Scribd Logo
Upload

Welcome to Scribd!

  • Lebowitz-Penrose - Modern Ergodic Theory - PT1973 PDF

    Uploaded by

    concebays
    19 views7 pages

    Original Title

    Lebowitz-Penrose - Modern Ergodic Theory - PT1973.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    19 views7 pages

    Lebowitz-Penrose - Modern Ergodic Theory - PT1973 PDF

    Uploaded by

    concebays

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 7
    Modern ergodic theory There is much more to the mathematical study of Gibbs ensembles than the question of whether or not time averages and ensemble averages are equal Joel L. Lebowitz and Oliver Penrose "The founding fathers of statistical me- chants, Boltzmann, Maxwell, Gibbs fend Einstein, invented the concept of ensembles to'deseribe equilibrium and nonequilibrium macroscopic systems. In tying to justify the use of ensem- Dies, and to determine whether the ensembles evolved as expected from ronequilibrium to equilibrium, they introduced further concepts such as “ergodicity” and “coarse graining.” The use” of these concepts raised mathematical problems that they could not solve, but like the good physicists they were they assumed that everything was or could be made all right mathematicelly and went on with the physics. ‘Their mathematical worries, how- fever, became the seeds mm’ which grew @ whole beautiful subject called Fergodic theory.” Here we describe some recent (and some not so recent) Sevelopments that partially solve some of the problems thst worried the Founding Fathers; these results appear nat to be widely known in the physics community, and we hope to remedy the situation. (For @ more thorough review of the subject, see reference 1.) “oot, Lebowits i chaitman ofthe physics “deparimant at the Beller Graduate School of Selence, Yeshiva University, in New York Olver Penrose, = member of the ‘mathematics faculty at the Open Univers- 1y, Wallon, UK, worked oa tis etc while (04 leave ‘at the Appliee. Physics Depart- ‘mnt, Stanford ‘At the op of the page we see Boltzmann, Manwall, Gibbs and Einstein, the "Founding 1 Although results are not yet well enough developed to answer all the ‘questions in this area that are of inter- st to physicists today, such as the der- vation of kinetic equations or the gen eral problem of irreversibility, they do makeastart. Ergodic theory is concerned with the time evolution of Gibbs ensembles. It has revealed that there is more to the subject than the simple question of whether # dynamical system is ergodic (hich means, roughly, whether the system, if left to itself for long enough, trill pass close to nearly all the dynm- ical states compatible with conserva- tion of energy). Instead there is @ hi erarchy of properties that a dynamical system may have, each property imply ing the preceding one, and of which ergodicity is only the first (see figure 1). The next one, called “mixing,” provides a clear-cut formulation of the type of irreversible behavior that peo- ple try to obtain by introducing coarse-grained ensembles. At the top ff our hierarchy is @ condition (the Bernoulli condition) ensuring that in 3 certain sense the system, though ‘deterministic, may appear to behave as, randomly as the numbers produced by a roulette wheel ‘Some of the mathematical results we shall be diseussing have established the positions of some model physical systems in this hierarchy. Of particu lar interest to physicists is the work of Ya, Sinai? on the hard-sphere system which shows that this system is both ergodic and mixing. We shall also dis css somme work by A. N. Kolmogorov, vactoun kia Lelawi fe oa Ref 1 V, 1. Amold and J. Moser on systems of coupled anharmonic oscillators, Which shows that, contrary to a com: mon assumption,” these systems may hot even reach the “ergodie" rung on the ladder, (G. H. Walker and J. Ford have described this work for physi- ists) “Al the physical eystems_we_shall ‘pte numberof eres feed ‘sid_are_ confined. ta. finite region of, physical space. ‘The reason for our Fe EEEGr Us cleksea! mechanics fo that States qumstare system can never or tibie an of the properties higher than hte egadis in our hierarchy al eget omeae elaroe gaan ee tem ay appretinate lee the Be Kir ehaPtcernd by thee con cepts), ‘This is because the spectrin ofa ste quantum system i neve. yee ee To tite cs Sh Sytem the opecttm (othe Li he Sprache eotinvus- Tae reason for our restfeton to fie a ea that ery te Snowe about the eeaie properties of infinite ay tine Ths Tock of vowledge Te (potable ‘because nly for at nite Stem (oy mich term we mean the Baar nahe gnem aris see be formes tite) can one expect to find Stil ireverbte behave in gun tum hechanca Moreover, the ds tineuon beeneen mlcescopic and mae tencopeebsorable which appease Teanil co any complete theory of trey Sign 20 Mined equations can Shy Be formulated. precise for inf SRY stem, "Mack research remains PHYSICS TODAY / FEBRUARY 1973, wy) History of ergodic theory The “ergodic hypothesis” was into: ced by Boltzmana in 1871.” To cust Maxwell”. it) that the systom, ett to ifeat in ie actual slato et motion ‘wil, sooner of later, pass through every phase which is. consistent with the Equation of energy." In our notation “hase” means dynamical stato and the conginal ergosic hypathesis moans that ty and x are any two points on the ‘energy surface Sp then y = dh) tor tome {The ergodic hypothesis thus ftated was proven tobe false, whenover ‘Sp has dimensionality geste than ono by A. fosenthal and NM. Planchorsl in 3O13. |S. G. Brush ghes @ nice ac. Count ofthe early work on ths problem (Gee reteronce 5) ‘The. Gefniton of an ergodic systom given in equation 1 (page 25) can be Shown to be equivalent to what is sore- times called the. "quasi-ergodic™ hy- pothesis, which replaces “every phase In Manwel's statement by “every region A on Sp of tte aes,” withthe frtner qualification tat this Is te for “almost ai" dynamical states. Indeed as we point out in this article, the fraction of Time thatthe system wil spend in FIs fequal, for an ergodic system, 10 the fraction of tho area of Se that is occu- pos by, Tt wat shown by G. 0. Bitkhott in 1927 That ergodicity #6 equivalent to the energy surface being "metrically tanst thes Stated precisely this means that {system Is ergedic on S it and only it 4a tha regions Fn S let invariant by the time evolution. ¢4(F) = A. ether have zara area or have an erea equal to the area ofS. “The diicut part of Bikhotl's Theo em is 10 show that (x), which ine ‘oles taking the time average over ini rite tines, actualy exists for almost all when fe) fs an integrable function, It {5 then relatively easy 9 show that F(x) time invariants Hat Is, (6, (2)) Fy, and that ergodicly fs equivalent 498 being meticalyinvanstive to be done on the ergodic theory of in finite systems, both classical and ‘quantum, but we can be sure that the concepts to be discussed here will play ‘an important part init Surfaces and ensemble densities Before we go on to discuss the new results, we review some mathematical Gefinitions.* If our dynamical system hhas n degrees of freedom, we can think ofits possible dynamical states geome. trically, as points in @ 2n-dimensional space (phase space), with 1 position coordinates and m momentum coordi- nates. Ifthe energy of the system is E, then its dynamical state x [= (gs Gms Px «+. Po)] must lie on the energy surface Hx) = E, where H is the Hamiffonian function. We donote the energy surface, which is (22 — 1)-di- mensional, by Se or just S and assume that S is Smooth snd of finite extent; for example in the case of system of Harmonic oscillators, for which the Hamiltonian is a quadratic form, the energy surfaces. are (20 ~ 1)-dimen: sonal ellipsoids, ‘The time evolution of the system causes x to move in phase space, but ince we are assuming our system to be conservative the point x always stays fon the energy surface. Ifthe system is in some state x at some time to then its State at anyother time fg + f is tuniquely determined by x and ¢ (only) Let us call the new state dy(x). This defines a transformation @ from S tonto itself, ‘There is one such fanetion foreach value of. We shall want to integrate dynam: cal functions (that is, functions of the PHYSICS TODAY /FESRUARY 1873, dynamical staté) over S. When doing ‘this it ie convenient not to measure (201 ="ij-dimensional “areas” on the sur face $ in the ususl way ut to weight the areas in such a way that the natu- ral motion of the system on S carries ‘any region (after eny time t) into @ region ¢<(R) of the same area. This can be accomplished by defining the ‘weighted area of a small surface ele- iment near x, dx, to be such that dxdé is the correct Euclidean 2n-dimension- ‘al volume element of a “pill box” with base de and height dE. ‘By a Gibbs ensemble we mean an in finitely Targe hypothetical cllection of systems, all having the same Hamilto- rian but not necessarily the same dy- hhamical state. We shall only consider ensembles whose systems all have the Same energy, #0 that their dynamical States are distributed in some way over some energy surface S.- It may happen that this distribution can be described by an ensemble density; by this we mean a yeal-valued function p on S Such that the fraction of members of the ensemble whose dynamical states lie in any region ® on the surface S is J exits with dx the weighted area defined above, The simplest ensemble density ‘on Sis given by Ax) = C (all in) where C is a constant, which can be Getermined from the normalization condition fyC.dx = 1. This is called the mierocanonica! ensemble on S “The systems constituting the ensem ble evolve with time, so that the en semble density. ill depend on time. The rule connecting the ensemble densities pr describing the same en- semble at different times ¢ is Liou- ville's theorem, which can be written px) = pli (2Y] (all all x in S) where po(x) is the ensemble density at time zero, For example, Liouvlle’s theorem shows that the density of the Imierocsnonical ensemble does not change with time: If pds) = C for all x in S, then Liouville’s theorem gives, forany é, pax) = C forall xin. ‘The ergodic problem ‘The principal success of ensemble ‘theory has been in its application to ‘equilibrium. To calculate the equilib- rium value of any dynamical function wwe average it over a suitable ensemble. ‘The same ensemble also enables us to estimate the magnitude of the fluctua tions of our dynamical function. To ensure that the calculated averages are independent of time, we use an invar- fant ensemble; that is, one for which the fraction of members of the ensem: ble in every rogion Ron the energy sur- face S is independent of time. We al- ready know one invariant ensemble the ‘microcanonical, whote ensemble Gensity is uniform ‘on S. Before we can use it confidently to calculate ‘equilibrium values, however, we would Tike to be sure that this is the only i variant ensemble: If other invariant fensembles exist then, in principle, they could do just as well for the calculation of equilibrium properties, and we ‘would have to choose which to use in a particular situation. ‘There are two questions to settle: the first is whether there are_am variant ensembles on 8 That” do-it Wave an ensemble density. —Tn-general there ate for example in ike case of @ lard-ephere system in @ box one could have an invariont ensemble where every particle moves on the same ‘straight line being reflected at each fend from a perfectly smooth parallel wall (see figure 2) ‘The obviously exceptional character of this motion is reflected mathemati- cally in the fact that this ensemble, though invariant, is confined to a re gion of zero “area” on S and therefore hhas no ensemble density. To set up sucha motion would presumably be physically impossible because the ‘lightest inaccuracy would rapidly de. stroy the perfect. alignment. It is, therefore natural to rule out such ex ceeptional ensembles by adopting the Principle that any ensemble corre- 2 sponding to @ physically realizable sic {aNOU ECAR eRe Tena. “There remains the second pa¥t of the question: Are there any invariant ensembles on S that do have a density but differ from the microcanonical en. semble? This is equivalent to the er adic problem® in which ome compares the time averages of a dynamical func tion f, Peay = tim pf Keosanae with its microcanonial ensemble aver: eon face f ‘A system is said to be ergodic on its energy surTace SW Cmne averages are in general equal to ensemble averages; that is, if for every integrable function fwehave ra) fa a for almost all points x on S, “Almost all” means that if M is the set_of points x for which equation 1 is false, wwe have Suydr = 0. ‘The answer to our ‘second question is given by a theorem,¢ which we shall not prove: the micro canonical ensemble density i period of time, the fraction of time Garing which event A will be observed is given by the time average £*(xo), tshere xo is the initial dynamical state and gis defined by aay = fiifxisin Ry GB 0if not ‘The ergodic theorem tells us that for almost all initial dynamical states this fraction of time is equal to the ensem- Dle average og, which is te Sa) Sex ‘This is just the “probability” of the event A as calculated in the micro ‘canonical ensemble on S ‘Another way of defining ergodicity is to aay that" any” integrable: invariant function is constant almost every- where. ‘That is to say, iff is an inte- frable function satisfying the eondition that Mx] = fla) for all x in S, then there is @ constant ¢ such that (2) equals ¢ for almost all x: In other words, the set M of points < for which f(s} does not equal ¢ satisfies Syde = 0. This has the physical in ierpretation that for a Hamiltonian system ergodic on S every integrable constant of the motion is constant on S. Furthermore if ergodicity holds on each Sp then there are no integrable constants of the motion other than fanctions of the energy E. Indeed, if there were other constants of the to Gon (for example angular momentum if the Hamiltonian had an axis of sym: metry) we could construct invariant densities that were not mierocanonieal by taking p(x) to be a function of one of these other constants of the motion, and so the system would clearly be nonergodie. When such additional constants of the motion exist they must he taken into account in the sta istical mechanics and thermodynam: ies of the system; the standard meth ‘ods, based on the mierocanonieal en semble, must then be generalized for ‘these systems. Some special systems We now illustrate the ides of ergodi city by considering some specific me chanical systems. ‘The simplest. of these is the harmonic oscillator, whose fivarlantensemble—that is, the only OME SAETVIAE p1Ge(8)] = pC) for all x in Sif and only if the system is ergo. diconS. ‘The physical importance of ergodi city is that it can be used to justify the tse of the mierocanonical ensemble for cealculating equilibrium values and fluctuations. Suppose f is some mac roscopie observable and the system is started at time zero from a dynamical state x, for which f(x) has a value that may be very far from its equilibrium value.” As time proceeds, we expect the current value of f, which is 16.(2)), to approach and mosily stay very close to an equilibrium value with only very rare large fluctuations away from this value. ‘This equilibrivm value should therefore be equal to the time average fF because the initial period during which equilibrium is established con: tributes only negligibly to the formule defining /*(2).."The theorem tells us ‘that this equilibrium value is almost always equal to (), the average of f in ‘the microcanonical ensemble, provided the system is ergodic. ‘To justify the use of the microcanon ical ensemble in calculating equilib. ‘um fluctuations we proceed in a simi lar way. For some observable event A (Guch as the event that the percentage ‘of gas molecules in one half of a con: tainer exceeds 5196) let R be the region in phase space consisting of all phase points compatible with the event A; that is the event A is observed i and only if the phase point is in at time Inthe system is observed over a long. 3 Arrows denote implication: Sernoult systom equivatent to a roulette wheel, and £9 0M, Every mining system is ergodic, every "The mide column bats the twee sjstoms clscussed inthis avticlo—with the “strongest” al the top-—as well a& the Kesysiem (atter A. N. Kolmogorov). which is nor discussea te Systems end a the eght physical interpretations a Impiestions. PHYSICS TODAY /FESRUARY 1973, ae (aot Hamiltonian (in some suitable units) is 1 Hiq.p) = p09 +9) where w is the angular frequency. ‘The transformation ¢; for this system is a rotation through angle wt in the (.7) plane. ‘The trajectories, which here Coincide with the energy surfaces Se, are circles of radius (2E)", (The sur face element dx is here proportional to the ordinary length of an are segment.) "To be invariant under the transforma tion ¢: an ensemble density on S must bbe unaffected by rotations and is there fore a constant. It follows, then, that the only invariant density isthe taiero- ccanonical density and so. the simple harmonic oscillators ergodic. ‘Almost as simple is the multiple harmonic oscillator (physically, sey, an ideal enystal), that is, a eystem with two or more degrees of freedom whose potential energy is a quadratic form in the position coordinates, Unlike the simple harmonic oscillator it cannot be ergodic, because it has constants of the ‘motion (the energies of the individual normal modes) that are not constant ‘on the energy surfaces (the surfaces of constant fotal energy) 1 used o-be thought that this lack of Sadat nik sy agtgeat end hat fay small anharmonicity. (uch as would inevitably be present in. real tntem) rust make the system ergodic by permitting transfer of energy from ‘one mode to another. In 1964, how. ever, Kolmogorov announced results that contradicted this belief." Tn 1955, Enrico Fermi, J Pasta and. W. Ulam carried out @ computer simulation of such a system, Initially, they excited fone mode only, and instesd of the equipartition of the energy between all modes predicted by the microcanonical ensemble they found that most of it fappeared to remain concentrated in a few modes; this indicated that anhar- monic oscillator systems may not be ergodic The lack of ergodicity was. proved gorously by Kolmogors, Arsalan They” found that. ifthe frajueaces of the unperturbed oxilla tore are not “rationally connected" (Chat is, if no rational linear combi tion of them is zero) then, in general, ‘adding to the Hamiltonian an anher. ‘monic perturbation sufficiently small compared to the total energy does not make the system ergodic. ‘The unper- tured trajectories (possible paths of the phase point) all lie on n-dimen: sional surfaces in S (which has 2n~ 1 dimensions) called “invariant. tori,” and “KAM” prove that under @ weak perturbation most trajectories continue {0 lie within smooth n-dimensional tori, £0 that the perturbed system is also nonergodie. ‘The trajectories that’ do rot lie on the new invariant tor are, on Pwvsies TODAY /FEBRUABY 1973 Ensemble with no ensemble density. Hard spheres mova up and down the dotted i Which moots the perfect smooth hard wails at right angles. Colisions between Bariles and colsions withthe was do not Setlect ne partes irom th in thoy are pertectyalgned at tho star. An ensemble ff such systoms har no ensemble density Because tte conconated ona region on fhe energy surlace with 2eco area (just 22 the area of a tne or of 8 ine sogmant fa 8 plane ero} Figure 2 the other hand, very erratic indeed and may fill some (2n ~1)-dimensional region densely. One consequence of this very compliested behavior is that even though the system is not ergodic ‘the motion can no longer be decom. posed into independent normal modes. ‘Similar results probably hold also for rationally connected Irequencies (which cannot be treated rigorously, although they are of more physical in: Helles!© carried out computer ealcula- ee wel (nts p+ of tai) + 2 1,3) (rufa) ees Pd ey oe ee sci eh ca oh gare, see ne ee Th the case of gases, the situation is T t t smooth nardvatts | t i somewhat different, If there were no interaction at all between the mole coules then the energy of each molecule ‘would be an invariant of the motion, 0 that the system (an ideal gas) would be nonergodic. The KAM theorem would therefore lead us to expect nonergodi city to persist in the event of a sufi ciently Weak interaction between the particles. ‘The actual interactions, hhowever, are not weak because two molecules very close together repel each other strongly; consequently the theorem does not apply. A. simple model of this type is the hard-sphere gas enclosed in a cube with perfectly reflecting walls or periodic boundary conditions. Sinai has outlined a proof that this system i ergodic; he bas pub: lished a detailed proof, based on the same idess, for a particle moving in a Periodic box containing any number of Tigid convex elastic scatterers, We shall refer again to this important re- sult icing 24 10% p 139, 17 We have seen how to formulate a condition to ensure that the equilib uum properties of a dynamical system are determined by ite energy alone and ean be calculated from a microeanoni cal ensemble. This ergodicity condi tion does not, however, ensure that if wwe start from’e nonequilibrium ensem ble the expectation values of dynami cal functions will approach their equi- brium values as time proceeds. An illustrative example is given by the harmonic otcillator. For the har monie-oseillator system, Liouville’ theorem shows that the ensemble den sity repeats itself regularly at time in tervals of 2x/u therefore all ensemble Tos ° oa averages alzo have this periodicity, and s0 cannot irreversibly approach thelr alibriom values “Sito ensure that our ensembles ap- proach equilibrium in the way we ‘would expect of ensembles composed of Teal systems, we need a stronger condi tion than ergoiity. To see what is required, let us start at = 0 with some ensemble density g2) on. S, ‘which is supposed to represent the ini I nonequilibrium state, At a later time the ensemble density 1s, by Liouville’: theorem, ole]. The expectation value of any’ dynamical ‘ariable fat time tis therefore J ferndo.coldx Ast becomes large, we would like this integral. to approach the equilibrium ‘ale off, which i (for an ergodic sys tem). falls) de] fede. A efit Condition for this s that the system Should easy the eondition elled mir ing which is that for every pai of functions and g whose squares are in teqrableon Swe require e fa)eeda) dx = [sores f ec» ‘The special case where § is po shows that integral 2 will approach the equi: Iibrium value of f for large ¢ when the system is mixing. Another way of looking at this condition is that it re quires: every equilibrium time-depen Gent correlation function suchas (laeler(2)) to approach a limit () (e) 5 Toa ° oy moe ° oa Nonergodicity of an anharmonic oscilator system with rationsly connected troquoncies ‘The Hamiltonian fortis system i given on page 26. Michael Heron and Cart Helios (soe Teteconce 10) did computor calcuiations for this aystem and found that energy surfaces ‘Se win € equal to 1/12, 1/8 and probably 1/6 ar0 not ergodic. The planes shown here ‘re intersections af the surface gy equal to Zoro with Spfor E equal to 1/12 (a) 1/8 (0) ane {7/6 (ch, and the pnts are the intereections ofa trajectory with this plane. When the tra Jectory es on a smooth two-dimensional iavatlant torus, the intersection points form & Emecth curve, bul the Intersections of an “ereaic” trajectory (ane that does not ie on & Emooth curve) are more or less ransom.” Note thatthe fraction of area corresponding 10 smooth curves (which are responsible for the novergedic Dehavier) decreases with ‘creasing energy. fs t approaches + ©. ‘The condition can be shown to be equivslent to the following requirement: if Q and R are arbitrary regions in S, snd an ensemble is initially distributed uniformly over , then the fraction of members of the ensemble with phaée points in Rat time will approach a limit as ¢ ap- proaches ©; this limit equsls the frac. tion of the area ofS occupied by R. Mixing is a stronger condition than ‘ergodicity: it can easily be shown to imply ergodicity but is not implied by it, as we have seen in the case of a sim ple barmonicoccillator. ‘The mathematical definition of mix ing was introduced by John von New mann’? in 1982 and developed by E. Hopf, but goes back to J. Willard (Gibbs, who diseusses it by means of an analogy: “the effect of stirring an incompressible liquid... Let us sup- ppose the liquid to contain certain amount of coloring matter which doos not affect its hydrodynamic properties {end] thst the coloring matter is dis- isibuted with variable density. IFwe give the liquid any motion whatever... the density of the coloring matter at any same point of the liquid will be un- changed ... Yet ... stirring tends to bring a liquid toa state of uniform mixture.” Gibbs saw clearly that the ensemble density p, of a mixing system does not approach its limit in the usual “fine. grained” or “pointwise” sense of p(x) approaching a limit as ¢ ~+ © for each fixed x, “Rather, it is a“ rained” or “weak” limit, verage of p(x) over a region R in S approaches a limit as ¢ —~ = for each fixed R. (A similar distinction applies in defining the entropy. ‘The fine. Figures grained entropy —k Sox) log piix)dx, ‘where Kis Boltzmann's constant, re: tains its initial value forever, but the coarse-grained entropy ~k J9,*(2) (log perG}dx, where g(x) 18 ‘a. coarse grained ensemble density obtained by averaging p)(2) over cells in phase space, does change for a nonequil brium ensemble, and approaches as its limit the equilibrium entropy. value helog f six.) tis sometimes argued that one can- not have a proper approach to equilibri- uum for any finite mechanical system because of a theorem, due to Poincaré,1> that every. such system eventually retums arbitrarily close to its initial state, (The time involved, however, will be enormously large for a macro scopic system. Boltzmann, for exam ple, estimated a typical Poincaré peri fd for 100em? of gas to be enormously long compared to 10 raised to the power 10 raised to the power 10 years.) Here, however, we are considering ensembles, not individual systems, and the mixing ‘condition guarantees that the ensemble density eventually becomes indistin fuishable from the microcanonical Gensity and remains so forever after. Te is true that individual systems in the ensemble will return to their initial dynamical states, as required by Poin- caré’s theorem, but this will happen at different times for different systems, so that at any particular time only a very smal fraction of the systems in the en: semble are close to their initial dynam ical states, ‘The reason for the irrelevance of Poincaré recurrences in mixing systems is that the motion of the phase point is ‘very unstable. Dynamical states that start very close to each other in phase PHYSICS TODAY /FEBRUARY 1975, 1 famiiar example of “mixing.” According to V. |. Arnold and A. Avex, the two lauids are fum {wenty percent) and cola (eighty percent), with he result ofthe mixing process known ‘asa "Cuba tbr space become widely separated as time progresses, 0 that the recurrence time depends extremely sensitively on the initial conditions of the motion. (The importance of this instability in statis tical mechanics was first recognized by N.S. Krylov, a Russian physicist who died in his twenties in 1947; a recent paper gives a theorem that shows this Statement about instability to be a rig forous consequence of mixing.) This type of instability appears to be char acteristic of teal physical systems, and Teads to one sort of irreversibility even if we could reverse the velocities of every particle in a real system that hhas been evolving towards equilibrium, the system would not necessarily. re tum or even come close to its initial, dynamical state with the velocities re versed because the unavoidable small external perturbations would be mag- nified.. This instability is noticeable in molecular-dynamies calculations with hard-sphere systems: if we numerical- ly integrate the equations of motion from time 0 to ¢ and then try to recover {See reference I fr details ofthe process.) Figure’ the initial state by integrating back- wards from time ¢ to time 0, we obtain instead a completely new state, This ig because the numerical integration is unstable to small rounding.off errors made during the computation, which play the same role as extemal pertur- bations in areal system, ‘Only a few physical systems have been proven so far to be mixing. The most Important is the hard-sphere gas, mentioned above. Sinai's proof that this system is ergodic? also gives the stronger result. that itis” mixing. Roughly, Sinai’s method of proving mixing is to show that the hard-sphere system i unstable in the sense dis. ccusted above. Physically this instabil ity comes from the fact that a slight change in direction of motion of any particle is megnified at each collision Inith the convex surface of another par. ticle, The full proof for the simplest ease of a “single” particle moving ‘among fixed convex scatterers occupies some 70 pages of difficult mathemat. jes. (G, Gallavotti has provided an el Bernoulli systems For lack of space we shall not discuss the next concept in our sequence, that of the “K-system” (after Kolmogorov) For this and related topics such as the Kolmogorov-Sinai “entropy” the read. er is referred to reference 1. Instead ‘we jump to the last concept in our hi: erarchy, that of a Bernoulli system. By a Bernoulli dynamical system we mean one that is, in @ sense we shall Gefine, equivalent to an infinite se ‘quence of spins of a suitably designed Toulette wheel "To make this equivalence more pre: cise, we define a finite partition of the tenergy surface Sof our dynamical system fas any finite collection of " nonover Tapping regions Mo,-. Rn» which to gether cover the whole of S. Suppose that some device ean be made that will, Getermine which of these regions the phase point is in at any time, but gives, rho information whatever about, which part of the region it is in. That is, every time we use this measuring de vice we obtain an outcome that is a positive integer—the label of the region the phase point of the system is in at that time. ‘Suppose we use the device repeated- ly at intervals of, say, one second. Its ‘outcome will be & sequence of positive integers from the set [0,-.. — ll, ‘which can be extended indefinitely. In general, we would expect theve integers {fo be correlated; that is, the micro: canonical probability for each new ob: servation depends on what has been observed before (as ina Markov pro cess, for example). This correlation ‘comes about because the dynamical States of the system at different times fre deterministically related, through the equations of motion. "The astonishing fact is, however, that for a certain class of dynamical “Toe baker transformation recalls the kneading ofa ploce of dough. rigiel height and twice ss orginal with, and then cut the resulting Fectangle in al! and move the right haf of he rectangle above the let Figures PRYSICS TODAY / FEBRUARY 1873 We first squash tha systems itis possible to choose the re- digit after the Binary point from p and Ri, at time ~2 in Ryz, and s0 on. se ate oat ay AM CO Bary) = teeter times by a roulette wheel, At the cal ensemble for this system has a uni line time, he eons so chosen gine where dhe 2 and qu take onthe valoes fern deny in the equate tf nt di {rough information to discriminate be- 2nd {This transformation invert” heult to ste that the micoeanonical ‘tween dynamical states: if two sys. iple: and from it we can define #2 #8 probability of each of these digits in tems have different dynamical states the inverse of ¢ and 4, as the t th ite- the binary expression for (p.q) is 1/2 at some time, then the observations "ation of dai (Only intoger values of gnd is uncomelated. with ail the other ‘made on them cannot yield identical the time ate used here, rather than all digits, ‘The observations made at dif results for the observations at every eal values as in our discussion of dy- ferent times ¢ (= integer) are therefore Gite. When such regions can be cho- amics earlier in this article but we do uncorrelated, and so the baker's trans fen, we call the system a Berouli sys. NOt regard this distinction’ as-impor- fermation model is a Hemnull system. tem, tant.) Moreover, the transformation “Possibly ne should ot read too Recently D. Ornstein and B. Weiss! Preserves geometrical area, and so the much physical meaning into this type showed that the class of Bernoulli sys- @8alog of the microcanonical distribu- of result, for with a more complicated tems includes a type of dynamical sys-* tion isjust a uniform density. dynamical system the regions Ro, Ry, tem (the geodesic flows on a space of To see how this completely Rs... ft, would probably be exceed: Constant megative curvature) whose er- deterministic system can at the same ingly complicated tets in phase space, fodie properties are very similar to time behave like a roulette wheel, we but. trom a “philosophicel™ point ef these of the hard-sphere system stud. take the regions Ro and Ry to be the view it is very interesting to see how ied by Sinai, and it appears likely that two rectangles 0 < p <'1/2,1/2 = p< the same dynamical system can show the hard-sphere system too is @ Ber. 1 shown in figure 6. Suppose the perfect determinism on the mwieroscop. soul system phase point at time is ie level and at the same time perfect ‘As an illustration of @ simple Ber- yale Ora randomness on a “macroscopic” level sot aem, coterie whos pp, 20 OPO Te taal ee treat phase space is the square 0

    You might also like