1: \documentclass[prb,twocolumn,showpacs,showkeys]{revtex4}
2: %\documentclass[preprint,showpacs,showkeys]{revtex4}
3: \usepackage{graphicx}% Include figure files
4: \usepackage{graphics}
5: \usepackage{dcolumn}% Align table columns on decimal point
6: \usepackage{bm}% bold math
7: \newcommand{\beq}{\begin{equation}}
8: \newcommand{\eeq}{\end{equation}}
9: \newcommand{\beqnar}{\begin{eqnarray}}
10: \newcommand{\eeqnar}{\end{eqnarray}}
11: \newcommand{\bfig}{\begin{figure}[!hbp]}
12: \newcommand{\efig}{\end{figure}}
13: \begin{document}
14: %\preprint{APS/123-QED}
15:
16: \title{Scaling properties of one-dimensional off-diagonal disorder}% Force line breaks with \\
17:
18: \author{Hosein Cheraghchi$^{1,2}$} %, Keivan Esfarjani$^{1}$}
19:
20: \affiliation{$^{1}$Department of Physics, Sharif University of
21: Technology,P.O.Box 11365-9161, Tehran, Iran \\ $^{2}$Department of
22: Physics, Damghan University of Basic Sciences, Damghan, Iran}
23: \email{cheraghchi@mehr.sharif.edu}
24: \date{\today}
25:
26: \begin{abstract}
27: Validity of the single parameter scaling (SPS) in one dimensional
28: Anderson model with purely off-diagonal disorder is being studied.
29: It is shown that the localized region with standard symmetry is
30: divided into two regimes: SPS and non-SPS. Scaling relations of
31: the Lyapunov Exponent are proposed for these two regimes. In the
32: non-SPS regime, in additional to the localization length, there
33: exists a new length scale which is related to the integrated
34: density of states. A physical interpretation of the new length is
35: the cross-over length which separates regions with chiral symmetry
36: from those that have standard symmetry.
37:
38: \end{abstract}
39:
40: \pacs{72.15.Rn, 71.23.An, 71.30.+h}
41: \keywords{localization, off-diagonal, single parameter scaling}
42:
43: \maketitle
44: \section{Introduction}
45: It is well-known for about forty years that all electron states in
46: standard one dimensional (1D) disordered models are localized for
47: any strength of disorder, and there is no localization transition
48: in 1D systems [\onlinecite{anderson}]. However, in the case of
49: off-diagonal disorder, there is an anomalous localized state at
50: the band center [\onlinecite{soukoulis,cohen,inui,ziman}]. It has
51: been proposed that the Lyapunov Exponent (L.E.) is the appropriate
52: scaling variable to describe fluctuations of the conductivity.
53:
54: \beq\gamma(N) = \frac{1}{2N}\ln(1+\frac{1}{g}) = -
55: \frac{1}{2N}\ln(T) \label{eq:gama} \eeq
56: where $g (= T / R)$ and $T$ are conductance and transmission coefficients through the system
57: with length $N$.
58:
59: A reason for the revival of the interest in 1D disordered model
60: is due to the revision of the well-known Single Parameter Scaling
61: (SPS) hypothesis. According to this hypothesis
62: [\onlinecite{abrahams}], there exists a single parameter,
63: conductance $g$, which determines scaling properties of $g(N)$.
64: Soon after the report of SPS, it became clear that one should
65: consider scaling of the full probability distribution of
66: conductivity.
67:
68: In order to take fluctuations of conductance into account, one
69: should consider a parameter $\gamma$ (L.E.) defined in
70: Eq.(\ref{eq:gama}) instead of conductance $g$ itself. In the
71: thermodynamic limit, $\gamma_{0} = \gamma({N\rightarrow\infty})$
72: has a non-random value which is the inverse of the localization
73: length $\lambda$. This parameter has normal distribution for $ N
74: \gg \lambda$, and its dispersion $\sigma_{\gamma}$ obeys the law
75: of large numbers and is not an independent variable.
76:
77: \beq \frac{1}{\tau} =
78: \frac{\gamma_{0}}{N \sigma^{2}_{\gamma}}=1
79: \label{eq:measure}\eeq
80: where the parameter $\tau$ is usually defined as a conventional scaling
81: parameter in literature. The above equation was originally
82: derived by Anderson {\it et al}.[\onlinecite{Andersonscale}] by using
83: the random phase hypothesis.
84:
85: However, as shown in [\onlinecite{Deych}] without the assumption
86: of phase randomization, 1D SPS (Eq.[\ref{eq:measure}]) is
87: violated, where states are much far apart from each other than the
88: localization length ($\lambda$).
89:
90: This new characteristic length scale $\ell_{s}$ which is related to the distance between states,
91: for the states near the band center is defined in terms of total
92: number of states ($N(E)$) whose energy is less than $E$
93: [\onlinecite{Deych2}]. This criterion was initially extracted from
94: the exact calculation of the variance of L.E. for the Anderson
95: model with Cauchy distribution of the site energies
96: [\onlinecite{Deych}]. However in contradiction with violation of
97: SPS in 1D [\onlinecite{schomerus}], it has been recently shown
98: using exact diagonalization [\onlinecite{kantelhardt}] and
99: transfer matrix method [\onlinecite{Queiroz}] that at special
100: point $E=0$, SPS holds perfectly in 1D.
101:
102: In 2D case, the interest is motivated by the experimental
103: observations of a metal-insulator transition which is at odds with
104: the SPS for noninteracting electrons [\onlinecite{kravchenko}].
105: The validity of SPS in 2D is currently very controversial. There
106: exists some numerical analysis of 2D Anderson model which confirms
107: the SPS hypothesis [\onlinecite{Kramer, Scheriber,slevin}]. Other
108: studies suggest a two-parameter scaling
109: [\onlinecite{kantelhardt,Queiroz,Prior}]. It has been shown
110: [\onlinecite{Prior}] that 2D SPS does not follow
111: Eq.(\ref{eq:measure}).
112:
113: In 1D systems with off-diagonal disorder (random hopping model),
114: it is clear that an anomalously localized state at $E=0$, results
115: in a violation of SPS. Divergence of the localization length and
116: density of states at the band center in this model
117: [\onlinecite{Brouwer1},\onlinecite{Brouwer3}], is in contradiction
118: with the scaling theory. Unusual properties of this model are due
119: to chiral symmetry [\onlinecite{Brouwer2,Mudry}]. In an interval
120: close to an anomalous state, SPS does not hold
121: [\onlinecite{Deych2}]. The main objective of the present paper, is
122: to answer how far from the anomalous state (at $E=0$), 1D SPS will
123: again be held.
124:
125: In this paper, with care of some debates on the validity of 1D
126: SPS, we reexamine the scaling properties of one-dimensional system
127: with purely off-diagonal disorder by using transfer matrix method.
128: Our attention is on a region near the band center which contains
129: strongly localized states with standard symmetry. In the strong
130: localization limit, it will be shown that the L.E. distribution
131: function is normal. In this region, it is shown that there exists
132: a new length scale which is the same as the length scale
133: ($\ell_{s}$) defined in Eq.(\ref{eq:ells}) [\onlinecite{Deych}].
134: The SPS exists as long as the localization length $\lambda$
135: exceeds $\ell_{s}$. In the SPS region, it is shown that the L.E.
136: only depends on the disorder strength as
137: $\gamma_0\propto\sigma_{\ln(t)}^2$. The scaling properties of the
138: non-SPS region which has been reported in Ref.
139: [\onlinecite{cheraghchi}], is confirmed by a data collapse. The
140: variance and mean of the L.E. for different disorder strengths,
141: system sizes and also for various range of energy spectrum, lie on
142: a single curve when they are expressed in terms of the scaling
143: parameter ($\tau$) defined in Eq.(\ref{eq:measure}) as a function
144: of the ratio $\kappa=\lambda/\ell_{s}$. It can also provide a
145: physical interpretation for $\ell_{s}$ as a cross-over length
146: between two chiral and standard symmetries.
147:
148: This article is organized as follows: Section {\bf II} describes our model and the
149: exact calculation of all L.E. moments at the band center. Section
150: {\bf III} describes the transition from chiral symmetry region to
151: localized region by the calculation of L.E. distribution function
152: and its mean. In this Section, it will be shown that the localized
153: region is divided into non-SPS and SPS regimes. A new length
154: scale($\ell_{s}$) which controls the scaling theory is defined in
155: Section (IV). We try to find a meaningful physical interpretation
156: for the new length scale as a cross-over length in Section ({\bf
157: V}). Discussions and conclusions are finally presented in Section
158: {\bf VI}.
159:
160: \section{Model and Moments of Lyapunov Exponent}
161: %\subsection{\label{sec:level11} Model}
162: We consider non-interacting electrons in 1D disordered systems
163: within a tight binding approximation. The Schroedinger equation
164: with the assumption of nearest-neighbor hopping becomes
165: \beq
166: \varepsilon_{i}\psi_{i}+t_{i,i+1}\psi_{i+1}+t_{i-1,i}\psi_{i-1}=E\psi_{i}
167: \eeq where E is the energy corresponding to the electron wave
168: function. ${\bf |\psi_{i}|}^{2}$ is the probability of finding the
169: electron at site i, ${\varepsilon_{i}}$ are the site potentials
170: and ${t_{i-1,i}=t_{i,i-1}=t_{i}}$ the hopping terms. Using the
171: transfer matrix method, one can relate the electron wave functions
172: at the two ends of the system to each other. In our model, we
173: consider all site energies to be zero and a periodic boundary
174: condition on hopping terms as $t_{1} = t_{N+1}$. All energies
175: which appear, are scaled by typical mean of hoppings terms $t_0$,
176: where $\ln(t_{0}) = <\ln(t_{i})>_{c.a.}$. Here, {\bf c.a.} refers
177: to the configurational average. The L.E. can be extracted from the
178: eigenvalues of the total transfer matrix
179: [\onlinecite{cheraghchi}].
180:
181: As proved in Ref.[\onlinecite{cheraghchi}], the L.E.
182: at $E=0$ has a {\it semi-Gaussian} distribution with a mean which
183: can be derived in terms of the pair correlation function. By having the distribution function, higher powers of the L.E. can
184: be simply derived in the case of correlated and uncorrelated disorder at the band center ($E=0$) as:
185:
186: \begin{eqnarray} <\gamma^2>=\frac{\pi}{2}<\gamma>^{2} ;
187: <\gamma^3>=\pi <\gamma>^{3} ; ... ; \nonumber\\
188: <\gamma^n>\propto<\gamma>^{n}
189: \end{eqnarray}
190: Therefore, higher moments of the L.E. can be written as:
191:
192: \begin{eqnarray}\sigma^{2}_{\gamma}=<(\gamma-<\gamma>)^{2}>=(\frac{\pi}{2}-1)<\gamma>^{2}
193: \nonumber \\
194: <(\gamma-<\gamma>)^{3}>=(2-\frac{\pi}{2})<\gamma>^{3}
195: \label{eq:moments}\end{eqnarray}
196: %
197: % FIG.1
198: %
199: \bfig
200: \includegraphics[width=8cm]{fig1.eps}
201: %\includegraphics[width=21 cm, angle=90]{GAMMA(N).pdf}
202: \caption{ The size dependency of the Lyapunov Exponent for
203: different energies with $\eta=0.18$.\label{fig:gamma(N)}}
204: \efig
205: This can be generalized to the n'th moment of the L.E. which will
206: be as $<(\gamma-<\gamma>)^{n}>\propto<\gamma>^{n}$. By paying our
207: attention to the L.E. form at $E=0$ ($<\gamma>
208: \propto\sigma_{\ln(t)}/N^{1/2}$), it can be seen that the variance
209: of the L.E., scales according to the law of large numbers for
210: uncorrelated disorder. As it has been mentioned by Anderson. et al
211: [\onlinecite{Andersonscale}], the localization properties can be
212: described by a variable (such as L.E.) whose width of distribution
213: function follows the law of large numbers.
214: \beq
215: \sigma_{\gamma}^{2}=(1-\frac{2}{\pi})\frac{\sigma_{\ln(t)}^{2}}{N}\label{eq:unvar}\eeq
216: This equation and its equivalence in Eq.[\ref{eq:moments}] are
217: consistent with the result of Ref.[\onlinecite{Mudry}], where it
218: was derived for a weak disorder by solving the Fokker-Planck
219: equation. However, the size dependence of the L.E. variance will
220: change when the disorder becomes correlated.
221: As a result, the L.E. distribution function and all
222: its higher moments converge for large system sizes. So, L.E. is a
223: good variable to describe statistical properties of disordered
224: systems.
225:
226: \section{Scaling and distribution function of Lyapunov Exponent}
227: \subsection{\label{sec:level31} Scaling of Lyapunov Exponent}
228: We calculate the mean and variance of L.E. by using
229: the transfer matrix method when randomness is imposed on
230: $\ln(t)$'s. The study of L.E. close to the band center results in the coexistence
231: of two symmetries in this system. It can be shown that there is a chiral symmetry at
232: $E=0$. This is a significant property of purely
233: off-diagonal disorder with nearest-neighbor approximation.
234: However, at energies close to the band center, and for lengths
235: greater than a cross-over length ($N_{cr}$), chiral symmetry is
236: broken. At sufficiently long lengths, localization properties will
237: flow to those of the standard symmetry class. All states in this
238: regime are strongly localized.
239: % characterized by a length scale which is namely localization length.
240:
241: For any realization of the disorder, the energy density of states
242: is symmetric around the band center. This symmetry, which
243: originates from the fact that the disorder preserves the
244: bipartite structure of the lattice, is referred to as chiral
245: symmetry. The chiral symmetry is broken by, e.g., on-site
246: randomness or next-nearest-neighbor hopping
247: [\onlinecite{inui,Mudry}].
248:
249: In the case of purely onsite disorder case which was originally
250: considered by Anderson [\onlinecite{anderson}], one distinguishes
251: three universality classes, corresponding to the presence or
252: absence of time reversal and spin-rotation symmetry. These three
253: classes are called orthogonal, unitary, and symplectic
254: [\onlinecite{Mudry}]. Here, we will refer to these as the three
255: standard universality classes. In this paper, it will be shown
256: that the standard symmetry region (strongly localized region) is
257: also divided into two regimes (non-SPS and SPS regimes defined in
258: section (III.C) and Fig.(\ref{fig:Nvar-gam})), which depend on the
259: number of scaling parameters.
260:
261: Fig.(\ref{fig:gamma(N)}) shows the scaling properties of L.E. near
262: the band center. All data with various energies lie on a single
263: curve when ($\gamma\times E^{-\eta}$) is plotted in terms of the
264: dimensionless variable ($N/N_{cr}$). It confirms a power law
265: divergence of the localization length where energies belong to the
266: non-SPS regime (Fig.(\ref{fig:Nvar-gam})). Therefore, the
267: following scaling law of L.E. at $E \neq 0$ can be proposed
268: [\onlinecite{cheraghchi}].
269:
270: \beq (\gamma\times E^{-\eta}) \propto \left\{
271: \begin{array}{c} \sigma_{ln(t)}^2(\frac{N}{N_{cr}})^{-1/2}
272: \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\ N /N_{cr}\ll1
273: \\ \\ \sigma_{ln(t)}^{2}
274: \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, N/N_{cr}\gg1\end{array} \right.
275: \label{eq:cr-energy}
276: \eeq
277: where the cross-over length is as
278: \beq N_{cr}\propto{E^{-2\eta}}/{\sigma_{ln(t)}^{2}}\eeq
279: and $\eta \approx 0.18 \pm 0.03$. However, for energies
280: very close to the band center, $\eta$ has a small energy dependence. Fig.(\ref{fig:gamma(N)})
281: shows a transition from the region with chiral symmetry ($N \ll N_{cr}$) to the region with standard
282: symmetry ($N \gg N_{cr}$). It was also checked that each of the
283: data sets do not collapse on each other when one uses the
284: logarithmic energy dependence of the localization length as seen
285: in Refs.(\onlinecite{cohen,ziman}).
286:
287:
288: \subsection{\label{sec:level32} Distribution Function of Lyapunov Exponent}
289:
290: The localization properties of different symmetry regions can be
291: also characterized by the distribution function of L.E. As it was
292: mentioned in section (II), at the band center, the distribution
293: of L.E. is semi-Gaussian. For zero energy ($E=0$), numerical
294: evidence in Fig.(\ref{fig:distribution}.a) confirms such
295: analytical distribution function for all system sizes.
296: %
297: % FIG.2
298: %
299:
300: \bfig
301: \begin{center}
302: \includegraphics[width=8 cm]{fig2.eps}
303: % \includegraphics[width=21 cm, angle=90]{dist.pdf}
304: \end{center}
305: \caption{Lyapunov Exponent Distribution Function for different
306: system sizes at a) $E=0$ and b) $E=10^{-5}$. Disorder strength is
307: considered to be 0.1. The number of samples is $2\times10^4$
308: configurations. \label{fig:distribution}}
309: \efig
310:
311: For energies near the band center ($E\neq0$), and for system sizes smaller
312: than the cross-over length ($N \ll N_{cr}$), the distribution of
313: L.E. is {\it semi-Gaussian}. In this region, the band center
314: behavior is dominant. However, as the system size increases, the
315: distribution
316: function becomes more \textit{Gaussian}-like for $N \gg N_{cr}$. %changes and for system size greater than the cross-over length ($N \gg N_{cr}$), distribution function will be as a {\it Gaussian} form.
317: This region has a regular Anderson-like behavior with standard
318: symmetry class. Fig.(\ref{fig:distribution}b) shows distribution
319: function of the L.E. for the same system sizes as the band center
320: case. It can be seen that there exists again a transition between
321: these two symmetries; from chiral to standard symmetry or from
322: semi-Gaussian to Gaussian distribution. For the sake of
323: completeness, the skewness of the distribution functions has been
324: calculated as a measure of the symmetry of the distribution. It
325: can be defined as [\onlinecite{press}]: \beq {\rm
326: Skewness}=\frac{<(\gamma-{\overline{\gamma}})^3>}{{<(\gamma-{\overline{\gamma}})^2>}^{3/2}}\eeq
327: Distribution close to the normal form, has a skewness equal to zero. A distribution whose skewness has absolute
328: value less than $0.5$ is considered fairly symmetrical. Therefore,
329: distributions of long enough systems in
330: Fig.(\ref{fig:distribution}b), are very close to the Gaussian
331: form. In the SPS regime (Fig.(\ref{fig:Nvar-gam})) where the L.E.
332: is independent of size and energy, distribution function of L.E.
333: is exactly Gaussian (the skewness order of $10^{-3}$) and
334: independent of size. All distribution curves have been softened by
335: the Kernel smoothing method [\onlinecite{Kernel}] without changing
336: any statistical characteristic of distributions.
337: \subsection{\label{sec:level33} SPS and Non-SPS Regimes}
338: For a fixed length ($N$), Eq.(\ref{eq:cr-energy}) proposes a
339: critical energy point which separates two regions with different
340: symmetries. In fact, for energies greater than
341: $\varepsilon_{cr.}^{(1)}\propto (N
342: \sigma_{\ln(t)}^{2})^{\frac{-1}{2\eta}}$, the system is in the
343: localized region. Fig.(\ref{fig:Nvar-gam}) shows the energy
344: dependence of L.E. in the localized region. It can be seen that
345: there is a second critical point ( $\varepsilon_{cr.}^{(2)}\approx
346: 10^{-2}$ in Fig.(\ref{fig:Nvar-gam})) where the energy spectrum is
347: divided into SPS and non-SPS regimes. In the SPS regime, L.E. is
348: independent of energy and the scaling theory is valid
349: (Eq.(\ref{eq:measure})). Although, near the band edges anomaly,
350: L.E. will become energy dependent. In the SPS regime, it has been
351: checked (Figs.(\ref{fig:gamma(N)},\ref{fig:Nvar-gam})) that in
352: contradiction
353: to the result of Ref.[\onlinecite{izrailev}], the
354: L.E. only depends on disorder strength. Fig.(\ref{fig:Nvar-SN})
355: shows that the L.E. is proportional to the square of disorder
356: strength.
357: \beq
358: \gamma_0\propto\sigma_{\ln(t)}^2\label{eq:sps_scale}\eeq
359: The line Fitted on data in Fig.(\ref{fig:Nvar-SN}) has a slope equal to the value 2.
360: The coefficient of the above scaling law is ($\frac{1}{3.00\pm0.2}$)
361: which can be extracted for a
362: fixed disorder strength ($\sigma_{\ln(t)}=0.1$) and for an energy
363: ($\varepsilon=0.1 \geq\varepsilon_{cr.}^{(2)}$) in the SPS regime.
364: The skewness of this point is about 0.005. This skewness shows an exactly
365: Gaussian form for L.E. in this regime.
366:
367: It can be seen that matching of two
368: Eqs.(\ref{eq:cr-energy},\ref{eq:sps_scale}) at the boundary of
369: non-SPS to SPS regime ($E=\varepsilon_{cr}^{(2)}$) leads to a
370: second critical point ($\varepsilon_{cr}^{(2)}$) which is
371: independent of all system parameters (a constant).
372: %
373: % FIG.3
374: %
375: \bfig
376: \includegraphics[width=8 cm]{fig3.eps}
377: %\includegraphics[width=21 cm, angle=90]{Nvar_gam.pdf}
378: \caption{SPS to non-SPS transition with plotting Lyapunov Exponent
379: and $N\sigma_{\gamma}^{2}$ versus energy for system size
380: $10^5$. $\varepsilon_{cr.}^{(1)}$ which depends on the disorder strength,
381: is about $5\times10^{-9}$ and $3\times10^{-7}$ for $\sigma_{\ln(t)}=0.1$ and $0.075$
382: , respectively. $\varepsilon_{cr.}^{(2)}$ is equal to $10^{-2}$.\label{fig:Nvar-gam}}
383: \efig
384:
385: Now, we numerically study the variance of L.E. as a function of
386: system parameters. Fig(\ref{fig:Nvar-gam}) shows
387: $N\sigma_{\gamma}^{2}$ versus energy at fixed system size. As it
388: can be seen, the variance of L.E. is approximately independent of
389: energy at energies in the SPS and also non-SPS regimes. Figs.(\ref{fig:Nvar-gam},\ref{fig:Nvar-SN}) show that size
390: and disorder strength dependence of the variance of L.E. follows
391: from Eq.(\ref{eq:unvar}). As it is clear from a fitted line (with slope $2$)
392: on data in Fig.(\ref{fig:Nvar-SN}),
393: the quantity $N\sigma_{\gamma}^2$ is
394: proportional to $\sigma_{\ln(t)}^{2}$. The size dependence of the
395: L.E. variance ($\sigma_{\gamma}^{2}(N)$) is shown in the inset
396: Fig.(\ref{fig:Nvar-SN}). The L.E. variance decreases with the
397: inverse of the system size similar to the size dependence of
398: variance at the band center (Eq.(\ref{eq:unvar})). So, the
399: variance of L.E. at the band center can be generalized to other
400: energies near the band center.
401: \section{Violation of Single Parameter Scaling}
402: According to Eq.(\ref{eq:measure}), the two parameters of the
403: distribution reduce to only one. Parameter $\tau$ can be defined
404: as a measure of SPS in that equation.
405: %
406: % FIG.4
407: %
408: \bfig
409: \includegraphics[width=8 cm]{fig4.eps}
410: %\includegraphics[width=21 cm, angle=90]{Nvar(S&N).pdf}
411: \caption{ Standard deviation of the Lyapunov Exponent versus
412: disorder strength for system with length $10^5$. Dependency of
413: $\gamma_0$ to disorder strength in the SPS regime ($E=0.1$). Inset
414: figure shows the system size dependence of the variance of
415: Lyapunov Exponent.\label{fig:Nvar-SN}}
416: \efig
417: Now, in this Section, we try to find the independent parameters of
418: the system in non-SPS regime. It is again stressed that the random
419: hopping model away from the $E=0$ and in the localized region is
420: being studied. The quantity of interest is $\ell_{s}$ that is
421: related to the integral density of states from the band center to
422: a given energy, normalized by the total number of states in the
423: band.
424:
425: \beq \ell_{s}=\frac{1}{\sin(\pi N(E))} \label{eq:ells}\eeq
426:
427: For the Anderson model, $N(E)$ can be computed by the node-counting
428: theorem [\onlinecite{node}]. By starting an initial vector in
429: transfer matrix method, we count the number of wave function nodes
430: as the length is scanned.
431:
432: Fig.(\ref{fig:SPS-D}) shows the numerical result of inverse
433: scaling parameter ($\tau (\kappa)$) in terms of the dimensionless
434: parameter ($\kappa=\frac{\lambda}{\ell_{s}}$) for different values
435: of disorder strength and energy.
436:
437: The data included in this graph correspond to the localized regime
438: with standard symmetry class where the L.E. has a Gaussian distribution.
439: The inset Fig.(\ref{fig:SPS-D}) shows that the skewness of
440: the data are less than 0.25 for $\kappa>0.1$. All data are in the
441: region $N\gg\ell_{s}$. What is important, is that all data with
442: different values of energy and disorder strength and also system
443: size collapse to a single curve when they are expressed in terms
444: of $1/\tau$ and $\kappa$. Therefore, for $\kappa \ll1$ (non-SPS
445: regime) variance of L.E. depends on two parameters; $\kappa$ and
446: the mean of L.E. In the case of $\kappa \gg1$,
447: the inverse of the scaling parameter ($1/\tau$) in the present model
448: goes to the value 0.5 which is different from unity. Therefore, our expression
449: from non-SPS is only the deviation of Eq.(\ref{eq:measure}) from 0.5.
450: %
451: % Fig.5
452: %
453: \bfig
454: \includegraphics[width=8 cm]{fig5.eps}
455: %\includegraphics[width=21 cm, angle=90]{SPS_D_smalk.pdf}
456: \caption{ Measure of the validity of the Single Parameter Scaling
457: by inverse Scaling Parameter ($1/\tau =\gamma_{0}/(N
458: \sigma^{2}_{\gamma})$) in terms of $\kappa=\lambda/\ell_{s}$ for
459: disorder strengths from $\sigma_{\ln(t)}=0.05$ to $0.2$. Energies
460: depending on disorder strength, are scanned from
461: $E=10^{-7}-10^{0}$. Here $N \gg \ell_{s}$. Inset figure shows the skewness of the data.
462: The number of samples is $10^4$ configurations.
463: \label{fig:SPS-D}}
464: \efig
465: Energies are scanned from near the band center to the middle of
466: the conduction band, and far from the second anomaly in the band
467: edges. In fact, close to any anomaly (here a delocaliztion at the
468: band center), a violation of SPS will happen
469: [\onlinecite{Deych2}] in an interval close to the anomalous
470: state. It can be seen that for $\kappa \ll 1$, SPS is clearly
471: violated (non-SPS region). For $\kappa \gg 1$, independent of
472: disorder strength, standard SPS is restored again. In the SPS
473: spectral region, the localization length is a macroscopic length
474: as only one parameter of system. The conductance can be defined
475: as a function of this variable.
476:
477: Both length scales $\ell_{s}$ and the localization length are
478: decreased when energy is swept from the vicinity of the band
479: center to the band edges. In energies close to zero, the new
480: length scale is greater than the localization length. There is a
481: critical energy where both length scales are of the same order
482: ($\lambda\approx \ell_s$). Far from the band center, the
483: localization length is independent of energy and the new length
484: scale steeply decreases so that it would be much smaller than a
485: macroscopic localization length.
486:
487: This result confirms the general conjecture of the authors
488: [\onlinecite{Deych}] that the second moment of the distribution
489: function of L.E. can be universally described in terms of
490: variables $\tau$ and $\kappa$ regardless of the microscopic nature
491: of the models under consideration. The form of the function
492: $\tau(\kappa)$ may differ for different models and its essential
493: behavior is not universal. All models follow $\tau(\kappa)=1 $
494: for $\kappa\gg 1$, while in the hopping disorder model, it is $\tau(\kappa)=2$.
495: In the model studied in the present paper, for
496: $\kappa\ll 1$, it can be seen an exceptional behavior compared to
497: other models such as Lloyd model
498: [\onlinecite{Deych,Deych2,Deych3}] and Anderson (onsite disorder)
499: and superlattice models [\onlinecite{Deych3}]. The scaling
500: parameter $\tau$ increases with $\kappa$ in the hopping disorder
501: model for $\kappa \ll 1$, while in the above models, $\tau$
502: steeply decreases with $\kappa$. As an example, analytical
503: calculations carried out in Ref.[\onlinecite{Deych}] for the
504: Lloyd Model produced $\tau=(\pi/2)\kappa$.
505:
506: The power law form is the best fitted curve for $\kappa\ll1$.
507: \beq
508: \frac{1}{\tau}\propto \alpha \kappa^{\beta} \eeq
509: Coefficients are estimated by using a linear regression in log-log plot. The fitted power of
510: $\kappa$ and its coefficient are $\beta= 0.815\pm0.008$ and
511: $\alpha=1.64\pm0.03$, respectively. Since there is a kind of
512: delocalization at the band center, the L.E. sharply decreases near
513: the band center compared to $N\sigma_{\gamma}^{2}$
514: (Fig.(\ref{fig:Nvar-gam})). Therefore, the inverse scaling
515: parameter decreases in energies close to the band center.
516: %
517: % Fig.6
518: %
519: \bfig
520: \includegraphics[width=8 cm]{fig6.eps}
521: %\includegraphics[width=21 cm, angle=90]{SPS_Ncr.pdf}
522: \caption{ Inverse scaling parameter $1/\tau =\gamma_{0}/(N
523: \sigma^{2}_{\gamma})$ versus localization length in the ratio of
524: the cross-over length ($\kappa' = \lambda/N_{cr}$) for different
525: disorder strengths. Energies
526: depending on disorder strength, are scanned from
527: $E=10^{-7}-10^{0}$.\label{fig:SPS}} \efig
528: \section{Cross-over Length as a physical interpretation of $\ell_{s}$}
529: Numerical results in Fig.(\ref{fig:Nvar-gam}) show that as a crude
530: approximation, one can consider the variance of L.E. independent
531: of energy in the non-SPS and SPS regimes (not near their
532: transition point). It was shown that the form of variance near the
533: band center is similar to its form in the band center
534: (Eq.(\ref{eq:unvar})). By using this form of the variance and the
535: scaling relations derived in
536: Eqs.(\ref{eq:cr-energy},\ref{eq:sps_scale}), the scaling parameter
537: function ($\tau$) can be proposed in the non-SPS and SPS regimes.
538:
539: First, we investigate the scaling parameter in the non-SPS
540: regime. According to the infinite L.E. in the localized region
541: (Eq.(\ref{eq:cr-energy})) $\gamma_{0}\propto\sigma_{\ln(t)}^{2}
542: E^{\eta}$ and its variance as
543: $N\sigma_{\gamma}^2\propto\sigma_{\ln(t)}^2$, it can be seen that
544: the measure of SPS (scaling parameter $\tau$) can be proposed to
545: have a linear relation with the dimensionless variable
546: $\kappa'=\lambda/N_{cr}$.
547:
548: \beq\frac{1}{\tau} \propto E^{\eta} \propto
549: \frac{\lambda}{N_{cr}} \label{eq:proposed}\eeq
550:
551: This equation shows a deviation from the SPS value (unity) in the
552: non-SPS regime. The cross-over length $N_{cr}$ plays the role of
553: a length scale like $\ell_{s}$ in this system. The above form for
554: the scaling parameter is independent of the system parameters
555: such as disorder strength when it is expressed in terms of $\tau$
556: and $\kappa'$. This expression is confirmed by Fig.(\ref{fig:SPS})
557: which shows the inverse scaling parameter versus $\kappa'$. It
558: can be seen that for $\kappa'\ll1$, all data for different
559: disorder strengths, coincide with each other on a single curve.
560: However, since in the second critical point ($\kappa'\approx1$),
561: the variance of L.E. is energy dependent and also, the scaling
562: form of L.E. (Eq.(\ref{eq:cr-energy})) is not correct in this
563: point (Fig.(\ref{fig:Nvar-gam})), curves with different disorder
564: strengths are separated from each other in the transition point.
565:
566: In the SPS regime, the scaling of L.E. proposed in
567: Eq.(\ref{eq:sps_scale}) $\gamma_0\propto\sigma_{\ln(t)}^2$ and its
568: variance form as Eq.(\ref{eq:unvar}), show that the scaling
569: parameter $\tau$ is independent of system parameters and a
570: constant (Fig.(\ref{fig:SPS})).
571:
572: Therefore, both the new length scale and the
573: cross-over length that are of the same order in the non-SPS regime,
574: can characterize the scaling properties. Fig.(\ref{fig:Ls})
575: compares these two length scales. It shows the ratio of the new
576: length to the cross-over length in terms of energy. In the Non-SPS
577: regime, they both weakly depend on energy, although, in the
578: transition region, the difference is remarkable.
579:
580: As a result, the cross-over length is proposed as a physical and
581: meaningful interpretation for the new length scale ($\ell_{s}$).
582: On the other hand, we showed that the statistical distribution of
583: L.E. is different for sizes lower or greater than this length
584: scale.
585: %
586: % Fig.7
587: %
588: \bfig
589: \includegraphics[width=8 cm]{fig7.eps}
590: %\includegraphics[width=21 cm, angle=90]{L_sN_cr.pdf}
591: \caption{The ratio of the new length scale ($\ell_{s}$) to the
592: cross-over length in terms of energy for different values of
593: disorder strength.\label{fig:Ls}}
594: \efig
595: \section{Conclusion}
596: In this paper, we study the scaling theory in the hopping
597: disorder model. The main result of this paper is to show the
598: single parameter scaling (SPS) is violated not only in a region
599: with chiral symmetry, but also in the localized region where there
600: exists a standard symmetry class. The localized region is also
601: divided into two regimes: SPS and non-SPS regimes. We proposed
602: the scaling relations for the Lypunov Exponent in these two
603: regimes. The criterion of the SPS is controlled by a new length
604: scale which is related to the integral of density of states,
605: $\ell_{s}$ defined in Ref.[\onlinecite{Deych}]. The SPS holds when
606: the localization length $\lambda$ exceeds the new length
607: ($\lambda\gg\ell_{s}$). In $\lambda\ll\ell_{s}$ regime, standard
608: deviation of the Lyapunov Exponent ( $\gamma$) distribution can
609: be described by two independent scaling parameters: the mean of
610: $\gamma$ and $\kappa=\lambda/\ell_{s}$.
611:
612: We showed that all data related to the variance and mean of the
613: Lyapunov Exponent with different values of disorder strengths,
614: system sizes and also the data extracted from various energy
615: regions, lie on a single curve, when they are expressed in terms
616: of the inverse scaling parameter
617: $1/\tau=\gamma_{0}/N\sigma^2_{\gamma}$ and the dimensionless
618: variable $\kappa$.
619:
620: The cross-over length ($N_{cr}$) which separates the region with
621: chiral symmetry from that of standard symmetry, is proposed as a
622: meaningful physical interpretation for $\ell_{s}$.
623:
624: \begin{acknowledgments}
625: I wish to acknowledge Prof. Keivan Esfarjani for a critical reading of the manuscript
626: and his effective comments and suggestions.
627: I would like to thank Dr. S. Mahdi Fazeli for useful discussions on the initial
628: manuscript. I would like also to thank Dr. Nima Ghal-Eh for his
629: review and editing of the manuscript.
630: \end{acknowledgments}
631:
632: \begin{thebibliography}{99}
633:
634: \bibitem{anderson}
635: P. W. Anderson, Phys. Rev. {\bf 109}, 1492 (1958).
636: \bibitem{soukoulis}
637: C. M. Soukoulis, E. N. Economou, Phys. Rev. B. {\bf 24}, 5698
638: (1981).
639: \bibitem{cohen}
640: G. Theodorou, M. H. Cohen, Phys. Rev.B. {\bf 13}, 4597 (1976).
641: \bibitem{inui}
642: M. Inui, S. A. Trugman and E. Abrahams, Phys. Rev. B {\bf 49}, 3190 (1994).
643: \bibitem{ziman}
644: T. A. L. Ziman, Phys. Rev. Lett. {\bf 49}, 337, (1982)
645: \bibitem{abrahams}
646: E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V.
647: Ramakrishnan, Phys. Rev. Lett. {\bf 42}, 673 (1979).
648: \bibitem{Andersonscale}
649: P. W. Anderson, D. J. Thouless, E. Abrahams, D. S. Fisher, Phys.
650: Rev. B, {\bf 22}, 3519, (1980)
651: \bibitem{Deych}
652: L. I. Deych, A. A. Lisyansky, and B. L. Altshuler,
653: Phys. Rev. Lett, {\bf 84}, 2678 (2000); Phys. Rev. B, {\bf 64},
654: 224202 (2001).
655: \bibitem{Deych4}
656: L. I. Deych, M. V. Erementchouk, A. A. Lisyansky, Phys. Rev. B, {\bf 67},024205 (2003).
657: \bibitem{Deych2}
658: L. I. Deych, M. V. Erementchouk, A. A. Lisyansky, and B. L.
659: Altshuler, Phys. Rev. Lett. {\bf 91}, 096601 (2003)
660: \bibitem{schomerus}
661: H. Schomerus, M. Titov, Phys. Rev. B, {\bf 67}, 100201(R) (2003).
662: \bibitem{kantelhardt}
663: J.W. Kantelhardt and A. Bunde, Phys. Rev. B 66, 035118 (2002)
664: \bibitem{Queiroz}
665: S. L. A. de Queiroz, Phys. Rev. B 66, 195113 (2002)
666: \bibitem{kravchenko}
667: E. Abrahams, S. V. Kravchenko, and M. P. Sarachik, Rev. Mod.
668: Phys. {\bf 73}, 251 (2001)
669: \bibitem{Kramer}
670: M. MacKinnon, B. Kramer, Phys. Rev. Lett. {\bf 47} 1546 (1981)
671: \bibitem{Scheriber}
672: M. Scheriber, M. Ottomeier, J. Phys.: Condens. Matter {\bf 4},
673: 1959 (1992)
674: \bibitem{slevin}
675: K. Slevin, Y. Asada, and L.I. Deych, Phys. Rev. B 70, 054201
676: (2004)
677: \bibitem{Prior}
678: J. Prior, A. M. Somoza, M. Ortuno, Phys. Rev. B, {\bf 72}, 024206
679: (2005)
680: \bibitem{Brouwer1}
681: P. W. Brouwer, C. Mudry, and A. Furusaki, Nucl. Phys. B, {\bf
682: 565} (2000)
683: \bibitem{Brouwer3}
684: P. W. Brouwer, C. Mudry, A. Furusaki, Phys. Rev. Lett, {\bf 84}, 2913, (2000)
685: \bibitem{Brouwer2}
686: C. Mudry,P. W. Brouwer and A. Furusaki, Phys. Rev. B, {\bf 62},
687: 8249, (2000)
688: \bibitem{Mudry}
689: C. Mudry, P. W. Brouwer, A. Furusaki, Phys. Rev. B, {\bf 62},
690: 8249, (2000)
691: \bibitem{Deych3}
692: L. I. Deych, M. V. Erementchouk, A. A. Lisyansky, A. Yamilov, and
693: H. Cao, Phys. Rev. B, {\bf 68}, 174203 (2003); L. I. Deych, M. V.
694: Erementchouk, and A. A. Lisyansky, Phys. Rev. Lett. {\bf 90},
695: 126601 (2003).
696: \bibitem{cheraghchi}
697: H. Cheraghchi, S. M. Fazeli, K. Esfarjani, Phys. Rev. B. {\bf
698: 72}, 174207,(2005)
699: \bibitem{press}
700: W. H. Press, A. A. Teukolsky, W. T. Vetterling and B. P.
701: Flannery. {\it Numerical Recipes in Fortran} (Cambridge Uni-
702: versity Press, Cambridge, 1992).
703: \bibitem{Kernel}
704: A. Mugdadi, E. Munthali, Journal of Statistical Research, {\bf
705: 37}, 2,203-218 (2003).
706: \bibitem{izrailev}
707: L. Tessieri, F. M. Izrailev, Physica E {\bf 9}, 405 (2001)
708: \bibitem{kappus}
709: M. Kappus, F. Wegner, Z. Phys. B: Condens. Matter {\bf 45}, 15
710: (1981)
711: \bibitem{node}
712: F. R. Gantmacher, {\it Theory of Matrices} (Chelsa Publishing
713: Company, New York (1956)
714: \end{thebibliography}
715: %
716: %%
717: %%% FIG.1
718: %%
719: %\bfig
720: %%\includegraphics[width=8cm]{fig1.EPS}
721: %\includegraphics[width=21 cm, angle=90]{fig1.eps}
722: %\caption{ The size dependency of the Lyapunov Exponent for
723: %different energies with $\eta=0.18$. Variance is considered to be
724: %0.1.\label{fig:gamma(N)}}
725: % \efig
726: %%% FIG.2
727: %%
728: % \bfig
729: % \begin{center}
730: %%\includegraphics[width=8 cm]{fig2.eps}
731: % \includegraphics[width=21 cm, angle=90]{fig2.eps}
732: %\end{center}
733: %\caption{Lyapunov Exponent Distribution Function for different
734: %system sizes at a) $E=0$ and b) $E=10^{-5}$. Variance is
735: %considered to be 0.1. The number of samples is $2\times10^4$
736: %configuration. \label{fig:distribution}}
737: % \efig
738: %
739: %%
740: %%% FIG.3
741: %%
742: %\bfig
743: %%\includegraphics[width=8 cm]{fig3.EPS}
744: %\includegraphics[width=21 cm, angle=90]{fig3.eps}
745: %\caption{SPS to non-SPS transition with plotting Lyapunov Exponent
746: %and $N\sigma_{\gamma}^{2}$ versus energy for system size
747: %$10^5$.\label{fig:Nvar-gam}}
748: % \efig
749: %%
750: %%% FIG.4
751: %%
752: %\bfig
753: %%\includegraphics[width=8 cm]{fig4.EPS}
754: %\includegraphics[width=21 cm, angle=90]{fig4.eps}
755: %\caption{ Standard deviation of the Lyapunov Exponent versus
756: %disorder strength for system with length $10^5$. Dependency of
757: %$\gamma_0$ to disorder strength in the SPS regime ($E=0.1$). Inset
758: %figure shows the system size dependence of the variance of
759: %Lyapunov Exponent.\label{fig:Nvar-SN}}
760: % \efig
761: %%
762: %% % Fig.5
763: %%
764: %\bfig
765: %%\includegraphics[width=8 cm]{fig5.EPS}
766: %\includegraphics[width=21 cm, angle=90]{fig5.eps}
767: %\caption{ Measure of the validity of the Single Parameter Scaling
768: %by inverse Scaling Parameter ($1/\tau =\gamma_{0}/(N
769: %\sigma^{2}_{\gamma})$) in terms of $\kappa=\lambda/\ell_{s}$ for
770: %disorder strengths from $\sigma_{\ln(t)}=0.05$ to $0.2$. Energies
771: %depending on disorder strength, are scanned from
772: %$E=10^{-7}-10^{0}$. Here $N \gg \ell_{s}$. Inset figure shows the
773: %skewness of the data with disorder strengths (0.05 to 0.15) and
774: %for system size $10^5$. \label{fig:SPS-D}}
775: % \efig
776: %%
777: %% % Fig.6
778: %%
779: %\bfig
780: %%\includegraphics[width=8 cm]{fig6.EPS}
781: %\includegraphics[width=21 cm, angle=90]{fig6.eps}
782: %\caption{ Inverse scaling parameter $1/\tau =\gamma_{0}/(N
783: %\sigma^{2}_{\gamma})$ versus localization length in the ratio of
784: %the cross-over length ($\kappa' = \lambda/N_{cr}$) for different
785: %disorder strengths.\label{fig:SPS}} \efig
786: %%% Fig.7
787: %%
788: %\bfig
789: %%\includegraphics[width=8 cm]{fig7.eps}
790: %\includegraphics[width=21 cm, angle=90]{fig7.eps}
791: %\caption{The ratio of the new length scale ($\ell_{s}$) to the
792: %cross-over length in terms of energy for different values of
793: %disorder strength.\label{fig:Ls}}
794: % \efig
795:
796: \end{document}
797: