HAL Id: jpa-00208162
https://hal.archives-ouvertes.fr/jpa-00208162
Submitted on 1 Jan 1974
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, estdestinée 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.
Clusters of atoms coupled by long range interactions
J.P. Gayda, H. Ottavi
To cite this version:
J.P. Gayda, H. Ottavi. Clusters of atoms coupled by long range interactions. Journal de Physique,
1974, 35 (5), pp.393-399. �10.1051/jphys:01974003505039300�. �jpa-00208162�
LE JOURNAL DE PHYSIQUE
CLUSTERS OF ATOMS COUPLED BY LONG RANGE INTERACTIONS
J. P. GAYDA and H. OTTAVI
Département d’Electronique (*),
Université de Provence, Centre deSaint-Jérôme,
13397 Marseille Cedex4,
France(Reçu
le 13 novembre1973)
Résumé. 2014 Nous présentons une étude statistique de la distribution de la taille des amas formés par des atomes répartis aléatoirement dans un milieu continu. Les résultats statistiques sont obtenus
par une méthode de Monte Carlo. r0 mesurant la portée des interactions, nous trouvons qu’un amas
infini apparaît quand le nombre moyen d’atomes, W, contenus dans une sphère de rayon r0, atteint
- une valeur légèrement
supérieure
à 2,6. Pour rendre compte de ces résultats statistiques, nous avonsélaboré des modèles de processus de ramification. Quelques
propriétés
essentielles sont dégagées,et on obtient une expression approchée pour une grande gamme de concentration du nombre d’atomes appartenant à des amas de petites tailles.
Abstract. 2014 We present a study on cluster size distribution for atoms randomly distributed in
a three-dimensional continuous medium. Statistical results are obtained by a Monte Carlo method.
r0 being the range of interaction, it is found that the infinite cluster appears when the mean number of atoms, W, in a sphere of radius r0 reaches a value slightly higher than
2.6.
Besides this, the statis-tical results are compared to the results deduced from branching process models : some essential features are then outlined and an approximate expression is proposed for the number of atoms
belonging to clusters of small sizes, for a large range of concentration.
Classification Physics Abstracts
1.660
1. Introduction.. - In
preceding
papers[1, 2, 3],
we
study
EPR lineshapes
for ions diluted in a diama-gnetic
matrix. Accent is put on theparticular
roleplayed by
clusters of ionscoupled by long
rangeexchange
interactions. - We present here a statisticalstudy
on the distribution of theclusters,
versus concentration. Thisproblem
is one aspect of thegeneral percolation
process(for
abibliography
onpercolation
andphysical applications,
see[4]).
We consider atoms or
points randomly
distributed in a three dimensional continuous medium andcoupled by
an interaction of range ro. The statistical lawsgoverning
the cluster size distributiondepend only
on one parameterW,
the mean number ofpoints
in a
sphere
of radius ’0:where p is the number of
points
per unit volume.Holcomb,
Iwasawa and Roberts(H.
I.R.) [5]
usethe parameter t
proportional
to W :Clusters of infinite size
appear
above the criticalvalue
Wc.
In the three dimensional continuous case, Domb and Dalton[6]
propose the value 2.7 forWc.
This value is an
asymptotic
limit for critical values calculatedby
the series method[7]
in lattices where ro islarger
andlarger.
Anapproach
to the continuouscase is obtained
by
Holcomb and Rehr[8]
who makeuse of a Monte Carlo calculation on a
simple
cubiclattice with ro
equal
to 3 times the lattice parameter.They
obtainWc -
2.4(tc - 0.07).
Morerecently,
the same result if obtained
by
H. I. R.by
a linearextrapolation
of the mean cluster sizes obtainedby
a Monte Carlo method in the continuous case.
The discrepancy
with our value(Wc > 2.6)
willbe discussed in the
following.
Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:01974003505039300
394
In the first part of this paper, we present a method of cluster construction which
gives
statistical resultsdirectly
in the continuous medium. These resultsare called
experimental
results.In the other parts, we use the
analogy
between theexperimental
results and statistical lawsgoverning
a
branching
process to obtain on the onehand,
an evaluation ofWc,
on the otherhand, analytic
expres- sions forcounting
the clusters of small size in alarge
range of concentration.
Notation. - We call the
probability
that apoint belongs
to a cluster of size N,PN,
to a cluster of size N or more,P/,
to a cluster of infinitesize, P 00
= limP;-.
N --+- 00 N
2. The Monte Carlo calculations. - 2.1 THE ME - THOD. - The
principle
of cluster construction is thefollowing:
we choose apoint
and examine thesize of the cluster
containing
it. Forthat,
weexplore
the
sphere
of radius ro centered on it. The random number n ofneighbours
in thesphere
is determinedaccording
to a Poisson law ofparameter W and
theirpositions according
to a statistical law with uniformdensity. (This
needs utilization of random numbers which aregenerated by
amultiplicative congruential method.)
If n =
0,
the choosenpoint belongs
to a clusterof size 1.
If n #
0,
we say we found nneighbours
in the 1 stgeneration.
Then we look to see if thepoints
of the1 st
generation
haveneighbours themselves, by exploring
thesphere
of each one. However we notethat the
points
found in aregion previously explored
must now be eliminated. The
neighbours
so obtainedbelong
to the 2ndgeneration
and the construction of the cluster ispursued
until we find an empty genera- tion. The chosenpoint belongs
then to a cluster of sizeN,
where N is the total number ofpoints
whichhave been found,
plus
theorigin.
Inpractice,
anupper limit for N is
imposed by
thecomputer.
In ourcalculations,
this limit is 256. When the cluster size goesbeyond
thisvalue,
the firstpoint
is said tobelong
to a cluster of size 256 or more.In
fact,
in order to avoid thespherical coordinates,
eachsphere
isprovisionally replaced by
a cube withedges of length
2 ro. Thefilling
of the cube isperformed
in accordance to a Poisson law of parameter
(6/n)
W.But a
point
will beignored
if it fails to fall in thesphere.
So,
to be retained andbelong
to thegeneration
number
k,
apoint
must be :a)
inside thesphere
centered on thepoint
of thegeneration k -
1 from whom it isdescended,
b)
outside thespheres previously explored,
centeredon the
points
of thegeneration k - 1,
c)
outside thespheres
centered on thepoints
ofthe
generations
numbered from 0 to k - 2.FIG. 1. - Construction of a cluster.
These rules are
pictured
infigure
1 where the cons-truction of a cluster is shown up to the 2nd
generation.
The 1 st
generation
contains the twopoints 1,
2.We have found for the
point 1,
the threepossible neighbours 3, 4, 5 ; however, following
thepreceding criteria,
4 and 5 must beeliminated ;
likewise the threeneighbours 6, 7,
8 found forpoint
2 must beeliminated ; finally,
the 2ndgeneration
containsonly
the
point
3.Our method seems less
expensive
than the methodof H. I. R.
Indeed,
in this last case, in order to builda cluster
containing
185points,
it is necessary to examine 4 000candidates ;
in our methodonly
about400 candidates must be
analysed. So,
for W =2.3,
we have been able to obtain five clusters
containing
500
points
or more, and for each studied value ofW,
at least 500 clusters have been built.
2.2 THE RESULTS. - The results are
presented
infigure 2,
under the formThis
representation
mustput
in evidence the critical valuefl.
Indeed when WWc,
all thepoints
arein clusters of finite size and
PN
must tend toward zéro ; on the otherhand,
when W> Wc
a non nullproportion
ofpoints
is in the infinite cluster andPN
has a horizontal asymptote with ordinateP oo(W).
Each value obtained for
PN+
isrepresented
withits error
interval,
the halflength
of which is twicethe standard deviation.
It appears from these results that the
Wc
value iscertainly
located between 2.3 and3 ;
this iscompatible
with the value 2.7
given
in[6].
We have verified that the values of ê
(the
meancluster
size)
calculated from our results are ingood
agreement with the valuesgiven by
H. I. R. It is aconfirmation of the
validity
of the two clusterbuilding
methods.
FIG. 2. - PN = f(N). Dots : experimental results. - results given by a branching process with fertilities depending on the generation number. The value W of the
fertility
of the ancestor isindicated on each curve.
3.
Analogy
with abranching
process. - Our method of cluster construction suggests thedevelopment
ofa
family according
to abranching
process. In thesimplest
process, a man has r sons withprobability
pr
(r
=0, 1, 2...)
and each of his sons has the sameprobabilities
ofhaving
agiven
number of sons ofhis own. The
fertility
of every man isW,
the mean number of sons. If W1,
theprobability
of extinc-tion of the
family
is one; however if W >1,
theprobability
of an infinite descendance becomes nonnull
[9].
- -In the construction of a
cluster,
thefertility
iscertainly
not the same for allpoints.
Inparticular
the
fertility
of the firstpoint
is greater than the fertilities of thefollowing points.
Thus we consider a branch-ing
process more elaborate where thefertility Wk
of a man
depends
on itsgeneration
number k. Let usassume that
Wk
has anasymptotic
limitWa.
Onecan show that :
- If
Wa
>1,
theprobability
of an infinite descen- dance is non null(P 00 i= 0).
The meanpopulation Gk
of the kthgeneration
increasesindefinitely
with k.- If
Wa 1,
the meanpopulation
tends towardzero and
P 00
= 0.- If
Wa
=1, P 00
is null or notdepending
onwhether the convergence of
Wk
is fast or not.In
fact,
for theproblem
ofclusters,
we can convince ourselves that theasymptotic
valueWa (if
itexists)
cannot be greater than one even if W >
Wc’
3 .1
Let gk
be the number ofpoints
in thegeneration
k for a
given trial,
andG,
the mean value of gk.By
definitionand
log Gk
versus k has anasymptotic
direction witha
slope equal
tolog Wa.
3 . 2 On the other
hand,
for anycluster,
the distance between apoint
of the kthgeneration
and theorigin
is at most
kro.
The number ofpoints (belonging
ornot to the
cluster)
inside thesphere
of radiuskro, N(kro),
is an upper bound to the number ofpoints
in the
generations
0 to k of thecluster,
and afortiori
This remains true for the mean values
Then
on a semi-log scale, log G,
versus k is boundedby
a function of k which has anasymptotic
directionwith a null
slope.
Thus theslope log Wa
of the asymp- totic direction oflog Gk
is less than orequal
to zeroso that
3. 3 Furthermore it is easy to show
that,
whenWa 1,
theprobability P 00
that a cluster growsindefinitely
is zéro ;Gk
is anupperbound
to theproba- bility
that gk is nonnull,
thisprobability
is itself anupperbound
toPoo
and we can writefor any value of k .
Now,
ifWa 1, lim
k- 00oGk
= 0 so thatAll these
results, i. e. :
1) Wa
1 for any value of W,2)
ifWa 1, P 00
=0,
or WW,,
suggest that the critical value
Wc
is the smallestvalue of W for which
Wa
= 1.Following
thispicture,
we haverepresented
infigure
3 the results of the Monte Carlo method under the form of the mean number ofpoints
in each396
FIG. 3. - GK : mean population of each generation according to experimental results.
generation Gk
up to the 20thgeneration.
The halflength
of error intervals is twice the standard devia- tion. Theprevious
estimates ofWc
arestrongly
confirmed
by
thisrepresentation.
Inparticular,
thecritical value
Wa
= 1 appears to the reached for aWc
valueslightly higher
than 2.6.This value is rather different from the value
given by
H. I. R.(Wc ~ 2.4).
Aspreviously
notedby
Domb
[10]
the linearextrapolation
of the function1 /ê
usedby
H. I. R. mustgive
an underestimate ofWc,
because it
neglects
the curvature which is still apparentnear the critical value. A better
extrapolation
wouldbe attained if one could calculate ê for values of t closer
to tc; unfortunately
the value of ê is thenessentially
due tolarge
clusters and thecomputation
is limited
by
the computercapabilities.
The situationis
quite
different in our method where the computa- tion of the fertilitiesWk
isalways possible
even inthe nearest
vicinity
of tc.4. Validity
of the model ofbranching
process with fertilitiesdepending
on thegénération
number. -We want to examine whether the
growing
of clustersis well discribed
by
abranching
process of this type i. e. ; a process where the number of sons of a manbelonging
to the kthgeneration
is in accordance toa Poisson law of parameter
Wk.
For
that,
we compute the theoreticalP;
valuesgiven by
this process where the valuesWk
are deducedfrom the
experimental Gk (Fig. 3) :
Then we compare with
PN experimental
values ofthe
figure
2.We failed to obtain the theoretical
PN+
in thegeneral
case. However the calculus is easier if onesupposes that
Wk
is constant above agiven generation
number 1
If 1 = 1
(two
fertilitiesprocess)
an exactexpression
has been obtained
(see § 6).
If 1 > 1 recurrent formulae can be derived in this
manner :
Let
Ak(s)
be thegenerating
function of theproba- bility PkN
that thefamily
descended from a man of the kthgeneration (fertility Wk)
contains N men :In
particular, k
= 0gives
thegenerating
functionof the
probabilities P N :
Consider now a man of the kth
génération ;
hehas
Z,
sons and thegenerating
function ofZ,
isThe total progeny of this man can be considered
as the sum of
Zk
ensembles. With the aid of the theoremsconcerning
thegenerating
functions ofsums of a random number of
variables,
we can writeThe recurrent formulae are obtained from these
expressions: (4a)
is writtenand
by taking
thederivative,
we getFinally Ak(s)
andA[(s)
arereplaced by
their power series and we deduce the recurrent relations betweenthe
PkN
coefficientsThese relations have been used to compute
PN
N-1
and
PN -
1- L PN
with thefollowing
valuesk=l
of
Vfi
evaluated from the results offigure
3 :W1
=W2o = 0.8, 0.9,
0.98respectively
for W =2, 2.3, 2.6,
W1 == Wg ==
1.07 for W=3.According
to apreceding remark,
this last value isprobably
toohigh ;
however thePN
values arechecked
only
for moderate values of N and we expect that the resultsdepend essentially
on the fertilities of the firstgenerations.
The standard error on
PN
calculatedby
thismethod,
isgiven by
where (Jk
is the standard deviation onWk.oP; /oWk
wasnumerically computed.
To obtain an estimate of 6k,we assume that the clusters grow
according
to abranching
process with Poisson law. Thenwhere M is the total number of constructed clusters.
The values
PN given by
thebranching
processare
presented
infigure
2. Theprecision
is shownby
the error interval centered on
P64+ (its
halflength
istwice the standard error
a’).
The agreement with the
expérimental
values isvery
good
up to W = 2.3. For the upper values of W, the deviation islarger
than the deviation allowedby
theinacurracy
of the results.This deviation may be understood if we note that in the
growing
of a cluster thefertility
of apoint depends strongly
on the number ofpoints
of itsgénération :
because of thepossible overlap
of thespheres
centered dn thesepoints,
the volume to beexplored
perpoint (which
is similar to a vitalspace)
is
certainly decreasing
with their number.Now,
the model considers
only
a meanfertility Wk.
Theresults show that this is not allowed for
large
concen-trations. In order to
clarify
thisremark,
we considerin the
following
part abranching
process where thefertility
of a mandepends
on the number of men in itsgeneration.
We call this process a vital space process.5. Vital space process. - We
proceed
as in thepreceding
part, i. e. we compute the theoreticalPN
values
according
to the vital space process with fertilities deduced from the Monte Carlo results.For
that,
we calculate for eachgeneration the experi-
mental
fertility
of apoint
versus the number n ofpoints
in thisgeneration.
