty}f_s(x);\quad x\equiv n/l_{\infty} 
168: \ee
169: %
170: where~$l_\infty$ is determined by a fit according to Eq.~(\ref{Gogolin}).
171: The very good agreement between the numerical results and the analytical curve over a 
172: broad range of $x$-values suggests that a properly modified theory including the 
173: effect of an intermediate magnetic field should be able to account for the asymptotic
174: profile of wavepackets in quasi-1D systems. Furthermore the scaling relation~(\ref{prof}) 
175: implies that the wavefunction in the crossover regime shows the same gross structure 
176: (envelope) as in the pure symmetry cases, on scales comparable to the 
177: localization length. This is in contrast to the occurrence of a second localization 
178: scale for the far tails proposed in~\cite{KE99,KE00}. Hence we expect that all the scaling 
179: laws for the eigenstates dominated by the fluctuations of the "envelope" (e.g., moments 
180: like the inverse participation ratio), also hold for the crossover regime.
181: 
182: 
183: \begin{figure}
184: \begin{center}
185: \epsfig{figure=bfigp2.eps,width=8cm}\hfill
186: \end{center}
187: \noindent
188: {\footnotesize {\bf FIG.~2.}
189: (a) Asymptotic average profile of the wavepackets for $b=5,10,15$ and various
190: crossover parameters~$\alpha$ rescaled according to Eq.~(\ref{prof}). The light 
191: smooth line is the theoretical expression~(\ref{Gogolin}). (b) The localization 
192: lengths~$l_{\infty}$ (open symbols) and $l_{\infty}^{\rm as}$ (full symbols) as 
193: a function of the time reversal symmetry breaking parameter~$\alpha$. Various
194: symbols correspond to various values of~$b$; diamonds to~$b=5$, circles to~$b=10$
195: and triangles to~$b=15$.
196: } 
197: \end{figure}
198: 
199: In Fig.~2b the asymptotic localization length~$l_{\infty}$, as determined from 
200: the best fit with~(\ref{Gogolin}) is plotted versus the crossover parameter~$\alpha$. 
201: One can clearly see that the transition from preserved to broken time-reversal 
202: symmetry is rather smooth and thus we have
203: %
204: \be\label{lalpha}
205: l_{\infty}(\alpha) = \beta(\alpha) l_{\infty}(0)
206: \ee
207: %
208: where~$\beta(\alpha)$ is a smooth function that interpolates between the values~$1$ 
209: and~$2$ for preserved ($\alpha=0$) and totally broken time reversal symmetry 
210: ($\alpha\simeq 1$) respectively. In Fig.~2b we also present the localization 
211: length derived by a direct fit of the average of the logarithm of the asymptotic 
212: wavefunction $l_{\infty}^{\rm as}= \lim_{N\rightarrow \infty} \langle\ln(f_s)
213: \rangle/(2N)$. It clearly shows the same smooth behaviour as $l_{\infty}$. 
214: 
215: Our numerical results are in contrast to the supersymmetry results~\cite{KE99,KE00} 
216: and agree nicely with the transfer matrix calculations~\cite{SB99}. Moreover, they
217: give a definite answer to the question, whether the assumed discrepancy is due 
218: to the fact that in~\cite{KE99,KE00} the logarithm of the averaged wavefunction was 
219: investigated, while in~\cite{SB99} it was the logarithmic average. Based 
220: on our calculations we are now able to exclude both possibilities suggested 
221: in~\cite{KE00,SB99}; namely that one could attribute the two-scale localization phenomenon 
222: to anomalously localized states, that dominate the asymptotic average profile~$f_s(n)$ or 
223: to the non-applicability of the Borland conjecture in the crossover regime.
224: 
225: \begin{figure}
226: \begin{center}
227: \epsfig{figure=bfigp3.eps,width=8cm}\hfill
228: \end{center}
229: \noindent
230: {\footnotesize {\bf FIG.~3.}
231: Distribution of $y\equiv \ln f_s(n=6b^2\pm 2)$ for~$b=5$ and various values for
232: the time-reversal symmetry breaking parameter: (a) $\alpha=0.003$; (b) $\alpha=0.03$; 
233: (c) $\alpha=0.3$; (d) Variance of~$y$ for $b=5,10,15$ and various $\alpha$. The data are rescaled 
234: according to Eq.~(\ref{varsc}).}
235: \end{figure}
236: 
237: As mentioned above, the localization of all eigenstates implies that for $t\rightarrow 
238: \infty$ the quantum steady state~$f_s(n)$ is localized and fluctuates around the average 
239: profile~(\ref{Gogolin}). Thus, one can ask about the distribution of~$f_s(n)$. Having in 
240: mind that the asymptotic profile is exponentially localized (see Eq.~(\ref{Gogolin})), 
241: we computed the distribution of the logarithm of the wavefunction $y\equiv \ln(f_s)$ for 
242: various values of the crossover parameter~$\alpha$. In Figs.~3a-c we report some 
243: representative distributions ${\cal P} (y)$, which refer to the case~$b=5$ with 
244: $\alpha= 0.003,0.03,$ and $\alpha =0.3$. In all cases, we have found that the distribution of the
245: logarithm of the wavefunction ${\cal P} (y)$, is a Gaussian with good accuracy. This is 
246: in perfect agreement with the analytical calculation of~\cite{KE00} for ${\cal P} (y)$.
247: In particular a useful indicator is, the spatial growth of the asymptotic variance
248: %
249: \be\label{varexp}
250: \sigma ^2(n)=\langle(\ln f_s(n))^2\rangle-\langle\ln f_s(n)\rangle^2  \;.
251: \ee
252: %
253: The results for the cases~$b=5,10$ and various values~$0\le\alpha\le1$ are reported in Fig.~3d 
254: under the scaling assumption 
255: %
256: \be\label{varsc}
257: \sigma^2(n) = \sigma^2(x)/\sqrt{l_\infty};\,\, n/l_{\infty}.
258: \ee
259: %
260: This further confirms that not only the mean asymptotic profile~$f_s(n)$ but also 
261: higher moments change smoothly as a function of the crossover parameter~$\alpha$.
262: 
263: A global characterization of the diffusion and localization properties of a wavepacket is 
264: provided by the evolution of the mean-square displacement 
265: %
266: \be\label{m2}
267: M(t,\alpha)\equiv \left\langle \sum_mm^2|c_m(t)|^2\right\rangle \quad , 
268: \ee
269: %
270: where~$\langle\cdot\rangle $ denotes the average over different realizations of the 
271: disorder. On the basis of the localization properties of the asymptotic wavefunction 
272: (see Eqs.~(\ref{Gogolin},\ref{lalpha})) one expects that, for~$b\gg 1$, $M_{\infty}$~grows as 
273: $M_{\infty}\sim l_{\infty}(\alpha)^2$. The quantity~$M$ reaches its maximum value~$M_\infty$
274: at $t\simeq t_D$. Up to that time the evolution of the packet is diffusive, i.e.
275: $M_\infty \simeq  M(t_D) \simeq D t_D$, where~$D$ is the diffusion constant. From
276: supersymmetry~\cite{E97} it is known that $l_{\infty} \Delta = D$ with~$\Delta$ being the mean
277: level spacing. The latter depends on~$\alpha$ as $\Delta(\alpha) = {\sqrt {1+0.5
278: \alpha^2}} \Delta(0)$. Accordingly the following relations hold:
279: %
280: \be\label{tdD}
281: \frac{t_D(\alpha)}{t_D(0)}= \frac {\beta(\alpha)}{\sqrt{1+0.5\alpha^2}} \,\,;\,\,\,
282: \frac{D(\alpha)}{D(0)}=\beta(\alpha){\sqrt{1+0.5\alpha^2}} 
283: \ee
284: %
285: where $t_D(0)\sim b^{3/2}$ and $D(0)\sim b^{5/2}$~\cite{IKPRT96}. In the inset of Fig.~4 
286: we show the scaling of the diffusion constant~$D(\alpha)$ for $b=30,40$ and various~$\alpha$. 
287: Our numerical results are in perfect agreement with~(\ref{tdD}).
288: Equation~(\ref{tdD}) suggests that the mean-square displacement~$M(t)$ follows the scaling 
289: relation
290: %
291: \be\label{variance}
292: M(t,\alpha) = M_\infty(\alpha) {\tilde M}(t/t_D(\alpha)).
293: \ee
294: %
295: The numerical results obtained for the cases $b=5,10,15$ and various values of the 
296: crossover parameter~$\alpha$ are reported in Fig.~4 according to the above ansatz. The 
297: close coincidence of data reveals the presence of a scaling regime already at 
298: moderately large $b$-values. 
299: 
300: \begin{figure}
301: \begin{center}
302: \epsfig{figure=bfigp4.eps,width=8cm}\hfill
303: \end{center}
304: \noindent
305: {\footnotesize {\bf FIG.~4.}
306: Mean-square displacement~$M(t)$ for $b=5,10,15$ and various values of~$\alpha$,
307: confirming the scaling law~(\ref{variance}). 
308: In the inset we present the diffusion constant~$D(\alpha)$ as a function of the time-reversal 
309: symmetry breaking parameter~$\alpha$ for $b=30$ (squares), $40$ (filled circles). The 
310: solid line is the theoretical expectation~(\ref{tdD}).}
311: \end{figure}
312: 
313: In conclusion, we have studied scaling properties of both diffusion and strong localization 
314: of wavepackets in quasi-1D (i.e., in 1D random media with long-range interactions). Our 
315: numerical data show that the decay rate of the asymptotic wavefunction profile changes 
316: smoothly, in contradiction to the supersymetry results~\cite{KE99,KE00}. We have also found 
317: that the asymptotic shape of the wavepacket is well reproduced by the analytical 
318: expression~(\ref{Gogolin}) derived for the pure symmetry cases (i.e.~$\alpha=0,1$). Moreover, 
319: in agreement with~\cite{KE00} we find, that the logarithm of the asymptotic wavefunction is 
320: normally distributed for every~$\alpha$, and the variance is scaled according to Eq.~(\ref{varsc}). 
321: Another issue addressed in this Letter concerns the wavepacket evolution. In particular, 
322: we found how the diffusion constant~$D(\alpha)$ and the diffusion time~$t_D(\alpha)$ change 
323: as a function of the time reversal symmetry breaking parameter~$\alpha$. As a result we 
324: were able to establish a scaling law~(\ref{variance}) for the variance~$M(t)$ of the packet.
325: 
326: We are grateful to Y. Fyodorov and A. Kolesnikov for useful discussions and suggestions.
327: 
328: \begin{thebibliography}{99}
329: 
330: \bibitem{FM94}  Y.F.~Fyodorov and A.D.~Mirlin, Int. J. Mod. Phys. {\bf 8}, 3795 (1994);
331: F.M.~Izrailev, Chaos, Solitons and Fractals {\bf 5}, 1219 (1994).
332: 
333: \bibitem{B91} J.P.~Bouchaud, J. Phys. I (Paris) {\bf 1}, 985 (1999).
334: 
335: \bibitem{LI95} I.V.~Lerner and Y.~Imry, Europhys. Lett. {\bf 29}, 49 (1995).
336: 
337: \bibitem{PSSD90} J.L.~Pichard, M.~Sanquer, K.~Slevin and P.~Debray,
338: Phys. Rev. Lett. {\bf 65}, 1812 (1990).
339: 
340: \bibitem{SB92} R.~Bl\"umel and U.~Smilansky, Phys. Rev. Lett. {\bf 69}, 217 (1992).
341: 
342: \bibitem{KGB98}  Yu.B.~Khavin, M.E.~Gershenson and A.L.~Bogdanov, 
343: Phys. Rev. Lett. {\bf 81}, 1066 (1998); Phys. Rev. B {\bf 58}, 8009 (1998).
344: 
345: \bibitem{KE99} A.V.~Kolesnikov and K.B.~Efetov, Phys. Rev. Lett. {\bf 83},
346: 3689 (1999). 
347: 
348: \bibitem{KE00} A.V.~Kolesnikov and K.B.~Efetov, cond-mat/0005048 (2000);
349: A.V.~Kolesnikov and K.B.~Efetov, to be published (2000).
350: 
351: \bibitem{SB99} H.~Schomerus and C.W.~Beenakker, Phys. Rev. Lett. {\bf 84},
352: 3927 (2000).
353: 
354: \bibitem{E97} K.B.~Efetov, {\it Supersymmetry in Disorder and Chaos}
355: Cambridge University Press, New York, (1997).
356: 
357: \bibitem{PRT00} A.~Politi, S.~Ruffo and L.~Tessieri, Europ. Phys. J. B, {\bf 14}, 
358: 673 (2000).
359: 
360: \bibitem{IKPRT96}  F.M.~Izrailev, T.~Kottos, A.~Politi, S.~Ruffo and G.P.~Tsironis, 
361: Europh. Lett. {\bf 34}, 441 (1996); F.M.~Izrailev, T.~Kottos, A.~Politi, and G.P.~Tsironis, 
362: Phys. Rev. E {\bf 55}, 4951 (1997).
363: 
364: \bibitem{loc} G.~Casati, L.~Molinari and F.M.~Izrailev, Phys. Rev. Lett. {\bf 64} 16 (1990);
365: Y.F.~Fyodorov and A.D.~Mirlin, Phys. Rev. Lett. {\bf 67}, 2405 (1991); T.~Kottos, A.~Politi, 
366: F.M.~Izrailev, S.~Ruffo, Phys. Rev E., {\bf 53}, R5553 (1996)
367: 
368: \bibitem{Z97} O.V.~Zhirov, chao-dyn/9712001.
369: 
370: \bibitem{G76}  A.A.~Gogolin, Sov. Phys. JETP, {\bf 44}, 1003 (1976).
371:  
372: \end{thebibliography}
373: 
374: %\newpage
375: 
376: 
377: 
378: \end{multicols}
379: \end{document}
380: 
381: 
382: 
383: 
384: 
385: 
386: