• Aucun résultat trouvé

Surface tension and interface fluctuations in immiscible lattice gases

N/A
N/A
Protected

Academic year: 2021

Partager "Surface tension and interface fluctuations in immiscible lattice gases"

Copied!
19
0
0

Texte intégral

(1)

HAL Id: jpa-00246889

https://hal.archives-ouvertes.fr/jpa-00246889

Submitted on 1 Jan 1994

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Surface tension and interface fluctuations in immiscible lattice gases

Christopher Adler, Dominique d’Humières, Daniel Rothman

To cite this version:

Christopher Adler, Dominique d’Humières, Daniel Rothman. Surface tension and interface fluctu- ations in immiscible lattice gases. Journal de Physique I, EDP Sciences, 1994, 4 (1), pp.29-46.

�10.1051/jp1:1994119�. �jpa-00246889�

(2)

Classification Pllysics Abstracts

68.10C 05.70F 47.55K

Surface tension and interface fluctuations in inuniscible lattice gases

Christopher

Adler

(~), Dominique

d'Humières

(~)

and Daniel H. Rothman (~>*)

(~ Department of Earth, Atmospheric, and Planetary Sciences, Massachusetts Institute of Tech-

nology, Cambridge, MA 02139, U-S-A-

(~) Laboratoire de Physique Statistique, C.N.R.S.,

École

Normale Supérieure, 24 rue Lhornond, 75005 Paris, France

(Recel

vert 3 August 1993, accepted 8 October

1993)

Résumé. partir de l'approximation de Boltzrnann, nous calculons la tension superficielle

en fonction de la densité pour un modèle de gaz sur réseau décrivant à deux dimensions des fluides irnrniscibles avec conservation de l'impulsion. Les résultats des calculs, qui prédisent que

la tension de surface disparaît en dessous d'une densité critique, sont comparés aux mesures

faites à partir de la simulation d'interfaces planes et de bulles; l'accord entre la théorie et les données est qualitativement bon. Les fluctuations à l'équilibre d'une interface plane sont

aussi étudiées. Empiriquement, les fluctuations observées se comportent de manière classique,

décroissant

comme l'inverse du carré du vecteur d'onde et suivant qualitativement une relation

d'équipartition de l'énergie de surface.

Abstract. Using a Boltzmann approximation, we calculate the surface tension as a function of population density in a momentum-conserving lattice-gas model of immiscible fluids in two

dimensions. The calculation, which predicts that the surface tension vanishes below a critical

density, is compared to measurements made from simulations of flat interfaces and bubbles;

the fit of theory to data is qualitatively good. Equilibrium fluctuations of flat interfaces

are

also studied. The fluctuations

are empincally observed to be classical, decaying like the inverse square of wavenumber and obeymg quafitatively the equipartition of surface energy.

1. Introduction.

Since trie introduction of

lattice-gas

models of

hydrodynamics il,

2], considerable effort bas been devoted to trie

adaptation

of these models for trie simulation of interfaces in

hydrodynamic

(*) Permanent address: Department of Earth, Atmosphenc, and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139, U-S-A-

(3)

flows,

both for the case of interfaces

separating

different

species

of

partides

[3-6] and interfaces

separating

different

thermodynamic phases

of the same

species

[7-10]

Although

ail of these models bave been shown

empirically

to

reproduce

the

physics

of interfacial tension, and various arguments have been

given

to support the presence of surface tension in trie

two-species

models [5,

11],

no theoretical

predictions

of trie surface tension bave yet been made. Here we present

such a

study

for a

particular example

[3] of a

two-species,

or immiscibie

?attice-gas,

model.

A

study

of surface tension in

lattice-gas

models of

hydrodynamics

is

interesting

for several

reasons. First, from a

practical point

of view, it

gives

one a better

understanding

of llow this

parameter may be varied for

applications

and

provides

a basis for

understanding

any

anisotropy

in interfacial tension.

Second,

from the

viewpoint

of

theory,

it

gives

additional

understanding

of

phase

transitions in models of this type. As bas been

previously

discussed

[11-13]

,

immiscible lattice gases

(ILG'S)

exhibit a

phase

transition from a

mixed, one-phase

state, to an

unmixed, two-phase

state. This transition bas been understood in trie past via a calculation of trie

diffusivity [11],

but never

directly

in terms of surface tension. Our

prediction

of surface tension

as a function of particle

density

shows

dearly

a

phase

transition from

vanishing

to finite surface tension at a critical

density

de, and confirms

quahtatively

the previous estimate of de made from the calculation of

diffusivity.

To

predict

the surface tension m a lattice gas, we consider

planar

interfaces in a two- dimensional fluid, onented either

parallel

or

thirty degrees

from a lattice direction. The in- terface is

given

finite width, and is described

dynamically by coupled

Boltzmann

equations.

The

magnitude

of the surface tension is then derived from the

stationary

solution to these

coupled

Boltzmann

equations.

As is

typical

in trie estimation of transport coefficients in lat- tice gases, trie Boltzmann

approximation

of no correlations aids trie calculation

considerably.

However,

because surface tension arises in part from trie correlations

themselves,

we obtain

only

a

qualitative,

rather thon

quantitative,

match between trie surface tension obtained via

the Boltzmann

approximation

and that observed in simulations.

As a kind of

application

of what we bave learned about

lattice-gas

surface tension from our Boltzmann

approximation

and

empirical

measurements, we condude our

study

with a brief

investigation

of interface fluctuations.

(This subject

has been

previously

discussed in Ref. [14]

in trie context of a

lattice-gas

model with

explicitly

constructed interfaces. We find that trie

equilibrium

fluctuations of ILG interfaces

decay

like the inverse of wavenumber

squared,

and

are thus,

by

virtue of an

assumption

of

equipartition

of surface energy, in

qualitative

agreement with dassical

theory (e.g.,

[15]

).

In what

follows,

we first

precisely

define a

microdynamical

equation for trie evolution of the ILG. We then

approximate

the

dynamics

of this

equation by neglecting

correlations

(the

Boltzmann

approximation),

and seek the

stationary

solution of a set of these

equations

cou-

pled

across an interface. The

stationary

solution is obtained

numerically,

and is

compared

to

measurements obtained

directly

from ILG

simulations,

from both bubbles and

plane

interfaces.

The

study

of interface fluctuations then follows.

2.

Microdynamics.

In the usual manner, the two-dimensional lattice gas

il,

2] is defined on a

triangular

lattice with vertices

pointing

in directions

given by

trie unit vectors

c~ =

(cos 27ri16,sin 27ri16),

= 1,.

,

6.

(1)

Thus the lattice is oriented with one of the axes

parallel

to the horizontal direction, as shown in

figure

1.

(4)

~ ~

~2 ~l

~ ~

c~ c~

~ ~

~4 ~5

Fig.

1. Correspondence of trie vectors c, to trie vertices on the hexagonal lattice.

In trie immiscible lattice gas [3], each site on this two-dimensional lattice contains trie 14-bit state defined

by

s =

(r, b)

=

(ro,

ri r6 bo bi b6

(2

where the bits ri and b~ represent the presence

(1)

or absence

(0)

of a "red"

partide

and "blue"

particle, respectively,

moving with

velocity

one in the direction c~, <1 < 6, or with

velocity

zero in the direction c0 = 0 when

= 0. Since r~ and b~ cannot both

equal

one, we may define the additional Boolean variable

n~ = ri + b~

(3)

to mdicate the presence of either a red or blue

particle moving

with

velocity

c~.

The collisions of

particles

at time t and their motion from sites located at

position

x to sites at x + c~ are

expressed symbolically by

the

coupled microdynarnical equations

r~(x

+ c~,t +

ii

=

r((x,t) (4)

b~(x

+ ci,t +

ii

=

b[(x,t), (Si

where the result of a

collision,

ri ~

r[,

b~ - b[, is

given by rl(x,t)

=

C[(S(x,t), f#) (6)

bj(x,t)

=

c)js(x,t), f~). ii)

The collision operator C( E

(0,1)

takes as

input

the 14-bit state

six)

and the

discrete,

scalar-

valued,

color-field

angle f~

and

gives

as output the state of the ith element of species

j

E

(r, b)

after collisions have occurred. Trie discrete color-field

angle f~

enters the

dynamics

in the

followmg

way. Trie collision operators

C[

and

Cl

are constructed such that collisions maximize trie flux of

color,

6

qlr', b')

=

LCz(rl bl), (8)

z=i

in trie direction of trie

gradient

f of

color,

or, m other

words,

such that q f is maximized.

(If

the choice of r' and b' is

non-unique,

then trie result of a collision is chosen

randomly

from this

set.)

The color

gradient

f is

given by (within

an irrelevant

constant)

6

f(X)

"

~

C~#j,

(9)

i=1

(5)

where the relative color

density

#~ at the site located at

position

x + c~ is

~ =

~ ÉÎrJ(x +c~)

b J=o

J x + c~)j

(io~

=

fi4(S(x+Cz)). (ii)

Here the

color-counting look-up

table fi4 has been

implicitly

defined. Since #~ E

(-7,

-6,

,

6,

7),

there are 15~

possible

color distributions

(#~ )~=i,...,6

Because 1) many of these distributions

are

redundant; 2) only

trie orientation, not the

magnitude,

off determines the outcome of a

collision;

and

3)

small differences m orientation are

insignificant

for trie creation of surface tension, f is

typically

discretized further to a set of discrete field

angles,

denoted

by

trie

angle

code

f~, uniformly

distributed from 0 to 2x.

Here,

as elsewhere [3], we allow for 36

possible angles,

and therefore 37

possible

values of

f~. (Trie

extra

angle

code allows for trie case f

=

0.)

More severe

discretizations, however,

are

possible.

Trie transformation of trie caler distribution

(#~)

to trie discrete

angle f~

is

symbofically represented by

trie operator

(or look-up table)

T such that

y(ji~j)

=

f~. (12)

We note that other forms of trie immiscible lattice gas do not approximate f

by equations (11)

and

(12),

but instead use colored "holes" in addition to colored

partides,

and obtain an estimate of trie

gradient

from trie information

propagated by

both trie

partides

and hales [4-6].

3. Boltzmann

approximation.

To obtain an estimate of surface tension, we first need to express

equations (4)

and

(5)

in terms of the evolution of a

probability

field. In an ensemble of

systems prepared

with different initial

conditions but

subjected

to the same extemal constraints, we define the average

quantities

R~ =

jr~i,

B~ = jb~i,

j13)

and also

N~ = (n~) = R~ +

B~, (14)

which are,

respectively,

the

probability

of

observing

r~

= 1, b~ = 1, and n~

= 1. We then

make the assumption that ail

partides

are uncorrelated with trie others. This is the Boltzmann approximation of molecular

chaos,

well-founded for

estimating lattice-gas

transport coefficients such as viscosity

[16],

but of

questionable validity

for the estimation of surface tension smce trie surface tension itself arises in part from correlations. Nevertheless such an

approximation

allows us to make progress, and serves as a useful reference for better

approximations

that may follow.

Thus, specifically,

for the evolution of the red

partides

we write

~~~~

~ ~~~ ~ ~~ =

sÎ, ~~~~~'~"

~*~~'~~~~>

t)Q(f#;

x,

t),

~~~~

and for evolution of the blue

particles

we have

B~(x

+ c~, t +

1)

=

~j b[A(s, s', f~ )P(s;

x,

t)Q(f*

ix,

t). (16)

s,s>, f.

Here the sums are taken over ail

possible

states s that may enter a collision, ail

possible

states s' that may result from a

collision,

and all

possible

discrete field

angles f~.

The factor

A(s, s',

f~

(6)

represents the

probability

of

obtaining

state s' when state s enters a collision. The

probability

that state s

actually

enters the collision at time t at the site located at position x is

given by

6

P(s;

x,

t)

=

fl Rl'Bl'(1 N~)~~~'~~' (17)

1=0

The

probability

that the discrete field

angle

is

f~

is

6

Q(f#; x)

=

L fl W(#z;x)

,

(18)

iw,iT(iw,ii=f. i=1

where the relative color

density

#~ was defined in

equation (10).

Here the sum is taken over all

possible

color distributions

(#~)

that

correspond

to

f~,

and the

product

is taken over the

probabilities W(#~)

of

observing

the relative color

density

#~ at the ith

neighbor. Specifically, W(#~)

is

given by

tlle sum of tlle

probabilities

of ail states that

yield

the caler

density

#~:

W(#~; x)

=

~j P(s;

x + c~).

(19)

S=fi4(S)=4,

4. Surface tension calculation.

To calculate the surface tension, we note that in the

vicinity

of an interface the pressure is

locally

anisotropic, since the pressure in the direction

parallel

to the interface is reduced

by

the tension on the interface itself. For the case of a flat interface

perpendicular

to the

z-axis,

the surface tension a is given

by

the

integral

over z of the difference between the component FN of pressure normal to the interface and the component PT transverse to the interface

[17]:

a =

/°° p~(z) p~(z)dz. (20)

In mechanical

equilibrium

one has

PN(z)

=

P,

the

(isotropic)

pressure far from the interface;

at this juncture however we retain the

z-dependence

for

generality. Equation (20) gives

the surface tension as a function of the pressure. The pressure in lattice gases is

given

in tensorial form

by

[2]

6

~afl

"

~

CmC~pfil~

(21)

1=0

where

a and

fl

are tensor indices. Prediction of the surface tension is thus a

problem

of

predicting

the distribution of the

populations

N~ near an interface.

To determine surface tension m the

ILG,

we

study

two cases: one in which the interface is

parallel

to a lattice direction

(say,

c6)> and the other in which the

perpendicular

to the interface

is

parallel

to a lattice direction. See

figure

2. The former case is called the

"0-degree interface,"

while the latter is called the

"30-degree interface,"

where the number of

degrees

refers to the smallest

angle

the interface makes with a lattice

axis.

Below,

we describe the surface-tension

calculation for the

0-degree

interface for the

simplest

case m which it is

composed

of

only

two

layers.

An extension of this calculation to an interface which is 4

layers

thick is given in

Appendix

A. The

equations

for the

30-degree

interface are

given

in

Appendix

B.

As shown m

figure

2a, the center of the

0-degree

interface is taken to be between and

parallel

to two

(horizontal)

lattice hnes. The upper line is labeled yi and the lower line is labeled y-1.

(7)

interface

Yi

4- interface x~

Y-i

(ai

jbi Fig.

2. a) The 0-degree interface. b) The 30-degree interface.

We assume an average of id

particles

per site for from the

interface;

in

equilibrium,

therefore,

we must have

partides arriving

at interface sites with

probability d, independent

of time. This fixes the

boundary

conditions

N~(yi, t)

= d, = 4, 5 Vi

(22)

at all sites in

layer

yi and

N~(y-i t)

= d, = 1,2 Vi

(23)

at all sites in

layer

y-i These

boundary

conditions

require

that the

populations

N~ be symmet-

nc across the the center of interface after rotation

through

180

degrees.

Thus for the moving

partides

N~(yi,t)

=

N~+3(Y-i>t),

= 1,.

,

6,

(24)

where the circular shift 1+ 3

=

j

such that cj = -c~,

j

= 1,.

.,

6,

while for the rest

partiales No(Yi>t)

=

No(Y-i>t). (25)

Thus m this two

layer

case, the

dynamics

of the interface can be

completely

determmed

by solving only

for the

populations

in

layer

yi Within this

layer, requirements

of symmetry and

mechanical

stability

further reduce the

remaining

rive

populations

to

only

two

mdependent populations. Specifically,

mechanical

stability requires

that the pressure be

divergence-free,

and therefore that FN

" P

= 3d. This

gives, by

virtue of the

boundary

condition

(22),

Ni"N2"N4"Ns"d.

(26)

(8)

In addition, since there is no current

parallel

to the interface

(or, equivalently, by

symmetry with respect to the

perpendicular

to the

interface),

we have

N~ =

N6 (27)

Thus the two free

population

variables are the

rest-partide population,

No and one of the

laterally moving populations,

say

N3.

These

populations

evolve

according

to

N~(yi>t +1)

=

N((yi,t),

1= 0,3,6,

(28)

where the

post-collision

state is denoted

by

in which we have also used

n[

=

r[

+ b[.

The evolution of color must also be

specified.

We work in terms of the red concentration [ =

R~/N~.

In addition to the symmetry

given by equations (24)

and

(25),

we also have

Ù~(yl>t)

" Î

Ùj+3(Y-1>t),

# Î,...,6, (3Ù)

and

Ùo(Yi,t)

"

Ùo(Y-i,t). (31)

Additional

symmetries

and

stationarity

of the interface allow the seven concentration vari- ables in layer vi to be reduced to three

independent

variables. First, we note that symmetry with respect to the

perpendicular

to the interface

gives

63 = 66, ôi

= 62, 64

= ôs

(32)

Together

with the concentration ôo for the

rest-partiale population,

the first of these three pairs evolve

according

to

ô~(gi,t

+ 1)

=

à((yi>t),

= 0, 3,6

(33)

where we have used

à[

=

R[/N].

The evolution of the second

pair

of concentrations is deter- mined

by partides

that cross the

interface; using equation (30),

we obtain

ô~(yi,t +1)

=

à[~~(yi>t),

= 1,2.

(34)

Since the

stationarity

of the interface requires that no net concentration crosses it, in

steady

state we must have

~~~~~~ ~~~~~~

Î'

~~~~

and

therefore, by equation (34),

~~~~~~ ~~~~~~

Î'

~~~~

It remains

only

to

specify

the red concentration coming m from afar. Since m

equilibrium

the concentration that leaves the interface must be

equal

to the concentration that enters

it,

we

set

ô~(yi,t

+ 1) =

à(~~(yi>t),

=

4,

5.

(37)

Thus two of the three free concentration variables may be taken to be ôo and 63, which enter via equation

(33),

while the third may be taken to be 64> which enfers equation

(37)

above.

(9)

To

complete

the

specification

of the

problem

we need an

expression

for

Q( f~),

the

probability

of trie discrete field

angle f~.

From

equation (18),

one sees tllat all tllat is

required

is

required

is

knowledge

of

W(#~),

for

=

1,...,6.

Tllese

quantities

may each be obtained from the

symmetries

and

boundary

conditions of the

problem. Noting

that

W(#o)

is the

probability

distribution for relative color

density

for the interface site in

loyer

yi, one finds

w(i~)

=

w(i~)

=

w(io) (38)

for the

neighboring

sites in

loyer

yi> and

Wl#4)

=

Wl#s)

=

Wl-#o) (39)

for the

neighbonng

sites across the interface in

layer

y-i For the sites on the

boundary,

one has,

assuming

that the chosen site in

layer

yi is at position xi,

N~

(xi

+ cj,

t)

=

d,

=

0,.

.,

6,

j

= 1,2

(40)

for the

populations,

and

ô~(xi

+ cj,

t)

=

ôj +3(x, t),

= 0,..., 6,

j

= 1,2

(41)

for the concentrations.

W(#i

and

W(#2)

may then be calculated

directly

from

equations (17)

and

(19).

The

dynamics

of the

two-layer 0-degree

interface is thus

fully specified by equation (28)

for the two free

populations

No and

N3, equations (33)

and

(37)

for the three free concentrations

ôo, 63,

and 64, and equations

(38)-(41)

for the determination of the color field

angle.

Note that the

remaining populations

and concentrations m the two

layers,

23 variables m

all,

may eacll be obtained

directly

from trie considerations of symmetry,

stability,

and

stationanty

detailed

above.

To determine trie surface

tension,

one may

simply

seek trie fixed point, or

steady

state, of trie à-variable map given

by

trie free

population

and concentration

equations. However,

since trie

use of collision tables mandates that the calculation be

performed numerically (to

evaluate the

sums and

products

m

Eqs. (15)-(19)),

we instead choose to seek the

steady

state of trie entire 14-variable system

(1.e.,

one

layer),

and then check to see that trie relevant

symmetries

are satisfied

by

trie calculation.

Specifically,

after

mitializing

trie calculation with N~

=

d, [

= 1

for 1

= 0,.

,

6, we

successively

evaluate trie evolution

equations

until trie time t when

É(iNi(Yi,t) Ni(Yi,t -1)i~

+

iÙi(Yi,t) Ùi(Yi,t -1)i~)

<

E,

(42)

where we choose e

=

10~~°

The surface tension may then be calculated from

equations (20)

and

(21).

In trie present

case, we bave

~~ ~~~

~

~~

~

~~

~

~~~

~~~~

and

~~

~~

~

~~

~

Î~~

~

~~

~

~~

~

~~~

~~~~

where

NI

denotes the

steady-state post-collision populations.

The surface tension is then

given by

°

~ÎÎ~~~ ~~~

~~~~

where the factor of 2 comes from the symmetry of the interface and tlle factor of

và/2

anses

from the lattice geometry.

(10)

5. Results: Boltzmann

theory

vs. simulation.

We first

provide

a discussion of the results of the theoretical calculation.

Then,

after

providing

an

empirical

demonstration of

Laplace's law,

we compare the theoretical

predictions

with results from simulations. The simulation results are obtained from both bubbles and flat

interfaces.

5.1 THEORETICAL cuRvEs.

Figure

3 summarizes the results from the theoretical calcula- tion of surface tension. There are two sets of curves. The first set

(smooth curves)

is from the

prediction

of a for the

0-degree interface,

both for two

layers

and four

layers.

One sees that

0.S ;.---.-.,

,/ 30,J. ", 30,6." ,/"' ",".

0A

g

0.3 e

)

©

~ 0.2

W

o-1

~~0.0 0.2 0.4 0.6 0.8 1-o

reduced density

Fig.

3. Theoretical prediction of surface tension as a function of reduced density d in the irn- miscible lattice gas. Smooth curves: Boltzmann approximation for 0-degree, 2-layer interface (0, 2)

and 0-degree, 4-layer interface (0, 4). Dotted curves: Boltzmann approximation for 30-degree, 4-loyer

interface

(30, 4)

and 30-degree, 6-loyer interface (30, 6). In each case, the thicker interface results in the greater surface tension.

the thicker interface

always yields

a

larger

surface tension,

though

the two curves

effectively

converge for d > o-1- Thicker interfaces

yield

greater surface tension because the difference between P and PT is not forced to go to zero as

rapidly.

At

high densities, however,

there is

virtually

no

mixing

of red and

blue,

and thus the interface

dynamics

is

adequately

descnbed

by just

two

layers.

One sees

essentially

the same behavior for the second set of curves, which represent

predic-

tions for

30-degree interfaces,

for four

layers

and six

layers, respectively.

We note,

however,

that the

o-degree

surface tension and the

30-degree

surface tension are in

general

not

equal,

thus showing that the surface tension is

anisotropic.

The gross features of the curves that

they

vanish below d

=

de

m

0.2,

rise to a

single

maximum, and then

dropoff

towards zero at d

= I.o is

perhaps

of the greatest interest. The appearance of the critical

density

de is the signature of the ILG'S

phase

transition from the mixed state

(d

<

de)

to the

unmixed, two-phase

state

(d

>

de). Broadly speaking,

there is no surface tension for d < de because the discrete model offers insufficient

degrees

of freedom to

(11)

create a stable interface at low densities.

(That

the theoretical estimate of de decreases as the interface becomes thicker results from the additional

degrees

of freedom available to form a thicker

interface.)

The situation at d

= I.o is somewhat

analogous:

in this case, the fact that N~ = I.o for all offers no

possibility

for an anisotropic pressure. The theoretical maximum

surface tension at d m o-S thus occurs at a

density

that

optimally provides

an anisotropic pressure while

minimizing

any

mixing

between the

phases.

We note that the

phase

transition at de m 0.2 was

previously

located in two different ways [11].

First,

a Boltzmann calculation and

empirical

measurement of the ILG diffusion coefficient

showed that the

diffusivity changes sign

from

positive

to

negative

as the

density

is increased past de.

Second,

simulations initialized as

homogeneous

mixtures of concentration à

= o-S were

observed to

spontaneously undergo phase separation

above de but not below.

5.2 LAPLAcE'S LAW. Before companng the results from the Boltzmann

approximation

and

simulation, we first

provide

an

empirical

demonstration that surface tension exists.

Figure

4

is a

graph

of the difference between the pressure inside a

bubble, fin,

and the pressure outside

a bubble, P~ut> as a function of the inverse of the bubble's radius R, for

density

d

= 0.7.

Laplace's

law [17] states that

fin P~ut " °

(46)

R

m two dimensions. The pressure is calculated from

equation (21), which, sufficiently

far from the

interface, gives

P

=

Papô~p

=

3p/7,

where p is the average number of partides

(of

either

color)

per site. To obtain fin and P~ut> we locate the center of the bubble

by computing

the average location of each

partide

of the bubble's

color,

and then define fin to be the average

pressure within 0.7Ro of the bubble's center, and P~ut to be the average pressure at sites further than 1.3Ro from the bubble's center, where

Ro

is the

input

radius. Bubbles with radii rangmg from Ro

" 4 to Ro " 64 were

placed

in boxes with

penodic boundary

conditions and lmear

O.io

o.08

w

(

0.06

~ i

ce E .Oz

0.00

0.00 1/R

Fig.

4. Venfication of Laplace's law m trie immiscible lattice gas. Bubble radii R range from

4 to 64 lattice umts. An estimate of surface tension is given by the slope of the best fitting fine thon passes through trie engin. Error bars

(based

on bubble size and number of lime steps used for averagmg) mcrease finearly with 1/R, but are always less than 10~~, and thus srnaller than the size of the symbols.

(12)

dimension of at least

4Ro.

After

being

allowed 1000 time steps to

relax,

AP

= fin P~ut was

averaged

over the

following

2 x 10~ time steps.

(Two independent

measurements were

averaged

for the cases Ro

" 4 and RD "

5.) By assuming

that AP is due

entirely

to an increase of the

pressure inside the

bubble,

we calculate the

equilibrium

radius R of the bubble from

~~

Po ÎAP~~'

~~~~

where Po

" 3d.

(Though

more

appropriate, using

R instead of Ro

changes

trie estimate of

the surface tension

by

less thon 1Yo.)

Figure

4 shows excellent agreement witll tlle

Laplace

law of

equation (46).

Here the

best-fitting

line

passing through

the

origin gives

the

empirical

estimate a

= o.382 + 0.col for the surface

tension(~).

Note that one expects

departures

from the

Laplace

law as R - o, in part because trie finite-state automaton cannot support

large

pressure contrasts.

5.3 COMPARISON OF SIMULATION AND THEORY.

Figure

5 compares trie

0-degree, 2-loyer

Boltzmann

approximation

and the

30-degree, 4-layer

Boltzmann

approximation

to two types of

empirical

measures of surface tension. The first

empincal

measure,

given by

the

circles,

comes

from measurements made from bubble tests as described above.

(For

densities d

#

o-1the fit

to

Laplace's

law was

performed

with 4 points,

using

radii RD " 8, 12, 24, and

64).

The second

set of measurements are made from flat

interfaces, by empincally computing

trie

integral

given

o.5

~'~

;"

~

à,

C .~

° O

0.3

~©

)

O.Z

ce

o-1 ".

~'~0.0 0.2 0.4 0.6 0.8 1.0

reduced denslty

Fig.

5. Companson of Boltzmann approximations for 2-layer, 0-degree interface

(smooth curve)

and 4-layer, 30-degree interface

(dotted curve)

with three empincal measures of surface tension. Cir- des: bubble tests. Squares: measurements from flan interfaces oriented at 0 degrees. Triangles:

measurements from flan interfaces oriented at 30 degrees. Errer bars

are smaller than the

size of the

symbols.

(~) Details of this measurement, and therefore the results, dioEer from the bubble tests descnbed in reference [3].

(13)

o.5

o

Îf 0.5 OE

<

Oé 2.5

~

ogl

0

cycles/latt>ce unit)

Fig.

6. Comparison of interface fluctuations ([Aq[~) obtained from simulations

(circles)

and those

predicted from classical theory

(straight

fine, given by Eq.

(54)).

by

equation

(20),

which now takes the form

a =

Ii

~

L

~~

(CII

lii

ni(zi>

zii

(48)

~+<~ <_

~i~À<~

Here c~i and cqj are the components of c~

perpendicular

and

parallel

to the

interface,

respec-

tively, (.)

indicates

averaging

over time, Li and Ljj represent the

length,

m lattice units, of the lattice m the directions

perpendicular

and

parallel

to the

interface, respectively,

and xi and zjj are the

physical

coordinates of lattice sites.

Specifically,

a computational box of size Li = 128ai

by

Ljj =

64ajj

was mitialized with a flat interface m the center, where ai and ajj

are the lattice

spacings

in the

perpendicular

and

parallel directions, respectively (e.g.,

ajj

equals

v5/2

for

30-degree

interfaces and unity for

o-degree interfaces). Using no-slip

"bounce-back"

boundanes on the two walls

parallel

to the interface and

penodic boundary

conditions in the other

direction,

we allowed the interface 1000 time steps to

relax,

after which we

computed

(n~) over 2.5 x 10~ time steps.

The fit of the Boltzmann calculation to the

empirical

measurements is

qualitatively,

but not

quantitatively, good. Qualitatively,

one sees that the

empincal

and theoretical curves have

approximately

the same

shape

and their maxima are at

approximately

the same

density.

Moreover the anisotropy

predicted by

the Boltzmann

theory

is

evident,

with the same

sign

and

approximately

the same

magnitude,

m the measurements made from

interfaces,

while the

measurements made from bubbles fall

approximately

within the measurements made from 0-

degree

and

30-degree

interfaces. The poor

quantitative

fit can be attributed to several factors.

First, and most

obviously,

the

neglect

of correlations m the

theory

is a severe

deficiency,

not the least because the interfaces themselves are created

by

correlations.

Second,

because interface fluctuations

(see below)

cause the true values of c~i and

c~jj to differ from the assumed values used in equation

(48),

measurements made from interfaces are

necessarily

approximate.

Lastly,

we note that companson of the

empincal

data with the Boltzmann calculations made from the thicker interfaces is somewhat less

good

at low densities than the companson with the Boltzmann calculation made from the thinner interfaces.

(14)

6. Interface fluctuations.

Having investigated

above the

physics

of

stationary

interfaces m immiscible lattice gases, we

now tum to the

question

of

fluctuating

interfaces. Interface fluctuations in ILG'S are

interesting

for several reasons.

First,

for

applications,

one would like to know if

quantitative

aspects of ILG interface fluctuations are in agreement with theoretical

predictions

for dassical systems.

Second,

the

question

of agreement with dassical

theory

is an

interesting

issue in and of itself.

Because the

microdynamics

of the ILG lacks semi-detailed

balance,

the model is

microscopically time-irreversible,

and therefore lacks a dassical

description

in which fluctuations are distributed

m

equilibrium according

to a Gibbs distribution. Thus it cannot be

expected

a priori that the dassical

theory

would hold.

In the

following,

we first review what is

expected

for interface fluctuations in dassical sta- tistical systems in

equilibrium,

and then compare these

predictions

with simulations of ILG interfaces.

6.1 CLASSICAL EQUILIBRIUM THEORY. Classical interface fluctuations may be understood

m terms of fluctuations of surface energy [15]. For a one-dimensional interface m a two-

dimensional space, the energy,

H,

of the interface is

proportional

to its

length

L. Assume that the interface has

length

Lo when it is flat. If

h(z)

is the

single-valued

interface

height

as a

function of

position

z, then the

length

of the interface is

given by

~ ~ j

L =

/

~ dz

+

()) (49)

o 1

For

sufficiently

small fluctuations such that

dh/dz

< 1,

L m Lo +

/ ~

~ dz

~)) (50)

1

~

Now express the interface as the sum of Fourier modes of

amplitude

(Aq(

corresponding

to wavenumber q:

h(z)

=

~j

Aqe~~~~

(51)

q

Substituting

equation

(51)

into equation (SO)

above,

the interfacial energy H

= aL is, to

leading

order,

H m aLo +

°Lo ~ q~ÎAq(~, (52)

2

q

where a is

again

the surface tension.

By

the

equipartition theorem,

one has

aLoq~ÎA~(~

=

(kT, (53)

or

(Aq(~ =

~~~

~.

(54)

oaq

Thus for a dassical interface m

equilibrium,

one expects the average

amplitude

of fluctuations to be

inversely proportional

to the square of the wavenumber.

(15)

6.2 RESULTS FROM SIMULATIONS. TO compare the

prediction

of

equation (54)

with the fluctuations of ILG

interfaces,

the ILG was initialized with a flot interface of

length Lo

=

Nv5/2 dividing

the red fluid from trie blue fluid in a box of

height

N, where N

= 256 and

the

partide density

d

= o-1- The

boundary

conditions were

periodic

in the direction

parallel

to the

interface,

while walls were

placed

above and below trie

interface,

with

no-slip boundary

conditions. After

allowing

trie system looo time steps to relax to

equilibrium,

trie power spectrum

11_~ 2

lAql~(~) "

j £ h(~,~) ~~P(~iq~/N) (55)

~=0

was

computed

from measurements

h(z, t)

of the interface

heights

at each time step t and lateral location z, and then

averaged

over 10~ time steps. The interface

height

was determined at each discrete value of z

by searching

from above

(the

red

half)

for the

height

of the location of the first site

occupied by only

blue

partides,

then

searching

from below

(the

blue

half)

for the

height

of the first site

occupied by only

red

partiales,

and then averaging these two

heights.

If any

spikes

of

height

greater than or

equal

to rive lattice units were found in any interface

h(z, t),

this interface was eliminated from the

averaging

to avoid any measurement errors due to the small dissolution of each

phase

in trie other. This aspect of the data

processing

ehminated

approximately

1.5% of trie interfaces from the

averaging.

Figure

6 shows

log((Aq(~)

as a function of

log

q,

compared

to the theoretical curve obtained from equation

(54).

To compute the theoretical curve, we bave used a

= o.403, trie prediction obtained from the Boltzmann

approximation

for the

o-degree, 2-loyer

interface. For an estimate of

kT,

we have chosen the

equal-time

momentum-momentum correlation function

L L(lez«(nz Nz)llcjp(nj NJ

III

=

3d(1 dlô«p,

(561

or,

equivalently,

the vanance of momentum fluctuations at a site [18].

Comparing

trie two curves, one finds that the

slope

of the empirical curve is indeed

approximately

-2 for wavenum- bers below a

high-wavenumber cutoff,

and that the

prefactor (and

thus our estimate of kT from trie

equipartition theorem)

is

close,

within about 25%. We

condude, then,

that ILG interface

fluctuations are

qualitatively

in accord with the statistical mechanics of dassical interfaces.

That we have recovered the relation (Aq(~ oc

q~~ probably only

mdicates that ILG interfaces

are

"rough,"

in the sense that their

slopes

are random and uncorrelated. In other words,

h(z)

is

just

a random walk in one

dimension,

and (Aq(~ oc q~~ results from the usual considerations of diffusive processes.

However,

that the

prefactor

in equation

(54)

is also

approximately

recovered in the simulations is

relatively

remarkable. This agreement indicates net

only

that

our estimate of kT from equation

(56)

is

approximately

correct, but that the

equipartition

of

energy aise appears to be honored by the simulation. Such a result

implies

that the states of

trie system are distributed with Gibbs probabilities, which itself

implies

that the microscopic

dynamics effectively

behaves as if it were time-reversible. Such a behavior could

possibly

be

due to the fact that the noise arises

mostly

m the bulk

phases,

which are time-reversible, while the irreversible noise

generated

at the interface is bounded

by

the finite thickness of the

interface,

and is thus

possibly negligibly

small.

7. Conclusions.

The most important result of this paper is contained in

figure

5. There we have

compared

the

predictions

of a Boltzmann

approximation

for surface tension m an immiscible lattice gas with

Références

Documents relatifs

La raison : on trie bien les grandes bouteilles à la maison, mais on a moins le réflexe avec les plus petits formats que l’on emporte dans les transports, en balade, au sport…

Our final implementation uses a red-black tree to store fil- tered transactions, item order is ascending according to their support, simultaneous traversal is used as a routing

• Pour que la visibilité du bouton dépende de l'état de la case pointM, ouvrir le panneau des propriétés du bouton, et, dans l'onglet Avancé , inscrire dans le champ Condition

➋ L'image initiale pour insérer une image nommée image1 dans la zone de travail : deux points A et B.. sont automatiquement créés aux coins inférieurs

Placer ces points sur la figure

00 PRIX DU SYNDICAT DES MARCHANDS DE (Groupe A) Paris Simple Gagnant, Simple Placé, Couplé Gagnant, Couplé Placé, TrioD.

Although a compact suffix trie has a bit more nodes than the corresponding suffix tree, all of its arcs are labeled by single symbols rather than factors (substrings).. Because of

Largement répandu dans les établissements d’enseignement, le photocopillage menace l’avenir du livre, car il met en danger son équilibre économique et prive les auteurs