1: \documentclass[aps,prb,twocolumn,showpacs,twoside,amsmath, amsfonts]{revtex4}
2: %\usepackage{graphicx}
3: \usepackage{psfrag}
4: \usepackage{graphicx}
5: \usepackage{epsfig}
6: \usepackage{amssymb}
7:
8:
9: %\newcommand{\mybf}[1]{{\bf #1}}
10: %\renewcommand{\vec}{\mybf}
11: \begin{document}
12: \title{One-dimensional Anderson Localization: Devil's Staircase of Statistical Anomalies.}
13: \author{V.E.Kravtsov$^{1,2}$ and V.I.Yudson$^{3}$}
14: \affiliation{$^{1}$The Abdus Salam International Centre for
15: Theoretical Physics, P.O.B. 586, 34100 Trieste, Italy.\\
16: $^{2}$Landau Institute for Theoretical Physics, 2 Kosygina
17: st.,117940 Moscow, Russia.\\$^{3}$Institute for Spectroscopy,
18: Russian Academy of Sciences, 142190 Troitsk, Moscow reg., Russia.}
19:
20: \date{today}
21: \begin{abstract}
22: The statistics of wavefunctions in the one-dimensional (1d) Anderson
23: model of localization is considered. It is shown that at any energy
24: that corresponds to a rational filling factor $f=\frac{p}{q}$ there
25: is a statistical anomaly which is seen in expansion of the
26: generating function (GF) to the order $q-2$ in the disorder
27: parameter. We study in detail the principle anomaly at
28: $f=\frac{1}{2}$ that appears in the leading order. The
29: transfer-matrix equation of the Fokker-Planck type with a
30: two-dimensional internal space is derived for GF. It is shown that
31: the zero-mode variant of this equation is integrable and a solution
32: for the generating function is found in the thermodynamic limit.
33: \end{abstract}
34: \pacs{72.15.Rn, 72.70.+m, 72.20.Ht, 73.23.-b}
35: \keywords{localization, mesoscopic fluctuations} \maketitle
36:
37: --{\it Introduction.} Anderson localization (AL) enjoys an unusual
38: fate of being a subject of advanced research during a half of
39: century. The seminal paper by P.W.Anderson \cite{Anderson} opened up
40: a direction of research on the interplay of quantum mechanics and
41: disorder which is of fundamental interest up to now
42: \cite{Mirlin2008}. The one-dimensional tight-binding model with
43: diagonal disorder --the Anderson model (AM)-- which is the simplest
44: and the most studied model of this type, became a paradigm of AL:
45: \begin{equation}
46: \label{Ham}
47: H=\sum_{i}\varepsilon_{i}\,c^{\dagger}_{i}c_{i}-\sum_{i}t_{i}\left(\,c^{\dagger}_{i}c_{i+1}+
48: c^{\dagger}_{i+1}c_{i}\right).
49: \end{equation}
50: In this model the hopping integral is deterministic $t_{i}=t$ and
51: the on-site energy $\varepsilon_{i}$ is a random Gaussian variable
52: uncorrelated at different sites and characterized by the variance
53: $\langle(\delta\varepsilon_{i})^{2}\rangle=w$.
54: %distribution function:
55: %\begin{equation}
56: %\label{Gauss} {\cal P}(\varepsilon_{i})=\frac{1}{\sqrt{2\pi
57: %w}}\,{\rm exp}\left[{-\frac{\varepsilon_{i}^{2}}{2w}}\right].
58: %\end{equation}
59: The dimensionless parameter $\alpha^{2}=w/t^{2}$ describes the
60: strength of disorder.
61:
62: The best studied is the continuous limit of this model in which the
63: lattice constant $a\rightarrow 0$ at $ta^{2}$ remaining finite
64: \cite{Ber, AR, Mel, Kolok}. There was also a great deal of activity
65: \cite{1dRev, Pastur} aimed at a rigorous mathematical description of
66: 1d AL. However, despite considerable efforts invested, some subtle
67: issues concerning 1d AM still remain unsolved. One of them is the
68: effects of commensurability between the de-Broglie wavelength
69: $\lambda_{E}$ (which depends on the energy $E$) and the lattice
70: constant $a$. The parameter that controls the commensurability
71: effects is the filling factor $f=\frac{2a}{\lambda_{E}}$ (fraction
72: of states below the energy $E$).
73:
74: It was known for quite a while \cite{Derrida} that the Lyapunov
75: exponent (which is essentially the inverse localization length)
76: takes anomalous values at the filling factors equal to $\frac{1}{2}$
77: and $\frac{1}{3}$ (compared to those at filling factors $f$ beyond
78: the window of the size $\alpha^{2}\ll 1$ around $f=\frac{1}{2}$ and
79: $f=\frac{1}{3}$, see Fig.1). Recently \cite{Titov, AL} it was found
80: that the statistics of conductance in 1d AM is anomalous at the
81: center of the band that corresponds to the filling factor
82: $f=\frac{1}{2}$. We want to stress that all these anomalies were
83: observed for the AM Eq.(\ref{Ham}) in which the on-site energy
84: $\varepsilon_{i}$ is random. This Hamiltonian does not possess the
85: {\it chiral symmetry} \cite{Dyson, Mirlin2008} which is behind the
86: statistical anomalies at the center of the band $E=0$ in the {\it
87: Lifshitz model} described by Eq.(\ref{Ham}) with the deterministic
88: $\varepsilon_{i}=0$ and a random hopping integral $t_{i}$. Thus the
89: statistical anomalies in the 1d AM raise a question about {\it
90: hidden symmetries} which do not merely reduce to the two-sublattice
91: division \cite{Dyson, AL, Mirlin2008}.
92: \begin{figure}
93: \label{fig1}
94: \includegraphics[width=6cm, height=5cm]{GF11.eps}
95: \caption{Schematic representation of statistical anomalies in the
96: localization radius. The dashed line represents the "bare"
97: localization radius $\ell_{0}=a\frac{2t^{2}}{w}\,\sin^{2}(\pi f)$;
98: the circles give the localization length at a filling factor being a
99: simple fraction and solid lines give a sketch of behavior in the
100: perturbed window.}
101: \end{figure}
102: %We also note that anomalies at the fractional values of $f$ cannot
103: %be caused by the Bragg scattering off the perfect lattice. The Bragg
104: %singularities take place when the transferred momentum $2k_{F}$ is a
105: %{\it multiple} of the inverse lattice constant $b=2\pi/a$ and not a
106: %{\it fraction} of it.
107: Finally, the {\it sign} of the anomaly is different for the center
108: of the band $f=\frac{1}{2}$ and the filling factor $f=\frac{1}{3}$.
109: All these observations point out to a new phenomenon of the devil's
110: staircase type which is essentially due to {\it disorder} and its
111: interplay with the Bragg scattering off the underlying lattice.
112:
113: In this Letter we study the {\it generating function} (GF)
114: $\Phi(u,\phi;x)$ that allows to compute {\it all} local statistical
115: properties of 1d AM. The simplest of them is the statistics of the
116: wavefunction amplitude $|\psi(x)|^{2}$ characterized by the moments
117: $I_{m}=\langle |\psi(x)|^{2m}\rangle\,\ell_{0}^{m}$:
118: \begin{eqnarray}
119: \label{moment1}
120: I_{m}&=&\frac{2}{(m-2)!}\,\int_{0}^{\pi}\frac{d\phi}{\pi}\,
121: \cos^{2m}\phi\,\\
122: \nonumber&\times&\int_{0}^{\infty}du\,u^{m-2}\,\Phi(u,\phi;x)\,
123: \Phi(u,-\phi-2\pi f;L-x),
124: \end{eqnarray}
125: where $L$ is the total length of the system and
126: $\ell_{0}=a\frac{2t^{2}}{w}\,\sin^{2}(\pi f)$ is the "bare"
127: localization length.
128:
129: We derive the corresponding transfer-matrix equation (TME) for the
130: GF
131: \begin{equation}
132: \label{TME}
133: \partial_{x} \Phi(u,\phi;x)=[\hat{L}_{f}(u,\phi)-u]\,\Phi(u,\phi;x),
134: \end{equation}
135: which {\it zero-mode} variant ($\Phi(u,\phi;x)\equiv \Phi(u,\phi)$
136: is independent of the space coordinate $x$) appears to be a partial
137: differential equation (PDE) depending on {\it two} variables. One of
138: them (denoted by $u$)is associated with the amplitude of the
139: wavefunction $\psi\sim \sqrt{u} \cos\phi$ while the other (denoted
140: by $\phi$)has a physical meaning of its phase. The well known
141: \cite{Kolok} TME in the continuous limit $f\ll 1$ can be obtained by
142: the averaging of this PDE over the phase variable $\phi$ thus
143: reducing it to an ODE in the {\it single} variable $u$.
144:
145: We show that there are statistical anomalies at {\it any} rational
146: filling factor $f=\frac{p}{q}$. Namely, the operator
147: $\hat{L}_{f}(u,\phi)$ in Eq.(\ref{TME}) expanded in the disorder
148: parameter $\alpha^{2}$
149: \begin{equation}
150: \label{expan}
151: \hat{L}_{f}(u,\phi)=\hat{L}_{f}^{(0)}(u,\phi)+\alpha^{2}\,\hat{L}_{f}^{(1)}(u,\phi)+
152: \alpha^{4}\,\hat{L}_{f}^{(2)}(u,\phi)+...
153: \end{equation}
154: is such that
155: \begin{equation}
156: \label{deltaL} \hat{L}_{f}^{(n)}=\hat{L}_{f}^{(n,{\rm
157: reg})}+\sum_{p=1}^{n+1}\Delta\hat{L}^{(n)}_{p}\,\delta\left(f,\frac{p}{n+2}\right)
158: \end{equation}
159: contains a regular part $\hat{L}_{f}^{(n,{\rm reg})}$ with a smooth
160: dependence on $f$ and an anomalous part that appears only at
161: $f=\frac{1}{2+n},\frac{2}{2+n},...\frac{n+1}{2+n}$. In the leading
162: order $(n=0)$ in $\alpha^{2}$ the anomalous term appears only at
163: $f=\frac{1}{2}$. In the next order one can observe anomalies at
164: $f=\frac{1}{3}$ and $\frac{2}{3}$, etc. Though anomalous terms
165: corresponding to the denominator $q>2$ are small at weak disorder,
166: they have an abrupt dependence on $f$. This allows to speak about
167: the "devil's staircase of anomalies".
168:
169: We study in detail the principal anomaly at $f=\frac{1}{2}$.
170: Remarkably, the corresponding zero-mode TME appears to be {\it
171: integrable}. We find a unique solution to this equation which
172: describes any local statistics of wavefunctions in the center of the
173: band.
174:
175:
176: -- {\it Derivation of the TM equation.} The starting point of our
177: analysis is the TM equation for the generating function
178: $\Phi_{j}(u,\phi)$ on the lattice site $j$:
179: \begin{equation}
180: \label{op}
181: \Phi_{j+1}(u,\phi)=\left(1+\frac{2a}{\ell_{0}}\,\left[{\cal L
182: }(u,\phi)-c_{1}(\phi)\,u\right]\right)\,\Phi_{j}(u,\phi-\pi f),
183: \end{equation}
184: where ${\cal
185: L}(u,\phi)=c_{2}(\phi)\,u^{2}\partial^{2}_{u}+c_{3}(\phi)\,(u\partial_{u}-1)+c_{4}(\phi)\,
186: u\partial_{u}\partial_{\phi}+c_{5}(\phi)\,\partial_{\phi}+c_{6}(\phi)\,\partial^{2}_{\phi}$.
187: The coefficients $c_{i}(\phi)$ are all combinations of $\cos(2\phi)$
188: and $\sin(2\phi)$ which at first glance do not show any nice
189: structure: $c_{1}(\phi)=\frac{1}{2}(1+\cos(2\phi))$,
190: $c_{2}(\phi)=1-\cos^{2}(2\phi)$,
191: $c_{3}(\phi)=-(1-\cos(2\phi)-2\cos^{2}(2\phi))$,
192: $c_{4}(\phi)=\sin(2\phi)(1+\cos(2\phi))$,
193: $c_{5}(\phi)=-\frac{3}{2}\sin(2\phi)(1+\cos(2\phi))$,
194: $c_{6}(\phi)=\frac{1}{4}(1+\cos(2\phi))^{2}$.
195:
196: This equation has been derived in Ref.\cite{OsK} by expansion to the
197: first order in $\alpha^{2}$ of the exact integral TM equation
198: obtained by the super-symmetry method \cite{Efet-book}. By
199: construction \cite{OsK} the function $\Phi_{j}(u,\phi)$ must be
200: periodic in $\phi$ with the period of $\pi$ which corresponds to the
201: phase factor $\cos\phi$ of the wave function sweeping all possible
202: values in the interval $[0,\pi]$. However, the shift in the argument
203: $\phi$ in the r.h.s. of Eq.(\ref{op}) is by a {\it fraction} $f$ of
204: $\pi$. For a rational $f=\frac{p}{q}$ one has to make $q$ iterations
205: in Eq.(\ref{op}) in order to get a closed equation for the GF. In
206: the leading order in $\alpha$ we obtain:
207: \begin{eqnarray}
208: \label{lin} &&\Phi_{j+q}(u,\phi)-\Phi_{j}(u,\phi)
209: =\frac{2a}{\ell_{0}}\\ &\times&\left[\sum_{r=0}^{q-1}{\cal
210: L}(\phi-r\,\pi p/q)-u\sum_{r=0}^{q-1}c_{1}(\phi-r\,\pi p/q)
211: \right]\,\Phi_{j}(u,\phi).\nonumber
212: \end{eqnarray}
213: %A remarkable identity that gives rise to the anomaly is the
214: %following:
215: The reason for the anomaly is the following identity that shows a
216: jump at $q=2$:
217: \begin{equation}
218: \label{id1} \sum_{r=0}^{q-1}e^{2i\phi-2i r\,\pi p/q}=0,\;\;
219: \sum_{r=0}^{q-1}e^{4i\phi-4i r\,\pi p/q }=\left\{\begin{matrix}0, &
220: q>2\cr 2e^{4i\phi},& q=2\cr
221: \end{matrix} \right.
222: \end{equation}
223: One can see from this identity that for $q>2$ the summation in
224: Eq.(\ref{lin}) is the same as averaging over $\phi$: all the
225: $\phi$-dependent terms vanish in both cases. Assuming $q\ll
226: \ell_{0}/a$, expanding the l.h.s. of Eq.(\ref{lin}) and introducing
227: the dimensionless coordinate $x=ja/\ell_{0}$ we obtain:
228: \begin{equation}
229: \label{ord}
230: \partial_{x}\Phi=\hat{L}_{f}^{(0,{\rm reg})}\Phi=\left[u^{2}\partial^{2}_{u}-u+\frac{3}{4}\partial^{2}_{\phi} \right]
231: \,\Phi.
232: \end{equation}
233: This equation admits the independent of $\phi$ {\it stationary}
234: solution:
235: \begin{equation}
236: \label{stat} \Phi(u,\phi)=e^{-\epsilon
237: x}\,\sqrt{u}\,K_{\sqrt{1-4\epsilon}}(2\sqrt{u}).
238: \end{equation}
239: This solution has been earlier obtained \cite{Kolok} in the
240: continuous limit $f\ll 1$. It also arises in the theory of a
241: multi-channel disordered wire \cite{Efet-book, Mirlin2000}. For a
242: system of the size $L\rightarrow\infty$ only zero mode solution
243: corresponding to $\epsilon=0$ is relevant. Substituting
244: Eq.(\ref{stat}) with $\varepsilon=0$ into Eq.(\ref{moment1}) we
245: found the following distribution function of the eigenfunction
246: amplitude in a long {\it strictly} one-dimensional system
247: (amazingly, this result was not known before):
248: \begin{equation}
249: \label{1d-dist} {\cal P}(|\psi|^{2})=\frac{\ell_{0}}{L}\,\frac{{\rm
250: exp}\left(-|\psi|^{2}\ell_{0}\right)}{|\psi|^{2}}.
251: \end{equation}
252: This distribution is valid for $|\psi|^{2}\ell_{0}\gg
253: e^{-L/\ell_{0}}$ and should be cut off at very small $|\psi|^{2}$ to
254: ensure normalizability \cite{rem1}.
255:
256: At $q=2$ (and only at $q=2$ in the leading order in $\alpha$) the
257: $\phi$-dependence in Eqs.(\ref{id1}),(\ref{lin}) survives and gives
258: rise to the anomalous term \cite{rem4}:
259: \begin{eqnarray}
260: \label{del-2}
261: \Delta\hat{L}^{(0)}&=&\cos(4\phi)\,\left[-u^{2}\partial^{2}_{u}+2u\partial_{u}
262: +\frac{1}{4}\partial^{2}_{\phi}-2 \right]\nonumber \\
263: &+&\sin(4\phi)\,\left[u\partial_{u}\partial_{\phi}-\frac{3}{2}\partial_{\phi}
264: \right].
265: \end{eqnarray}
266: Again, like in Eq.(\ref{op}), there is apparently no nice structure
267: in Eq.(\ref{del-2}). Moreover, because of the anomalous term the
268: entire operator $\hat{L}_{\frac{1}{2}}^{(0)}(u,\phi)$ acquires an
269: explicit $\phi$-dependence and thus the {\it zero-mode} TME becomes
270: a two-variable second-order PDE which no longer admits a
271: $\phi$-independent solution. Yet it appears exactly solvable!
272:
273: --{\it Separation of variables.} The integrability of the zero mode
274: TME for $f=\frac{1}{2}$ is shown in three steps. The step one is to
275: introduce new set of variables $u$ and $v=u\cos(2\phi)$ instead of
276: $(u,\phi)$ and a new function
277: $\tilde{\Phi}(u,v)=u^{-1}\,\Phi(u,\frac{1}{2}\arccos(v/u))$. In
278: these variables the stationary TME $[\hat{L}_{\frac{1}{2}}^{(0,{\rm
279: reg})}+\Delta\hat{L}^{(0)}]\Phi=-\epsilon \Phi$ takes a very
280: symmetric form:
281: \begin{eqnarray}
282: \label{stat-xy}
283: %&&\sqrt{u^{2}-v^{2}}\,\left\{\partial_{u}\,\,\sqrt{u^{2}-v^{2}}\,\,\partial_{u}
284: %+\partial_{v}\,\,\sqrt{u^{2}-v^{2}}\,\,\partial_{v}\right\}\,\tilde{\Phi}=\nonumber
285: [D_{1}^{2}+D_{3}^{2}]\,\tilde{\Phi}=
286: \frac{u}{2}\,\,\tilde{\Phi}-\epsilon\,\tilde{\Phi},
287: \end{eqnarray}
288: where the operators $D_{1}$ and $D_{3}$ belong to the family of
289: three operators from the representation of the $sl_{2}$ algebra:
290: \begin{equation}
291: \label{A-xy} D_{1}=\sqrt{u^{2}-v^{2}}\,\,\partial_{u},\;
292: D_{2}=u\,\partial_{v}+v\,\partial_{u},\;
293: D_{3}=-\sqrt{u^{2}-v^{2}}\,\,\partial_{v}
294: \end{equation}
295: obeying the commutation relations:
296: \begin{eqnarray}
297: \label{algebra} [D_{1},D_{2}]=-D_{3},\;
298: [D_{3},D_{1}]=D_{2},\;
299: [D_{2},D_{3}]=D_{1}.
300: \end{eqnarray}
301: Now it is clear that there is a hidden order in a set of
302: $\phi$-dependent terms in Eq.(\ref{del-2}) and the way they match
303: the regular part $\hat{L}_{f}^{(0,{\rm reg})}$ in r.h.s. of
304: Eq.(\ref{ord}).
305:
306: The next step is to transform Eq.(\ref{stat-xy}) to the
307: Schroedinger-like equation $
308: -(\partial_{u}^{2}+\partial_{v}^{2})\,\Psi+U(u,v)\,\Psi=0$ for the
309: function $\Psi(u,v)=(u^{2}-v^{2})^{\frac{1}{4}}\,\tilde{\Phi}$,
310: where
311: \begin{eqnarray}
312: \label{Schroedinger}
313: %\hat{H}\Psi&\equiv&
314: %-(\partial_{u}^{2}+\partial_{v}^{2})\,\Psi+U(u,v)\,\Psi=0,\\
315: U=-\frac{3}{4}\,\frac{u^{2}+v^{2}}{(u^{2}-v^{2})^{2}}+\frac{1}{2}\,
316: \frac{u}{u^{2}-v^{2}}-\frac{\epsilon}{u^{2}-v^{2}}.
317: \end{eqnarray}
318: Finally we introduce the variables
319: \begin{equation}
320: \label{xi-eta} \xi=\frac{u+v}{2}=u\,\cos^{2}\phi,\;\;\;\;
321: \eta=\frac{u-v}{2}=u\,\sin^{2}\phi.
322: \end{equation}
323: It is easy to see that the kinetic energy and the first two terms in
324: Eq.(\ref{Schroedinger}) become the sum of two identical Hamiltonians
325: $\hat{H}_{\xi}+\hat{H}_{\eta}$ where $\hat{H}_{\xi}$ is given by:
326: \begin{equation}
327: \label{H-1d}
328: \hat{H}_{\xi}=-\partial_{\xi}^{2}-\frac{3}{16}\,\frac{1}{\xi^{2}}+\frac{1}{4\xi}.
329: \end{equation}
330: Thus at $\epsilon=0$ we explicitly separated the variables in
331: Eq.(\ref{stat-xy}) reducing the two-dimensional PDE to two ODE's of
332: the Schredinger type
333: $\hat{H}_{\xi}\varphi_{\lambda}(\xi)=\lambda\varphi_{\lambda}(\xi)$
334: and
335: $\hat{H}_{\eta}\varphi_{-\lambda}(\eta)=-\lambda\varphi_{-\lambda}(\eta)$.
336: Each of this equations reduces to the well known Weber's
337: differential equation which solution is given in terms of the
338: hypergeometric functions (Whittaker functions) \cite{GR}.
339:
340: Albeit the above procedure does not work for $\varepsilon\neq 0$
341: because of the last term in Eq.(\ref{Schroedinger}), the
342: integrability of the zero-mode TME is a remarkable fact that allows
343: to describe anomalous statistics in an infinitely long system.
344:
345: --{\it Uniqueness of the solution.} The general solution to the
346: "Schroedinger equation" with $\epsilon=0$ in Eq.(\ref{Schroedinger})
347: is given by the integral over the parameter $\lambda$:
348: \begin{equation}
349: \label{gen} \Psi=\int d\lambda
350: d\bar{\lambda}\;c(\lambda,\bar{\lambda})\;\varphi_{\lambda}(\xi)\,\varphi_{-\lambda}(\eta),
351: \end{equation}
352: where integration is generically over the complex plane of $\lambda$
353: and $c(\lambda,\bar{\lambda})$ is an arbitrary function \cite{rem2}.
354: How does this huge degeneracy comply with the intuitive expectation
355: that the statistics of wavefunctions in an infinite disordered chain
356: should be unique and independent of the boundary conditions? Below
357: we show that the natural physical requirements on $\Phi(u,\phi)$
358: help to determine GF up to a constant factor which can be further
359: fixed using the wave function normalization $\langle
360: |\psi|^{2}\rangle=\frac{1}{L}$.
361:
362: First of all we note that
363: $F(\lambda;\xi,\eta)=\varphi_{\lambda}(\xi)\,\varphi_{-\lambda}(\eta)$
364: is a holomorphic function of $\lambda$, i.e. it depends only on
365: $\lambda=\rho e^{i\sigma}$ but not on $\bar{\lambda}=\rho
366: e^{-i\sigma}$. The idea is to represent the integral over the
367: complex plane as an integral over $\rho$ and $\sigma$ and then
368: rotate the contour of integration $\rho\rightarrow t e^{-i\sigma}$
369: so that the dependence on $\sigma$ remains only in
370: $c(\lambda,\bar{\lambda})$ and in the integration measure but not in
371: $F(\lambda;\xi,\eta)$. Then performing integration over $\sigma$ one
372: obtains a new function $C(t)=t\int
373: d\sigma\,e^{-2i\sigma}\,c(t,te^{-2i\sigma})$ which stands for
374: $c(\lambda,\bar{\lambda})$ in an expression similar to
375: Eq.(\ref{gen}) but involving only a one-dimensional {\it contour
376: integral}. This contour can be further rotated to make the
377: expression more symmetric. Thus without loss of generality we write
378: a solution to the zero-mode TM equation Eq.(\ref{stat-xy}) for
379: $f=\frac{1}{2}$:
380: \begin{eqnarray}
381: \label{canon}
382: &&\Phi(\xi,\eta)=\frac{\xi+\eta}{(\xi\eta)^{1/4}}\int_{0}^{\infty}d\lambda\,
383: C(\lambda)
384: \\
385: &\times&\left[W_{-\lambda\epsilon,\frac{1}{4}}\,\left(
386: \frac{\bar{\epsilon}\xi}{4\lambda}\right)\,W_{-\lambda\bar{\epsilon},\frac{1}{4}}\,\left(
387: \frac{\epsilon\eta}{4\lambda}\right)+ c.c\right].\nonumber
388: \end{eqnarray}
389: Here $W_{\kappa,\mu}(z)$ is the Whittaker function \cite{GR};
390: $\epsilon=e^{i\pi/4}$, $\bar{\epsilon}=e^{-i\pi/4}$, and
391: $C(\lambda)$ is a real function yet to be determined.
392:
393: Before we proceed with determining this function it is important to
394: establish its properties as $\lambda\rightarrow 0$. To this end we
395: note that the integral over $u$ in Eq.(\ref{moment1}) with $m=1$
396: must be divergent at small $u$. Indeed, if it is convergent then the
397: factor $\frac{1}{(m-2)!}=\frac{1}{\Gamma(m-1)}$ makes the first
398: moment equal to zero which contradicts the normalizability of wave
399: function $\int dx\langle |\psi(x)|^{2}\rangle=1$. At the same time
400: for $m>1$ the integral should converge to ensure finite higher
401: moments. This implies that $\Phi(u,\phi)$ must tend to a constant
402: as $u\rightarrow 0$ \cite{rem3}. Given the asymptotic behavior of
403: Whittaker functions this is equivalent to:
404: \begin{equation}
405: \label{tile-non}
406: C(\lambda)=\lambda^{-\frac{3}{2}}\;\tilde{C}(\lambda),\;\;\;\tilde{C}(0)={\rm
407: const}.
408: \end{equation}
409: GF defined by Eq.(\ref{canon}) is periodic in $\phi$ with the period
410: $\frac{\pi}{2}$ as it should be for $q=2$. This is guaranteed by the
411: adding of the {\it c.c} term in Eq.(\ref{canon}). What is not
412: automatically guaranteed is that $\Phi(\xi,\eta)$ is {\it smooth} as
413: a function of $\phi$ at $\phi=0$. We will see that it is the
414: requirement of {\it smoothness} at $\phi=0$ which fixes (up to a
415: constant factor) the unknown function $\tilde{C}(\lambda)$.
416:
417: Indeed, the discontinuity of derivatives at $\phi=0$ may arise from
418: the branching of the expression in Eq.(\ref{canon}) at a small
419: $\eta$. From the representation of the Whittaker function in terms
420: of the hypergeometric functions one concludes that the general
421: solution Eq.(\ref{canon}) is a sum of a part which is regular in the
422: vicinity of $\eta=0$ and a part which has a square-root singularity
423: $\sqrt{\eta}\approx \sqrt{u} |\phi|$. The condition that this latter
424: part cancels out in the solution Eq.(\ref{canon}) is the following
425: ($t$ is real):
426: \begin{eqnarray}
427: \label{rot} &&\Im
428: \left[\frac{\tilde{C}(\bar{\epsilon}t)}{\Gamma\left(\frac{1}{4}-it\right)}\,
429: e^{-\frac{i\eta}{8t}}\,_{1}F_{1}\left(
430: \frac{3}{4}-it,\frac{3}{2},\frac{i\eta}{4t}\right)\right]=0.
431: %\\
432: %&-&\left.\frac{\tilde{C}(\epsilon t)}
433: %{\Gamma\left(\frac{1}{4}+it\right)}\,
434: %e^{+\frac{i\eta}{8t}}\,_{1}F_{1}\left(
435: %\frac{3}{4}+it,\frac{3}{2},\frac{-i\eta}{4t}\right)
436: %\right]=0.\nonumber
437: \end{eqnarray}
438: The crucial fact for the possibility to fulfil this condition is the
439: identity for the hypergeometric functions \cite{GR}:
440: %\begin{equation}
441: %\label{gr-iden} e^{-z/2}\,_{1}F_{1}\left(\alpha,\gamma,z
442: %\right)=e^{z/2}\,_{1}F_{1}\left(\gamma-\alpha,\gamma,-z \right).
443: %\end{equation}
444: %It leads to:
445: \begin{equation}
446: \label{gr-iden-part}
447: e^{-z/2}\,_{1}F_{1}\left(\frac{3}{4}-it,\frac{3}{2},z
448: \right)=e^{z/2}\,_{1}F_{1}\left(\frac{3}{4}+it,\frac{3}{2},-z
449: \right).
450: \end{equation}
451: Now one can immediately guess the solution for $\tilde{C}(\lambda)$:
452: \begin{equation}
453: \label{GG} C_{0}(\lambda)=\Gamma\left(\frac{1}{4}+\epsilon\lambda
454: \right)\,\Gamma\left(\frac{1}{4}+\bar{\epsilon}\lambda \right).
455: \end{equation}
456:
457:
458: It is easy to see that the general solution to Eq.(\ref{rot})is
459: \begin{equation}
460: \label{gen-C}
461: \tilde{C}(\lambda)=C_{0}(\lambda)S(\lambda)=C_{0}(\lambda)\,\sum_{k=0}^{\infty}a_{k}\,\lambda^{4k},
462: \end{equation}
463: where the function $S(\lambda)$ must be regular in the entire
464: complex plane of $\lambda$. Now we apply the condition of
465: convergence of the integral over $\lambda$ in Eq.(\ref{canon}) at
466: large $\lambda$ to find the allowed asymptotic behavior of
467: $S(\lambda)$ at $\lambda\rightarrow\infty$. Substituting
468: Eq.(\ref{gen-C}) into Eq.(\ref{canon}) and using the asymptotics of
469: the Whittaker and $\Gamma$-functions we find that the integrand
470: behaves as $\lambda^{-3}S(\lambda)$ at $\lambda\rightarrow\infty$.
471: This means that $|S(\lambda)|$ should increase not faster than
472: $\lambda^{2}$. There is only one such entire function with the
473: structure of Eq.(\ref{gen-C}): this is a constant
474: $S(\lambda)=a_{0}={\rm const}$.
475:
476: --{\it Conclusion and discussion.}
477: Eqs.(\ref{canon}),(\ref{tile-non}),(\ref{GG}) is the main result of
478: the paper. They give an exact and unique (up to a constant factor)
479: solution for the generating function at $f=\frac{1}{2}$ anomaly
480: which can be used to compute all local statistics of the
481: one-dimensional Anderson model at $L\rightarrow\infty$. The
482: integrability of TME Eq.(\ref{TME}) suggests that there is a hidden
483: symmetry of the problem at $f=\frac{1}{2}$. We make a conjecture
484: that this symmetry is naturally formulated in the three dimensional
485: space rather than in the two-dimensional space $(\xi,\eta)$ and that
486: it has to do with the symmetry of the 3d harmonic oscillator. This
487: conjecture is based on an analogy between our main result
488: Eq.(\ref{canon}) and the expression for the Green's function of the
489: 3d harmonic oscillator problem \cite{BV}. This analogy concerns the
490: parameter ($\lambda$ in our problem and $k$ in Ref.\cite{BV})
491: entering both in the argument of the Whittaker functions and in its
492: first index in a mutually reciprocal way, as well as the second
493: index of the Whittaker functions being $\frac{1}{4}$ in both cases.
494: Establishing this symmetry would also be useful for studying the
495: anomalies at $f=\frac{p}{q}$ with $q>2$. We have obtained
496: \cite{rem4} the operator $\hat{L}_{f}^{(1)}(u,\phi)$ in
497: Eq.(\ref{expan}) and shown that the mechanism similar to
498: Eq.(\ref{id1}) is responsible for the anomaly at $f=\frac{1}{3}$.
499: The results of this study will be published elsewhere.
500:
501: --{\it Acknowledgement.} We appreciate stimulating discussions with A.Agrachev,
502: Y.V.Fyodorov, A.Kamenev and A.Ossipov and a support from RFBR grant
503: 06-02-16744.
504: \begin{thebibliography}{99}
505: \bibitem{Anderson} P.W.Anderson, Phys.Rev. {\bf 109}, 1492 (1958).
506: \bibitem{Mirlin2008} F.Evers and A.D.Mirlin, {\it Anderson
507: transition}, ArXiv:0707.4378 (2007).
508: \bibitem{Ber} V.L.Berezinskii, Zh.Exp.Teor.Fiz.{\bf 65}, 1251
509: (1973)[Sov.Phys.JETP {\bf 38}, 620 (1974)].
510: \bibitem{AR} A.A.Abrikosov and I.A.Ryzhkin, Adv.Phys.{\bf 27}, 147
511: (1978).
512: \bibitem{Mel} V.I.Melnikov, JETP Lett. {\bf 32}, 255 (1980).
513: \bibitem{Kolok} I.V.Kolokolov, Zh.Exp.Teor.Fiz.{\bf 103}, 2196
514: (1993)[JETP {\bf 76}, 1099 (1993)].
515: \bibitem{Pastur} I.M.Lifshitz, S.A.Gredeskul and L.A.Pastur
516: {\it Introduction to the theory of disordered systems} (Wiley, New
517: York, 1988).
518: \bibitem{1dRev} J.Frohlich, F.Martinelli, E.Scoppola and T.Spencer,
519: Comm.Math.Phys. {\bf 101}, 21 (1985).
520: \bibitem{Derrida} B.Derrida and E.Gardner, J.Phys. (Paris) {\bf 45}, 1283
521: (1984).
522: \bibitem{Titov} H.Schomerus and M.Titov, Phys.Rev.B {\bf 67},
523: 100201(R) (2003).
524: \bibitem{AL} L.I.Deych, et al. Phys.Rev.Lett. {\bf 91}, 096601
525: (2003).
526: \bibitem{Dyson} F.J.Dyson, Phys.Rev. {\bf 92}, 1331 (1958).
527: \bibitem{OsK} A.Ossipov and V.E.Kravtsov, Phys.Rev.B {\bf 73},
528: 033105 (2006).
529: \bibitem{Efet-book} K.B.Efetov, {\it Supersymmetry in chaos and
530: disorder} (Cambridge University Press, Cambridge, England, 1977).
531: \bibitem{Mirlin2000} A.D.Mirlin, Phys.Rep.{\bf 326},259 (2000).
532: \bibitem{rem1} All the positive moments of this distribution are
533: finite, in particular $\langle |\psi|^{2} \rangle=\frac{1}{L}$.
534: \bibitem{rem4} In the next order of expansion in $\alpha^{2}$ there
535: appear anomalous terms at $f=\frac{1}{3}$ and $f=\frac{2}{3}$
536: proportional to $\sin(6\phi)$ and $\cos(6\phi)$ containing up to the
537: third derivative wrt $u$ and $\phi$.
538: \bibitem{GR} I.S.Gradshtein and I.M.Ryzhik {\it Table of integrals series and products}
539: (Academic Press, New York, 1996).
540: \bibitem{rem2} Note that the "Hamiltonian" in Eq.(\ref{H-1d}) is a
541: non-Hermitean operator. This is a consequence of the singular
542: inverse-quare potential. To make it Hermitean one has to impose a
543: condition $\varphi(0)=0$ assuming a hard wall at $\xi<0$. There is
544: no such condition in our problem.
545: \bibitem{rem3} One can show that
546: $\Phi(u,\phi)=h_{0}(\phi)+\sum_{k=1}^{\infty} u^{k}\,[h_{k}(\phi)+
547: g_{k}(\phi)\,\ln u]$.
548: \bibitem{BV} V.L.Bakhrakh and S.I.Vetchinkin, Theor. Math.Phys.{\bf
549: 6},283 (1971)[Sov.Phys: Teor.Mat.Fiz.{\bf 6},392(1971)].
550: %\bibitem{SkvorPasha} M.A.Skvortsov and P.M.Ostrovsky, Pis'ma v ZhETF
551: %{\bf 85}, 79 (2007)[JETP Letters {\bf 85}, 72 (2007)].
552: \end{thebibliography}
553: \end{document}
554: