• Aucun résultat trouvé

Nonlinear theory of traveling wave convection in binary mixtures

N/A
N/A
Protected

Academic year: 2021

Partager "Nonlinear theory of traveling wave convection in binary mixtures"

Copied!
21
0
0

Texte intégral

(1)

HAL Id: jpa-00211128

https://hal.archives-ouvertes.fr/jpa-00211128

Submitted on 1 Jan 1989

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.

Nonlinear theory of traveling wave convection in binary

mixtures

D. Bensimon, A. Pumir, B.I. Shraiman

To cite this version:

(2)

Nonlinear

theory

of

traveling

wave

convection

in

binary

mixtures

D. Bensimon

(1),

A. Pumir

(1,3)

and B. I. Shraiman

(2)

(1)

LPS,

Ecole Normale

Supérieure,

24 rue Lhomond, Paris

75005,

France

(2) AT

& T Bell Laboratories, 600 Mountain Ave.,

Murray

Hill, NJ07974, U.S.A.

(3)

SPhT, CEN

Saclay,

F-91191 Gif-sur-Yvette, France

(Reçu

le 27

février

1989, révisé le 25 mai 1989,

accepté

le 15

juin

1989)

Résumé. 2014 Les ondes

propagatives

dans les fluides binaires en convection sont étudiées par un

développement

perturbatif

en

puissance

du rapport de

séparation

(| 03C8 |~

1 )

à

partir

de l’état convectif d’un fluide pur

(

|03C8|

= 0).

Pour des conditions aux limites libres-libres et

perméables

la transition de l’état conductif vers l’état de convection

propagative

est

supercritique.

Pour les conditions aux limites

plus

réalistes

expérimentalement

(rigide-rigide-imperméable)

avec

03C8 ~ 2014

L2,

on obtient la courbe

sous-critique

décrivant la transition de l’état conductif vers l’état de convection

propagative

et on évalue les

profils

de concentration. La transition secondaire de l’état de convection

propagative

vers la convection stationnaire est

surcritique

et la valeur du nombre de

Rayleigh

à

laquelle

elle

apparaît

est calculée

analytiquement

et

numériquement.

Ces résultats sont

comparés

avec les observations

expérimentales

et d’autres

expériences

sont

suggérées.

Abstract. 2014

Traveling

wave

(TW)

convection in

binary

mixtures is studied

by

an

expansion

in the

séparation

ratio

(|03C8| ~ 1)

around the pure fluid convective state

(03C8 = 0 ).

For

free-free-permeable boundary

conditions the bifurcation from the

conducting

state to TW convection is critical. For the

experimentally

relevant

rigid-rigid-impervious

boundary

conditions with

03C8 ~ 2014

L 2,

the sub-critical bifurcation curve from the

conducting

state to TW convection is obtained and the concentration

profiles

are evaluated. The transition from

traveling

waves to

steady

convection is

predicted

to be critical and the value of the

Rayleigh

number at which it

happens

is calculated

analytically

and

numerically.

Comparison

with available data and

suggestion

for further

experiments

are discussed. Classification

Physics

Abstracts

45.25Q

Introduction.

Thermal convection in a

binary

fluid mixture is an

interesting experimental

[1-3]

and theoretical

[4-8]

model

system,

for which the destabilization of the

quiescent (conducting)

steady

state occurs at finite

frequency through

a

Hopf

bifurcation

[4].

This leads. to a rich

dynamical

behavior near onset

[1-3].

The

conducting

state is observed to

undergo

a transition to a

steady

state of uniform

travelling

convective rools

[2a, la].

Then upon

increasing

the

temperature

difference across the

cell,

this state of

traveling

waves

undergoes

a

hysteretic

(3)

transition to uniform

steady overturning

convection

[2a, la].

The

spatial

structure can be

quite complex involving

the coexistence of left and

right traveling

waves

(TW),

fast and slow TW and TW and

conducting

state

[1-3].

The linear

instability

of the

binary

mixture

conducting

state is

by

now well understood

[6].

However the

present

non-linear

analysis

[5,

7,

8]

is

only

valid near the so called

codimension-two

point

[5, 8],

where

weakly

non-linear studies

[9]

are

justified.

As most

experiments

are

done very far from this

point,

the

agreement

between the

theory

and the observations is at

best

qualitative.

The purpose of this work is to extend the theoretical

analysis,

to describe

finite

amplitude

convection and to

provide

some

quantitative

results to compare with the

experimental

data. In the

following

we are

going

to review

briefly

the mechanisms

reponsible

for convection in pure fluids and in mixtures. We will introduce the basic

equations describing

convection in mixtures. We will then solve them for

free-free-permeable

and

rigid-rigid-impermeable boundary

conditions. We will solve

analytically

when the

traveling

wave

velocity

vo is much

larger

than the convective

velocity

u (vo > 1 u

and when

vo « 1 u|

1

and

numerically

between those limits.

Convection in pure fluids results from the

competition

between the

stabilizing

effect of heat diffusion and

viscosity

[10] (on

time scales of Td =

d2/ K

and T v =

d2/ v

respectively)

and the destabilization

by

the

buoyancy

forces

resulting

from the unstable thermal stratification on a

time scale Tb =

(

a8dTg ) 1/2,

, where g

is the

gravitational

acceleration,

a the coefficient of

a&Tg )

thermal

expansion,

K the heat

diffusivity, v

the kinematic

viscosity,

8 T the

imposed

temperature

difference across the cell and d its

height.

When Tv Td :>

Tb,

temperature

and

vorticity

fluctuations do not

damp

fast

enough

and convection sets in. In terms of the dimensionless

Rayleigh

number :

the convection

instability

occurs when R exceeds some critical value

Rc,

the

precise

value of

which

depends

on the thermal and

hydrodynamic boundary

conditions

[10].

In

binary

mixtures,

a second

stabilizing

mechanism may exist due to the

coupling

between

temperature

and concentration fluctuations

[4a].

This

coupling

is

parametrized by

the

separation

ratio t/J = -

(a’1a ) ST co(l - co),

where cl ’ is the coefficient of solutal

expansion,

ST

the Soret coefficient and co the average concentration.

When Vi

0,

the

coupling

is such that the

lighter

fluid

migrates

to the colder

region,

thus

establishing

a

stabilizing

concentration

stratification,

which opposes the

destabilizing

thermal stratification. A linear

stability analysis

of the

conducting

state

[6]

correctly

predicts

the existence of a

Hopf

bifurcation at the critical value :

characterised

by

a

frequency w ~

(-03C8 / (1

+ 03C8 )

)1/2

and hence

corresponding

to convective rolls

traveling

with a

velocity

Vo =

103C8 I

11/2.

The linear

theory

[6]

also

correctly predicts

the

dependence

of the rate of

growth

and the

frequency

of the unstable mode as a function of its

wavenumber

[2b].

However the

experiments

show that the unstable mode fails to saturate at

small

amplitude

even

arbitrarily

close to onset

[1-3].

Thus the bifurcation is backward and the transition

hysteretic

[1-3].

This can be understood if one observes that in the

experiments

the

Lewis number L =

D 1 K

is very small

(L =

10- 2)

and as a result in the presence of a

(4)

is obliterated and the

conducting

state with its linear

profile

of concentration ceases to be a

good approximation.

This limits the

validity

of the linear and non-linear

analysis

done around

the

conducting

state, to a

regime

where convective motions are not

destroying

the concentration stratification. The destruction of this concentration stratification

happens

close

to the codimension-two

point

[5]

where the

mixing

time due to

convection,

is of the order of the molecular diffusion time Tmol =

d21 D,

needed to establish a concentration

gradient

(D

is the molecular diffusion

constant).

This

corresponds

to

separation

ratios :

Since L =

10- 2,

such a

regime requires

Il

= - 10- 4,

which is very difficult to achieve

experimentally.

Most

experiments

[1-3]

are done for values

of t/1

« -

L 2.

The concentration

gradient

is therefore

effectively destroyed by

the convection flow and in order to

explain

the existence of the observed

traveling

waves one must understand the influence of the Soret effect in the

boundary layers,

where a concentration stratification may still

persist.

The purpose of this paper is to

develop

a

theory correctly accounting

for the effect of

convective

mixing

of concentration and to

present

results relevant to the

experiments

done at

values

of Ji

« -

L 2.

In

particular,

we would like to understand the very

rapid

decrease of the TW

period

(by

almost 2 orders of

magnitude)

along

the

hysteretic

branch when the control

parameter

changes

[2d]

by only

10 %. We would like to know the form of the concentration

profiles

in the TW

regime

and understand the observed transition from TW to

steady

overturning

convection

[2a, la].

The

analysis

we are

going

to

present

is done in an infinite

medium,

but its results are valid for

experiments performed

with

periodic boundary

conditions,

i.e. in an annulus

[2d].

The basic idea behind our method is to treat the

separation

ratio Il

as a small

parameter,

and

analyse

the convection in

binary

mixtures

by

an

expansion

around the pure fluid convective state and not, as is

usually

done

[4-8],

by expanding

around the

binary

mixture

conducting

state. Thus the

temperature

and

velocity

field have the same form as in a pure

fluid,

though

maybe

not the same

amplitude.

Moreover we shall take into account the fact that the Lewis number L satisfies : L

B/201303C8,

, and that the Péclet number :

p =

1 u 1 IL > 1.

More

precisely,

we introduce two small

parameters,

B/2013 03C8

and

L/

B/2013 03C8,

thus

taking

into account

the ratios between the control

parameters

most

commonly

used in

experiments.

It

happens

that the

limit L

B/201303C8 ~

0 is very

singular

and must be dealt with

carefully.

On the

contrary,

the limit

B/2013 03C8 ~

0 is well

behaved,

and it is

possible

to do a

systematic expansion

of the

equations

in powers of

B/- 03C8.

We will elaborate these ideas in turn. The

equation

for the concentration field can be solved

analytically

in two limits :

vo > 1 u

1

by

an

expansion

in

1 u 1 lvo «

1,

and

vo «

1 u 1

by

a

boundary layer analysis.

We have determined

numerically

the

concentration field

by

a Galerkin method in between those values. The

velocity

of the

TW,

vo, and their

amplitude

is obtained as

solvability

conditions on our

equations.

Below we will first describe the basic

equations,

which will be solved for the case of

free-free-permeable boundary

conditions where many

analytical

results can be obtained. In

particular,

one can show that the

Hopf

bifurcation in this case is forward. We will then deal

with the

experimentally

relevant

rigid-rigid-impermeable boundary

conditions. Close to onset

(vo > 1 DI),

one can show that the

Hopf

bifurcation is inverted. One can also

study

analytically

the transition from TW to

steady overturning

convection

(vo > 1 u 1)

and find the critical value Pc =

1 uc 1 /L at

which the transition

happens.

In between those

values,

our

equations

are solved

numerically

and checked in the relevant limits

against

the

analytical

(5)

Basic

equations.

In the limit of infinite Prandtl

number,

which

maybe

the relevant limit for

binary

fluids like water-ethanol

[1, 2]

but not

He3 - He4 mixtures

[3d],

the

equations

describing

convection are

[4] :

a)

the Navier-Stokes

equations

in the

Boussinesq approximation :

where w is the vertical

velocity

component,

T the

temperature,

y the

concentration,

R the

Rayleigh

number

and If

the

separation

ratio. In this paper the unit of

length

is

d,

of time

d2/K,

of

temperature

6 T and of concentration

8 T ST co(l - co ).

b)

the heat and concentration diffusion

equations :

where

Dt

is a convective derivative :

Dt

=

at

+ u . V. Notice that

equation (lc)

can be

rewritten as

with

y = y’+ L TI (1 - L ).

Since L 1 and T is of

0(1)

and

nonsingular,

we will

approximate

y = y’.

