1: \documentclass[doublespacing]{elsart}
2: \oddsidemargin 0pt
3: \evensidemargin 0pt
4: \textheight 21.5cm
5: \textwidth 15.25cm
6: \topmargin 0cm
7: \renewcommand{\theequation}{\arabic{section}.\arabic{equation}}
8: \renewcommand{\baselinestretch}{1.5}
9: \usepackage{amssymb}
10: \usepackage{graphics}
11: \usepackage{epsfig}
12: \def\onlinecite{\cite}
13: \def\cal{\mathcal}
14:
15: \pagestyle{plain}
16: \begin{document}
17:
18: \begin{frontmatter}
19: \title{Amplitude scaling behavior of band center
20: states of Frenkel exciton chains with correlated off-diagonal disorder}
21: \author{Ibrahim Avgin}
22: \address{Department of Electrical and Electronics
23: Engineering, Ege University,
24: Bornova 35100, Izmir, Turkey}
25:
26: \author{David L Huber}\footnote{Corresponding author: e-mail address
27: dhuber@mail.src.wisc.edu, Tel. No. (608) 238-8880, Fax. No. (608)
28: 265-2334} \\
29: \address{Department of Physics, University of Wisconsin-Madison,
30: Madison, Wisconsin 53706 USA}
31:
32: %PACS No: 78.20.Bh, 78.40.Pg, {\bf 78.67.$-$n} \\
33: %Keyword: exciton, one dimension, correlated off-diagonal disorder,
34: %amplitude scaling
35:
36: \begin{abstract}
37: We report the amplitude scaling behavior of Frenkel exciton chains with
38: nearest-neighbor correlated off-diagonal random interactions. The band
39: center spectrum and its localization properties are investigated through
40: the integrated density of states and the inverse localization length. The
41: correlated random interactions are produced through a binary sequence
42: similar to the interactions in spin glass chains. We produced sets of data
43: with different interaction strength and "wrong" sign concentrations that
44: collapsed after scaling to the predictions of a theory developed earlier
45: for Dirac fermions with random-varying mass. We found good agreement as
46: the energy approaches the band center for a wide range of concentrations.
47: We have also established the concentration dependence of the lowest order
48: expansion coefficient of the scaling amplitudes for the correlated
49: case. The correlation causes unusual behavior of the spectra,
50: i.e., deviations from the Dyson-type singularity.
51: \end{abstract}
52:
53: \end{frontmatter}
54:
55: \begin{keyword}
56: exciton, one dimension, correlated off-diagonal disorder,
57: amplitude scaling
58: \PACS 78.20.Bh, 78.40.Pg,{\bf 78.67.$-$n}
59: \end{keyword}
60:
61: \section{Introduction}
62: Low-dimensional disordered excitonic systems have generated renewed interest
63: recently. For example, the possibility of using optically active polymers in
64: various photonic applications such as flat panel displays and light emitting
65: diodes has led to increased interest in their electronic and optical
66: properties \cite{orel,jpc,cpa}.
67: We study here numerically the dynamics of Frenkel excitons near the band
68: center in a system with nearest-neighbor off-diagonal random interactions.
69: This study is, we believe, the first to treat correlated disorder in the
70: off-diagonal dipolar interactions in an exciton system which focuses on the
71: behavior near band center $E=0$. However, an analysis of a similar problem
72: in the correlated electronic system has appeared \cite{irn} but the authors
73: considered correlations those were different from ours, and the interactions
74: were not of the dipolar type \cite{av99}. Many efforts have also been spent
75: on the same off-diagonal tight-binding system using supersymmetric methods
76: (SUSY) \cite{fisher} where the interactions are formally similar to our
77: dipolar interactions but the authors eventually worked with continuous
78: variables whereas ours are discrete. Later, correlated disorder, in
79: particular the exponential type, was included for the same tight binding
80: model using the SUSY \cite{jp1,jp2}. Our results can thus be compared with what
81: has been developed for the SUSY methods.
82: This study is
83: important because most realistic random disordered systems
84: have nonlocal correlations.
85:
86: The nature of the excitations at the band center of random off-diagonal
87: exciton models has been discussed for a long time, and it has been found
88: that the spectrum has a singularity of the Dyson type \cite{dyson53} in
89: which spectral properties in the vicinity of the band center are functions
90: of $\ln (E)$ where $E$ is the energy relative to the center of the band.
91: The question of the band center localization has also been argued at
92: length, and it was claimed that the band center mode may be weakly localized
93: because of the strong fluctuations \cite{lic7, delyon83,av98}.
94: Ziman \cite{ziman7,eggarter} analyzing various uncorrelated distributions of
95: the random coupling found the low-energy behavior of the integrated density
96: of states ($IDOS$)
97:
98: \begin{equation} \label{idos}
99: IDOS-{\frac{1 }{2}}=\frac {V^U} {{\vert ln(E^{2}) \vert}^{2}},
100: \end{equation}
101:
102: \noindent and the inverse localization length ($ILL$)
103:
104: \begin{equation} \label{ill}
105: ILL=\frac {V^U} {\vert ln(E^{2}) \vert},
106: \end{equation}
107:
108: \noindent in which $E$ is\ the energy and the logarithmic variance
109: \cite {ziman7,eggarter} $V^{U}={\frac{1}{2}\left[ {\langle (\ln
110: (J^{2}))^{2}\rangle }-{\langle \ln (J^{2})\rangle }^{2}\right] }$ is the
111: amplitude factor, $J$ being the off-diagonal coupling specified below. For
112: definitions of the $ILL$ and the $IDOS$ see Ziman's work \cite{ziman7}
113: particularly his discussion after the equation of the motion. Notice that
114: the averages above are independent of the lattice site $n$ since
115: uncorrelated random variables are involved; otherwise they are site
116: dependent. Concerning the shape of the $IDOS$ and the $ILL$, according to
117: Ziman's analysis \cite{ziman7}, they should depend on disorder only through
118: a multiplicative constant, the variance in the distribution of $ln(J^{2})$.
119: As a result, the spectra near $E = 0$ have the same form
120: for all sets of independent, identically distributed random variables $J$.
121: In this context, such a property is referred to as scaling.
122: When there is scaling, the data can be collapsed onto a
123: single curve by dividing through by a factor that depends on the properties
124: of the disorder but is independent of the energy. Although the above
125: mentioned results were developed for uncorrelated disorder, our analysis
126: shows that qualitatively similar scaling
127: behavior for limited range of concentrations
128: and energy can be obtained for a system with
129: exponential correlations but with a modified amplitude factor $V^{C}$.
130: Moreover, for a certain range of correlation lengths a higher order
131: expansion of the Dyson singularity can reproduce the numerical data better
132: than just a single term.
133:
134: A disordered 1D exciton model with nearest-neighbor interactions has the
135: following equation of motion
136:
137: \begin{equation} \label{tb}
138: J_{n+1}U_{n+1} + J_{n-1}U_{n-1} = EU_{n},
139: \end{equation}
140:
141: \noindent where $U_{n}$ denotes the usual exciton creation operator and $
142: J_{n}$ are the couplings. This is equivalent to the XY model in a strong
143: applied field in the $z$-direction ($<Sz>=S$).
144: In our numerics we calculated
145: iteratively the Lyapunov exponent
146: \cite{ziman7,eggarter}
147: $\gamma (E)=\frac{1}{N}\ln {\frac{U_{N}}{ U_{1}}}$ (see appendix),
148: where $N$ is the number of
149: sites \cite{ziman7,eggarter},
150: whose real part is related to the $ILL$ while
151: the imaginary is related to the $IDOS$. We consider couplings that have
152: correlated or uncorrelated (for comparison) disorder and are of a dipolar
153: type. Since the displacements of
154: the atoms in quasi-one dimensional systems
155: are likely to be small compared to the lattice constants, it is usually
156: enough to consider the first order term in the expansion of $J$ in terms of
157: the displacements \cite{dip}.
158: For the first order expanded coupling we have $
159: J_{n}=1+A\xi _{n}$ with unit
160: lattice constant where $A$ is the strength of
161: the disorder and ${\xi _{n}}$ is a random sign variable that is
162: exponentially correlated $<\xi _{n}\xi _{m}>=e^{-\frac{|m-n|}{l(c)}}$
163: where $l(c)$ is the correlation length as a function of
164: "wrong sign" concentration $c$ (see below).
165: Thus the dipolar coupling $J_n$ has a non
166: fluctuating part and a fluctuating part as in Dirac fermions with
167: random-varying mass \cite{jp1,jp2}.
168:
169: The generation of random binary sequences with correlation was deviced
170: by implementing the convolution method of constructing a random squence.
171: Such squence can be used in various applications as in designing
172: low-dimensional devices with desired properties
173: \cite{jp1,izr99}. A detailed discussion on this topic recently
174: appeared \cite{izr07}. However, a more
175: efficient and easily handled correlated sequence, related to the spin glass
176: chain problem \cite{avgin93, boukahil89}, can be obtained through
177: uncorrelated random numbers as outlined below. The correlated random number
178: at site $n$ follows the relation $\xi _{n}=\xi
179: _{n-1}x_{n}=x_{1}x_{2}x_{3}\ldots x_{n-1}x_{n}$
180: with the distribution (of $x_{i}$ $i$ within $[1,n]$)
181:
182: \begin{equation}
183: P(x_{i})=(1-c)\delta (x_{i}-1)+c\delta (x_{i}+1) , \label{dist}
184: \end{equation}
185:
186: \noindent The $x_{i}$ are uncorrelated between different sites and $c$ is
187: the concentration of "wrong signs" such that $c=0$ and $c=1$ mean no
188: disorder. Clearly $\xi _{n}$ is exponentially correlated $\langle \xi
189: _{n}\xi _{m}\rangle =(1-2c)^{|m-n|}$ where one can define a correlation
190: length \cite{vulp89} $l(c)=-1/ln|1-2c|$. For the numerical computation
191: uncorrelated $x_{i}$ are obtained from the uniformly distributed random
192: number generator sequence. For the uncorrelated results we set $\xi
193: _{n}=x_{n}$ at the particular concentration $c=0.5$ where both correlated
194: and uncorrelated cases produce identical results since $l(0.5)=0$.
195:
196: It should be emphasized that while the model for uncorrelated disorder
197: and the model for correlated disorder involve the parameter $c$,
198: in the latter model $c$ determines both the fraction of ‘
199: wrong sign’ bonds and the correlation length. The two models ‘
200: intersect’ at $c = 0.5$ where there are equal numbers of
201: $\pm$ interactions and the correlation length is $0$.
202: We have carried out extensive numerical studies to verify the
203: equation for the correlation length given above.
204:
205: We note here that unlike the off-diagonal case above,
206: many works have recently appeared on the
207: diagonal tight binding model with correlated disorder dealing with various
208: aspects of the problem \cite{yan05,tit03,deyc03}. It was revealed that even
209: short range correlated disorder at the band edge caused the anomalies in the
210: spectrum \cite{scho05}. Frenkel excitons in off-diagonal and diagonal models
211: have markedly different spectra, i.e., the former has a
212: Dysonian singularity \cite{ziman7,eggarter} at the band center while the
213: latter has a power-law singularity at the band edge. Hence the short-range
214: correlations are expected to influence off-diagonal spectra more than in the
215: diagonal case.
216: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
217: \setcounter{equation}{0}
218: \section{Results}
219: The distribution and localization of the exciton modes are characterized by
220: the Lyapunov exponent. In the case of one-dimensional arrays with
221: nearest-neighbor interactions, the Lyapunov exponent can be determined by
222: making use of mode-counting techniques \cite{dean,hub73}. Recently we
223: studied \cite{avh05} similar excitons with orientationally disordered
224: couplings in conjugated polymers and calculated the full spectrum and
225: optical line shapes. The goal here is to look at the effects of
226: exponentially correlated disorder on the behavior of the $IDOS$ and $ILL$
227: near the center of the band ($E=0$). We have produced three sets of data for
228: each case varying the correlation length i.e., c) and the
229: strength of the disorder, $A,$ with $A=0.25,0.50,0.75 $.
230:
231: \begin{figure}
232: \unitlength=1cm
233: \begin{picture}(15,5)
234: \unitlength=1mm
235: \centerline{
236: \epsfysize=7.5cm
237: \epsfbox{g1.eps}
238: }
239: \end{picture}
240: \caption{The integrated density of states ($IDOS$, left panel) and the
241: inverse localization length ($ILL$, right panel) plots for the correlated
242: distributions of the dipolar type couplings with $A=0.25,0.50, 0.75$ bottom
243: to top. Symbols are data for a chain of $10^{8}$ sites for respective
244: concentrations divided by $f(c)=\frac{c}{1-c}$.
245: The lines, shown as guides to the eye, are plots of the amplitude function
246: $f(0.5)(\sqrt 2 \tanh^{-1}A)^2$ divided by $\ln(E^2)$ for the $ILL$
247: or its square for the $IDOS$.
248: The parameter $c$ is related to the correlation
249: length $l(c)$ by the equation $l(c) = - 1/\ln|1-2c|$.}
250: \label{fig:ids0}
251: \end{figure}
252:
253: Figure~\ref{fig:ids0} displays the results of the $ILL$ and $IDOS$
254: for exponentially correlated distributions of the couplings.
255: For fixed $A$, we divided the data produced for a particular
256: concentration $c$ by a scaling factor $f(c) = c/(1-c)$
257: (see next section and the appendix).
258: The data for selected concentrations are shown in Fig.~\ref{fig:ids0}
259: for the indicated values of $A$.
260: We expect first that for a range of concentrations
261: the data should collapse to a Dysonian singularity for fixed $A$.
262: However, Fig.~\ref{fig:ids0} reveals that as $c$ increases,
263: the $ILL$ and the $IDOS$ take values that differ increasingly from
264: predictions based on Ziman's analysis \cite{ziman7} and a slow approach to
265: the band center is observed. In this limit, the data clearly show
266: qualitative scaling behavior but deviations with varying magnitude exist as $%
267: c\rightarrow 1$ and $c\rightarrow 0$ where the correlation length is
268: infinite. Notice also that the deviations are stronger for the $IDOS$ than
269: for the $ILL$.
270:
271: \noindent Clearly as energy decreases, the scaling gets better; however, the
272: deviations become worse as one gets closer to $c\rightarrow 0,1$. To
273: see whether longer chains and/or lower energies improve the situation, we
274: checked further and found that the longer chain did not improve the
275: agreement appreciably, but as the energy is lowered, the results follow the
276: scaling behavior more closely up to a certain limiting $c$. Similar
277: disagreement between the data and the Dysonian (with a single term)
278: singularity were encountered in the SUSY \cite{jp1} off-diagonal tight
279: binding case for continuous exponential correlations. The analytical
280: investigations \cite{jp2} revealed that the spectra for the exponential
281: correlation can be expanded for the $IDOS$ as
282:
283: \begin{equation} \label{idosc}
284: IDOS-{\frac{1 }{2}}=a_1/|\ln(E^2)|^2 + a_2/|ln(E^2)|^3 + a_3/|ln(E^2)|^4
285: +\dots,
286: \end{equation}
287:
288: \noindent and for the inverse localization length ($ILL$)
289:
290: \begin{equation} \label{illc}
291: ILL=b_1/|ln(E^2)|+b_2/|ln(E^2)|^2+b_3/|ln(E^2)|^3 \dots ,
292: \end{equation}
293:
294: \noindent where $a_{1},a_{2},a_{3},b_{1},b_{2}$ are given as of functions of
295: the correlation length multiplied by the strength \cite{jp1,jp2} of the
296: exponential correlations. The equations\ (\ref{idos})--(\ref{ill}) are only
297: the first order term in the expansion and $a_{1}$ and $b_{1}$ should
298: correspond to correlated amplitude factor $V^C(c)$. Figure~\ref{fig:idsf}
299: clearly shows the improved results obtained using the
300: Eqs.\ (\ref{idosc})--(\ref{illc})
301: for the selected concentrations where the
302: worst deviations followed from the first order result: $%
303: c=0.01,0.05,0.1,0.9,0.95$. Unfortunately, for $c<0.01$ and $c>0.95$, this
304: type of expansion appears to fail; however,
305: other forms of fitting with fractional
306: powers can give better results such as
307: tried in the SUSY case \cite{jp1,jp2}. We will not here
308: discus significance of this fractional fitting since
309: there doesn't seem to be any analytic theory that predicts it.
310:
311: \begin{figure}
312: \label{fig:idsf}
313: \unitlength=1cm
314: \begin{picture}(15,5)
315: \unitlength=1mm
316: \centerline{
317: \epsfysize=7.5cm
318: \epsfbox{g2.eps}
319: }
320: \end{picture}
321: \caption{The integrated density of states ($IDOS$, left panel) and the
322: inverse localization length ($ILL$, right panel) plots for the correlated
323: distributions of the dipolar type couplings with $A=0.50$. Symbols are data
324: for a chain of $10^{8}$ sites for particular concentrations (where
325: significant deviations occur) $c=0.01,0.05,0.1,0.9, 0.95$ from bottom the
326: top and the lines are the scaling function $f(c)(\sqrt 2
327: \tanh^{-1}0.5)^2$ divided by $\ln(E^2)$ for the $ILL$ or
328: its square for the $IDOS$ plus the higher terms given
329: in Eqs.\ (\ref{idosc})--(
330: \ref{illc}) where $f(c)=\frac{c}{1-c}$.
331: The parameter $c$ is related to the correlation length $l(c)$
332: by the equation $l(c) = - 1/\ln|1-2c|$.}
333: \end{figure}
334:
335: %%%%%%%%%%%%%%%%%%%%%
336: \setcounter{equation}{0}
337: \section{Discussion}
338: To shed some light on the scaling behavior for the correlated case, we first
339: compute the binary uncorrelated case. A simple calculation of the
340: logarithmic variance, the amplitude of the spectrum, for the uncorrelated
341: case given in Eqs.\ (\ref{idos})--(\ref{ill}) yields the result
342:
343: \begin{equation}
344: V^U(c)=4c(1-c){\left( \sqrt 2 \tanh ^{-1}A\right) }^{2}. \label{uvar}
345: \end{equation}
346:
347: \noindent The amplitude factor for the uncorrelated disorder clearly
348: displays a scaling behavior in concentration $c$ as well as in $A$. We can
349: identify the scaling functions for the concentration $g(c)=4c(1-c)$ and the
350: strength of the deviation $h(A)=(\sqrt{2}\tanh ^{-1}A)^{2}$. The top inset
351: in Fig.~\ref{fig:scl} displays the uncorrelated data vs concentrations. The
352: three lines are the calculated amplitudes $V^{U}(c)=h(A)g(c)$ presented in
353: Eq.\ (\ref{uvar}) for $A=0.25,0.50,0.75$. As expected, the theory for the
354: uncorrelated case works well
355:
356: The amplitude factors in Fig.~\ref{fig:scl} (large panel) are found from the
357: best fit to the Dysonian singularity. They are obtained matching the
358: smallest value of the energy rather than the value obtained by fitting over
359: a range of energies. In this way, we will be more likely to get a value
360: characterizing the asymptotic region. The calculated amplitude scaling
361: function is also shown in Fig.~\ref{fig:scl} for the correlated case $%
362: V^C(c)=c(1-c)^{-1}{\left( \sqrt{2}\tanh ^{-1}A\right) }^{2}$ for the three
363: specified $A$ values above. In the appendix we will argue why this formula
364: holds. We notice immediately that the concentration dependence has a
365: different form $f(c)=\frac{c}{1-c}$ as compare to $g(c)=4c(1-c)$ \ of the
366: uncorrelated case while the scaling on $A$ is the same for both cases. $h(A)=%
367: {\left( \sqrt{2}\tanh ^{-1}A\right) }^{2}$. This asymptotic fit can further
368: be checked by plotting the $a_{1}$ and $b_{1}$ in the expansion above since
369: they correspond the coefficient of the Dysonian singularity. The bottom
370: inset displays the $a_{1}$ and $b_{1}$ obtained by the best fit as compare
371: to the theoretical line $V^C(c)$ (for $A=0.5$). Both plots clearly support
372: the asymptotic amplitude scaling.
373:
374: \begin{figure}
375: \label{fig:scl}
376: \unitlength=1cm
377: \begin{picture}(15,5)
378: \unitlength=1mm
379: \centerline{
380: \epsfysize=7.5cm
381: \epsfbox{g3.eps}
382: }
383: \end{picture}
384: \caption{The asymptotic amplitude factors for
385: correlated disorder vs concentration $c$
386: with $A=0.25,0.50, 0.75$ bottom to top.
387: In the main plot, the lines show the correlated amplitude
388: $V^C(c)=f(c){\left (\sqrt 2 \tanh^{-1} A \right )}^2$
389: where $f(c)=\frac{c}{1-c}$.
390: The top inset is the same plot for
391: uncorrelated case for the same $A$ values.
392: The lines are a plots of the uncorrelated amplitude $V^U(c)=g(c){\left
393: (\sqrt 2 \tanh^{-1} A \right )}^2$ where $g(c)=4c(1-c)$.
394: The bottom inset shows
395: $f(c){ \left ( \sqrt 2 \tanh^{-1} A \right )}^2$,
396: the expansion coefficient of the $IDOS$ and the
397: $ILL$ $a_1$ and $b_1$ , respectively vs $c$ for $A=0.5$. The parameter
398: $c$ is related to the correlation length $l(c)$
399: by the equation $l(c) = - 1/\ln|1-2c|$.}
400: \end{figure}
401:
402: Although the results obtained by asymptotic fit are clearly reproduced by
403: the theoretical amplitude factor, there are striking differences between
404: correlated and uncorrelated cases. What is particularly interesting and not
405: expected is that the results for correlated disorder is not symmetric about
406: $c=0.5$ like those of the uncorrelated case. Also, from our scaling analysis
407: the $ILL$ coefficient becomes large as $c\rightarrow 1$, but vanishes as $%
408: c\rightarrow 0$. This is surprising since the modes for $c=1$ are extended
409: just like the modes for $c=0$. The divergence of the amplitude factor as $%
410: c\rightarrow 1,$ is not inconsistent with the results for $c=1$ since the
411: energy interval over which scaling holds shrinks to zero in the same limit.
412: This can be seen from Eqs.\ (\ref{idos})--(\ref{ill}) by setting either the $%
413: IDOS$ or the $ILL$ to a constant value and solving for the cut-off energy as
414: as a function of $c$.
415: As shown in the appendix, the amplitude factors perform a
416: correlated random walk in parameter space. For our case, the square of
417: the length of the random walk is proportional to $cN/(1-c)$ not $N$
418: alone as in the uncorrelated case. That is the source of the marked
419: difference between the two cases. As the concentration approaches $1$,
420: divergent behavior occurs. However, we note that we cannot take c
421: arbitrarily close to $1$; beyond a certain point, the random walk sequence
422: can not be constructed \cite{eggarter}.
423:
424: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
425: \setcounter{equation}{0}
426: \section{Summary}
427:
428: We have investigated the dynamics of 1D Frenkel exciton systems with
429: correlated off-diagonal disorder. We have used a negative eigenvalue
430: counting technique \cite{dean} which provides a simple and physically
431: transparent analysis of the $IDOS$ and the $ILL$. We investigated the
432: question of scaling when there was correlated off-diagonal disorder such as
433: might occur, for example, when the off-diagonal interaction depended on the
434: distance between two ions and thus would be affected by small displacements
435: of the ions from their equilibrium positions \cite{dip}. Our numerical data
436: indicate that in the correlated case the scaling behavior found by Ziman
437: \cite{ziman7} is followed only for a limited range of $c$ and very small $E$
438: . We compared our results with the scaling predictions of Ziman \cite{ziman7}
439: and with the similar tight binding correlated electronic case \cite{jp1,jp2}%
440: . We computed the asymptotic amplitude factors
441: as a function of concentration and found
442: that they are equal to the uncorrelated variance at $c=0.5$ multiplied by
443: the factor $\frac{c}{1-c}$ (see appendix). This factor \cite{avgin93,av05}
444: has played a very interesting scaling role in number of unrelated problems.
445: We found that with the new variance for correlated distributions rather
446: unexpected behavior was obtained as compare to that of the uncorrelated
447: case. In particular, the asymptotic amplitude
448: factor is not symmetric about $c=0.5$
449: while the $ILL$ diverges for concentrations approaching $1$ in contrast to
450: what is observed in the uncorrelated case. This amplitude factor can explain
451: only the first order expansion in a Dyson-type singularity as supported by
452: the SUSY model \cite{jp1,jp2}. As shown in the appendix, if the random walk
453: observation of Eggarter \emph{et}$\,$\emph{al. is} implemented, with
454: correlation, the obtained amplitude factor can reproduce the data within a
455: shrinking energy range as $E\rightarrow 0$ as $c$ is increased. Finally, we
456: should emphasize that our work is a numerical study supplemented by the
457: approximate theoretical analysis outlined in the Appendix. As mentioned
458: previously, the surprising result is the discovery that near the center of
459: the exciton band, a model with correlated disorder showed asymptotic
460: behavior that is
461: similar to the behavior of systems without correlations, the only
462: difference being in the amplitude factor appearing in the limiting
463: expressions for the IDOS and ILL. Like those of essentially all numerical
464: studies, our results are approximate. We hope that the availability of the
465: numerical findings together with our analytical results will stimulate
466: rigorous analyses of the model that may shed light on the origin of the
467: similarity.
468:
469: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
470: \setcounter{equation}{0}
471: \section{Acknowledgement}
472: This work is partially sponsored by the Scientific and Technical Research
473: Council of Turkey (TUBITAK) and Ege university research Grant (BAP).
474:
475: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
476: \setcounter{equation}{0}
477: \section{Appendix}
478: In this section we present the amplitude scaling related to the first order
479: expansion coefficients $a_{1}$ and $b_{1}$ in Eqs.\ (\ref{idosc})--(\ref
480: {illc}). Similar scaling but in different cases has appeared before \cite
481: {rus,stinch,westerberg,avgin93,av05}. However the off-diagonal problem here is
482: more involved than those cases. First the Lyapunov exponent at the band
483: center $E=0$ is calculated using Eq.\ (\ref{tb}) that can be rearranged \cite
484: {irn,ziman7,eggarter}
485:
486: \begin{equation}
487: \gamma(0)=\frac {1}{N}\ln(U_N/U_1)=\frac {1}{N} \sum_{n=1}%
488: \ln(U_{2n+1}/U_{2n})= \frac{1}{N}\sum_{n=1}\ln(J_{2n+1}/J_{2n})+\frac {i \pi
489: }{2}. \label{l1}
490: \end{equation}
491:
492: \noindent The bipartite nature (the chiral symmetry) of the Eq.\ (\ref{tb})
493: is responsible for the logarithmic variance as rigorously shown in \cite
494: {inui} and eventually the Dyson- type singularity \cite{eggarter} since $
495: \gamma (0)$ executes a random walk with a step $\Delta _{n}=\ln
496: (J_{2n+1})-\ln (J_{2n})$. For the uncorrelated case the averages yield $
497: <\Delta _{n}>=0$ and $<{\Delta _{n}}^{2}>=\frac{1}{2}[{<(\ln (J^{2}))^{2}>-{
498: <\ln (J^{2})>}^{2}}]$. The Lyapunov exponent \cite{eggarter,inui} takes this
499: form $\gamma (0)=\frac{1}{N}\sqrt{<(\sum_{n=1}\Delta _{n})^{2}>}+\frac{i\pi
500: }{2}$. The imaginary part here $IDOS=\frac{\Im {\gamma (0)}}{\pi }=0.5$ and
501: the real part is related to the $ILL$. To make our point, we can attack the
502: problem with a different angle. The average of the summation $
503: <\sum_{n=1}\Delta _{n}>=0$ for both correlated and uncorrelated cases;
504: whereas, the average of the squared summation $\sum_{n=1}<{\Delta _{n}}
505: ^{2}>+2\sum_{m\not=n}<\Delta _{n}\Delta _{m}>$. It further reduces to $N<{
506: \Delta }^{2}>$ since the second sum is zero for the independent random
507: numbers but non zero for the correlated random numbers. For the correlated
508: disorder the summation is not easy. The complication arises when correlation
509: present since $<\ln (J_{n})>$ is not equal to its ensemble average rather it
510: dependents on $n$. But this complication can be avoided in a following way.
511: Let us calculate\ the Taylor expansion of $\ln (J_{n})=\ln (1+A\xi _{n})$
512: which reads
513: \begin{equation}
514: \ln(1+A\xi_n)=A\xi_n-(A\xi_n)^2/2+(A\xi_n)^3/3- (A\xi_n)^4/4+(A\xi_n)^5/5+
515: \dots. \label{a1}
516: \end{equation}
517:
518: \noindent As ${\xi _{n}}^{2k}=1$ but
519: for odd powers ${\xi _{n}}^{2k+1}=\xi _{n}$. Arranging even and odd powers
520: we get
521:
522: \begin{equation}
523: \ln(1+A\xi_n)=(A+A^3/3+A^5/5+A^7/7+ \dots) \xi_n-(A^2/2+A^4/4+A^6/6+ \dots) .
524: \label{a2}
525: \end{equation}
526:
527: \noindent The first sum (with odd powers of A) is just the expansion of $
528: tanh^{-1}A$ and let the second sum (with even powers) be $D$ which will be
529: eliminated when it is inserted in $\Delta _{n}=tanh^{-1}A(\xi _{2n+1}-\xi
530: _{2n})$. The same expansion can be used to calculate the uncorrelated case
531: as well; the step takes the form $\Delta _{n}=tanh^{-1}A(x_{2n+1}-x_{2n})$
532: (recall that number of steps $N/2$). Using Eq.\ (\ref{dist}), the variance
533: is given by $V^{U}(c)=<\frac{2}{N}\sum_{n}{\Delta _{n}}^{2}>=2<{\Delta }
534: ^{2}>=4c(1-c)(\sqrt{2}tanh^{-1}A)^{2}$ so that the $ILL$ becomes $\Re {
535: \gamma (0)}=\sqrt{V^{U}(c)/N}=\sqrt{2}tanh^{-1}A\sqrt{4c(1-c)/N}$. For the
536: correlated case, we have $V^{C}(c)=\frac{2}{N}<(\sum_{n}\Delta
537: _{n})^{2}>=(tanh^{-1}A)^{2}<(\sum_{n}(\xi _{2n+1}-\xi _{2n}))^{2}>$. Taking
538: the square, we get three terms $<(\sum \xi _{odd/even})^{2}>$ and twice $
539: <\sum \xi _{odd}\sum \xi _{even}>$, hence the summation results in $N\frac{c
540: }{1-c}$ where inverse of this factor is encountered in previous works \cite
541: {av05,avgin93,rus,westerberg}. The correlated random walk results in $V^{C}(c)=%
542: \frac{c}{1-c}(\sqrt{2}tanh^{-1}A)^{2}$ and the $ILL=\Re \gamma (0)=\sqrt{
543: V^{C}(c)/N}=\sqrt{2}tanh^{-1}A\sqrt{\frac{c}{(1-c)N}}$. In contrast, the $
544: IDOS$ value at the band center is not affected by the correlations since the
545: distribution in eigenvalues is symmetric about the mid-point. \ The
546: calculation above holds at the band center; however Eggarter \emph{et}$\,$
547: \emph{al.}, \cite{eggarter} have shown that for a certain range of $E$ close
548: to the band center, the random walk behavior still held and the $ILL$ and $
549: IDOS$ can be represented as in Eqs.\ (\ref{idosc})--(\ref{illc}) where the
550: amplitude factor is the same as calculated at the band center.
551:
552:
553: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
554: \begin{thebibliography}{99}
555: \bibitem{orel} G. Malliaras and R. Friend Phys. Today, May (2005) 53.
556:
557: \bibitem{jpc} S. C. J. Meskers, M. Bender, J. Hubner, Y. V. Romanovskii, M.
558: Oestreich, A. P. H. J. Schenning, E. W. Meijer, and H. Bassler J. Phys.
559: Chem. A \textbf{105} (2001) 10220.
560:
561: \bibitem{av99} I. Avgin and D.L. Huber J. Lumn. \textbf{83-84}, 193 (1999)
562: and J. Lumn. \textbf{96} (2002) 149.
563: \bibitem{cpa} D. B. Balagurov, G. C. La Rocca and V. M. Agranovich Phys.
564: Rev. B \textbf{68} (2003) 045418.
565:
566: \bibitem{irn} H. Cheraghci, S. M. Fazeli, and K. Esfarjani Phys. Rev. B
567: \textbf{72} (2005) 174207.
568:
569: \bibitem{fisher} L. Balents and M. P. A. Fisher Phys. Rev. B \textbf{56}
570: (1997) 1297.
571:
572: \bibitem{jp1} K. Takeda and I. Ichinose J. Phys. Soc. Japan \textbf{70}
573: (2001) 3623.
574:
575: \bibitem{jp2} Ikuo Ichinose and M. Kimura Nucl. Phys. \textbf{B} \textbf{554}
576: 554, 607 (1999), ibid \textbf{554} (1999) 627.
577:
578: \bibitem{dyson53} F. J. Dyson Phys. Rev. \textbf{92} (1953) 1331.
579:
580: \bibitem{av98} I. Avgin and D. L. Huber Phys. Rev. Phys. Rev. B \textbf{60}
581: (1999) 7646.
582:
583:
584: \bibitem{lic7} L. Fleishman and D. C. Licciardello J. Phys. C \textbf{10}
585: (1977) l215.
586:
587: \bibitem{delyon83} F. Delyon, H. Kunz and B. Souillard J. Phys. A \textbf{16}
588: (1983) 25.
589:
590: \bibitem{ziman7} T. A. L. Ziman Phys. Rev. Lett. \textbf{49} (1982) 337.
591:
592: \bibitem{eggarter} T. P. Eggarter and R. Reidenger Phys. Rev. B \textbf{23},
593: (1981) 4791.
594:
595: \bibitem{dip}
596: According to the Frenkel exciton model,
597: the transfer integral usual dipole-dipole
598: interaction has the inverse cube of the distance between the dipoles.
599: Offdiagonal transfer integral disorder is due to the distribution
600: of bond length. Assuming the transition dipoles, the intersite transfer
601: dipole-dipole interaction is given as $J_{n,m}={\frac{1
602: } {{|(a_0n+\delta_n)-(a_0m+\delta_m)|}^{3}}}$ where $\delta_n$ is
603: the positional deviation from their equilibrium
604: position $a_0n$ of transition dipoles. The nearest distance between the atoms
605: is $a_0$ and $n$ is the site index. Since the
606: interaction decreases rapidly over the $n$ to $m$ distance,
607: regarding nearest-neighbor couplings is sufficient as a first approximation.
608: Assuming unit neighbor distance the tranfer integral reduces to
609: $J_{n,n+1}={\frac{1} {{|1-B\xi _{n}|}^{3}}}$
610: where $B$ is the strength and $\xi_n=(\delta_{n+1}-\delta_n)/B$ is the
611: random sign of the deviation from the regular lattice point.
612: Notice that $J_{n,n+1}$ is used in the text as $J_{n}$.
613: The deviations from the equilibrium positions introduces a
614: correlation. If $B$ is small, one can expand the coupling
615: as $J_{n} \simeq 1+3B\xi _{n} $. Since B
616: is a free parameter, one can replace 3B with A without loss of generality.
617: Note that using the approximate form in place of the cubic form does not
618: make an essential change in the spectra apart from a numerical factor since
619: the spectra are related to averages of the logarithm, $<\ln({J_{n}}^{2})>$
620: and our numerical study confirmed this.
621:
622: \bibitem{izr99} F. M. Izrailev and A. A. Krokhin Phys. Rev. Lett. \textbf{82}
623: (1999) 4062.
624:
625: \bibitem{izr07} F. M. Izrailev,A.A.Krokhin, N.M.Makarov and O.V.Usatenko
626: arXiv:cond-mat/0702231 and Sh. Hod and U. Keshet
627: Phys. Rev. E \textbf{70} (2004) 15104.
628:
629: \bibitem{avgin93} I. Avgin and D. L. Huber Phys. Rev. B \textbf{48}
630: (1993) 13625 and Phys. Rev. B \textbf{57} (1998) 8269.
631:
632: \bibitem{boukahil89} A. Boukahil and D. L. Huber Phys. Rev. B \textbf{40}
633: (1989) 4638 and Phys. Rev. B \textbf{50} (1984) 2978.
634:
635: \bibitem{yan05} Yan-Yang Zhang and Shie-Jie Xiong Phys. Rev. B \textbf{72},
636: (2005) 132202.
637:
638: \bibitem{tit03} M. Titov and H. Schomerus Phys. Rev. Lett. \textbf{91}
639: (2003) 1766001 and H. Schomerus and M. Titov Phys. Rev. B, \textbf{67}
640: (2003) 100201.
641:
642: \bibitem{deyc03} L. I. Deyc, M. V. Erementchouk, , A. A. Lyansky, and B. L.
643: Altshuler Phys. Rev. Lett. \textbf{91}, (2003) 96601.
644:
645: \bibitem{scho05} M. Titov and H. Schomerus Phys. Rev. Lett. \textbf{95}%
646: (2005) 126602.
647:
648: \bibitem{vulp89} A. Crisanti, G. Paladin and G. Vulpiani Phys. Rev. A
649: \textbf{39} (1989) 6491.
650:
651: \bibitem{dean} P. Dean Proc. Phys. Soc. Lond. \textbf{73} (1959) 413.
652:
653: \bibitem{hub73} D. L. Huber Phys. Rev. B \textbf{8} (1978) 2124.
654:
655: \bibitem{avh05} I. Avgin and D. L. Huber J. Lumn.\textbf{122-123}, (2007) 389.
656:
657: \bibitem{av05} I. Avgin Phys. Rev. B \textbf{73} (2006) 52201.
658:
659: \bibitem{rus} S. Russ (2002) Phys. Rev. B \textbf{66}, (2002) 12204 and S.
660: Russ, S. Havlin and I. Webman Phil. Mag. B \textbf{77}, (1998) 1449.
661:
662: \bibitem{stinch} R. B. Stinchcombe and I. R. Pimentel, Phys. Rev. B \textbf{
663: 38}, (1988) 4980.
664:
665: \bibitem{westerberg} E. Westerberg, A. Furusaki, M. Sigrist, and P. A. Lee,
666: Phys. Rev. Lett. \textbf{75}, (1995) 4302 and Phys. Rev. B \textbf{55},
667: (1997) 12578.
668:
669: \bibitem{inui} M. Inui, S. A. Trugmani and Elihu Abrahams Phys. Rev. B
670: \textbf{49}, (1994) 3190.
671:
672: \end{thebibliography}
673:
674: \end{document}
675: