cond-mat0208018/mfa.tex
1: \documentclass[prl,twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4} 
2: %\documentclass[prl,twocolumn]{revtex4}
3: \usepackage{amsmath}
4: \usepackage{graphics}
5: \usepackage{graphicx}% Include figure files 
6: \usepackage{dcolumn}% Align table columns on decimal point 
7: \usepackage{bm}% bold math 
8: 
9: \begin{document}
10: 
11: 
12: %\Css{body { border : solid black }
13: %body { border-width: 2em }
14: %body { padding : 5em 5em 5em 5em }
15: %body { width : 30em }
16: %body { align : centre }}
17: %
18: %\CutAt{section}
19: 
20: 
21: 
22: \title{Disorder-Induced Vibrational Localization}
23: 
24: \author{J.~J.~Ludlam, S.~N.~Taraskin, S.~R.~Elliott }
25: \affiliation{ Department of Chemistry, University of Cambridge,
26: 	      Lensfield Road, Cambridge, CB2 1EW, UK }
27: 
28: 
29: \date{\today}
30: 
31: \begin{abstract}
32: 
33: The vibrational equivalent of the Anderson tight-binding
34: Hamiltonian has been studied, with particular focus on the 
35: properties of the eigenstates at the transition from 
36: extended to localized states. The critical energy has been
37: found approximately for several degrees of force-constant disorder using
38: system-size scaling of the multifractal spectra of the
39: eigenmodes, and the spectrum at which there is no system-size 
40: dependence has been obtained. This is shown to be in good
41: agreement with the critical spectrum for the electronic
42: problem, which has been derived both numerically and by 
43: analytic means. Universality of the critical states
44: is therefore suggested also to hold for the vibrational problem.
45: 
46: \end{abstract}
47: 
48: \pacs{63.50.+x, 63.20.Pw}
49: 
50: \maketitle
51: %\tableofcontents{}
52: 
53: The Anderson electron localization problem \cite{Anderson_58} is one that has attracted
54: much attention over the last 40 years. The fact that the problem
55: can be stated so simply, and yet have startlingly complex 
56: consequences, has made it a challenging topic to work on \cite{Kramer_93}. Indeed,
57: only recently has it been possible to verify numerically
58: many of the theoretical results on powerful supercomputers \cite{Schreiber_96:Book}. However, the 
59: closely related vibrational problem has not been explored to the same degree, 
60: despite being similar enough to use the same techniques 
61: yet different enough to produce new and interesting results.
62: 
63: %\section{Introduction}
64: The phenomenon of localization is a second-order phase transition 
65: between eigenstates that are spatially localized and those that are 
66: delocalized, or extended \cite{Janssen_98:review}. In the thermodynamic limit, extended 
67: eigenmodes would cover the whole of space whereas localized 
68: eigenstates are those which only involve a local subset of the 
69: system within a typical localization length. In the crystalline 
70: case for both the electronic and vibrational problems, 
71: the eigenstates are simple Bloch states due to translational 
72: invarience, and are therefore extended.
73: For the electronic problem, disorder is generally 
74: introduced either in the on-site energy terms (diagonal disorder) 
75: or the interaction terms (off-diagonal disorder) \cite{Kramer_93}. 
76: In a 3d lattice with weak diagonal disorder, 
77: there are two critical energies, at the top and bottom of the 
78: band, at which the Localization-Delocalization (LD) transition 
79: occurs. As the degree of disorder is increased, these two critical 
80: energies approach, and finally meet. At this point, all the 
81: eigenmodes are localized and the system becomes an electrical 
82: insulator. This transition is termed the Metal-Insulator 
83: Transition (MIT) \cite{MacKinnon_81}. Off-diagonal disorder produces fundamentally
84: different behaviour: at no level of disorder are all the eigenmodes
85: localized, and hence there is no MIT \cite{Cain_99}.
86: 
87: Our approach to the problem of vibrational localization has been numerical, 
88: applying high-perfomance computers to the task of obtaining the eigenmodes. 
89: The Anderson
90: electron Hamiltonian can be expressed in a site basis, giving a
91: sparse matrix representation of the problem, for which the eigenvectors
92: can then be 
93: found by using standard Lanczos methods. 
94: Modern computers can solve such eigenproblems for many millions of 
95: atomic sites. A bigger problem is how to recognise quantitatively
96: the difference between localized and extended states.
97: 
98: There are several methods for distinguishing extended from
99: localized states, e.g. by looking at the properties of the Hamiltonian, 
100: such as the transfer matrix method 
101: \cite{MacKinnon_81,Pichard_81}, 
102: observing differences in the level-spacing statistics \cite{Carpena_99},
103: the Thouless criterion \cite{Thouless_72}, or by looking at
104: the eigenstates themselves. The latter is not trivial though, since
105: as the critical energy is approached from the localized regime, 
106: the localization length diverges. Thus, for a finite system size, 
107: the eigenmodes quickly become extended over a larger range than 
108: the system size and it becomes difficult to assess whether a state 
109: is truly localized or extended. These states are known as 
110: prelocalized states \cite{Mirlin_00:review}, and to characterize
111: these as localized or extended, we can use multifractal analysis (MFA)
112: \cite{Janssen_94}. 
113: 
114: It has been suggested that the eigenmode at exactly the LD critical 
115: energy will show multifractal characteristics 
116: \cite{Schreiber_91}. The standard way of characterising the 
117: multifractality is the singularity spectrum, which has been shown 
118: for electrons \cite{Grussbach_95} to be universal for an isotropic
119: system (see Ref.\cite{Milde_97} for treatment of an anisotropic system) 
120: and independent 
121: of the probability distribution of the disorder. 
122: The analytic predictions for the singularity spectrum \cite{Wegner_89}, 
123: based on the $d=2+\epsilon$ expansion of the non-linear $\sigma$ model, 
124: are in good agreement with numerics \cite{Grussbach_95}.
125: 
126: The aim of this paper is two-fold. Firstly, we use MFA in order to 
127: identify the threshold energy of the LD transition for different
128: degrees of force-constant disorder and thus obtain the ``phase diagram'' 
129:  in the frequency-disorder
130: plane for vibrational excitations in disordered models.
131: Secondly, we demonstrate the universal features of the multifractal
132: critical states at the LD transitition for the vibrational problem.
133: 
134: We can use the idea of critical multifractality to determine whether  
135: the states are extended or localized by looking at how the 
136: singularity spectrum, characterising the eigenmode, changes with 
137: simulation-system size. For a true multifractal state, assuming
138: that finite-size effects are small, the 
139: singularity spectrum will not depend on the simulation box size, 
140: whereas the spectra for states on either side of the LD 
141: transition will vary. Hence, by calculating the singularity 
142: spectrum for different system sizes, we can locate the critical 
143: energy \cite{Grussbach_95}. 
144: 
145: %\section{Vibrations}
146: The harmonic vibrational problem that is addressed in this Letter can be 
147: formulated in a very similar way to the Anderson electron problem \cite{Elliott_74}.
148: For vibrations, the equivalent to the electronic Hamiltonian 
149: is the symmetric dynamical operator: 
150: \begin{equation}
151: \label{dynmat}
152: {\hat D}=\sum\limits_{(i\alpha)(j\beta)}
153: D_{(i\alpha)(j\beta)}
154: (\left| i,\alpha\right\rangle - \left|j,\alpha\right\rangle)
155: (\left\langle i,\beta\right| - \left\langle j,\beta \right|)~,
156: \end{equation}
157: with $\left| i,\alpha \right\rangle$ being the site basis 
158: describing the displacement of atom $i$ ($i=1\ldots N$) along the Cartesian 
159: direction $\alpha$ ($\alpha=1\ldots d$, with $d$ the 
160: dimensionality). The matrix elements
161: $ D_{(i\alpha)(j\beta)} 
162: = (\kappa_{ij}/2) \left( 
163: {\hat r}_{ij} 
164: \right)_\alpha  
165:  \left({\hat r}_{ij}
166: \right)_\beta $ 
167: are defined in terms of force constants $\kappa_{ij}$, and unit vectors, 
168: ${\hat r}_{ij}$, connecting the atoms $i$ and $j$ (for simplicity, 
169: all masses are taken to be equal, $m_i=1$).  The dynamical matrix 
170: consists of $ d\times d$ blocks with strong lattice symmetry-dictated 
171: correlations inside the blocks. Additionally, 
172: all the elements of the 
173: on-diagonal blocks are the sums (with opposite sign) of the 
174: similar elements of off-diagonal blocks, 
175: reflecting the sum-rule correlations in the dynamical matrix. Therefore, 
176: in the case of nearest-neighbor interactions considered below, the
177: number of correlations between the elements in the dynamical matrix
178: is comparable with the number of independent random variables.
179: 
180: There are three main differences between the Anderson electron and 
181: the vibrational problems. Firstly, when there are no negative 
182: values of $\kappa_{ij}$, the system is mechanically stable, so there
183: are no negative eigenvalues, unlike the electron case. 
184: Secondly, the basic Anderson formulation 
185: gives a symmetric band structure. The vibrational case is  asymmetric 
186: for the third reason: there are $d$ zero-frequency modes that cannot 
187: be localized since they correspond to bulk translational 
188: displacements of the system (Goldstone modes). Since the lower bound
189: of the spectrum is therefore constrained to be extended in character,
190: we expect in a single-band model that there will be only one LD 
191: transition near the high-frequency band edge.
192:  
193: %\section{Models}
194: There are two major classes of model which can be used in studying 
195: localization: structures based on an underlying crystalline lattice 
196: with introduced disorder, and structures which have been 
197: created in an effort to recreate the distribution of atomic positions 
198: and bond angles found in real amorphous materials. 
199: For our study, we have chosen to analyse lattice models from the first class, 
200: with an underlying f.c.c. geometry and with the $\kappa_{ij}$ in 
201: Eq.~(\ref{dynmat})
202: taken from a probability distribution $\rho(\kappa)$. 
203: This is one of the 
204: simplest models and can be easily compared with the established 
205: results for the electron-localization problem for similar models. 
206: The distribution $\rho(\kappa)$ has been chosen to be a uniform (box) distribution, centered at 
207: $\kappa_0=1$ with a full width $\Delta < 2\kappa_0$ in order to give 
208: both a simple random 
209: distribution and one where there are no negative force constants.
210: Our models are face-centered cubic and range in size from $L=16$ with 4096 atoms
211: up to $L=48$ 
212: with $N= 110592$ atoms. 
213: 
214: 
215: %\section{Multifractal Analysis}
216: 
217: A multifractal is a generalization of a standard geometric fractal
218: for the case
219: when a single fractal dimension cannot characterize the
220: system \cite{Peitgen_92:book}. For each point in our measure, we can define
221: a value $\alpha(r)$ that describes the scaling of the
222: measure with $L$ around that point. If we now take
223: the set of all points with a specific $\alpha$, that
224: itself is a fractal, with dimensionality $f(\alpha)$.
225: The curve $f(\alpha)$ is known as the multifractal spectrum, or singularity spectrum, and can be used to characterize eigenmodes as localized or extended, as
226: shown below.
227: To calculate the scaling exponents, we define the measure
228: \(P_i(L_{b})=\sum _{j \in \text{Box}_i(L_b)} \left| \mathbf{u_j}\right| ^{2} \)
229: as the sum of the squared displacements $\left| \mathbf{u_j}\right|^2$ of
230: all the atoms $j$ within the $i$th box of size \( L_{b} \le L \) for a 
231: particular eigenmode, and examine
232: how this measure scales with $L_b$, or equivalently, with $\lambda=L_b / L$. 
233: We split our system up into $N(\lambda)$ boxes which completely and exactly
234: cover the system, so that $N(\lambda)=\lambda^{-d}$. 
235: The standard normalization of the eigenmodes 
236: leads to a scaling law for the measure of
237: the form \( \left\langle P(L_{b})\right\rangle _{L}\propto \lambda ^{d} \), 
238: averaging over all boxes.
239: 
240: The assumption underlying multifractal analysis is that, for a finite
241: interval of \( \lambda  \), the $q$th moments of the \( P(L_{b}) \)
242: also scale with power laws:
243: $\left\langle P^{q}(L_{b})\right\rangle _{L}\propto \lambda ^{d+\tau (q)}$
244: where \( \tau (q) \) is independent of \( \lambda  \). The range
245: of \( \lambda  \) in our case has a lower bound at the interatomic
246: spacing, since we are dealing with a discrete rather than a continuous
247: system. The upper bound $L/2$ is dictated by finite-size effects.
248: In the thermodynamic limit, as \( L\rightarrow \infty  \) ($\lambda\rightarrow 0$), 
249: the states which satisfy the multifractal condition are only found exactly at the critical energy,
250: and thus the exponents are defined uniquely as
251: 
252: \begin{equation}
253:  \label{tau}
254: \tau (q)=\lim _{\lambda \rightarrow 0}\frac{\ln (\left\langle P^{q}(L_{b})\right\rangle _{L})}{\ln \lambda }-d.
255: \end{equation}
256: 
257: In practice, $\tau (q)$ is found by performing a linear regression of the
258: calculated exponents with \( \ln \lambda  \). From this we can obtain
259: the singularity spectrum, \( f(\alpha ) \), where \( \alpha  \) is
260: defined as $\alpha (q)=\textrm{d}\tau(q)/\textrm{d} q$
261: and \( f(\alpha ) \) is obtained from the Legendre transformation
262: of \( \tau (q) \), \( f(\alpha (q))=\alpha (q)q-\tau (q) \).
263: Calculation of the singularity spectrum using the Legendre transformation
264: suffers from numerical errors, so it is more convenient to calculate
265: \( f(\alpha ) \) as a function of \( P(L_{b}) \) explicitly \cite{Peitgen_92:book}:
266: \begin{equation}
267: \label{ParametricAlpha}
268: \alpha (q)=\frac{1}{\ln \lambda }\sum _{\textrm{boxes}}\frac{P_{i}^{q}(L_{b})}{Z(q,L_{b})}\ln P_{i}(L_{b}),
269: \end{equation}
270: \begin{equation}
271: \label{ParametricF}
272: f(q)=\frac{1}{\ln \lambda }\sum _{\textrm{boxes}}\frac{P^{q}_{i}(L_{b})}{Z(q,L_{b})}\ln P_{i}^{q}(L_{b}),
273: \end{equation}
274: where $Z(q,L_b)=\sum_{\textrm{boxes}} P^{q}_i(L_b) $.
275: 
276: Since we cannot take the limit $\lambda \rightarrow 0$ in Eqs.(\ref{ParametricAlpha})-(\ref{ParametricF}), the values of \( f(q) \) 
277: and \( \alpha (q) \) are calculated
278: by performing a linear regression of the respective sums with respect to
279: \( \ln \lambda \). The linearity of these graphs is a good check of the 
280: multifractal nature of the measure.
281: 
282: 
283: %, and is shown for $\alpha(0)$ in 
284: %Fig. \ref{alpharegs}.
285: 
286: 
287: Care has to be taken over the box sizes used in the analysis.
288: For example, taking the box to include just one atomic site proved to 
289: skew the regression, as did taking the box size be that of the entire
290: system. The reason for the former is that the multifractality must break
291: down at some point, certainly for box sizes on the order of the atomic
292: spacing. Finite-size effects account for the discrepancy for the largest
293: box size.
294: 
295: The singularity spectra of the eigenmodes around the critical energy fluctuate
296: strongly, and so it becomes necessary to take an average. Ideally, we would
297: like to average over different realizations of disorder, but in practice this
298: is only realistic for the smaller size models. For larger models, we take the
299: computationally cheaper option of averaging consecutive eigenmodes, which can
300: be obtained simply in the Lanczos algorithm. In order to reduce errors, we have
301: used the gliding-box method, averaging over all possible origins when dividing
302: the system into boxes \cite{Peitgen_92:book}.
303: 
304: %\section{Locating the edge}
305: Once we have the spectra, we can find the frequency at which there is no change
306: with system size to locate the mobility edge. 
307: Empirically, it was noted \cite{Milde_97} that, for the Anderson
308: case, a plot
309: of \( \alpha(q) \) against \( (\log{L})^{-1} \) gave a good linear fit with
310: a different sign of gradient $g(\omega^2)=\textrm{d}\alpha(q;\omega^2)/\textrm{d}(log{L})^{-1}$ 
311: on either side of the critical frequency. The same
312: holds true for vibrational models, as clearly demonstrated in 
313: Fig. \ref{regressions}. We have therefore performed a 
314: linear regression on these curves, and the gradients of these lines have been
315: plotted at different energies to find the point where the singularity
316: spectrum is size independent, at which $g(\omega^2)$ crosses the abscissa. 
317: We can get additional information by looking at different values of $q$. 
318: In practice, since $\alpha(q)$ is strongly correlated for similar $q$, we
319: have looked at the representative values $q=0$ and $1$, 
320: for which the $g(\omega^2)$ have opposite
321: signs (see Fig. \ref{gradsvse}). 
322: 
323: \begin{figure}
324: \centerline{\includegraphics[width=\columnwidth]
325: {fits-1.5.eps}} 
326: \caption{\label{regressions}Estimation of the localization edge for $\Delta=1.5$. 
327: Each line is 
328: at a different frequency, from $\omega^2=9.3$ at the bottom to $9.5$ at the top in steps of $0.02$. The critical frequency is that at which this line has
329: zero gradient.
330: Notice the bold line shown, with approximately zero gradient, 
331: corresponds to $\omega^2=9.4$ and is at $\alpha(0)=4.0$}
332: \end{figure}
333: 
334: \begin{figure}
335: \centerline{\includegraphics[width=\columnwidth]
336: {gradients-vs-e.eps}} 
337: \caption{\label{gradsvse}Plot of $g=\textrm{d}\alpha(q)/\textrm{d}(log{L})^{-1}$ for $q=0$ and $1$.
338: The squared critical frequency $\omega_*^2$ is given by the zero-crossing point of the graph.
339: In this case, $\omega_*^2$ is between 9.4 and 9.44. } 
340: \end{figure}
341: 
342: \begin{figure}
343: \centerline{\includegraphics[width=\columnwidth]
344: {PhaseDiagram.eps}} 
345: \caption{\label{phasediagram} 
346: Phase diagram showing the boundary between 
347: extended and localized states. 
348: The VDOS for the crystal and the lattice with $\Delta=2.0$ 
349: have also been plotted to show the location of 
350: the mobility edge, $\omega_*^2$, within the band tail. 
351: The band edge calculated with CPA is also shown for reference.}
352: \end{figure}
353: 
354: \begin{figure}
355: \centerline{\includegraphics[width=\columnwidth]
356: {vib-critical.eps}} 
357: \caption{\label{critspecs} Critical spectra for the force-constant disordered 
358:  models. 
359: The parabolic approximation (PA) to Wegner's result 
360: \cite{Wegner_89} is shown for comparison.} \end{figure}
361: 
362: 
363: 
364: Initially, the analysis was undertaken throughout the acoustic band. We did 
365: not expect
366: to find localization at the lower (zero-frequency) band edge \cite{Taraskin_01:PRL}, and indeed 
367: it was found that there was only one LD phase transition, located in the 
368: far high-energy band tail. The band edge calculated within the coherent potential
369: approximation (CPA) was found to be quite close to the true localization
370: threshold, as can be seen in Fig.\ref{phasediagram}, and therefore it can be 
371: used as a rough estimate for the frequency
372: of the actual LD transition .
373: 
374: Having found the mobility edge for several values of force-constant 
375: disorder $\Delta$, we
376: can plot these to produce a 'phase diagram' of the eigenmodes. This is shown
377: alongside the VDOS and the CPA band edge in Fig.\ref{phasediagram}. As the
378: localization edge is in the band tail and we are limited to finite-size 
379: systems, few states are localized. With increasing $\Delta$, the mobility edge
380: decreases in frequency with respect to the CPA band edge. However, since the band
381: is broadening, the result is that the critical frequency actually increases 
382: with disorder, %It can also be seen that the mobility edge,
383: %although coming down in frequency relative
384: %to the band edge, numerically increases with $\Delta$, 
385: and thus there is no vibrational analogue to the electronic MIT. Similar 
386: behaviour of the mobility
387: edge with disorder can be seen in the phase diagram of the Anderson model with
388: off-diagonal disorder \cite{Cain_99}. 
389: 
390: For each degree of disorder, we obtain a new critical spectrum which
391: is constant for each size. These critical spectra have been
392: plotted in Fig.\ref{critspecs}, showing that for positive values of q, i.e. the left
393: hand side of the graph, all the spectra fit onto a master curve. The parabolic
394: approximation (PA) to Wegner's analytic result \cite{Wegner_89} 
395:  is one which goes through the 
396: critical
397: points $f(\alpha=4)=3$ and $f(\alpha=2)=2$, where the latter corresponds to
398: the information dimension of the eigenmode. 
399: This PA has also been plotted on the graph
400: for comparison. Note that the Wegner result is for the electronic Anderson model, 
401: yet it still fits well with the vibrational data, indicating a universality
402: across the two different systems. 
403: The
404: large error bars at high $\alpha$ are in the region where $q$ is negative, 
405: where $f$ and $\alpha$ are strongly dependent on the smallest values of the
406: measure and where the errors in the eigenmodes themselves are largest.
407: 
408: %The singularity spectrum is a peak-shaped function, with a maximum value
409: %of \( f=3 \), the Euclidean dimension of the system. It can be approximated
410: %with a quadratic function, known as the Parabolic Approximation (PA) which 
411: %corresponds to a broad, log-normal distribution of \( P_{i}(L_{b}) \). 
412: %\section{Conclusions}
413: %We have calculated the singularity spectrum for the eigenmodes of a
414: %disordered vibrational hamiltonian, and verified that an energy can be found
415: %at which the spectrum is system size independent. This can be used to obtain
416: %an estimate of the frequency of the mobility edge at which the transition from
417: %localized to delocalized eigenstates occurs. The calculated spectrum at this
418: %point fits within the error bars to the universal spectrum calculated by Wegner
419: %for the Anderson model. There is only one LD transition, which occurs towards
420: %the top of the single band we have analysed. The trajectory of the mobility
421: %edge is up in energy with increasing disorder up to at least the point where
422: %springs with unphysical negative force constants are introduced. Thus there
423: %can be no metal insulator transition. 
424: To conclude, we have investigated the localization phenomenon for vibrational
425: excitations in disordered structures, using an f.c.c. lattice model with force-constant
426: disorder for analysis. Using MFA, we have confirmed the existence of only one 
427: LD transition in the acoustic band, and found the energy at which it occurs for different degrees
428: of disorder. The eigenmodes at the threshold have been shown to be multifractal
429: states exhibiting a quantitatively similar distribution function to that of the
430: critical states in the electron Anderson model.
431: 
432: We are grateful to R.R\"omer for supplying us with MFA code \cite{romercode}, 
433: and to M.Schreiber for instructive communications.
434: 
435: \begin{thebibliography}{10}
436: \expandafter\ifx\csname bibnamefont\endcsname\relax
437:   \def\bibnamefont#1{#1}\fi
438: \expandafter\ifx\csname bibfnamefont\endcsname\relax
439:   \def\bibfnamefont#1{#1}\fi
440: \expandafter\ifx\csname url\endcsname\relax
441:   \def\url#1{\texttt{#1}}\fi
442: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
443: \expandafter\ifx\csname bibinfo\endcsname\relax \def\bibinfo#1#2{#2}\fi
444: \expandafter\ifx\csname eprint\endcsname\relax \def\eprint#1{#1}\fi
445: 
446: \bibitem{Anderson_58}
447: \bibinfo{author}{\bibfnamefont{P.~W.} \bibnamefont{Anderson}},
448:   \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{109}},
449:   \bibinfo{pages}{1492} (\bibinfo{year}{1958}).
450: 
451: \bibitem{Kramer_93}
452: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Kramer}} \bibnamefont{and}
453:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{MacKinnon}},
454:   \bibinfo{journal}{Rep. Prog. Phys.} \textbf{\bibinfo{volume}{56}},
455:   \bibinfo{pages}{1469} (\bibinfo{year}{1993}).
456: 
457: \bibitem{Schreiber_96:Book}
458: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Schreiber}}, in
459:   \emph{\bibinfo{booktitle}{Computational Physics}}, edited by
460:   \bibinfo{editor}{\bibfnamefont{M.~S.} \bibnamefont{K.H.Hoffmann}}
461:   (\bibinfo{publisher}{Springer}, \bibinfo{year}{1996}), pp.
462:   \bibinfo{pages}{147--165}.
463: 
464: \bibitem{Janssen_98:review}
465: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Janssen}},
466:   \bibinfo{journal}{Phys. Rep.} \textbf{\bibinfo{volume}{295}},
467:   \bibinfo{pages}{1} (\bibinfo{year}{1998}).
468: 
469: \bibitem{MacKinnon_81}
470: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{MacKinnon}} \bibnamefont{and}
471:   \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Kramer}},
472:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{5}},
473:   \bibinfo{pages}{1546} (\bibinfo{year}{1981}).
474: 
475: \bibitem{Cain_99}
476: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Cain}},
477:   \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Romer}}, \bibnamefont{and}
478:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Schreiber}},
479:   \bibinfo{journal}{Ann. Phys. (Leipzig)}
480:   \textbf{\bibinfo{volume}{8}}, \bibinfo{pages}{507}
481:   (\bibinfo{year}{1999}).
482: 
483: \bibitem{Pichard_81}
484: \bibinfo{author}{\bibfnamefont{J.~L.} \bibnamefont{Pichard}} \bibnamefont{and}
485:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sarma}}, \bibinfo{journal}{J.
486:   Phys. C} \textbf{\bibinfo{volume}{14}}, \bibinfo{pages}{L127}
487:   (\bibinfo{year}{1981}).
488: 
489: \bibitem{Carpena_99}
490: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Carpena}} \bibnamefont{and}
491:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Bernaola-Galvan}},
492:   \bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{60}},
493:   \bibinfo{pages}{201} (\bibinfo{year}{1999}).
494: 
495: \bibitem{Thouless_72}
496: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Edwards}} \bibnamefont{and}
497:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Thouless}},
498:   \bibinfo{journal}{J. Phys. C} \textbf{\bibinfo{volume}{5}},
499:   \bibinfo{pages}{807} (\bibinfo{year}{1972}).
500: 
501: \bibitem{Mirlin_00:review}
502: \bibinfo{author}{\bibfnamefont{A.~D.} \bibnamefont{Mirlin}},
503:   \bibinfo{journal}{Phys. Rep.} \textbf{\bibinfo{volume}{326}},
504:   \bibinfo{pages}{259} (\bibinfo{year}{2000}).
505: 
506: \bibitem{Janssen_94}
507: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Janssen}},
508:   \bibinfo{journal}{Int. J. Mod. Phys. B} \textbf{\bibinfo{volume}{8}},
509:   \bibinfo{pages}{943} (\bibinfo{year}{1994}).
510: 
511: \bibitem{Schreiber_91}
512: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Schreiber}} \bibnamefont{and}
513:   \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Grussbach}},
514:   \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{67}},
515:   \bibinfo{pages}{607} (\bibinfo{year}{1991}).
516: 
517: \bibitem{Grussbach_95}
518: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Grussbach}} \bibnamefont{and}
519:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Schreiber}},
520:   \bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{51}},
521:   \bibinfo{pages}{663} (\bibinfo{year}{1995}).
522: 
523: \bibitem{Milde_97}
524: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Milde}},
525:   \bibinfo{author}{\bibfnamefont{R.~A.} \bibnamefont{R\"omer}},
526:   \bibnamefont{and}
527:   \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Schreiber}},
528:   \bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{55}},
529:   \bibinfo{pages}{9463} (\bibinfo{year}{1997}).
530: 
531: \bibitem{Wegner_89}
532: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Wegner}},
533:   \bibinfo{journal}{Nucl. Phys. B} \textbf{\bibinfo{volume}{316}},
534:   \bibinfo{pages}{663} (\bibinfo{year}{1989}).
535: 
536: \bibitem{Elliott_74}
537: \bibinfo{author}{\bibfnamefont{R.~J.} \bibnamefont{Elliott}},
538:   \bibinfo{author}{\bibfnamefont{J.~A.} \bibnamefont{Krumhansl}},
539:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.~L.} \bibnamefont{Leath}},
540:   \bibinfo{journal}{Rev. Mod. Phys.} \textbf{\bibinfo{volume}{46}},
541:   \bibinfo{pages}{465} (\bibinfo{year}{1974}).
542: 
543: \bibitem{Peitgen_92:book}
544: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Peitgen}},
545:   \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Jurgens}}, \bibnamefont{and}
546:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Saupe}},
547:   \emph{\bibinfo{title}{Chaos and Fractals - New Frontiers of Science}}
548:   (\bibinfo{publisher}{Springer-Verlag}, \bibinfo{year}{1992}).
549: 
550: \bibitem{Taraskin_01:PRL}
551: \bibinfo{author}{\bibfnamefont{S.~N.} \bibnamefont{Taraskin}},
552:   \bibinfo{author}{\bibfnamefont{Y.~L.} \bibnamefont{Loh}},
553:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Natarajan}},
554:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{S.~R.}
555:   \bibnamefont{Elliott}}, \bibinfo{journal}{Phys. Rev. Lett.}
556:   \textbf{\bibinfo{volume}{86}}, \bibinfo{pages}{1255} (\bibinfo{year}{2001}).
557: 
558: \bibitem{ARPACK}{We have used the package ARPACK (http://www.caam.rice.edu/software/ARPACK) with sparse matrix inversion routines from the HSL (http://hsl.rl.ac.uk/)}
559: 
560: \bibitem{romercode}{Both the German MFA and our own MFA code gave practically
561: identical results.}
562: 
563: \end{thebibliography}
564: 
565: \end{document}
566: 
567: 
568: 
569: 
570: 
571: 
572: 
573: 
574: 
575: 
576: 
577: 
578: 
579: