1: %\documentclass[pss,fleqn,embeddedheads]{w-art}
2: \documentclass[aps,superscriptaddress,showpacs,floatfix]{revtex4}
3:
4: \usepackage{amsmath}
5: \usepackage{amssymb}
6: \usepackage{bm}
7: \usepackage{graphicx}
8:
9: \bibliographystyle{apsrev}
10:
11: \begin{document}
12:
13: \title{Local level statistics for optical and transport properties of disordered
14: systems at finite temperature
15: }
16:
17: \author{A. Malyshev
18: \footnote{Corresponding author: e-mail: {\sf a.malyshev@valbuena.fis.ucm.es},
19: Phone: +34\,91\,394\,50\,12,
20: Fax: +34\,91\,394\,45\,47}
21: }
22: \affiliation{
23: GISC, Departamento de F\'{\i}sica de Materiales, Universidad Complutense, Madrid, Spain
24: }
25:
26: \begin{abstract}
27: It is argued that the (traditional) global level statistics which determines
28: localization and coherent transport properties of disordered systems at zero
29: temperature (e.g. the Anderson model)becomes inappropriate when it comes
30: to incoherent transport. We define local level statistics which proves to be
31: relevant for finite temperature incoherent transport and optics of
32: one-dimensional systems (e.g. molecular aggregates, conjugated
33: polymers, etc.).
34: \end{abstract}
35:
36: \pacs{ 78.30.Ly, %Disordered solids
37: 36.20.Kd,
38: 72.20.Ee,
39: 71.55.Jv
40: 71.35.Aa
41: }
42:
43:
44:
45: \maketitle % Produces the title.
46:
47: %% If there is not enough space inside the running head
48: %% for all authors including the title you may provide
49: %% the leftmark in one of the following three forms:
50:
51: %% \renewcommand{\leftmark}
52: %% {First Author et al.: A Short Title}
53:
54: %% \renewcommand{\leftmark}
55: %% {First Author and Second Author: A Short Title}
56:
57: \renewcommand{\leftmark}
58: {A. Malyshev: Local level statistics}
59:
60: \section{Introduction}
61:
62: Level statistics has extensively been studied in the
63: connection with localization and transport properties of disordered systems
64: (see e. g. Ref.~\cite{Mirlin} and references therein). In particular,
65: nearest level spacing (NLS) distribution function $P(S)$ has a clear
66: physical meaning and its relationship with localization properties is easily
67: established: in the localized phase, energy levels are uncorrelated and
68: can be infinitesimally close in energy (the corresponding wave
69: functions can be localized far away from each other with a vanishing overlap),
70: which gives rise to the Poisson NLS distribution. On the other hand, in
71: the extended phase wave functions overlap well, which leads to strong level
72: repulsion and results in the Wigner-Dyson NLS statistics (time-reversal
73: symmetry is assumed hereafter). Various quantities associated with $P(S)$
74: were proposed as scaling variables to analyze metal-insulator transition
75: point, critical exponents, mobility edges, and other characteristics
76: relevant for coherent transport (see e. g. Ref.~\cite{Janssen} and
77: references therein). In the case of incoherent transport a quasi-particle can
78: hop between different localized states because of interactions with
79: environment (e.g. with a thermal bath as in the case of the phonon-assisted
80: hopping). The rate (or probability) of hopping between two states is then
81: proportional to the square of the matrix element of the interaction
82: calculated between wavefunctions of the states. Hopping probability between
83: states with vanishing overlap is therefore negligible (localized phase is
84: considered hereafter); scattering rate is relatively large only for the
85: states that overlap well. The latter introduces correlation between pairs
86: of levels that are relevant for hopping.
87: No such correlation is accounted for in the traditional {\it global} level
88: statistics, we therefore define {\it local} level statistics that takes the
89: correlation into account and proves to be adequate for incoherent transport
90: in 1D disordered systems.
91:
92: In the next section, we introduce a sample model of phonon-assisted
93: hopping, section 3 presents a definition of the {\it local} NLS
94: statistics in 1D, numerical results are discussed in section 4, while
95: section 5 concludes the paper.
96:
97: \section{Diffusion model}
98:
99: In this section we present an example model of quasi-particle diffusion
100: in a random uncorrelated potential \{$\varepsilon_n$\} mediated by weak
101: coupling to a phonon system. The quasi-particle spectrum and wavefunctions
102: \{$E_\nu, \Psi_\nu$\} are defined as:
103: \begin{equation}
104: \hat{H}\,\Psi_\nu = E_\nu\,\Psi_\nu,\quad
105: \hat{H}=
106: \sum_n^N \varepsilon_n\,|n\rangle\langle n| -
107: t\sum_{n}^{N}\left(|n\rangle\langle n+1|+|n+1\rangle\langle n|\right)
108: \label{H}
109: \end{equation}
110: On-site energies $\varepsilon_n$ are randomly distributed within the box
111: $[-W/2,W/2]$. All wave functions are localized at $W\neq 0$; incoherent
112: transport can then be modeled by means of the
113: Pauli master equation for the populations $P_\nu$ of the quasi-particle
114: states:
115: \begin{equation}
116: {\dot P}_\nu = R_\nu - \Gamma_\nu P_\nu + \sum_{\mu}
117: (W_{\nu\mu}P_{\mu} - W_{\mu\nu}P_\nu) \ ,\qquad
118: W_{\mu\nu} = S(|\omega_{\mu\nu}|)\,G(\omega_{\mu\nu})
119: \; I_{\nu\mu}\ .
120: \label{Pnu}
121: \end{equation}
122: where $R_\nu$ is the source term,
123: $\Gamma_\nu $ is the total decay rate of state $\nu$ (and describes the
124: drain), $W_{\nu\mu}$ is the scattering rate from the state
125: $|\mu\rangle$ to the state $|\nu\rangle$, which is due to the weak
126: coupling to a phonon system. The rates $W_{\nu\mu}$ are taken to be
127: proportional to the one-phonon spectral density $S(\omega)$ and
128: the probability overlap integral $I_{\nu\mu} = \sum_{n} \Psi_{\mu n }^2
129: \Psi_{\nu n}^2$ (see e.g. Refs.~\cite{Malyshev03,Bednarz03}
130: and \cite{Shimizu98} for details),
131: %\begin{equation}
132: %\label{Wnumu}
133: % W_{\mu\nu} = S(|\omega_{\mu\nu}|)\,G(\omega_{\mu\nu})
134: % \; I_{\nu\mu}\ .
135: %\end{equation}
136: %Here,
137: $\omega_{\mu\nu} = E_\mu-E_\nu$, and $G(\omega)=n(\omega)$ if
138: $\omega>0$ while $G(\omega)=1+n(-\omega)$ if $\omega<0$, with
139: $n(\omega)=[\exp(\omega/T)-1]^{-1}$ being the mean thermal occupation
140: number of a phonon mode with energy $\omega$ (the Boltzmann constant is set to
141: unity while $T$ is the temperature).
142: %$S(\omega)$ is usually approximated by a power-law function.
143: The model proved to be relevant for J-aggregates~\cite{Malyshev03,Bednarz03,Knoester02,Heijs05}.
144:
145: \section{Local level statistics in 1D}
146:
147: \begin{figure}[ht]
148: \includegraphics[clip,width=.53\textwidth]{FirstStates-diffusion.eps}
149: \caption{
150: Wavefunctions and a diffusion trajectory: baselines of each wavefunction
151: represent the energy of the state, wavefunctions are in arbitrary units.
152: Wavefunctions forming the Lifshits tail of the density of states
153: ($E_\nu \leq E_b = -2t$, $E_b$ is the bare band bottom [at $W=0$])
154: are localized at segments of typical size $N^*=N^*(W)$.
155: Dashed line --- a typical quasi-particle diffusion trajectory: because the hopping
156: probability is proportional to the the overlap integral $I_{\mu\nu}$ the
157: particle hops mostly over well-overlapping states. $\Delta E$ is the
158: smallest energy separation between two distant states, which is relevant for
159: global level statistics
160: ($\lim_{N\to\infty}\Delta E=0$), $\epsilon_i$ is the energy spacing for
161: the hop $i$ with the typical spacing $\epsilon^*$ such that
162: $\lim_{N\to\infty}\epsilon^*=const$.
163: }
164: \label{fig:1}
165: \end{figure}
166:
167: A typical realization of wavefunctions of the Hamiltonian (\ref{H})
168: for a chain of size $N$ and a possible diffusion trajectory is
169: shown in Fig.~1. Note, that the overlap integral $I_{\mu\nu}$ that enters
170: the hopping rate $W_{\mu\nu}$ selects hops
171: between states that overlap well (see also the figure caption for comments).
172: The corresponding energy spacing at hop $i$ between such correlated
173: levels is $\epsilon_i$. To account for the correlation we propose the
174: following procedure that is relevant for 1D case: (i) from the full sorted energy set \{$E_\nu$\}
175: select energies within the window of interest
176: (e.g. $(-\infty,E_c]$ where the upper cut-off energy can reasonably be
177: set as $E_c \sim E_b$ or $E_c \sim E_b + T$, $E_b$ being the bare band bottom
178: [at $W=0$]), (ii) sort the selected energy set in the averaged
179: radius-vector for the state $\nu$, $\bar{x}_\nu = \sum_n|\Psi_{\nu n}|^2\,n$
180: (in 1D $\bar x$ is a number and such sorting is well defined);
181: let \{$\bar E_i:\,\bar{x}_{i+1}\geq\bar{x}_{i}$\} be the $\bar x$-sorted set,
182: then (iii) construct a new energy sequence
183: \{${\cal E}_n:\,{\cal E}_n=\sum_i^n s_i,\;
184: s_i=|\bar{E}_{i+1}-\bar{E}_i|$\}, i. e. new energies are
185: cumulative sums of absolute spacings between $\bar x$-sorted energies.
186: The set \{${\cal E}_i$\} takes into account the above-mentioned correlations
187: because sequential states from the set \{${\cal E}_n$\} are always close
188: in $r$-space, so that overlap integral between them is large.
189: In the next section we
190: calculate the {\it local} NLS distribution functions $P_{E_c}(S)$
191: for $\bar x$-sorted energy sequences and discuss the results.
192:
193: \section{Numerical results and discussion}
194:
195: Fig.~2 demonstrates the {\it local} NLS statistics results calculated for $N=4096$ and
196: averaged over 1000 realizations of disorder. Left panel shows the dependence of the mean
197: NLS on the disorder magnitude $W$. In contrast to the global NLS statistics with vanishing
198: mean spacing in the thermodynamic limit, the mean {\it local} NLS is independent of the
199: system size. Calculated for $E_c=E_b$ it determines the characteristic
200: energy scale for the states at the Lifshits
201: tail of the density of states, e.g., the diffusion activation energy at low
202: temperature (the latter was demonstrated for the case of the Frenkel exciton
203: diffusion, see Ref.~\cite{Malyshev03}, where the statistics of the hidden energy
204: structure\cite{Malyshev01} was used; such statistics can be viewed as a
205: limiting (less general) case of the proposed {\it local} NLS statistics).
206: The characteristic energy $\epsilon^*(W)$ of the {\it local} NLS for $E_c=E_b$
207: is in good agreement with the energy spacing in the hidden energy
208: structure~\cite{Malyshev01}.
209:
210: In order to analyze universal correlations between energy levels and
211: compare NLS distribution functions to the standard ones (Poisson and Wigner-Dyson) it is
212: necessary to unfold the raw energy sequence \{${\cal E}_n$\}
213: (see Ref. \cite{Unfolding} for details). Right panel of Fig. 2 shows the distribution functions
214: $P_{E_c}(S)$ calculated from the unfolded energy sequence
215: for $W=0.3t$ and different cut-off energies $E_c$.
216: Poisson and Wigner-Dyson distributions are shown for
217: comparison. When $E_c$ is well below the bare band edge $E_b=-2t$ ($E_c$ is deep in the Lifshits tail)
218: the distribution tends to the Poisson distribution and demonstrates almost no level repulsion.
219: In this case the states are localized far
220: away from each other and there is a vanishing overlap and correlation between
221: them. As the cut-off energy approaches the bare band bottom $E_b$ clear sign of
222: level repulsion appears in the {\it local} NLS distribution function. However, the
223: distribution does not tend to the Wigner-Dyson one upon increase of $E_c$,
224: rather it tends to some limiting distribution. The latter is not surprising
225: because although the $\bar x$-sorted sequence is correlated it is a sequence of localized states and no
226: true Wigner-Dyson statistics can be expected. The reason for this is the following:
227: it turns up that there is a non-negligible amount of neighboring segments of the chain with
228: very similar fluctuations of the random potential (averaged over the typical localization
229: volume of the wavefunction, $N^*$);
230: energies of the states localized at such fluctuations are therefore very close to each other.
231: Such pairs of fluctuations are similar to the system of two identical potential wells or a
232: two-atom molecule; as the result, the "eigenstates" of such pairs of fluctuations remind bonded and anti-bonded
233: states of two-atom molecule and can be very close in energy (subject to the wavefunction overlap).
234: Positions of both states are also very close to each other, so they almost always
235: follow each other in
236: the $\bar x$-sorted sequence making even the {\it local} NLS statistics less repulsive.
237: Quantitative contribution of such pairs of states to the NLS statistics will be analyzed elsewhere.
238:
239: \begin{figure}[ht]
240: \includegraphics[clip,width=.49\textwidth]{EpsW.eps}
241: \includegraphics[clip,width=.49\textwidth]{LocalNLS.eps}
242: \caption{
243: Left panel --- the dependence of the characteristic
244: energy scale $\epsilon^*$ on the disorder magnitude $W$ for $E_c=E_b$.
245: Right panel --- {\it local} NLS distribution function $P_{E_c}(S)$ for the
246: unfolded $\bar x$-sorted sequence compared to the Poisson distribution
247: $P_P(S)$ (completely uncorrelated levels) and the Wigner-Dyson one $P_{WD}(S)$
248: (correlated levels).
249: }
250: \label{fig:2}
251: \end{figure}
252:
253: The dependence of the {\it local} NLS statistics on the cut-off energy
254: allows for definition of the diffusion mobility edge in a similar way
255: the conventional mobility edge that separates extended and localized
256: phases is defined. Although no scaling variables or rigorous
257: argumentation is available in the present case (to the best of the author's knowledge),
258: by analogy, we formally define the diffusion mobility edge
259: as the energy $E_\mu$ such that:
260: \begin{equation}
261: P_{E_{\mu}}(S_0)) - \frac{P_P(S_0)+P_{WD}(S_0)}{2} = 0\ ,
262: \end{equation}
263: where $P_P(S)$ and $P_{WD}(S)$ are Poisson and Wigner-Dyson distributions
264: respectively, and the spacing $S_0\sim 1$. In practice, we
265: averaged $E_\mu$ over various $S_0:0.1\leq S_0\leq 3$. For all values of
266: disorder magnitude the diffusion mobility edge coincided with the
267: bare band bottom $E_b=-2t$. In order to interpret the later result it is
268: useful to compare the number of states below the mobility edge to the number
269: of the Lifshits tail states $N/N^*$ (see Fig.~1),
270: it turns up that for all values of $W$
271: \begin{equation}
272: \int_{-\infty}^{E_\mu}\rho(E)\, dE \approx \frac{N}{N^*}\ , \qquad \rho(E)
273: =\left\langle \sum_\nu^N \delta(E-E_\nu)\right\rangle\ .
274: \end{equation}
275: The diffusion mobility edge has therefore a clear physical meaning: it separates
276: the spectral region with states that do not overlap with neighbors
277: (hopping rates between such neighbors are vanishing)
278: from the energy region where wavefunctions begin to overlap well: higher (band)
279: states cover one or more tail states and overlap well with them and between each
280: other as well (see Fig.~1). In the latter case hopping rates and
281: diffusion build up drastically.
282: The importance of two-step hops via higher states was discussed in detail in
283: Ref.~\cite{Malyshev03} in connection with the Frenkel exciton diffusion.
284:
285: \section{Conclusions}
286:
287: We have defined {\it local} NLS statistics for 1D disordered systems. The
288: {\it local} statistics grasps correlation between levels that are
289: involved in hopping transport and is generally
290: more repulsive than the global one. The {\it
291: local} NLS distribution in the localized phase changes from the Poisson one to
292: the distribution that
293: demonstrates clear level repulsion, depending on the spectral region and the energy window.
294: Monitoring the change we determine the diffusion mobility edge that has
295: a clear physical meaning: it separates the states with vanishing probability
296: of hopping between neighbors (slow incoherent transport) from the states with large inter-state
297: hopping rate (fast transport regime). Using similar argumentation {\it local} level statistics can
298: be defined for higher-dimensional systems for which the common mobility edge can be defined.
299: The mobility edge determined by the {\it local} statistics would however be different from the
300: common one because the former is always the diffusion mobility edge.
301: Analysis of higher-dimensional cases is beyond the scope of
302: this paper and will be discussed elsewhere.
303:
304: \vspace{0.5in}
305:
306: %\begin{acknowledgement}
307: AM is grateful to J.\ Knoester and V.\ Malyshev
308: for fruitful discussions. Support by INTAS through YSF 03-55-1545 and
309: Spanish Ministry of Education and Science through Ram\'{o}n y Cajal Program is
310: acknowledged. AM is on leave from Ioffe Institute, St. Petersburg, Russia.
311: %\end{acknowledgement}
312:
313: \begin{thebibliography}{10}
314:
315: \bibitem{Mirlin} A. D. Mirlin, Phys. Rep. {\bf 326}, 259 (2000)
316:
317: \bibitem{Janssen} Martin Janssen, Phys. Rep. {\bf 295}, 1 (1998)
318:
319: \bibitem{Malyshev03} A. Malyshev, V. Malyshev, and F. Dom\'{\i}nguez-Adame, Chem. Phys. Lett. {\bf 371}, 417 (2003)
320:
321: \bibitem{Bednarz03} M. Bednarz, V. A. Malyshev, and J. Knoester,
322: Phys. Rev. Lett. {\bf 91}, 217401 (2003); J. Chem. Phys.
323: {\bf 120}, 3827 (2004).
324:
325: \bibitem{Shimizu98} M. Shimizu et al, Phys. Rev. B {\bf 58}, 5032 (1998); M. Shimizu,
326: S. Suto, and T. Goto, J. Chem. Phys. {\bf 114}, 2775 (2001).
327:
328: \bibitem{Knoester02} J. Knoester, in {\it Organic Nanostructures:
329: Science and Application} (IOS Press, Amsterdam, 2002),
330: pp. 149-186.
331:
332: \bibitem{Heijs05} D. J. Heijs, V. A. Malyshev, and J. Knoester,
333: Phys. Rev. Lett. {\bf 95}, 117402 (2005);
334: J. Chem. Phys. {\bf 123}, 144507 (2005).
335:
336: \bibitem{Malyshev01} V. Malyshev and P. Moreno, Phys. Rev. B {\bf 51}, 14587 (1995);
337: A. Malyshev and V. Malyshev, Phys. Rev. B {\bf 63}, 195111 (2001).
338:
339: \bibitem{Unfolding} M. L. Mehta, {\it Random Matrices}, 2nd ed. (Academic Press 1991)
340:
341: \end{thebibliography}
342:
343: \end{document}
344:
345: \endinput
346: