Skip to main content

Full text of "Thermal Fluctuations in Quantized Chaotic Systems"

See other formats

November 1995 

Thermal fluctuations in quantized chaotic systems 

Mark Srednicki* 

Department of Physics, University of California, Santa Barbara, CA 93106 


We consider a quantum system with degrees of freedom which is 
classically chaotic. When is large, and both fi and the quantum 
energy uncertainty Ai? are small, quantum chaos theory can be used 
to demonstrate the following results: (1) given a generic observable A, 
the infinite time average A of the quantum expectation value {A{t)) is 
independent of all aspects of the initial state other than the total energy, 
and equal to an appropriate thermal average of A; (2) the time variations 
of {A{t))—A are too small to represent thermal fluctuations; (3) however, 
the time variations of {A^{t)) — {A{t))'^ can be consistently interpreted 
as thermal fluctuations, even though these same time variations would 
be called quantum fluctuations when is small. 

Typeset using REVT^ 

*E— mail: markQtpau . physics . ucsb . edu 


In this paper we examine the compatibihty of certain results in quantum chaos theory 
with standard results in statistical mechanics. We consider a bounded, isolated, many- 
body quantum system whose classical limit is chaotic. Given an initial state \ip{0)) and 
a generic observable A, we ask the following questions. What is the infinite time average 
of {A(t)) = {'i/j(t)\A\ilj{t))7 Is it independent of the initial state |V'(0))? If so, is it equal 
to an appropriate thermal average of A7 What are the root-mean-square fluctuations, in 
time, of {A{t)) about its infinite time average? Are these fluctuations correctly predicted by 
statistical mechanics? 

We begin by noting that the energy spectrum of a bounded quantum system is purely 
discrete; if the system is classically chaotic, and also has no discrete symmetries, then the 
energy eigenvalues Ea are almost always nondegenerate |jl|] . Since we assume that the system 
is isolated, its state at time t is 

|^(t)) = E^^ae-^^"*/^|a), (1) 


where the C^'s specify the initial state, and we assume the usual normalization 

T.\Ca\' = l. (2) 


The expectation value of an observable A at time t is 

{A{t)) ^ m)\A\m) 

= $:c:C^e^(^"-WM„^, (3) 



A^p = {a\A\(3) (4) 

are the matrix elements of A in the energy eigenstate basis. The infinite time average of 
{A(t)) is given by 

A= lim - r dt (A(t)) 

T^OO 1 Jq 


The time averaged fluctuations of {A{t)) about A are given by 

(A(t))-AY = lim - r dt \(A(t))-AY 

J r^oo r Jo L ^ ^ " J 

= E \Ca\'\Cp\'M'. (6) 

We now turn to a discussion of what can be inferred about (^) and (P) from quantum chaos 


Quantum chaos theory is largely based on semi classical arguments; to make use of it, 
we will have to assume that Planck's constant is "small." This means that there is some 
dimensionless combination of parameters, with a single power of Planck's constant in the 
numerator, which serves as an expansion parameter for quantities such as Aap- The relevant 
combination of parameters, which we will call h, obviously depends on the system under 
consideration. How small h has to be depends on both A and the range of energies which are 
of interest. It is particularly difficult to determine the dependence of h on A^, the number of 
degrees of freedom in the system. This question is irrelevant when is small, but crucial 
when N is large. We will not discuss this important problem any further here, however; we 
will simply assume that the correct expansion parameter h, whatever its dependence on N, 
is sufficiently small. 

Given a classically chaotic system with degrees of freedom, we consider an observable 
A which is a smooth function of the classical coordinates and momenta, and which has no 
explicit dependence on h. Then quantum chaos theory predicts that the matrix elements 
Aai3 are given by 

Ao^p = A{E^)6o.p + n^''~^^I^R^p . (7) 

Here A{E) is a smooth function of energy whose leading term in the % expansion is 0{?f). 
The matrix elements Rap are also 0{h^) at leading order, and their values are characterized 
by a smooth distribution, often assumed to be gaussian. Eq. has not been demonstrated 
rigorously, but it follows from a variety of different arguments, including Berry's random- 
wave conjecture for the energy eigenfunctions the analogy between quantized chaotic 
systems and random matrix theory f^, and the semiclassical periodic orbit expansion, 
assuming a certain randomness for the periodic orbits 0. There is, however, one aspect of 
(^) which has been proven rigorously; specifically, 



where H{p, q) is the classical hamiltonian, and A{p^ q) is the classical form of the operator 
A 1^. The limit holds for all energy eigenstates \a) except possibly a subsequence of density 
zero. The right-hand side of (||) is the 0(/i°) contribution to A{Ea). 

For later use, we must also examine the matrix elements of A^. Consider first the diagonal 
elements {A^)aa = Y^pA^pApa] using gives 



A\Ea) + h''-'Y.\Rc.P? +n^''-'^/^2A{Ea)Raa. (9) 

We have grouped the terms as shown because the second term in square brackets is actually 
0{h^), despite the explicit factor of h^~^ . This is because the sum over (3 can be converted 
to an integral over the quantum density of states, and the quantum density of states is 
0{h-^) [0]. One more factor of ?i then arises from converting a quantum energy integral 
into a classical frequency integral 0. Thus, the diagonal matrix elements of A^ have the 
same general structure (|^) as the diagonal matrix elements of A; this is of course required 
for internal consistency, since there was nothing special about A. 

Now consider the off-diagonal elements (v4^)q,^ = J2p AapAp^; using gives 


when a 7^ 7. This time, however, the sum over 13 in the last term does not contribute a 
factor of h~^^^ , because RapRp-y is not positive definite. Instead, we expect Raf^Rfi-y to have 
a phase (or perhaps just a sign) which varies erratically with (3. This implies that the sum 


over (3 of RapRp^ is the same order in h as the square root of the sum over (3 of iRapRp-yl 
this latter sum is 0{fr^^^). Thus we conclude that, overall, the second term on the right- 
hand side of (p!OD is 0{TS^~^''^'^)^ just like the first term, and just like the off-diagonal matrix 
elements of A. Again this is required for the consistency of (|^) with the generic character 
of v4. 

Returning to (H), if we insert (|^) we get 

A = Y.\Co.?A{E^) + 0{h^''-'^/'). (11) 


We now assume that the expected value of the total energy 

{E)=Y.\C^\^E^ (12) 

has a quantum uncertainty 




which is small, in a sense which we will make more precise shortly. This is a natural 
assumption if is large, since states of physical interest typically have A^E ~ N~^^'^{E). 
Note, however, that in this case the smallness of AE does not imply or require the smallness 
of h. 

Assuming AE is small, we can expand A{Ea) about (E) to get 

A{E^) = A{{E)) + {E^ - {E))A'{{E)) + i(i?„ - {E)fA"{{E)) + .... (14) 

Substituting this expansion into (|ll|), we find 

A = A{{E)) + l{AEfA"{{E)) + OUAEf) + Oih^"'-'^'^) . (15) 

Thus, the infinite time average A depends on the expected value of the total energy (i?), 
but is independent of all other aspects of the initial state, provided that % is small enough 
to make the 0{h!'^~^^^'^) term negligible, and provided that 




< 1 . (16) 

This is the more precise criterion for the smallness of AE. 

We are now able to make a connection with statistical mechanics. Mathematically, we 
can choose the |Cq,P's to represent a microcanonical average over an energy range AE 
centered on [E). If this AE is chosen to satisfy ([T6|), then A is equal to this microcanonical 
average of A. Alternatively, we can choose the |Cq,P's to be canonical Boltzmann weights; 


the canonical energy dispersion AE is usually smaller than {E) by a factor of A^~^/^, and 
therefore the canonical AE should satisfy (|T^ when N is large. If so, then A is equal to the 
canonical thermal average of A at whatever temperature results in a total energy of (E). 
Thus, the function A{E) can in principle be calculated, at least up to corrections which are 
Oi^h^^^^^^"^) and 0{N^^), by the methods of canonical statistical mechanics. 

Some time ago, Jaynes ||^ pointed out that a canonical calculation of the size of the 
thermal fluctuations in some observable A must ultimately be based on demonstrating that 
A exhibits time variations with the same root-mean-square amplitude. To study this issue 
in the present context, we first consider the time variations of {A{t)) — A. From (|]), (^, 
and (0), we find 

{A{t)) - Af = 0{h''-') . (17) 

We see that the fluctuations of {A{t)) about A are small. This tells us that, whatever the 
initial value (^4(0)) happens to be, {A{t)) must eventually approach its thermal average A, 
and then remain near A most of the time. (We do not, however, learn anything about the 
time scale of this approach.) Apparently, under appropriate circumstances quantum chaos 
can serve as the dynamical underpinning of certain basic results of statistical mechanics, an 
idea which has already appeared in various guises [||,^. 

On the other hand, (p!7| ) is too small to represent the expected thermal fluctuations of 
A, which are 0{h^). To find thermal fluctuations, we must look at the infinite time average 
of {A'^{t))] this is given by 

A^= lim- r dt (A\t)) 

r^oo T Jq 

= Y.\Ca\HAX^. (18) 


We have already seen that the matrix elements of A^ have the same general structure as 
the matrix elements of A. Therefore, we can immediately conclude that is equal to a 
thermal average of A^, up to corrections which are C'(^'-^~^-'''^) and 0{N~^). 

Putting everything together, we conclude that, up to corrections which are 
and 0{N~^), the infinite time average of (A^lt)) — (A(t))^ is equal to a thermal average of 
{A — Ay. Thus, variations with time of (A^(t)) — (A(t))^ can be interpreted as representing 
thermal fluctuations. It is interesting to note that, in a few-body system, these same time 
variations would be called quantum fluctuations. 

To summarize, results from quantum chaos theory are compatible with results from 
statistical mechanics; quantum chaos theory can even be used as a basis from which one can 
demonstrate, e.g., that the quantum expectation value of an observable must approach its 
thermal average, at least when the number of degrees of freedom is large, the quantum 
energy uncertainty AE is small, and the semiclassical expansion parameter h is small. Just 
how small h needs to be is a question to which we hope to return. Also, we have seen that 
the variations with time of a quantum expectation value are too small to account for the 
expected thermal fluctuations; instead, what would be called quantum fluctuations when A^ 
is small have just the right amplitude to be identified as thermal fluctuations when A is 


I would like to thank John Prcskill for helpful discussions. This work was supported in 
part by NSF Grant PHY-91-16964. 



[1] M. V. Berry, in Les Houches XXXVI, Chaotic Behavior of Deterministic Systems, edited 
by G. looss, R. H. G. Helleman, and R. Stora (North- Holland, Amsterdam, 1983). 

[2] M. V. Berry, J. Phys. A 10, 2083 (1977); R Pechukas, Phys. Rev. Lett. 51, 943 (1983). 

[3] M. Feingold and A. Peres, Phys. Rev. A 34, 591 (1986); T. Prosen, Ann. Phys. 235, 115 

[4] M. Srednicki, Phys. Rev. E 50, 888 (1994). 

[5] O. Bohigas, M.-J. Giannoni, and C. Schmit, Phys. Rev. Lett. 52, 1 (1984); O. Bohigas, 
in Les Houches LII, Chaos and Quantum Physics, edited by M.-J. Giannoni, A. Voros, 
and J. Zinn- Justin (North-Holland, Amsterdam, 1991). 

[6] B. Eckhardt and J. Main, Phys. Rev. Lett. 75, 2300 (1995); B. Eckhardt, S. Fishman, 
J. Keating, O. Agam, J. Main, and K. Miiller, Los Alamos Archive Report No. |chao- 

I dyn/ 9509011 , to be published. 

[7] A. I. Shnirelman, Ups. Mat. Nauk 29, 181 (1974); S. Zelditch, Duke Math J. 55, 919 
(1987); Y. Colin de Verdiere, Comm. Math. Phys. 102, 497 (1985); in Les Houches LII, 
Chaos and Quantum Physics, edited by M.-J. Giannoni, A. Voros, and J. Zinn- Justin 
(North-Holland, Amsterdam, 1991); B. Helffer, A. Martinez, and D. Robert, Comm. 
Math. Phys. 109, 313 (1987). 

[8] E. T. Jaynes, in The Maximum Entropy Formalism, edited by R. D. Levine and M. Tribus 
(MIT, Cambridge, 1979). 

[9] N. G. van Kampen, in Chaotic Behavior in Quantum Systems, edited by G. Casati 
(Plenum, New York, 1985); J. M. Deutsch, Phys. Rev. A 43, 2046 (1989); P. Gaspard, 
in Quantum Chaos - Quantum Measurement, edited by P. Cvitanovic, I. Percival, and 
A. Wirzba (Kluwer, Dordrecht, 1992).