epl draft

On statistics of molecular chaos

arXiv:0911.0651v1 [nlin.CD] 3 Nov 2009

PACS PACS PACS

05.20.Dd Kinetic theory 05.20.Jj Statistical mechanics of classical uids 05.40.Fb Random walks and Levy ights

Abstract. - It is shown that the BBGKY equations for a particle interacting with ideal gas imply exact relations between probability distribution of path of the particle, its derivatives in respect to the gas density and irreducible many-particle correlations of gas atoms with the path. These relations visualize that the correlations of any order always signicantly contribute to evolution of the path distribution, so that the exact statistical mechanics theory does not reduce to the classical kinetics even in the low-density (or Boltzmann-Grad) limit.

Introduction. In the seminal book [1] Bogoliubov formulated powerful tools for statistical theory of transport processes and noises in many-particle systems, in particular, the exact hierarchy of evolution equations (now referred to as the Bogolyubov-Born-Green-Kirkwood-Yvon, or BBGKY, hierarchy [2]) for s -particle ( s = 1, 2, . . . ) distribution functions. But in practice the theory was canceled out by common prejudice that it must conrm, at least for dilute gases, the Boltzmanns molecular chaos hypothesis, - the Stozahlansatz [3], - and thus reduce to the Boltzmann equation. This idea is so much attractive that Bogoliubov introduced own hypotheses which allow to truncate the BBGKY hierarchy at s = 2 and perform the desired reduction [1]. In fact, however, until now neither Boltzmanns nor Bogoliubovs assumptions have a rigorous substantiation based on the BBGKY hierarchy itself 1 , and hence the Boltzmann equation also stays without substantiation 2 . Here, I suggest a short visual proof of that all this never will be substantiated, even for arbitrary dilute gas (or
1 We take in mind spatially inhomogeneous gas evolutions, of course, since in respect to strictly homogeneous (translation invariant) dilute gas the Stozahlansatz is undoubtedly true. 2 The frequently mentioned Lanford theorem [4] about gas of hard spheres under the Boltzmann-Grad limit concerns the so called hard-sphere BBGKY hierarchy (about it see also e.g. [2, 5]) which is not a true BBGKY hierarchy since represents interactions of the spheres by invented terms like the Boltzmann collision integrals (i.e. postulates what should be proved, if any). We here consider regular smooth interaction potentials only and thus the true BBGKY equations where the interaction is represented by the forces (potential gradients) like in the parent Liouville equation. But our conclusions naturally extend to the limit of singular hard-sphere interaction.

Boltzmann-Grad gas). At that, in order to simplify the proof and at once strengthen it, instead of the usual gas I consider a test particle in thermodynamically equilibrium ideal gas whose atoms interact with this particle only but not with each other (thus concentrating on situation least favorable for inter-particle correlations). Our rst step will be representation of the BBGKY hierarchy of equations for this system in terms of irreducible s-particle correlations between the test particle and s 1 gas atoms ( s = 1 corresponds to probability distribution of variables of the test particle). Then, we discover that these equations produce a set of formally exact relations between any s-particle correlation and derivative of the previous (s 1)-particle one in respect to the gas density. Next, make sure that all the correlations keeps non-zero even under the low-density (or Boltzmann-Grad) limit and, consequently, no truncation of the BBGKY hierarchy can be made without qualitative detriment to its solution. Finally, we touch upon important properties of the correlations. The BBGKY equations and cumulant distribution functions. We are interested in random walk R(t) of a test molecule (TM) which is placed into thermodynamically equilibrium gas and starts at initial time moment t = 0 from a specied position R(0) = R0 . Let P and M denote momentum and mass of TM, m , rj and pj ( j = 1, 2, ... ) are masses, coordinates and momenta of gas atoms, (r) is potential of (short-range repulsive) interaction between any of them and TM, and n is gas density (mean concentration of atoms). At arbitrary time t 0 , full statistical description of this system

p-1

Yu. E. Kuzovlev is presented by the chain of (k + 1)-particle distribution functions ( k = 0, 1, 2, ... ): F0 (t, R, P| R0 ; n ) which is normalized (to unit) density of probability distribution of TMs variables, and Fk (t, R, r(k) , P, p(k) | R0 ; n ) (where r(k) = {r1 ... rk } , p(k) = {p1 ... pk } ) which are probability densities of nding TM at point R with momentum P and simultaneously nding out some k atoms at points rj with momenta pj . A rigorous denition of such distribution functions (DF) was done in [1]. In respect to the coordinates rj they are not normalized, but instead (as in [1]) satisfy the conditions of vanishing of inter-particle correlations under spatial separation of particles (in other words, DF satisfy a cluster property with respect to spacial variables). Subject to the symmetry of DF in respect to xj = {rj , pj } these conditions can be compactly written as follows: Fk Fk1 Gm (pk ) at rk , where Gm (p) is the Maxwell momentum distribution of a particle with mass m . The enumerated DF obey the standard equations [1]: Fk = [ Hk , Fk ] + n t P (R rk+1 ) Fk+1 ,
k+1

Apparently, from viewpoint of the probability theory, Vk are a kind of cumulants, or semi-invariants, or cumulant functions (CF). It is important to notice that if all these CF were zeros then all conditional DF of gas, Fk /F0 , would be independent on initial position R0 of TM and thus on its displacement RR0 . This fact makes visible very interesting speciality of the CF Vk : they are not mere correlations between instant dynamic states of TM and k gas atoms but simultaneously their irreducible correlations with the total past TMs displacement. Relation of many-particle correlations to probability law of diusion of the test particle. In terms of CF the BBGKY hierarchy acquires tridiagonal structure (we omit uninteresting algebraic details): Vk = [Hk , Vk ] + n t P
k

(R rk+1 )Vk+1 +
k+1

+T
j =1

Pkj Gm (pk ) E (rk R)

P Vk1 . (4) + MT P

(1)

Here E (r) = E(r) , and Pkj symbolizes transposition with k = 0, 1, ... and along with obvious initial conditions of the pairs of arguments xj and xk . At that, initial conditions (2) and the above-mentioned clustering condiFk |t= 0 = (R R0 ) exp (Hk /T ) = tions [1] take extremely simple form: (2) = (R R0 ) GM (P) k =1 E(rj R) Gm (pj ) , j V0 (0 , R, P| R0 ; n) = (R R0 ) GM (P) , where Hk is Hamiltonian of subsystem k atoms + Vk (0 , R, r(k) , P, p(k) | R0 ; n) = 0 , (5) ... = TM , [..., ...] means the Poisson brackets, Vk (t, R, r(k) , P, p(k) |R0 ; n) 0 at rj k ... drk dpk , (r) = (r) , and E(r) = exp [(r)/T ] . Notice that one can treat TM as molecule ( 1 j k ). Thus, as it should be with cumulants, Vk of non-uniformly distributed impurity and (1) as equations (k > 0) vanish under removal of already one of atoms. of two-component gas [1] in the limit of innitely rare imFrom equations (4) as combined with the boundary and purity with main component being in spatially uniform initial conditions (5) it is clear that passage to the limit and thermodynamically equilibrium state. in (5) must realize in an integrable way, so that integrals Equations (1) together with (2) unambiguously deter- V = V take nite values. Let us consider them. k k+1 k+1 mine evolution of F0 and eventually probability distriBy applying the operation k to equations (4) one obtains bution of total TMs displacement, or path, R R0 . These equations will become more transparent if we make Vk a proper linear change of DF Fk by new functions Vk , = [Hk , Vk ] + n (R rk+1 ) Vk+1 + t P k+1 namely, with the help of recurrent relations as follow: (R rk+1 ) Vk+1 + (6) + F0 (t, R, P| R0 ; n) = V0 (t, R, P| R0 ; n) , P k+1 F1 (t, R, r1 , P, p1 | R0 ; n) = k (3) P = V0 (t, R, P| R0 ; n) f (r1 R, p1 ) + Vk1 + +T Pkj Gm (pk ) E (rk R) MT P + V1 (t, R, r1 , P, p1 | R0 ; n) , j =1 where f (r, p) = E(r) Gm (p) , F2 (t, R, r , P, p |R0 ; n) = = V0 (t, R, P|R0 ; n) f (1 , p1 ) f (2 , p2 ) + + V1 (t, R, r1 , P, p1 |R0 ; n) f (2 , p2 ) + + V1 (t, R, r2 , P, p2 |R0 ; n) f (1, p1 ) + + V2 (t, R, r(2) , P, p(2) | R0 ; n) , where j rj R , and so on.
(2) (2)

(with k = 0, 1, ... ). Because of (5) initial conditions to these equations are zero: Vk (t = 0) = 0 at any k . Now, in addition to Vk , let us consider derivatives of CF in respect to the gas density, Vk = Vk /n . It is easy to see that dierentiation of (4) in respect to n yields equations for the Vk which exactly coincide with (6) after changing there Vk by Vk . Besides, in view of (5), initial conditions to these equations again all are equal to zero: Vk (t = 0) = 0 at any k 0 . These observations strictly p-2

On statistics of molecular chaos imply the equalities Vk = Vk , or Vk (t, R, r(k) , P, p(k) | R0 ; n) = n =

the integrated CF under the Boltzmann-Grad limit must be instead of n and rewrite (8) in the form (7) W0 (t, ; ) +
k=1

Vk+1 (t, R, r(k+1) , P, p(k+1) | R0 ; n) .

uk Wk (t, ; ) = W0 (t, ; /(1 + u)) , k!

This is main, original and formally exact, result of the present paper. Evidently, it conrms the assumed niteness of integrals Vk = k+1 Vk+1 . Notice also that it anticipates similar but more complicated general statements of statistical kinetics of uids. Discussion and resume. The equalities (7) contain the proof promised in the beginning of this paper. Indeed, they show, rstly, that all the many-particle correlations between gas atoms and total path, or displacement, of the test molecule (TM) really exist, i.e. dier from zero. Secondly, integral values of all the correlations, represented in the natural dimensionless form, have roughly one and the same order of magnitude. Indeed, multiplying equalities (7) by nk and integrating them over TMs momentum and all gas variables, we have nk Vk (t, ; n) nk ...
1 k

Vk dP =

= nk

k V0 (t, ; n) ck V0 (t, ; n) , nk

where V0 (t, ; n) = V0 (t, R, P|R0 ; n) dP is just the probability density distribution of the TMs displacement, = R R0 , and ck some numeric coecients obviously comparable with unit. Equivalently, unifying all CF into one generating function, we can write

V0 (t, ; n) +
k=1

u k nk Vk (t, ; n) = k! (8)

= V0 (t, ; (1 + u) n) .

We see that distribution of total of the TMs random walk is made of its correlations with gas atoms like the walk itself is made of collisions with them. Hence, none of the REFERENCES correlations can be neglected if we aim at adequate anal[1] Bogolyubov N.N., Problems of dynamical theory in staysis of solution of the BBGKY equations. tistical physics (North-Holland, Amsterdam) 1962. We see also that characteristic spacial volume occupied [2] Resibois P. and de Leener M., Classical kinetic theory of by the correlations has an order of the specic volume: uids ( Wiley, New York) 1977. ( | 1 ... k Vk dP |/V0 )1/k n1 . In the Boltzmann- [3] Boltzmann L., Vorlesungen uber Gastheorie (Leipzig) Grad limit, n , r0 0 ( r0 is eective radius of the 1896-1898. 2 interaction), r0 n = 1/ = const , this volume becomes [4] Lanford O.E., Lecture Notes in Physics, 38 (1974) 1. vanishingly small as measured by the TMs mean free path [5] van Beijeren H., Lanford O.E., Lebowitz J. L. and Spohn H., J. Stat. Phys., 22 (1979) 237. . But, nevertheless, it remains on order of the eective 2 [6] Balescu R., Statistical dynamics (ICP, London) 1997. volume of collision, r0 . This observation prompts that k-particle correlations are concentrated just at those [7] Kuzovlev Yu. E., Sov. Phys. JETP, 67 (1988) 2469. particular subsets of k-particle phase space which correspond to collisions. Therefore eects of the correlations completely hold under the limit. The same is said by the 3 Reprint of this article is available from the arXiv: 0907. 3475 . 4 See also [ Yu. Kuzovlev, Theoretical and Mathematical Physics, equality (8) which also holds out. This becomes obvious if we take into account that actually important parameter of 160 (3), 1301 (2009)] and references therein. p-3

where Wk (t, ; ) = lim nk Vk (t, ; n) . Thus, in essence nothing changes under the Boltzmann-Grad limit. It is necessary to underline that the correlations under our attention are qualitatively dierent from correlations which appear in standard approximations of the BBGKY hierarchy and connect velocities of particles after collision (see e.g. [6]). In our notations, a pair correlation of such the kind would look nearly as V1 (t, R, r, P, p) = F1 (t, R , r , P , p ) F1 (t, R, r, P, p) , where the primed variables describe the pre-collision state. Clearly, because of the phase volume conservation during collision, integration of this expression over = r R and the momenta results in zero. This observation shows that correlations under our interest do live not so much in momentum space as in the conguration space. Their salt is that they connect coordinates and walks of particles and may coexist with statistical independence of particles velocities. By their very nature, they are attributes of spatially inhomogeneous states and evolutions (evolution of V0 (t, ; n) gives an example). More profoundly, statistical and physical meanings of such correlations were discussed already in [7] (so for detail see [7] 3 ). In view of the aforesaid 4 we have to conclude that the Boltzmann-Lorentz equation [2, 5] and, moreover, the Boltzmann equation in itself do not represent a (lowdensity) limit of the exact statistical mechanical theory. The classical kinetics is only its approximate probabilistic model (may be good in one respects but caricature in others). Of course, in the exact theory also molecular chaos does prevail. But here it is much more rich, even (and rst of all) in case of dilute gas, and does not keep within naive probabilistic models.