1: %----------------------- LaTex article-----------------------------------
2: \documentstyle[aps,prl,epsfig]{revtex}
3: %\documentstyle[aps,prl,preprint,epsfig]{revtex}
4:
5: \def\mps{M_{\rm PS}(t)}
6: \def\mint{M_{\rm int}}
7: \def\mintt{M_{\rm int}(t)}
8: \def\mext{M_{\rm ext}}
9: \def\mextt{M_{\rm ext}(t)}
10: \def\mmax{M_{\rm int}^{\rm max}}
11: \def\ton{t_{\rm on}}
12: \def\toff{t_{\rm of\/f}}
13: \def\locl{l_{\infty}}
14:
15: \begin{document}
16: \draft
17:
18: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname @twocolumnfalse\endcsname
19:
20: \title{Superballistic spreading of wave packets}
21: \author{L. Hufnagel, R. Ketzmerick, T. Kottos, and T. Geisel}
22: \address{ Max-Planck-Institut f\"ur Str\"omungsforschung
23: und Institut f\"ur Nichtlineare Dynamik der Universit\"at G\"ottingen,\\
24: Bunsenstra{\ss}e~10, 37073 G\"ottingen, Germany}
25:
26: \date{\today}
27:
28: \maketitle
29:
30: \begin{abstract}
31: %
32: We demonstrate for various systems that the variance of a wave packet
33: $M(t)\propto t^\nu$, can show a {\it superballistic} increase with $2<\nu\le3$,
34: for parametrically large time intervals. A model is constructed which explains
35: this phenomenon and its predictions are verified numerically for various
36: disordered and quasi-periodic systems.
37: %
38: \end{abstract}
39:
40: \pacs{PACS numbers: 03.65.-w, 05.60.Gg, 72.20.Dp}
41:
42: ]
43:
44: The time evolution of wave packets in one dimensional (1D) and quasi-1D
45: lattices is described by the time-dependent Schr\"odinger equation
46: \begin{equation}
47: \label{schr}
48: i\,{\frac{dc_n(t)}{{dt}}}=\,\sum_{m=n-b}^{n+b}H_{nm}c_m\quad ,
49: \end{equation}
50: where $c_n(t)$ is the probability amplitude for an electron to be at site $n$,
51: $b$ is the number of channels, and $H_{nm}$ is a tight-binding Hamiltonian.
52: When translational symmetry is present, the eigenstates of the system are
53: plane waves and the variance of wave packets increases quadratically in time
54: ("ballistic spreading"). On the other hand, since the pioneering work of
55: Anderson \cite{A58}, it is known that disorder usually tends to suppress
56: propagation and leads to localization. In the one-dimensional case, even a
57: small amount of disorder leads to localization of all eigenstates\cite{A58,book},
58: and therefore asymptotically the spreading of a wave packet remains bounded.
59: In higher dimensions a localization to delocalization transition can occur
60: which leads to diffusive or subdiffusive spreading of a wave packet\cite{book,HS94}.
61: In addition, there are quasi-periodic systems which even in 1D show fractal
62: energy spectra and eigenfunctions leading to a power law spreading \cite{KKKG97}
63: that is reminiscent of anomalous diffusion in classical systems.
64: %
65:
66: %
67: A global characterization of the dynamical evolution of a wave packet is provided
68: by its variance
69: %
70: \begin{equation}\label{defvar}
71: M(t)\equiv\sum_nn^2|c_n(t)|^2 \propto t^{\nu} \quad .
72: \end{equation}
73: %
74: Its time dependence gives a quantitative description of the dynamics:
75: $\nu=0$ corresponds to localization, $\nu=1$ to diffusion, $\nu=2$ to
76: ballistic motion, and $\nu\in(0,2)$ to anomalous diffusion.
77: It was shown that a ballistic upper bound, $M(t)\le A\cdot t^2$, exists for all
78: times with a system specific constant $A$\cite{statement}.
79: Although this statement gives no restriction on $\nu$ for
80: finite time intervals, to our knowledge in all studies up to now $\nu\le2$
81: was found for any time.
82: %
83:
84: %
85: In this paper we show that there can be superballistic spreading with exponents
86: $\nu\in(2,3]$
87: for parametrically large time intervals.
88: Examples with $\nu=3$ can be seen in Fig.~\ref{fig1} where the variances of
89: wave packets, while staying below the ballistic upper bound, show a cubic growth.
90: Each system consists of a perfect lattice with a disordered region of finite length.
91: Moreover, extending the length of the disordered region {\it increases} the time interval
92: of the cubic growth.
93: %
94: %
95: \begin{figure}
96: \begin{center}
97: \epsfxsize=8.5cm
98: \leavevmode
99: \epsffile{fig1.eps}
100: \caption[model]{
101: Variance $M(t)$ of a wave packet in a 1D disordered model of size $L=50$, 70, 80, 100
102: (from left to right) with a perfect lattice attached to both ends.
103: The dashed lines indicate the time range over which the cubic growth appears
104: before the asymptotic ballistic spreading sets in.
105: The inset shows the cubic growth for a Band Random Matrix model ($b=10$, $L=100$, $T=0.001$).
106: }
107: \label{fig1}
108: \end{center}
109: \end{figure}
110: %
111:
112: %
113: We will show that this unexpected behavior of the variance is associated with the rate
114: $\Gamma$ for the emission of an electron from the disordered region into the perfect lattice.
115: This parameter determines the time scales
116: %
117: \begin{equation}
118: \label{tonoff}
119: \ton\sim \left({1\over \Gamma}\right)^{1/\nu}\quad{\rm and}\quad\toff\sim{1\over\Gamma}\quad,
120: \end{equation}
121: %
122: when the superballistic growth starts and ends, respectively. The different
123: $\Gamma-$dependence of $\ton$ and $\toff$ ensures that this time interval becomes
124: arbitrarily large as $\Gamma$ decreases. We would like to mention that such
125: intermediate superballistic regimes are well known in other contexts like in
126: hydrodynamic turbulence (Richardson law) or in plasma physics, but are unrelated.
127: %
128:
129: %
130: In order to understand the appearance of the cubic growth of the variance in
131: Fig.~\ref{fig1} we consider a simple probabilistic model:
132: The disordered part is replaced by a point source and anything emitted from it
133: moves with a constant velocity $v$ modelling the dynamics of a perfect lattice.
134: Initially, all probability is confined to the point source and decays with a constant
135: rate $\Gamma$, such that the probability at the point source is given by
136: %
137: \begin{equation}\label{eqstay}
138: P(t) = \exp{(-\Gamma t)}.
139: \end{equation}
140: %
141: The variance $\mps$ of the point-source model is then given by
142: %
143: \begin{equation}\label{eqmps}
144: \mps = \int_0^\infty\!\!\!\!\!dx x^2\int_0^t\!\!\!dt^{'}
145: (-\dot{P}(t^{'}))\:\delta(x-v(t-t^{'}))\quad,
146: \end{equation}
147: %
148: where $-\dot{P}(t)$ is the flux emitted from the point source.
149: Substituting Eq.~(\ref{eqstay}) for $P(t)$ we get
150: %
151: \begin{equation}\label{eqmpspoft}
152: \mps = v^2\Gamma\int_0^t dt^{'} e^{-\Gamma t^{'}} (t-t^{'})^2\quad,
153: \end{equation}
154: %
155: which yields after integration
156: %
157: \begin{equation}
158: \label{var3}
159: \mps= v^2\left( t^2- {2\over \Gamma} t + {2\over \Gamma^2}
160: -{2\over \Gamma^2} e^{-\Gamma t}\right)\quad.
161: \end{equation}
162: %
163: As expected, the variance grows quadratically, $\mps\sim v^2 t^2$,
164: for asymptotically large times.
165: Expanding the exponential term in Eq.~(\ref{var3}) one finds
166: %
167: \begin{equation}
168: \label{varfin}
169: \mps = v^2\Gamma\left({1 \over 3} t^3-{\Gamma\over 12} t^4+...\right)\quad.
170: \end{equation}
171: %
172: Under the condition $t<1/\Gamma$ the cubic term dominates all higher orders.
173: Thus we find a cubic increase of the variance in the point-source model
174: starting from the time $\ton=0$ up to the time $\toff\approx 1/\Gamma$.
175: At the same time scale the crossover to the asymptotic ballistic spreading
176: starts, as can be seen from Eq.~(\ref{var3}).
177: %
178:
179: %
180: An intuitive understanding is based on the fact, that the linear decrease of $P(t)=1-\Gamma t$
181: for small times is responsible for the cubic growth of the variance [Eq.~(\ref{eqmps})].
182: During this time the norm of the wave packet outside the point source increases linearly. This
183: linear increase of the norm combined with the usual quadratic increase of the variance
184: due to the ballistic spreading yields the cubic growth of the variance.
185: %
186:
187: %
188: A more realistic model should take into account the length $L$ of the disordered region
189: and the time $\tau$ at which the norm of the wave packet outside the disordered region
190: starts to increase linearly.
191: The total variance is $M(t) = \mintt + \mextt$, where $\mintt$ is the contribution
192: of the internal region and $\mextt$ originates from the perfect lattice, which is given by
193: %
194: \begin{equation}\label{eqmextps}
195: \mextt=\int_L^\infty\!\!\!\!\!dx x^2\int_\tau^t\!\!\!\!dt^{'}
196: (-\dot{P}(t^{'}))\:\delta(x\!-\!L-\!v(t-t^{'}\!-\!\tau)).
197: \end{equation}
198: %
199: This reduces to Eq.~(\ref{eqmps}) of the point-source model
200: for times $t$ fulfilling the following three conditions:
201: (i) $t\gg\tau$, (ii) $vt\gg L$, and (iii) $\mextt\gg\mintt$.
202: These conditions set the time scale $\ton$ at which the cubic law may start, in contrast
203: to $\ton=0$ for the point-source model.
204: Conditions (i) and (ii) give the time scales $\tau$ and $L/v$, respectively.
205: The internal variance $\mintt$ is bounded by $\mmax\le L^2$.
206: Thus condition (iii) together with $\mextt\approx v^2\Gamma/3t^3$
207: valid for $t<\toff$ leads to the time scale $(\mmax/(v^2\Gamma))^{1/3}$.
208: The maximum of these three time scales defines the onset of the cubic law
209: %
210: \begin{equation}\label{eqton}
211: \ton= {\rm max}\{\tau,\frac{L}{v},\left(\frac{\mmax}{v^2\Gamma}\right)^{1/3}\}\quad.
212: \end{equation}
213: %
214: In order to see the cubic law over a large time interval $\toff\approx 1/\Gamma$
215: has to be large implying a small $\Gamma$.
216: For sufficiently small $\Gamma$ the third time scale in Eq.~(\ref{eqton})
217: will dominate leading to the scaling presented in Eq.~(\ref{tonoff}) for $\nu=3$.
218: The ratio $\toff/\ton$ scales as $\Gamma^{-2/3}$ and therefore can be made
219: arbitrarily large by decreasing $\Gamma$.
220: %
221: %
222: \begin{figure}
223: \begin{center}
224: \hspace*{-0.5cm}\epsfig{figure=fig2.eps,height=9cm,width=8cm,angle=270}
225: \caption[model]{
226: (a) Norm of the wave packet inside the disordered region of the 1D model vs. t
227: for $L=80$, 90, 100 showing linear decays.
228: (b) Decay rate $\Gamma$ vs. $L$ following an exponential law (dashed line).
229: }
230: \label{fig2}
231: \end{center}
232: \end{figure}
233: %
234:
235: %
236: One can extend the above analysis to other dynamical exponents $\nu<3$
237: by embedding the internal region in a lattice showing anomalous diffusion.
238: These lattices are characterized by a variance scaling as $t^{\mu}$, $\mu\in(0,2)$.
239: This yields the general expression $\nu=\mu+1$ for Eqs.~(\ref{defvar},\ref{tonoff}).
240: Again, the intuitive understanding is that the linear increase of the norm of the
241: wave packet outside the internal region increases the exponent by 1.
242: %
243: %
244: \begin{figure}
245: \begin{center}
246: \epsfxsize=8.5cm
247: \leavevmode
248: \epsffile{fig3.eps}
249: \caption[model]{
250: Times $\ton$ (circles) and $\toff$ (squares) vs. $\Gamma$
251: for the 1D model (filled symbols) and for the Band Random Matrix
252: model (open symbols, data for $\ton$ and $\toff$ are shifted to account
253: for different prefactors of the power laws for the two models).
254: }
255: \label{fig3}
256: \end{center}
257: \end{figure}
258: %
259:
260: %
261: In the remainder of this paper we will give numerical evidence supporting the
262: above analysis.
263: We use perfect or quasi-periodic lattices with a finite 1D or quasi-1D
264: disordered region, although there are many other possible settings.
265: Our first example consists of a 1D disordered
266: region of size $L$ attached with semi-infinite perfect lattices at both ends.
267: In Eq.~(\ref{schr}) this corresponds to a tridiagonal Hamiltonian $(b=1)$,
268: with $H_{n,n\pm 1}=1$ and $H_{nn}=0$ except for a region of size $L$ where
269: $H_{nn}$ is a random number. We fix the disorder strength and we use sample
270: sizes $L=50,\dots,100$ such that $L>\locl$, where $\locl$ is the localization
271: length of the corresponding infinite disordered system.
272: In all cases the initial $\delta-$like wave packet is launched in the middle of the
273: disordered region.
274: Fig.~\ref{fig1} shows the variance averaged over 10 disorder realizations.
275: It should be noted, that without averaging we get the same qualitative behavior.
276: For small times all wave packets spread ballistically until the variance starts to
277: saturate when $\mintt\approx\locl^2$.
278: Then the cubic increase of the variance can be observed before finally the asymptotic
279: ballistic spreading sets in.
280: There is a ballistic upper bound for all times in agreement with Ref.~\cite{statement}.
281: The range over which the cubic law holds increases with the size of the disordered sample.
282: This is in agreement with the predictions of the point-source model:
283: It is applicable since the norm of the wave packet inside the disordered region initially decays
284: linearly (Fig.~\ref{fig2}a).
285: From this linear decay $P(t)=1-\Gamma t$ we determine the rate $\Gamma$ which decreases
286: exponentially with the sample size $L$ (Fig.~\ref{fig2}b), as expected for the localized regime\cite{BGZ98}.
287: Thus extending the disordered region decreases $\Gamma$ and together with Eq.~(\ref{tonoff}) explains
288: the increase of the superballistic time interval.
289: The time scales $\ton$ and $\toff $ are determined from the times when $M(t)$ deviates
290: from the fitted cubic increase by more than 10\%.
291: The scaling of the time scales $\ton$ and $\toff$ with $\Gamma$ are shown in Fig.~\ref{fig3}
292: confirming Eq.~(\ref{tonoff}).
293: %
294:
295: %
296: We now show that our considerations also apply to a Band Random Matrix model that
297: describes quantum wires\cite{FM94}.
298: The Hamiltonian matrix $H_{nm}$ is real and the entries are different from
299: zero in a stripe of width $b$ around the diagonal ($|n-m|\le b$), only.
300: The parameter~$b$ defines the hopping range between neighbouring sites, or,
301: in the quasi-1D interpretation, the number of transverse channels along a thin
302: wire\cite{FM94}.
303: The non-zero matrix elements are independent Gaussian random numbers with variance
304: 1 within a region of size $L$ and are equal to 1 outside this region.
305: The matrix elements $H_{nm}$ that couple the sample to the perfect lattice
306: are random numbers with variance $T$ such that $\Gamma\sim T$\cite{weiden93}.
307: In order to study an example where the internal variance $\mintt$ is bounded by the
308: sample size rather than the localization length, we choose a sample size $L\le\locl$.
309: Thus the initial wave packet spreads diffusively over the disordered region before it
310: leaks out to the leads.
311: We find a cubic increase of the variance (Fig.~\ref{fig1}, inset) and by varying the
312: coupling strength $T$ we confirm Eq.~(\ref{tonoff}) for $\ton$ and $\toff$ (Fig.~\ref{fig3}).
313: %
314:
315: %
316: Finally, we use a Fibonacci chain model\cite{firstfibo} outside the disordered region,
317: which allows to vary the exponent $\mu$ of its anomalous diffusion by changing the
318: potential strength $V$\cite{GKP95}.
319: Fig.~\ref{fig4} shows the superballistic increase of the external variance in nice agreement
320: with the expected exponent $\nu=\mu+1$.
321: %
322: \begin{figure}
323: \begin{center}
324: \epsfxsize=8.5cm
325: \leavevmode
326: \epsffile{fig4.eps}
327: \caption[model]{
328: External variance $\mextt$ for a 1D disordered sample ($L=50$) attached to Fibonacci
329: chain models with $V=0.1$, 0.4, and 0.7 (from top to bottom), showing the expected
330: power laws (dashed lines).
331: }
332: \label{fig4}
333: \end{center}
334: \end{figure}
335: %
336: We thank R.~Fleischmann, I.~Guarneri, and A.~Politi for helpful discussions.
337: %--------------------------------------------------------------------------
338: \begin{thebibliography}{99}
339:
340: %
341: \bibitem{A58} P.~W.~Anderson, Phys. Rev. {\bf 109}, 1492 (1958).
342: %
343: \bibitem{book} E.~N.~ Economou, {\it Green's Functions in Quantum
344: Physics}, Springer Series in Solid State Physics, Vol. 7
345: Springer-Verlag, Berlin, (1979).
346: B. Kramer and A. Mackinnon, Rep. Prog. Phys. {\bf 56},1469 (1993).
347: %
348: \bibitem{HS94} B. Huckestein and L. Schweitzer, Phys. Rev. Lett. {\bf 72},
349: 713 (1994);
350: %
351: \bibitem{KKKG97} I.~Guarneri, Europhys. Lett. {\bf 10}, 95 (1989);
352: ibid {\bf 21}, 729 (1993);
353: R. Ketzmerick, K. Kruse, S. Kraut, and T. Geisel, Phys. Rev. Lett. {\bf 79},
354: 1959 (1997).
355: %
356: \bibitem{statement} I.~Guarneri, J. Math. Phys., {\bf 37}, 5195 (1996);
357: I.~Guarneri and H.~Schulz-Baldes, Rev. Math. Phys. {\bf 11}, 1249, (1999).
358: %
359: \bibitem{BGZ98} A. Buchleitner, I. Guarneri, and J. Zakrzewski, Europhys. Lett.
360: {\bf 44}, 162 (1998);
361: O. A. Starykh, P. R. J. Jacquod, E.~E.~Narimanov, and A. D. Stone, cond-mat/0001017 (2000).
362: %
363: \bibitem{FM94} Y. F. Fyodorov and A. D. Mirlin, Int.J.Mod.Phys., {\bf 8} 3795 (1994);
364: T. Kottos, A.~Politi, and G.~P.~Tsironis, Phys.~Rev.~E, {\bf 55}, 4951 (1997);
365: M. Weiss, T. Kottos, and T. Geisel, Phys. Rev. B, in press; [cond-mat/0005339].
366: %
367: \bibitem{weiden93} H.~A.~Weidenm\"uller, Physica A, {\bf 167}, 28, (1990).
368: %
369: \bibitem{firstfibo} M.~Kohmoto, L.~P.~Kadanoff, and C.~Tang, Phys. Rev. Lett. {\bf 50}, 1870 (1983);
370: S.~Ostlund, R.~Pandit, D.~Rand, H.~J.~Schellnhuber, and E.~D.~Siggia, Phys. Rev. Lett. {\bf 50}, 1873 (1983).
371: %
372: \bibitem{GKP95}T. Geisel, R. Ketzmerick, and G. Petschel, in {\it Quantum Chaos;
373: Between Order and Disorder}, G. Casati and B. Chirikov, eds. (Cambridge University Press, Cambridge, 1995),
374: page 634, (1995).
375: %
376: \end{thebibliography}
377:
378: \end{document}
379: