• Aucun résultat trouvé

Kinetic roughening by exceptional fluctuations

N/A
N/A
Protected

Academic year: 2021

Partager "Kinetic roughening by exceptional fluctuations"

Copied!
5
0
0

Texte intégral

(1)

HAL Id: jpa-00246306

https://hal.archives-ouvertes.fr/jpa-00246306

Submitted on 1 Jan 1991

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.

Kinetic roughening by exceptional fluctuations

Joachim Krug

To cite this version:

Joachim Krug. Kinetic roughening by exceptional fluctuations. Journal de Physique I, EDP Sciences,

1991, 1 (1), pp.9-12. �10.1051/jp1:1991112�. �jpa-00246306�

(2)

J

Phys.

I 1

(1991)

9-12 JANVIER1991, PAGE 9

Classification

Pliysics

Abstracts

05.40 61.50C 05.70L

Show Communication

Kinetic roughening by exceptional fluctuations

Joachim

Krug(*)

Theoretische

Physik, Ludwig

Maximilians Universitlt, TheresiensIrasse 37, D-80W Miinchen 2, FR.G.

(Receii,ed

26

September

199fl

accepted

30 October1990)

Abstract.

Using simple scaling

argumenIs it is shown that a

sufficiently slowly decaying

power law noise distnbuIion leads Io nonuniversal

scaling properties

for a

kinetically roughened

interface.

Good agreement is found with recent simulations of

Zhang (J Phys.

France 51

(1990) 2129).

In a recent paper

Zhang [I] presented

numerical evidence that the

scaling properties

of an interface driven

by

an uncorrelated random local

growth

rate

~(x,

t with a power law distribution

PI??)

"

~ll~~"~~~,

ll

2

1

(1)

depend continuously

on the noise

exponent

j While the effect of power law

spatiotemporal

noise correlations h well understood in the framework of various renormalization group treatments of the

problem [2],

the influence of the noise disnibufion comes as a

surprise,

since such

microscopic

details are

traditionally expected

to be irrelevant for the

large scale, long

time

properties.

It is the

purpose

of this note to

propose

a

simple physical

mechanism that could account for the observed

nonuniversal behavior.

Specifically,

I use

scaling arguments

in the

spirit

of Villain

[3,

4] to derive the

following expression

for the interface

roughness exponent

x as a function of ~,

x =

) 12)

where d denotes the

dimensionality

of the interface. For small ~

(2

< ~ <

3)

this formula is in

good

agreement with the numerical results

ill

and the derivation suggests that it

gives

a

lower bound on ; for all p. The essence of my

approach

is to concentrate on the

largest

noise fluctuations that occur on a

gi,,en length

and time scale [5]. This is motivated

by

the numerical

[I]

and

experimental

[6] observation that the interface advances in occasional

large

thrusts which

then

rapidly spread

in the lateral direction.

(*)

Present address IBM Thomas J. Watson Research CenIer, P O. Box 218, Yorktown

HeighIs,

New York 10598, U-S-A-

(3)

lo JOURNAL DE PHYSIQUE I N°1

The model

investigated by Zhang

is a discrete version of the

Kardar-Parisi-Zhang (KPZ)

equa-

tion

[7,

8]

~~ "~~~

~

~~~~~

~ ~' ~~~

The

height

of the interface above a d-dimensional reference

plane

at time t is

given by

a

single

valued

height

function

h(x, t).

The first term on the

right

hand side of

(3)

describes

smoothing

due to an interfacial tension v and the

second,

nonlinear term reflects the lateral

displacement dynamics [7, 8].

The stochastic

growth

rate

~(x, t)

has a nonzero mean which

gives

the "bare"

interface

velocity

in the absence of

fluctuations,

and short range correlations in space and time. I have in mind a discretization of

(3)

such that the random

height

increments at different

(discrete) positions

and times are drawn

independently

from the distribution

(I).

All

lengths

and times are

measured in units of the discretization.

The

quantity

of interest h the

height

difference correlation function

Gllx x'l,t)

=

ilhlx,t) h(x',t)li 14)

where the average extends over all realizations of the process with a flat initial

configuration, h(x, 0)

= 0. Scale invariant

roughness implies

the

scaling

form [8]

Glr, t)

=

r~g lr~ /t) 15)

for r, t >

I,

with g

ix

-

0)

= const.,

g(x

-

co)

~w

x~X'~.

This means that the interface is

rough,

I.e. the

typical amplitude ii

of transverse fluctuations increases

algebraically

with the

distance, ii

~w

rX,

on

length

scales below a

dynamical

correlation

length (jj (t)

~w

t~'~.

The nonlinear term

in

(3)

enforces the

scaling

relation [9] X + z = 2. For a one-dimensional interface the continuum

equation (3)

satisfies detailed balance

[7],

which

implies

x

=

1/2

and hence z

=

3/2.

The

question

at stake is the

degree

of

universality

of the continuum results. The

predictions

x

=

1/2,

z

=

3/2

have been verified [8]

by

many simulations of dhcrete

models,

where the noise h

typically

drawn from a distribution of bounded

support.

In contrast, the simulations of

Zhang [I]

for the noise

distribution

(I)

indicate that x increases

continuously

with

decreasing

jt,

reaching

the

limiting

value x

= I at the

point

~ = 2 where the distribution ceases to have a finite second moment and hence any continuum

description

must break down.

Zhang

argues that these results may be

relevant to several recent

experiments [6, lo,

11] on one-dimensional

moving

interfaces where

roughness exponents

x >

1/2

have been observed. In the

following

I

attempt

to

explain

his

findings.

We consider a

patch

of the interface Of linear extension r and assume that the

roughness

is

stationary

on the scale r, I-e- r «

(jj it ).

The number of random events within the

patch

in a time interval Of duration T is N

=

r~T.

We want to estimate the size

~max(N)

of the

largest

of these events.

Clearly

Prob

(~max(N)

<

~)

=

(Prob(~

< x)16'~

=

(1- x~"16'~ (6)

which

yields,

for

large

N

(llmaxlN))

'~

N~'" 17j

This corrects

Zhang's equation (3).

The time

required

to

spontaneously

create a fluctuation of transverse

amplitude ii

is thus

Tc ~w

( [ /r~. (8)

Once created the fluctuations

sprcads laterally

across the

patch,

driven

by

the nonlinear term in

(3). Balancing

this term

against

the timc dcrivative we obtain the

spreading

time scale

[4,

8]

Ts ~

r~ Iii. (9)

(4)

N° I KINETIC ROUGHENING BY EXCEPTIONAL FLUCTUATIONS II

In a

Stationary

situation the

largest

fluctuations that contribute to the

roughness

are those with Tc ~w T~.

Larger

fluctuations will

occasionally

appear, but

they spread

too

rapidly

to have a

signifi-

cant statistical

weight.

Hence the

spontaneously

created

large

fluctuations

give

rise to a transverse

roughness 11

~

rX with the

roughness exponent given by (2).

As stated the

argument completely neglects

the cumulative effect of small

fluctuations,

which becomes all

important

for noise distributions of finite

support.

It is therefore not

surprhing

that x as

given by (2)

vanishes for jt - co instead of

tending

to some

positive

limit

(e,g.,

x =

1/2

for d =

I).

For a power law noise distribution the

roughness generated by exceptional

fluctuations

superimposes

the cumulative

roughness

and dominates it for small jt. I

consequently expect (2)

to be more accurate for smaller values of ~ and to

provide

a lower bound on the true

roughness exponent

for all p.

This is borne out

by

a

comparison

with

Zhang's

simulations of one-dimensional interfaces

(d

=

I).

For 2 < p < 3 my formula is within the error bars of the numerical results. In

particular,

the

limiting

behavior x

- I for jt

- 2 is

reproduced,

and

x(jt

=

3)

=

3/4

as

compared

to the numerical value x

= 0.75 + 0.01. For d = I the

roughening exponent

for bounded noise is X =

1/2

and hence

(2)

cannot be valid for jt > 5. In fact the

prediction x(jt

=

4)

=

3/5

is

already significantly

lower than the numerical result x

= 0.66 + 0.02.

Zhang

also observes that the

exponent identity

[9] x + z = 2 holds

independently

of jt. Here this relation is inherent in the

expression (9)

for the

spreading time,

which

implies

Ts

~w

r~ with z

= 2 x.

As a check of the

present scaling approach

I now consider the linearized KPZ

(or

Edwards- Wilkinson

[12]) equation,

I,e.

(3)

with 1

= 0. It is

trivially solvable,

with the result that x = xo =

(2 d)/2

and z = zo = 2 for any noise distribution with a finite second moment. The linear

problem

thus has a

higher degree

of

universality

than the nonlinear one, lb see how this comes

about,

we

repeat

our

argument

for the linear case. In the absence of lateral

growth

fluctuations

spread diffusively,

so

(9)

is

replaced by

Ts

~w

r2.

More

importantly,

the

integrated height

of a

spreading

fluctuation is conserved under the diffusive

dynamics fih/fit

=

vv~h,

and hence in contrast to the nonlinear case the

amplitude

of the fluctuation decreases

during

the

spreading

process. After the

spreading

time Ts the

amplitude

of a fluctuation created

by

a noise event of size ~max is

(i

+~

lJmax/Td.

As a

consequence (8)

is

replaced by

T~ +~

( (r("~

~)~ and

balancing

with Ts +~

r2 yields

x =

(2 (~ l)d)/~.

This

exponent

is smaller than the continuum

prediction

Xo =

(2 d)/2

whenever jt >

2, I-e-,

whenever the distribution

(I)

has a finite second moment.

We therefore recover the

(obvious)

conclusion that the linear

theory

is insensitive to the none distribution. The same

argument applies

to linear

equations

[4] with a

general

relaxation term

V2~h.

Summarizing,

I have demonstrated how

continuously varying

kinetic

roughness exponents

can arise from the

exceptionally large

fluctuations that are characteristic of the power law noise dis- tribution

(I).

It would be of interest to extend the simulations of

Zhang [I]

to

higher

substrate dimensions in order to test the

dimensionality dependence predicted by (2). Concerning

the ex-

periments [6, lo, 11]

that motivated

Zhang's original study

the

question

remains of how to relate the noise distribution

exponent

to some

physical [6,

11] or

biological [10] properties

of these sys-

tems.

Acknowledgements.

thank Y.-C.

Zhang,

T Vicsek and J. Villain for

communicating

their results

prior

to

publication.

This work was

supported by

Deutsche

Forschungsgemeinschaft.

(5)

12 JOURNAL DE PI.IYSIQUE I N° I

References

[Ii

ZHANG Y.-C., J Pl~ys. France Sl

(1990)

2129.

[2] MEDINA

E.,

HWA

T,

KARDAR M. and ZHANG

Y.-C., Phys.

Rev A39

(1989)

3o53 NAITERMANN T,

Europhys.

Lett. 4

(1987)

1241

HALPIN-HEALY

T, Phys,

Rev A42

(1990)

7 Ii- [3] VILLAIN

J.,

I

Phys.

Lett, Frarre 43

(1982)

L551

GmNSrEIN G. and AIR

S,Xl, Phys.

Rev Leit. 49

(1982)

685.

[4] VILLAIN

J.,

to appear in the

proceedings

of the

Carg£se

Summer School on Nonlinear

phenomena

related to

growth

and form.

[5j The

significance

of the

largest

fluctuation on a

given length

scale has been noted

previously

in the

context of deterministic

growth

on

rough

substrates, see TANG C., ALEXANDER S. and BRUINSMA

R., Phys.

Rev Lett. 64

(1990)

772, and section 5A of [8].

[6j RUBIO M. A., EDWARDS C. A., DOUGHERiY A. and GOLLUB J.

P, Phys.

Rev Lett. 63

(1989)

1685 RUBIO M. A~, DOUGHERrY A. and GOLLUB J. P, Pliys. Rev Lett. 65

(1990)

1389.

[7j KARDAR M., PARISI G. and ZHANG Y.-C.,

Phys.

Rev Lett. 56

(1986)

889.

[8] For a recent review see KRUG J. and SPOHN

H.,

in Solids Far From

-Equilibrium

:

Growth, Morphology

and Defects, Ed. C. Gear£ctle

(Cambridge University Press)

1990.

[9] KRUG J.,

Phys,

Rev A36

(1987

5465.

[10] VICSEK T, CSERzO M. and HoRvhTH V

K., preprint.

iii

HoRvhTH V K., FAMILY F and VICSEK T,

Phys.

Rev Lett. 65

(1990) 1388,

and

preprint.

[12] EDWARDS S. E and WILKINSON D.

R.,

hoc. Roy. Sac, London A381

(1982)

17 FAMILY

E,

J

Phys.

A19

(1986)

L441.

Cet article a 6t6

imprim6

avec le Macro

Package

"Editions de

Physique

Avril 1990".

Références

Documents relatifs

Zeng and Chiu’s weighting scheme as described in section 1.4 is firstly used in an attempt to describe the indentation-unloading curve (in fact, the upper 50% of the unloading

We demonstrate the existence of a simple scaling form which describes the crossover from anomalous to Gaussian exponents as a function of cutoff in the Zhang model of surface

diffraction pattem obtained after the growth of a thick silicon film m the temperature range 300 °C-400 °C, is very different from that given in figure 1. The spotty diagram

Moreover, since the random times T n are sums of positive variables with finite expectation, one can prove, using the arguments in [3], that (n lnn) − 1/2 Y (n · ) converges to a

Free boundary problems, front propagation, combustion, Arrhenius law, travelling wave solutions, periodic solutions, homogenization, curvature effects, asymptotic analysis.. During

Consequently, we are not interested in, for example, a visible condition that is constant in every marking of the composed system since this would be a situation of (local)

In summary, we have proposed a growing network model based on the mechanism of the step-by-step random walk, which generates a power-law clique- degree distribution without

However, the cost function depends on the position of the agent in economic configuration space and reciprocally, the motion of the agents in economic configuration space depends