1: %% LyX 1.1 created this file. For more info, see http://www.lyx.org/.
2: %% Do not edit unless you really know what you are doing.
3: \documentclass[twoside,twocolumn,english,prb,a4paper]{revtex4}
4: \usepackage[T1]{fontenc}
5: \usepackage[latin1]{inputenc}
6: \usepackage{geometry}
7: \geometry{verbose,a4paper,tmargin=2cm,bmargin=1cm,lmargin=1cm,rmargin=1.5cm}
8: \usepackage{babel}
9: \usepackage{graphics}
10:
11: \makeatletter
12:
13: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
14: \providecommand{\LyX}{L\kern-.1667em\lower.25em\hbox{Y}\kern-.125emX\@}
15:
16: \makeatother
17: \begin{document}
18:
19: \title{The importance of friction in the description of low-temperature
20: dephasing}
21:
22:
23: \author{by Florian Marquardt}
24:
25:
26: \email{Florian.Marquardt@unibas.ch}
27:
28:
29: \affiliation{Department of Physics and Astronomy, University of Basel, Klingelbergstrasse
30: 82, CH-4056 Basel, Switzerland}
31:
32: \begin{abstract}
33: We discuss the importance of the real part of the Feynman-Vernon influence
34: action for the analysis of dephasing and decay near the ground state
35: of a system which is coupled to a bath. Using exactly solvable linear
36: quantum dissipative systems, it is shown how the effects of the real
37: and the imaginary part (describing friction and fluctuations, respectively)
38: may cancel beyond lowest-order perturbation theory. The resulting
39: picture is extended to a qualitative discussion of nonlinear systems
40: and dephasing of degenerate fermions. We explain why dephasing rates
41: will, in general, come out finite at zero temperature if they are
42: deduced from the imaginary part of the action alone, a procedure which
43: is reliable only for highly excited states.
44: \end{abstract}
45:
46: \date{July 30th, 2002}
47:
48: \maketitle
49:
50: \section{Introduction}
51:
52: The Feynman-Vernon influence functional\cite{feynvern,feynhibbs}
53: plays a prominent role in discussions of dephasing that aim to go
54: beyond a simple master equation approach. In principle, the influence
55: functional takes into account every effect of the environment (the
56: {}``bath'') on the system under consideration. This includes heating,
57: dephasing, friction and renormalization effects (changing the external
58: potential or the effective mass of a particle). Its popularity arises
59: not only from the fact that it constitutes an exact approach, but
60: also from the direct physical meaning which it acquires in some situations.
61: In particular, this holds for typical interference experiments, where
62: two wave packets describing a single particle follow two different
63: trajectories in order to be recombined later on. For such a case,
64: the influence functional is equal to the overlap between the two different
65: bath states which result due to the particle moving along either one
66: of the semiclassical trajectories. If the bath can distinguish between
67: the two paths, it acts as a which-way detector, and the diminished
68: magnitude of the overlap directly gives the ensuing decrease of the
69: visibility of the interference pattern. Writing the overlap in the
70: form of an exponential, \( \exp (iS_{R}-S_{I}) \), one can obviously
71: conclude that only the imaginary part \( S_{I} \) of the {}``influence
72: action'' is responsible for this decrease and, therefore, \( S_{I} \)
73: alone describes dephasing \emph{in such a situation}. Furthermore,
74: if the system is subject to a fluctuating external \emph{classical}
75: field, only \( S_{I} \) remains, while \( S_{R} \) vanishes identically.
76: This is because \( S_{I} \) is due to the fluctuations of the bath,
77: while \( S_{R} \) stems from the back-action of the bath onto the
78: system (including friction effects), which is absent for an external
79: noise field. The approach of evaluating \( S_{I} \) along semiclassical
80: trajectories has been successfully applied to the calculation of dephasing
81: rates in many situations\cite{AAK,sai,cohen}.
82:
83: However, near the ground state of a system, an analysis along these
84: lines is likely to fail. Qualitatively, this may already be deduced
85: from the fact that dropping \( S_{R} \) means replacing the environment
86: by an artificial fluctuating classical field whose correlation function
87: includes the zero-point fluctuations of the original quantum bath.
88: Therefore, even at zero temperature, this fluctuating field will,
89: in general, heat the system, which directly leads to dephasing. The
90: role of \( S_{R} \) is to counteract this effect. In this article,
91: it is our aim to display explicitly, in a detailed manner, the necessity
92: of keeping the real part \( S_{R} \) of the influence action in discussions
93: of dephasing and decay near the ground state of a system.
94:
95: This issue derives its importance partly from the fact that the evaluation
96: of \( S_{I} \) along semiclassical trajectories has proven to be
97: an efficient way of extracting dephasing rates in the problem of weak
98: localization\cite{AAK,sai} in the limit of high temperatures, when
99: zero-point fluctuations of the bath may be omitted. In contrast, at
100: low temperatures, a single-particle semiclassical calculation may
101: become invalid, since it neglects the Pauli principle which is known
102: to play an important role for the inelastic scattering of electrons
103: and is not included in the Feynman-Vernon influence functional. However,
104: recently an extension of the influence functional to the case of a
105: many-fermion system has been derived, using an exact procedure\cite{GZ,GZ_CL,GZ_PB}
106: and including the Pauli principle. This permitted a discussion of
107: dephasing in a disordered metal even for the case of low temperatures.
108: Following the general strategy of earlier works\cite{chakravarty}
109: dealing with the high-temperature case, the dephasing rate was deduced
110: from \( S_{I} \) in a semiclassical calculation and found to be finite
111: even at zero temperature. Since the {}``orthodox'' theory\cite{AAK}
112: had predicted a vanishing rate in the limit \( T\rightarrow 0 \),
113: the new results prompted considerable criticism\cite{AAGcritique,VavAmbeg,cohenimry,BelitzKirckpatrick},
114: which mostly emphasized technical aspects of impurity averaging or
115: used perturbation theory to arrive at different conclusions. In the
116: present work, we want to clarify some essential aspects of the roles
117: of \( S_{R} \) and \( S_{I} \), using physically much more transparent
118: exactly solvable models.
119:
120: We want to emphasize that {}``zero-temperature dephasing'' as such
121: is perfectly possible: If one prepares a system in any superposition
122: of excited states and couples it to a bath, it will, in general, decay
123: towards its ground state by spontaneous emission of energy into the
124: bath (at \( T=0 \)). This will destroy any coherent superposition,
125: thus leading to dephasing. Considerations of this kind are particularly
126: relevant for quantum-information processing, where one necessarily
127: deals with nonequilibrium situations involving excited qubit states
128: of finite energy. The situation is different for the weak-localization
129: problem (and similar transport interference effects): There, one is
130: interested in the zero-frequency limit of the system's linear response,
131: which depends on the coherence properties of arbitrarily low-lying
132: excited states. It is the subtleties associated with a path-integral
133: description of these situations which we want to address in this work.
134:
135: The article is organized as follows: After a brief review of the influence
136: functional and its meaning in semiclassical situations (sec.~\ref{section2}),
137: we will rewrite the expressions for \( S_{I} \) and the Golden Rule
138: decay rate, in order to compare the two (sec.~\ref{section3}). In
139: doing so, we will closely follow the analysis of Cohen and Imry\cite{cohen,cohenimry}.
140: Then, we will show how and why the effects of \( S_{R} \) and \( S_{I} \)
141: may compensate each other in the derivation of a decay rate starting
142: from the path integral expression for the time-evolution of the density
143: matrix (sec.~\ref{section4}), even though they cannot cancel each
144: other in the influence action. There, it may be observed that drawing
145: conclusions about decay directly from the exponent \( iS_{R}-S_{I} \)
146: of the influence functional is usually not possible. The crucial cancellation
147: takes place at a later stage of the calculation, after proper integration
148: over the fluctuations around the classical paths and after averaging
149: over the initial state. In order to prove that this compensation takes
150: place not only in lowest order perturbation theory, we will then specialize
151: to exactly solvable linear dissipative systems, i.e. the damped harmonic
152: oscillator (sec.~\ref{linearmodels}) and the free particle (sec.~\ref{qubrownmotion}).
153: We will also point out that there is an important difference between
154: the oscillator and the free particle: Dropping \( S_{R} \) has a
155: much more drastic effect on the former, leading to an artificial finite
156: decay rate of the ground state at zero temperature. However, in order
157: to discuss the importance of \( S_{R} \) for the calculation of dephasing
158: rates (involving decay of excited states), we have to extend our analysis
159: to nonlinear models (sec. \ref{section7}). We will explain in a more
160: qualitative fashion why the essential insights gained from the exactly
161: solvable models should remain valid both for the nonlinear models
162: as well as for systems of degenerate fermions.
163:
164:
165: \section{The influence functional and dephasing in simple situations}
166:
167: \label{section2}We are interested in the time-evolution of the reduced
168: density matrix of a system with coordinate \( q \) which is coupled
169: to some environment (the bath). If system and bath had been uncoupled
170: prior to \( t=0 \), then the density matrix at a later time \( t \)
171: is linearly related to that at time \( 0 \):
172:
173: \begin{equation}
174: \label{rhoJ}
175: \rho (q^{>}_{t},q^{<}_{t},t)=\int dq_{0}^{>}dq_{0}^{<}\, J(q_{t}^{>},q_{t}^{<}|q_{0}^{>},q_{0}^{<};t)\rho (q_{0}^{>},q_{0}^{<},0)\, .
176: \end{equation}
177:
178:
179: The propagator \( J \) on the right-hand-side of this equation is
180: given by
181:
182: \begin{equation}
183: \label{prop}
184: J=\int \! \! \! \int Dq^{>}Dq^{<}\exp \left[ i(S_{0}^{>}-S_{0}^{<})\right] \exp \left[ iS_{R}-S_{I}\right] \, .
185: \end{equation}
186:
187:
188: Here we have set \( \hbar \equiv 1 \) and introduced an abbreviated
189: notation: The path integral extends over all {}``forward'' paths
190: \( q^{>}(\cdot ) \) running from the given value of \( q^{>}_{0} \)
191: to \( q^{>}_{t} \), likewise for the {}``backward'' paths \( q^{<}(\cdot ) \).
192: The value of the action of the uncoupled system, evaluated along \( q^{>(<)}(\cdot ), \)
193: is denoted by \( S_{0}^{>(<)} \). The second exponential in Eq.~(\ref{prop})
194: is the Feynman-Vernon influence functional\cite{feynvern,feynhibbs}.
195: Both \( S_{R} \) and \( S_{I} \) are real-valued functionals that
196: depend on both paths simultaneously. The influence functional is the
197: overlap of bath states which have time-evolved out of the initial
198: bath state \( \chi _{0} \) under the action of either \( q^{>}(\cdot ) \)
199: or \( q^{<}(\cdot ) \):
200:
201: \begin{equation}
202: e^{iS_{R}-S_{I}}\equiv \left\langle \chi [q^{<}(\cdot )]|\chi [q^{>}(\cdot )]\right\rangle \, .
203: \end{equation}
204:
205:
206: At zero temperature, the state \( \chi _{0} \) is the ground state
207: of the unperturbed bath, while at finite temperatures an additional
208: thermal average over \( \chi _{0} \) has to be performed. From this
209: representation, it follows\cite{feynvern,feynhibbs} that \( S_{R} \)
210: changes sign on interchanging \( q^{>} \) and \( q^{<} \) while
211: \( S_{I} \) always remains nonnegative - the magnitude of the overlap
212: can only be decreased compared with its initial value of one.
213:
214: The meaning of \( S_{R} \) and \( S_{I} \) becomes particularly
215: transparent for an interference setup, where two wave packets \( \Psi _{>} \)
216: and \( \Psi _{<} \) travel along two different paths \( q_{cl}^{>}(\cdot ) \)
217: and \( q^{<}_{cl}(\cdot ) \). We want to assume a situation which
218: can be described semiclassically, i.e. the wave-length is supposed
219: to be much smaller than the size of a wave packet and this again is
220: much smaller than the typical separation between the paths and the
221: typical dimensions over which the external potential changes. Then,
222: it suffices to evaluate \( S_{R} \) and \( S_{I} \) just for the
223: combination of these two paths, since the fluctuations around them
224: are comparatively unimportant. The environment will affect the interference
225: pattern on the screen mainly by changing the interference term\cite{sai}:
226:
227: \begin{eqnarray}
228: \rho (x,x,t) & \approx & \left| \Psi _{>}(x,t)\right| ^{2}+\left| \Psi _{<}(x,t)\right| ^{2}+\nonumber \\
229: & & 2\, Re\left[ \Psi _{>}(x,t)\Psi ^{*}_{<}(x,t)e^{iS_{R}-S_{I}}\right] \, .\label{simpleinterference}
230: \end{eqnarray}
231:
232:
233: In this equation, \( \Psi _{>(<)}(x,t) \) are assumed to represent
234: the unperturbed time-evolution of the wave packets. To a first approximation,
235: \( S_{R,I} \) do not enter the {}``classical'' terms in the first
236: line of Eq. (\ref{simpleinterference}), since \( S_{R,I}[q_{cl}^{>},q^{>}_{cl}]=0 \).
237: Deviations from this first approximation stem from the integration
238: over fluctuations away from \( q^{>(<)}_{cl} \) and describe, for
239: example, mass- and potential renormalization as well as slowing-down
240: of the wave packet due to friction. Obviously, \( S_{I}\equiv S_{I}[q^{>}_{cl},q^{<}_{cl}] \)
241: determines directly the decrease in visibility of the interference
242: pattern while \( S_{R}\equiv S_{R}[q^{>}_{cl},q^{<}_{cl}] \) only
243: gives a phase-shift. Averaging the interference pattern over different
244: configurations of the external potential (appropriate for impurity
245: averaging in mesoscopic samples) may further decrease the visibility.
246: However, if the two paths are time-reversed copies of each other (as
247: is the case in discussions of weak-localization\cite{AAK,sai}), the
248: phase difference between \( \Psi _{>} \) and \( \Psi _{<} \) vanishes
249: in a time-reversal invariant situation, so that the impurity-average
250: over the corresponding phase factor does not lead to a suppression
251: in the situation without the bath. On the other hand, the average
252: over \( \exp (iS_{R}) \) will, in general, decrease further the magnitude
253: of the interference term\cite{notesravg}. In any case, given the
254: simple physical picture presented here, one would not expect \( S_{I} \)
255: and \( S_{R} \) to be able to cancel each other's effects, since
256: they represent, respectively, the real and imaginary part of an exponent.
257:
258: For the special case of a force \( \hat{F} \) deriving from a bath
259: of harmonic oscillators (linear bath), with vanishing average \( \left\langle \hat{F}\right\rangle =0 \)
260: and a linear interaction \( \hat{V}=-\hat{q}\hat{F} \), we have:
261:
262: \begin{eqnarray}
263: S_{I} & = & \int _{0}^{t}\! \! \! dt_{1}\! \! \! \int _{0}^{t_{1}}\! \! \! dt_{2}\left( q^{>}_{1}-q^{<}_{1}\right) Re\left\langle \hat{F}_{1}\hat{F}_{2}\right\rangle \left( q^{>}_{2}-q^{<}_{2}\right) \label{silin} \\
264: S_{R} & = & -\! \int _{0}^{t}\! \! \! dt_{1}\! \! \! \int _{0}^{t_{1}}\! \! \! dt_{2}\left( q^{>}_{1}-q^{<}_{1}\right) Im\left\langle \hat{F}_{1}\hat{F}_{2}\right\rangle \left( q^{>}_{2}+q^{<}_{2}\right) \label{srsi}
265: \end{eqnarray}
266:
267:
268: Here we have used the notation \( q^{>}_{1}\equiv q^{>}(t_{1}) \).
269: The angular brackets denote averaging over the equilibrium state of
270: the unperturbed bath. \( S_{I} \) only depends on the symmetrized
271: part of the bath correlator \( Re\left\langle \hat{F}(\tau )\hat{F}(0)\right\rangle =\left\langle \left\{ \hat{F}(\tau ),\hat{F}(0)\right\} \right\rangle /2 \),
272: which becomes the classical correlator \( \left\langle F(\tau )F(0)\right\rangle \)
273: for the case of classical Gaussian random noise. In the latter case,
274: \( S_{R} \) vanishes and the influence functional \( \exp (-S_{I}) \)
275: is simply the classical average of the phase factor
276:
277: \begin{equation}
278: \exp \left[ i\int _{0}^{t}(q^{<}(\tau )-q^{>}(\tau ))F(\tau )d\tau \right] \, .
279: \end{equation}
280:
281:
282: If one drops \( S_{R} \) in a dephasing calculation (noting that
283: it does not enter dephasing in semiclassical situations such as those
284: that can be described by Eq.~(\ref{simpleinterference})), one effectively
285: replaces the quantum bath by a classical fluctuating force whose correlator
286: is determined by the symmetrized part of the quantum correlator. This
287: contains the zero-point fluctuations, since \begin{eqnarray}
288: \left\langle \left\{ \hat{Q}(\tau ),\hat{Q}(0)\right\} \right\rangle \propto & & \nonumber \\
289: \left( 2n(\omega )+1\right) \cos (\omega \tau )=\coth (\omega /2T)\cos (\omega \tau ) & &
290: \end{eqnarray}
291: for the coordinate \( \hat{Q} \) of a single bath oscillator of
292: frequency \( \omega \), with \( n(\omega ) \) being the Bose distribution
293: function that vanishes at \( T=0 \).
294:
295: Although replacing the quantum bath by a classical noise force seems
296: to be a drastic step, it can lead to correct predictions for the dephasing
297: rate in semiclassical situations such as the one discussed above,
298: even at zero temperature. It is instructive to observe how this comes
299: about in an exactly solvable model. A particularly simple situation
300: is the one analyzed by Caldeira and Leggett\cite{clwavepackets} (see
301: also Ref.~\onlinecite{lossmullen}). They considered a damped quantum
302: harmonic oscillator, where the initial state consisted of a superposition
303: of Gaussian wavepackets, one centered at the origin, the other at
304: a distance \( z \). In the course of time, the displaced wavepacket
305: oscillates back and forth in the oscillator potential well. Whenever
306: the packets overlap, an interference pattern results (due to the difference
307: in the respective momenta). The environment leads both to damping
308: and dephasing, where the latter typically proceeds at a much faster
309: rate. For our purposes, we are interested in the limit of small damping,
310: where, to a first approximation, the center-of-mass motion of the
311: wave packets is not appreciably altered by friction in the period
312: of the oscillation. In this case, it turns out that, indeed, the result
313: predicted by the approximation Eq.~(\ref{simpleinterference}) for
314: the attenuation of the interference pattern is correct. This can be
315: seen in Ref.~\onlinecite{clwavepackets} by taking the limit of weak
316: damping (\( \gamma \rightarrow 0,\, \omega \rightarrow \omega _{R} \))
317: in their exact result for the exponent of the attenuation factor (see
318: Eq.~(2.13) of Ref.~\onlinecite{clwavepackets}) and comparing this
319: to \( S_{I} \) (see Eq.~(\ref{silin})), which is to be evaluated
320: for the pair of classical paths followed by the two wave packets,
321: \( q_{cl}^{>}(t)=z\, \cos (\omega t) \) and \( q_{cl}^{<}(t)\equiv 0 \).
322: In terms of the quantity \( C_{00} \) listed in appendix \ref{appc00}
323: of the present work, both results may be obtained by multiplying \( C_{00} \)
324: by \( \sin (\tilde{\omega }t)^{2} \) and setting \( \gamma =0,\, \tilde{\omega }=\omega \).
325: In addition, in order to obtain \( S_{I}[q^{>}_{cl},q^{<}_{cl}] \),
326: the factors \( \sin (\omega (t-t_{1})) \) and \( \sin (\omega (t-t_{2})) \)
327: must be replaced by \( \cos (\omega t_{1}) \) and \( \cos (\omega t_{2}) \).
328: The results obviously coincide for the points in time when the wave-packets
329: meet (\( \omega t=\pi /2+n\pi \)). We emphasize that the correspondence
330: to the semiclassical result holds only for a situation far away from
331: the ground state of the system, where at least one of the wave packets
332: is in a superposition of highly excited oscillator states.
333:
334: At zero bath temperature, the dephasing is purely due to spontaneous
335: emission, as pointed out in the discussion of Ref.~\onlinecite{clwavepackets}.
336: Energy is transferred from the oscillating wave-packet into the bath
337: and the system relaxes towards lower-energy states. The dephasing
338: rate is proportional to the total rate \( \Gamma _{out} \) at which
339: the system leaves a given energy level\cite{clwavepackets}. Of course,
340: this rate may be obtained using the Golden Rule (i.e. a master equation
341: description) only for weak coupling, but the qualitative picture seems
342: to be general\cite{clwavepackets}. In the correct description of
343: the physical situation considered here, the rate \( \Gamma _{out} \)
344: is given entirely by the rate of spontaneous emission, \( \Gamma ^{em}_{sp} \).
345: However, if \( S_{R} \) is neglected, then \( S_{I} \) describes
346: a classical noise force (equivalent to the zero-point fluctuations
347: of the bath at \( T=0 \)) and there will be both induced emission
348: and absorption, proceeding at equal rates \( \Gamma ^{em}_{ind}=\Gamma ^{abs}_{ind} \).
349: Therefore, the system is also excited by the bath in that approximation.
350: Nevertheless, we have pointed out above that the total dephasing rate
351: comes out right. The reason is the following: The rate \( \Gamma _{out} \)
352: is the same in both cases, because the rate of spontaneous emission
353: in the correct description is exactly twice that of induced emission
354: in the approximation:
355:
356: \begin{equation}
357: \Gamma _{out}=\Gamma ^{em}_{sp}\equiv \Gamma ^{em}_{ind}+\Gamma ^{abs}_{ind}\, .
358: \end{equation}
359: At this point, it is easy to see the physical reason why such an approximation
360: may fail near the ground state of the system. Then, the transitions
361: downwards in energy may be blocked, which completely suppresses \( \Gamma ^{em}_{sp} \),
362: but not \( \Gamma ^{abs}_{ind} \). We will make this argument more
363: precise in the following sections.
364:
365:
366: \section{Decay rates from \protect\( S_{I}\protect \) and Golden rule: Dependence
367: on the spectra of bath and system motion}
368:
369: \label{section3}The growth of \( S_{I} \) with time depends on the
370: spectral density of the bath fluctuations at frequencies which appear
371: in the system's motion. Formally, this can be seen by introducing
372: the spectrum of the system motion related to the given pair of paths
373: \( q^{>(<)} \) (following the ideas of Refs.~\onlinecite{cohen,cohenimry}):
374:
375: \begin{eqnarray}
376: P(\omega ,s) & \equiv & \frac{1}{2\pi }\int _{-\infty }^{+\infty }d\tau \, e^{i\omega \tau }(q^{>}(s+\frac{\tau }{2})-q^{<}(s+\frac{\tau }{2}))\nonumber \\
377: & & \times (q^{>}(s-\frac{\tau }{2})-q^{<}(s-\frac{\tau }{2}))\label{systemspec}
378: \end{eqnarray}
379:
380:
381: Here, \( s\equiv (t_{1}+t_{2})/2 \) and \( \tau \equiv t_{1}-t_{2} \)
382: are sum and difference times. Therefore, \( P \) is the Fourier transform
383: with respect to \( \tau \) of the \( q \)-dependent terms in \( S_{I} \)
384: (Eq. (\ref{silin})). We take \( q^{>(<)}(t') \) to be zero whenever
385: \( t' \) falls outside the range \( [0,t] \).
386:
387: Furthermore, we define \( \left\langle FF\right\rangle _{\omega } \)
388: to be the Fourier-transform of the symmetrized correlator of \( \hat{F} \),
389:
390: \begin{equation}
391: \left\langle FF\right\rangle _{\omega }\equiv \frac{1}{4\pi }\int _{-\infty }^{+\infty }d\tau \, e^{i\omega \tau }\left\langle \left\{ \hat{F}(\tau ),\hat{F}(0)\right\} \right\rangle \, ,
392: \end{equation}
393:
394:
395: so it is real and symmetric in frequency. The same holds for the system
396: spectrum \( P(\omega ,s) \).
397:
398: Using this, \( S_{I} \) can be expressed in the following way:
399:
400: \begin{equation}
401: \label{overlapspectra}
402: S_{I}=\pi \int _{0}^{t}ds\int _{-\infty }^{+\infty }d\omega \, \left\langle FF\right\rangle _{\omega }P(-\omega ,s)\, .
403: \end{equation}
404:
405:
406: If the dependence of \( P \) on \( s \) is not essential, then \( S_{I} \)
407: grows linearly with time \( t \), at a rate given by the {}``overlap
408: of bath and system spectra''. Similar expressions can be derived
409: for a spatially inhomogeneous fluctuating force\cite{cohen,cohenimry},
410: leading to an additional \( k \)-dependence.
411:
412: Applying this kind of reasoning to a \emph{single} electron moving
413: in a dirty metal\cite{cohen}, the system motion is found to contain
414: frequencies up to (at least) \( 1/\tau _{el} \), such that the growth
415: of \( S_{I} \) with time depends on the bath-spectrum up to this
416: cutoff, including the zero-point fluctuations of the corresponding
417: high-frequency bath modes (which become important at low temperatures).
418: As explained above (sec. \ref{section2}), the description of dephasing
419: in terms of \( S_{I} \) alone may be trusted in a semiclassical situation,
420: where the electron is in a highly excited state. This holds even at
421: zero temperature, when the qualitative physical picture is essentially
422: the same as in the model of oscillating wave packets due to Caldeira
423: and Leggett, discussed above: Dephasing is due to spontaneous emission
424: of energy into the bath. The contribution of frequencies up to \( 1/\tau _{el} \)
425: then implies that the spontaneous emission (and the resulting dephasing)
426: is facilitated by the impurity scattering. The physics behind this
427: is well-known in another context: In quantum electrodynamics, the
428: emitted radiation would be called {}``bremsstrahlung'', since it
429: is the scattering off an external potential that induces the electron
430: to emit radiation.
431:
432: On the other hand, one can express in a similar manner decay rates
433: from a simple Golden Rule calculation\cite{cohenimry}. For the decay
434: of an initial state \( \left| i\right\rangle \) , we have
435:
436: \begin{equation}
437: \label{decayrate}
438: \Gamma _{i}=2\pi \int d\omega \, \left\langle \hat{F}\hat{F}\right\rangle _{\omega }\left\langle \hat{q}\hat{q}\right\rangle ^{(i)}_{-\omega }\, .
439: \end{equation}
440:
441:
442: Here \( \left\langle \hat{F}\hat{F}\right\rangle _{\omega } \) and
443: \( \left\langle \hat{q}\hat{q}\right\rangle ^{(i)}_{-\omega } \)
444: are the Fourier transforms of the \emph{unsymmetrized} correlators
445: of \( \hat{F} \) and \( \hat{q} \), taken in the equilibrium state
446: of the bath or the initial state of the system, respectively:
447:
448: \begin{eqnarray}
449: \left\langle \hat{F}\hat{F}\right\rangle _{\omega } & \equiv & \frac{1}{2\pi }\int d\tau \, e^{i\omega \tau }\left\langle \hat{F}(\tau )\hat{F}(0)\right\rangle \nonumber \\
450: \left\langle \hat{q}\hat{q}\right\rangle ^{(i)}_{\omega } & \equiv & \frac{1}{2\pi }\int d\tau \, e^{i\omega \tau }\left\langle i\right| \hat{q}(\tau )\hat{q}(0)\left| i\right\rangle
451: \end{eqnarray}
452:
453:
454: The important point to notice is that, near the ground state of the
455: system, the quantum correlator \( \left\langle \hat{q}\hat{q}\right\rangle _{\omega }^{(i)} \)
456: is very asymmetric in frequency space, as the system can mostly be
457: excited only (\( \omega >0 \)). At low temperatures, the same holds
458: for the bath. Thus, the decay rate (\ref{decayrate}), containing
459: the product of correlators evaluated at \( \omega \) and \( -\omega \),
460: is very much suppressed below the value that it would acquire if either
461: the bath-correlator or the system-correlator were symmetrized (becoming
462: symmetric both in time and frequency). Since dropping \( S_{R} \)
463: is equivalent to symmetrizing the bath correlator and, furthermore,
464: semiclassical calculations give a symmetric spectrum \( P(\omega ,s) \)
465: of the system motion as well, it becomes clear why there are situations
466: when the decay rate, as deduced from \( S_{I} \), is finite at low
467: temperatures, while the Golden-rule decay rate vanishes. The question
468: then arises whether any procedure that amounts to symmetrization of
469: the correlators, thus leading to drastically wrong results for Golden
470: Rule decay rates at zero temperature, may be justified to discuss
471: dephasing using a path-integral approach. Observations such as this
472: have led Cohen and Imry to conclude that, in their semiclassical analysis
473: of electron dephasing inside a metal\cite{cohenimry}, the contribution
474: of the bath's zero-point fluctuations to the dephasing rate should
475: be dropped. Their argument was not a mathematical proof, but rather
476: drawn from physical intuition and analogies. In the following sections,
477: we try to elucidate the importance of \( S_{R} \) in descriptions
478: of dephasing and decay near the ground state of a system, demonstrating
479: exactly how a cancellation between \( S_{R} \) and \( S_{I} \) may
480: arise.
481:
482:
483: \section{Cancellation of \protect\( S_{R}\protect \) and \protect\( S_{I}\protect \)
484: in lowest-order perturbation theory}
485:
486: \label{section4}We consider a system which is in its ground state
487: at \( t=0 \) before being coupled to a linear bath. From the Golden
488: Rule, we expect there to be no finite decay-rate for the ground state
489: at zero-temperature. It is well-known\cite{feynhibbs} how to derive
490: a master equation from the influence functional, by expanding it to
491: lowest order in the exponent, \( iS_{R}-S_{I} \). We display that
492: calculation here in order to point out where and how \( S_{R} \)
493: and \( S_{I} \) do cancel. The time-evolution of the probability
494: to find the system in its unperturbed ground state is:
495:
496: \begin{eqnarray}
497: & & P_{0}(t)\equiv \left\langle \Psi _{0}\right| \hat{\rho }(t)\left| \Psi _{0}\right\rangle \nonumber \\
498: & = & \int dq_{t}^{>}dq_{t}^{<}\rho _{0}(q^{<}_{t},q^{>}_{t})\rho (q_{t}^{>},q_{t}^{<},t)\nonumber \\
499: & = & \int Dq^{>}Dq^{<}\rho _{0}(q^{<}_{t},q^{>}_{t})e^{i(S_{0}^{>}-S_{0}^{<})+iS_{R}-S_{I}}\rho _{0}(q_{0}^{>},q_{0}^{<})\nonumber \\
500: & \approx & 1+\int Dq^{>}Dq^{<}e^{i(S^{>}_{0}-S^{<}_{0})}\rho _{0}(q^{<}_{t},q^{>}_{t})\nonumber \\
501: & & \, \times (iS_{R}-S_{I})\rho _{0}(q_{0}^{>},q_{0}^{<})\, .\label{eq13}
502: \end{eqnarray}
503:
504:
505: Here \( \rho _{0} \) is the density matrix of the unperturbed ground
506: state, and the path integration also includes integrals over the initial
507: and final coordinates. Now one can insert the expressions for \( S_{R} \)
508: and \( S_{I} \) from Eq.~(\ref{srsi}). The integration over the
509: trajectories and the endpoints produces the (\emph{unsymmetrized})
510: correlators of the system coordinate, such as (for \( t_{1}>t_{2} \)):
511:
512: \begin{eqnarray}
513: \int Dq^{>}e^{iS_{0}^{>}}\Psi ^{*}_{0}(q_{t}^{>})q^{>}(t_{1})q^{>}(t_{2})\Psi _{0}(q_{0}^{>})= & & \nonumber \\
514: \left\langle \Psi _{0}\right| \hat{q}(t_{1})\hat{q}(t_{2})\left| \Psi _{0}\right\rangle e^{-iE_{0}t}\, . & &
515: \end{eqnarray}
516:
517:
518: Evidently, both the unperturbed action \( S_{0} \) and the integrations
519: involving the state \( \Psi _{0} \) are essential for obtaining the
520: correct correlator. In the long-time limit, we obtain the decay rate
521: which also follows from the Golden Rule (\ref{decayrate}): \begin{widetext}
522:
523: \begin{equation}
524: \label{pathmaster}
525: P_{0}(t)\approx 1-\int _{0}^{t}dt_{1}\int _{0}^{t_{1}}dt_{2}\, 2\, Re\left[ \left\langle \hat{F}(t_{1})\hat{F}(t_{2})\right\rangle \left\langle \Psi _{0}\left| \hat{q}(t_{1})\hat{q}(t_{2})\right| \Psi _{0}\right\rangle \right] \approx 1-\Gamma _{0}t\, .
526: \end{equation}
527:
528:
529: \end{widetext}Only the term growing linearly in time has been retained
530: at the end of this equation.
531:
532: Note that, of course, there will always be a small reduction in the
533: probability of finding the unperturbed ground state after switching
534: on the interaction, since the ground state of the coupled system contains
535: contributions from other system states as well. This point has been
536: discussed in more detail recently for the case of the damped harmonic
537: oscillator\cite{buettHO}. It does not contribute to \( \Gamma _{0} \)
538: which describes the decay linear in time.
539:
540: We emphasize that the decay of \( P_{0}(t) \) is not governed by
541: \( S_{I} \) alone: If \( S_{R} \) were dropped, then only the (real-valued)
542: symmetrized version of the bath correlator would appear in Eq.~(\ref{pathmaster}).
543: In that case, the decay rate would only depend on the (real-valued)
544: symmetric part of the system correlator as well. Therefore, a finite
545: decay rate of the ground state would result even at zero temperature,
546: where really there would have been none. The physical reason is the
547: heating introduced by the classical noise field, whose correlator
548: contains the zero-point fluctuations of the original quantum bath,
549: as we have discussed before.
550:
551: On a formal level, we may argue that the decay of the density matrix
552: is not governed by \( S_{I} \) alone since the {}``weighting factor''
553: that is used when {}``averaging'' over many paths contains the \emph{phase}
554: factor \( \exp (i(S_{0}^{>}-S_{0}^{<})) \). Therefore, such an average
555: is not the same as a classical average, for which the decay could
556: not be overestimated by dropping \( S_{R} \), since then we could
557: use \( \left| \left\langle \exp (iS_{R}-S_{I})\right\rangle \right| \leq \left\langle \exp \left( -S_{I}\right) \right\rangle \).
558: Furthermore, the integration over the density matrix of the initial
559: state is obviously essential, as it is needed to produce the correct
560: form of the system correlator. Both facts mean that it would be premature
561: to draw any conclusions about dephasing and decay at an early stage
562: of the calculation, by merely looking at the influence action.
563:
564:
565: \section{Exactly solvable linear systems: {}``Cancellation to all orders''
566: for the damped oscillator}
567:
568: \label{linearmodels}In order to show how \( S_{R} \) and \( S_{I} \)
569: can cancel also beyond lowest-order perturbation theory, we will now
570: turn to linear quantum dissipative systems. Although the exact solutions
571: for the damped harmonic oscillator and the free particle have been
572: well-known for a long time\cite{callegg}, we will review the essential
573: steps in the derivation, pointing out where \( S_{R} \) and \( S_{I} \)
574: do enter. We will first turn to the oscillator, for which the Golden
575: Rule result suggests that one would obtain an artificial decay of
576: the ground state at zero temperature if \( S_{R} \) were neglected.
577: For simplicity, our discussion is restricted to \( T=0 \), since
578: that is the limit where these effects show up most clearly.
579:
580: For any linear system which is linearly coupled to a bath of oscillators,
581: the propagator \( J \) for the density matrix (Eq.~(\ref{prop}))
582: can be evaluated exactly, since the integration over system paths
583: is Gaussian. In fact, \( J \) is found to be given by the {}``semiclassical''
584: result, i.e. an exponential containing the action along stationary
585: paths, multiplied by a prefactor which does not contain the endpoints
586: of the paths (a specialty of linear systems):
587:
588: \begin{equation}
589: \label{Jsemi}
590: J(q^{>}_{t},q^{<}_{t}|q^{>}_{0},q^{<}_{0};t)=\frac{1}{N(t)}e^{iS[q^{>}_{cl},q^{<}_{cl}]}\, ,
591: \end{equation}
592:
593:
594: with \( S=S_{0}^{>}+S_{0}^{<}+S_{R}+iS_{I} \). The paths \( q^{>(<)}_{cl} \)
595: make the full action stationary,
596:
597: \begin{equation}
598: \label{stat}
599: \frac{\delta S}{\delta q^{>}_{cl}}=\frac{\delta S}{\delta q^{<}_{cl}}=0\, ,
600: \end{equation}
601:
602:
603: and fulfill boundary conditions of the form \( q^{>}_{cl}(0)=q^{>}_{0} \).
604: The prefactor \( N(t) \) can be obtained most easily from the condition
605: for normalization of the density matrix,
606:
607: \begin{equation}
608: \label{normal}
609: \int dq_{t}\, J(q_{t},q_{t}|q_{0},q_{0};t)=1\, .
610: \end{equation}
611:
612:
613: \( N(t) \) is found to be independent of \( S_{I} \), but it does
614: depend on \( S_{R} \) (see, for example, the general proof in app.
615: E of Ref.~\onlinecite{GZ_CL}).
616:
617: In the following, we will turn to the special case of the \emph{Ohmic}
618: bath, which leads to a velocity-proportional friction force and has
619: a power spectrum rising linearly at low frequencies (for \( T=0 \)):
620:
621: \begin{equation}
622: \label{ohmbathdef}
623: \left\langle FF\right\rangle _{\omega }=\frac{\eta |\omega |}{2\pi }\, .
624: \end{equation}
625:
626:
627: We will argue below that, in the case of the damped harmonic oscillator,
628: no essential qualitative result will be changed by using other bath
629: spectra, unless these have an excitation gap exceeding the oscillator
630: frequency.
631:
632: As usual, we introduce the center-of-mass coordinate \( R(\tau )\equiv (q^{>}(\tau )+q^{<}(\tau ))/2 \)
633: and the difference coordinate \( r(\tau )\equiv q^{>}(\tau )-q^{<}(\tau ) \)
634: in order to write down the equations obtained from (\ref{stat}),
635: for the case of the Ohmic bath:
636:
637: \begin{eqnarray}
638: \frac{d^{2}r}{d\tau ^{2}}-\gamma \frac{dr}{d\tau }+\omega _{0}^{2}r & = & 0\label{req} \\
639: \frac{d^{2}R}{d\tau ^{2}}+\gamma \frac{dR}{d\tau }+\omega _{0}^{2}R & = & \! \! \! -i\! \! \! \int _{0}^{t}\! \! \! d\tau '\, Re\! \left\langle \hat{F}(\tau )\hat{F}(\tau ')\right\rangle \! r(\tau ')\label{Req}
640: \end{eqnarray}
641:
642:
643: Here \( \gamma \equiv \eta /m \) is the damping rate and \( \omega _{0} \)
644: is the unperturbed frequency of the oscillator. Note that the second
645: equation leads to a complex-valued solution \( R(\cdot ) \). It is
646: also possible to formulate the calculation slightly differently\cite{callegg,hakimambeg},
647: by using stationary solutions with respect to the real part of \( S \)
648: only. In any case, inserting the solutions \( R \) (\( \equiv R_{cl} \))
649: and \( r \) (\( \equiv r_{cl} \)) into the action \( S \) shows
650: that the imaginary part of the action \( S_{cl}\equiv S[q_{cl}^{>},q^{<}_{cl}] \)
651: is determined directly only by \( S_{I} \). As expected, \( S_{cl} \)
652: turns out to be a bilinear expression in the endpoints \( R_{t},\, R_{0},\, r_{t},\, r_{0} \):
653:
654: \begin{eqnarray}
655: Re\, S_{cl} & = & S_{0}^{>}+S_{0}^{<}+S_{R}=\nonumber \label{res} \\
656: & & R_{t}r_{t}L_{tt}+R_{t}r_{0}L_{t0}+R_{0}r_{t}L_{0t}+R_{0}r_{0}L_{00}\label{res} \\
657: Im\, S_{cl} & = & S_{I}=r_{0}^{2}C_{00}+2r_{0}r_{t}C_{0t}+r_{t}^{2}C_{tt}\, .\label{ims}
658: \end{eqnarray}
659:
660:
661: All the coefficients are real-valued functions of time. For our purposes,
662: it is sufficient to know that the entries of the matrix \( L \) are
663: independent of \( S_{I} \) (while they do depend on the damping \( \gamma \)
664: from \( S_{R} \)). In contrast, the entries of \( C \) depend both
665: on \( S_{I} \) and \( S_{R} \). The dependence on \( S_{R} \) arises
666: only because the path \( r \), which is inserted into Eq.~(\ref{silin}),
667: is affected by the friction described by \( S_{R}, \) see Eq.~(\ref{req}).
668: The entries of \( C \) are proportional to the strength of the bath
669: fluctuations (i.e. the magnitude of \( \left\langle FF\right\rangle _{\omega } \)).
670: Explicit expressions for all of these coefficients\cite{5} can be
671: found in Ref.~\onlinecite{callegg} (cf. their section 6). For reference,
672: the quantities used in the following discussion have been listed in
673: appendix \ref{appc00}.
674:
675: At this point, evaluation of the coefficients for \( Im\, S_{cl} \)
676: shows that they will grow in time beyond all bounds, regardless of
677: whether the path \( r \) is calculated by taking into account \( S_{R} \)
678: or neglecting it (setting \( \gamma =0 \) in Eq.~(\ref{req})).
679: However, no conclusion about dephasing and decay can be drawn from
680: this, since the foregoing discussions lead us to expect that any potential
681: cancellation between \( S_{R} \) and \( S_{I} \) will take place
682: only \emph{after} proper integration over the initial density matrix.
683: This can be deduced from the derivation of the Golden Rule decay rate
684: given in the preceding section. It is also very reasonable that \( S_{I} \)
685: grows without bounds for any semiclassical path \( r(\cdot ) \),
686: since all paths contribute more or less to the time-evolution of all
687: eigenstates, and decay of excited states is perfectly correct in the
688: model of the damped harmonic oscillator.
689:
690: The initial ground state density matrix has the form
691:
692: \begin{eqnarray}
693: \rho _{0}(q_{0}^{>},q_{0}^{<})\propto \exp \left( -(q_{0}^{>2}+q_{0}^{<2})/(4\left\langle \hat{q}^{2}\right\rangle _{0})\right) & & \nonumber \\
694: =\exp \left( -(R_{0}^{2}+r_{0}^{2}/4)/(2\left\langle \hat{q}^{2}\right\rangle _{0})\right) \, , & & \label{indens}
695: \end{eqnarray}
696:
697:
698: with \( \left\langle \hat{q}^{2}\right\rangle _{0}=1/(2m\omega _{0}) \)
699: as the square of the ground state width. After performing the Gaussian
700: integrals over \( R_{0} \) and \( r_{0} \), which are needed to
701: evaluate Eq.~(\ref{rhoJ}), we obtain the final result for the density
702: matrix \( \rho (q^{>}_{t},q^{<}_{t},t) \) at a time \( t \) after
703: switching on the interaction with the bath. It has the general form
704:
705: \begin{equation}
706: \frac{1}{\tilde{N}(t)}\exp \left( -(aR_{t}^{2}+bR_{t}r_{t}+cr_{t}^{2})\right) \, .
707: \end{equation}
708:
709:
710: In particular, the prefactor of \( R_{t}^{2} \) gives the width of
711: the probability distribution at time \( t \),
712:
713: \begin{equation}
714: \left\langle \hat{q}^{2}(t)\right\rangle =\left( 2a\right) ^{-1}\, .
715: \end{equation}
716:
717:
718: Let us now look at the behaviour of the width in order to analyze
719: the effect of dropping \( S_{R} \) in the calculation. The expression
720: for \( \left\langle \hat{q}^{2}(t)\right\rangle \) has been derived
721: already by Caldeira and Leggett\cite{callegg} (cf. their Eq.~6.34):
722:
723: \begin{equation}
724: \label{width}
725: \left\langle \hat{q}^{2}(t)\right\rangle =\left( 2C_{00}+m\omega _{0}/2+L_{00}^{2}/(2m\omega _{0})\right) /L_{t0}^{2}\, .
726: \end{equation}
727:
728:
729: This expression clearly contains quantities both from the imaginary
730: part (\( C_{00} \)) and from the real part (\( L_{00},\, L_{t0} \))
731: of the action. Formally speaking, real and imaginary parts have become
732: intermixed when integrating over \( R_{0} \), due to quadratic completion
733: of the expression \( iR_{0}(r_{t}L_{0t}+r_{0}L_{00})-m\omega _{0}R_{0}^{2} \)
734: in the exponent (the former term is from Eq. (\ref{res}), while the
735: latter term is from the initial density matrix, Eq. (\ref{indens})).
736: Physically, this result is to be expected, since the growth of the
737: width of the probability distribution will be governed by the balance
738: between the bath fluctuations (\( S_{I} \)) and the friction (\( S_{R} \)).
739:
740: The complete time-evolution of \( \left\langle \hat{q}^{2}(t)\right\rangle \)
741: after switching on the interaction at \( t=0 \) is displayed in Fig.~\ref{figeins}.
742: This includes cases where the value of the damping constant \( \gamma \)
743: has been artificially set to zero or to other values different from
744: that prescribed by the fluctuation-dissipation theorem (FDT), which
745: connects \( \gamma \) to the strength of the bath fluctuations described
746: by \( \left\langle FF\right\rangle _{\omega } \) (see Eq. (\ref{ohmbathdef})
747: for the case of \( T=0 \)). If \( S_{R} \) is neglected completely
748: (no friction, \( \gamma =0 \)), heating takes place, causing the
749: variance to grow linearly in time, at a rate given by the Golden Rule
750: expression involving the \emph{symmetrized} bath correlator. Formally,
751: the growth of \( C_{00} \) with time cannot be compensated in that
752: case, since \( L_{00} \) and \( L_{t0} \) acquire their original
753: unperturbed values. For small \( \gamma >0 \), the width saturates
754: at a value much larger than that of the ground state. On the other
755: hand, if the friction is too strong, the width saturates at a value
756: below that of the ground state, and the Heisenberg uncertainty relation
757: is violated (since \( \left\langle \hat{p}^{2}\right\rangle \) also
758: shrinks). This point is discussed in the book of Milonni\cite{milonni}
759: in the context of an atom interacting with the vacuum electromagnetic
760: field. If, however, \( \gamma \) has the correct value prescribed
761: by the FDT, \( \left\langle \hat{q}^{2}\right\rangle \) changes
762: only slightly from its unperturbed value.
763:
764: In fact, for a linear system like the quantum harmonic oscillator,
765: its behaviour under the action of the linear bath can be described
766: and understood entirely using a classical picture\cite{weiss}. The
767: reason for this is as follows: one starts out with an initial state
768: for system and bath that has a positive definite Gaussian Wigner-density,
769: which can be interpreted directly as a classical phase space density
770: whose time-evolution then corresponds one-to-one to the evolution
771: of the Wigner-density (see appendix \ref{appWigner}). Thus, the quantum
772: dissipative dynamics may be described by a \emph{classical} Langevin
773: equation including friction and a fluctuating driving force:
774:
775: \begin{equation}
776: \label{langevin}
777: \frac{d^{2}q}{dt^{2}}+\gamma \frac{dq}{dt}+\omega _{0}^{2}q=\frac{1}{m}F(t)\, .
778: \end{equation}
779:
780:
781: In this equation, the damping term incorporates the effects of \( S_{R} \).
782: It counteracts the fluctuations of \( F(\cdot ) \) that are described
783: by \( S_{I} \) in the path-integral picture and whose correlation
784: function is, therefore, given by the \emph{symmetrized} part of the
785: bath correlator.
786:
787: The behaviour of the width has been discussed above using the influence
788: functional, but it can be understood more easily in this picture.
789: We can use the susceptibility of the damped oscillator to find its
790: response to the force fluctuations \( F \), which, at zero temperature,
791: are entirely due to the zero-point fluctuations. In particular, for
792: the limit \( t\rightarrow \infty \), we obtain
793:
794: \begin{equation}
795: \label{sigmalimit}
796: \left\langle \hat{q}^{2}(t)\right\rangle \rightarrow \int _{-\infty }^{+\infty }d\omega \, \left| \chi (\omega )\right| ^{2}\left\langle FF\right\rangle _{\omega }\, ,
797: \end{equation}
798: where
799:
800: \begin{equation}
801: \label{susc}
802: \chi (\omega )=\frac{1/m}{\omega _{0}^{2}-\omega ^{2}-i\omega \gamma }
803: \end{equation}
804: is the susceptibility of the damped oscillator, and \( \left\langle FF\right\rangle _{\omega } \)
805: is the symmetrized power spectrum (see Eq. (\ref{ohmbathdef})). For
806: the correct equilibrium state (obtained by fully keeping \( S_{R} \)
807: and \( S_{I} \)), the width at \( t\rightarrow \infty \) can also
808: be found by applying the FDT to the oscillator coordinate (cf. Eq.~6.37
809: of Ref.~\onlinecite{callegg}). In contrast, the effect of dropping
810: \( S_{R} \) is obtained by letting \( \gamma \rightarrow 0 \) in
811: Eq. (\ref{langevin}), while \( \left\langle FF\right\rangle _{\omega } \)
812: is kept fixed.
813:
814:
815: \begin{figure}
816: {\centering \resizebox*{0.95\columnwidth}{!}{\includegraphics{fig1.eps}} \par}
817:
818:
819: \caption{\label{figeins}The variance \protect\( \sigma ^{2}(t)\equiv \left\langle \hat{q}^{2}(t)\right\rangle \protect \)
820: of the probability density \protect\( \rho (q,q,t)\protect \) for
821: the damped harmonic oscillator, plotted in units of the unperturbed
822: ground state variance \protect\( \sigma _{0}^{2}\protect \), as a
823: function of time \protect\( t\protect \) for the following friction
824: strengths (top to bottom): \protect\( \gamma /(2\pi \left\langle FF\right\rangle _{\omega =1})=0,\, 0.5,\, 1,\, 2\protect \)
825: (other parameters are \protect\( m=\omega _{0}=1\protect \)). A ratio
826: of \protect\( 1\protect \) corresponds to the correct value prescribed
827: by the FDT, while \protect\( \gamma =0\protect \) means \protect\( S_{R}\protect \)
828: has been neglected completely, such that \protect\( \sigma ^{2}(t)\protect \)
829: grows without bound. The dashed lines give the limit for \protect\( t\rightarrow \infty \protect \)
830: (see Eq.~(\ref{sigmalimit})).}
831: \end{figure}
832:
833:
834: The example of the damped harmonic oscillator demonstrates that the
835: proper behaviour near the ground state of the system cannot be observed
836: at the early stages of the calculation, by looking at the action evaluated
837: along classical paths (Eqs.~(\ref{res}),~(\ref{ims})), regardless
838: of whether these paths properly include the damping or not. Only after
839: correct integration over the initial state density matrix the contributions
840: from \( S_{R} \) and \( S_{I} \) can compensate each other (in Eq.~(\ref{width})),
841: such that no artificial finite decay rate of the ground state results,
842: if \( S_{R} \) is kept.
843:
844:
845: \section{Power-law behaviour in quantum Brownian motion}
846:
847: \label{qubrownmotion}We now turn to a discussion of the free particle,
848: i.e. the Caldeira-Leggett model of quantum Brownian motion. There,
849: the reasoning is the same as before in principle: Neglecting \( S_{R} \)
850: means that the zero-point fluctuations of the bath contained in \( S_{I} \)
851: may heat up the particle. However, there is an important difference:
852: The spectrum of the free motion does not contain any resonance peak,
853: unlike that of the harmonic oscillator. Therefore, only the low-frequency
854: components (\( \omega \rightarrow 0 \)) of the bath spectrum will
855: contribute to heating and decay (compare Eq.~(\ref{overlapspectra})),
856: which results in a peculiar long-time behaviour\cite{callegg,hakimambeg,schramm}
857: that cannot be captured using the Golden Rule, as has been emphasized
858: in Ref.~\onlinecite{GZ_CL}.
859:
860: One can observe the distinction between oscillator and free particle
861: most easily in the picture of the classical Langevin equation (\ref{langevin})
862: introduced above. In appendix \ref{appWigner}, it is shown how the
863: density matrix propagator \( J \) may be obtained in general via
864: the time-evolution of the Wigner density. This approach is physically
865: more transparent than the path-integral calculation. For the purposes
866: of our present discussion, however, it will suffice to consider the
867: time-evolution of the diagonal elements of the density matrix in momentum
868: space. These have also been analyzed in Ref.~\onlinecite{GZ_CL},
869: in order to demonstrate the failure of the Golden Rule calculation
870: to describe the decay of the original ground state at zero temperature.
871: We will confirm and explain the outcome of this analysis using a simple
872: argument based on the Langevin equation (\ref{langevin}). Then we
873: will point out the role of \( S_{R} \) and describe the important
874: difference to the damped oscillator.
875:
876: If the particle has momentum \( p_{0} \) at time \( 0 \), its momentum
877: \( p \) at time \( t \) will be determined by the fluctuating force
878: \( F(\cdot ) \) in the following way (by solving Eq. (\ref{langevin})
879: for \( \omega _{0}=0 \)):
880:
881: \begin{equation}
882: \label{pevol}
883: p=p_{0}e^{-\gamma t}+\int _{0}^{t}ds\, e^{-\gamma (t-s)}F(s)\, .
884: \end{equation}
885:
886:
887: Since \( F(\cdot ) \) is a Gaussian random process, the probability
888: density of finding a momentum \( p \) at time \( t \) is a Gaussian,
889: of variance
890:
891: \begin{equation}
892: \label{pgaussian}
893: \left\langle \delta p^{2}\right\rangle \equiv \int _{0}^{t}ds_{1}\int _{0}^{t}ds_{2}e^{-\gamma (2t-s_{1}-s_{2})}\left\langle F(s_{1})F(s_{2})\right\rangle \, .
894: \end{equation}
895:
896:
897: The behaviour of \( \left\langle \delta p^{2}\right\rangle \) as
898: a function of time is as follows (at \( T=0 \)): As long as \( t\ll 1/\gamma \),
899: friction is unimportant and it is the double time-integral over the
900: correlator of \( F \) which is to be evaluated. For the Ohmic bath
901: considered here, the power spectrum is relatively strong at low frequencies
902: (rising linearly with \( \omega \), see Eq. (\ref{ohmbathdef})).
903: This leads to a slow decay in time, \( \left\langle F(t)F(0)\right\rangle \propto 1/t^{2} \),
904: which results in a logarithmic growth of \( \left\langle \delta p^{2}\right\rangle \),
905:
906: \begin{equation}
907: \label{p2log}
908: \left\langle \delta p^{2}\right\rangle \approx 2\frac{\eta }{\pi }\ln (\omega _{c}t)\, .
909: \end{equation}
910: Here \( \omega _{c} \) is the cutoff frequency of the bath spectrum
911: and we have assumed \( \omega _{c}t\gg 1 \). It is this logarithmic
912: behaviour which cannot be obtained using the Golden Rule approximation,
913: to be discussed below. At later times, \( t\gg 1/\gamma \), the
914: growth saturates at a constant value,
915:
916: \begin{equation}
917: \left\langle \delta p^{2}\right\rangle \rightarrow \frac{\eta }{\pi }\ln (\omega _{c}/\gamma )\, .
918: \end{equation}
919:
920:
921: The perturbation expansion in the coupling between system and bath
922: can be carried out by expanding the influence action in the propagator
923: \( J \) (Eq. (\ref{prop})) in powers of \( iS_{R}-S_{I} \) (see
924: Eq. (\ref{eq13}) in the present article or Eq. (E13) of Ref. \onlinecite{GZ_CL}).
925: We want to discuss the time-evolution of the diagonal elements \( \rho _{pp}(t) \)
926: of the density matrix in momentum space. The evolution starts from
927: the equilibrium density matrix, which is a Gaussian momentum distribution
928: of variance \( \left\langle p_{0}^{2}\right\rangle =mT \) (the following
929: formulas will be analyzed for arbitrary \( T \)). After a time \( t \),
930: the density matrix is still a Gaussian, but of variance (compare Eq.
931: (\ref{pevol})):
932:
933: \begin{equation}
934: \label{momvar}
935: \left\langle p^{2}\right\rangle =\left\langle p_{0}^{2}\right\rangle e^{-2\gamma t}+\left\langle \delta p^{2}\right\rangle \, .
936: \end{equation}
937:
938:
939: From this we can easily derive the result for \( \delta \rho ^{(1)}_{pp}(t) \)
940: which could alternatively be obtained by first expanding \( J \)
941: to first order in \( iS_{R}-S_{I} \), going over to the momentum
942: representation and finally integrating over the initial density matrix
943: (as was done in Ref.~\onlinecite{GZ_CL}; see Eq. (E18), which is,
944: however, still written in terms of matrix elements of the coordinate
945: operator):
946:
947: \begin{eqnarray}
948: & & \rho _{pp}(t)=\frac{1}{\sqrt{2\pi \left\langle p^{2}\right\rangle }}\exp \left[ -\frac{p^{2}}{2\left\langle p^{2}\right\rangle }\right] \nonumber \\
949: & & \approx \rho _{pp}(0)\cdot (1+\frac{\left\langle p^{2}\right\rangle -\left\langle p_{0}^{2}\right\rangle }{2\left\langle p_{0}^{2}\right\rangle }\left( \frac{p^{2}}{\left\langle p_{0}^{2}\right\rangle }-1\right) +\cdots )\, .\label{rhoexpansion}
950: \end{eqnarray}
951:
952:
953: In evaluating \( \left\langle p^{2}\right\rangle \), only the terms
954: up to first order in \( \gamma \) and \( \left\langle FF\right\rangle _{\omega } \)
955: must be kept for the purposes of this expansion. At finite temperature
956: \( T \), the behaviour of \( \left\langle p^{2}\right\rangle \)
957: at long times \( t\gg 1/T \) is governed by the linear decrease stemming
958: from \( \left\langle p_{0}^{2}\right\rangle e^{-2\gamma t}\approx \left\langle p_{0}^{2}\right\rangle (1-2\gamma t) \)
959: and the linear increase from \( \left\langle \delta p^{2}\right\rangle \approx 2\eta Tt \).
960: If the initial density matrix really describes the equilibrium distribution,
961: then detailed balance holds and these terms cancel. For times \( t\ll 1/T \)
962: (or arbitrary times at \( T=0 \)), the time-evolution of \( \left\langle p^{2}\right\rangle \)
963: is governed by \( \left\langle \delta p^{2}\right\rangle \), evaluated
964: for the correlator \( \left\langle FF\right\rangle _{\omega } \)
965: taken at \( T=0 \) (which only contains the zero-point fluctuations
966: of the bath). For consistency of the expansion, \( \left\langle \delta p^{2}\right\rangle \)
967: in Eq. (\ref{pgaussian}) has to be evaluated by setting \( \gamma =0 \).
968: Therefore, only the discussion given above for times \( t\ll 1/\gamma \)
969: turns out to be relevant. The logarithmic growth of \( \left\langle \delta p^{2}\right\rangle \)
970: given in Eq. (\ref{p2log}) corresponds to a power-law behaviour of
971: the (exact) time-evolution of \( \rho _{pp}(t) \) for \( t\ll 1/\gamma \).
972: This logarithm, that appears in the perturbation expansion (\ref{rhoexpansion}),
973: is {}``overlooked'' in the Golden Rule approximation, where only
974: terms growing linearly with time are kept. Therefore, the Golden Rule
975: rate turns out to vanish\cite{GZ_CL}. These power-laws are characteristic
976: of the density matrix propagator \( J \) of quantum Brownian motion
977: at zero temperature (see appendix \ref{appWigner}).
978:
979: On the other hand, the effect of neglecting \( S_{R} \) may be analyzed
980: by setting the friction constant \( \gamma \) to zero in Eq. (\ref{pgaussian}).
981: Then, the spread \( \left\langle \delta p^{2}\right\rangle \) of
982: the momentum distribution grows without bounds, which is obviously
983: not the correct physical behaviour. It is qualitatively similar to
984: the heating produced in the damped oscillator model. However, since
985: the growth proceeds only logarithmically in time, the artefacts of
986: this approximation, when applied to the free particle, are not nearly
987: as drastic as the finite decay rate of the ground state observed for
988: the oscillator. In particular, within perturbation theory, no qualitative
989: change is obtained by dropping \( S_{R} \) for the free particle
990: (no finite decay rate is produced).
991:
992: Furthermore, if we had considered a super-Ohmic bath (e.g. due to
993: phonons), whose spectrum decays faster than \( \omega ^{1} \) for
994: \( \omega \rightarrow 0 \), the growth of \( \left\langle p^{2}\right\rangle \)
995: would saturate even without friction (as it should, since there is
996: no velocity-dependent friction for such a bath). In contrast, the
997: behaviour of the harmonic oscillator discussed above would remain
998: qualitatively the same as for the Ohmic bath, since its decay depends
999: on the bath spectrum at the resonance frequency of the oscillator
1000: and is not affected in any essential way by the details of the spectrum
1001: at low frequencies. This is consistent with the fact that in the oscillator
1002: case already the Golden Rule is sufficient to obtain an essentially
1003: correct picture of the artificial decay produced by dropping \( S_{R} \),
1004: while, for quantum Brownian motion, it fails to describe the subtle
1005: power-laws associated with the low-frequency (long-time) properties
1006: of the Ohmic bath at zero temperature.
1007:
1008: The difference between oscillator and free particle also shows up
1009: clearly in the behaviour of the imaginary part of the influence action:
1010: For the free particle, \( S_{I} \) \emph{always} grows only logarithmically
1011: with time \( t \) (at \( T=0 \)), independent of whether one keeps
1012: \( S_{R} \) or sets it to zero (i.e. \( \gamma \equiv 0 \) in the
1013: equations of motion). This can be derived from the formulas given
1014: in appendix \ref{appWigner} or those of Ref.~\onlinecite{GZ_CL},
1015: Appendix E (or Ref.~\onlinecite{callegg}, section 6, with the proper
1016: limit \( \omega _{R}\rightarrow 0 \) for the free particle).
1017:
1018: Hence, we conclude that for the oscillator (or any motion containing
1019: an extended spectrum) the effects of dropping \( S_{R} \) are much
1020: more pronounced than for the free motion, since they lead to an artificial
1021: finite decay rate. We have also observed that this effect can already
1022: be understood within the framework of the Golden Rule approximation,
1023: whereas that approximation is incapable of describing the more subtle
1024: {}``power-law decay'' found for the free particle.
1025:
1026:
1027: \section{Qualitative discussion of other models: Nonlinear coupling and Pauli
1028: principle}
1029:
1030: \label{section7}It could be argued that our discussion of the damped
1031: oscillator has only demonstrated the importance of \( S_{R} \) in
1032: preventing an artificial decay of the \emph{ground state} and has
1033: little to do with dephasing. Indeed, if one wants to discuss dephasing,
1034: one should rather consider the decay of a coherent superposition of
1035: the ground state and some excited state. This is relevant both for
1036: arbitrary nonequilibrium situations as well as for the calculation
1037: of the system's linear response, where the perturbation creates a
1038: superposition of excited states and the ground state. In the example
1039: of the damped harmonic oscillator, full relaxation into the ground
1040: state will \emph{always} take place (at \( T=0 \)), for any such
1041: excited state. This holds regardless of the bath spectrum, provided
1042: the latter does not vanish around the resonance frequency of the oscillator.
1043: For weak coupling, the decay of the density matrix (and, therefore,
1044: the linear response) may be described conveniently using a simple
1045: master equation approach. What is more, even if one neglects \( S_{R} \)
1046: in a path-integral calculation, the predicted decay rates will be
1047: correct, excepting only that for the ground state, which we have discussed
1048: above. The reason for this, however, is a specialty of the linearly
1049: damped harmonic oscillator. Since the bath couples to the coordinate
1050: operator, whose matrix elements only connect adjacent oscillator levels,
1051: the relevant {}``system correlator'' (in Eq.~(\ref{systemspec}))
1052: turns out to be symmetric for any state but the lowest one. Therefore,
1053: the decay rate will depend only on the symmetric part of the bath
1054: correlator as well, see Eq. (\ref{pathmaster}). Hence, for all the
1055: excited states, the evaluation of the total decay rate is not affected
1056: by dropping \( S_{R} \). This point has already been discussed in
1057: connection with the model of two oscillating wave packets, at the
1058: end of section \ref{section2}.
1059:
1060: Since we are interested in demonstrating the importance of \( S_{R} \)
1061: for the dephasing of a superposition of low-lying levels, the natural
1062: choice is to consider a bath with an excitation gap larger than the
1063: resonance frequency of the oscillator (see Fig.~\ref{fig2}). In
1064: that case, we expect the lowest levels to remain coherent in the correct
1065: description. Unfortunately, in the model of the linearly damped oscillator
1066: discussed above, we would not obtain any finite decay rate, \emph{regardless}
1067: of whether \( S_{R} \) is kept or not. This is because the corresponding
1068: classical noise force is out of resonance and cannot heat the oscillator.
1069:
1070: Therefore, in order to analyze a situation where the importance of
1071: retaining \( S_{R} \) is displayed even for excited states, we have
1072: to go beyond exactly solvable linear models. As a consequence, the
1073: following discussion is necessarily incomplete insofar as we cannot
1074: give exact proofs of the statements to be made below. These statements
1075: will be based on the experience acquired in the simpler models discussed
1076: above. We will also make use of some Golden Rule type arguments, noting
1077: that the Golden Rule has been sufficient to understand the behaviour
1078: of the damped oscillator in a qualitatively correct way. In any case,
1079: we believe it is useful to contrast the results obtained by dropping
1080: \( S_{R} \) with the {}``commonly accepted'' picture.
1081:
1082: Let us, therefore, consider the following model: We retain the harmonic
1083: oscillator, but change the coupling between system and bath from \( \hat{V}=-\hat{q}\hat{F} \)
1084: to \( \hat{V}=-f(\hat{q})\hat{F} \), with a \emph{nonlinear} function
1085: \( f \). This is called state-dependent friction\cite{weiss}. In
1086: that case, the transition matrix elements of the system operator \( f(\hat{q}) \)
1087: will, in general, be nonzero between any two states, which is the
1088: important difference to the linear case. If we introduce a bath spectrum
1089: with a gap larger than \( n\omega _{0} \), then, for sufficiently
1090: weak coupling and according to a simple Golden Rule calculation, the
1091: first \( n \) excited states will not decay at zero temperature.
1092: Any coherent superposition of these states will therefore remain coherent.
1093: The full, non-perturbative picture will be slightly more complicated,
1094: but the essential point should remain the same: The first excited
1095: states will acquire an admixture of other levels and become entangled
1096: with the states of the bath. They will experience some frequency shifts
1097: and the transition matrix elements will be renormalized. Of course,
1098: the new eigenstates of the coupled system will remain in a coherent
1099: superposition forever (by definition). The \( n+1 \) lowest ones
1100: still have a discrete spectrum and are in direct correspondence to
1101: the initial unperturbed states. When switching on the interaction
1102: at \( t=0 \), a partial decay will result. Physically, this corresponds
1103: to the relaxation of the initial state into the selfconsistent coupled
1104: state of system and bath (see Ref.~\onlinecite{hakimambeg} for a
1105: discussion of these issues in the case of the free particle). However,
1106: the initial decay will saturate in time on a short time-scale. During
1107: this transient adjustment, the reduced system density matrix becomes
1108: mixed to a small extent (if the coupling is weak), but no non-saturated
1109: long-term decay (indicated by a finite decay rate) results. In any
1110: case, the transient decay is an artefact of the procedure of suddenly
1111: switching on the interaction. A full calculation of, for example,
1112: the linear response properties of the dissipative system would start
1113: out with the selfconsistent ground state of the fully interacting
1114: system (as was done in Ref.~\onlinecite{schramm} for the free particle).
1115:
1116:
1117: \begin{figure}
1118: {\centering \resizebox*{0.95\columnwidth}{!}{\includegraphics{fig2.eps}} \par}
1119:
1120:
1121: \caption{\label{fig2}The models described in the text. \emph{Left}: A bath
1122: with a gap would lead to the upward transitions indicated by grey
1123: arrows if (and only if) \protect\( S_{R}\protect \) were neglected.
1124: \emph{Right}: A sea of fermions may block downward transitions (dashed)
1125: even though they are possible in the single-particle picture. }
1126: \end{figure}
1127:
1128:
1129: In contrast, dropping \( S_{R} \) would lead to a completely different
1130: picture, since then \( S_{I} \) would correspond to a classical noise
1131: force. Starting from any of the first \( n \) excited states, the
1132: fluctuating field would always be able to induce transitions \emph{up}
1133: in energy, thus leading to a finite decay rate. The role of \( S_{R} \)
1134: in the correct description of a quantum noise force is precisely to
1135: cancel these upward transitions, as has been demonstrated in the case
1136: of the linearly damped oscillator. Due to this finite decay rate,
1137: any initially coherent superposition would get destroyed, leading
1138: to complete dephasing. No saturation of the decay could be observed,
1139: in spite of the gap in the bath spectrum.
1140:
1141: It is \emph{this} behaviour that is described qualitatively correctly
1142: by evaluating \( S_{I} \) along semiclassical paths of the oscillator.
1143: The expression for \( S_{I} \), Eq.~(\ref{silin}), is now to be
1144: changed simply by replacing \( q \) by \( f(q) \). While an (unperturbed)
1145: oscillator path of the form \( q(\tau )\propto \cos (\omega _{0}t) \)
1146: only contains frequencies \( \pm \omega _{0} \) which cannot couple
1147: to the gapped bath, the function \( f(q(\tau )) \) will, in general,
1148: contain all higher harmonics as well. This directly corresponds to
1149: the character of the \emph{symmetrized} system spectrum, i.e. the
1150: Fourier transform of \( \left\langle \left\{ f(\hat{q}(\tau )),f(\hat{q}(0))\right\} \right\rangle \).
1151: These frequencies contained in the system motion then will couple
1152: to the bath spectrum, leading to an unbounded growth of \( S_{I} \)
1153: with time.
1154:
1155: Similarly, if we were to use a bath without a gap, but where the spectrum
1156: falls off fast towards low frequencies, then the correct description
1157: would also yield decay rates that quickly become smaller when going
1158: towards the ground state, while they would saturate at a finite, comparatively
1159: large level if \( S_{R} \) were dropped.
1160:
1161: The present work has been motivated in part by the ongoing controversial
1162: debate on low-temperature dephasing in an interacting disordered fermion
1163: system\cite{mohantywebb,GZ,GZ_CL,GZ_PB,cohenimry,VavAmbeg,AAGcritique,BelitzKirckpatrick},
1164: which has led to a revival of interest for the general question of
1165: dephasing in mesoscopic systems\cite{CedrBuett,21,seelig,buettHO,Guinea,ghz}.
1166: Since the usual Feynman-Vernon influence functional deals only with
1167: a single-particle situation, it cannot be applied directly to this
1168: problem. However, the authors of Refs.~\onlinecite{GZ,GZ_CL,GZ_PB}
1169: have succeeded in deriving an extension of the usual influence functional
1170: to the many-fermion situation, using an exact procedure. In their
1171: result, the Fermi distribution (and, therefore, the Pauli principle)
1172: only enters \( S_{R} \), while \( S_{I} \) is unaffected (Eqs. (54)
1173: and (55) of Ref.~\onlinecite{GZ}). Therefore, the dephasing rate,
1174: as read off from \( S_{I} \), is found to equal the rate which would
1175: also be obtained in a purely single-particle calculation. Although
1176: we do not analyze the full problem of weak-localization here, we will
1177: use the insights gained above in order to explain why, in our opinion,
1178: the \emph{assumption} that the dephasing rate can be derived from
1179: \( S_{I} \) alone must be proven instead of being taken for granted.
1180: This holds even when \( S_{R} \) vanishes on the relevant pairs of
1181: classical time-reversed trajectories, since, in general, the integration
1182: over the fluctuations away from these trajectories will be essential
1183: for obtaining a cancellation between the effects of \( S_{R} \) and
1184: \( S_{I} \) (see sec. \ref{section4}), in addition to properly taking
1185: into account the initial density matrix. Furthermore, we note that
1186: in these calculations the detailed form of the bath spectrum at low
1187: frequencies turns out to be unimportant for the essential result of
1188: a finite zero-temperature dephasing rate. This is in marked contrast
1189: to the case of the free particle (where the Ohmic bath plays a special
1190: role\cite{schramm}; see also Refs. \onlinecite{Guinea,ghz}), but
1191: similar to what is observed for the damped harmonic oscillator when
1192: \( S_{R} \) is neglected. It is for this reason that we have chosen
1193: a bath with an excitation gap in order to demonstrate the importance
1194: of \( S_{R} \), since there the effects come out most clearly, although
1195: they are also present for other bath spectra (compare the discussion
1196: above).
1197:
1198: The model situation of the damped oscillator described above already
1199: contains one key ingredient related to the description of dephasing
1200: for an electron moving inside a disordered metal, namely an extended
1201: system spectrum: Due to the impurity scattering, the system spectrum
1202: (e.g., of the velocity operator) contains frequencies up to (at least)
1203: the elastic scattering rate, which may couple to the zero-point fluctuations
1204: contained in the bath spectrum. This is in contrast to the free particle,
1205: which can only couple to the low-frequency bath modes (see the discussion
1206: at the end of section \ref{qubrownmotion}). As before, the importance
1207: of \( S_{R} \) will be visible most easily for a bath containing
1208: an excitation gap. For such a bath, the \emph{free} motion (without
1209: impurity scattering) will not show any nontransient decay, \emph{regardless}
1210: of whether \( S_{R} \) is taken into account or not. In contrast,
1211: the model to be discussed below, which is more similar to the situation
1212: in the disordered metal, will show a finite decay rate of low-lying
1213: levels if and only if \( S_{R} \) is neglected.
1214:
1215: Apart from the extended system spectrum, we have to take into account
1216: another important feature of the calculations\cite{GZ,GZ_CL,GZ_PB,cohenimry,AAK}
1217: concerning electrons in a disordered metal: the use of the semiclassical
1218: analysis. For linear systems the semiclassical result for the path-integral
1219: is exact, such that it can even be used near the ground state of the
1220: system, as we have done it here. However, in such a case, the size
1221: of the fluctuations around the path is comparable to the amplitude
1222: of the path itself, which is not the situation in which semiclassics
1223: is usually applicable. In order to gain intuition for a typical semiclassical
1224: situation, we could reconsider the example with the nonlinear coupling
1225: to a gapped bath given above: Imagine an initial superposition of
1226: fast wavepackets, whose wavelength is much smaller than the packet
1227: size but whose excitation energies are still below the bath threshold.
1228: They would show a finite decay or dephasing rate (for the reasons
1229: given above), if only \( S_{I} \) were used for the analysis, but
1230: not in the full calculation. Still, for this scenario one might argue
1231: that it has been clear from the outset that the semiclassical analysis
1232: cannot be trusted, since those high-frequency components in the system
1233: spectrum that are responsible for the decay are necessarily larger
1234: than the energetic distance to the ground state.
1235:
1236: It is only in a system of degenerate fermions that the following two
1237: conditions can be fulfilled all at once: On the one hand, the semiclassical
1238: analysis of a single (non-interacting) electron moving at the Fermi
1239: level is valid for an external potential which is sufficiently slowly
1240: varying, such that the system spectrum (concerning the motion of the
1241: single electron) only contains frequencies much smaller than the Fermi
1242: energy. On the other hand, the whole many-particle system may be near
1243: (or in) its ground state.
1244:
1245: While the first feature would lead one to believe that \( S_{R} \)
1246: may be omitted, we have learned from the examples discussed before
1247: that this is likely to be incorrect whenever the second condition
1248: holds as well. We stress once again that the two conditions are mutually
1249: exclusive in a single-particle problem, which is the reason why such
1250: considerations have not played any role in influence-functional calculations
1251: up to now.
1252:
1253: In order to render the discussion concrete, we can once again make
1254: use of the example with a nonlinear coupling \( f(\hat{q}) \) given
1255: above, provided we suppose the \( N \) lowest oscillator states to
1256: be filled up with fermions initially (see Fig.~\ref{fig2}). For
1257: a relatively smooth function \( f \), the matrix elements of \( f(\hat{q}) \)
1258: only connect states within a range much smaller than \( \epsilon _{F}=N\omega _{0} \),
1259: which, in our example, should still be larger than the gap \( n\omega _{0} \)
1260: of the bath. Dephasing of a \emph{single} particle near \( \epsilon _{F} \)
1261: can then certainly be described fully within the semiclassical analysis,
1262: and, to a good approximation, using \( S_{I} \) alone, as has been
1263: discussed above. The same holds for the many-particle system, if one
1264: explicitly considers a \emph{classical} noise force, where \( S_{R} \)
1265: is absent. In that case, the many-particle problem can really be treated
1266: as a collection of independent single-particle problems, as is the
1267: case for any external time-dependent potential. Only in the end an
1268: average of the full Slater determinant over all possible realizations
1269: of the external noise has to be carried out.
1270:
1271: If the dephasing rate is calculated solely from \( S_{I} \), it turns
1272: out to be finite\cite{GZ,GZ_CL,GZ_PB} and not to depend at all on
1273: the distance to the Fermi surface. This is consistent with the fact
1274: that the the value of \( \epsilon _{F} \) does not even appear in
1275: the single-particle calculation. On the other hand, for a quantum
1276: bath \( S_{R} \) does not vanish and should be included in the influence
1277: action. It is true that for the case of a highly excited \emph{single}
1278: particle the dephasing rate comes out correct, regardless of whether
1279: \( S_{R} \) is kept or not, as we have discussed before. However,
1280: we have also pointed out that there is an important physical difference
1281: between the two calculations. Since in the correct approach (including
1282: \( S_{R} \)), the transitions induced by the bath are purely downwards
1283: in energy (accompanied by spontaneous emission into the bath), it
1284: is reasonable to expect that, in the \emph{many-particle} problem,
1285: they will be blocked by the Pauli principle. In any case, this is
1286: what is found using the Golden Rule. Judging from these arguments,
1287: the contribution to \( S_{I} \) from the nonzero overlap of the (symmetrized)
1288: system spectrum with the high frequencies of the bath spectrum does
1289: not imply dephasing and decay, but rather a renormalization, similar
1290: to that obtained for an electron interacting with optical phonon modes
1291: (a gapped bath), leading to the formation of a polaron. The formation
1292: of the polaron will be visible as an initial transient decay of the
1293: single-electron density matrix (for the artificial case of factorized
1294: initial conditions), which saturates on a short time-scale. Therefore,
1295: to lowest order, for a bath with a gap no finite relaxation rate of
1296: the lowest-lying single-particle excitations above the Fermi sea is
1297: expected, just as for the lower levels of the harmonic oscillator
1298: in the single-particle model with nonlinear coupling to a gapped bath.
1299:
1300: We have to qualify this statement by taking into account the fact
1301: that the coupling between electrons and bath always also induces an
1302: effective interaction between the electrons. In this way, a given
1303: electron becomes coupled indirectly to the bath of other electrons,
1304: such that scattering processes will, indeed, lead to a finite decay
1305: rate even for those low-lying excited levels. This precludes any rigorous
1306: proof demonstrating the complete absence of decay and dephasing for
1307: low-lying levels even in such a rather simple situation, in spite
1308: of the assumption of an excitation gap in the bath spectrum. However,
1309: the consequences of the effective interaction can be distinguished
1310: easily from the finite dephasing rate that would be predicted by looking
1311: at \( S_{I} \) alone. If the coupling to the bath is of strength
1312: \( g \), then the latter rate would go as \( g^{2} \). In contrast,
1313: the effective interaction (obtained after integrating out the bath)
1314: will itself be of strength \( g^{2} \), such that the resulting relaxation
1315: rate (due to coupling of an electron to the density fluctuations of
1316: other electrons) will be of fourth order in \( g \). In addition,
1317: of course, the rate will depend strongly on the distance to the Fermi
1318: surface, vanishing when the Fermi energy is approached.
1319:
1320:
1321: \section{Conclusions}
1322:
1323: We have tried to demonstrate that the real part \( S_{R} \) of the
1324: Feynman-Vernon influence action cannot be neglected in an analysis
1325: of dephasing and decay \emph{near the ground state of a system}, in
1326: spite of the fact that, in the simplest situations (involving highly
1327: excited states and the semiclassical analysis), it is only the imaginary
1328: part \( S_{I} \) which enters the dephasing rate. To this end, we
1329: have discussed how \( S_{R} \) and \( S_{I} \) may cancel each other's
1330: effects not only in lowest-order perturbation theory but also to all
1331: orders, by examining exactly solvable linear quantum dissipative systems.
1332: In general, the cancellation is only found after proper integration
1333: over the fluctuations away from semiclassical trajectories, taking
1334: into account the action of the unperturbed system and its initial
1335: density matrix. Furthermore, we have pointed out an essential difference
1336: between the damped oscillator and the free particle with respect to
1337: this issue. We have argued that the insight obtained in the case of
1338: the damped oscillator is also applicable to nonlinear systems and
1339: important for discussions of dephasing in systems of disordered degenerate
1340: fermions.
1341:
1342: In summary, it may be possible to discuss dephasing and decay without
1343: considering \( S_{R} \) either if the noise is nearly classical (external
1344: nonequilibrium radiation or bath in the high-temperature limit) or
1345: if the system itself is in a highly excited state. Otherwise, dephasing
1346: rates obtained solely from \( S_{I} \) are bound to come out finite
1347: at zero temperature in most cases, even when a simpler Golden Rule
1348: calculation gives vanishing results. Judging from the examples discussed
1349: above, this is probably not because such a nonperturbative procedure
1350: goes beyond the Golden Rule approximation, but because it neglects
1351: some physics already contained even within this approximation.
1352:
1353: Still, although the primary message of this paper is that special
1354: care has to be taken in extracting dephasing rates from calculations
1355: using the influence functional, we emphasize at the same time that
1356: there is no \emph{general} proof showing the impossibility of zero-temperature
1357: dephasing near the ground state. There cannot be such a proof, since
1358: there is evidently at least one counter-example (the Caldeira-Leggett
1359: model of quantum Brownian motion and similar models\cite{callegg,Guinea,ghz}),
1360: and, furthermore, there is no generally applicable definition of {}``dephasing''
1361: that is useful under all conceivable circumstances.
1362:
1363: \begin{acknowledgments}I would like to thank Christoph Bruder, Jan
1364: von Delft, Dimitri Golubev and Andrei Zaikin for stimulating discussions.
1365: This work has been funded by the Swiss National Science Foundation.\end{acknowledgments}
1366:
1367: \appendix
1368:
1369:
1370: \section{Some quantities for the damped oscillator}
1371:
1372: \label{appc00}In this appendix, we list, for purposes of reference,
1373: the quantities relevant to our discussion of the time-evolution of
1374: the quantum damped harmonic oscillator. These can also be found in
1375: Ref. \onlinecite{callegg} (in a slightly different notation).
1376:
1377: The quantity \( C_{00} \) arises in evaluating \( S_{I} \) along
1378: a pair of semiclassical paths, by inserting the solution \( r(\cdot ) \)
1379: of Eq. (\ref{req}) for the boundary conditions given by \( r_{t},\, r_{0} \)
1380: and determining the coefficient of \( r_{0}^{2} \) in the result
1381: (see Eq. (\ref{ims})). Necessarily, \( C_{00} \) contains only the
1382: symmetric part of the bath correlator, since \( S_{I} \) depends
1383: only on that:
1384:
1385: \begin{eqnarray}
1386: C_{00}=\frac{1}{4}\left( \sin (\tilde{\omega }t)\right) ^{-2}\int _{0}^{t}dt_{1}\int _{0}^{t}dt_{2}\, \exp \left( \frac{\gamma }{2}(t_{1}+t_{2})\right) & & \nonumber \\
1387: \times \sin (\tilde{\omega }(t-t_{1}))\left\langle \{\hat{F}(t_{1}),\hat{F}(t_{2})\}\right\rangle \sin (\tilde{\omega }(t-t_{2}))\, . & & \label{C00formula}
1388: \end{eqnarray}
1389:
1390:
1391: Here \( \tilde{\omega }\equiv \sqrt{\omega _{0}^{2}-(\gamma /2)^{2}} \)
1392: is the renormalized frequency of the (underdamped, \( \gamma <2\omega _{0} \))
1393: oscillator.
1394:
1395: The other quantities needed for calculating \( \left\langle \hat{q}^{2}(t)\right\rangle \)
1396: arise from the evaluation of \( Re\, S_{cl} \) (Eq. (\ref{res})).
1397: In contrast to \( C_{00} \), which has to be evaluated numerically,
1398: they can be given in closed form:
1399:
1400: \begin{eqnarray}
1401: L_{00} & = & m(\gamma /2+\tilde{\omega }\cot (\tilde{\omega }t))\\
1402: L_{t0} & = & -\frac{m\tilde{\omega }e^{\gamma t/2}}{\sin (\tilde{\omega }t)}
1403: \end{eqnarray}
1404:
1405:
1406: The exponential increase of \( L_{t0} \) cancels that of \( C_{00} \)
1407: when calculating the width in Eq. (\ref{width}). However, if the
1408: damping rate \( \gamma \) is set to zero, \( C_{00} \) still grows
1409: beyond all bounds while \( L_{t0} \) remains bounded.
1410:
1411:
1412: \section{Density matrix propagator from the Wigner density evolution}
1413:
1414: \label{appWigner}Using the classical Langevin equation (\ref{langevin}),
1415: the kernel \( J \) which relates the reduced density matrix of a
1416: linear damped system at time \( t \) to that at time \( 0 \) can
1417: be found in a way which is physically more transparent than the corresponding
1418: derivation using path-integrals (see Eqs. (\ref{rhoJ}), (\ref{prop}),
1419: (\ref{Jsemi}), (\ref{res}) and (\ref{ims})). One first solves for
1420: the time-evolution of the classical phase space density (i.e. the
1421: Wigner density) under the action of friction and the Gaussian random
1422: force \( F(\cdot ) \). Starting from a \( \delta \) peak located
1423: in phase space, the phase space density evolves into a two-dimensional
1424: Gaussian distribution, whose covariances are related to the correlator
1425: of the force. This gives the propagator \( J^{W} \) of the Wigner
1426: density, which only needs to be Fourier transformed with respect to
1427: the momenta in order to obtain the density matrix propagator \( J \),
1428: expressed via center-of-mass and difference coordinates \( R \) and
1429: \( r \):
1430:
1431: \begin{eqnarray}
1432: J(R_{t},r_{t}|R_{0},r_{0};t)= & & \nonumber \\
1433: \int \, dp_{t}dp_{0}\, e^{i(p_{t}r_{t}-p_{0}r_{0})}J^{W}(R_{t},p_{t}|R_{0},p_{0};t)\, . & & \label{propWigner}
1434: \end{eqnarray}
1435:
1436:
1437: Here we show how the propagator \( J \) for the density matrix of
1438: a free damped particle subject to the Ohmic bath may be obtained in
1439: this way. Everything works the same for the damped oscillator, only
1440: the resulting expressions are slightly more lengthy.
1441:
1442: The propagator \( J^{W}(R_{t},p_{t}|R_{0},p_{0};t) \) of the Wigner
1443: density is found by solving the classical equations of motion for
1444: \( R \) and \( p \) for a given initial condition \( (R_{0},p_{0} \)),
1445: taking into account friction and the action of the force \( F \):
1446:
1447: \begin{eqnarray}
1448: \frac{dp}{dt} & = & -\gamma p+F(t)\\
1449: \frac{dR}{dt} & = & \frac{p}{m}
1450: \end{eqnarray}
1451:
1452:
1453: This yields the solutions
1454:
1455: \begin{eqnarray}
1456: p_{t} & = & p_{0}e^{-\gamma t}+\xi _{p}\label{Wigpev} \\
1457: R_{t} & = & R_{0}+\frac{p_{0}}{\eta }(1-e^{-\gamma t})+\xi _{R}\, ,\label{WigRevol}
1458: \end{eqnarray}
1459:
1460:
1461: where \( \xi _{p} \) and \( \xi _{R} \) are given as integrals over
1462: the force \( F(\cdot ) \):
1463:
1464: \begin{eqnarray}
1465: \xi _{p} & = & \int _{0}^{t}ds\, e^{-\gamma (t-s)}F(s)\label{xip} \\
1466: \xi _{R} & = & \frac{1}{\eta }\int _{0}^{t}ds\, (1-e^{-\gamma (t-s)})F(s)\, .\label{xir}
1467: \end{eqnarray}
1468: Since \( F(\cdot ) \) is a Gaussian random process, \( \xi _{p} \)
1469: and \( \xi _{R} \) are Gaussian random variables as well. Therefore,
1470: the phase space density evolving out of \( \delta (R-R_{0})\delta (p-p_{0}) \)
1471: is a two-dimensional Gaussian distribution in phase space \( (R,p) \):
1472:
1473: \begin{equation}
1474: J^{W}(R_{t},p_{t}|R_{0},p_{0};t)=\left\langle \delta (R_{t}-\bar{R}_{t}-\xi _{R})\delta (p_{t}-\bar{p}_{t}-\xi _{p})\right\rangle \, .
1475: \end{equation}
1476:
1477:
1478: The average values \( \bar{p}_{t} \) and \( \bar{R}_{t} \) may be
1479: read off from eqs. (\ref{Wigpev}) and (\ref{WigRevol}). \( J^{W} \)
1480: has to be Fourier transformed with respect to \( p_{t} \) and \( p_{0} \)
1481: in order to arrive at the density matrix propagator \( J(R_{t},r_{t}|R_{0},r_{0};t) \)
1482: (see Eq. (\ref{propWigner})). In order to do this, we express \( J^{W} \)
1483: as a Gaussian density in terms of \( p_{t},\, p_{0} \), for fixed
1484: \( R_{t},R_{0} \):
1485:
1486: \begin{equation}
1487: J^{W}\propto \exp \left[ -\frac{1}{2}\delta P^{t}K^{-1}\delta P\right] \, ,
1488: \end{equation}
1489:
1490:
1491: with
1492:
1493: \begin{equation}
1494: \delta P=\left[ \begin{array}{c}
1495: p_{t}-\bar{p}_{t}\\
1496: p_{0}-\bar{p}_{0}
1497: \end{array}\right]
1498: \end{equation}
1499:
1500:
1501: and the covariance matrix
1502:
1503: \begin{equation}
1504: K=\left[ \begin{array}{cc}
1505: \left\langle \delta p_{t}^{2}\right\rangle & \left\langle \delta p_{t}\delta p_{0}\right\rangle \\
1506: \left\langle \delta p_{t}\delta p_{0}\right\rangle & \left\langle \delta p_{0}^{2}\right\rangle
1507: \end{array}\right] \, .
1508: \end{equation}
1509:
1510:
1511: Making use of (\ref{Wigpev}) and (\ref{WigRevol}), the average values
1512: and the deviations of \( p_{t} \) and \( p_{0} \) are found to be
1513: given by:
1514:
1515: \begin{eqnarray}
1516: \bar{p}_{t} & = & \lambda (R_{t}-R_{0})\\
1517: \bar{p}_{0} & = & e^{\gamma t}\lambda (R_{t}-R_{0})\\
1518: \delta p_{t} & = & \xi _{p}-\lambda \xi _{R}\label{dpt} \\
1519: \delta p_{0} & = & -e^{\gamma t}\lambda \xi _{R}\label{dp0} \\
1520: \lambda & \equiv & \frac{\eta }{e^{\gamma t}-1}
1521: \end{eqnarray}
1522:
1523:
1524: \( \bar{p}_{0} \) and \( \bar{p}_{t} \) are the momenta at time
1525: \( t \) and \( 0 \) which the particle must have if it is to go
1526: from \( R_{0} \) to \( R_{t} \) in time \( t \), provided no fluctuating
1527: force acts. \( \delta p_{t} \) and \( \delta p_{0} \) are the deviations
1528: from these values needed to compensate the effects of \( F(\cdot ) \).
1529:
1530: Quadratic completion in the exponent immediately yields the result
1531: of the Fourier integration over \( p_{t} \) and \( p_{0} \), which
1532: is the desired density matrix propagator \( J \):
1533:
1534: \begin{eqnarray}
1535: J(R_{t},r_{t}|R_{0},r_{0};t)\propto & & \nonumber \\
1536: \exp \left[ i(R_{t}-R_{0})\lambda (r_{t}-e^{\gamma t}r_{0})-\frac{1}{2}\left\langle \left( r_{t}\delta p_{t}-r_{0}\delta p_{0}\right) ^{2}\right\rangle \right] \, . & & \label{jprop}
1537: \end{eqnarray}
1538:
1539:
1540: This reproduces the result given in Ref.~\onlinecite{callegg} (or
1541: Ref.~\onlinecite{GZ_CL}, App. E). The prefactor can be determined
1542: from the normalization condition, Eq. (\ref{normal}), and only depends
1543: on the time \( t \). The term in angular brackets still has to be
1544: averaged over the force \( F(\cdot ) \). The resulting real part
1545: of the exponent equals \( -S_{I} \) evaluated along the semiclassical
1546: paths (compare the general structure given in Eq. (\ref{ims})). In
1547: terms of \( F(\cdot ) \), \( \delta p_{t} \) and \( \delta p_{0} \)
1548: read explicitly:
1549:
1550: \begin{eqnarray}
1551: \delta p_{t} & = & \int _{0}^{t}ds\, \frac{e^{\gamma s}-1}{e^{\gamma t}-1}F(s)\\
1552: \delta p_{0} & = & \int _{0}^{t}ds\, \frac{e^{\gamma s}-e^{\gamma t}}{e^{\gamma t}-1}F(s)\, .
1553: \end{eqnarray}
1554:
1555:
1556: Note that
1557:
1558: \begin{equation}
1559: \delta p_{t}-\delta p_{0}=\int _{0}^{t}ds\, F(s)\, .
1560: \end{equation}
1561:
1562:
1563: The averages \( \left\langle \delta p_{t}^{2}\right\rangle \), \( \left\langle \delta p_{0}^{2}\right\rangle \)
1564: and \( \left\langle \delta p_{t}\delta p_{0}\right\rangle \) to
1565: be evaluated in Eq. (\ref{jprop}) contain time-integrals over the
1566: force correlator \( \left\langle F(s_{1})F(s_{2})\right\rangle \).
1567: At zero temperature, these integrals lead to terms growing logarithmically
1568: in time, which are characteristic for the free particle coupled to
1569: an Ohmic bath (compare the discussion in the main text, sec. \ref{qubrownmotion}).
1570:
1571: \begin{thebibliography}{10}
1572: \bibitem{feynvern}R.~P. Feynman and F.~L. Vernon, Ann. Phys. (N.Y.) \textbf{24}, 118
1573: (1963).
1574: \bibitem{feynhibbs}R.~P. Feynman and A.~R. Hibbs, \emph{Quantum Mechanics and Path
1575: Integrals}, (McGraw-Hill, New York, 1965).
1576: \bibitem{AAK}B.~L. Altshuler, A.~G. Aronov, and D.~E. Khmelnitsky, J. Phys.
1577: C Solid State \textbf{15}, 7367 (1982).
1578: \bibitem{sai}A. Stern, Y. Aharonov, and Y. Imry, Phys. Rev. A \textbf{41}, 3436
1579: (1990).
1580: \bibitem{cohen}D. Cohen, Phys. Rev. E \textbf{55}, 1422 (1997); D. Cohen, J. Phys.
1581: A \textbf{31}, 8199 (1998).
1582: \bibitem{GZ}D.~S. Golubev and A. Zaikin, Phys. Rev. B \textbf{59}, 9195 (1999).
1583: \bibitem{GZ_CL}D.~S. Golubev and A. Zaikin, Phys. Rev. B \textbf{62}, 14061 (2000).
1584: \bibitem{GZ_PB}D.~S. Golubev and A. Zaikin, Physica B \textbf{280}, 453 (2000).
1585: \bibitem{chakravarty}S. Chakravarty and A. Schmid, Phys. Rep. \textbf{140}, 195 (1986).
1586: \bibitem{AAGcritique}I.~L. Aleiner, B.~L. Altshuler, and M.~E. Gershenson, Waves in
1587: Random Media \textbf{9}, 201 (1999).
1588: \bibitem{VavAmbeg}M. Vavilov and V. Ambegaokar, \eprint{cond-mat/9902127} (1999).
1589: \bibitem{cohenimry}D. Cohen and Y. Imry, Phys. Rev. B \textbf{59}, 11143 (1999).
1590: \bibitem{BelitzKirckpatrick}T.~R. Kirckpatrick and D. Belitz, Phys. Rev. B \textbf{65}, 195123
1591: (2002); Comment by D.~S. Golubev, A. Zaikin and G. Schön in \eprint{cond-mat/0111527},
1592: and reply in \eprint{cond-mat/0112063}.
1593: \bibitem{notesravg}Note that the impurity average of \( S_{R} \) (which corresponds
1594: to a properly weighted average over all possible pairs of trajectories)
1595: vanishes \emph{in general}, simply because \( S_{R}[q^{>},q^{<}]=-S_{R}[q^{<},q^{>}] \).
1596: \bibitem{clwavepackets}A.~O. Caldeira and A.~J. Leggett, Phys. Rev. A \textbf{31}, 1059
1597: (1985).
1598: \bibitem{lossmullen}D. Loss and K. Mullen, Phys. Rev. B \textbf{43}, 13252 (1991).
1599: \bibitem{buettHO}K.~E. Nagaev and M. Büttiker, Europhys. Lett. \textbf{58}, 475 (2002).
1600: \bibitem{callegg}A.~O. Caldeira and A.~J. Leggett, Physica \textbf{121}A, 587 (1983).
1601: \bibitem{hakimambeg}V. Hakim and V. Ambegaokar, Phys. Rev. A \textbf{32}, 423 (1985).
1602: \bibitem{5}Note that the damping constant \( \gamma \) used here corresponds
1603: to \( 2\gamma \) in Ref.~\onlinecite{callegg}, starting from their
1604: Eq.~(6.8).
1605: \bibitem{milonni}P.~W. Milonni, \emph{The Quantum Vacuum}, (Academic Press, San Diego,
1606: 1994).
1607: \bibitem{weiss}U.~Weiss, \emph{Quantum Dissipative Systems}, (World Scientific,
1608: Singapore, 2000).
1609: \bibitem{schramm}P.~Schramm and H.~Grabert, Journal of Stat. Phys. \textbf{49}, 767
1610: (1987).
1611: \bibitem{mohantywebb}P. Mohanty, E.~M.~Q. Jariwala, and R.~A. Webb, Phys. Rev. Lett.
1612: \textbf{78}, 3366 (1997).
1613: \bibitem{CedrBuett}P. Cedraschi, V.~V. Ponomarenko, and M. Büttiker, Phys. Rev. Lett.
1614: \textbf{84}, 346 (2000).
1615: \bibitem{21}F. Marquardt and C. Bruder, Phys. Rev. B \textbf{65}, 125315 (2002).
1616: \bibitem{seelig}G. Seelig and M. Büttiker, Phys. Rev. B \textbf{64}, 245313 (2001).
1617: \bibitem{Guinea}F. Guinea, Phys. Rev. B \textbf{65}, 205317 (2002).
1618: \bibitem{ghz}D.~S.~Golubev, C.~P.~Herrero, and A.~D.~Zaikin, \eprint{cond-mat/0205549}.\end{thebibliography}
1619:
1620: \end{document}
1621: