HAL Id: jpa-00247126
https://hal.archives-ouvertes.fr/jpa-00247126
Submitted on 1 Jan 1995
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.
Spatial and Nearest Neighbor Correlations
Norbert Masbaum
To cite this version:
Norbert Masbaum. Simulation of Ostwald Ripening in Two Dimensions: Spatial and Nearest Neighbor Correlations. Journal de Physique I, EDP Sciences, 1995, 5 (9), pp.1143-1159. �10.1051/jp1:1995188�.
�jpa-00247126�
Classification Physics Abstracts
64.60My 64.60Cn 64.75+g
Simulation of Ostwald Ripening in Two Dimensions: Spatial and Nearest Neighbor Correlations
Norbert Masbaunl
Mathematisches Institut, Universitàt zu KôIn, 50931 KôIn, GerJnany
(Received
6 May 1995, received in final form 20 May 1995, accepted 2 June1995)
Abstract. Recent experimental results on fate stage coarsening
(Ostwald ripening)
in two di- mensionswere reinvestigated by means of numerical siJnulation using the Cahn-Hilliard equation
(mortel B).
We determine the spatial partiele-partide and charge-charge correlation functionsaccording to the experimental results of Krichevsky and Stavans. We find that our numerical results correspond well with these experiments. We aise investigate partiale-partiale correlations
between nearest neighbors
(defined
by the Voronoi diagram of partialecentroids)
following theexperiments of Seul, Morgan and Sire and compare our numerical results with their Maximum
Entropy mortel of Ostwald ripening.
1. Introduction
In the fate stage of
phase
separation in abinary
mixture thelarge partides
of the nlinorityphase
growby
material diffusion at the expense of theshrinking
small ones. This process,generally
known as Ostwaldripening iii,
is drivenby
the reduction of the total interface between theparticle
and the nlatrixphase.
Since 1961, whenLifshitz, Slyozov
andWagner [2,3]
developed
thenowadays
classicalLSW-theory,
thephysical
laws of Ostwaldripening
have beenextensively
studied. While the result obtainedby
LSW is correct in the hmit of avanishing
volume fraction
ç§ of the
droplet phase,
theunderstanding
of Ostwaldripening
for a nonzero ç§is still
incomplete.
In two dimensions the situation is even more
diflicult,
smce theLSW-theory
tutus out to be notapplicable
due to alogarithmic singularity
m the diffusionequation
[4].Moreover,
similar ta the situation in 3D [5], theexisting
2D-theories of Ardell[6,7], Marqusee
[4],Zheng
andGunton [8] and Yao et ai. [9]
give quite
diiferent results for the#~dependence of,
for instance, trie asymptotic distribution functionf(p,
ç§) of trie reducedpartiale
radii p =R/(R)
or triegrowth
rateK(ç§)
in triegenerally
observedgrowth
law(R)~
= Ro +K(ç§)
t.From the
experimental
point of view,however,
two-dimensional work has trie great advan- tage ofallowing
a directimaging
of thecoarsening
process.Recently Krichevsky
and Stavans(KS) [10, iii
andSeul, Morgan
and Sire(SMS) [12,13] performed
experiments of Ostwald ripening at diiferent area fractions of thedroplet phase
in thinlayers
of succionitril andLang-
muir filJns of
binary anlphiphilic
systems. Bath groupsanalyzed
their data of thevideotaped
Q Les Editions de Physique 1995
coarsening
process in view ofspatial
correlations of thecoarsening partiales.
These correlationsare
generally
believed toplay
animportant
rote in trieunderstanding
of Ostwaldripening.
In this work we present trie
corresponding
results of numerical simulations on trie basis of trie Cahn-Hilliardequation.
In the last years, modem computer facilities have made this nonlinear diffusionequation popular
for a simulation ofphase
separation.Large
scale computations for the case of Ostwaldripening
have beenperformed by Rogers
and Desai [14] andChakrabarti,
Toral and Gunton
(various
papers, see Ref. [15] for areview).
These worksmainly
deal witha numerical determination of trie
particle
size distributionfunctions,
triegrowth
law and trie structure function. Here we focus on triespatial partiale-partiale correlations,
1-e-, trieprob- ability
offinding
otherparticles
in trie environment of apartiale,
as well as on triespatial
correlations between the
charges
of thepartiales
[16]. Here thecharge
q=
R~~
of apartiale
is defined
proportionally
to the rate ofchange
of the areaoccupied by
thepaÀlcle, leading
to thequestion
howprobable
it is to findgrowing (or shrinking) partiales
in the environment ofgrowing (or shrinking) particles.
We conlpare thenumerically
obtained correlation functions with triecorresponding experimental
results of KS[10, Iii.
In addition we follow trie work of SMS and
analyze
our numericalcoarsening
data in view of correlations between nearestneighbors.
SMSdeveloped
a maximum entropytheory (ME)
for these correlations
leading
to laws similar to the famous Lewis law and the Aboav-Weaire law of cellular patterns. We will see that our numerical results show somediscrepancies
with thistheory.
A diiferent and
numerically
less expensive method for the simulation of Ostwaldripening
bas been
developed by
Voorhees and GlicksmanIi?i.
Instead ofsolving
for trie timedependent
diffusion field
according
to trie Cahn-Hilliardapproach
used in our workthey developed
analgorithm
eut of themonopole approximation
of thequasistationary
diffusionequation,
which allows to calculate the individualgrowth
rate of eachpartiale provided
that trie sizes and locations of ailpartiales
aregiven. Recently
this formalism has been extended to the so-calleddipole approximation,
which aise considers themoving
ofpartiales
in the process ofcoarsening (which
can be observed inexperiments [1Ii). Recently
thisapproach
has been usedby
Akaiwa and Voorhees [181 and Akaiwa and Meiron [19] for a determination ofspatial
correlations in 3D and 2D,respectively.
The 2D-results will be shown later on in context with the correlationfunctions we obtained.
The paper is structured as follows: the second section
gives
adescription
of the simulationtechnique.
In Section3.1,
we compare the results of the numerical simulationsconcerning
thespatial partiale-partiale
andcharge-charge
correlation functions with theexperimental
results of KS. In Section 3.2, we discuss the simulationsanalyzed
in view of nearestneighbor
correla- tionsfollowmg
SMS and compare the results with their MaximumEntropy
model.Finally,
in Section4,
we summanze our results and give a conclusion.2. Trie Model and trie Numerical Method
In the framework of the Cahn-Hilliard model [20] the
phase
separation process in abinary
system is formulatedby
the evolution equationwhere
c(r,t)
represents the conserved order parameter of the concentration of one of the twospecies.
M is amobility
factor and F denotes the total free energy of the systemgenerally
assumed in the
Ginzburg-Laudau
formF =
/lflc)
+ ~l?Cl~ldr (2)
Here
f(c)
is a double-wellpotential
and ~ apositive
parameter related to the interfacial energy.The thermal noise term
~(r,t)
in theLangevin equation (1),
aise known as model B in theliterature,
has beendeveloped by
Cook [21]. Since the numerical simulation of the thermal noise is very timeconsuming,
we will trot consider it in this work.Generally,
the noise isbelieved trot to affect the
growth
of thedroplets
in the fate stages ofphase separation,
at least at low temperatures[14,15].
It should benoted, however,
that a strict validation of thisassumption
is stilloutstanding [25].
It has been shown
by
Grant et ai. [23] that for the Mass of double-wellpotentials f(c)
=~c~ rc~
aftera suitable
rescaling
ofconcentration,
space and time to ~fi, x, and 2r,
respectively, equation (1)
can be put in the dimensionless form)
=
V~(1fi~
-1fi
V~1fi), ifi = 1fi(x,T), (3)
which is
generally
used for simulations of the Cahn-Hilliard model. The solutions of equa- tion(3)
exhibitcoarsening
domains with values of the rescaled concentration ~b close to 1 and -1,respectively,
which are identified with the twophases.
At the interfaces a steepgradient
of the concentration cari be observed.
We salve equation
(3)
on a square latticeusing
anexpirait
Euler method as describedby Rogers
et ai. [22]. In order to reduceboundary
eifectsperiodic boundary
conditions wereemployed
in that scheme. As discussed in reference [22] the mesh size hz has to be chosen smallenough
to reflect thesharp
concentrationgradient
at thephase
boundaries. As a consequence, the time step ht has to be chosencarefully
m order to avoid instabilities of the numericalscheme
[22].
We workedsuccessfully
with the values hz= 1 and AT
= 0.04 in ail simulations
reported
in this work.As
usual,
weidentify
thepartiales
with the clusters ofgridpoints
with a concentration ~fi < 0.In view of a determination of
dR/dr,
1-e-, the variation of thepartiale
size, the radius R of apartiale
has to becarefully
defined. It tumed eut to be necessary netonly
to count the cluster size but aise to look for the concentration ~fi at thegrid points belonging
to a cluster. Wefound the definition
~j~2
~
~
~fi~
z
an
appropriate
measure and calculated the variation of sizeby dR(r)/dr
GÎ(R(r
+5Ar) R(r)) /5Ar.
The location of apartiale
was definedby
the center of massrcm =
£
~~~
~fiz(rz)rz~
For the
study
ofpartiale
coarsenmg each simulation was startedby
an initial concentration field ~fi(x, r =0), consisting
of circular clusters with concentration values ~b= -1 distributed
on the
grid,
andsetting
~fi to +1 at ailgrid points
outside thepartiales. By varying
the number ofpartiales
diiferent values of the area fraction ç§ of thepartiale phase
could be studied. Theadvantage
of this method(see [14,15]
for a diiferentapproach)
is thepossibility
ofcreating
anintial condition with a
given
distribution ofpartide
radii. As mentionedabove,
anasymptotic
partiale
size distribution(depending
on ç§) is reached in the fate stage ofcoarsening.
Here weTable I. Parameters ~lsed in the sim~llations.
ç§ d furax size
0.13 0.3 23.000 2048 x 2048
0.25 0.3 30.000 1024 x 2048
0.40 0.15 23.000 1024x2048
follow an idea of Akaiwa and Voorhees [18]
starting
ourlarge
scale simulations withpartiale
size distributions near to the
asymptotic
form(known
fromprevious
smaller ruas at the samearea
fraction).
It isgenerally
found thatthereby
thescaling regime
is reached much faster thanby starting
with anarbitrary
distribution.Another important parameter for the initial condition of the simulation is the mean
partiale
radius(Ro).
On the onehard,
a small(Roi
is desired in order to be able to start with alarge
number of
partiales
to reduce the scatter in the data. On the otherhard,
it is well known [24]that the Cahn-Hilliard model reflects the
principle
of a criticalpartiale
radius in the sense that smallerpartiales
are trot stable in the matrix and tend to dissolve. A reasonable value was found to be(Ro)
" 6.2, which is used
throughout
ail simulations.Using
this meanparticle radius,
wesuccessively
tilt up thegrid
withpartides according
to the distribution function at random locations until theprescribed
area fraction is reached.Following
a method of Akaiwa and Voorhees [18], we do trotonly
exdudepartide overlap
but more than that create a
depletion
zone, 1-e-, aspherical
shell of thickness d R with aparameter d around every
partide
of radiusR,
which is forbidden for otherpartides.
In agreement with[18,19]
we did trot find an influence of thesedepletion
zones in the initialcondition on the fate stage
scaling
functions. Theadvantage
ofusing depletion
zones is,however,
thatthey
hinderpartiales
toinstantly
coalesce so that we cari start with apartide
number as
large
aspossible.
Forlarger
values of the area fraction ç§ the parameter d isnaturally
limited to smallervalues,
otherwise theparticles
could trot besuccessfully placed.
In Table I ail parameters for the diiferent simulations are summarized. The mainproblem
instudying
trie statistical laws of Ostwald
ripening
in trieasymptotic
fate stageby
numerical simulation is thedecreasing
number ofpartides, causing problems
in the statistics of theanalyzed
quantity.One method to overcome this
problem
is to average over diiferent ruas with initial randomconfigurations [14,15].
Since we had theopportunity
ofusing large
computer memory, wepreferred
in this work a second method ofperforming single
ruas of verylarge
systems(Tab. I).
The
advantage
of this methodis,
of course, the reduction of finite sizeeifects,
which tumed out to be aproblem
in thestudy
ofspatial
correlations at small system sizes. The simulationsperformed
in thiswork, however,
start with+~ 2600 3100
partides (slightly diifering
with thearea fraction ç§) and end up with
+~ 600 850
particles
at furax,allowing
a preciseanalysis
of the diiferent correlation functions even in the fate stage.The calculations have been carried out on
partitions
of 128 or 256 processors of the 1024 nodeparallel
computerGcel-3/1024
of PARSYTECusing geometric parallelisation,
1-e-,dividing
the
grid
in partsassigned
to the dilferent processors. It turned out that with thegrid
sizes and the number of processors that we used the finite diiference scheme could be calculated withan
eflicieny
of more than90il,
wheredividing
into squares does asignificantly
bernerjob
thandividing
intostripes [25]. Finally,
ii should be mentioned that due to technicalprovisions
for data reduction the cluster identification for the calculatiolî ofpartiale
sizes and locations has also beenperformed
inparallel, using
atechnique
similar to trie host model of Hackl [26] basedon the
Hoshen-Kopelmann algorithm
[27].3. Results of Numerical Simulation
3.1. SPATIAL CORRELATIONS. In this
chapter
we present trie results ofspatial particle- partiale
andcharge-charge
correlationsanalogously
to theexperimental
work ofKrichevsky
and Stavans(KS) [loi.
LikeKrichevsky
and Stavaus westudy
the two area fractions ç§= 0.13
and ç§ = 0.40. For the calculation of the
spatial
correlation functions we follow the definitiongiven by
Akaiwa and Voorhees [18], 1-e-, the radial distribution functionGir)
is defined as theratio of the number of
partiales
in a circular shell of radius r and thickness drsurrounding
a
partiale
to the number ofpartiales
in trie shellexpected
from thepartiale density
in the system. The two type correlation functions are defined in the same mariner.Gir, LS),
forexample,
measures theprobability
offinding
a smallpartiale (S)
at the distance r of alarge partide IL).
The idea ofdividing
thepartides
in two classes oflarge (R
>(R))
and smallpartiales (R
<IRI)
has been introducedby
KS. We follow this concept in order to allow a directcomparison
with theirexperimental
results.In
Figure
1 the radial distribution functionG(r)
is shown for the two values of trie area fractionç§.
Throughout
thiswork,
trie distance r between theparticles
is measured in units of the mean radiusIRI.
Wegenerally
found ascaling
in the correlation functions in triefate stages of the simulations
(Fig. l), allowing
to follow KS inaveraging
over 10 times(r
= 5000,7000,..., 23000)
in order to reduce the scatter in the data. Weperformed
this av-eraging throughout
this work.Figure
1 shows that trie correlationfunctions, diifering strongly
for trie two area
fractions,
could be wellreproduced by
trie numerical simulation of trie Cahn- Hilliard model.Very
similar results bave been obtainedby
Akaiwa and Meiron(AM) [19], calculating
amultiple approximation
of thequasistationary
diffusion equation. Since theirmonopole
andcorresponding dipole
approximation results are very close [19] wereplot only
thedipole
results in ourfigures.
There are two
typical
values m the radial distribution functionG(r).
That is first trie maximumdefining
trie mean distance to trie nextparticle, being
more distinct athigher
areafractions with
partiales packed
moredensely,
and second trie size of thedepletion
zone sur-rounding
apartide.
Bothtypical
distances cari bereproduced
veryprecisely by
simulation.A doser look can be taken
by studying
trie correlations between triepartiales
of diiferentor of same Mass as shown in
Figure
2.Again
we observe agood
agreement of our simulation with trieexperiment
and the simulation method of AM[28].
At trie lower area fraction trie two correlation functions arequalitatively
similarshowing slightly higher
values forparticles
of diiferent size
compared
withpartiales
of trie samesize,
while at thehigh
area fraction ofç§ = 0A this diiference is very strong,
showing
ahigh probability
offinding
smallparticles
in trie environment oflarge
ores and ~ice versa. This of course is a consequence of trie fact thatlarge partiales
grow at the expense of small ones in trieneighborhood.
According
to the definition of the correlation function it froids thatG(r, LS)
=
G(r, SL)
=
G(r,
SLILS).
Like AM we found a strongdiiference, however,
in the correlations of twolarge partides
and two smallpartides,
1-e-,G(r, LL)
andG(r, SS).
While weaveraged
these two functions inFigure
2 in order to compare them with triecorresponding
results ofKS,
we show both functions inFigure
3together
with trie simulation results of AM.Qualitatively
both methodsgive
the sameresults, especially
in view of trieposition
of the maximum and triesize of the
depletion
zone. On trie otherhaud,
thequantitative
diiferences in trie correlation values exceed the statistical errorssignificantly,
which isprobably
a consequence of trie diiferentapproximation
methods forstudying
themultiparticle
diffusionproblem
of Ostwaldripening.
We now tum to the
charge-charge correlations,
1-e-, to thespatial
correlations ofgrowing
andi-o a
O 1
~ ~'
a fl~is work
/
o KS
é
i AM
~'~
O
~ ~
~~~~u
, ~
o i
1
~ ,
o /
~
a)
~'~ 0 2 4 6 r&4.40
0
,
o ~=5000
° ~=7000
o ~=9ooo
6 ~=I1000
0.5 ~ ~=13000
V ~=15000
~ ~=17000
+ ~=19000
x ~=210W
~ ~ *
w23000
b)
0 2 4 6 rFig. 1.
a)
Numerical siJnulation of trie radialpartiale-partiale
correlation functionG(r) (r
in units of < R>)
for the area fraction # = 0.13(empty
symbols and dashed fine) and # = 0.40(fuit
symbolsand solid
une)
in the fate stage. The experiJnental results by KS [10] as wellas the numerical results by AM [19] are also shown; b) The salue correlation function
G(r)
Jneasured at ter times m trie fatestage of simulation, showmg a dynaJnical scaling. The solid une corresponds to the averaging shown
m
a).
shrinking partides.
Here thegrowing partides
have a positive(+) charge,
q= R
~~,
whiledr the
shrinking partides
havenegative charge (-).
In order to make trie resultscomparable
wefollow KS and derme trie correlation of the
charges by G(r,...)
=
(q(0)q'(r)), assuming
triecharge q'
to beuniformly
distributedalong
theboundary
of theparticles.
The numerical results for the correlations of two
particles
of samesign
and twopartiales
of diiferent
sign
are shown mFigure
4,together
with theexperimental
results of KS[29].
For bath area fractions there is a strong correlation of
partiales
of oppositecharge, especially
&4.13
j
i.o
,j'
OE ©fl~is work
0.5 > ~ KS
' - fl~is work
-- KS
AM
a)
0 2 4 6 r2.0 w a dûs work
~'~~
~Î
~ls
work1.5 ~~j
i o
o.5
b) ~~0
2 4 6 r
Fig. 2. The radial correlation function
G(r, LL/SS) (empty syJnbols)
for two partides of the same dass and the correlation functionG(r,SL/LS) (fuit syJnbols)
for two partiales of different class mcomparison with the expenJnental results of KS [10]. The nuJnerical results of AM [19] are aise shown.
interesting
in the case of ç§= 0.13 where such a distinct maximum could net be observed for the S-L correlation.
Obviously
there is a strongscreening
eifect even for small areafractions,
1-e-,growing (shrinking) partides
ar surroundedby
a shell ofshrinking (growing) particles.
Trie
good correspondence
between trie numerical simulation and trieexperiment
indicates that trie Cahn-Hilliard model is alsoappropriate
to describe triecharge-charge
correlations.Again
trie size of triedepletion
zone as well as trieposition
of the maximum, 1-e-, trie first cell of interactions, arereproduced
veryprecisely. Finally,
trie four diiferentcharge-charge
correlation functions
G(r, ++), G(r, -), G(r, -+), G(r,
+- are shown mFigure
5. Notethat,
unlike
G(r,LS)
=
G(r, SL),
the correlation functionsG(r, +-)
andG(r, -+)
are diiferent due to theunsymmetric
definitiongiven by
KS.However,
it isinteresting
to see that the correlations of twogrowing
and twoshrinking partides
are very close, which was not the case with trie L L&4.40
1.5 çq
pf
~
Ù~,
» ~
,
l-O ,
' w
~ --a fl~'sw rk
, °
~'~
j
~ ~~lwork
'
-- AM
i i
'W
~ r
~~
~~
0 2 4 ~
0m0.13
,a
?'~' ',
' S,
,~-~ ~,
i-o
w B
a ~ÎÎ~Î~~~~
'
- fl~is work
Î
i
-_ AM
j i
5
1 w
i i
i i
i w
i i
i~
~,
~
°.°
o 2 4 ~ ~
Fig. 3. The radial large-large correlation function G(r, LL)
(empty
symbols) and the small-small correlation functionG(r,
SS) (fuit syJnbols). The numerical results of AM [19] are also shown.or S S correlations.
3.2. CORRELATIONS oF NEAREST NEIGHBORS. In this section we present results of numer-
ical simulation
according
to the correlations ofadjacent partides during coarsening,
motivatedby
recentexperimental
and theoretical results ofSeul, Morgan
and Sire(SMS) [12]. Following
SMS,
twopartiales
are defined as nearestneighbors
ifthey belong
toneighboring
cells of trie cellular patterngiven by
the Voronoidiagram
of the partiales centroids. Anexample
of this construction [30] is shown inFigure
6. As in the experiments ofSMS,
we simulated the area fraction ç§= 0.25, but we also looked for an area fraction
dependence
of ail measured quantitiesby analyzing
the simulationsconcerning
ç§ = 0.13 andç§ = 0.40 studied in the last section now in view of nearest
neighbor
correlations.TO ascertain that there is a
dynamical scaling
in trieconfigurations
ofpartiales
in trie fate2.° OE#0.40
5
1.o
0.5 * ~ fl~is work
> ~ KS
- fl~is work
-- KS
a)
~ ~ 0 2 4 6 8 r0m0.13
2.0
1.5
1.0
@~
i
'
~
OE B ~s work
0.5 > ~ KS
- fins work
-- KS
b) ~~0
2 4 6 8 r
Fig. 4. The radial charge-charge correlation functions
G(r,
+ +/ -)
andG(r,
+/ +)
forpartides with charges of salue sign
(eJnpty syJnbols)
and opposite sign(fuit symbols)
in comparison with the experimental results of KS [10].stages of Ostwald
ripening
we measured trieprobability
densitiesP(d
N N/ (d
N N of the normed distances dNN between trie boundaries ofadjacent partiales.
In agreement with yetunpublished
results
by
SMS [12] we found that the distribution becomes timeindependent
in the fate stage(Fig. 7).
We observed that thescaling
functions showed nosignificant dependence
on the areafraction with a characteristic maximum at
approximately 0.8(dNN),
and almost no pairs ofpartiales
with distanceslarger
than2.5(dNN).
A more detailed view on the structure of the relations between nearest
neighbors
will begiven
inFigure 8,
where the distributionPin)
of the coordination number n, le-, the mean number ofneighbors
of apartiale,
areplotted. Especially
for the coordination number n= 6
we found a
significant
lowerprobability
in trie simulation (ç§=
0.25)
than in the experiments of SMS. Moreover, we obtained the value 0.84~0.01 for the second moment ~12 "L(n 6)~ pin),
aOEA.40
~ a G(r,+ +)
~'~
> ~ G(r,- -)
- G(r,- +)
2 o -- G(r,+ -)
1.5
o
o.5
a)
0 2 4 6 8 rOEA.13
~ a G(r ++~
~'~
> ~G(rÎ- -)
- G(r,-+)
~ ~ -- G(r,+-)
1.5
0
aa~~od
o.5
b)
O.O o 2 4 6 8 rFig. 5. The different radial charge-charge correlation functions in the fate stage.
measure for the deviation from the
hexagonal
orderpin)
e 6, while SMS report smaller valuesranging
from 0.64 to 0.85. The reason for thishigher (hexagonal)
order in thecoarsening
experiments of SMS has to be seen in additionallong-range repulsive
electrostatic interactions between thedroplets
in theamphilic
system. We will come back to this point later. Triedistribution of coordination numbers for the other area fractions are also shown in
Figure
8.We observe a lower
hexagonal
order for the low area fraction ç§= 0.13, while there is no
significant
diiference in the distribution between ç§= 0.25 and ç§
= 0.40. For the second
moment we found ~12
" 1.08 ~ 0.02 for
ç§ = 0.13 and
~12 " 0.86 ~ 0.02 in the case of area
fraction ç§
= 0.40.
We now tum to the maximum entropy
theory (ME)
forpartiale coarsening developed by
Sire and Seul
(SMS)
[13]. Since the pattem formedby
thepolydisperse droplets
are connected via the Voronoi construction to a cellular pattern with each cellcontaining
onedroplet,
trieauthors had the idea to translate a maximum entropy
analysis
of cellular patternsby
RivierFig. 6. Determmation of the nearest neighbors for a
droplet configuration
with an area fraction#
= 0.40. Trie Voronoi diagram
(bold fines)
and trie dual graph, the Delaunay triangulation(fine unes)
are shown. Two partides are defined as nearest neighbors if they belong to adjacent cells, orequivalently, if they are connected by an edge of the Delaunay triangulation.
et ai. [31] into the
droplet
case.Conceming
correlations ofcoarsening droplets,
SMS stated two laws well known from thetheory
of cellular pattems, that is the Lewis lawzn = b +
Ain 6),
xn=
jAnj /jA) j4)
and the Aboav-Weaire law
arum)
= 16
a)in 6)
+ 36 + t12.là)
Here the Lewis law describes a linear relation between the
topological charge
C
= n 6 of a
partiale
and the normed mean area An ofparticles
with coordination number n.The Aboav-Weaire law states that the
product
of the coordination number n times the meancoordination number
min)
of the nneighbors
of apartiale depends linearly
on thetopological charge
C= n 6.
Investigating
these lawsby
numericalsimulation,
we observedscaling
in both cases. Theaveraged
fate stagedependencies
are shown inFigure
9. Our simulation results areprecisely
described
by
thehnearity according
to the Aboav-Weaire law. While SMS measured the constant to a +~ I.I for ç§= 0.25, we obtained a
= 1.15 ~ 0.07, not
significantly depending
on the area fraction ç§.
Moreover,
thevalidity
of the Aboav-Weaire law seems not to be a characteristic of the fate stagespatial partide
correlations as assumedby
SMS. We found thateven a random
configuration
ofpartiales
with the samepartiale
size distribution(like
trie initialconfiguration
used in triesimulations)
follows trie Aboav-Weaire law withhigh precision.
In contrast to trie Aboav-Weaire law a strong
dependence
on trie area fraction can be seenin the case of trie Lewis
law, leading
to ahigher
variation of trie number ofneighbors
with thepartiale
size athigher
area fractions.Moreover, significant
deviations from thelinearity
~i'* -- 0Em0.13
o.8 !." . . 0Em0.25
iÎ +----+ 0Em0.40
0.6
0.4
0.2
1 t 1
~~0.0
1-ù 2.0 dNN/<~j~~~>Fig. 7. Probability densities of trie distances between trie boundaries of nearest neighbors in the fate stage of coarsening for different values of area fraction #.
P(n)
o.5 o
o.4 ,"~, -- &4.13
' "
. . &4.25
o.3 +----+ &4.40
o SMS
0.2
o-1
~'~
2 4 5 6 7 8 9 n
Fig. 8. Probability densities of the coordination numbers
Pin)
m the fate stage of coarsening. TheexperiJnental values of SMS [12] for the area fraction # = 0.25 are aise shown.
could be observed. For that reason we did trot try to determine the constant in
equation (4).
Figure
9shows,
however, that theexperimental
value of SMS could be confirmed in trie case of the area fraction ç§= 0.25.
Analogous
to thestudy
ofspatial
correlations of thepartiales
sizes in Section 3.1 it is alsopossible
tostudy
these size correlations between nearestneighbors
as has beenproposed by
SMS. InFigure
10 the mean normalizedpartide
area(ANN) /(A)
of theneighbors
is showno
i.o a
O 1
~ ~'
a MS Work
'
o KS
é i~
Î AM
. fl~is work g
0.5 1
~~
o
j AM
~ i
o i
i ,
o
),/
~
~°
o 2 ~ ~
&4.40
~ l.0
,
o ~=5000
° ~=7000
o w9000
6 ~=I l0lX~
0.5 ~ ~=13000
V ~=150lX~
~ ~=170lX~
+ W190lX~
x W2IOIX~
* w230lX~
b)
0 2 4 6 rFig. 9. Mean coordination nuJnber
m(n) (Aboav-Weaire
law(5))
and the mean particle area< An >
/
< A >(Lewis
law(4))
of the nearest neighbors of a partiale with topological chargeC = n 6 in the fate stage [32]. The dotted unes reflect trie experimental values of trie slope À(0.2 < < 0.25) determined by SMS
(#
=
0.25).
as a function of the normalized
partide
areaA/(A) (under averaging
over ailpartides
of thesystem). Strong
anticorrelations of thesequantities
could beobserved,
that is smallpartides
are surrounded more
likely by large neighbors
thanby
smallneighbors (and
~ice~ersa),
aphenomenon
that was also found in the experiments of SMS.According
to trie numericalsimulation this screemng eifect is more pronounced at
higher
area fractions withparticles packed
moredensely (Fig. 10).
Combining
trie Aboav-Weaire law(5)
and the Lewis law(4),
SMS deduced theanalytic expression
(ANN
1a(1 ~)
+ À~12 AIA)
z ~ 6 +lx 1) là
' ~IA)
~~~<ANN>/<A>
-- &4.13
",,
. . &4.25
~',,
+----+ &4.40w
",~
Î",-~_
".
'---x-__
1
~'~~o.o o.5 o .5
Fig. 10. Numerical results for the correlations of the area A of a partide and the mean area
< ANN > of its nearest neighbors m the fate stage [32].
for these correlations shown in
Figure
10. However, we were unable to confirm this expressionon the basis of our numerical
results, probably
a consequence of the fact thatalready
trielinearity
in trie Lewis law was violated [34].Finally
weanalyzed
our simulationsaccording
to aquestion
raisedby
SMS[13],
that is whether the von Neumann law of cellular patterns is alsoappropriate
to describe thepartiale
coarsening in the fate stage. This law states that the
growth
of a cell isproportional
to the number of itsadjacent
cells minus 6, 1-e-, cells with more than 6neighbors
grow, cells with less than 6neighbors shrink,
and cells with 6neighbors
maintain their area. In trie case oftwo dimensional soap froth
growing by
gas diffusionthrough
trie watts of triecells,
this law bas been deducedby
von Neumann [33] aud it is also verypopular
fordescribing grain growth
in
metallurgic
cellular structures. Translated into trie framework of trie MEtheory
ofSMS,
itgives
triegrowth
law~j"
=k(n 6) (7)
for trie area A of trie
partiales,
thus aproportionality
ofcharge ~~"
andtopological charge
dr
C = n 6. In
particular
the averagegrowth
rate of aparticle
would be determinedonly by
a
topological quantity
of the number ofneighbors. Figure
II shows triecorresponding
results of numerical simulation. Indeed we found ascaling
of ~~with
topological charge
and could confirm alinearity
between thesequantities
with astrolldependence
on the area fraction. It should be noted that there is a
significant
non-zero intercept of the curves,leading
to a moreappropriate
formulation ~~"= const +
k(ç§)(n 6)
of equation(7).
With thiscompletion
dr
of the
growth law, however,
the coarseningdynamics
seem to be well described.4. Discussion
In this work we have studied the
spatial
correlations ofpartiales
in the fate stages of coars-ening by
numerical simulation of the Cahn-Hilliard equation(model B).
A comparison withrecent experiments on 2D Ostwald
ripening [10, iii
shows that within this model thespatial
particle-partiale,
as well as thecharge-charge
correlationfunctions,
can be well understood andreproduced
withhigh
accuracy.Especially
we were able to confirm adynamical scaling
ofthese correlation functions in the fate stage of
particle coarsening. By comparing
the obtainedscaling
functions in the case ofparticle-particle
correlations with simulation results of Akaiwa and Meiron [19] we found that thedipole approximation
of thequasistationary
diffusion equa-tion that
they
usedgives quite
similar results. Seen from theviewpoint
of the numericaleffort,
this simulation method is less expensive than ours, however
accepting
restrictions of circularpartides,
for instance.Especially
athigher
areafractions,
wherepartides
tend tochange
theirshapes,
trieimportance
of thisapproximation
is undear. We believe that a directcompari-
son of both
methods, possibly following
aripening
of trie same initialcondition,
couldhelp
to
darify
thisquestion.
Anothermajor
part of this work was concemed with trie correlation of nearestneighbors, following
anexperimental study
ofSeul, Morgan
and Sire[12].
As inthese experiments, we found
pronounced
anticorrelations in trie sizes ofadjacent partides.
We were
unable, however,
to confirm atheory given by
SMS for theseanticorrelations,
basedon maximum
entropy principles.
Trie numerical simulation showed that even triepostulated
Lewis
law,
1-e-, a lineardependence
ofparticle
size and number ofneighbors,
was violated in contradiction to theexperimental
results of SMS. A reason for this differencemight
be seen in an additional repulsive electrostatic interaction of thepartides
in theamphilic
system, assupposed by
SMS. This conjecture issupported by
thesignificautly higher
values oftopologi-
cal
defects,
i e.,partides
with coordination number n#
6 in the simulation. It is well known~ l'
~ ~~ ~
~Î~~Ù. Î ~
'
'~
~
~Î~~.~~ ~'
/~ ~ ''
:"
~Î~~Ù.~~ ~
~'
Ô-ÔÔ~ ,' ~'
' '
'
.~
~,'
~Ô-ÔÙS ,'
#
' ' / / ' /
~~~~ ~
-3 -2 -1 0 2
n-6
Fig. Il. Relation between the variation of the area
(dAn/dT)
and the topological charge C = n 6of a particle m the fate stage corresponding to the van Neumann law (7) [32].