1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \catcode`\@=11
4: \topmargin 0pt
5: \oddsidemargin 0pt
6: \headheight 0pt
7: \headsep 0pt
8: \textheight 9in
9: \textwidth 6.25in
10: \marginparwidth 0.875in
11: \def\numberbysection{\@addtoreset{equation}{section}
12: \def\theequation{\thesection.\arabic{equation}}}
13: \def\baselinestretch{1.05}
14: \numberbysection
15:
16: \newcommand{\beq}{\begin{equation}}
17: \newcommand{\beqa}{\begin{eqnarray}}
18: \newcommand{\eeq}{\end{equation}}
19: \newcommand{\eeqa}{\end{eqnarray}}
20: \newcommand{\abs}[1]{\vert#1\vert}
21: \newcommand{\bigstat}[1]
22: {\left\langle\!\!\!\left\langle#1\right\rangle\!\!\!\right\rangle}
23: \renewcommand{\d}{{\rm d}}
24: \newcommand{\dstar}{{\star\star}}
25: \newcommand{\e}{{\rm e}}
26: \newcommand{\eps}{\varepsilon}
27: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}
28: \newcommand\gammaratio{\frac{\Gamma(2/3)}{\Gamma(1/3)}}
29: \renewcommand{\i}{{\rm i}}
30: \newcommand{\mean}[1]{\langle#1\rangle}
31: \newcommand{\stat}[1]{\left\langle\!\left\langle#1\right\rangle\!\right\rangle}
32: \newcommand{\Ai}{{\rm Ai}}
33: \newcommand{\Bi}{{\rm Bi}}
34: \newcommand{\E}{{\cal E}}
35: \newcommand{\I}{^{\rm I}}
36: \newcommand{\R}{^{\rm R}}
37:
38: \begin{document}
39: \centerline{\Large\bf Non-monotonic disorder-induced enhanced tunneling}
40: \vspace{1.6cm}
41: \centerline{\large J.M.~Luck\footnote{luck@spht.saclay.cea.fr}}
42: \vspace{.4cm}
43: \centerline{Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},
44: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}
45: \vspace{1cm}
46: \begin{abstract}
47: The quantum-mechanical transmission through a disordered
48: tunnel barrier is investigated analytically in the following regime:
49: (correlation range of the random potential) $\ll$ (penetration length)
50: $\ll$ (barrier length).
51: The mean and/or the width of the potential
52: can either be constant, or vary slowly across the barrier.
53: The typical transmission is found
54: to be a non-monotonic function of the disorder strength,
55: increasing at weak disorder, reaching a maximum in the crossover
56: from weak to strong disorder, and decreasing at strong disorder.
57: This work provides a quantitative analysis of the
58: phenomenon of disorder-induced enhanced tunneling,
59: put forward by Freilikher et al.
60: [Phys.~Rev.~E {\bf 51}, 6301 (1995); B {\bf 53}, 7413 (1996)].
61: \end{abstract}
62:
63: \vfill
64: \noindent To be submitted for publication to the Journal of Physics A
65:
66: \noindent P.A.C.S.: 03.65.Xp, 73.23.-b, 03.65.Nk, 73.20.Fz, 03.65.Sq.
67:
68: \newpage
69: \section{Introduction}
70:
71: The Anderson localization is one of the most spectacular
72: disorder-induced phenomena~\cite{rev}.
73: The one-dimensional situation is especially well-understood~\cite{pen}.
74: Consider for definiteness the Schr\"odinger equation
75: for an electron moving on a line, in a disordered potential~$V(x)$.
76: Even in the presence of an infinitesimal amount of disorder,
77: all eigenstates are exponentially localized,
78: with a localization length $\xi=1/\gamma$,
79: where~$\gamma$ is the Lyapunov exponent.
80: As a consequence, the typical conductance of a disordered sample
81: falls off exponentially with its length $L$.
82: More precisely, the zero-temperature conductance $g$ of a one-channel sample
83: is related to the transmission $T$ across the sample
84: by the two-probe Landauer formula~\cite{lan}:
85: \beq
86: g=\frac{2e^2}{h}\,T.
87: \label{lanfor}
88: \eeq
89: The theory of one-dimensional localization predicts that
90: the transmission $T$ is a widely fluctuating quantity
91: in the insulating regime $(\gamma L\gg1)$,
92: so that the meaningful quantity to consider
93: is its typical (i.e., most probable) value $\exp(\mean{\ln T})$.
94: The mean $\mean{\ln T}$ grows linearly with the sample length,
95: \beq
96: \mean{\ln T}\approx -2\gamma L,
97: \label{tintro}
98: \eeq
99: and the ratio $(\ln T)/L$ is self-averaging,
100: in the strong sense that all the cumulants of $(\ln T)$
101: grow linearly with $L$~\cite{pen}.
102: In other terms, the statistics of $T$ (resp.~of $\ln T$)
103: is similar to that of the partition function
104: (resp.~of the total free energy)
105: of a disordered thermodynamical system.
106: This deep analogy appears clearly in the framework of
107: the transfer-matrix formalism,
108: especially for discrete (tight-binding) models~\cite{cpv,alea}.
109:
110: So far, it was implicitly assumed that the energy $E$
111: of the incoming electron is above the mean of the disordered potential
112: (usually taken to be zero).
113: Much less is known on the converse situation
114: of {\it tunneling through a disordered barrier},
115: where the mean potential~$\mean{V(x)}$ inside the sample is non-zero,
116: and higher than the energy~$E$.
117: More generally, a tunneling situation is met
118: if the disordered sample is periodic on average,
119: and if the energy is in a gap
120: of the underlying average structure~\cite{f1}.
121:
122: It has been put forward, seemingly in~\cite{f1} for the first time,
123: that a weak disorder enhances the transmission in such a tunneling situation.
124: In their subsequent work~\cite{f2},
125: the authors show that a weak disorder increases
126: both the mean conductance (proportional to~$\mean{T}$)
127: and the mean resistance (proportional to~$\mean{1/T}$) of a tunnel barrier.
128: This disorder-induced enhanced tunneling effect is paradoxical,
129: because random impurity potentials usually lead to additional scattering,
130: which hinders transport.
131: In any case, a strong enough disordered potential is expected to have
132: the usual effect of reducing the transmission.
133: Putting together these observations, it can therefore be anticipated
134: that the transmission reaches a maximum in an intermediate crossover regime,
135: corresponding to a moderate amount of disorder.
136: This non-monotonic behavior of the transmission as a function
137: of the disorder strength seems to have been overlooked so far
138: (see, however,~\cite{hein} for attempts earlier to~\cite{f1,f2},
139: and~\cite{dey} for a recent discussion of the effects
140: of a weak disorder on gap states).
141:
142: Our aim is to provide a quantitative analysis of the non-monotonic behavior
143: of the disorder-induced enhanced tunneling transmission.
144: We restrict this study to the regime of most physical interest:
145: \beq
146: a\hbox{ (correlation range of potential)}
147: \ll 1/K\hbox{ (penetration length)}
148: \ll L\hbox{ (barrier length)}.
149: \label{regime}
150: \eeq
151: The first inequality implies that the fluctuations of the disordered potential
152: are short-ranged, so that the latter can be modeled as a Gaussian white noise.
153: The second inequality implies that the transmission of the barrier
154: is exponentially small, even in the absence of disorder.
155: This suggests that $\mean{\ln T}$ will be the right quantity to consider.
156:
157: For completeness,
158: we first give in Section~2 an overview of tunneling through a clean barrier.
159: Section~3 then deals with tunneling through a square disordered barrier
160: (the mean and the width of the potential are constant),
161: while Section~4 is devoted to the general case
162: (the mean and/or the width of the potential vary smoothly across the barrier).
163: A summary and an outlook are presented in Section~5.
164:
165: \section{Tunneling through a clean barrier}
166:
167: We begin with a reminder on the well-known problem
168: in Quantum Mechanics~\cite{ll} of tunneling through a clean barrier.
169: In reduced units ($\hbar=2m=1$),
170: the one-dimensional Schr\"odinger equation reads
171: \beq
172: -\psi''(x)+V(x)\psi(x)=E\psi(x).
173: \label{sch}
174: \eeq
175:
176: \subsubsection*{Square clean barrier}
177:
178: Consider first the simple case of a square barrier of length $L$.
179: The potential is constant in the barrier:
180: \beq
181: V(x)=V_0\qquad(0\le x\le L),
182: \eeq
183: and vanishes elsewhere.
184:
185: In the situation of interest,
186: the energy $E$ of the incoming particle is in the range $0<E<V_0$.
187: The wavevector $p$ of the particle
188: and its inverse penetration length $K$ in the barrier read
189: \beq
190: p=\sqrt{E},\qquad K=\sqrt{V_0-E}.
191: \label{defpk}
192: \eeq
193: The reflection and transmission amplitudes $r$ and $t$
194: are determined by looking for a solution to~(\ref{sch}) of the form
195: \beq
196: \psi(x)=\left\{\matrix{
197: \e^{\i px}+r\,\e^{-\i px}\hfill&&(x\le0),\hfill\cr
198: a\,\e^{Kx}+b\,\e^{-Kx}\hfill&&(0\le x\le L),\hfill\cr
199: t\,\e^{\i p(x-L)}\hfill&&(x\ge L).\hfill
200: }\right.%}
201: \label{wave}
202: \eeq
203: Expressing the continuity of $\psi(x)$ and of its derivative
204: at $x=0$ and $x=L$ provides four linear equations, whose solution yields
205: \beqa
206: &&r=\frad{(p^2+K^2)\sinh KL}{(p^2-K^2)\sinh KL+2\i pK\cosh KL},\label{r}\\
207: &&t=\frad{2\i pK}{(p^2-K^2)\sinh KL+2\i pK\cosh KL}.\label{t}
208: \eeqa
209:
210: Throughout the following, we will be mostly interested in
211: the transmission intensity coefficient (or transmission for short),
212: \beq
213: T=\abs{t}^2,
214: \eeq
215: which enters the Landauer formula~(\ref{lanfor}).
216: In the present case,~(\ref{t}) yields
217: \beq
218: T=\frac{4p^2K^2}{4p^2K^2+(p^2+K^2)^2\sinh^2KL}.
219: \eeq
220: In the regime~(\ref{regime}),
221: where the barrier length is much larger than the penetration length,
222: the transmission falls off exponentially, as
223: \beq
224: T\approx\frac{16p^2K^2}{(p^2+K^2)^2}\,\exp(-2KL).
225: \label{tcarreasy}
226: \eeq
227: All subsequent results for the transmission will be given
228: {\it with exponential accuracy}, in analogy with~(\ref{tintro}).
229: Neglecting the prefactor, we thus rewrite~(\ref{tcarreasy})~as
230: \beq
231: \ln T\approx-2KL.
232: \label{tcarre}
233: \eeq
234:
235: \subsubsection*{Arbitrary clean barrier}
236:
237: Let us now consider tunneling through an arbitrary clean barrier.
238: The potential $V(x)$ is larger than the energy $E$ for $0\le x\le L$,
239: so that the inverse penetration length reads
240: \beq
241: K(x)=\sqrt{V(x)-E}.
242: \eeq
243: We assume that the potential has a smooth profile across the barrier,
244: i.e., the length scale over which $V(x)$ or $K(x)$ varies
245: is of the order of the barrier length $L$ itself.
246:
247: The transmission can be determined along the lines of the previous case,
248: by seeking a solution to~(\ref{sch}) of the form~(\ref{wave}),
249: where $\exp(\pm Kx)$ are replaced by the two elementary
250: solutions $u(x)$ and $v(x)$,
251: with initial values $u(0)=v'(0)=1$, $u'(0)=v(0)=0$,
252: whose Wronskian reads $u(x)v'(x)-u'(x)v(x)=1$.
253: We thus obtain
254: \beq
255: t=\frac{2\i p}{p^2v(L)+\i p(u(L)+v'(L))-u'(L)}.
256: \label{tuv}
257: \eeq
258:
259: The hypothesis of a smoothly varying potential implies that it is legitimate
260: to use the well-known W.K.B. approximation~\cite{ll,wkb}.
261: Indeed, the condition for this scheme to be valid,
262: \beq
263: \frac{1}{K(x)^2}\,\frac{\d K(x)}{\d x}\sim \frac{1}{K(x)L}\ll1,
264: \eeq
265: is automatically satisfied in the regime~(\ref{regime}) of long barriers.
266: Within this framework, a basis of solutions to~(\ref{sch}) in the barrier reads
267: \beq
268: \psi_\pm(x)\sim\exp\left(\pm\int_0^x K(y)\,\d y\right)\qquad(0<x<L).
269: \label{psiwkb}
270: \eeq
271: At least one of the elementary solutions $u(x)$ and $v(x)$
272: (and generically both of them)
273: is proportional to the growing solution $\psi_+(x)$.
274: Hence~(\ref{tuv}) leads to the estimate
275: \beq
276: T\sim\frac{1}{\abs{\psi_+(L)}^2},
277: \label{tsim}
278: \eeq
279: which will be sufficient hereafter, in order to work with exponential accuracy.
280:
281: The transmission therefore again falls off exponentially:
282: \beq
283: \ln T\approx-2\int_0^L K(x)\,\d x,
284: \label{twkb}
285: \eeq
286: where the integral is the action of the classical imaginary-time trajectory,
287: or {\it instanton}, crossing the barrier at energy $E$~\cite{ll,wkb,jzj}.
288: The estimate~(\ref{twkb}) generalizes~(\ref{tcarre}) to a potential barrier
289: with an arbitrary (smooth) profile.
290:
291: \section{Tunneling through a square disordered barrier}
292:
293: We now turn to the case of a square disordered barrier of length $L$.
294: The barrier potential,
295: \beq
296: V(x)=V_0+W(x)\qquad(0\le x\le L),
297: \eeq
298: is the sum of a constant $V_0$ and of a disordered component $W(x)$
299: with zero mean.
300:
301: As stated in the Introduction, in the regime~(\ref{regime})
302: it is legitimate to model $W(x)$ as a Gaussian white noise, such that
303: \beq
304: \mean{W(x)W(y)}=2D\,\delta(x-y).
305: \label{ww}
306: \eeq
307: The estimate~(\ref{tsim}) still holds in the presence of disorder.
308: We are therefore led to consider the situation where the random potential
309: extends over the half-line $x>0$, and to investigate the growth rate
310: of the generic solution to~(\ref{sch}).
311:
312: The most salient effect of the disordered potential
313: is that the wavefunction in the barrier now changes sign many times
314: in the regime under consideration,
315: so that neither the concept of a single instanton trajectory,
316: nor the W.K.B. estimate~(\ref{psiwkb}), make sense anymore.
317: The most efficient approach to this problem is the invariant-measure method,
318: initiated long ago by Dyson~\cite{dy} and Schmidt~\cite{sch}.
319: For technical reasons,
320: the energy $E-\i0$ and the inverse penetration length $K+\i0$
321: are respectively endowed
322: with infinitesimal negative and positive imaginary parts.
323: The method consists in introducing the Riccati variable
324: \beq
325: z(x)=\frac{\psi'(x)}{\psi(x)},
326: \label{zdef}
327: \eeq
328: so that
329: \beq
330: \psi(x)=\psi(0)\exp\int_0^x z(y)\,\d y.
331: \label{psiexp}
332: \eeq
333:
334: The Schr\"odinger equation~(\ref{sch}) is equivalent to the Riccati equation
335: \beq
336: z'=K^2-z^2+W(x).
337: \label{ric}
338: \eeq
339: The key property of this equation is the following:
340: the complex random variable $z(x)$ has a well-defined
341: limit probability distribution, irrespective of the position $x$,
342: provided it is deep enough in the sample $(Kx\gg1)$,
343: and of the initial condition $z(0)$.
344: Let us denote averages with respect to this invariant measure as $\stat{...}$.
345: Equation~(\ref{psiexp}) implies that the growing solution to~(\ref{sch})
346: typically grows exponentially, as
347: \beq
348: \ln\psi_+(x)\approx\Omega x,
349: \label{omedef}
350: \eeq
351: where
352: \beq
353: \Omega=\stat{z}
354: \eeq
355: is the complex characteristic exponent.
356: When the energy variable is just below the real axis,
357: $\Omega$ splits according to~\cite{tn,alea}
358: \beq
359: \Omega(E-\i0)=\gamma(E)+\i\pi H(E).
360: \label{split}
361: \eeq
362:
363: The real part of~(\ref{split}) is the Lyapunov exponent $\gamma$.
364: Inserting the behavior~(\ref{omedef}) into the estimate~(\ref{tsim}),
365: we therefore obtain at once the prediction
366: \beq
367: \mean{\ln T}\approx -2\gamma L,
368: \label{tdis}
369: \eeq
370: recalled in the Introduction~[see~(\ref{tintro})].
371: The imaginary part of~(\ref{split}) is proportional
372: to the integrated density of states of the problem per unit length,
373: \beq
374: H(E)=\int_{-\infty}^E\rho(E')\,\d E',
375: \eeq
376: so that $1/H(E)$ is the mean distance between any two consecutive zeros
377: of $\psi_+(x)$.
378:
379: The present case of a white-noise potential has been investigated
380: by several authors~\cite{fl,h,sulem}.
381: Their findings can be recast into the following
382: formula for the characteristic exponent:
383: \beq
384: \Omega=D^{1/3}\,F(X),
385: \label{char}
386: \eeq
387: with
388: \beq
389: X=\frac{K^2}{D^{2/3}}=\frac{V_0-E}{D^{2/3}}
390: \label{xdef}
391: \eeq
392: and
393: \beq
394: F(X)=\e^{-2\i\pi/3}\frac{\Ai'(\e^{-2\i\pi/3}X)}{\Ai(\e^{-2\i\pi/3}X)}
395: =\frac{\Ai'(X)+\i\,\Bi'(X)}{\Ai(X)+\i\,\Bi(X)},
396: \label{fdef}
397: \eeq
398: where $\Ai(z)$ and $\Bi(z)$ are Airy functions~\cite{as}.
399: As a consequence of~(\ref{split}), we have
400: \beqa
401: &&{\hskip 1.4mm}\gamma=D^{1/3}\,F\R(X),
402: {\hskip 8.6mm}F\R(X)=
403: \frac{\Ai(X)\Ai'(X)+\Bi(X)\Bi'(X)}{\Ai(X)^2+\Bi(X)^2},\label{charr}\\
404: &&H=\frac{D^{1/3}}{\pi}\,F\I(X),
405: {\hskip 10mm}F\I(X)=\frac{1}{\pi\left(\Ai(X)^2+\Bi(X)^2\right)}.
406: \label{chari}
407: \eeqa
408: References~\cite{tn,fl,h,sulem} deal separately
409: with the real and imaginary parts of $\Omega$,
410: and therefore rather derive~(\ref{charr}) and/or~(\ref{chari}).
411: For completeness, we give in the Appendix
412: a self-contained derivation of~(\ref{char})--(\ref{fdef}).
413:
414: Inserting~(\ref{charr}) into~(\ref{tdis}) leads us to the prediction
415: \beq
416: \mean{\ln T}\approx -2D^{1/3}\,F\R(X)\,L,
417: \eeq
418: where the scaling variable $X$ is real and positive
419: in a tunneling situation.
420:
421: In order to emphasize the dependence of the transmission on the disorder
422: strength, we recast the above prediction as
423: \beq
424: \mean{\ln T}\approx(\ln T)_0\,G(Y),
425: \label{main}
426: \eeq
427: where $(\ln T)_0=-2KL$ is the result~(\ref{tcarre})
428: in the absence of disorder, while
429: \beq
430: Y=\frac{D}{K^3}=\frac{D}{(V_0-E)^{3/2}}=X^{-3/2}
431: \label{ydef}
432: \eeq
433: is the reduced disorder strength, and the scaling function~$G$ reads
434: \beq
435: G(Y)=Y^{1/3}\,F\R(Y^{-2/3}).
436: \label{gdef}
437: \eeq
438:
439: \subsubsection*{Weak-disorder regime}
440:
441: The weak-disorder regime corresponds to $D\ll K^3$, i.e., $X\gg1$ or $Y\ll1$.
442: The differential equation
443: \beq
444: F^2+F'=X
445: \eeq
446: obeyed by the function $F(X)$ easily yields the asymptotic expansion
447: \beq
448: F(X)=X^{1/2}-\frac{1}{4X}-\frac{5}{32X^{5/2}}+\cdots,
449: \eeq
450: This result is formally real, so that $F\R$ has the same asymptotic expansion
451: (while $F\I$ is exponentially small as $X\to+\infty$).
452: We thus obtain
453: \beq
454: G(Y)=1-\frac{Y}{4}-\frac{5Y^2}{32}+\cdots,
455: \label{gweak}
456: \eeq
457: i.e., more explicitly
458: \beq
459: \mean{\ln T}\approx2L\left(-K+\frac{D}{4K^2}+\frac{5D^2}{32K^5}\cdots\right).
460: \label{tweak}
461: \eeq
462: The first correction of order $D$
463: is in agreement with the perturbative result of~\cite{f2}:
464: the leading effect of a weak disorder is found to enhance transmission.
465:
466: \subsubsection*{Strong-disorder regime}
467:
468: Consider now the opposite regime of strong disorder
469: ($D\gg K^3$, i.e., $Y\gg1$ or $X\ll1$).
470: In this regime, the scaling function
471: \beq
472: G(Y)\approx F\R(0)\,Y^{1/3}
473: \label{gstrong}
474: \eeq
475: grows as the power $1/3$ of the strength of disorder,
476: with the explicit prefactor
477: \beq
478: F\R(0)=\frac{3^{1/3}}{2}\,\gammaratio=0.364506.
479: \eeq
480: As a consequence, we have
481: \beq
482: \mean{\ln T}\approx-2F\R(0)\,D^{1/3}\,L.
483: \label{tstrong}
484: \eeq
485: The transmission decreases with the disorder strength
486: in the strong-disorder regime, as anticipated in the Introduction.
487: The result~(\ref{tstrong}) is independent of $K$:
488: the damping due to the mean barrier $V_0$
489: becomes negligible with respect to localization effects.
490:
491: \subsubsection*{Non-monotonic crossover behavior}
492:
493: The function $G(Y)$ which enters the scaling law~(\ref{main})
494: decreases first from its value $Y(0)=1$, according to~(\ref{gweak}),
495: in the weak-disorder regime,
496: and then increases according to~(\ref{gstrong}), in the strong-disorder regime.
497: It must therefore reach a minimum,
498: somewhere in the crossover from weak to strong disorder.
499:
500: Figure~\ref{fig1} shows plots (full lines) of the scaling function $G(Y)$,
501: given by~(\ref{charr}), (\ref{gdef}).
502: This function passes through a minimum, $G(Y^\star)=0.7284$ for $Y^\star=1.695$,
503: before it crosses the value $G(Y^\dstar)=1$ for $Y^\dstar=14.168$.
504: The non-monotonic behavior of the transmission as a function
505: of the disorder strength in the crossover regime,
506: anticipated in the Introduction,
507: is therefore confirmed at a quantitative level.
508: The tunneling transmission is enhanced for a weak enough
509: reduced disorder strength ($Y<Y^\dstar$),
510: the enhancement being maximal for~$Y=Y^\star$.
511:
512: \begin{figure}[ht]
513: \begin{center}
514: \includegraphics[angle=90,width=.495\linewidth]{fig1a.eps}
515: \includegraphics[angle=90,width=.49\linewidth]{fig1b.eps}
516: \caption{\small
517: Plots of the scaling functions describing the effect of disorder
518: on tunneling transmission, against the reduced disorder strength $Y$.
519: Full lines: $G$ entering~(\ref{main}) (square barrier).
520: Dashed lines: $G_1$ entering~(\ref{exa1})
521: (parabolic barrier with parabolic disorder).
522: Dash-dotted lines: $G_2$ entering~(\ref{exa2})
523: (parabolic barrier with uniform disorder).
524: (a): Circles show the values of $Y$ where the scaling functions cross unity:
525: $Y^\dstar=14.168$, $Y_1^\dstar=11.469$, $Y_2^\dstar=6.522$.
526: (b) (enlargement): Circles show the values of $Y$ where the
527: scaling functions have their minima:
528: $G=0.7284$ for $Y^\star=1.695$,
529: $G_1=0.7331$ for $Y_1^\star=1.408$,
530: $G_2=0.7704$ for $Y_2^\star=0.982$.}
531: \label{fig1}
532: \end{center}
533: \end{figure}
534:
535: \section{Tunneling through an arbitrary disordered barrier}
536:
537: We finally turn to the general situation
538: where both the deterministic part $V_0(x)$
539: and the strength of the disordered part $W(x)$ of the potential
540: may have a smooth dependence on the position across the barrier.
541: We set
542: \beq
543: V(x)=V_0(x)+W(x)\qquad(0<x<L),
544: \eeq
545: with
546: \beq
547: K(x)=\sqrt{V_0(x)-E},
548: \eeq
549: and
550: \beq
551: \mean{W(x)W(y)}=2D(x)\,\delta(x-y).
552: \eeq
553:
554: We assume that $K(x)$ and $D(x)$ vary smoothly across the barrier,
555: and we again focus our attention onto the regime~(\ref{regime}).
556: In this situation, the Riccati variable $z(x)$
557: is approximately distributed according to the local invariant measure,
558: characterized by the parameters $K(x)$ and $D(x)$.
559: The reason why this adiabatic approach is legitimate
560: is the same as for the W.K.B. scheme in the absence of disorder:
561: the length over which parameters vary, i.e., the barrier length~$L$ itself,
562: is much larger than the characteristic length
563: of the relaxation of the distribution of the Riccati variable,
564: i.e., the localization length $1/\gamma(x)$.
565:
566: Equations~(\ref{tsim}), (\ref{psiexp}) yield the following expression
567: \beq
568: \mean{\ln T}\approx-2\int_0^L\gamma(x)\,\d x
569: \label{twd}
570: \eeq
571: for the mean logarithm of the transmission.
572: The prediction~(\ref{twd}) includes all the previous results~(\ref{tcarre}),
573: (\ref{twkb}), (\ref{tdis}) as special cases.
574: By means of~(\ref{charr}), it can be recast into the more explicit form
575: \beq
576: \mean{\ln T}\approx-2\int_0^L
577: D(x)^{1/3}\,F\R\!\left(\frac{K(x)^2}{D(x)^{2/3}}\right)\d x.
578: \label{texpli}
579: \eeq
580:
581: In the weak-disorder regime, this prediction behaves as
582: \beq
583: \mean{\ln T}\approx2\int_0^L\left(-K(x)+\frac{D(x)}{4K(x)^2}+\cdots\right)\d x,
584: \eeq
585: while in the strong-disorder regime we have
586: \beq
587: \mean{\ln T}\approx-2F\R(0)\int_0^LD(x)^{1/3}\,\d x.
588: \eeq
589: These two expressions respectively generalize~(\ref{tweak})
590: and~(\ref{tstrong}).
591:
592: To close up, let us illustrate more explicitly the result~(\ref{texpli})
593: on two examples.
594:
595: \subsubsection*{Example~1: Parabolic barrier with parabolic disorder}
596:
597: In our first example, both the deterministic potential
598: and the disorder strength have a parabolic shape:
599: \beq
600: K(x)^2=V_0(x)-E=K_0^2\,\frac{4x(L-x)}{L^2},\qquad
601: D(x)=D_0\,\frac{4x(L-x)}{L^2},
602: \label{para}
603: \eeq
604: with maximum values $K_0^2$ and $D_0$ at the center of the barrier ($x=L/2$).
605:
606: In this situation,~(\ref{texpli}) yields the scaling law
607: \beq
608: \mean{\ln T}\approx(\ln T)_0\,G_1(Y),
609: \label{exa1}
610: \eeq
611: similar to~(\ref{main}), with
612: \beq
613: (\ln T)_0=-\frac{\pi}{2}\,K_0L,
614: \label{tvar}
615: \eeq
616: \beq
617: Y=\frac{D_0}{K_0^3},
618: \eeq
619: and
620: \beq
621: G_1(Y)=\frac{2}{\pi}\,Y^{1/3}\int_0^\pi
622: F\R\left(Y^{-2/3}(\sin\theta)^{2/3}\right)(\sin\theta)^{5/3}\,\d\theta.
623: \label{g1def}
624: \eeq
625: The latter expression is obtained by setting $x=L(1+\cos\theta)/2$,
626: so that $4x(L-x)/L^2=\sin^2\theta$, with $0\le\theta\le\pi$.
627:
628: The scaling function $G_1(Y)$ has been evaluated numerically by means of the
629: integral~(\ref{g1def}), and plotted in Figure~\ref{fig1} (dashed lines).
630: Its qualitative dependence on the disorder strength $Y$
631: is similar to that of $G(Y)$,
632: with the following behavior at weak and strong disorder:
633: \beq
634: \matrix{
635: G_1(Y)\approx1-\frad{Y}{\pi}\hfill&(Y\ll1),\cr\cr
636: G_1(Y)\approx\frad{12^{1/3}\Gamma(1/3)}{5\pi}\,Y^{1/3}\approx0.390454\,Y^{1/3}
637: \qquad&(Y\gg1),}
638: \eeq
639: with a minimum $G_1(Y_1^\star)=0.7331$ for $Y_1^\star=1.408$,
640: and with $G_1(Y_1^\dstar)=1$ for $Y_1^\dstar=11.469$.
641:
642: \subsubsection*{Example~2: Parabolic barrier with uniform disorder}
643:
644: In our second example, the deterministic potential still has
645: the parabolic form~(\ref{para}),
646: while the disorder strength $D(x)=D$ is constant.
647:
648: The prediction~(\ref{texpli}) now yields the scaling form
649: \beq
650: \mean{\ln T}\approx(\ln T)_0\,G_2(Y),
651: \label{exa2}
652: \eeq
653: again with~(\ref{tvar}), and with
654: \beq
655: Y=\frac{D}{K_0^3},
656: \eeq
657: and
658: \beq
659: G_2(Y)=\frac{2}{\pi}\,Y^{1/3}\int_0^\pi
660: F\R\left(Y^{-2/3}\sin^2\theta\right)\sin\theta\,\d\theta.
661: \eeq
662:
663: The scaling function $G_2(Y)$ is also plotted in Figure~\ref{fig1}
664: (dash-dotted lines).
665: Its qualitative dependence on the disorder strength $Y$
666: is again similar to that of $G(Y)$,
667: with the following behavior at weak and strong disorder:
668: \beq
669: \matrix{
670: G_2(Y)\approx1-\frad{Y}{3\pi}\ln\frad{1}{Y}\hfill&(Y\ll1),\cr\cr
671: G_2(Y)\approx\frad{2\,3^{1/3}}{\pi}\gammaratio\,Y^{1/3}\approx0.464103\,Y^{1/3}
672: \qquad&(Y\gg1),}
673: \eeq
674: with a minimum $G_2(Y_2^\star)=0.7704$ for $Y_2^\star=0.982$,
675: and with $G_2(Y_2^\dstar)=1$ for $Y_2^\dstar=6.522$.
676:
677: \section{Discussion}
678:
679: In this paper, we have investigated by analytical means
680: the transmission through a disordered tunnel barrier,
681: in the regime~(\ref{regime}) of most physical interest.
682: We have thus provided a quantitative analysis of the
683: phenomenon of disorder-induced enhanced tunneling,
684: put forward by Freilikher et al.~\cite{f1,f2}.
685: The most salient outcome of the present work
686: is that the enhancement effect is a non-monotonic
687: function of the strength of disorder, and that it is maximally efficient
688: at some well-defined intermediate value $Y^\star$
689: of the reduced disorder strength~$Y$.
690:
691: The key point of our approach consists in utilizing
692: the scaling law~(\ref{char})--(\ref{fdef})
693: for the complex characteristic exponent $\Omega$.
694: It is worth noticing that this formalism encompasses,
695: and treats on the same footing,
696: both the usual situation of the Anderson localization,
697: where the energy is above the mean of the disordered potential
698: (corresponding to negative values of the scaling variable $X$),
699: and the tunneling situation,
700: where the energy is below the mean of the disordered potential
701: (corresponding to positive values of~$X$).
702: As a consequence, the general results on the statistics of the transmission,
703: recalled in the Introduction, still hold in the tunneling situation.
704: In particular, $\mean{\ln T}$ is the right quantity to consider.
705:
706: In the case of a white-noise potential, considered throughout this work,
707: the form~(\ref{xdef}) of the scaling variable $X$
708: is merely dictated by dimensional analysis,
709: while the explicit formula~(\ref{fdef}) for the scaling function $F(X)$
710: is given a self-contained derivation in the Appendix.
711: Somewhat equivalent results had been obtained
712: in several earlier works~\cite{fl,h,sulem}.
713: The scaling law~(\ref{char})--(\ref{fdef})
714: also describes the spectra of other one-dimensional disordered systems,
715: such as the diffusion of classical particles in a random force field~\cite{bg}.
716: An analogous scaling formula holds
717: for discrete models near their band edges.
718: Consider the tight-binding equation
719: \beq
720: \psi_{n-1}+\psi_{n+1}+V_n\psi_n=\E\psi_n.
721: \label{tbm}
722: \eeq
723: For a clean chain ($V_n=0$), the dispersion relation reads $\E=2\cos p$.
724: In the regime of a weak disorder ($\mean{V_n^2}=\Delta\ll1$),
725: and near the upper band edge ($p\ll1$), one has~\cite{dg}
726: \beq
727: \Omega=\Delta^{1/3}\,f(x),\qquad
728: x=-\frac{p^2}{\Delta^{2/3}}\approx\frac{\E-2}{\Delta^{2/3}}.
729: \label{dgres}
730: \eeq
731: This scaling law turns out to play a central role in various problems,
732: such as the spreading dynamics of a wave packet~\cite{toro}.
733: The results~(\ref{char})--(\ref{fdef})
734: can be viewed as the formal continuum limit of~(\ref{dgres}).
735: The identification between our continuum problem
736: and the tight-binding model~(\ref{tbm}) has to take place near the band edge.
737: Indeed, introducing explicitly the lattice spacing $a$,
738: which plays the role of the correlation range of the potential,
739: the first inequality of~(\ref{regime}) implies $\abs{pa}=Ka\ll1$,
740: as $p=\i K$, whereas $D=\Delta/(2a^3)$.
741: The scaling variables and functions match between~(\ref{dgres})
742: and~(\ref{char})--(\ref{fdef})
743: (up to powers of $2$ due to different conventions):~$x=2^{2/3}X$, $f=2^{1/3}F$.
744:
745: For a square disordered barrier,
746: the main result~(\ref{main}) for $\mean{\ln T}$,
747: i.e., the logarithm of the typical transmission,
748: can be generalized to all the moments of $T$ in the regime~(\ref{regime}).
749: It can indeed be shown, along the lines of~\cite{pen},
750: that these moments scale as
751: \beq
752: \mean{T^n}\approx\exp\left(-2n\,G_n(Y)\,KL\right),
753: \eeq
754: where $Y$ is the reduced disorder strength of~(\ref{ydef}),
755: while the scaling function $G_n(Y)$ depends on the order $n$
756: (not necessarily an integer).
757: Equation~(\ref{main}) is recovered as $G_0(Y)=G(Y)$.
758: Skipping every detail, let us mention the weak-disorder expansion
759: \beq
760: G_n(Y)=1-\frac{2n+1}{4}\,Y+\cdots\qquad(Y\ll1),
761: \eeq
762: hence
763: \beq
764: \mean{T^n}\approx
765: \exp\left[2L\left(-nK+\frac{n(2n+1)D}{4K^2}+\cdots\right)\right].
766: \eeq
767: In agreement with~\cite{f2}, both $\mean{T}$ and $\mean{1/T}$,
768: and indeed all the moments whose order is not in the range $-1/2<n<0$,
769: are increased by a weak amount of disorder.
770: In the opposite strong-disorder regime, we have
771: \beq
772: G_n(Y)\approx a_n\,Y^{1/3}\qquad(Y\gg1),
773: \eeq
774: with some $n$-dependent amplitude $a_n$, hence
775: \beq
776: \mean{T^n}\approx\exp\left(-2n\,a_n\,D^{1/3}L\right).
777: \eeq
778: The evaluation of the full scaling function $G_n(Y)$ is a difficult task
779: in general, except for~$n$ a negative integer,
780: corresponding by means of~(\ref{tsim}) to positive integer
781: powers of $\abs{\psi_+(L)}^2$.
782: In this case, $G_n(Y)$ can be derived
783: by means of the algebraic approach of~\cite{pen}:
784: it is (the real part of) an algebraic function with degree $2\abs{n}+1$.
785:
786: For a disordered barrier with an arbitrary profile,
787: the general prediction~(\ref{texpli}) has been illustrated on two
788: realistic examples of parabolic barriers.
789: These examples demonstrate that the qualitative features of
790: disorder-induced enhanced tunneling, and chiefly its non-monotonic behavior
791: as a function of the disorder strength,
792: are rather insensitive to the shape of the mean and/or the width of the
793: random potential.
794: Even quantitative characteristics, such as $Y^\star$ or $Y^\dstar$,
795: do not depend too much on the profile of the barrier.
796:
797: Besides the present situation of thick tunnel barriers,
798: the non-monotonic enhancement of transmission may be a more general phenomenon.
799: Somewhat similar features have indeed been observed recently~\cite{g}
800: in a problem inspired by nuclear fission.
801:
802: It is also worth noticing that,
803: in the somewhat dual case of the total reflection
804: of an electron by a semi-infinite disordered sample,
805: universal features of the distribution of the Wigner time delay
806: has received much attention recently~\cite{tc}.
807: The effects of disorder-induced enhanced tunneling
808: on the Wigner time delay are also potentially of interest
809: in transmission~\cite{bolt}, even though this concept has been criticized
810: as being somewhat ambiguous~\cite{dwell}.
811:
812: Finally, the disorder-induced enhancement of transmission
813: is a quantum-mechanical phenomenon, and more generally a wave phenomenon.
814: It is indeed due to the existence of non-trivial localized states
815: in a weakly-disordered barrier,
816: which themselves originate in the interferences between the multiply
817: scattered waves.
818: Therefore, no disorder-induced enhancement is to be expected
819: e.g. in the Kramers problem~\cite{mel} of the thermally activated hopping
820: of a classical particle over a potential barrier.
821:
822: \subsubsection*{Acknowledgements}
823:
824: It is a pleasure to thank Bertrand Giraud for discussions related
825: to~\cite{g} which motivated this work,
826: and Boris Shapiro for valuable correspondence.
827:
828: %\newpage
829: \appendix
830:
831: \section{Derivation of expressions~(\ref{char})--(\ref{fdef})}
832:
833: This Appendix presents a self-contained derivation
834: of expressions~(\ref{char})--(\ref{fdef})
835: for the complex characteristic exponent $\Omega$, introduced in~(\ref{omedef}).
836:
837: To do so, we determine the invariant measure
838: of the complex Riccati variable $z$, introduced in~(\ref{zdef}).
839: As this distribution has a complex support,
840: instead of writing a Fokker-Planck equation for its density,
841: it is advantageous to consider the linear transforms
842: \beq
843: \Phi(y)=\stat{\ln(y-z)},\qquad\phi(y)=\Phi'(y)=\bigstat{\frac{1}{y-z}}.
844: \eeq
845:
846: For definiteness, we assume that $K$ has positive real and imaginary parts.
847: Equation~(\ref{ric}) shows that $z(x)$ keeps a positive imaginary part,
848: so that $\Phi(y)$ is analytic in the lower half plane.
849:
850: In the spirit of the derivation of the Fokker-Planck equation~\cite{vank},
851: we consider a small increment $\eps$ of $x$, and set $z(x+\eps)=z(x)+\eta$.
852: Both
853: \beq
854: \mean{\eta}\approx(K^2-z^2)\eps,\qquad\mean{\eta^2}\approx2D\eps
855: \label{cums}
856: \eeq
857: are proportional to the increment $\eps$,
858: while higher cumulants are negligible.
859:
860: Along the lines of the Dyson-Schmidt approach~\cite{dy,sch},
861: we then look for a stationarity condition by comparing
862: the expressions of $\Phi(y)$ corresponding
863: to the points $x$ and $x+\eps$: $\stat{\ln(y-z)}=\stat{\ln(y-z-\eta)}$.
864: By expanding the right side of this equality in powers of $\eta$,
865: and using~(\ref{cums}), we obtain the condition
866: \beq
867: \bigstat{\frac{K^2-z^2}{y-z}+\frac{D}{(y-z)^2}}=0,
868: \eeq
869: i.e.,
870: \beq
871: D\phi'(y)+(y^2-K^2)\phi(y)=y+\Omega.
872: \eeq
873: The strength of disorder $D$ can be scaled out by setting
874: $K^2=D^{2/3}X$, $\Omega=D^{1/3}F$,
875: proving thus the scaling formulas~(\ref{char}), (\ref{xdef}),
876: and $y=D^{1/3}u$, $\phi=D^{-1/3}\psi$.
877:
878: We are thus left with a differential equation for $\psi(u)$,
879: \beq
880: \psi'(u)+(u^2-X)\psi(u)=u+F,
881: \label{ode}
882: \eeq
883: which can be solved by {\it varying the constant}:
884: \beq
885: \psi(u)=\exp(-u^3/3+Xu)\,C(u),\qquad C(u)=\int(v+F)\,\exp(v^3/3-Xv)\,\d v.
886: \eeq
887: The existence of a regular solution,
888: such that $\psi(u)\to0$ as $\abs{u}\to\infty$, determines $F(X)$.
889: We must have $C(u)\to0$ as $\abs{u}\to\infty$
890: in all the directions of the lower half plane where $\exp(-u^3/3)$ diverges.
891: This happens in two Stokes sectors, represented by the directions
892: $u\to-\infty$ and $u\to\e^{-\i\pi/3}\infty$.
893: We thus obtain
894: \beq
895: F=\frac{I'(X)}{I(X)},
896: \eeq
897: with
898: \beq
899: I(X)=\int_{-\infty}^{\e^{-\i\pi/3}\infty}\exp(v^3/3-Xv)\,\d v.
900: \eeq
901: Finally, $I(X)$ is equal to the Airy function $\Ai(\e^{-2\i\pi/3}X)$~\cite{as},
902: up to a multiplicative constant.
903: This observation leads to~(\ref{fdef}).
904:
905: \newpage
906: \begin{thebibliography}{99}
907:
908: \bibitem{rev}
909: See e.g. I.M. Lifshitz, S.A. Gredeskul, and L.A. Pastur, {\it Introduction to
910: the Theory of Disordered Systems} (Wiley, New-York, 1988);
911: B. Kramer and A. MacKinnon, Rep. Prog. Phys. {\bf 56}, 1469 (1993);
912: Y. Imry, {\it Introduction to Mesoscopic Physics} (Oxford University Press,
913: Oxford, 1997);
914: K. Efetov, {\it Supersymmetry in Disorder and Chaos} (Cambridge University
915: Press, Cambridge, 1997).
916:
917: \bibitem{pen}
918: J.B. Pendry, Adv. Phys. {\bf 43}, 461 (1994).
919:
920: \bibitem{lan}
921: Y. Imry and R. Landauer, Rev. Mod. Phys. {\bf 71}, S306 (1999),
922: and the references therein.
923:
924: \bibitem{cpv}
925: A. Crisanti, G. Paladin, and A. Vulpiani, {\it Products of Random Matrices in
926: Statistical Physics} (Springer, Berlin, 1992).
927:
928: \bibitem{alea}
929: J.M. Luck, {\it Syst\`emes d\'esordonn\'es unidimensionnels} (in French)
930: (Collection Al\'ea-Saclay, 1992).
931:
932: \bibitem{f1}
933: V.D. Freilikher, B.A. Liansky, I.V. Yurkevich, A.A. Maradudin, and A.R. McGurn,
934: Phys. Rev. E {\bf 51}, 6301 (1995).
935:
936: \bibitem{f2}
937: V.D. Freilikher, M. Pustilnik, and I.V. Yurkevich, Phys. Rev. B {\bf 53}, 7413
938: (1996).
939:
940: \bibitem{hein}
941: J. Heinrichs, Phys. Rev. B {\bf 33}, 5261 (1986); {\bf 36}, 2867 (1987).
942:
943: \bibitem{dey}
944: L.I. Deych, A.A. Lisyansky, and B.L. Altshuler, Phys. Rev. Lett. {\bf 84}, 2678
945: (2000).
946:
947: \bibitem{ll}
948: See e.g. L.D. Landau and E.M. Lifshitz, {\it Quantum Mechanics:
949: Non-Relativistic Theory} (Pergamon, Oxford, 1959).
950:
951: \bibitem{wkb}
952: See e.g. N. Fr\"oman and P.O. Fr\"oman, {\it Physical Problems Solved by the
953: Phase-Integral Method} (Cambridge University Press, Cambridge, 2002).
954:
955: \bibitem{jzj}
956: See e.g. J. Zinn-Justin, {\it Quantum Field Theory and Critical Phenomena}
957: (Clarendon, Oxford, 1989).
958:
959: \bibitem{dy}
960: F.J. Dyson, Phys. Rev. {\bf 92}, 1331 (1953).
961:
962: \bibitem{sch}
963: H. Schmidt, Phys. Rev. {\bf 105}, 425 (1957).
964:
965: \bibitem{tn}
966: Th.M. Nieuwenhuizen, Physica {\bf A 120}, 468 (1983).
967:
968: \bibitem{fl}
969: H.L. Frisch and S.P. Lloyd, Phys. Rev. {\bf 120}, 1175 (1960).
970:
971: \bibitem{h}
972: B.I. Halperin, Phys. Rev. {\bf 139}, A 104 (1965).
973:
974: \bibitem{sulem}
975: P.L. Sulem, Physica {\bf 70}, 190 (1973).
976:
977: \bibitem{as}
978: M. Abramowitz and I.A. Stegun, {\it Handbook of Mathematical Functions} (Dover,
979: New York, 1974).
980:
981: \bibitem{bg}
982: J.P. Bouchaud, A. Comtet, A. Georges, and P. Le Doussal, Ann. Phys. {\bf 201},
983: 285 (1990).
984:
985: \bibitem{dg}
986: B. Derrida and E. Gardner, J. Phys. (France) {\bf 45}, 1283 (1984).
987:
988: \bibitem{toro}
989: S. De Toro Arias and J.M. Luck, J. Phys. A {\bf 31}, 7699 (1998).
990:
991: \bibitem{g}
992: B.G. Giraud, K. Amos, S. Karataglidis, and B.A. Robson (work in progress).
993:
994: \bibitem{tc}
995: C. Texier and A. Comtet, Phys. Rev. Lett. {\bf 82}, 4220 (1999),
996: and the references therein.
997:
998: \bibitem{bolt}
999: C.J. Bolton-Heaton, C.J. Lambert, V.I. Fal'ko, V. Progodin, and A.J. Epstein,
1000: Phys. Rev. B {\bf 60}, 10569 (1999).
1001:
1002: \bibitem{dwell}
1003: R. Landauer and T. Martin, Rev. Mod. Phys. {\bf 66}, 217 (1994).
1004:
1005: \bibitem{mel}
1006: See e.g. V.I. Mel'nikov, Phys. Rep. {\bf 209}, 1 (1991).
1007:
1008: \bibitem{vank}
1009: N.G. van Kampen, {\it Stochastic Processes in Physics and Chemistry}
1010: (North-Holland, Amsterdam, 1992).
1011:
1012: \end{thebibliography}
1013: \end{document}
1014: