Skip to main content

Full text of "Chaos, Dissipation and Quantal Brownian Motion"

See other formats


Chaos, Dissipation and Quantal Brownian Motion 

(Lecture notes of course in the international school of physics Enrico Fermi, 
Session CXLIII New Directions in Quantum Chaos, Varenna Italy, July 1999) 



DoRON Cohen 

Department of Physics, Harvard University 



Summary. — Quantum dissipation, the theory of energy spreading and quantal 
Brownian motion are considered. In the first part of these lecture-notes we discuss 
the classical theory of a particle that interacts with chaotic degrees of freedom: 

• The Sudden and the Adiabatic approximations; 

• The route to stochastic behavior; 

• The fluctuation-dissipation relation; 

• And the application to the 'piston' example. 

In the second part of these lecture-notes we discuss the restricted problem of classical 
particle that interacts with quantal (chaotic) degrees of freedom: 

• Limitations on quantal-classical correspondence; 

• The perturbative core-tail spreading profile; 

• Linear response theory versus Fermi-golden-rule picture; 

• Random-matrix-theory considerations; 

• And the quantal Sudden and Adiabatic approximations. 

In the third part of these lecture-notes we discuss the problem of quantal particle 
that interacts with an effective harmonic bath: 

• Classical Brownian motion; 

• The ZCL model and the DLD model; 

• The white noise approximation; 

• The reduced propagator and master-equation formulation; 

• And the two mechanisms for dephasing. 

We conclude with explaining the main open question in the theory of quantum 
dissipation and quantal Brownian motion. That question concerns the problem of 
quantal particle that interacts with quantal (chaotic) degrees of freedom. 



© Societa Italiana di Fisica 



1 



2 



DoRON Cohen 



1. — Definition of the problem 

We are interested in the redueed dynamics of a slow degree of freedom (x, p) that 
interacts with a 'bath'. The Hamiltonian is 

(1) n = Wo(x,p)+Wenv(x,Qa,Pa) 

In Fig.l we display a list of our main assumptions and an illustration of our leading 
example. In this example the slow degree of freedom is the 'piston', and the 'bath' 
consist of one gas particle. 

Assumptions: 

• He„v(Q, -P; x) with x=const generates classically chaotic dynamics. 

• Wo(x, p) = Tp^ /{2m) describes a free particle, (optional is linear driving). 

• Initially the bath is characterized either by an energy or by a temperature T. 

• In a classical sense x(f) is a slow degree of freedom. 




xit) 



Fig. 1. - List of main assumptions and an illustration of the 'piston' example. 



Classically the reduced dynamics of the slow degree of freedom is described by the 
Langevin equation 

(2) mx = T{t) 

where T{t) is a stochastic force. The average force on the particle is 

(3) {Ht)) = P(x) - Mi 

Without loss of generality, just for the sake of simplicity, we assume from now on that the 
velocity-independent force F(x) is equal to zero (see Fig.l). Thus there is no reversible 



Chaos, Dissipation and Quantal Brownian Motion 



3 



energy change due to conservative work. However, there is still systematic irreversible 
change of energy due to the friction force. The energy dissipation rate is: 

(4) ^ ^ ^mx^^ Ri — /ix^ for non- thermal motion. 

The fluctuating component of the stochastic force is like white noise, and it is character- 
ized by its intensity v. The dissipation constant ^ is related to the noise intensity f via 
a universal Fluctuation-Dissipation (FD) relation 

(5) /X 



2k^T 

We are interested in developing a corresponding quantum mechanical theory of dis- 
sipation and quantal Brownian motion. 'Dissipation' means from now on that energy is 
absorbed by the bath degrees of freedom due to the time dependence of x. The dissi- 
pation coefficient is ^. 'Brownian motion' refers from now on to the reduced dynamics 
of the (x, p) degree of freedom. We shall use the term 'piston' rather than 'Brownian 
particle' if the motion is constrained to one dimension. Note that the term 'piston' is 
not used in its literal sense. 

2. — Restricted versions of the problem 

The quantum mechanical treatment of the general problem is extremely complicated. 
Therefore it is a good idea to analyze restricted versions of the general problem. Treating 
x{t) as a classical degree of freedom, we can consider the time dependent Hamiltonian 

(6) H = Knv(<3,P;x(t)) 

For simplicity we can further assume that x{t) describes a motion with a constant velocity 
X = V. Consequently we can treat x in (6) as a one-degree-of-freedom variable. This 
restricted problem is the main issue of the following lectures (parts 1 and 2). 

In consistency with the terminology that has been introduced at the end of the pre- 
vious section, we shall refer to the restricted problem defined above as 'the problem of 
quantum dissipation'. Obviously, in other physical examples x does not have to be the 
position of a particle. It can be any controlling parameter that appears in the Hamilto- 
nian, e.g. electric field. Of particular interest is the case where x is the magnetic flux 
via a ring. The velocity V = x has then the meaning of electro-motive-force. Let us 
assume that the ring contains one charged particle (Q, P) that performs diffusive mo- 
tion. The charged particle gains kinetic energy, and the dissipation coefficient /i is just 
the conductivity of the ring. Note that in actual circumstances the charged-particle is 
an electron, and its (increasing) kinetic energy is eventually transfered to the vibrational 
modes (phonons) of the ring, leading to Joule heating. The latter process is 'on top' of 
the generic dissipation problem that we are going to analyze. 



4 



DoRON Cohen 



We come back to the Brownian particle / 'piston' example, and shift the focus from 
the 'bath' to the reduced- dynamics of the (x, p) degree-of-freedom. We can study the 
effect of the fluctuating force by considering an Hamiltonian of the type 



where t) is an effective stochastic potential the mimics the noisy character of the 
environment. Obviously, with such Hamiltonian wc cannot mimic the effect of dissipation. 
In order to have dissipation the interaction should be with dynamical degrees of freedom. 
We can introduce an effective harmonic-bath as follows: 



The effective-noise model (7) as well as the effective harmonic-bath model (8) can be 
treated analytically using the Feynman- Vernon (FV) formalism. The reduced dynamics 
of the particle is obtained after averaging over realizations of the stochastic potential 
(in case of (7)), or by elimination of the environmental degrees of freedom (in case of 
(8)). It leads to a unified description of diffusion localization and dissipation (DLD). It is 
found indeed that the formal solution of (8) reduces to that of (7) if dissipation effect is 
neglected. We shall refer to (8), and more generally to its formal solution, as the 'DLD 
model'. It is possible to introduce an effective RMT-bath instead of an effective harmonic 
bath, but then there is no general analytical solution. It has been demonstrated however 
that at high temperatures an approximate treatment of the effective RMT-bath coincides 
with the exact (high temperature) solution of the DLD model. 

3. — 'History' of the problem (possibly biased point of view [1]) 

There is a well established classical theory for dissipation and Brownian motion. In 
the pre-chaos literature (~1980) the dissipation constant for a moving 'piston' is given 
by the 'wall formula'. See [4, 5] and followers. A more general point of view, which 
is discussed in the first part of these lectures (Sections 4-11), has been adopted in the 
post-chaos literature (~1990). There, the emphasis is on relating the dissipation constant 
to the intensity of fluctuations [14, 15, 17, 18]. 

Various methods such as 'linear response', 'Kubo-Grcenwood formalism', and 'mul- 
tiple scale analysis' have been used [5, 15, 18] in order to make a quantum mechanical 
derivation of the FD relation. All these methods are essentially equivalent to a naive 
application of the Fermi- Golden-Rule (FOR) picture. The validity of the naive FOR 
result has been challenged in by Wilkinson and Austin [16]. They came up with a sur- 
prising conclusion that we would like to paraphrase as follows: A proper FGR picture, 
supplemented by an innocent-looking RMT assumption, leads to a modified FGR result; 
In the classical limit the modified FGR result disagrees with the FD relation and leads 
to violation of the quantal-classical correspondence principle. Obviously this conclusion 



(7) 





Chaos, Dissipation and Quantal Brownian Motion 



5 



should be regarded as a provocation for constructing a proper theory for Quantum Dis- 
sipation [19, 20]. This is the issue of the second part of these lectures (Sections 12-24). 
The theory is strongly related to some studies of parametric dynamics [25, 26, 27, 28] 
and wavepackct dynamics [29, 30, 31]. 

Most of the literature that comes under the heading 'Quantum Dissipation' is con- 
cerned with the more general problem where a; is a dynamical variable [12]. A direct 
handling of the 'bath' degrees of freedom is usually avoided. As an exception see for ex- 
ample [13]. It is common to adopt an 'effective-bath' approach. See [2, 3, 6] and followers. 
Specific discussion of quantal Brownian motion has been introduced in [7] and later in 
[8, 9]. It leads to the DLD model, that unifies the treatment of diffusion localization and 
dissipation. This is the issue of the third part of these lectures (Sections 25-31). The 
high temperature version of the DLD model is obtained also by considering coupling to 
an effective RMT-Bath [10]. 



4. — Fluctuations: intensity and correlation time 

We consider the Hamiltonian H{Q,P;x) with a;=const. The phase space volume 
which is enclosed by the energy surface H{Q,P;x) = E will be denoted by Cl{E; x). For 
the classical density of states we shall use the notation g{E) = 9eO(£J). We define 

(9) m = -^{Q{t),P{t)-x) 

The conservative force F{x) = {J^{t))^ is obtained by performing microcanonical aver- 
aging over {Q,P). It is equal to zero if and only if fl{E;x) is independent of x. For 
simplicity we assume from now on that this is indeed the case. Now we define a correla- 
tion function 



(10) Ce(t) = {J^{t)J^{t+T))^ 

The fluctuating force is characterized by an intensity 

/oo 
-oo 

and by a correlation time t^,. The power spectrum of the fluctuations Ce{i-o) is the Fourier 
transform of Ce(t). For chaotic bath the stochastic force is like white noise. We always 
have Tel < ierg, where t„g is the ergodic time. In the general discussion we shall not 
distinguish between the two time scales. However, in specific examples the distinction is 
meaningful. For example, in case of the 'piston' example t^, is the collision time with the 
walls of the piston, while <erg is of the order of the ballistic time Tbi. 



6 



DoRON Cohen 



5. — Fluctuations: time dependent Hamiltonian 

We consider from now on the time-dependent Hamiltonian 'H{Q,P;x{t)) with con- 
stant non-zero velocity x = V. We define 



The statistical properties of the fluctuating force J^{t) are expected to be slightly differ- 
ent from the V=Q case. The average {T{t)) is no longer expected to be zero. Rather, we 
expect to have {T{t)) = —^V. This implies that the correlator {T{ti)!F{t2)) acquires an 
offset {iJiVY. We shall argue that if the velocity V is small enough then the 'offset cor- 
rection' can be ignored for a relatively long time which will be denoted by tf^^. Obviously 
it is essential to have 

(13) Tci <^ tfrc [non-trivial slowness condition]. 

There is another possible reason for the correlator {T{t\)T{t2)) to be different from 
CE{t2—ti)- Loss of correlation may be either due to the dynamics of {Q{t), P{t)) or 
else due to the parametric change of x{t). Wo can define a parametric correlation scale 
(5x°'. For the piston example it is just the penetration distance into the piston upon 
collision (the effective 'thickness' of the wall). The associated parametric correlation 
time is t°' = 6x'^/V. We assume that 

(14) ^ T°' [trivial slowness condition]. 

meaning that loss of correlations is predominantly determined by the chaotic nature 
of the dynamics rather than by the (slow) parametric change of the Hamiltonian. For 
the piston example application of the above requirements leads to the obvious condition 
V <^Vb, where Ve is the velocity of the gas particle. 

6. — Actual, Parametric and Reduced energy changes 

For the time dependent Hamiltonian H{Q,P;x{t)) energy is not a constant of the 
motion. Changes in the actual energy £{t) reflect 'real' dynamical changes as well as 
parametric changes. Therefore it is useful to introduce the following deflnitions: 



(12) 



•^(*) = -^{Q{t),P{t);x{t)) 



(15) 
(16) 



£{t) = n{Q{t),P{ty,x{t)) 



The actual energy change can be calculated as follows: 



(17) 



(18) 




Chaos, Dissipation and Quantal Brownian Motion 



7 




Fig. 2. - Schematic illustration of the dynamics. An initially localized distribution is launched 

in phase space. For a limited time (loft plot) it travels upon the initial energy surface. But then 
it departs from it. After much longer times (right plot) the evolving distribution is concentrated 
across an instantaneous energy surface. 



The actual energy change 5£ can be viewed as a sum of parametric-energy-change 5£o, 
and reduced-energy-change 5£' . 

(19) 5£ = n{Q{i),P{iy,x{i))-n{Q{Q),P{Q);xm 

(20) 5£o = niQ{t),P{t);x{t))-n{Q{t),P{t);x{0)) 

(21) 6£' = n{Q{t),P{ty,x{0))-n{QiO),P{0);x{0)) 

The reduced energy change d£' reflects the deviation of {Q, P) from the original energy 
surface. It can be calculated as follows: 



(22) 5£' = V- 



T{t) X t 



Jo 



On the other hand, the actual energy change S£ reflects the deviation of {Q, P) from the 
instantaneous energy surface. 

7. — The Sudden and the Adiabatic approximations 

By inspection of the expressions for the reduced energy change we arrive at the con- 
clusion that for short times we have the so called 'sudden approximation': 

(23) 5£' w fort«:Td 

By inspection of the expression for the actual energy change we arrive at the conclusion 
that for longer times we have the so called 'adiabatic approximation': 

(24) ~ fori<t,,. 



8 



DoRON Cohen 



The time evolution of an initially localized phase-space distribution po{Q, P) is illustrated 
in Fig. 2. For short times we have in general non-stationary time evolution: 

(25) U{t) po{Q,P) po{Q,P) fort<T,i 

Here U{t) is the classical propagator of phase space points. However, if we operate with 
the same U{t) on a microcanonical distribution, then 

(26) U{t) Pe,x(o) (Q, P) ~ /Oe,x(o) (Q, P) for t < t,. 

It is as if the microcanonical state does not have the time to adjust itself to the changing 
Hamiltonian. The sudden approximation implies that for short times U{t) can be replaced 
by unity if it operates on an initial microcanonical state. For longer times we have 

(27) U{t) po{Q, P) ~ PE,x(t)(Q, P) for < t < 

Here there is enough time for the evolving distribution to adjust itself to the changing 
Hamiltonian. If we start with a microcanonical state, then the above similarity will hold 
for any t <^ ff„. The adiabatic approximation becomes worse and worse as time elapses 
due to the transverse spreading across the energy surface. We shall see that is the 
breaktime for the adiabatic approximation. 

8. — Ballistic and Diffusive energy spreading 

We recall the formula for the actual energy change 

(28) £{t) - £{0) = -V [ J^{t')dt' 

Jo 

We Assume an initial microcanonical preparation which is characterized by an energy E. 
The energy spreading after time t is: 

(29) mt) - S{0)f) = f f{T{h)T{t2)) dtidt2 

Jo Jo 

Now we make the following approximation 

(30) {T{ti)T{t2)) « C^{t) applicable if t<^U,, 

The validity of this approximation is restricted by the condition t <C ttrc which will be 
discussed later. The energy spread is 

(31) {{£{t) - £{Q)f) « C^{0)-{Vtf fori«T., 

(32) {{£{t) - E{Q)f) « 2D^t for Tel < f < 



Chaos, Dissipation and Quantal Brownian Motion 



9 



The ballistic spreading on short time scales just reflects the parametric change of the 
energy surfaces. The diffusive spreading on longer times reflects the deviation from the 
adiabatic approximation. The diffusion coefficient is 



(33) = iy^y c4T)dT ^ I'^^v^ 



9. — Energy spreading and dissipation 

It is possible to argue that the energy distribution p{E) obeys the following diffusion 
equation: 



(341 ^ - A 



for t > L, 



If we transform to the proper phase space variable n = i^{E), this diffusion equation gets 
the standard form. The average energy is calculated via 

(35) {£{t)) = E p{E)dE 

Jo 

and consequently 

Using the expression for We obtain the following FD relation 
(37) = i^^feW-.) 

The actual value of p is obtained by averaging p^ according to the distribution p{E). The 
standard version of the FD relation fj, = y/{2k^T) is obtained if p{E) is of the canonical 
type. 



10. — Application to the 'piston' example 

It is instructive to apply the FD relation for the 'piston' example. The bath degrees 
of freedom is a gas particle whose mass is m. The faces of the piston are characterized by 
their total Area. The piston is moving with a constant velocity V inside a d-dimensional 
cavity. For simplicity wc treat the collisions as 'one-dimensional' and omit d, dependent 
pre-factors [20]. An illustration of one collision with the piston is presented in Fig. 3. 
Each collision can be either from the left or from the right side. The resultant stochastic 
force which is experienced by the piston is a sum over short impulses: 



(38) 



= ^[2m(^;,„, - y)] ^(t-teoi) 

col 



10 



DoRON Cohen 



The duration of each impulse is equal to the penetration-time upon a collision with the 
(soft) faces of the piston. If successive collisions with the piston are uncor related, then 
the correlation time is equal to the average duration of an impulse. The main steps 
in the analysis of the multi-collision process are summarized in Fig. 3 as well. It is easily 
verified that the FD relation between and i^e is satisfied. Note however that a proper 
application of kinetic theory is essential in order to obtain the correct geometrical factors 
involved [20]. 

The FD relation is a very powerful tool . This becomes most evident once we consider 
a variation of the above example: If successive collisions with the piston are correlated, 
for example due to bouncing behavior, then it is still a relatively easy task to estimate 
Ve for the V = case, and then to obtain /Ie via the FD relation. On the other hand, 
a direct evaluation of /Xe using kinetic considerations is extremely difficult, because in 
calculating {J^{t)) it is essential to take into account the correlations between successive 
collisions. 

11. The route to stochastic behavior 

The derivation of the FD relation consists of two steps: The first step establishes 
the local diffusive behavior for short {t <C ftrc) time scales, and De is determined; The 
second step establishes the global stochastic behavior on large {t ;:g> t^rg) time scales. The 
various time scales involved are illustrated in Fig. 4. The classical breaktime t^^ is defined 
as follows: 

(39) tf., = u/{/iVf 

For t > if„ the systematic energy change {6£) = jiV'^t becomes larger than the energy 
spreading {{5£'^)y'^ = {vV'^ty^^, and the local analysis becomes meaningless. Therefore 
ffrc can be regarded as the breakdown of the adiabatic approximation. 

From the discussion above it follows that the validity of the classical derivation is 
restricted by the non- trivial slowness condition <^ ttrc- Alternatively, it is more useful 
to regard the non-trivial slowness condition using the point-of-view of Section 5. This 
point-of-view is further explained below, and can be easily generalized to the quantum- 
mechanical case. For V ^ the average of !F{t) does not vanish, and therefore its 
correlator should include an 'offset' term. Namely, {J^{ti)J^{t2)) ~ CE{t2—ti) + {fJ-Vy. 
The offset term can be neglected for a limited time t <^ t^,^ provided [i^VY C'e(O). 
The latter condition is equivalent to the non-trivial slowness condition Td <C tt^c- 

In the quantum-mechanical analysis we can use the same two-steps strategy in order 
to derive a corresponding FD relation. We are going to concentrate on the first step. 
It means that our objective is to establish a crossover from ballistic to diffusive behav- 
ior at the time t ^ t^. The classical analysis in Section 8 is essentially 'perturbation 
theory'. We can follow formally the same steps in the quantum- mechanical derivation, 
using Heisenberg picture. However, quantum-mechanical perturbation-theory is much 
more fragile than the corresponding classical theory, and we have typically t-p^t ttrc- 



Chaos, Dissipation and Quantal Brownian Motion 



11 





penetration 
distance 


/ 


'piston' ) j 

displaced 
piston 








\P . 




n 

turning 
point 








Q 





F(t) 














tcl 




'^col 




t 



• Assume that the velocities |?;coi| ~ Ve are uncorrelated. 

• The average time between conisions is t^„i ~ (Volume/ Area) / v^- 

• For y = we have {T{t)) = and u^, = (2mvE)^ • (l/r,,,,)- 

• For F ^ we have {^{t)) = —n^V where /Ue = 2m • 

• The phase space volume is Q.[E) = Volume • (mvE)'^. 

Fig. 3. - Illustration of one collision with the piston, and a list of steps in the analysis of a series 
of such collisions. The Area of the moving faces, the Volume of the cavity, the mass m of the 
gas-particle, and its kinetic energy E should be specified. Geometric d-dependent prefactors 
have been dropped. See [20] for exact calculations. 



12 DoRON Cohen 



t 


t 


t 






^scl ^ 




Stochastic 

diffusive 








ballistic 
t=0 



perturbation semiclassical classical 
theory theory theory 



ifrc = breakdown of classical perturbation theory 


iprt = breakdown of quantal perturbation theory 


taci = breakdown of semiclassical approximation 


Tci < tf„ 




classical definition of slowness 


Tci < iprt 




quantal definition of slowness 


Tel <C tscl 




not restrictive condition 


Tsc < Td 




quantal definition of fastness 



Fig. 4. - Illustration of the various time scales involved in constructing cither classical, semi- 
classical or perturbation theory of dissipation. We use the notation tsc = Sxsc/V. The ac- 
companying table summarizes the associated requirements for the applicability of each of those 
theories. Note that the quantum mechanical definitions of slowness and of fastness are not 
complementary. In any case, slowness in the classical sense is always assumed. 



Therefore, the quantum mechanical definition of slowness (Tci <^ t^rt) is much more re- 
strictive than the classical requirement (r^j <C tfrc)- In the regime where perturbation 
theory fails we have to use a non-perturbative theory. In particular we are going to 
find the sufficient conditions for having detailed quantal-classical correspondence (QCC) 
using semiclassical considerations. 

12. — The transition probability kernel 

In order to go smoothly from the classical theory to the quantum mechanical theory 
it is essential to use proper notations. From now on we shall use the variable n = Cl{E) 
instead of E. Given x, the energy surface that corresponds to a phase-space volume n 



Chaos, Dissipation and Quantal Brownian Motion 



13 



will be denoted by: 

(40) \n{x)) = {{Q,P)\ n{Q,P;x) = En} 

where £"„ is the corresponding energy. The microcanonical distribution which is sup- 
ported by \n{x)) will be denoted by Pn,x{Q>P)- Upon quantization the variable n be- 
comes a level-index, and Pn,x{Q,P) should be interpreted as the Wigner function that 
corresponds to the eigenstate \n{x)). With these definitions we can address the quantum 
mechanical theory and the classical theory simultaneously. We shall use from now on 
an admixture of classical and quantum-mechanical jargon. This should not cause any 
confusion. 

The transition probability kernel Pt{n\m) is the projection of the evolving state on 
the instantaneous set of energy surfaces. It is also possible to define a parametric kernel 
P{n\m). See illustrations in Fig. 5. The definitions are: 

(41) Pt{n\m) = trace( 

Pn,x{t) Pm,x{0) ) 

(42) P{n\m) = trace( Pn,a:{t) Pm,x(o) ) 

The phase-space propagator is denoted by U{t). The parametric kernel P{n\m,) depends 
on the displacement 5x but not on the actual time that it takes to realize this displace- 
ment. The trace operation is just a dQdP integral over phase-space. 

Before we go to the quantum mechanical analysis, let us summarize the classical 
scenario. The classical sudden approximation is 

(43) Pt{n\m) w P{n\m) for t «: t,, 

For longer times we have the classical adiabatic approximation, or more precisely we 
have diffusive spreading: 

(44) Pt{n\m) w Gaussian(n— m) for Td <C f <IC tfrc 

For t ^ tf,.(. the kernel Pt{n\m) is no-longer a narrow Gaussian that is centered around 
n = m. However, we can argue [17, 20] that its profile can be obtained as the solution 
of a stochastic diffusion equation (see Sec. 9). 

The kernel P{n\m) reflects the parametric correlations between two sets of energy 
surfaces. Consequently Non-Gaussian features may manifest themselves. An important 
special non-Gaussian feature is encountered in many specific examples where x affects 
only a tiny portion of the energy surface (Fig.5). In the 'piston' example this is the case 
because (dH/dx) = unless Q is near the face of the piston. Consequently P{n\m) has 
a (5-singularity for n = m. 




Fig. 5. - Upper Left: Phase space illustration of the initial and of the instantaneous set of 
parametric energy surfaces; Plot of the associated P(n|m), where the classical behavior is indi- 
cate by the black lines, and the quantum-mechanical behavior is represented by the grey filling. 
Detailed QCC is assumed. In the quantum-mechanical case classical sharp-cutoffs are being 
smeared. Upper Right; Illustration of a typical non-generic feature. In the quantum-mechanical 
case the classical delta-singularity is being smeared. Lower Right; The same non-generic feature 
manifests itself in the 'piston' example. Lower Left; In the perturbative case there is no detailed 
QCC. The kernel is characterized by a core-tail structure. The tail is limited by the bandwidth 
of the coupling matrix-elements. If 5x is sufficiently small the core is just a Kronecker's delta. 



Chaos, Dissipation and Quantal Brownian Motion 



15 



13. — Limitations on quantal-classical correspondence (QCC) 

The main objects of our discussion are the transition probability kernel Pt{n\m) and 
the parametric kernel P{n\m) which have been introduced in the previous section. We 
refer now to Equations (41) and (42). In the classical context pn,x{Q, P) is defined as the 
microcanonical distribution that is supported by the energy-surface H{Q, P: :f:{t)) = En- 
The energy En corresponds to the phase-space volume n=fl(E). In the QM context 
Pn,x{Q, P) is defined as the Wigner-function that represents the energy-eigenstate \n{x)). 
The phase-space propagator is denoted symbolically by U{t). In the classical case it 
simply rc-positions points in phase-space. In the QM case it has a more complicated 
structure. It is convenient to measure phase-space volume {n=U,{E)) in units of {2iThY 
where d is the number of degrees of freedom. This way we can obtain a 'classical ap- 
proximation' for the QM kernel, simply by making n and m integer variables. If the 
'classical approximation' is similar to the QM kernel, then we say that there is detailed 
QCC. If only the second-moment is similar, then we say that there is restricted QCC. 
In the present section we are going to discuss the conditions for having detailed QCC, 
using simple semiclassical considerations. 

Wi gner function pn.x 

{Q,P), unlike its classical microcanonical analog, has a non- 
trivial transverse structure. For a curved energy surface the transverse profile looks like 
Airy function and it is characterized by a width [22] 



(45) 



where e^i is a classical energy scale. For the 'piston' example e^i = E is the kinetic energy 
of the gas particle. In the next paragraph we discuss the conditions for having detailed 
QCC in the computation of the parametric kernel P{n\m). Then we discuss the further 
restrictions on QCC, that are associated with the actual kernel Pt{n\m). 

Given a parametric change 5x we can define a classical energy scale SE^i cx 5x via 
(31). This parametric energy scale characterizes the transverse distance between the 
intersecting energy-surfaces \m{x)) and \n{x+Sx)). In the generic case, it should be 
legitimate to neglect the transverse profile of Wigner function provided SE^ ^ ^sc- 
This condition can be cast into the form 6x Sxsc where 

(46) 5xsc = oc h^^^ 

V^e/TcI 

Another important parametric scale is defined in a similar fashion: We shall see that it 

is not legitimate to ignore the transverse profile of Wigner function if 6E^t < A;,. This 
latter condition can be cast into the form 6x <C Sx-p^ where 




(47) Sx, 



prt 



16 



DoRON Cohen 



Typically the two parametric scales are well separated {5xp^t ^ Sxsc)- If we have 5x <C 
Sxprt then the parametric kernel P{n\m) is characterized by a perturbative core-tail 
structure which is illustrated in Fig. 5 and further discussed in the next sections. We do 
not have a theory for the intermediate parametric regime Sxp^^ Sx ^ Sx^^^. But for 
6x ^ 6xsc we can argue that there is a detailed QCC between the quantal kernel and 
the classical kernel. Obviously, 'detailed QCC does not mean complete similarity. The 
classical kernel is typically characterized by various non-Gaussian features, such as sharp 
cutoffs, delta-singularities and cusps. These features are expected to be smeared in the 
quantum-mechanical case. 

We turn now to discuss the actual transition probability kernel Pt{n\m). Here we 
encounter a new restriction on detailed QCC: The evolving surface Z^(t)|m) becomes more 
and more convoluted as a function of time. This is because of the mixing behavior that 
characterizes chaotic dynamics. For t > t^ci the intersections with a given instantaneous 
energy surface |n) become very dense, and associated quantum-mechanical features can 
no longer be ignored. The time scale t^^i can be related to the failure of the stationary 
phase approximation [23]. 

The breaktime scale tsd of the semiclassical theory is analogous to the breaktime scale 
iprt of perturbation theory, as well as to the breaktime scale ttrc of the classical theory. 
In order to establish the crossover from ballistic to diffusive energy spreading using 
QCC considerations we should satisfy the condition <IC t^d- This velocity-independent 
condition is not very restrictive. On the other hand we should also satisfy the condition 
Sx ^ Sxsc, with Sx = Vt^. The latter condition implies that the applicability of the 
QCC considerations is restricted to relatively fast velocities. We can define: 

(48) t^sc = scaled velocity = ^/ / Age 

If Vsc ^ 1 then the classical approximation is applicable in order to analyze the crossover 
from ballistic to diffusive energy spreading. 

14. The parametric evolution of P{n\m) 

Detailed QCC between the quantal P{n\m) and the classical P{n\m) is not guaranteed 
if 5x < Sxsc- For sufficiently small parametric change 6x, perturbation theory becomes a 
useful tool for the analysis of this kernel. A detailed formulation of perturbation theory 
is postponed to later sections. Here we are going to sketch the main observations. We 
are going to argue that for small 6x there is no detailed QCC between the quantal and 
the classical kernels, but there is still restricted QCC that pertains to the second moment 
of the distribution. Only for large enough Sx we get detailed QCC. These observations 
are easily extended to the case of Pt{n\m) in the next section. 

For extremely small Sx the parametric kernel P{n\m) has a standard "first-order" 
perturbative structure, namely: 



(49) 



P(n\m) Snm + Tail(n— m) for Sx <C Sx'^' 



Chaos, Dissipation and Quantal Brownian Motion 



17 



where Ja;^" is defined as parametric change that is needed in order to mix neighboring 
levels. For larger values of Sx neighbor levels are mixed non-perturbatively and conse- 
quently we have a more complicated spreading profile: 

(50) P{n\m) ~ Core(n— m) + Tail(n— to) for 6x <C Sx^^t 

In the perturbativc regime (Sx ^ (^ajprt) the second moment of P{n\m) is generically 
dominated by the 'tail'. It turns out that the quantum- mechanical expression for the 
second-moment is classical look-alike, and consequently restricted QCC is satisfied. The 
core of the quantal P{n\m) is of non-perturbative nature. The core is the component that 
is expected to become similar (eventually) to the classical P{n\m). A largo perturbation 
Sx ^ Sx-prt makes the core spill over the perturbative tail. If we have also Sx ^ Sxsc^ 
then we can rely on detailed QCC in order to estimate P{n\m). 

For the piston example wc can easily get estimates for the various scales involved. 
These are expressible in terms of De-Broglie wavelength Ae = 2TTh/{mvE) where m and 
Ve are defined as in Section 10. The displacement which is needed in order to mix 
neighboring levels, and the displacement which is needed in order to mix core and tail, 
are respectively: 

(51) SxT « (A^+VArea)^/^ « A^ 

(52) SX^,, « (Teol/Te,)''" Ae » Ae 

We have, in the generic case as well as in the case of the 'piston', the hierarchy Sx'^'^ <C 
Sxpri <^ 6xsc- Thus there is a 'gap' between the perturbative regime {6x <C Sx^rt) and 
the semiclassical regime {5x 6xsc)- 

15. — The time evolution of Pt{n\m) 

The dynamical evolution of Pt{n\m) is related to the associated parametric evolution 
of P(n I to). We can define a perturbative time scale tp^t which is analogous to Sxp^t- For 
t <C tprt the kernel Pt{n\m) is characterized by a core-tail structure that can be analyzed 
using perturbation theory. In particular we can determine the second moment of the 
energy distribution, and we can establish restricted QCC. If the second moment for the 
core-tail structure is proportional to t^, we shall say that there is a ballistic-like behavior. 
If it is proportional to t, we shall say that there is a diffusive-like behavior. In both cases 
the actual energy distribution is not classical-like, and therefore the term 'ballistic' and 
'diffusive' should be used with care. We are going now to give a brief overview of the 
various scenarios in the time evolution of Pt{n\m). These are illustrated in Fig. 6. In 
later sections we give a detailed account of the theory. 

For slow velocities such that Td ^ tprt ^ in, there is a crossover from ballistic- like 
spreading to diffusive-like spreading a,t t ^ t^. In spite of the lack of detailed QCC 
there is still restricted QCC as far as this ballistic-diffusive crossover is concerned. The 
breakdown of perturbation theory before the Heisenberg time {tp^^ <C tn) implies that 



18 



DoRON Cohen 



i 




x(t) 



5jCprt-- 



\^,R=1 Vsc=l 



Fig. 6. - The various crossovers in the time evolution of Pt{n\m). The vertical axis is x{t) = Vt. 
The parametric scales Sx'^"' and Sxp^t are indicted by horizontal lines. The horizontal axis is the 
velocity V. It is divided by vertical dashed lines to various velocity regimes. In each velocity 
regime there is a different dynamical route. The various crossovers are explained in the text and 
the various symbols are easily associated with having either Gaussian or some non-Gaussian 
spreading profile. In particular the perturbative spreading profile is either with or without 
non-trivial core, and its tail is either band-limited or resonance-limited. 



there is a second crossover at t ~ t^^t from a diffusive-like spreading to a genuine diffusive 
behavior. 

extrem,ely slow velocities arc defined by the the inequality ^ tp,t. This inequality 
implies that there are quantum-mechanical recurrences before the expected crossover from 
diffusive-like spreading to genuine-diffusion. This is the quantum-mechanical adiabatic 
regime. In the t ^ oo limit Landau-Zener transitions dominate the energy spreading, 
and consequently neither detailed nor restricted QCC is a-priori expected [15]. 

For fast velocities we have t^i- There is a crossover at t ~ tp^t from ballistic- 

like spreading to a genuine ballistic behavior, and at t ~ there is a second crossover 
from genuine-ballistic to genuine-diffusive spreading. The description of this classical- 
type crossover is out-of-reach for perturbation theory, but we can use the semiclassical 
picture instead. Note that the semiclassical definition of 'fastness' and the perturbative 
definition of 'slowness' imply that there is a 'gap' between the corresponding regimes. 
However, the interpolation is smooth, and therefore for simple systems surprises are not 
expected. 



Chaos, Dissipation and Quantal Brownian Motion 



19 



16. — Linecir response theory 

The classical derivation in Section 8 applies also in the quantum-mechanical case 
provided J^{t) is treated as an operator. This is known as 'linear response theory', 
and the expression for is essentially the 'Kubo-Greenwood' formula. Obviously the 
restriction t <C tf^^ is replaced by the more restrictive condition t ^ tp^^, which will 
be discussed later. For the purpose of concise presentation, the formula for the energy 
spreading can be written as follows: 

(53) 6E^ = y2 / / c^(t2-ti)dhdt2 = VH / — Ce(w) Ftico) 

Jo Jo J-oo ^''^ 



(54) 



where 



Ft{oj) = tismc{ojt/2)f 



The power spectrum of the classical fluctuations looks like white noise. It satisfies 
C-e{uj) « i'e for < 1/tji and decays to zero outside of this regime. Thus, still con- 
sidering the classical case, for t <C r^i we can make the replacement Ft{uj) ^ t, and we 
obtain the ballistic result SE"^ = CE{0)-{Vt)'^, while for t » Td we can make the replace- 
ment Ft{uj) 2tt5{uj), and we get then the diffusive behavior SE"^ = lysV^t. Now we 
turn to the quantum-mechanical case. The power-spectrum of the quantum-mechanical 
fluctuations is given by the formula 



(55) 



En — En 

h 



Semiclassical reasoning [24] applied to (55) leads to the immediate conclusion that energy 
levels are coupled by matrix elements provided |S„ — Em\ < where 



(56) 



band width 



The discrete nature of the power spectrum is of no significance as long as t ^ t^. 
Therefore we have a crossover from ballistic to diffusive behavior as in the classical case. 
On the other hand, if t » we have SE"^ ~ const. This is due to quantum-mechanical 
recurrences [18]. We shall argue that the latter result is valid only for extremely slow 
velocities, for which ts <^ tprt, and provided Landau-Zener transitions are ignored. This 
is the quantum-mechanical adiabatic regime. For non extremely slow velocities we have 
*H ^ iprti and consequently there is a second crossover a.t t = t^^t from diffusive- like 
behavior to genuine diffusion. In the latter case there are no recurrences, and QCC holds 
also for t > tn- 



20 



DoRON Cohen 



17. — Actual and Parametric Dynamics 

The simplicity of linear response theory is lost once we try to formulate a controlled 
version of it. Therefore it is better to use a more conventional approach and to view the 
energy spreading as arising from transitions between energy levels. The transition prob- 
ability kernel and the parametric kernel can be written using standard Dirac notations 
as follows: 

(57) PMm) = |U„„(OP = |(n(x(i))|U(i)|m(0))|2 

(58) P(n|m) = |T„„(a;)|2 = |(n(a;)|m(0))|' 

The evolution matrix Unmit) can be obtained by solving the Schroedinger equation 

(59) = -liJ„ .„ 

(60) where W„to 

The derivation of (59) follows standard procedure [20] . The transformation matrix T„m 
can be obtained by considering the same equation with the first term on the right hand 
side omitted. We shall refer to Tnm{x{t)) as describing the parametric dynamics (PD), 
while Unm{t) describes the actual dynamics (AD). For PD the velocity x = V plays no 
role, and it can be scaled out from the above equation. Consequently, for PD, parametric 
scales and temporal scales are trivially related via the scaling transformation 6x = Vt. 
For short times, as long as the energy differences between the participating levels are 
not yet resolved, the AD coincided with the PD. This is the quantum-mechanical sudden 
approximation, which we are going to further discuss later on. 

The generic parameters that appear in the quantum-mechanical theory are summa- 
rized in Table I. The specification of the mean level spacing A is not dynamically signif- 
icant as long as t ^ tn- Longer times arc required in order to resolve individual energy 
levels. Thus we come to the conclusion that in the time regime t < t^ there is a sin- 
gle generic dimensionless parameter, namely VpR, that controls QCC. We shall see that 
the quantum mechanical definition of slowness, namely t^, ^ tp^t, can be cast into the 
form <C 1. On the other hand in the classical limit we have iipn ^ Vsc ^ 1- Thus 
the dimensionless parameter Vp^ marks a border between two regimes where different 
considerations are required in order to establish QCC. 

18. — Perturbation theory 

We can use Equation (59) as a starting point for a conventional first-order perturba- 
tion theory. For short times, such that Pt(m\m,) ^ 1, the transition probability from level 
m to level n is determined by the coupling strength IW^mp! by the energy difference 



an 



fix 

En — Err 



dx 



Chaos, Dissipation and Quantal Brownian Motion 



21 



Classical Parameters: 

T^i = correlation time 

Quantum- Mechanical Parameters: 

A5 = band width 
A = level spacing 

Primary Dimensionless Parameters: 

= scaled velocity = \/ D^^ T^i / 

b — scaled band width — A5 / A 

Secondary Dimensionless Parameters: 

1! — h^''^ V — T I T'''" 

•^RMT ^ *^PR ' cl / ' c 



Table I. - Generic parameters in the classical and in the quantum-mechanical theories of dis- 
sipation. Note that quantum-mechanics requires the specification of two additional param- 
eters. However, for the purpose of QCC considerations only Aj, is significant. Note that 
v-Lz S> wrmt ^ VpR. In the classical limit all of them ^ 1. 



{En— Em) and by the correlation function F{t). The latter describes loss of correlation 
between 'Wnm{x{Q)) and '^nm{x{t)). It is defined via 

(61) ( W:„(i+T) W„™(t) ) = \Wnm? F{t) 

with the convention -F(O) = 1. It is now quite straightforward [20] to obtain, using 
first-order perturbation theory, the following result: 



(62) PMm) « tPt 



h 



2 



for n 7^ m 



The function Ft{uj) describes the spectral content of the perturbation. For a constant 
perturbation (-F(r) = 1) it is just given by equation (54). For a noisy perturbation F{t) 
is characterized by some finite correlation- time Tc, and therefore the function Ft{Lo) is 
modified as follows: 

(63) Ft{aj) = F{aj) for t > Tc 



22 



DoRON Cohen 



where F{u) is the Fourier transform of the correlation function F{t). In order to use (62) 
we should determine how F{t) look like, and in particular we should determine what is 
the correlation-time Tc- We postpone this discussion, and assume that F{t) and hence Tc 
are known from some calculation. The total transition probability is p{t) = P{n\m), 
where the prime indicates omission of the term n = m. First order perturbation theory is 
valid as long as p{t) <^ 1. This defines the breaktime t'^^^ of the perturbative treatment. 
Wc use the notation t/^^^ rather than t^^^, since later we are going to define an improved 
perturbative treatment whore tp„ is defined differently. 

Using the above first-order perturbative result we can obtain the previously discussed 
expression (53) for the energy spreading: 



We see that the perturbative result would coincides with the linear-response result only 
if the coupling matrix-elements could have been treated as constant in time. The above 
derivation imply that wc can trust this formula only for a short time t < t'^^^. However, 
later we shall sec that it can be trusted for a longer time t < fp^. 

It is now possible to formulate the conditions for having restricted QCC. By 're- 
stricted' QCC we mean that only 5E'^ is being considered. We assume that the result 

(64) is valid for t < fp^. Wc also assume that as well as F{t) and hence Tc arc known 
from some calculation. The following discussion is meaningful if and only if the following 
Fermi-golden-rule (FGR) condition is satisfied: 

(65) FGR-condition: Either or Tc <C t^rt 
It is essential to distinguish between two different possible scenarios: 

(66) Resonance-limited transitions: Tc ^ 

(67) Band-limited transitions: Tc <C 

For resonance-limited transitions, finite Tc has no consequence as far as dE^ is concerned: 
The crossover to diffusive behavior SE'^ oc t happens at t ~ t^, and this diffusive behavior 
persists for t > Tc with the same diffusion coefficient. On the other hand, for band-limited 
transitions wc have at t ~ Tc a pre-mature crossover from ballistic to diffusive behavior. 
Consequently the classical result is suppressed by a factor {tc/t^i) <C 1. This is due 
to the fact that the transitions between levels are limited not by the resonance width 
(embodied by F{u;)), but rather by the band- width of the coupling matrix elements 
(embodied by Ce{u:)). 



(64) 




n 



Chaos, Dissipation and Quantal Brownian Motion 



23 



19. — The over-simplified RMT picture 

In order to have practical estimates for t'^^^ and Tc it is required to study the statistical 
properties of the matrix W„m. This matrix is banded, and its elements satisfy: 



(68) 



w 



1 \' 1 

for \n-m\ < 6/2 



qm 



(n— m)^ 

The time scale r^" is related via Sx = Vt to the parametric scale 



(69) Sxl 



/27rnA 



Vmi/dxUF) V 



This is the parametric change which is required in order to mix neighboring levels. Given 
two levels n and m one observes that Sx"^"" also determines the correlation scale of the 
matrix-element W„m(a;(t))- These observations can be summarized as follows: 

(70) t'^^^ = Tc = Tg" for standard perturbation theory. 

In the regime i^rmt 1 we have t^i . Therefore we have resonance limited transitions 
and we can relay on (64) in order to establish restricted QCC. On the other hand in the 
regime Vrmt '> 1 the FGR condition (65) is not satisfied, and the expected crossover at 
t = r^i is out-of-reach as far as the standard version of perturbation theory is concerned. 

In the spirit of random- matrix-theory (RMT) we may think of Wnm{x) of Equation 
(59) as a particular realization which is taken out from some large ensemble of (banded) 
random matrices. In order to have a well defined mathematical model we should specify 
the .T-corrclations as well. It looks quite innocent to assume that the only significant 
correlations are those expressed in Eq.(61). In other words, let us assume, following [16], 
that cross correlations between matrix elements can be ignored. One finds then that (64) 
should hold for classically long times. This claim can be summarized as follows: 

(71) tprt = tfpc and Tc = r^" for the over-simplified RMT picture. 

If the over-simplified RMT assumption were true it would imply that in the Urmt ^ 1 
regime the transitions would be band-limited and consequently classical diffusion would 
be suppressed by a factor (1/vrmt) "C 1. Note that in the semiclassical limit we have 
indeed iirmt ^ 1- Therefore, it is implied that the classical limit would not coincide 
with the classical result. Obviously, we expect this conclusion to be wrong, and indeed 
we shall demonstrate that cross correlations between matrix elements cannot be ignored. 
This will be done by transforming the Schroedinger equation to a more appropriate basis. 



24 



DoRON Cohen 



20. — The perturbative core-tail spreading profile 

The standard version of perturbation theory is vahd for an extremely short time. See 
(70). In order to formulate an improved version of perturbation theory it is essential to 
understand the perturbative structure of Pt{n\m). Once t > r^" neighboring levels are 
mixed and consequently the probability kernel acquires a non-trivial core-tail structure 
which is illustrated in Fig. 7. The 'improved' version of perturbation theory should be 
applicable as long as the core-tail structure is maintained. 



P(nlm) 







core 




b(t)^ 








s(t) ^ 






— ► 



tail 



Fig. 7. - Schematic illustration of a generic core-tail spreading profile. The core-width b{t) is 
defined by the participation-ratio. The second-moment should satisfy 6(f) <^ s{t) <^ b, where 

b is that bandwidth. In case of Pt{n\m) the tail becomes (for t > Td) resonance-limited rather 
than band-limited. In the resonance-limited case the bandwidth b in the above figure should be 
replaced by {fi/t)/A, and accordingly the requirement is b{t) <^ s{t) <C {fi/t)/A. 



Now we shall characterize the main features of a generic core-tail spreading profile. 
The expression for the probability kernel Pt{n\m) can be written schematically as follows: 

(72) Ptin\m) w Core(n-m) -|- Tail(n-m) for t < t^^t 
The kernel is characterized by two scales: 

(73) b{t) = core width = i'^{Pt{n\m)f 



(74) s{t) = spreading = j ^(n— m)^ Pt{n\m) 



-1/2 



such that b(t) <^ s{t) <C b. For t <§; r^'" wc have a trivial core with b(t) » 1. whereas 
for t ^ r^" we have a non-trivial core with b{t) ^ 1. The matrix elements satisfy 
(X l/(n— m)^. We shall see (see (78)) that in the 'band-limited tail' case we 
have Pt{n\m) ~ const/(n— m)^ up to the cutoff b, while for the 'resonance-limited tail' 
case we have Pt(n|m) ^ const/(n— m)^ up to the cutoff {h/t)/A. One should realize 
that the power-law behavior of the tail is 'fast' enough in order to guarantee that b{t) is 



Chaos, Dissipation and Quantal Brownian Motion 



25 



independent of the tail's cutoff. The cutoff does not have any effect on the evolving core. 
On the other hand, the second moment s{t), unlike b{t), is predominantly determined by 
the tail's cutoff, and it is independent of the core structure. 

21. — An improved perturbation theory 

In order to extend perturbation theory beyond r^?" it is essential to eliminate the non- 
perturbative transitions within the core. This can be done by making a transformation 
to an appropriate basis as follows: 

(75) Unit) = ^t„„ Cm{t) 

m 

(76) T„TO = Tnm if \n—m\ < b' /2, else zero. 

The amplitudes c„(t) satisfy the same Schroedinger equation as the a„(t), with a trans- 
formed matrix W. The general expression for W is quite complicated, but we are 
interested only in the core-to-tail transitions for which 

(77) (W)„„ = (f twf ) 

nm for |n— m| > b' 

(no approximation is involved). Once this transformation is performed the 'new' Schroedinger 
equation is characterized by a new correlation time Tc and by a new perturbative time 
ip^^. Both Tc and t'^^^ depend on the free parameter h' . Our choice of the course- graining 
parameter 6' is not completely arbitrary. The restrictions are: 

• Unitarity is approximately preserved: b{t) <C 6'. 

• Core-to- Tail transitions are preserved: b' <C b. 

• Long effective correlation time is attained: Td <C Tc 

The feasibility of the last requirement deserves further discussion. One should realize that 
for 6' = 6 we are actauUy transforming the Schrodinger equation to an a;-independent 
basis. Therefore the transformed matrix T^WT becomes correlated on a time scale r°', 
which has been discussed in Section 5. As we change b' from b to smaller values, we 
expect Tc to become smaller. By continuity, we expect no difficulty in satisftying the 
conditions 6' <C 6 and Tc ^ Td simultaneously. 

The usefullness of the above transformation stems from the fact that due to the 
elimination of non-perturbative transitions within the core, t'^^ becomes much longer 
than Tc'". At the same time the information which is required in order to determine the 
second moment s{t) is not lost. We have |W„„i| w |W„„i| for core-to-tail transitions, 
and a practical approximation for the 'renormalized' spreading profile would be 



26 



DoRON Cohen 



The behavior for |n— m| < b' is an artifact of the transformation and contains false infor- 
mation. However, for the calculation of the second moment only the tail is significant. 
The tail is not affected by our transformation and therefore we obtain the same result 
(64) for SE^ with one important modification: a different effective value for t^. Moreover, 
since b' is chosen such that Tc ^ Td, it follows that the transitions are resonant-limited 
and consequently restricted QCC is established also in the domain Vrmt > 1- 

The validity of the above QCC considerations is conditioned by having <C ip^- 
Breakdown of the present version of perturbation theory happens once the total transition 
probability in (78) becomes non- negligible (of order 1). Thus 

i'^^) *prt = {b'y^^ X T^"" for the improved perturbation theory. 

It is easily verified that the latter condition <C <p„ cannot be satisfied if Wpr > 1. This 
is not just a technical limitation of our perturbation theory, but reflects a real difference 
between two distinct routes towards QCC. This point is further illuminated in the next 
sections. 

22. — Consequences of the improved perturbative treatment 

Our perturbation theory is capable of giving information about the tail, and hence 
about the second moment. Given t, one wonders how much 6' can be 'pushed down' 
without violating the validity conditions of our procedure. It is quite clear that b' b{t) 
is a necessary condition for not having a breakdown of perturbation theory. If we assume 
that the energy-spreading-profile is characterized just by the single parameter b{t), then 
the condition b' b{t) should be equivalent to t <C t'^^^. Hence the following estimate is 
suggested: 



If we want to have a better idea about the core structure we should apply, in any special 
example, specific (non-perturbative) considerations. For example, in case of the piston 
example we can use semiclassical considerations in order to argue [21] that the core has a 
Lorentzian shape whose width is h/r^^i. This structure is exposed provided h/r^^, <C b{t), 
leading to the condition 6x Xe- Else we have a structure-less core whose width is 
characterized by the single parameter b{t). 

We turn now to determine the (5xp,, of the parametric dynamics (PD), and then the 
fprt of the actual dynamics (AD). Recall that PD is obtained formally by ignoring the 
differences (En— Em), which implies that we can make in (78) the replacement Ft t—^ t. 
Thus the tail of P(n|m) is band-limited and consequently the second moment is 



(80) 



b{t) = {t/rr? 



(81) 



s{tf = 6 X (l/r;?")^ <^ [band-limited tail] 



in agreement with the classical ballistic result. Our procedure for analyzing the core-tail 
structure of P{n\m) is meaningful as long as we have b{t) <C s{t) <C b. This defines an 



Chaos, Dissipation and Quantal Brownian Motion 



27 



upper time limitation 



'prt 



5x^^^/V, where 



(82) 



cl 



At t = tprt wc have b{t) ^ s{t) ^ b, and we expect a crossover from a ballistic-like 
spreading to a genuine ballistic spreading. 

The AD departs from the PD once the energy scale Ab is resolved. This happens 
when t ^ Tci . The perturbative approach is applicable for the analysis of the crossover at 
t ~ provided Vt^i <^ Sxp^t. This is precisely the condition Upi, ^ 1. For t 3> the tail 
becomes resonance limited (|n— to| < (h/t)/A) rather than band limited (|n— to| < b) 
and we obtain: 

(83) s{t)'^ = (l/r^°)^ t [resonance-limited tail] 

in agreement with the classical diffusive result. Our procedure for analyzing the core-tail 
structure of Pt{n\m) is meaningful as long as we have b{t) <C s{t) <C b. This defines a 
modified upper time limitation 



At t = tprt wc have h{t) s{t) ^ b, and we expect a crossover from a diffusive-like 
spreading to a genuine diffusive spreading. 

23. — The quantum mechanical sudden approximation 

It is now appropriate to discuss the quantum mechanical sudden approximation. For 

the perturbative scenario (upj, <C 1) wc have already mentioned that the AD departs 
from the PD at 4dn = 7"ci) which is the time to resolve the energy scale A;,. In case of 
the non-perturbative scenario (v^^ » 1) there is an earlier breakdown of the quantum 
mechanical sudden approximation. This is because wc have r^i 3> tprt and consequently 
at t = T„, we already have s{t) ^ b. Therefore t,^^ should be defined as the time to 
resolve the energy scale which is associated with s{t). It leads to 



The various time scales are summarized in Table II. The non-perturbative crossover from 
genuine-ballistic to genuine-diffusive behavior in not trivial. If v^^ 1 we can relay on 
semiclassical considerations in order to establish the existence of this crossover. More 
generally, for Vp^ ;:g> 1, we would like to have (but we do not have yet) an appropriate 

effective RMT model. This effective RMT model should support genuine-ballistic motion, 
and to be further characterized by an elastic scattering time t^. 



(84) 




[applies to Wpr <C 1] 



(85) 





28 



DoRON Cohen 



Perturbative route (fpn <^ 1): 

^sdn Tel ^prt 

At t = r,, b{t) < s{t) < 6 ~ {fi/t)/A 

At t^t^^, bit) s{t) ^ {h/t)/A<^b 



Non-perturbative route {v^^ ^1): 

^prt < ^sd„ < 

At t = t,,, ~ ~ b < (/i/t)/A 

At t = ta„ 6 < s(t) ~ (/i/t)/A 

At i = T„ b {h/t)/A <^ s{t) 



Table II. - Various time scales in the route to stochastic behavior. 



24. — The quantum mechanical adiabatic approximation 

The previous analysis has emphasized the role of core-to-tail transitions in energy 
spreading. Our assumption was that these transitions are not suppressed by recurrences. 
This is not true in the quantum- mechanical adiabatic regime {vi^z *C !)• Following [15] 
it is argued that energy spreading in the latter regime is dominated (eventually) by 
Landau-Zener transitions between near-neighbor levels. One obtains 

/ 1 \l-(/9/2) 

(86) ^ { — ] Dt for v^^^l 

where is the classical result. The non-trivial nature of Landau-Zener transitions and 

the statistics of the avoided-crossings is taken into account. One should use /? = 2 for 
the Gaussian unitary ensemble (GUE) and /3 = 1 for the Gaussian orthogonal ensemble 
(GOE). 



Chaos, Dissipation and Quantal Brownian Motion 



29 



25. — Classical Brownian motion 

It is possible to argue that the reduced motion of the particle ('piston' if the motion 
is constrained to one dimension) obeys the following Langevin equation: 

(87) mx + r?x = ^ 

where rj = is the friction coefficient. The stochastic force is redefined T := 
such that it is zero on the average. Its second moment should satisfy 

(88) (.F(W))at« = <A(i-i') 

where = C{t—t'). Usually, it is further assumed that higher moments are de- 

termined by Gaussian statistics. The correlation time is denoted as before by Td. It is 
better to view the stochastic force as arising from a stochastic potential 

(89) T{t) = J^(i,x(t)) = -VU{x,t) 

(90) {U{Tc",t")U{x',t')) = <l){t"-t')-w{x"-x') 

The spatial correlations of the stochastic potential are assumed to be characterized by a 
spatial scale £. The natural tendency is to identify £ with 5a;°', but this point deserves 
further non- trivial discussion. The normalization convention w"(0) = — 1 is used, and 
therefore there is consistency of (90) with (88). 

26. - The DLD Hamiltonian 

Formally, the Langevin equation (87) with (90) is an exact description of the reduced 
dynamics that is generated by the DLD Hamiltonian: 

(91) n = ^ + ^c„Q„tx(x-x„) + J2flL + ^rnu;lQl\ 

where Xq, is the location of the a oscillator, u(x — Xq) describes the interaction between 
the particle and the a oscillator, and Cq are coupling constants. It is assumed that the 
function u(v) depends only on |r|. The range of the interaction is I. The oscillators 
are distributed uniformly all over space. Locally, the distribution of their frequencies is 
ohmic. Namely, 

(92) o' yj — ~ — ^(w — Wet) (5(x — Xc) = ?7W for w < I/tci 

This distribution is uniquely determined by the requirement 0(r) = C(t). The spatial 
correlations are determined via 



(93) 



/OO 
-oo 



30 



DoRON Cohen 



For example, we may consider a Gaussian u{r) for which 
(94) w{r) = e^exp(^-^{^y^ 

Certain generalizations of this assumption has been considered in [9], but are of no 
interest here. In the formal limit i ^ oo the DLD model reduces to the well known ZCL 
model. The ZCL model is defined by the interaction term x ■ J2 CaQa- 




Fig. 8. - Illustration of the ZCL model (left upper drawing), versus the DLD model (left lower 
drawing). The instantaneous potential that is experienced by the particle is either linear (right 
upper drawing) , or of disordered nature (right lower drawing) correspondingly. If the fluctuations 
are uncorrelated in time (WNA), then the two models are classically equivalent. There is no 
such equivalence in the quantum-mechanical case. 



27. — The white noise approximation (WNA) 

One wonders whether the noise in Langcvin equation can be treated as white noise, 
meaning that Td is irrelevant and we can set ~ in any significant result. The 



Chaos, Dissipation and Quantal Brownian Motion 



31 



following condition defines the classical notion of white noise: 



(95) 



Generic Brownian Motion 




If the condition is not satisfied, then r^i is larger than and the particle performs a 
stochastic motion that depends crucially on the 'topography' of the stochastic potential. 
Note that upon the identification of £ with 5x°^ the condition above becomes equivalent 
to the trivial requirement of classical slowness (Sec. 5). 

In the quantum mechanical analysis it is found that 4>{t) is further characterized by 
a correlation time Ti/ik^T). Thus, the following additional requirement should be met if 
we wish to apply a white noise approximation. 

(96) High Temperatures <^ ^ < 

in 

Note that the smallest meaningful velocity is the thermal velocity, and therefore the above 
condition cannot be satisfied unless the thermal wavelength At = h/{mksT) is much 
smaller than £. From now on we assume that both classically and quantum mechanically 
we can use the white noise approximation. Thus we can use the formal substitution 
(^(r) = u6{t). 

28. - Consequences of the WNA 

In the general case, (88) is less informative than (90). However, in case of a classical 
particle that experiences white noise, the additional information is not required at all! 
Classically, the spatial correlations, and hence £, are of no importance. This is because 
at each moment a classical particle samples one definite point in space. The particle 
'docs not care' about the force elsewhere. In the quantum mechanical case the latter 
observation is wrong. A quantum mechanical particle samples each moment a finite 
region in space and therefore spatial correlations of the effective stochastic potential 
become important, even if the noise is white. 

Both classically and quantum mechanically the reduced dynamics that is generated 
by the DLD model can be obtained analytically and cast into the form 



where pt{R, P) represent either the classical-state or the quantum- mechanical-state of the 
particle. In the latter case it is a Wigner function. It is a consequence of the WNA that 
the propagator has a Markovian property. Namely, it can be written as the composition 
of smaller 'time steps'. Consequently pt{R,P) satisfies a master equation of the type 



(97) 



Pt 



dRodPo /C(i?, P|i?o, -Po)Pt=o(-Ro) -Po) 



(98) 




32 



DoRON Cohen 



The classical version of this equation is known as the Fokker-Planck equation. We shall 
discuss shortly its quantum-mechanical version. 

29. — The reduced propagator 

The reduced propagator of the DLD model can be obtained [8] classically as well as 
quantum mechanically using a path integral technique. In the quantum mechanical case 
it is known as the FV-formalism. The classical version of FV formalism can be regarded 
as a formal solution of Langevin equation. It can be obtained without going via the DLD 
Hamiltonian, but then one should assume Gaussian statistics. 

In the absence of coupling to the environment, the free-motion propagator of Wigner 
function is the same both classically and quantum mechanically: 

(99) IC(B.,P\Ro,Po) = /Clei = 27t6{P-Po) S{{R-Ro) ~ ^t) 
Taking the environment into account, one obtains, in the classical case, 

(100) K:{R,P\Ro,Po) = 4a2ped = Gaussian 

This result is, as a-priori expected, independent of i. The average velocity of of the 
particle goes to zero on a time scale = {r]/m)~^, while the spreading of the Gaussian is 

(101) V = ut iovt<^Tr, 

(102) Sx"^ = (l/(3m2))z/t3 fori«;T^ 

(103) Sp^ = mkj,T fort>T^ 

(104) Sx'^ = [v/rfY fort>T^ 



The above result holds also in the quantum mechanical case provided we take the limit 
^ — > 00. No genuine quantum- mechanical effects are found if the ZCL model is used to de- 
scribe Brownian motion! Recall again that we are considering here the high-temperature 
case where the WNA applies. Now we want to discuss the finite £ case. Here one obtains 
[8, 9] the following expression: 

(105) /C = T^R/,*4a2ped + e ^\l-Wn/t^) kW. 

As suspected, unlike in the classical case, the quantum mechanical result depends on £ 
in an essential way. W{R—R', P—P') is a smooth Gaussian-like kernel that has unit- 
normalization. Its spread in phase space is characterized by the momentum scale h/i, 
and by an associated spatial scale. The symbol -k stands for convolution. Thus, the 
classical propagator is smeared on a phase-space scale that correspond to Ap = h/i and 
there is an additional un-scattered component that decay exponentially and eventually 
disappears. The structure of the propagator is illustrated in Fig. 9. The significance of 
this structure will be discussed shortly. 



Chaos, Dissipation and Quantal Brownian Motion 



33 



O 



X 



i 




X 



Fig. 9. - Upper plot: Phase space illustration for the structure of the DLD propagator. Lower 
plot: The projected phase-space density. Note the existence of an un-scattered component. 
Such component is absent in the ZCL limit as well as in the classical analysis. 



30. — Master equation 

To write down an explicit expression for the propagator is not very useful. Rather, 
it is more illuminating to write an equivalent Master equation. This is possible since at 
high temperatures the propagator possess a Markovian property. The final result [9] is 



(106) 



dp 
dt 



Or —P + dp ^Gf*P + i^Gn* 
m m 



Gf = TT (^) 

Gn = ^J'T {w{r)-w{0)) 

For convenience the friction kernel Gf and the noise kernel Gn are expressed in term 
of smooth Gaussian-like scaling functions Gp and Gn that are properly normalized to 
unity. The notation J-'T stands for Fourier transform. 



1 

h 



Gf 



P-P' 

h/e 



n/i 



G 



N 



P-P' 

n/i 



5{P-P') 



34 



DoRON Cohen 



If Wigner function does not possess fine details on the momentum scale h/^, then the 
convolution with Gp can be replaced by multiplication with 1, and the convolution with 
Gn can be replaced by d"^ /dP^. These replacements are formally legitimate both in the 
classical limit Ti ^ Q, and in the ZCL limit £ oo. One obtains then the classical Fokker 
Planck equation 



(107) 



dp 
dt 



Or -P 
m 



dp^P 
m 



'qP2 



The Fokker Planck equation is just a continuity equation with an added noise term that 
reflects the effect of the stochastic force. The Fokker Planck equation is equivalent to 
solving Langevin equation as well as to the Gaussian propagator (100). 



31. — Brownian motion and dephasing 

Wigner function may have some modulation on a fine scale due to an interference 
effect. The standard text-book example of a two slit experiment is analyzed in [9]. In 
case of the ZCL model (£ = oo), the propagator is the same as the classical one, and 
therefore we may adopt a simple Langevin picture in order to analyze the dephasing 
process. Alternatively, we may regard the dephasing process as arising from Gaussian 
smearing of the interference pattern by the propagator. In case of the DLD model 
(finite t) we should distinguish between two possible mechanisms for dephasing: 

• Scattering (Perturbative) Mechanism. 

• Spreading (Non-Perturbative) Mechanism. 

Actually, it is better to regard them as mechanisms to maintain coherence. The first 
mechanism to maintain coherence is simply not to be scattered by the environment. The 
second mechanism to maintain coherence is not to be smeared by the propagator. The 
first mechanism is absent in case of the ZCL model. 

Let us discuss how coherence is lost due to the scattering mechanism. The discussion 
is relevant if Wigner function contains a modulation on a momentum scale much finer 
than fi/i, else I becomes non-rclcvant and we can take it to be infinite. One should 
observe that such modulation is not affected by the friction, but its intensity decays 
exponentially in time. This is based on inspection of either the propagator (105), or the 
equivalent Master equation (106). In the latter case note that the convolution with Gn 
can be replaced by multiplication with —{ijKf. The decay rate is 

(108) — = assuming WNA 

This is the universal result for the dephasing rate due to the 'scattering mechanism'. It is 

universal since it does not depend on details of the quantum-mechanical state involved. 
However, the validity of this result is restricted to the high temperature regime, where 



Chaos, Dissipation and Quantal Brownian Motion 



35 



the WNA can be applied. Extensions of this result, as well as discussion of dephasing at 
low temperatures can be found in [9, 11]. 

32. — The open question 

The effective-bath approach suggests that there is a universal description of quantal 
Brownian motion. If indeed the effective-bath approach is universally applicable, it is 
implied that 

(109) chaos dynamical disorder 

meaning that the motion under the influence of chaotic environment is effectively the 
same as the motion under the influence of dynamical disordered environment. 

At this stage the conditions for the validity of (109) are yet unclear. It is interesting 
however to emphasize the practical implications of such claim. The starting point should 
be a specification of the effective correlation scale i. For the 'piston' example the most 
obvious guess is 



One may estimate now the coherence time using the substitution v = rn^v^/L where m 
is the mass of the gas particle, and L = VeT^oi is the mean path-length between collisions 
with the piston. The result is 



where Ae is the De-Broglie wavelength of the gas particle. Note that we are assuming 
the WNA, and therefore we must satisfy Ab <C i, where Ab is the De-Broglie wavelength 
of the piston. 

The (effectively) disordered nature of the environment is significant only within the 
time domain t < T^p. The non-trivial effect that is 'predicted' by solving the DLD model 
is that the reduced propagator has a coherent 'unscattered-component' plus a smearing 
'scattered-component'. The latter is created due to the exchange of momentum-quanta 
of typical magnitude fi/t. Assuming 'hard walls' (in the sense of (110)) we get the result 
h/i ~ 2mvE and T^p = T^ai- This is a very 'funny' result since it has a trivial classical 
interpretation in terms of the actual ('piston') model, whereas within the effective-bath 
approach it appears as a genuine quantum- mechanical effect! 



(110) 




ff> A, 
else 



'E 



(111) 




* * * 



I thank Uzy Smilansky and Eric Heller for interesting discussions, and Shmuel Fish- 
man for fruitful interaction in intermediate stages of the study. 



36 



DoRON Cohen 



REFERENCES 

[1] For a detailed introduction sec [20] and [8]. 

[2] R.P. Feynman and F.L. Vernon Jr., Ann. Phys. (N.Y.) 24, 118 (1963). 
[3] R. Beck and D.H.E. Gross, Phys. Lett. 47, 143 (1973). 
R. Zwanzig, J. Stat. Phys. 9, 215 (1973). 

D. H.E. Gross, Nuclear Physics A240, 472 (1975). 

[4] J. Blocki, Y. Boneh, J.R. Nix, J. Randrup, M. Robel, A.J. Sierk and W.J. Swiatecki, 
Ann. Phys. 113, 330 (1978). 

[5] S.E. Koonin, R.L. Hatch and J. Randrup, Nuc. Phys. A 283, 87 (1977). 
S.E. Koonin and J. Randrup, Nuc. Phys. A 289, 475 (1977). 

[6] K. Mohring and U. Smilansky, Nuclear Physics A338, 227 (1980). 

[7] A.O. Caldeira and A.J. Leggett, Physica 121 A, 587 (1983). 

[8] D. Cohen, Phys. Rev. Lett. 78, 2878 (1997); Phys. Rev. E 55, 1422 (1997). 

[9] D. Cohen, J. Phys. A 31, 8199-8220 (1998). 
[10] A. Bulgac, G.D. Dang and D. Kusnozov, Phys. Rev. E 58, 196 (1998). 
[11] D. Cohen and Y. Imry, Phys. Rev. B 59, 11143 (1999). 

[12] W.H. Louiscll, Quantum Statistical Properties of Radiation, (Wiley, London, 1973). 
[13] A.R. Kolovsky, Phys. Rev. E 50, 3569 (1994). 

[14] E. Ott, Phys. Rev. Lett. 42, 1628 (1979). R. Brown, E. Ott and C. Grebogi, 

Phys. Rev. Lett, bf 59, 1173 (1987); J. Stat. Phys. 49, 511 (1987). 
[15] M. Wilkinson, J. Phys. A 21, 4021 (1988); J. Phys. A 20, 2415 (1987). 
[16] M. Wilkinson and E.J. Austin, J. Phys. A 28, 2277 (1995). 

E. J. Austin and M. Wilkinson, Nonlinearity 5, 1137 (1992). 

[17] C. Jarzynski, Phys. Rev. Lett. 74, 2937 (1995); Phys. Rev. E 48, 4340 (1993). 
[18] J.M. Robbins and M.V. Berry, J. Phys. A 25, L961 (1992). 

M.V. Berry and J.M. Robbins, Proc. R. Soc. Lond. A 442, 659 (1993). 

M.V. Berry and E.G. Sinclair, J. Phys. A 30, 2853 (1997). 
[19] D. Cohen, Phys. Rev. Lett. 82, 4951 (1999). 

[20] D. Cohen, Annals of Physics 283, 175 (2000). (Long detailed paper). 

[21] D. Cohen and E.J. Heller, Phys. Rev. Lett. 84, 2841 (2000). 

[22] M.V. Berry, Proc. R. Soc. Lond. A 423,219 (1989); 424,279 (1989). 

[23] M.A. Sepulveda, S. Tomsovic and E.J. Heller, Phys. Rev. Lett. 69, 402 (1992). 

[24] M. Feingold and A. Peres, Phys. Rev. A 34 591, (1986). M. Feingold, D. Leitner, M. 

Wilkinson, Phys. Rev. Lett. 66, 986 (1991); M. Wilkinson, M. Feingold, D. Leitner, J. 

Phys. A 24, 175 (1991); M. Feingold, A. Gioletta, F. M. Izrailev, L. Molinari, Phys. Rev. 

Lett. 70, 2936 (1993). 
[25] E. Wigner, Ann. Math 62 548 (1955); 65 203 (1957). 

[26] V.V. Flambaum, A.A. Gribakina, G.F. Gribakin and M.G. Kozlov, Phys. Rev. A 50 267 

(1994). 

[27] G. Casati, B.V. Chirikov, I. Guarneri and F.M. Izrailev, Phys. Rev. E 48, R1613 (1993); 

Phys. Lett. A 223, 430 (1996). 
[28] F. Borgonovi, I. Guarneri and F.M. Izrailev Phys. Rev. E 57, 5291 (1998). 

[29] D. Cohen, Phys. Rev. A 44, 2292 (1991). 

[30] F. M. Izrailev, T. Kottos, A. Politi, S. Ruffo and G. P. Tsironis, Europhys. Lett. 34, 441 
(1996). F.M. Izrailev, T. Kottos, A. Politi and G.P. Tsironis, Phys. Rev. E 55, 4951 (1997). 
[31] D. Cohen, F.M. Izrailev and T. Kottos, Phys. Rev. Lett. 84, 2052 (2000).