Mesoscopic model for solvent dynamics

Anatoly Malevanets and Raymond Kapral Citation: The Journal of Chemical Physics 110, 8605 (1999); doi: 10.1063/1.478857 View online: http://dx.doi.org/10.1063/1.478857 View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/110/17?ver=pdfcov Published by the AIP Publishing

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

JOURNAL OF CHEMICAL PHYSICS

VOLUME 110, NUMBER 17

1 MAY 1999

Mesoscopic model for solvent dynamics

Anatoly Malevanets
Department of Physics, Theoretical Physics, Oxford University, 1 Keble Road, Oxford OX1 3NP, England

Raymond Kapral
Chemical Physics Theory Group, Department of Chemistry, University of Toronto, Toronto M5S 3H6, Canada

Received 18 November 1998; accepted 1 February 1999 Complex uids such as polymers in solution or multispecies reacting systems in uid ows often can be studied only by employing a simplied description of the solvent motions. A stochastic model utilizing a synchronous, discrete-time dynamics with continuous velocities and local multiparticle collisions is developed for this purpose. An H theorem is established for the model and the hydrodynamic equations and transport coefcients are derived. The results of simulations are presented which verify the properties of the model and demonstrate its utility as a hydrodynamics medium for the study of complex uids. 1999 American Institute of Physics. S0021-96069952416-1

I. INTRODUCTION

The investigation of the dynamics of complex uids is a challenging task. In complex chemically reacting systems many chemical species dissolved in a solvent may undergo sequences of reactions leading to oscillations or chemical patterns.1 The reactive dynamics may depend on the solvent motions since the uid ow elds inuence the nature of the mixing of the species. Specic instabilities have been ascribed to ow effects, such as differential-ow-induced chemical instabilities.2 Reactions in porous media constitute another example. Here one is interested in how a ow eld in a complicated medium inuences the reactions that may themselves change the nature of the medium, for instance, through dissolution reactions.3 Similar considerations apply to the study of rheological properties of colloidal suspensions or polymers in solution.4 An especially important class of problems concerns the conformational dynamics of biopolymers such as proteins in solution.5 All of the above examples, and others like them, share the feature that one is interested in the detailed microscopic dynamics of some degrees of freedom of the system interacting with a solvent whose dynamics is essential for the phenomena but whose detailed properties are not of interest. The systems are sufciently complex that a full molecular dynamics MD simulation of the system plus solvent is impossible. Furthermore, the dynamics of interest often occurs on long time scales and over long distances; for example, the relaxation times for large polymers can be very long and the distance scales for many chemical pattern forming processes range from mesoscopic to macroscopic scales. In such circumstances one is led to consider mesoscopic models for the solvent dynamics that incorporate the essential dynamical properties, yet are simple enough to be simulated for long times and on long distance scales.6 These mesoscopic models must not only faithfully reproduce the main dynamical features of the solvent but must also be microscopic in character to permit coupling between microscopic
0021-9606/99/110(17)/8605/9/$15.00 8605

degrees of freedom, whose dynamics are of interest, and the solvent. For instance, one must be able to couple the monomer units of a polymer to the solvent or the reactive collisions between molecules to the ow of the solvent responsible for their mixing. A variety of mesoscopic models have been constructed for this purpose ranging from Langevin models that have been employed as simple heat baths in MD studies of biomolecule dynamics,7 to schemes such as Direct Simulation Monte Carlo DSMC methods,8 lattice Boltzmann methods,9 and extensions of hydrodynamic lattice gas automaton models.10 In this paper we present a mesoscopic model for simulating a uid that has the desirable features mentioned above. The uid is modeled by particles whose positions and velocities are treated as continuous variables. The system is coarse grained onto the cells of a regular lattice and there is no restriction on the number of particles that may reside in a cell. The dynamics is carried out synchronously at discrete time steps. Particle streaming is treated exactly while the cells are the collision volumes for a multiparticle collision dynamics that differs from that of DSMC schemes. The dynamics satises the mass, momentum and energy conservation laws and we shall show it yields the correct hydrodynamic equations. Since it is a particle model, collisional coupling to other microscopic degrees of freedom is easily incorporated and its application to complex system geometries is straightforward. The outline of the paper is as follows. Section II describes the streaming and multiparticle collision dynamics that forms the basis of the mesoscale description of the system. The Boltzmann approximation to the full dynamics is considered in Sec. III; a Boltzmann-type kinetic equation is derived, an H theorem for the evolution is proved and the ChapmanEnskog expansion is used to derive the Navier Stokes equations and determine the transport coefcients for a particular collision model. In Sec. IV simulations are carried out to conrm the utility of the model. The velocity
1999 American Institute of Physics

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

8606

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

distribution is shown to be Maxwellian and the hydrodynamic ow patterns for ow past a disk are examined. The conclusions of the study are presented in Sec. V.
II. STREAMING AND MULTIPARTICLE COLLISION DYNAMICS

i x V

vi 2

i x V

vi V 2 . V

We note that the elementary measure d d xi d vi is invariant with respect to streaming and collision transformations. For the deterministic streaming operator it follows from the identity, Jacobian

The system we study consists of N particles of unit mass with position xi and velocity vi coordinates, i 1,2, . . . , N . The evolution through a unit time interval is given by successive application of streaming and collision steps. During the streaming step the particle positions change in the standard way,11 xi x i v i , 1

xi t 1 1, xi t

and for the collision transformation it results from the semidetailed balance condition and phase space volume conservation during rotations, d vi i ,V V
N N VN

while in the collision step particle velocities transform according to vi V , vi V 2

N ) p( V

dv i i 7

d vi i ,V V
N

N VN

N, p V

is a random rotation from a set and V is the where average velocity of the colliding particles. In this multiparticle collision event, the velocity vector of particle i, relative to the mean velocity V, is rotated by a randomly chosen rotation operator to give the post-collision value of the velocity. We shall show below that the mass, momentum and energy collision invariants are preserved under this multiparticle collision dynamics. We imagine coarse graining the system into Wigner Seitz cells centered on the nodes of a regular lattice L. For a cubic lattice the Wigner-Seitz cell V is dened by x 1/2, where we introduce the norm x max(xx ,xy ,xz) and denotes a lattice coordinate. These cells dene the collision volumes for the multiparticle collision dynamics. The collisions are simultaneously performed on all particles in a , but may difWignerSeitz cell with the same rotation fer from cell to cell. The evolution governed by Eqs. 1 and 2 may be written in the form of a Liouville equation for the probability density, P V where CP V N , X N , t 1
L
N
N

which is valid if the choice of the rotations is independent of ( N ) ) is the conditional probability of V velocities. Here p ( i are given V ( N ) . In the above equation vi and v the rotation the post- and precollision velocities, respectively. Assuming the validity of the ergodic hypothesis, which is supported by the results of numerical simulations, we conclude that the stationary distribution is given by the microcanonical ensemble expression, P V N , X N A

2N
N N i1

vi 2

d 2

i1

vi u ,

where u is the mean velocity of the system. After integration over the coordinates and velocities of particles with i 2, . . . , N we arrive in the limit of large N at the Maxwell distribution, P m v1 , x1 1 V 2

,X

, t 1 CP V

,X

,t ,

d /2

exp v1 u 2 /2 ,

N P V N , X N , t dV 4

i1

i V , v vi V

where (1/k B T ) , V is the system volume and d is the dimension. Although the correct distribution results from the semidetailed balance condition, the computation of the transport coefcients is greatly simplied if we impose detailed balance conditions on the collision operator. In the current context this is tantamount to 1 .
III. BOLTZMANN APPROXIMATION

(N) where X( N ) ( x1 , x2 , . . . , xN ), VN ( v1 , v2 , . . . , vN ), V 1 ,v 2 , . . . ,v N ) and, L L is the number of lattice (v nodes. The evolution described by Eqs. 1 and 2 preserves total energy and momentum of the particles in each cell, as can be seen from the identities,
i x V

vi

i x V

vi V V

In Sec. II we found that the stationary distribution for the one-particle velocity distribution was Maxwellian. In order to study the relaxation to equilibrium we assume that there are no correlations among the colliding particles in Eq. 3. In this case the probability distribution P is a product of identical one-particle probability distributions, P V N , X N , t

and

i1

P 1 vi , xi , t .

10

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

8607

In order to derive the Boltzmann equation we multiply Eq. 3 by i ( v vi ) ( x xi ) and subsequently integrate over vi and xi , where i 1, . . . , N . Focusing on x in cell and accounting for all possible ways of assigning n particles to the cell, we obtain C( f ( t ))

Equation 15 preserves expectation values of integrals of motion such as the density, momentum and energy. Denoting these quantities by J 1 ( v) 1, J 2 ( v) v, and J 3 ( v) 1 v 2 , respectively, we have 2

n1

N n

in1

i P 1( v d xi d v i , xi , t )

d vd xJ v f v, x, t 1

( ( xi 1/2) where v, V n 1

Vn

n d X n dV 11

d vd xJ v C f t

i1

i P 1 v i , xi , t ( v, V n ), d xi d v

i , , n 1

e n!

Vn

n d X n J v iF V n , X n , t dV 18

d vd xJ v f v, x, t ,

where we used the following identities:

i,

i x xi . V v v V 12

1 1, i i i , i v V v V i i i 2 v i 2, V v V i i

19a 19b 19c

In writing this equation we have used the fact that we may express P 1 ( vi , xi , t ) as the sum of two terms, P 1 vi , xi , t P 1 vi , xi , t xi 1/2 1/2 xi , 13 where is the Heaviside function. Particles with coordinates outside the cell do not contribute to the collision term and this is accounted for by the Heaviside function in the rst factor of Eq. 11. We use the fact that the prefactor in Eq. 11 may be written as

for the density, momentum, and energy conservation laws, respectively.

A. H theorem

in1

i P 1( v d xi d v i , xi , t )( ( xi 1/2)

i P 1( v i , xi , t )( ( xi 1/2) d xi d v

Nn

The evolution Eq. 3 involves a multiparticle collision term and it not immediately clear that an H theorem exists. We dene the H functional in terms of the reduced oneparticle distribution function, H t

Nn

14

d vd x f v, x, t ln f v, x, t e n!

where is the total particle number density in the WignerSeitz cell . In the large N limit we have ( N n )(1 ( / N )) N n ( N n / n !) e . Incorporating the factor of N into the denition f( v, x, t ) N P 1 ( v, x, t ), we obtain the Boltzmann equation, f v, x v, t 1 C f t , where C f t

, n N

d V n d X n n

i1

f vi , xi , t ln 20

i1

f vi , x i , t ,

15

n1

e n!

where the second equality arises from the representation of the system in terms of phase space cells and makes use of the resolution of identity 1 n 0 ( e x x n / n !). In the Appendix we show that H ( t ) decreases on each evolution step so that H t

Vn

n d X n F V n , Xn , t v, V n, dV 16

d xf v , x, t 1 ln f v , x, t 1 . dv

21

and we have dened n , X n , t F V i , xi , t . f v i1

n

For the xed momentum and energy expectation values the Maxwell distribution, 17 f m 1 N V 2 k BT

d /2

v u 2 2kBT ,

22

The prefactor in Eq. 16 shows that the cell particle number is Poisson distributed.

provides a lower bound to the H functional. This follows from the identity,

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

8608

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

d xd v f v, x ln f m v

d xd v f m v ln f m v ,

23

and the inequality, f v, x d xd v f v, x ln f m v N

Time evolution of a functional of the conserved quantities is governed by an operator given as an expansion in the small parameter ,

n Dn , t n0

32

f m v f v, x f v, x d xd v ln 0. N f m v f m v

24

and the gradient term in Eq. 28 is scaled by . The expansion of the collision operator in a series in is written formally as C f
n0

Equations 18 show that the expectation values of the density, momentum and energy are conserved during evolution under the Boltzmann equation; thus, completing the proof of the H theorem.

n Cn f .

33

B. ChapmanEnskog asymptotic expansion

To ensure the identity J C( f ) J f we constrain the average of each term of the series by J Cn ( f ) J f n . By expanding the evolution Eq. 28 in powers of we arrive at the following set of equations: C0 f C f 0 f0, C1 f f 1 D0 v f0, 34a 34b

We derive hydrodynamical equations by using an expansion of the reduced probability distribution in slowly varying density elds. This ChapmanEnskog procedure12,13 is based on the assumption that any relevant functional can be expanded into a series of partial derivatives of the conserved elds. After scaling x x and t t the expansion is ordered in powers of . We rewrite Eq. 15 in terms of the average cell probability distribution f and averaged collision operator C, f v, , t C f

1 C2 f f 2 D1 f 0 D0 v f 1 2 f0. D0 v 2 34c The solution of Eq. 34a yields a local Maxwellian distribution,

f 0

x 1/2

d x f v, x, t ,

25 26

1 2 k BT

d /2

e v u

2 /2k T B

35

x 1/2

d xC f .

The average of the local collision invariants over v commutes with the operator Di so that integration of Eq. 34b yields D 0 J v f 0 . 36

If f is a smooth function the following relation holds

x 1/2

d xe

t v

f v, x, t e

t v

f v, , t ,

27

The average of Eq. 34c gives the second order correction to the Euler equations,
1 f 1 C1 f , D 1 2 J D0 v

and the evolution of the averaged cell probability distribution is described by a Galilean invariant, spherically isotropic equation e t v f C f . 28

and after transformations Euler equations are dissipationless,

1 D 1 2 J v f 1 C1 f ,

37

It is further assumed that the reduced probability distribution function f is dened by the instantaneous spatial distribution of local collision invariants J , f v, , t f v, , t
n0

where we used the conditions on J f 1 and J C1 ( f ).

C. NavierStokes equation

n f n v, , t .

29

The density of a local collision invariant is given by the average value

, t

d vJ f , v, t J f , v, t ,

30

In this section we apply general formulas 34a 34c to a specic collision model. Regardless of the collision model, the equation for the zeroth expansion term has the same form and constitutes the Euler equations of compressible ow. Evaluation of the averages in Eq. 36 yields the following results for the evolution of the conserved quantities:

and to ensure uniqueness an additional requirement is imposed, f n , v, t 0 for all and n 0, J 31

t u 0, t u u u k B T 0,
1 1 t 2 u 2 C v T u 2 u 2 C p T 0,

38

where J is the set of the density, momentum, and energy dynamical variables introduced earlier.

where C v dk B /2 and C p C v k B . The subscripts and refer to the Cartesian coordinates and the Einstein summa-

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

8609

tion convention is used. We have dropped the subscript to avoid proliferation of notation. Algebraic manipulations transform the above system into a set of evolution equations for , v, and T ,

1 s 1 c c c 2 d

and s 2

c 2 Cp c , 2T

t u 0 , tu u u
kB T0 ,

39a 39b 39c

are eigenfunctions of the collision operator and through their eigenvalues we evaluate values of the viscosity and thermal conductivity coefcients, respectively. Consider s 1 . Using Galilean invariance of the Boltzmann equation, without loss of generality we can choose a stationary frame with u 0. In this case c v and write 1

C v tT u T k BT u 0 .

With the use of Eqs. 39a 39c we rewrite Eq. 34b in the following form: f f 1 f0 C1

,i1

h 1 v i

,i1

V V V V

v u u u

log f 0 log f 0 v u u
kB T

1 vi vi vi 2 . 2

43

log f 0 kB v T u T T u T Cv

v x v y vx vy and ( v x ) 2 By using the relations /2 we arrive at the following equa ( vy ) 2 valid for tion: 1

,i1

h 1 v i

i1

We substitute the explicit form of f 0 given by Eq. 35 and, after algebraic transformation, arrive at the following equation: C1 f f 1 f0

1 vi vi vi 2 2

2 V V V 2

44

c Cp c log T 2 k BT k B
2

1 1 c c c 2 u , k BT d

We note that the cross terms with different particle index i vanish upon integration so that we have the following expression for the eigenvalue problem: C1 s 1 f 0 s 1 f0

40

n1

n1 e 2n n!
45

where c v u. We dene the function h 1 by the relation h 1 f 0 f 1 . The collision operator C1 has the following form: f v C1

2 1 e s 1 f 0 r s 1 f 0. e 1 s 2 f 0 r s 2 f 0.

An analogous calculation for s 2 yields, C1 s 2 f 0 46

n1

e n!
n

d Vn

j1

v v j
n

n R n dV

n P m V n V n , V

i, h 1 v i1

41

where R ( n ) is dened by Eq. A2. Using the invariance of the Maxwell distribution with respect to the collision transformation we obtain the following relation: f v C1 e d V n ,n1 n !

Using these results in Eq. 37 we may compute the shear viscosity and thermal conductivity coefcients in terms of the damping factors r and r as,

k BT
and k BT

1r 1 e k BT , 2 1r 2 e 1

47

j1

v v j P m V
n

i1

i, h 1 v

42

1 e 1r k BT . 1r 1 e

48

i V vi V . where v Equations 42 and 40 constitute a linear integral equation for h 1 , which can be split into two equations, one that involves terms log T and the other that depends on gradients of velocity elds. Explicit calculations for two dimensions. For d 2 and /2, /2 we may show that the functions,

Evaluation of the averages in Eq. 37 yields the following expressions for the second order terms in the Chapman Enskog expansion: D1 0, D1 u ,
1 u 2 C v T u q , D1 2

49 50 51

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

8610

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

where is thermal conductivity coefcient and the irreversible contribution to the pressure tensor and the heat ux are given by the following expressions:
1 u u 2 u ,

52 53

q T .

In the , u, T set of variables the resulting hydrodynamic equations have the following standard form:

t u 0 ,
1 1 t u u u T ,

54a 54b
FIG. 1. Velocity distribution: Maxwell velocity distribution solid line; histogram of v x distribution computed in simulation ( x -step line.

1 1 C v t T u T k B T u u q . 54c Similar calculations can be carried for other collision model. For example, in three dimensions, averaging the set of all random rotations yields, 1

,i1

i h 1 v

i1

1 V V V 2 , 3

55

v v where we used the fact that (1/ ) 2 1 v for the set of random rotations. 3 Combining Eqs. 55 and 42 we arrive at the following expressions for the damping factor r and viscosity coefcient: r 1e , 56 57

k BT

1 e
2 e 1

IV. SIMULATIONS OF HYDRODYNAMIC FLOWS

FIG. 2. Poiseuille ow prole: The solid line is a quadratic t to the computed velocity distribution triangles. This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:

In this section we present results of simulations implementing an algorithm based on the present model. The simulations validate the assumptions used in the derivation of hydrodynamic equations and show that the model correctly reproduces uid ow behavior. Simulations have also been performed to determine the equilibrium velocity distribution and the rate of relaxation to the equilibrium see Fig. 1. The nature of the viscous dissipation was determined by simulating ow in the rectangular domain of width L 30, density 10 and temperature k B T 1.0. In the y-direction we impose bounce-back boundary conditions and the ow is driven by assigning Maxwell-distributed velocities with prole u x ( y ) 1.2( L y ) y / L 2 to particles in the region x 5. The results are presented in Fig. 2 and show that the prole is quadratic to a good approximation. The small deviations are due to energy dissipation along the channel and the difference between local equilibrium distribution and the stationary distribution in the Poiseuille ow. The model also reproduces hydrodynamic ows on larger scales. In Fig. 3 we present the results of simulations of ow past a disk. The system size is 360 120 and the disk diameter 2 a 32. The ow is induced by assigning

Maxwell-distributed velocities with temperature T 1 and velocity u x 0.5 to particles in the region x 8. Periodic boundary conditions are imposed in y direction. For particle density 10.0, the kinematic viscosity coefcient 47 is 0.055 and the corresponding Reynolds number Re 2 au x / 288. The ow was sampled on a set of predened points by averaging velocities of nearby particles with weights 1/( r 4 100), where r is the distance between a test point and a particle. After a transient time a ow is established at short distances from the object; however, a ow tail grows indenitely until it occupies the entire system length. In experiments on uid ows for these values of the Reynolds number the existence of von Karman streets is documented14 see Fig. 4.12.6 of Batchelor. The density of the uid is nearly uniform and velocity randomization at the boundary eliminates the feedback due to the periodic boundary conditions in the system. As another illustration of the method in Fig. 4 we present the results of simulations of the stages in the development of the boundary layer at the rear of a disk suddenly set in motion. The system size is 400 400 and the disk diameter 2 a 100. The ow velocity at x 0, the density of the system and the Reynolds number are u x 0.4, 20.0,

14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

A. Malevanets and R. Kapral

8611

FIG. 3. Vortex shedding in a stationary ow past circular cylinder. The dark line segments represent tangents to the velocity eld at their median point. The length of a segment is proportional to the magnitude of the velocity.

FIG. 4. Development of the boundary layer in ow past disk set in motion.

and Re 1520, respectively. The system setup is the same as in the simulations of von Karman streets. The initially symmetric ow separates from the disk and a backow at the the far end of the disk develops. At the contact line between the normal ow and the oppositely-directed backow, a system of vortices appears which later expands into the full-scale boundary layer cf. Fig. 5.9.3. in Ref. 14.
V. CONCLUSION

The multiparticle-collision stochastic model combines advantageous features of both Direct Simulation Monte Carlo and lattice gas methods for simulations of uid ows. It differs from DSMC and its variants in the nature of the collision rule. In DSMC collisions in a cell are carried out sequentially on randomly chosen pairs of particles in the cell. This rule leads to exponentially distributed collision times, and the possibility of multiple collisions involving the same particle affects the efciency of the algorithm. In general, by allowing only pairwise collisions DSMC imposes a high lower bound on the accessible values of the kinematic viscosity coefcient, which may be important if higher Reynolds number ows are desired. Since it is a particle-based scheme the multiparticle-collision model does not suffer from numerical instabilities. The velocity distribution is Maxwellian and the hydrodynamic equations are isotropic. Because of these features it can serve as mesoscopic model for solvent dynamics which can be coupled to a full molecular dynamics treatment of solute degrees of freedom. The examples of simulations of uid ow around large ob-

jects presented in Sec. IV are relevant for studies of the interactions among large colloidal particles in solution. The method has also been used to study the coupling among small effective monomer units in a polymer through solvent dynamics.15 This is an example of the coupling of the full molecular dynamics of a polymer chain to the mesoscale dynamics of the solvent. Consequently one may study the validity of existing polymer models and the inuence of uid ow on polymer conformational dynamics. The collision model is easily extended to treat more complex situations such as the inclusion of interactions that give rise to phase segregation or reactive hydrodynamic ows. To simulate this effect the motion of a small number of complex solute molecules is coupled to the solvent through interaction potentials that affect motion of the ideal particles during the streaming step while preserving the integrals of motion and the phase space volume. The model also may be used to probe the dynamics on mesoscopic length scales where uctuations may play a role in determining the character of the phenomena. The emphasis in this paper was both on the formulation of the multiparticle collision model and the development of its theoretical underpinnings, namely, the existence of an H theorem and the derivation of hydrodynamic equations. The simulations have served to verify the theoretical predictions and demonstrate that the model can be used to simulate uid ows for applications to the dynamics of complex uids.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

8612

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999

n

A. Malevanets and R. Kapral

ACKNOWLEDGMENT

This work was supported in part by a grant from the Natural Sciences and Engineering Research Council of Canada.
APPENDIX: PROOF OF H-THEOREM

A B

i1

n f i ,

n , Xn , t f n V Z
i1

n f i

In this Appendix we show that H ( t ) dened in Eq. 21 decreases on each evolution step. The proof of the H-theorem relies on the following convexity inequality: A s B s ln B s A s B s s s

in the inequality A1 we obtain 1 Z

ln
s

A s B s , A1

n d Xn n , X n , t ln dV f n V n

n , X n , t f n V Z
i1

n f i

where A is normalized by s A ( s ) 1. Introducing the quantity R ( n ) dened as n R n V n , V 1

n d X n dV n

n , X n , t f n V ln Z

Vn

n d X n dV A9

i1

i , V v vi V A2
n

n , X n , t f n V 0. Z

whose integral over V H t

(n)

is unity, we may rewrite Eq. 20 as d V n d X n n

, n N

e n!
n

The last equality follows from the fact that the argument of the logarithm is unity in view of the denition of Z. From this inequality we deduce that

i1

f vi , xi , t

ln

i1

f vi , xi , t

Vn

n d X n n , X n , t ln n , X n , t dV f n V f n V

n R n Vn , V n . A3 dV

n d X n n , X n , t ln Z dV f n V n

i1

n f i .

A10

In each term in the above sum we exchange the order of the ( n ) and V( n ) integrations. Use of the convexity inequality V A1 leads to the following relation:

Finally, using the identity,

n d V n R n V n , V f n
V

i1

f vi , xi , t ln
V

i1

i1

Vn

n 1/n n i xi Z dv f f i vi , x i , t i vi , xi , t ln Z n

f vi , xi , t A4
n

, X , t ln f

,X ,t ,

n d X n n , X n , t ln Z dV f n V n

i1

n f i vi , xi , t ,

where n , X n , t f n V Hence, we may write H t

we establish the following relation from Eq. A6:

n d Vn R n V n , V

f vi , xi , t . i1

H t

i , , n N1

1 e n n!

n d xZ 1/n dv f i A

A5
B

n , x, t ln Z 1/n v f i v, x, t

, n N

e n!

n d X n n , X n , t dV f n V n A6

d xC f ln C f dv A11

d xf v , x, t 1 ln f v , x, t 1 . dv

n , X n , t . ln f n V If we dene the quantities Z and f ( n ) as Z

Vn

n d X n n , X n , t n dV f n V , 1 Z

A7

In the above expression the inequality A arises from applying inequality A1 with A given by the Poisson distribution 1 /( n 1)!) and equality B follows from A ( n ) ( e n the invariance of the integral with respect to translations by the streaming transformation. Consequently, the value of the H functional at time t 1 does not exceed its value at time t.
1

Chemical Waves and Patterns, edited by R. Kapral and K. Showalter Kluwer, Dordrecht, 1994. 2 A. Rovinsky and M. Menzinger, Phys. Rev. Lett. 70, 778 1993; 72, 2017 n n V , X , t , A8 1994. 3 J. Wells, D. Janecky, and B. Travis, Physica D 47, 115 1991. where the hat over the integration variables in denition A8 4 B. Dunweg and K. Kremer, J. Chem. Phys. 99, 6983 1993. 5 indicates that these variables should be omitted in the inteRecent Developments in Theoretical Studies of Proteins, edited by R. Elber World Scientic, London, 1996. gration, and let This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
V[ n 1]

n f i vi , xi , t

1 d x1 d i n d xn dv v d xi d v f n

14.139.227.196 On: Fri, 10 Jan 2014 09:10:56

J. Chem. Phys., Vol. 110, No. 17, 1 May 1999 Microscopic Simulations of Complex Flows, edited by M. Mareschal Plenum, New York, 1990; Y. Kong, C. W. Manke, W. G. Madden, and A. G. Schlijper, J. Chem. Phys. 107, 592 1997. 7 D. L. Ermak and J. A. McCammon, J. Chem. Phys. 69, 1352 1978; X. yan Chang and K. F. Freed, ibid. 99, 8016 1993. 8 G. A. Bird, Molecular Gas Dynamics, Clarendon, Oxford, 1976; Comput. Math. Appl. 35, 1 1998. 9 For a review, see, S. Chen and G. Doolen, Annu. Rev. Fluid Mech. 30, 329 1998. 10 D. H. Rothman and S. Zaleski, Lattice-Gas Cellular Automata: Simple
6

A. Malevanets and R. Kapral

8613

Models of Complex Hydrodynamics Cambridge University Press, Cambridge, 1997. 11 A stochastic streaming rule is described in A. Malevanets and R. Kapral, Europhys. Lett. 44, 552 1998. 12 S. Chapman and T. G. Cowling, The Mathematical Theory of NonUniform Gases Cambridge University Press, Cambridge, 1970. 13 C. Cercignani, The Boltzmann Equation and its Applications Springer, New York, 1988. 14 G. K. Batchelor, An Introduction to Fluid Dynamics Cambridge University Press, Cambridge, 1967. 15 A. Malevanets and J. M. Yeomans in preparation.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 14.139.227.196 On: Fri, 10 Jan 2014 09:10:56