Você está na página 1de 17

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

QUANTUM THERMODYNAMICS M. Hossein PARTOVI


Department of Physics, California State University, Sacramento, CA 95819, USA Received 28 February 1989; accepted for publication 6 April 1989 Communicated by J.P. Vigier

Equilibrium and nonequilibrium thermodynamics of microscopic systems are derived from quantum mechanics. An example ofa particle interacting with a heat reservoir is considered, and its rate ofapproach to equilibrium is calculated.

In a series of papers [13]concerned with a rigorous formulation of quantum measurements we have emphasized the fact that a pure quantum state is in general an idealization and that a fundamental understanding of the underlying physics must be based on a statistical treatment. Such a treatment was developed in ref. [2] on the basis of a maximum uncertainty/entropy principle rather as the statistical mechanics of large aggregates is based on some extradynamical hypothesis such as ergodicity, equal a priori probability or maximum entropy. The introduction of a statistical hypothesis into microscopic dynamics, however, immediately resurrects the century-old question of whether the extra assumption is a necessary consequence of the underlying dynamics or whether it is in fact an additional law of nature. A survey of the recent literature [4,5] indicates that in spite of a massive body of research since Boltzmanns pioneering work, the problem of the foundations of statistical mechanics and the origin of irreversibility is still basically unresolved. Thus, whereas the first and the third laws of thermodynamics are well-understood consequencesof dynamical principles, the second law is not. It is the purpose of this Letter to resolve this issue by showing that for microscopic systems, irreversibility and the second law follow from the dynamics and do not have an independent status. By so restricting our considerations to microscopic systems we are able to avoid inessential complications associated with large aggregates and at the same time underscore the fact that thermodynamic behavior is already present at the 440

quantum level and is not the exclusive domain of macroscopic systems. We shall start by introducing certain preliminary notions. The state of a quantum system is determined by a measurement/preparation process whereby a large number of copies of the system is produced, a fraction of these is subjected to interaction with measuring devices, and the information so obtained is used to define the state of the remaining copies. Since in general no finite measurement process can be exhaustive [2], the state so prepared is only incompletely specified and must therefore be represented by a density matrix, ~3, orc responding to a mixed state. The introduction of a statistical mixture directly raises the question of a measure of uncertainty, or entropy, for the ensemble. In many ways, the issue of the choice (or even the universal existence) of the entropy function lies at the heart of the irreversibility problem [5]. However, as shown in ref. [2], the choice for a quantum system is (up to a choice of scale) essentially unique, being given by the von Neumann formula tr , ln~5; e shall refer to this w quantity as the BoltzmannGibbsShannon (BGS) entropy [5]. Note that the BGS entropy is a constant of the motion for a closed system (which evolves unitarily according to a Hamiltonian), a fact which is often considered to contradict the principle of entropy increase and invalidate the BGS entropy as a universal (i.e., nonequilibrium) entropy function. For quantum systems, however, we shall consider this factas paramount and take it to imply that the origin

0375-960 1/89/$ 03.50 Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)

Volume

137, number 9

PHYSICS LETTERS A

5 June 1989

of irreversible behavior must be sought in interactions with external systems. The idea that entropy increase can only occur for open systems was emphasized in the fifties although the formulations that were based upon it have not proved conclusive [4]. The essential uniqueness, at the quantum level, of (a) the choice ofthe entropy function and (b) the origin of irreversibility is a significant aspect of the present formulation. A thermodynamic interaction for quantum sys~,

in the final state. To see this, consider the initial state ~ the final state f~,and the respective entropies S[f~J=S~~ and S[f~]=S~. Since t~~ f(~are unitarily related, S~=S~.On the and other hand, the entropies of the individual (or partial) systems, Sa and Sb, must be defined in terms of the individual density matrices Pa = trb F and Pb = tra I, where tra denotes the trace operation relative to the Hilbert space of system a. Thus S~ ab S[15~],and similarly for the initial states. Using

tems is in general a collision, i.e., a state for which the interacting systems are initially and finally out of each others interaction range. This characterization is basically the microscopic rendition of the standard definition for macroscopic systems. For a closed, or free, system we define a stationary state as one whose density matrix is constant, implying that p= i [J~, i5] =0, where Ris the Hamiltonian operator. Among the stationary states, a special class known as Gibbs states will play a distinguished role. The Gibbs states 1 exp(flui), where fland Zare are of the form(to insure the self-adjointness of real constants Z the normalization constant Z will be omitted where inessential ~t2 To lend substance to this definition, we consider an interaction between systems a and b which are initially in the Gibbs states A = exp ( flRa) and /b =exp( flAb). Since the S-matrix ~ for a collision commutes with the sum of the free Hamiltonians Ha +Hb, it follows that [S, PaPb] =0, implying that no change will result from the collision no matter what the interaction (as long as it is sufficiently regular to allowthe existence of an ~ matrix). Thus, systems in Gibbs states with the same value of /3, when allowed to interact, will emerge from the interaction with no change. It is not hard to see that in general the Gibbs states are the only ones with this property. With the preliminaries out of the way, we consider the interaction of two systems in arbitrary (but uncorrelated) initial states ~ and ~5 /). It is well~ known [51 that the sum of the BGS entropies will in general increase as a result of such an interaction, the difference arising from the correlations present
j);

these definitions, one can readily verify that


5(f)

+ 5~f) 5~) 5~)

tr f~ [ln(fr~4j3~,) ~ In

This last quantity is known to be non-negative as a result of a convexity property of the entropy function [5], and it can only be zero if ~ Pa Pb ~ i.e., if there is no correlation in the final state. Thus, in an obvious notation, L~Sa L~Sb 0. + ~ The general result just stated has an important consequence when one of the two systems, say b, is initially in a Gibbs state. With ~ x exp ( flAb), then, let us consider the quantity
~

1.\Sb fl~Ub,where Ub = tr pbHb

is (the mean value of) the energy of system b, and z~ denotes a change from the initial to the final state. We have
~5b

fl~Ub=

tr[~5~, ~3~/) In
~

In ~

fi(i~ /5~)Ab]
Using ln ~ + flHb = In (Z), one can reduce the right-hand side of the above equation to tr ~5~f) x (ln ~3~0 ln ~ The convexity property stated previously, on the other hand, shows that this last combination is non-positive, implying that Es (Sb flUb) ~ 0, with equality holding only if ~ =,3~,), .e., in general only when the two systems i are in equilibrium. It is worth remarking that the changes contemplated here are not necessarily small, and that therefore Es (Sb flUb) is not in general the change in the Helmholtz free energy ofsystem b, since there is no reason to suppose that /3~is a Gibbs state. The inequality just derived, when combined with ~a + EsSb ~ 0 established before and conservation of energy EsUa+EsUbO, implies that

~ ~2

See ref. [6]. This paper was in turn inspired by the pioneering workofBergmannandLebowitz [7]. This shows the distinguished role that the canonical ensemble plays in microscopic statistical mechanics.

1~(SaflUa)>~0.

(1)

This is an inequality that governs the interaction of 441

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

system a, initially in any state, with another (uncorrelated) system initially in a Gibbs state characterized by the parameter fl. Inequality (1) implies the zeroth and the second laws of thermodynamics. The first is obtained by considering both systems a and b to be in Gibbs states with corresponding parameters Pa and fib. Then (1) implies that Es (5a fit, Ua) ~ 0 as well as Es (Sb Pa Ub) ?~ Since EsSa + EsSb~ 0, it follows that 0. (Pa fit,) Es Ua ~ 0. Thus when two Gibbs states interact, the state with the higher value of fl~ loses energy to the other, while, as established earlier, with equal values of fi no flow of energy and no change of state occurs. This establishes the zeroth law and the existence of temperature, the latter may be taken to be fl~ as usual. To establish the second law, consider a cyclic transformation of any system a brought about by a series of interactions with an array of systems b~ which are initially in equilibrium at temperatures fl~ respectively. Then eq. (1) applies to each interaction, implying that EsSan ?~ Uan. Since for a finEs cyclic transformation EsS~ =0, we have the result that
~,,

~ finEs Uan
I

~ 0,

(2)

which is a precise statement of the Clausius pnncipie ~. Observe that the fl~are the initial temperatures of the systems with which system a interacts, so that the theorem expressed in (2) is in no way limited to processes where temperature is defined for system a, as often stated in textbooks. Thus far we have established the two laws of thermodynamics which are statistical in nature. Together with conservation ofenergy and the existence of a unique ground state (certainly true of systems with a few degrees of freedom), they constitute the laws of thermodynamics. The remaining task is to show how and why a system not in equilibrium can approach equilibrium (i.e., tend toward a Gibbs state). Approachto equilibrium is a resultof interactions, since a closed system can only evolve into unitarily
113

equivalent states, and an equilibrium state is not so equivalent to any nonequilibrium state. Clearly, the inequality expressed in (1) indicates that an interaction with a Gibbs state drives a system toward equilibrium. But the same interaction drives the other system initially in a Gibbs state out of equilibrium. To circumvent the latter difficulty, we will define an ideal heat reservoir as an infinite aggregate of microscopic systems in equilibrium at some temperature fl~.When a microscopic system in any state is allowed to interact with such a heat reservoir, it will undergo a series of interactions with the constituents of the reservoir with the probability ofa reencounter of being vanishingly small (essentially proportional to the inverse of the number of constituents in the reservoir). Clearly in each interaction eq. (1) must be satisfied, i.e., the quantity SflU must increase, where S and U are properties ofthe system while fl~ is the temperature of the reservoir. Given a sufficiently large number of interactions, the system will approach a limiting state for which SflU is as large as it can be. But this maximum property is precisely the recipe for the canonical ensemble, and the limiting state is none other than a Gibbs state with the temperature parameter equal to that of the reservoir. The argument just concluded tacitly assumes that the interactions among the constituents of the reservoir do not cause correlations to be established between the system and the as yet unencountered constituents ofthe reservoir. Stated simply, the argument essentially relies on the infinite size of the reservoir to guarantee that it will not be corrupted by a few interactions with the system. Clearly, if the constituents of the ideal reservoir are required to be sufficiently weakly interacting, the problem will not arise. It can be shown, however, that even in the case ofa strongly interacting reservoir, the correlation entropy ~ of the system relative to a constituent member of the reservoir is on the average no greater than (EsS flEsU) /N, where N is the number of the constituents in the reservoir, and EsS and EsU refer to the system as before. Clearly, for large N the correlations in question are vanishingly small. To see how these ideas work in practice, we shall
~ The difference between the sum of the individual entropies of the two systems and the entropy oftheir combination is known as their correlation entropy.

Note that in the absence ofchanges in the Hamiltonian of the system (which would give rise to exchange of work), IsU is just what we would identify asthe heat absorbed by the system.

442

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

consider an example of a system interacting with a heat reservoir in some detail. To simplify matters as much as possible (without making the example unrealistic), we shall consider a particle (e.g., a molecule) of mass ma, initially in the state ~ interacting with a reservoir composed of N particles of mass mt, in the Gibbs state /5~=Zj x exp ( fit,At,). All particles are confined to a volume V with perfectly reflecting walls, with N and V sufficiently large and N/V suitably small, and they will be taken to be without internal structure and moving nonrelativistically. The problem is to consider a collision between the system and the particles of the reservoir, and to calculate the change Es (Sfit, U) that results for the system. The initial state for the combined system is given by f(i) =/3~1~3~i), and the final state by where ~ is the S-matrix for the collision. The final state of the system is then given by ~ =trb ~(, and in general it depends on the details of the initialproceedHowever, just as in collision theory, one can states. to the (improper) limit of plane waves and still extract the desired information. In that limit, the density matrices are diagonal in the momentum representation, e.g., <pI,~~ Li> = (pp (p), is quantity tn this limitetc.A,Moreover, change of the the rate of the relevant state of the system; cf. the derivation of transition
)p~I)

p~,~(p)=N(flb/21tmb)

As mentioned above, we shall take 312exp( flbp2/2mb)

corresponding to a reservoir at temperature the system, we shall take pi~j~) (2it).2) 3/2 exp[ (pp 2/2A2]

fl~. or F

0)

and consider two limiting cases. For case I, we take Po = 0, corresponding to a system initially in equilibrium at temperature fly =22/me, and for case II, we consider the opposite limit of p~>>A2,p~>>m~/ mt,/lt,, corresponding to a system with a narrowly defined momentum Po, with p 0/m11 much larger than the rms speed of the reservoir particles. With this information substituted in eq. (3) and the integrals carried out, we finally arrive at the following expressions for the two cases I and II: ~i~i,
1I sf~t,U

2~2+2~ (392(1 +p)~)

1/2

18)3N

i7~

(4)

3p9(l+ u)(l 2(l9)2 9(9-I1u)+3u +11)2

probabilities in collision theory. Once A is obtained, the rate of change of Sfl,, U can be calculated from it as an expectation value. When the steps outlined above are carried out one finds, after some rearrangement,
~

(5) for case I (II), p= ma/ where 9(3/mbflb)I( /2flbA the rms velocity of the parm~,v= flt,/fla is tides of the reservoir, and v 0 =p0/m is the velocity of the system in case II. Moreover,
2/ma)

X (va/ti) 2N ~v0,

~(v)= jdss5exp(_s2/2)

x JdQe(l_coso)If(sv,O)12, ~=JdQo(l_cosO)If(po/(l+p),e)I2,
2p (p+q)p,~,~ [ (ab/aa)pq] where 2]2m,,ti.
=

x If(q, 0) J x{ln[p~(p+q)] +fl,,(,p+q)2/2m 5 ln[p~)(p+q )] fit,(p+q )2/2m

[~u(1u+)/3(l +p)

11}

(3)

where f is the scattering amplitude, q and q are the initial and final momenta of each 2, and aa,bthe cenparticle in = ma,b/ ter-of-mass Note that eq. (3) gives the steady-state (ma+ mt,). frame, cos 0= qq /q rate ofchange ofSAUat a moment when the states ofthe system and reservoir particles are specified by and ~ respectively,
~

Note that a and ff are basically weighted cross sections, favored by large-angle scattering as would be expected physically. Notice also that the bracketed combinations involving these cross sections in eqs. (4) and (5) are roughly of the order of the collision frequencies for the two cases. To interpret eqs. (4) and (5), it is useful to first recall the underlying physical situation, say for case
443

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

II: particles of mass m11 (the system) are prepared in the state (p) (this may be accomplished by preparing a sufficiently dilute within an enclosure that 2/magas of them in equilibrium at temperature )~ Pa / ma), and are subseis moving with velocity quently allowed to enter the reservoir and interact with it for a time Est, after which they are collected and momentum analyzed (assuming this to be possible in practice). Then the changes AS and EsU are calculated and the quantity (EsS fl,4U) /Est is cornpared to the prediction of eq. (5) for sufficiently small Est. With this picture in mind, first note that eqs. (4) and (5) have been organized as products of three factors, as in F,F 2F3. In both cases I and II, F, vanishes as p= ma/mb*O, reflecting the factthat an infinitely massive reservoir particle becomes a perfectly reflecting wall, and will not drive the system toward equilibrium. Note also that 9-+0 represents a (relatively) very cold system in case I, and a very sharply collimated system in case II, and that in both cases F~ behaves like 9, indicating a very large mitial rate of approach toward equilibrium. On the other hand 81100 represents a (relatively) very cold reservoir, wherefore F1F2 behaves like 93~2 in case I, again indicating a very large value for Sfl,, U. By contrast, as 9i~1 the rateof approach to equilibrium vanishes for case I, confirming the fact that two systerns in equilibrium at the same temperature2will not which change upon to be much larger factor (v0/IY) case II is stipulated interaction. The than unity for on the other hand shows that a system moving very fast through the reservoir will rapidly lose energy and move toward equilibrium. The third factor, F 3, depends on the systemreservoir interaction potential and has the structure of a weighted collision frequency, as remarked earlier. Remarkably, in spite of their quantum mechanical derivation, eqs. (4) and (5) do not involve Plancks constant. Finally, note that the quantity Sfit, U is manifestly positive for both equilibrium (case I) and nonequilibrium (case II) situations, in accordance with eq. (1) and the second law of thermodynamics. As already remarked, the idea that irreversible behavior is associated with openness has been extensively discussed in the literature [6,4]. Suffice it to
~~i)

point out here that eq. (1) is perfectly consistent with time reversal invariance, as are eqs. (4) and (5) althoughto extremely reversed versions would correspond their time improbable situations. One such situation would in fact occur if the system is confined to the enclosure of volume V (with perfectly reflecting walls, not real walls made of atoms) and forced to interact with the reservoir for a sufficient length of time (expected to be longer than the age of the universe for a macroscopic reservoir); the systern would then reverse its approach to equilibrium. It is clear that the practical impossibility of isolating macroscopic systems from external influence prevents such manifestations of reversibility ~. In this paper we have only dealt with microscopic systems. However, we expect no major modification in extending the present treatment to the macroscopic domain. Indeed as it may be verified, the arguments leading to eq. (1) and the Clausius primciple basically hold for systems of any size. It may also be recalled that the results given in eqs. (4) and (5) remain unchanged in the classical limit, mdidating that the second law is not a specifically quanturn mechanical result. This work was supported by the National Science Foundation under Grant No. PHY-85 13367 and by a grant from California State University, Sacramento. However, see ref. [8], and ref. [61 for a discussion of spinecho experiments.

References [11M.H. Partovi, Phys. Rev. Lett. 50


(1983)1883. [2] R. Blankenbecler and M.H. Partovi, Phys. Rev. Lett. 54 (1985) 373. [3] M.H. Partovi and R. Blankenbecler, Phys. Rev. Lett. 57 (1986) 2887, 2891. [4] 0. Penrose, Rep. Prog. Phys. 42 (1979) 1937. [5] A. Wehrl, Rev. Mod. Phys. 50 (1978) 221. [6]J.M. Blatt, Prog. Theor. Phys. 22 (1959) 745. [7] P.G. Bergmann and J.L. Lebowitz, Phys. Rev. 99 (1955) 578; J.L. Lebowitz and P.G. Bergmann, Ann. Phys. (NY) 1 (1955) 1. [8] E.L. Hahn, Phys. Rev. 80 (1950) 580.

444

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

IRREVERSIBILITY, REDUCTION, AND ENTROPY INCREASE IN QUANTUM MEASUREMENTS M. Hossein PARTOVI


Department ofPhysics, California State University, Sacramento, CA 95819, USA Received 28 February 1989; accepted for publication 6 April 1989 Communicated by J.P. Vigier

Using the strong subadditivity property of entropy, it is shown that the interaction ofthe measuring device with the environment brings aboutthe reduction process characteristic of a quantum measurement. This entails an entropy increase which has an inviolable lower limit. This limit is found and explicitly calculated for illustrative examples.

There are two long standing unresolved problems in the foundations ofphysics, the measurement problem of quantum mechanics and the reversibility problem of statistical mechanics, with notable parallels between them. Each dates back to the inception of the corresponding field, the latter to the introduction ofBoltzmanns H-theorem and the former to the advent of modern quantum theory. Both have been the subject of intense debate and activity ever since [1,2], resulting in a vast and often bewildering literature which, among other things, includes a considerable amount of qualitative, nearly metaphysical, discourse. Both problems are largely ignored by most physicists since there is in each case a more or less standard formalism which adequately accounts for the practical aspects of the problem. These are the reduction postulate for quantum mechanics and the ergodic hypothesis or the extremum principles of ensemble theory for statistical mechanics. However, these hypotheses are additional to the known laws of dynamics, serving only to explain the one problem they are dedicated to, rather like the aether of prerelativity electrodynamics. Naturally, many physicists have found this state of affairs unsatisfactory, seeking instead to find the underlying physical mechanism responsible in each case and from that a quantitative resolution in terms of the known laws of dynamics. These efforts now appear to have reached a successful conclusion on the recognition that macro-

scopic systems cannot be prevented from interaction with their environment, and that the loss of information which is at the heart of both the reduction process in quantum mechanics and the irreversible entropy increase in statistical mechanics is simply a consequence ofthis unavoidable coupling to the large number of degrees of freedom of the (unobserved) environment. The ideas of opennes and interaction with the environment are not new and occur in many writings, although only a few authors have seriously pursued the consequences. Among these, the contributions of Bergmann, Lebowitz, and Blatt in the context of statistical mechanics [3], and those ofZeh, Zurek, and Peres in the context of measurement theory [461 should be mentioned. In the preceding Letter [7] we have shown in a rigorous manner how the second law of thermodynamics, irreversibility, and approach to equilibrium for microscopic systems follow from the laws of quantum mechanics. The purpose of this paper is to present a rigorous proof for the reduction process as a consequence of interaction with the environment. Furthermore, it will be shown that a given quantum measurement necessarily entails an entropy increase which has an absolute lower bound The analysis follows methods and ideas introduced in a series of papers [10] concerned with the foundations of quantum mechanics, with entropy asthe central con~.

This was first considered by Szilard [8], see also ref. [9].

0375-960l/89/$ 03.50 Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)

445

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

cept. Together with the preceding Letter [7], the present analysis provides a unified and rigorous resolution of the irreversibility and measurement problems at the microscopic level, In the following we shall use the notation and definitions introduced in ref. [101. The state of quanturn system is in general represented by a density matrix /3 corresponding to a mixture. A measurement is a process whereby a fraction ofthe members of a similarly produced ensemble is subjected to interaction with a measuring device ~ designed to measure a physical observable A. Implicit in the construction of ~ is a partition of the spectrum of the operator A (which represents the observable A) into a (necessarily finite) number of subsets a,, i= 1, 2, N. With no loss in generality, one can take these to be disjoint intervals, or bins, with the corresponding projection operators ft, and the associated resolution of the identity >~,t, = 1. The end result of the f measurement is a set of relative frequencies, f, showing how often the value of the observable A turned up in the interval a1. These frequencies then serve to approximate the corresponding probabilities which are in turn given by the set of expectation values tr /3ft1. In addition to the quantum system and the measuringdevice there is a third system that plays a crucial role in the process of measurement, namely the environment, or that part of the rest of the universe which, without being directly observed, will unavoidably interact with the device while the measurement is in progress [46].Often the inessential parts ofthe measuring device, that is the parts which do not directly interact with the quantum system being measured, play the role of the environment. While the experimental arrangement is designed to isolate the quantum system from such spurious interactions with the environment, the measuring device cannot be so isolated on account of its large interaction cross section with the environment (resulting from its enormous number ofmicroscopic constituents); herein lies the reason why the measuring device acts as a classical system. The next step is to describe the dynamical arrangement for the measurement process. The measuring device ~ is in general an enormously complex systern on the quantum scale. However, certain aspects of this complexity are irrelevant to the measurement process and may be ignored. Thus with no loss in
,~,

generality, one can represent ~ as an assembly of N independent counters C1, one for each bin a1, so that the state of the measuring device can be represented by the product density matrix f, Now for ~ to function as a measuring device, each counter C, must possess a pair of (in general, mixed) states t~,to be referred to as the fired and quiescent states respectively, with the following properties: (a) t,-~ stable against interactions with the environare ment, (b) a quantum system with its value for the observable A known to be in the interval a, will, upon interaction with the counter C1, cause the latter to evolve from to i.e., will cause it to fire, and (c) the states f,~are macroscopically distinguishable, i.e., stable against, and resolvable by means of, macroscopic observation processes (typically involving an exchange of a large number of quanta). The last condition in particular implies the orthogonality of the pair of counter states, namely the requirement that tr f~= 0, which is a far weaker condition than macroscopic distinguishability. Condition (a), on the other hand, simply states that two (mixed) states ofthe counter are in equilibrium with the environment and will not undergo a macroscopic change of state as a consequence of interaction with it. Condition (b) simply characterizes the C as counters that detect and fire exactly as required. While the measuring device is in interaction with the environment throughout the measurement process, it is convenient to consider (following von Neumann) two distinct stages. During the first of these the quantum system interacts with the measuring device, with the environment as a spectator Then in the second stage all systemdevice interactions cease while the device is once again subject to interaction with the environment. The first stage establishes systemdevice correlations that serve to yield the desired information about the system. An interaction Hamiltonian that can accomplish this is ~, iv, j2~,where V~operates on the Hilbert space of counter C only. Thus the systemdevice Hamiltonian appears as

r=

~,

~.

r1 r~,

~.

i2

This can be realized by spatially separating the subensembles corresponding to different binsand directing them toward the corresponding counter. Examples are the SternGerlach cxperiment, the momentumanalyzer and scattering experiments. This simplifying assumption does not change the physics in any significant way.

446

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

H= R~ ~ +

0, + ~ iv, ~

of correlations. Intuitively, it is reasonable that the deviceenvironment interaction will tend to degrade the systemdevice correlations present in ~( T) 56 However, to establish this result in general, we need to use a highly nontrivial property ofentropy known as strong subadditivity. Consider0) =/3AB (0 )/3c(0). The C, initially in the 3ABC( three systems A, B, and notation implies state / that systems A and B are initially correlated while neither is initially correlated with C. Starting at t= 0, systems B and C interact for a time t, with A as spectator, culminating in the state PABC(t). The partial (or reduced) state I3AB ( t) is defined as usual by /3AB ( t) =trc /3ABC( t), where the trace is taken with respect to the Hilbert space of system C; other partial states are defined analogously. Next consider the (BoltzmannGibbsShannon) entropy S, defined for any state /3 by the von Neumann formula S= tr /3 ln /3. Now strong subadditivity ~ states that
SABC +58 ~SAB +SBc,

where 0. and 0, are the self-Hamiltonians of the systern and the counter C, respectively, with the commutator [ft1,0~J vanishing
~.

The evolution operator which results from (1) has the following form in the interaction picture: Ut ( t) = ~ ft~ (1) , i2J (2)
i

where J (t) = exp [ i (R~ t~)] is the (interaction + t picture) evolution operator that acts on counter c and, for sufficiently large t, say t ~ T, causes it to fire. Thus z2J ( T)T~-~ ( T) = [7. It is also necessarythat i J ( T)tT be traceless and orthogonal to t;, i.e., that tr J ( T)t; = tr t~ ( T)t~= 0 ~ ~ The dynamics of the first stage can now be described. The quantum system is initially in the state

/3 while the counters are all in the quiescent states t~ the combined state is then described by ~= /3H,t~.Under the action of (I), ~ evolves to
~(T)=U(T)~U~(T) the end of the first stage. by Using (2), one arrives at ~(T) = fl ft 1j3ft,t,~ fJ +~ ft ii
1,1

(4)

where SABC= tr/3ABc lfl/3ABc, SB= tr /3B ln /58. etc. Note that strong subadditivity implies subadditivity (i.e., the inequality SAB~SA+SB) for a pair of systems A and B. Moreover, subadditivity makes it pos5AB Note that CAB ~ CAB as the difsible to define the correlation entropy0, with equality ference SA+SB holding only if/5AB=jSA/SB, i.e., only if the two systems are uncorrelated. Consider applying (4) to /3ABC (t). The result is SABc(t)+SB(t)~SAB(t)+SBc(t). it follows that Since /SABc(t) is 5ABC (0), unitarily= SABc(0). to / the other hand, because related On SABc ( t) /3ABC (0) =/3AB (0 )/3~ (0), it follows that SABC (0) = SAB (0) + Sc(0). Furthermore, the factthat system A is a spectator while B and C interact implies the relations SA (t) = SA (0) and SBC (t) = SBC (0). The lack of correlations between systems B and C at time t=0, on the other hand, implies that SBC( 0) =
56

t7
k~,.J

1i3ft~j2~(T)t1trj~t(T) ~
k

(3)

The desired correlations are evident in the first (diagonal) sum ofeq. (3), while the second (off-diagonal) sum consists of correlations which mustobstripped away before the final results can be be tamed by means of macroscopic observations carried out on the measuring device. This selective removal of unwanted correlations, or reduction of the systemdevice density matrix, constitutes the second stage of measurement and is the essence of the measurement problem. To show that the deviceenvironment interactions bring about this reduction, we must first establish a general result on the decay
~ The condition [*, 1?] = 0 ensures that the measurement does not change the value ofthe observableA. Pauli referred to this kind of measurement as thefirst kind; see ref. [11]. ~ This condition together with the previously stated ones are generalizations ofthose commonly assumed for the idealized case of pure system and device states, and in fact they are weaker than the latter (see ref. [1]).

Model calculations confirm this expectation and provide an

estimate ofthe time scale involved; see ref. [9]. ~ According to Wehrl [2], D. Robinson and D. Ruelle (1967) first recognized the importance of strong subadditivity and proved it for the classical case. 0. Lanford and D. Robinson (1968) then conjectured it for the quantum case. After some partially successful attempts, the conjecture was finally proven by E. Lieb and M. Ruskai (1973); see ref. [2] for references to the original papers.

447

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

SB(O) +Sc(O). Combining all these relations, one arrives at the important inequality CAB(t) vi CAB (0). Thus strong subadditivity implies, among other things, that the correlation entropy of a pair of systems in general decreases when one of them interacts with a third system (which it is initially uncorrelated with either one of the pair) ~ Furthermore, if this process is repeated a large number of times (each time with a new, uncorrelated system), the pair will be driven toward a state of minimum correlation entropy consistent with the prevailing conditions. This establishes the intuitively reasonable result concerning the decay of correlations. The conditions just stated are precisely those of the quantum system (A), the device (B) and the environment (C) during the second stage of measurement as the device is subjected to a large number of interactions with various uncorrelated constituents of the environment. Thus the final systemdevice state QR evolves from ~( ) under the conditions that T (i) the counter states t~ do not change (the stability condition (a) stated above) and (ii) the systerndevice correlation entropy is a minimum. Referring to eq. (3), we note that the first (diagonal) sum will not undergo any change during the reduction stage owing to the stability of the counter states t~.The second (off-diagonal) sum, on the other hand, involves the combinations ~(T)t~(and its adjoint) which are traceless and orthogonal to the stable states. Under these conditions it follows (see below) that the minimum correlation entropy state corresponds to the vanishing of the second sum in eq. (3). In other words, the second stage reduces ~(T) to
j,~ i
= ~
I

~R

+~ I,]

where 1~,, which only involves the counter variables, evolves and changes under the action of the environment. By contrast, tr f,~ remains a constant, since it is an invariant quantity under the unitary evolution of the deviceenvironment complex. This trace can therefore be calculated from eq. (3), at t = T, whereupon it is found to be equal to zero. On the other hand, since tr ft,,5*~ c5,) tr iv1/3 (no sum on i) = while f,~ nonvanishing only when i ~j, it follows is that the trace of the second contribution to ~ with respect to the system variables also equals zero. This result together with the vanishing ofsame as those of shows that the two partial traces of ~are the tr f,~ ~. Thus they are found to be equal to
/3 =trD ~ = ~

and
~,,. I

P = tr5 ~ = ~ ~

fl

tj,

where, it will be recalled, ~ = tr ft1/3. Note that j5 and P stay fixed during the second stage. With the partial traces thus fixed, the systemdevice correlation entropy C~D now be calculated; can

CSD=s(/3~)+s(r)s(S~).

(6)

Clearly, CSD is minimized when S(~) is maximized, since the other two entropies in eq. (6) stay fixed during the second stage. The question then is what choice of f,~ maximizes S(~ To answer this )? question, we can use a well-known property of the BGS entropy: If & and ~ are density matrices, then

ft1/3ftit,~ fl

[j.

(5)

tr & ln vi tr & in

~.

To establish the result just stated, we first observe that the most general form of the systemdevice state during the second state is

Applying this inequality to ~ and ~R, we find tr~ln~vitr~n~R=tr~RlnS~R, (7) l where the last equality follows from the fact that tr ( ~R) in ~R vanishes (owing to the fact that ~R is diagonal with respect to the ~t, whereas ~ is off-diagonal). Inequality (7) now shows that the condition f~= 0, or ~= ~R, is the one that maximizes S(~). Thus the final systemdevice state is described by ~R; the environment has removed the un~ ~

68

This intuitively obvious result is a highly nontrivial theorem that follows from strong subadditivity. It can be generalized to the smmetric case where system A interacts with a (fourth) system D while Binteracts with C. Then CAB(t) ~ CAB(O)turn 5ABCO +SA +SB ~ SAB + SBc+ SAD, which in follows from a repeated application of strong subadditivity. follows from

448

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

wanted, off-diagonal correlations. This completes the proof of the reduction process (or the collapse) during the second stage of measurement. The subensemble consisting of the quantum systems that cause the counter C, to fire is described by /3, = ftI/3ftI /tr ftj5ft 1. Since tr ft11Sit, = one can write
.~,

To illustrate the entropy increase associated with reduction, we shall consider two examples. The first is the measurement of a spin-i/2 system initially in the state p= (1 +~p)/2by means of a SternGerlach apparatus set up in the direction n; & is the vector formed of Pauli matrices andp is the polarization vectorofthespinsystem;p=~pIviland lnI=1.The corresponding spin-up and spin-down projection operators are = (1 ~n)/2. Clearly, this is an example of a two-bin measurement. The entropy increase can now be calculated from eq. (8); 2 i~S=ln 1p 2 1 2 1 p,, 1 + ln ~ii) ~~ ] (9)

c-~ ~[(i)
it

where

describes the state of the device with counter C 1 fired and all others quiescent. Clearly, a macroscopic measurement (or reading) of the counters, repeated a sufficiently large number oftimes, will yield the set of probabilities The systemdevice evolution, ~r~.~( T) ~, is unitary during the first stage and (in general) nonunitary during the second stage. The entropy change for the process, which occurs during the second stage, is given by zS.S=S(4~) S(~).Using the fact that .31,1csc,1j51 (no sum on i), we find
,~.

____

S(~R)= ~ tr ~p[~ln

ei~p,t~~.

This expression in turn simplifies to


S(~R) =S(t)

where p,, =p n. Since I p~ vi I,I vi 1, it is evident that ~ 0, with equality holding only when p =p,,, i.e., only if the system is initially polarized in the direction is (including the trivial case of zero polarization). In such an event there are no off-diagonal elements to be removed during the reduction stage and i~.S=0 as can be seen from eq. (9). As a second example consider a momentum mea. surement (in one-dimension) by means of an analyzer whose channels (or bins) cover uniform, nonoverlapping intervals of width z\p centered at p~. hus T the operator
ir1

tr M/3~ln ~./3.

projects onto the interval

with S(t) equal to the initial entropy of the measuring device. The entropy change (per event measured) is then found to be ~S=~ tr ~/5, ln ~/31+tr/5ln/3. (8)

Let the system to be measured be in a Gaussian state 10, ,u><0, ,uI, with (P) KPI~~ ~>_(/~)_I/2exp(p2/2j~2) Let ,(p) denote the restriction of 0(p) to the interval a,. Then I d I~( ) 2
.~

A reasoning similar to the one leading to eq. (7) will show that zS.S~ reflecting the fact that the en0, vironment removes information from the system device complex during the reduction process. On the other hand, E~.S= is possible only when /5 is diag0 onal with respect to the ft1, i.e., when ft1j5ft~=0for i#1. This is so precisely because in this case there are no off-diagonal elements to be removed. It is important to realize that E~S an inviolable lower limit is which is far exceeded in actual experiments, since counters and other equipment used in such experiments are far from reversible,

and

~
The entropy increase (per event analyzed) is then found to be Li.S=

~ in

~,

(10)

a quantity which was named measurement entropy


449

Volume 137, number 9

PHYSICS LETTERS A

5 June 1989

in ref. [10]. Note that in this example the initial state is pure (an idealization) and has zero entropy. What are the limiting values of LsS for ji << ~p and it>> Ap? In the former case, nearly all systems will turn up in one of the channels, say channel k, so that 1, ~ 0 for i ~ k, and 4S~0, exactly as expected. In the latter case, one has ~p ( ji~/~) exp ( so that
~2/~2)

~(3+ln n)+ln~u/~p), Cu>> i\p).

(11)

the measurement. This three-way arrangement thus makes it essentially impossible for those systemdevice correlations that are off-diagonal with respect to the stable states to exist as macroscopically observable states. It is useful to recall here that this selectivity in the states of a system in interaction with an environment is a rather commonplace fact in the context ofthermodynamics and statistical mechanics. As stated at the outset, the measurement and the irreversibility problems are intimately related and can be understood in terms of known, fundamental dynamics. We believe the present work together with the preceding Letter provide a rigorous resolution of these problems at a fundamental, microscopic level. This work was supported by the National Science Foundation under Grant No. PHY-85 13367 and by a grant from California State University, Sacramento.

Thus the minimum entropy increase necessary for collapsing the wavepacket in this case grows indefinitely as the resolution of the momentum analyzer is increased. This is of course exactly as expected, and clearly points to the impossibility of producing pure states by means of a finite preparation procedure. Although we have been careful to preserve the features essential to the reduction process in the description of the measurement process, we have nevertheless introduced idealizations in describing the measuring device. For example, the stability conditions on imply, unrealistically, that the two counter states are in perfect equilibrium with the environment. In practice, one has a range of (mixed) states close to t,~ which are macroscopically equivalent and among which the counter state will fluctuate under the influence of the environment. Under typical conditions, the associated entropy fluctuations are far greater than the quantity calculated in eq. (8) owing to the macroscopic nature of the measuring device. This observation again shows how difficult it would be to realize the theoretical minimum E~.Sin actual measurements. Why is it necessary to use a macroscopic device to measure a microscopic system? A macroscopic systern has a large number of microscopic constituents and a large interaction cross section with its environment. This renders many of its possible states extremely improbable, hence unstable. Its stable states, on the other hand, provide the pointer basis ~ for

References
[1] J.A. Wheeler and W.H. Zurek, eds., Quantum theory and measurement (Princeton Univ. Press, Princeton, 1983); M. Jammer, The philosophy of quantum mechanics (Wiley, New York, 1977). [210. Penrose, Rep. Prog. Phys. (1978) 221. A. Wehrl, Rev. Mod. Phys. 5042 (1979) 1937; [3] P.G. Bergmann and J.L. Lebowitz, Phys. Rev. 99 (1955) and P.G. Bergmann, Ann. Phys. (NY) I (1955) 1; J.M. Blatt, Prog. Theor. Phys. 22 (1959) 745. [4] H.D. Zeh, in: Foundations of quantum mechanics, ed. B. dEspagnat (Academic Press, New York, 1971). [5] W.H. Zurek, Phys. Rev. D 26 (1982) 1862. [6] [7] [8] [9] A. Peres, Am. J. Phys. 54 (1986) 688. M.H. Partovi, Phys. Lett. A 137 (1989) 440. L. Szilard, Z. Phys. 53 (1929) 840. W.H. Zurek, in: Quantum optics, experimental gravitation

r,~

578; J.L. Lebowitz

and measurement theory, eds. P. Meystre and MO. Scully (Plenum, New York, 1983). [10] M.H. Partovi, Phys. Rev. Lett. 50 (1983)1883; R. Blankenbecler and M.H. Partovi, Phys. Rev. Lett. 54 (1985) 373. [11] S. Fl{igge, ed., Handbuch der Physik, Vol. V (Springer, Berlin, 1958) p. 73.

~ Zurek has advanced andemphasized this point ofview,

[5,9].

450

Volume 151, number 8

PHYSICS LETTERS A

24 December 1990

Entropic formulation of chaos for quantum dynamics


M. Hossein Partovi
Department of Physicsand Astronomy, California State University, Sacramento, CA 95819, USA Received 20 August 1990; revised manuscript received 10 October 1990; accepted for publication 19 October 1990 Communicated by A.R. Bishop

A general formulation of quantum chaos based on the asymptotic growth rate of measurement entropy is presented and shown to be closely related to the classical formulation in terms of Lyapunov characteristic exponents and the KolmogorovSinai invariant. Examples are used to illustrate the formalism.

Despite extensive activity in recent years on the existence or meaning of chaos for quantum systems and a large number of useful results ~, there is as yet no general formulation of the meaning of chaos for quantum dynamics, nor any indication of its existence in an actual physical system. Indeed there are good reasons to believe that the most common quantum dynamical models are essentially nonchaotic [2]. Nevertheless, the objective of formulating a universal definition of quantum chaos similar to those available for classical dynamical systems is an important problem deserving of serious attention, The purpose ofthis Letter is to present a formulation of quantum chaos using the concept of measurement entropy, a notion which is equally applicable to classical and quantum dynamical systems. The physical idea underlying the present approach originates in the realization that there is a natural and powerful measure of information in quantum mechanics which embodies virtually every statistical aspect of the theory *2 This is the above-mentioned measurement entropy, a quantity which provides an information-theoretic measure of the degree of scatter or indeterminacy in the measured values of a physical observable. Now for classical dynamical systems, the most characteristic feature of chaotic
~ Numerous papers in this and otherjournals have appeared in the recent past; for useful sources to the literature see ref. [1]. ~ The development of the concept and its applications can be found in ref. [3]. 0375-9601/90/s 03.50 1990

behavior is sensitivity to initial conditions, a property that for bounded systems gives rise to a strong mixing behavior. Accordingly, for both quantum and classical systems, one should expect chaotic behavior to be associated with a steady increase in the degree of indeterminacy in the measured values of the physical observables. We shall see below that this is indeed the case, and that in general the asymptotic rate at which measurement entropy increases is directly related to the characteristic exponents associated with the dynamical system. In particular, we shall find that in the case of classical dynamics, the present approach parallels the ergodic formulation of chaos in terms of Lyapunov characteristic exponents. For quantum dynamics, we shall define chaotic behavior to signify a positive value for the above asymptotic rate, and quantify the degree of stochasticity (or chaoticity) for such systems by means of an index which is the quantum version of the Kolmogorov Sinai (KS) invariant. Illustrative toy models will then be used to demonstrate various aspects of the formalism, including the calculation of characteristic indices, and to gain insight into the question of chaotic behavior for an actual quantum system. A general quantumclassical correspondence showing the (formal) reduction, in the classical limit, of the quantum indices to the corresponding classical ones will be presented to demonstrate the correct classical hmtt of the formalism. To start, we shall consider the general case of a measurement carried out on a quantum system, and
.

Elsevier Science Publishers B.V. (North-Holland)

389

Volume 151. number 8

PHYSICS LETTERS A

24 December 1990

define the corresponding measurement entropy. In general, the state of a quantum system is described by a density matrix, denoted by /5, and the results of the measurement of some observable A (a self-adjoint operator) by a set of probabilities, each corresponding to a group of possible values of A. Let the experimental arrangement used to measure A be referred to as the measuring device and be denoted by D. Then in general D4 entails a partitioning of the spectrum of A into a collection of (non-overlapping) subsets ~ct/ }, which we shall call bins, and a corresponding orthogonal decomposition of the underlying Hubert space given by {~ where ~ repre}, sents the projection operator corresponding to the bin a~.Clearly, we have u = spectrum of A ~,,it,1 and ~=i. The results of the measurement process are summarized in a set of probabilities {.~},where ~ is the probability that the outcome of the measurement turned up in bin a~. According to the rules of quantum mechanics, ~A =tr ,4I5. where tr denotes the trace operation. Among the various measures of information, the quantity
-

gressively refined by reducing the size of each bin toward zero. Assuming the existence of the limit, we arrive at ,~, a quantity that depends on the complex {H, /5, A} but not on the particular device used to measure A. Note the obvious but significant fact that if A, or /5, commutes with I-?, then A~vanishes (barring an explicit time dependence in A). Having carefully defined A~ means of a limiting by procedure, we now revert to a more compact (but less rigorous) notation in terms of infinite partitions and continuous distributions to facilitate the presentation. Thus in place of ~ we introduce the probability density ~ (a) which is related to /5 by means of ~ ~ (a) da = tr [ ftA (da)~3()], where ft4 (da) is the t projection operator corresponding to the bin ct~ (a. a+da]. In case the spectrum of A is sufficiently smooth, one can also write ~ (a) =tr(aA)/5(t), and in place of eq. (2) obtain

,~ =

~J
-~-

da ~3,4(a) n ~~a) I

(2)

S(p ID4) =

;?~ In ;~

~ Y,

(1)

named measurement entropy, plays a fundamental role [3]. It is in terms of this quantity that we shall define the characteristic indices. Note that S is a positive number, vanishing only if all but one of the ~A } vanish, and saturating at the value ln(number of bins) if all ~ are equal. Consider next a series of measurements carried out by DA at times 1,,, with measurement results ~ (t,,), and define the index
=

It is important to realize that the passage to the continuous form (2) will in general lead to a divergent result if carried out for the entropy S itself. We can now define quantum chaos: A bounded quantum systern ~H,/5} is chaotic if it possesses an observable with a positive index. To establish the basis for this definition and show the afore-mentioned relation to the characteristic cxponents for classical mechanics, we consider a general dynamical system whose evolution is given by a set of M first-order differential equations ~ = F~ ~ ( 1), where the vector ~= (~) represents the dynamical variables. Let the trajectories of this system be

lim
7-S(p~

lD~),

(2)

where /5,, =/5(t,,) represents the state ofthe system at time t,,. As usual, /5( t) = U( I )/5 U~t), where the evo( lutionoperator Uisdefinedby (i/t)U(t)_fiU(t); i? is the Hamiltonian operator for the system (11=1 throughout). Before proceeding to the interpretation of the index, let us note that the effect of refining the partition {a~}is, in general, to increase S(/5j~), hence to increase the value of the index in eq. (2). It is therefore advantageous to consider the supremum of the limit in eq. (2) as the partition {a~ is pro} 390

given by ~1(t)=R~(1I~0), corresponding to the initial conditions ~,(t0) We will assume the R1 to be differentiable and invertible, so that the equality ~=~j at time I implies ~= ~ at the earlier time 10 (a valid assumption for most systems of interest). Next we consider the probability density function Q(~, t) representing a distribution in the phase space of the system. The evolution of Q is given by
=~,.

Q(~,) t

d~0

M)(

R(tI~))Q(~0,

t0),

an obvious relation. From the full distribution function Q, one can derive probability distributions for

Volume 151, number 8

PHYSICS LETTERS A

24 December 1990

various dynamical variables. For example, the probability density function ~?1~(a) for the dynamical variable A(~,t) is given by

therefore expect that in the limit t~oo,the contribution of the logarithm to A~is equal to ~~~Q.maxt, so that A~ AQmax. In other words, the asymptotic growth rate ofthe measurement entropy ofa dynamical variable is in general equal to the largest characteristic exponent associated with the phase-space region that supports Q (under certain conditions, characteristic exponents can be associated with indecomposable regions in phase space; see ref. [4]). Higher-dimensional indices are similarly defined in terms of the joint measurement entropy of a set of dynamical variables. Thus associated with n independent variablesA, B,..., Z, n~<M, e have ajoint w probability distribution ~,B Z (a, b z) and the corresponding n-dimensional index ,~B Z Note that then independent variables form a subspace of dimension n in the M-dimensional t~-space. ollowF ing the reasoningfor the one-dimensional index, one can show that an n-dimensional index will in general converge to the sum of the n largest characteristic exponents. (As noted above, exceptional choices of the dynamical variables will invalidate the general rule. For example, for a system whose total energy H is conserved, ~~=0 and =~, where A is any dynamical variable.) Clearly, the largest among the higher-dimensional indices will equal the sum ofthe positive characteristic indices, a quantity that (under suitable conditions) equals the KS invariant [4]. For a Hamiltonian system of N degrees of freedom, M=2N, and the characteristic exponents are symmetric with respect to zero. In that case, the KS invariant will in general equal the joint index corresponding to a set of N dynamical variables (barring the exceptions noted above), e.g., the N position variables. For quantum systems, we have defined chaos to mean the existence of a positive index ,%~.We can also consider higher-dimensional indices corresponding to the simultaneous measurement entropy of sets of compatible observables. Since quantum systems are Hamiltonian in structure, one could define the spectrum of characteristic exponents in analogy with those stated above for classical systems. Note that the quantum mechanical prohibition against the joint measurement of incompatible observables is not a problem here (since, e.g., only the set of position operators need be measured simultaneously). While the usefulness of such a definition
~

dM~Q(~ t)

(aA (~ 1))

The index ~ is now defined from ~ (a) according to eq. (2), and can be expressed in terms of the trajectories by using the evolution equation for Q. After a few straightforward steps, one arrives at
=

urn

-~-

Q(i~, t~)ln(

dM

,~o Q(~,t~)

Xo(A[R(II~), t] A[R(t~

0), t])).

The condition imposed by the delta function defines an (M 1)-dimensional surface in the p0-space, and it can be used to reduce and reparametrize the integral over that surface. The result is

dM_I uJQ [~(

u, ~

I), t~] ,

where the Jacobian determinant is given by u1 au2 UMI J=~LIjk,~~-... ~


.

Here L,= (9A/9R~)D,~, is the matrix of partialdeD,, rivatives [(a/~01)R1(t~~j0)],and ~ is the totally antisymmetric symbol in M dimensions (repeated indices are summed in these expressions). At this point we recall that the matrix of partial derivatives D is precisely the object whose asymptotic behavior contains the relevant information on the characteristic exponents. In effect, the asymptotic behavior of D is dominated by exponential growth, with the exponents converging to characteristic values which are independent of Jci under suitable conditions *3 Therefore it follows that as t4o0, ln I DO ~ = largest characteristic exponent. Here 0 is a vector in M dimensions (as is well known, if 0 happens to be orthogonal to the eigenvector associated with AQ,max, the above limit converges to the next largest characteristic value). We
~ A detailed treatment, including the mathematical conditions under which various quantities exist, can be found in ref.

141.

391

Volume 151. number 8

PHYSICS LETTERS A

24 December 1990 I) =Q(q

must await detailed information on the existence and nature of the spectrum of characteristic exponents for quantum dynamical systems, we are in a position to define a global measure of chaos for quantum systerns that is the analog of the KS invariant of classical dynamics: For a bounded quantum system, the stochasticity index is equal to the supremum ofthe set ofindices {)~A.B Z }~ system is chaotic if ii has aposA itive stochasticity index, Before proceeding to examples, we will outline the derivation of a quantumclassical correspondence in the h~0limit between the quantum and classical indices defined above. It is important to realize that this can only be a formal correspondence, relying as it does on the interchangeability of the t~cxlimit in eq. (2) on the one hand, and the classical limit 15 0 on the other. Although there are good reasons to believe that the two limits in general fail to commute, it is still important to establish the correspondence for those cases where they do commute. Now what we wish to show is the reduction ,~~ in the classical of relating/5 and Q for aagiven system, and (b) limit. This requires (a) physically meaningful way the demonstration that the time evolution of /5 reduces to that of Q in the classical limit. We shall deem /5 and Q to be equivalent if they give rise to the same probability distributions (excluding, of course, the joint distributions involving incompatible observables) for the canonical (i.e., position and momenturn) observables. This requirement leads to the identification of Q with the Wigner distribution introduced a long time ago [5]. To establish (b), one starts with the evolution equation for/5, derives from it the evolution equations for the Wigner distribution, and takes the classical limit to verify that the latter in fact reduces to the classical law of evolution. The main quantum mechanical object in these equations is the evolution operator U introduced earlier. The essential ingredient in the derivation is the resuIt that as h0,
-~

0, Po 0) together with the condition that the phase point (q0, Po) at time zero evolves to (q, p) at time t. This statement, by way of Liouvilles theorem, is equivalent to time evolution according to classical dynamics. In recent years much has been learned about the spectral characteristics of quantum systems that are known to be chaotic classically, and examples have been constructed of open systems that obey nonHarniltonian evolution and exhibit chaotic excursions in some dynamical variable [11. However, as mentioned at the outset, there is as yet no known cxample of a realistic quantum system that exhibits chaos in an unambiguous way. Consequently, the examples considered below are only illustrative models, although they are Hamiltonian systems and they do illustrate the workings of the formalism. The first example will test whether the formalism works for linear systems, and whether it correctly produces a positive characteristic index at a point of linear instability. A convenient example is the har2/2m+k.*2/2. For monic oscillator, pure state ~i> <~I. /5, we choose the defined by ii=15 where ii(x, 0) is a normalized Gaussian function of width a/..J~. t is simple to verify that I = Ii~f(x, ) 2, and t not difficult to show that the latter equals iii(bx) 1/2 exp(a2x2/b), where b is the quantity (~2 C05 WI)2 + [sin (WI) / mw] 2, A little calculation then gives
. ~

statement that Q(q, p,

bo
2t n

,~

as expected of a periodic or quasiperiodic system. Now so far we have assumed k> 0, which corresponds to x = 0 being a stable equilibrium point (k = 0 describes a free particle, also with a vanishing index). The more interesting case, however, is that of k<0, corresponding to an unstable point at x=0. In that case we let w=iy, where y= (k/rn)2, and find that lnb.In[exp(2yt)], and from this, ~=y. Thus we find a positive index in case ofa linear instability, exactly as expected. The positive index notwithstanding, this system is not chaotic (it is not bounded) and does not even define a proper Hamiltonian operator. The second example involves a nonlinear system and exhibits certain attributes of a chaotic system.

U(q, q, 1) *1

(i/2it) (2/q 18q~)S(q, , t)II~~2 q

x exp [iS( q, q, t)], where S is the action function evaluated along the classical path from q at time 0 to q at time t [6]. After some manipulation (including an integration by the stationary-phase method), there emerges the

392

Volume 151, number 8

PHYSICS LETTERS A

24 December 1990

To construct it, we start with the differential equation 0 = w sin 0 it ~ ~ ~ it, where the endpoints it are identified and w is a positive constant. The solution is

O(t)=2arctan[tan(0/2)exp(wt)]

...). Once again one can use eq. (2) to calculate 2~ some choice of ~3; he result is w for both infor t dices, just as in the classical case. If this were an acceptable quantum system, we would identify w as the stochasticity index. However, this system is unbounded in momentum, both classically and quanturn mechanically, and cannot properly be considered a chaotic system. Nonetheless, it is an example of a spatially bounded quantum system that exhibits permanent diffusion in momentum space, albeit a diffusion that is accompanied by unbounded growth. Note also that the spectrum of I? consists of the entire set of real numbers, continuous and unbounded from below, a fact that explains the apparent contradictions with the quasiperiodicity results found by Hogg and Huberman [2]. The equality of the classical and quantum indices in the above models is the exception rather than the rule, and can be traced to the fact that the semiclassical approximation to the corresponding path integrals is actually an exact result for these cases. Indeed as emphasized above, the formal equality in the h~0limit will fail in general, since the various limiting processes involved may not commute. Thus a classically chaotic system may well produce a zero stochasticity index when quantized. The formulation presented in this paper relies on the structural similarity ofclassical and quantum dynamics, and on the key role measurement entropy plays in all statistical aspects of dynamical measurements. It provides a general and rigorous criterion for quantum chaos, one that applies equally naturally to classical and quantum dynamics. The entropic formulation thus standard classical definition. implementation of the produces a faithful quantum The reason for this strict adherence to the standard definition is that the concepts of randomness and unpredictability that underlie the notion of chaos concern the measured values of the observables of a system, not the particular dynamics that produces them. Whether this standard notion of chaos is of any use or relevance when dealing with quantum phenomena is a separate and as yet unanswered question. As already noted, most common quantum dynamical models are quasiperiodic and therefore nonchaotic, and it appears likely that a continuous energy spectrum is a necessary condition for a nonzero stochasticity index. Time-dependent Hamilto393

where O~=(0).It has two fixed points, the first, at 0=0, is unstable, and the second, at 0= it, is stable. The stable point has a basin of attraction which consists of the entire 0-space minus 0=0. Next we develop a prescription for quantizing the M-dimensional classical dynamical system ~=F(~, t) considered earlier. To each ~, we assign a conjugate momentum y, and define the Hamiltonian by H( ~, y, t) = ~,LL1F,. equations of motion for ~ reThe produce the original dynamical system, while those for y, constitute an additional set of M first-order equations not present originally. The resulting Hamiltonian system can now be quantized canonically (subject to the well-known hazards of the ordering problem). Following the procedure just outlined, we first arrive at the classical Hamiltonian H(, p)=Wpsin, where p is the momentum conjugate to 0. Using the constancy of H, we can then verify that p ( t) = p0sin 00/sin 0(t). The new fixed points are (0, 0) and (it, 0), both unstable; the rest of the phase space is attracted to ( it, cx), so that for almost all initial conditions, p(t) eventually tends to cx. At this point one can calculate the indices ~~P; the result is that for any Q(0, Po, 0) which is well-behaved at 0~=0,both indices are equal to w. Thus w is the largest characteristic exponent, a fact that can be verified by 0(Oo, Po) the matrix of partial denymeans of mentioned earlier, atives 8(0, p)/ version of the above system is an The quantized unusual rotor defined by the Hamiltonian ~w(p sin ~+ sin ~j3), with [0, jS] = i (jS is the usual angular momentum operator). The evolution operator U= exp ( i/it) can be calculated explicitly for this example; it is given by <01 U(t) I~> = C~~[O()], here C=(coshwt+cossinhWt)12 w and 0(0) is the angle 2 arctan [tan (0/2) exp ( Wt)]. These equations show that the 0-space wave function evolves by steadily piling up at 0= while it thinning everywhere else, and correspondingly the momentum wavefunction continually spreads out over the spectrum ofjS (which consists of 0, 1, 2,

Volume 151, number 8

PHYSICS LETTERS A

24 December 1990

nians and unorthodox models such as the one introduced in this paper appear to be fertile searching grounds for chaotic behavior, although it is already clear that chaos is not the omnipresent phenomenon in quantum dynamics that it has proved to be in nonlinear classical systems. Be that as it may, resolving the issue of the existence of physically sensible quantum models with positive stochasticity indices appears to be the main challenge now. References
[I

boss, R.H.G. Helleman and R. Stora, eds., Chaotic behavior in deterministic systems (North-Holland, Amsterdam, 1983); B. Eckhardt, Phys. Rep. 163 (1988) 205. [2] T. Hogg and B.A. Huberman, Phys. Rev. A 28 (1983) 22; MV. Berry, Proc. R. Soc.A 413 (1987) 183. [3] M.H. Partovi, Phys. Rev. Lett. 50 (1983) 1883; Phys. Lett. A 137 (1989) 445; R. Blankenbecler and M.H. Partovi, Phys. Rev. Lett. 54 (1985) 373. 141J.P. Eckmann and D. Ruelle, Rev. Mod. Phys. 57 (1985) 617. [51 E. Wigner, Phys. Rev. 40 (1932) 749; K. Gottfried, Quantum mechanics (Benjamin. New York, 1966) p. 440. [61 L.S. Schulman, Techniques and applications of path integration (Wiley, New York, 1981).

G.

] G. Casat, and J. Ford, eds., Lecture notes in physics, Vol. 93.


Stochastic behavior in classical and quantum Hamiltonian systems (Springer, Berlin, 1979);

394

Você também pode gostar