Let us introduce the

temperature

fluctuation 0 defined

by :

T = - z + 0 and stream

function 0 :

u =

(-

a,o,

0,

ax). Equations (la,

b,

d)

become :

We also need to

specify

the

boundary

conditions on the fields

0, 0

and y. We will consider

a) free-free-permeable

boundary

conditions :

b) rigid-rigid-impermeable

boundary

conditions on z =

0,1 :

We look for

traveling

wave solutions :

(6)

contrast with

previous

theoretical

analyses

of this

problem

which assumed

[4-8]

yo = z, we do not

expand

the concentration field y around the linear concentration field

existing

in the

conducting

state. We are thus able to

correctly

describe the concentration

boundary layers

existing

when convective

mixing

is efficient. However for the

expansion

around the pure fluid

convective state

(Eq.(3))

to be

valid, 03B5

must be small

( s « 1 ).

Since from the linear

stability

analysis

and from

experimental

data we know that in the domain of existence of

TW,

B =

0(B/2013 03C8),

we

require 1 03C81

1. This allows us to treat the

coupling

between the concentration and

velocity

fields

(the

term

R03C8raxy

in

Eq. (2a))

perturbatively.

Let us define

Inserting

the ansatz

equation

(3)

into

equations (2a,

b,

c)

yields

to order

0(03B5) :

with the concentration diffusion

equation :

where po =

vo/L.

This form is

appropriate

close to onset where it can be solved

by

an

expansion

in powers

of a - 1

|u|

1. Far from onset,

(a - 11

|u1| >

1 )

we shall use

boundary

layer

techniques

to solve

equation (5c)

rewritten as :

where ul =

Aûl,

A is the

amplitude

of the convection field

(see below),

p =

EAIL

is the

Péclet number and

g - 1 =

«A -1=

vo/ EA

(

1 )

is the ratio between the TW

velocity

and the

convective

velocity amplitude.

It

is important

to notice that in

equations (5c, d)

both

Po

1 and

p -1

are small,

of order

LI -.p.

However the limit

Po

1,

P 1-+

0 is

singular

and

must be treated

carefully.

To order

0 (e2) :

To order

0 (e3):

(7)

The zero modes of the

adjoint

are

(01, - 01)

and

(a"ol, - a"ol).

At

0 (E2),

the

solvability

condition is

(1) :

where the

brackets (... )

define the inner

product :

(h, f )

-

dx dz

h f .

Due to

symmetry

under reflection

(x -+ - x)

the last term on the left hand side is zéro.

Hence,

This

solvability

condition determines the TW

velocity.

At

0(e3)

the

solvability

condition is

(1) :

Which

yields

the

equation

for the

amplitude

of the TW. We will argue later that when convective

mixing

is efficient

(far

from

onset),

concentration

gradients

exist

only

in

boundary

layers

and therefore from

equation (8)

the TW

velocity

vo is small

(vo «.c 1 u

).

In this case the

right

hand side of

equation

(9)

is also small and its left hand side

yields

the same

amplitude

equation

as for a pure fluid. We will thus conclude that away from the onset of convection in

binary

mixtures,

one

expects

slow

traveling

waves with an

amplitude

equal

to the

amplitude

of

stationary

convection in a pure fluid at the same

Rayleigh

number.

In the

following

we are

going

to solve

equations

(5-9)

for two cases of interest :

free-free-permeable

and

rigid-rigid-impermeable boundary

conditions.

Free-free-permeable

boundary

conditions.

This case is

interesting

not because of its

experimental

relevance,

but because it is tractable

analytically

and instructive. The solution of

equations

(5a, b)

with

b.c ;

(2d)

is

[10] :

with

corresponding R, =

(k2 +

lr2 )’Ik2

and

choosing

k =

kc =

ir

/

/2.

This choice of

k is

suggested by

our

expectation

that the

velocity

field is close to the

velocity

field for

Rayleigh-Bénard

convection near onset. The

equation

for the concentration yo,

equation

(5c)

then becomes :

with g =- a -’A

= EA

/vo =e u1

/vo

and

Û1 ==

A-1

U1 =

(-

ir sin kx cos 7TZ,

0, k

cos kx sin 7rz

).

Let us look for a solution :

(1)

At each order

orthogonality

to the second mode of the

adjoint

induces a small

O (L2/1/1 )

phase

shift between the

velocity

and concentration fields. This can be taken into account, but

complicates

the

(8)

satisfying

the

boundary

condition

1’00 1 0

=

0,

yoo 1 1 = 1 ;

l’on

10,1

=

0,

for n

= 1, 2,

... One

obtains a set of

equations

which can be solved to any order. It has to be

emphasized

that,

although

po 1

is

small,

one cannot

neglect

it,

because of the

singular

nature of the

limit Po 1 -+

0. One has

immediately :

yoo = z. The solution can be written as :

with

mn * 1.

For

m = 0,

ymn

is

0 (p - 1)

and thus

negligible.

yR is

orthogonal

to

cp

1 and thus does not contribute to the

solvability

condition at next order

(Eq. (8)).

The

normalisation factor

f (g, po)

can be evaluated

analytically

to any desired order. To lowest

order

in g

it is :

We have also solved

equation

(11)

numerically, by writing

a Galerkin

expansion

of

yo and

inverting

the

(truncated)

system

of linear

equations.

Numerical results are

presented

in

figure

1,

and agree very well with

equation (14)

at small values of g. The

solvability

condition

determining

the TW

velocity equation

(8)

yields :

Fig.

1. - The function

f (g, po) in

the case of the

free-free-permeable

conditions as a function of g for po = 50. Curves

corresponding

to various values of p0 > 10 are

(9)

we recover the result of the linear

stability analysis

[6].

As

f

decreases with

increasing

g

(see

Fig. 1),

the TW

velocity

vo

decreases away from onset. Notice however

that,

since g -

eAlv

o, to

actually

calculate vo one has to solve for the

amplitude

of convection

A.

In order to determine the

amplitude

equation

for

A,

we need to find the

velocity,

temperature

and concentration fields at order

O(e2),

equation

(6a, b,c).

These are

(see

appendix

A) :

with . The

expressions

for OR, OR

and

y e)

are

given

in

appendix A.

At this order the fields can thus be written as :

Notice the

phase lag by

vol (k2 +

11’ 2)

between the

temperature

and

velocity

fields

[7].

Close

to onset

(g 1 1 ),

this

phase

lag,

due to the

coupling

between the

temperature

and concentration

fields,

is

responsible

for the

instability

to TW of the

binary

mixture

conducting

state

[7].

The second

solvability

condition,

equation

(9),

yields

the static

amplitude

equation :

with

bi

=

kB2/8 (k 2+ 7r 2)

and

i (g, po) -

f (g, po ) (see Appendix A). Equations (15)

and

(18)

form a set of two

coupled

non-linear

equations

in two unknowns vo and A. In

general they

must be solved

numerically,

however in two

limits g «

1

and g >

1,

one may

actually

obtain

analytical

results. Thus away from onset

(g > 1 ),

i.e. for slow

travelling

waves,

vo «

[ u 1

and

(as

one verifies

numerically) f (g,po)

1,

the

amplitude

of convection eA in a

binary

mixture is identical to the one in a pure fluid at the same

Rayleigh

number

[11] :

A =

/8 (k 2+

11’2)lk2

= J24.

However close to onset

(g « 1 )

as

f (g, po) -

1,

this is no

longer

true.

Combining equations

(15, 18)

one obtains :

And from the

expression

for

f(g,po), equation (14),

near onset one derives :

with

fi

=

(ir 2

+

k2)/8

and

f2

=

(k2

+

11’2)(k2

+ 9

ff2 )/64.

The bifurcation from the

conducting

state (g =

0)

to a convective state of

traveling

wave

(10)

subcritical if al

b1 - f,

0. For

free-free-permeable boundary

conditions,

al

bl - f 1=

0,

the

bifurcation is

supercritical

with the

amplitude

of convection eA

increasing

as the

quartic

root

of the deviation from onset

[12, 8c],

e2 + 03C8 :

This result

implies

that for

free-free-permeable

boundary

conditions,

the usual

perturbation

expansion

around the conduction state

[6-8]

is valid.

Rigid-rigid-impermeable

boundary

conditions.

This is the

experimentally

relevant case. The solution of

equations

(5a, b)

with b.c.

(2e)

is

[10] :

:

with £ =

z - 1/2,

Re

= 1707, k

=

3.117,

qo =

3.973,

q = 5.195 + i

2.126,

a =

-1

2

[0.0615

+ i

0.1038 ].

The solution for the concentration

field,

equation

q

(5c),

(5c

)

proceeds along

P

Fig. 1.

Fig. 2.

Fig.

2. -

The reduced

amplitude

of convection eA as a function of the reduced temperature

e2/ 1 03C8

1

for L

= 10-2 and 4r

= - 1/4, in the case of

rigid-impermeable boundary

conditions. The lower

branch

corresponds

to fast

(and unstable)

TW solutions, the upper one to slow and stable one.

Fig.

3. - The

phase velocity

of the

travelling

waves

vo/

/- 03C8

as a function of the reduced temperature

e2/1 t/J

1for L

=10-2 and 03C8 = - 1/4,

in the case of

rigid-impermeable boundary

conditions. The inset is

(11)

Fig.

4. -

Isoconcentration patterns for various

(unstable)

solutions for L = 10-

2,

and 03C8 = 2013 1/4

(a)

e2/03C8!

[

= 0.700 and

(b)

E 2/ 1 p

]

= 0.850. The difference between two isoconcentration lines is 6 x

10-2

for

figure

4a, and 8 x

10- 2 for figure

4b.

the same lines as for the

previous

case

equations (11-13),

although

the

algebra

is more

cumbersome. Close to onset,

(g « 1 )

one obtains

with

f,

= 7.856. From

equation

(22)

one calculates the value of al in the

equation

for the TW

velocity, equation

(15) :

al = (axq,1)2) 1 (âx6l)2)

= 40.344. From Manneville and

Piquem-al

[11],

one also knows the value of the coefficient

bl

in the

amplitude equation,

equation

(18) : bi

= 0.0612. One thus verifies

that,

since al bl - f

0,

the bifurcation from

the

conducting

state to the TW state is subcritical

(i.e. hysteretic).

To make further progress,

one has to

compute

f (g, po)

numerically.

Then

assuming equation

(18)

to hold to all g

(it

is known to hold

for g

1

and g >

1),

one can solve

numerically equations

(15)

and

(18)

(see

Appendix

B)

and determine the

amplitude

of convection : eA

(Fig. 2)

and the TW

velocity

vu

(Fig. 3)

as a function of the reduced

temperature £2

=

(R -

Rc)IRc

for various

values of the

separation

ratio

03C8,

and Lewis number L. The concentration

profile

yo(x - vo t, z)

is determined

simultaneously

(Figs. 4, 5).

Thus close to onset and on the

lower

(unstable)

branch of the

hysteretic loop,

the concentration

profile

deviates

only slightly

from the linear

gradient existing

in the

conducting

state

(Fig. 4)

and the TW

velocity

is fast

(0(B/03C8| )).

On the upper

(stable)

branch, however,

the concentration

profile

exhibits clear

boundary

layers

(Fig. 5)

and the TW

velocity

is slow

(vo «

B/|03C8| ).

When g >

1,

the

amplitude

of convection is

equal

to the

amplitude

of convection in a pure

Fig.

5. - Isoconcentration

pattems for various

(stable)

solutions for L

=10- 2,

and 4, = - 1/4,

(a)

E 2/ 1 03C8|

1

= 0.700

(b) e 2/ |

1 03C8 1

= 0.850. The difference between two isoconcentration lines is 3 x

10-2

for

(12)

fluid at the same

Rayleigh

number,

and the TW

velocity

can be evaluated

analytically

from

equation

(15)

by

a

boundary layer

calculation

[13]

for the concentration field which is

described next.

Slow

traveling

waves and concentration

boundary layers.

The characteristic scale for the

velocity

of convective flow is set

by

the thermal

diffusivity.

Since the later is much

larger

than the molecular

diffusivity

(L ~ D/K 1 )

the Péclet

number,

p

=

1 u IlL (2),

based on the convection

velocity

is

large

and the

mixing

due to the flow has a much

stronger

effect on the concentration

field,

than on the

temperature.

While

the uniform vertical

temperature

gradient

established in the

quiescent

conducting

regime

is

slightly

perturbed

and modulated

by

the

flow,

the uniform vertical concentration

gradient

is

completely

destroyed (in

the limit L -

0).

Instead,

the concentration

gradients,

imposed

by

the

coupling

to the

temperature

field

(via

the Soret

effect),

are confined to the

boundary

layers

near the walls

(where

the flow

velocity

vanishes),

and to the free

boundary layers along

the vertical

separatrices

of the flow

(see

Fig.

6).

This

picture

suggests

an alternative

approach

to

study

the TW convection :

instead

of

expanding

about the

conducting

state we start with the limit of

large

Péclet

number,

i.e. finite

amplitude

convection and L - 0. In that limit the

system

is in a state of

stationary

convection but as p goes down

(with

the

Rayleigh

number which is controlled in the

experiment)

it

undergoes

a transition to a

traveling

wave state. This transition can be

thought

of as an

instability

of the concentration

boundary layers

in the

convective flow.

The

strategy

of the calculation is as follows. First we determine the distribution of a

passive,

1/1

=

0,

impurity

in the convective flow with Soret

coupling

to the

temperature

field. For this

purpose we assume a

single

mode

rigid boundary

convection and use the

boundary layer

theory applicable for p >

1 to determine the concentration field.

Next,

we calculate the correction to the flow

(and temperature)

field due to the

finite 1/1

perturbatively

in

1 If 1

1. The

perturbation theory

involves a

solvability

condition

relating

the

asymmetric

distortion of the concentration

boundary layer

with the translational mode of the cellular

pattern.

We will find that such

asymmetric boundary layer

solution

corresponding

to

non-vanishing

TW

velocity

can occur

for p pc (and

hence for R _

R’)

with p,

(|03C8

11/2/L )8 /7

Thus,

we seek a

steady

state solution of

equation

(5d)

with

g-1=

0

(vo = 0 ) :

where p

=

sA /L

is the Péclet number and the flow field

(ul

=

Aûl)

is defined

by

the

stream-function 01

1 of

equation

(22)

corresponding

to the

single

mode

rigid boundary

convection.

The

boundary

conditions for the concentration field are :

The

general

solution of

equation

(24)

can be written as yo =

yh +

yp

where

yp

satisfies

equation

(24

a,

b)

and yh satisfies

equation

(24a)

with

homogeneous boundary

conditions :

âzl’h1z = 0,

1 = 0. The

inhomogeneous boundary layer

concentration

yp

is

generated by

the

simultaneous diffusion and advection

[14]

of the concentration

gradient

due to the Soret effect

existing

near the walls :

equation

(24b).

Since at a

distance e

« 1 from the

wall,

~l = A’ 2 sin kx,

balancing

the diffusion term

(p-1 a§yo)

with the advection terms

(ûl . oyo)

in

equation

(24a)

yields

an

inhomogenous

boundary layer

width

scaling

as

(13)

Fig.

6. -

Steady

flow pattern

(vo = 0 ).

The notation used coincides with the one in the text.

p-l/3.

This

boundary

layer

detaches at

points

A

(A’ )

giving

birth to a free

boundary layer

along

the

separatrix

AB

(A’B’).

The latter in turn extends

along

the streamlines into the wall

regions

forming

the

top

(bottom)

homogeneous boundary layers

(Fig. 6).

Since

along

the

separatrix çb 1 -- kAw (z) x,

balancing

the diffusion across streamlines

p -1 ax yh

with the

advection

along

streamlines

yields

a

boundary layer

width

scaling

as

p - 1/2.

As the

AB (A’B’ ) boundary

layer

extends

along

the streamlines into the

top

(bottom)

wall

regions,

the

expansion

of the streamlines will

change

the width of the

layer.

The concentration

y h in

the

boundary layer

varies with p 1/2

0,

and

since 0 ~

z 2,

we infer that the concentration

scales as

p1l4

z and

hence,

the

homogeneous boundary layer

width near the wall scales

[13a]

as

p-1I4.

The effect of diffusion

(p-1 a2 z Y h)

across this broadened

layer

is of 0

(p - 1/2)

and can

thus be

neglected

in

comparison

with advection

parallel

to the wall

(azo

axyh ’--.

zaxyh)

which is of order 0

(P-1/4).

Thus the

homogeneous boundary layers along

the walls are dominated

by

advection and the variation of yh

along

streamlines in these

regions

can be

neglected.

It is

important

to notice that the

homogeneous boundary layer

(of

width

p-

1/4)

is much thicker than the

inhomogeneous boundary layer

(of

width

p-1/3).

Thus,

the

inhomogenous

boundary layer

can be accounted for

by including

a source of

strength

F(- T )

at the

stagnation point

A (A’ )

in the

equation

for yh.

By

conservation of flux the

strength

of this

source is F=

7Tk-1p1l2

which is obtained

by integrating

az

"Yplz=l,O

along

AB’(A’B)

and

rescaling

with

pl/2

as

appropriate

for the

AB (A’B’)

boundary layer.

We now solve for the concentration field

along

the

separatrices. Introducing

streamlines

coordinates

[13] :

with ~

1 = kx

w (z )

one derives

[13]

from

equation

(24a)

the

equation

for the AB

boundary

layer

whose solution is :

(14)

To obtain an

equation

for

Yh(o-, 0)

we follow yh around the closed

loop

ABA’B’ and

remember that diffusion has no effect on yh in the wall

regions

BA’,

B’A. We find :

where To =

T (1 ).

Introducing

the Fourier transform

and fourier

transforming equation

(29)

yields :

Substituting

this into

equation

(27)

yields :

which back in real space has the form :

with :

TI;

=

2(T(z)

+

nTo)/p.

Having

found the concentration

profile

for the «

passive » impurity

we can turn back the

coupling

to the

flow, l.p 1 =F

0,

and calculate the corrections to the flow

perturbatively.

As has been shown

earlier,

the

perturbation theory

will involve the

solvability

condition :

equation

(8).

Provided that

y (x, z )

is

symmetric

under x - - x, the

solvability

condition is

trivially

satisfied.

However,

if the

boundary layers

were

asymmetric,

the

solvability

condition

would

require

a

non-vanishing

translational

velocity

of the rolls -

that the TW

velocity

Vo #

0. We can look for such a non-trivial self consistent solution

simply by replacing

the

streamfunction ~1

in

equation

(26)

by ~1 ~ g- 1 Z +

1 which

corresponds

to the TW

convective flow as seen in the

comoving

frame

(with

g-1~

vol~A

being

the rescaled TW

velocity).

As

long

as

g- 1

is

small,

i.e. smaller than all other small

parameters

in the

problem,

our

analysis

for the

stationary

convection

applies

without

change

and the concentration field is

given by equation

(32)

with ~1 replacing ~1.

The self consistent value of

g-1

is determined

by

the

solvability

condition of

equation

(8) :

with : à p

A - 1 0 1.

The inner

product: ôx’Yo, dxcÎ>

=

2 (ôx’Y AB’ ÔxcÎ>

(the

factor of 2

comes from the contribution of the A’B’

separatrix),

can be evaluated from

equation

(32)

(15)

with rc

= 16.98

a

positive

constant. The value we deduced from our numerical solution of

equations

(19, 15)

is K = 4.5. The

discrepancy

is due to the

approximations

in the evaluations of the

integrals,

as discussed in

appendix

C. The nontrivial

(g- 1 ~

0)

solution appears for

p pc with Pc

(g -

~

0 as p ~ pc from

below) :

The p

= pc

point

corresponds

to the transition from

stationary

to the TW convection

which,

in

terms of the reduced

Rayleigh

number is therefore

predicted

at

£TW = Pc L = 1 03C8/114n.

Equation

(35)

is confirmed

by

a numerical solution of

equations

(15, 19) (see

Fig.

7),

which

also shows the bifurcation from

steady

to TW convection to be critical

(Fig. 3).

Notice that the convection

amplitude

is smooth

through

this transition. This is in

partial

agreement

with the

experimental

results

[2a, la],

where a continuous transition from TW to

steady

convection

was observed on

heating.

On

cooling though,

the transition from the

steady

convective state to TW convection was

hysteretic.

However the

experiments

were done in a

rectangular

geometry

where the influence of the lateral walls is known

[1-3, 9]

to be

important.

These walls may stabilize the

steady

convective state

against

a TW

perturbation,

thus

leading

to

hysteretic

behaviour.

Fig.

7. - The critical Péclet

number Pc

at the transition from TW to

stationary

convection as a function of 1 =

L/

B/2013 03C8.

The

slope

of the curve is - 1.045, with an error of 0.005. The difference with the

analytic

result

(- 8/7)

may come from the fact that the

asymptotic regime

has not yet been reached

(the

values of 1 are not small

enough).

Discussion.

For

free-free-permeable

boundary

conditions,

we have recovered

previously

known

[8, 12]

results and showed that the bifurcation from the conduction state to a state of TW convection is

forward,

with the

amplitude

of convection

scaling

as

(R -

Rco)1/4.

(16)

convection is

subcritical,

i.e.

hysteretic,

in

agreement

with the

experiments

[1-3].

In addition

we have been able to evaluate

analytically

the critical

point

where the

secondary

bifurcation from TW to

steady

convection

happens.

This

system

is one of the few

examples

where such an

analysis

can be done.

We have calculated the full bifurcation

diagram.

Its upper branch

corresponds

to the observed uniform state of slow TW. In

agreement

with the

experimental

observations

[1-3],

the

amplitude

of the TW convection on the upper branch is found to be

equal

to the

amplitude

of

steady

convection in a pure fluid with similar

properties.

Our results exhibit a fast decrease

of the TW

velocity

(by

almost two orders of

magnitude)

as 03B5 is increased

by

a mere 10 %. This

is in

qualitative

agreement

with the

experimental

data

[2d].

The

analysis predicts

a critical

(forward)

transition from TW convection to

steady

convection. This however is at odds with the

experiments

[2a, la]

which find the transition to be

weakly hysteretic.

This

discrepancy

could be due to the presence of lateral walls. An

experiment

in an annular

geometry

is

clearly

needed. The

steady

convection state becomes unstable to TW convection at a value of the Péclet

number Pc =

1 UcIlL

given by equation

(35).

It would be

interesting

to

study

this transition as a function of the

separation

ratio t/J

experimentally

and make a

quantitative

comparison

with our

prediction.

It would also be of interest to

study

the concentration

profiles directly

and

verify

the existence of concentration

boundary layers. Experimental

comparison

should be

attempted

in a one-dimensional annular

geometry

[2d],

since the influence of the lateral walls in a

rectangular

cell,

which are not taken into consideration in

our

analysis,

is

experimentally

relevant

[le, 9].

On the lower

(unstable)

branch the TW

velocity

is fast

(of

0(V- 03C8)),

and the concentration

profile

is close to the linear concentration

profile existing

in the

conducting

state

(see

Fig.

4).

Although

such a state

(with

infinite

extent)

is

unstable,

one wonders if it could not exist as a stable state of

finite

length.

Due to the existence of fast

TW,

infinitesimal disturbances are advected and for a convective

region

of finite

length, they

may not

develop

to destabilize the structure. In other words the fast TW

existing

in the lower branch may be

only convectively

unstable,

not

absolutely

[9].

This may

explain

the

stability

of the observed confined states

(always

characterised

by

fast

TW),

which should then be understood as solitonic structures

[15].

To check for the

possibility

that the fast TW in a confined state

correspond

to the fast TW on the lower

branch,

one can think of various measurements. One could measure the concentration

profile

in a confined state and check whether it is as

predicted

for the lower

branch,

i.e. close to the concentration

profile

in the

conducting

state.

One could also measure the TW

velocity

and

amplitude

in the confined states as a function of

e and compare with the theoretical

predictions

(Figs.

2,

3).

Let us notice that the

analysis

presented

here can be extended to values

of tp

and L different from the ones we chose.

Appendix

A.

(17)

We seek a solution in the form :

Inserting

this ansatz into

equation

(Al)

yields :

The

arbitrary

parameter

À

corresponds

to a translation of the solution and doesn’t contribute to the

solvability

condition at next order. We thus set it to zero. Notice that the fields

02

and

62

can be written as :

The

equation

for the concentration field y 1 is now :

the solution to which can be written as :

where

l’Á1)

satisfies :

The solution

’YÁ1)

can be

sought

as a fourier

expansion.

The

solvability

condition at

(18)

Notice that :

Where we

identify :

Thus the

solvability

condition

equation

(A8)

yields :

with

f ~

f

+

h R.

It can be

readily

seen that

hR

is very small. Near onset

(g « 1)

one has

hR -

0

(g6)

f =

0

(1 ),

and far from onset

(g > 1 )

due to the existence of

boundary layers

f , hR --> 0.

We shall therefore use as an

approximate

amplitude equation, equation (A.10)

with

f =

f.

Appendix

B.

This

appendix

is devoted to the numerical solution of

equations

(15)

and

(19).

We will calculate the

amplitude

of convection

EA,

and the

velocity

vo of the

traveling

waves in the case

of

rigid-impermeable boundary

conditions,

as a function of the reduced

temperature

2

R - Rc

E2

=

Re,

, the other

physical

parameters

L

being

fixed.

Rc

p Y p

(03C8,L ) bein

g

Both

equations (15)

and

(19)

involve the function

f (g, po).

Our first task therefore consists in

computing

this function.

Using

the definition

(Eq. (13)) :

where yo is

the solution of

equation (11) :

(19)

In order to solve

numerically

equation (B.2)

one uses a Fourier series

decomposition

of the concentration field :

which satisfies

obviously

the

boundary

conditions

Ôz 1’01 z = 0,1

= 1.

Likewise,

the

streamfunc-tion,

which has to

satisfy

the

boundary

conditions CP1 = dzCP1 = 0 on z = 0, 1,

can be written

as a series of sines in z :

which satisfies

obviously 0

1

z = 0, 1

= 0 or as a series of cosines in z :

Which satisfies

trivially

âzcf> 11 z = 0,1 =

0.

Only

odd

(even)

terms contribute in the

decomposi-tion

(B.5a) (B.2b).

The coefficients of the

expansion 01

1 and t/J

can be

computed analytically

from

equation

(22)

and

expressed

in terms of

elementary

functions. Thus

equation

(B.2)

can

be rewritten as :

By using

the Fourier series

(B.5a)

in the evaluation of

û,

on the left hand side of

equation (B.6)

and the

decomposition

(B.5b)

on its

right

hand

side,

one

obtains,

after

straightforward manipulations,

an infinite set of

equations

for the

Am, n

and

Bm, n’s.

A

peculiarity

of the solution of this

system

is that all the coefficients

Am, n

and

Bm,

n’s,

such that

(m

+

n )

is odd are zero. A truncation of the

system

to a finite number of modes

yields

a finite set of

equations.

The

resulting

system

has been solved

by

an IMSL inversion routine up to 30 modes in x and 20 modes in z.

Of course, this truncation is valid as

long

as the

gradients

that build up in

boundary layers

(see

the

boundary layer

calculation)

are well resolved. We have

systematically

checked the resolution

by making

sure that the Fourier

spectrum

of the solution decreases fast

enough.

In

practice, problems

arise

when g

becomes

larger

than

1,

and the Péclet number p =

gpo

exceeds a certain

value,

for a

given

number of modes.

Once y has been

obtained,

it is

straightforward

to

compute

the scalar

product

ôxYo, ÔxcÎ>m

hence the function

f (g, po).

In order to obtain the

amplitude

of convection A and the

velocity

vo of the

traveling

waves as a function of the reduced

temperature

e2,

one first writes

in a sli tly

different way

(20)

This

system

can be solved

efficiently

in g

and po by

a standard Newton’s method. Once the solution has been

obtained,

the values of the

physical quantities

vo and A can be extracted :

The values we have chosen are L

=10- 2 ;

and

w

=-1/4;

tp

= - 1/16; 03C8 = - 1/36 ;

03C8

= - 1/64

and w

= - 1/100.

Appendix

C.

In this

appendix

we will show how to evaluate the inner

product :

To evaluate

equation

(C.1)

one notices that from

equation

(22) : ~

= sin kxw

(z).

Near the

walls at z =

0, 1

w(z)

=

K,2

where 03B6

is the distance from the wall

( 03B6 = z

or 1 - z and

rc =

15.45).

Introducing § m

kx and

6,, (z) =-

Tln(z)/w(z)

we rewrite

equation

(C.l) :

The main contribution to the

integrals

comes from the

region along

the

separatrix

AB (z.j. « z « zmax) for

which

8n (z ) 1,

and not from the

stagnation

zones near

A (0 «

z «

zm;n)

or B

(zmax « z « 1 ).

Since

8n

1,

we may

approximate

sin e -- e

and set the limits of

integration

on g

to ± oo . The

gaussian integration

is

straightforward, resulting

in :

Where the constant Notice that

by

setting

the limits of

integration

in

equation

(C.3)

to zm;n, Zmax we are

overestimating

the value

(21)

References

[1] a)

MOSES E. and STEINBERG V.,

Phys.

Rev. A 34

(1986)

693 ;

b)

MOSES E., FINEBERG J. and STEINBERG V.,

Phys.

Rev. A 35

(1987)

2757 ;

c)

STEINBERG V., MOSES E. and FINEBERG J., Nucl.

Phys.

Proc.

Suppl.

B 2

(1987) ;

d)

MOSES E. and STEINBERG V.,

Phys.

Rev. Lett. 60

(1988)

2030 ;

e)

FINEBERG J., MOSES E. and STEINBERG V.,

Phys.

Rev. Lett. 61

(1988)

838 ;

Phys.

Rev. A 38

(1988)

4939.

[2] a)

WALDEN R. W., KOLODNER P., PASSNER A. and SURKO C.,

Phys.

Rev. Lett. 55

(1985)

496 ;

b)

KOLODNER P., PASSNER A., SURKO C. and WALDEN R. W.,

Phys.

Rev. Lett. 56

(1986) ;

2621;

c)

SURKO C. and KOLODNER P.,

Phys.

Rev. Lett. 58

(1987)

2055 ;

d)

KOLODNER P., BENSIMON D. and SURKO C.,

Phys.

Rev. Lett. 60

(1988)

1723 ;

e)

KOLODNER P. and SURKO C.,

Phys.

Rev. Lett. 61

(1988)

842.

[3] a)

REHBERG I. and AHLERS G.,

Phys.

Rev. Lett. 55

(1985)

500 ;

b)

AHLERS G. and REHBERG I.,

Phys.

Rev. Lett. 56

(1986)

1373 ;

c)

HEINRICHS R., AHLERS G. and CANNELL D. S.,

Phys.

Rev. A 35

(1987)

2761;

d)

SULLIVAN T. S. and AHLERS

G.,

Phys.

Rev. Lett. 61

(1988)

78.

[4] a)

HURLE D. T. J. and JAKEMAN F., J. Fluid Mech. 47

(1971)

667 ;

b)

STEINBERG V., J.

Appl.

Math. Mech. 35

(1971)

335 ;

c)

GUTKOWICZ-KRUSIN D., COLLINS M. A. and Ross J.,

Phys.

Fluids 22

(1979)

1443, 1457 ;

d)

BRAND H. R. and STEINBERG V.,

Phys.

Lett. 93A

(1983)

333.

[5] a)

BRAND H. R., HOHENBERG P. C. and STEINBERG V.,

Phys.

Rev. A 27

(1983)

591 ;

Phys.

Rev.

A 30

(1984)

2548 ;

b)

ZIELINSKA B. J. A., MUKAMEL D., STEINBERG V. and FISHMAN S.,

Phys.

Rev. A 32

(1985)

702 ;

c)

SCHOPF W. and ZIMMERMANN W.,

Europhys.

Lett. 8

(1989)

41.

[6] a)

ZIELINSKA B. J. A. and BRAND H. R.,

Phys.

Rev. A 35

(1987)

4349 ;

b)

CROSS M. C. and KIM K.,

Phys.

Rev. A 37

(1988)

3909 ;

c)

KNOBLOCH E. and MOORE D. R.,

Phys.

Rev. A 37

(1988)

860.

[7] a)

LINZ S. J. and LUCKE M.,

Phys.

Rev. A 35

(1987)

3997 ; A 36

(1987)

2486 ;

b)

LINZ S. J., LUCKE M., MULLER H. W. and NIEDERLANDER J.,

preprint

(1988) ;

c)

LUCKE M.,

preprint

(1988).

[8] a)

KNOBLOCH E.,

Phys.

Fluids 23

(1980)

1918 ;

b)

KNOBLOCH E. and GUCKENHEIMER J.,

Phys.

Rev. A 27

(1983)

408 ;

c)

KNOBLOCH E.,

Phys.

Rev. A 34

(1986)

1538 ;

d)

DEANE A. E., KNOBLOCH E. and TOOMRE J.,

Phys.

Rev. A 36

(1987)

2862.

[9]

CROSS M. C.,

Phys.

Rev. Lett. 57

(1986)

2935 ;

Phys.

Rev. A 38

(1988)

3593 ;

[10] a)

CHANDRASEKHAR S.,

Hydrodynamics

and

Hydromagnetic

Stability (Dover,

New

York)

1981 ;

b)

PLATTEN J. K. and LEGROS J. C., Convection in

Liquids

(Springer,

New

York)

1984.

[11]

MANNEVILLE P. and PIQUEMAL J. M.,

Phys.

Rev. A 28

(1983)

1774.

[12]

BRETHERTON C. S. and SPIEGEL E. A.,

Phys.

Lett. 96A

(1983)

152.

[13] a)

SHRAIMAN B.,

Phys.

Rev. A 36

(1987)

261 ;

b)

ROSENBLUTH M. N., BERK H. L., DOXAS I. and HORTON W.,

Phys.

Fluids 30

(1987)

2636 ;

c)

YOUNG W., PUMIR A. and POMEAU Y.,

Phys.

Fluids A1

(1989)

462.

[14]

ACRIVOS A. and GODDARD J. D., J. Fluid Mech. 23

(1965)

273.

[15] a)

THUAL O. and FAUVE S., J.

Phys.

France 49

(1988)

1829 ;

Références

Documents relatifs

The bold periodic orbits near the center correspond to the lower and upper stability boundaries: in particular, all periodic traveling wave solutions of (2.1) corresponding to

These crises may be the disruptive entries of newcomers into a tightly regulated market.. Most of the time, these bring with them a new technology or a radically

In addition to per- formance considerations, ontology learning tools need to be fully integrated into the knowledge engineering life-cycle, working in the background and providing

Among these models, intestinal perfusion is the most common experiment used to study the in vivo drug permeability and intestinal metabolism in different regions of the

read all the information under the video and then do the 3 exercises and check your answers.. And you, what are you doing now apart from

Otherwise I’m going to stay at home and watch television. ) $ Jack and I are going to the Theatre.. There is a show of Mr

Atlantic Provinces Library Association Lisa Goddard.. Memorial University Libraries

Current French breeding schemes are pyramidal: at bottom, a large number of sheep breeders are only users of genetic progress, above, a selection nucleus is composed of