Infact,
theprecision
of theresults does not allow the effect of the
generation
number to be
perceived
and we take an averageon the first
generations.
The values aregiven
infigure
4.FIG. 4. - W(n) : mean value on 19 generations for W = 2.3,
18 générations for W = 2.6,8 generations for W = 3. ; --- : values used in the calculation of PN and PN with the vital space process.
We failed to obtain an
analytic expression
ofPN
or
PN
and we computePN directly
from its definition i. e.,PN
is the sum of theprobabilities
ofhaving
afamily
of N men characterisedby
agiven
set of thenumber of men in each successive
generation.
Thenumber of such sets is 2N - 2 and it is clear that
PN
can
only
be calculated for rather small values of N.The
P+N
values are shown infigure
5 up to N = 20.In this range of values of
N, they
agree very wellFIG. 5. - PN = f(N). Dots : experimental results. - process with fertilities depending on the generation number. --- vital space
process.
398
with the
experimental results,
even forlarge
valuesof W. Thus it is
clearly
shown that for thelarge
valuesof W, the concept of vital space of the
points plays
afundamental role in the statistical laws
governing
the cluster size distribution.
6.
Approximate analytical expression
ofPN
forsmall N. -
Figure
3 shows that in the range k = 1 to20, log Gk
is not very far from a linear func- tion of k and soWk
=Gk 11 IGk
isslowly varying
in this range.
This remark suggests that we consider a
branching
process with two fertilities as an
approximate
modelgiving
the essential features of statistics on the clusters at least for values of N which are not toolarge.
In this process, the
fertility
of the ancestor is Wand all the descendants have the
fertility
W’. Themain interest of this model is in the
possibility
ofobtaining
an exactanalytic expression
forPN.
Wefollow a method similar to that used in
[11] :
The
expression (5)
is reduced towhere
r is a
loop
enclosed in a circle of unit radius centeredon the
origin.
Then with the aid
of (4b) :
where F’ is a small closed
loop encircling
theorigin
and K =
W’/W.
The theorem of residue
gives finally :
A
priori,
the ratio K is a function of W. Infact,
agood approximation
is obtained for the constant value K = 0.5.The results are
presented
infigure
6 under theform WP and
WPN
versus W for W 3 and N = 1to 4.
They
maybe compared
to the resultsgiven by
the vital space process
represented by
dots(we
haveshown
previously
that this last process describes well the cluster construction for small values ofN).
For N =
2,
we have verified the agreement betwèen theapproximate expression (8)
and the exact oneFIG. 6. - Estimate of PN and PN given by the expression (8) of§6.
which is
The deviation is less than 3
%
for W inside the interval[0, 3.5].
Conclusion. - We have
presented
a statisticalstudy
on cluster size distribution based on statistical results
given by
a Monte Carlo method. These results arein
good
agreement with those ofHolcomb,
Iwasawa and Roberts[5]. However, our extrapolation
methodis
quite
différent andgives
alarger
critical value ofW, slighly higher
than 2.6. This value is in accordance with the value 2.7given by
Domb and Dalton[6].
We note that in a
previous
work[12],
the authorsattempted
to obtain an estimate ofWc
which wasexpected
to be near the value for which the meantotal curvature of the cluster frontier becomes nul.
They
obtained for this last value 3.018. All thepreced- ing
results show that this value iscertainly
toohigh.
In
addition,
we have examined theanalogy
betweenthe
statistical
lawsgoverning
the construction of clustersby
the Monte Carlo method and thedevelop-
ment of a
family according
to abranching
process : for values of W less than the criticalvaluesr
theexperi-
mental results can be fitted
by
a process where thefertility
of a mandepends only
on itsgeneration
number. For greater values of W, the effect of
overlap
of the
spheres
is described in a more correct mannerby
a process where thefertility
of a mandepends
onthe number of men in its
generation.
Finally,
we derive anexpression
forPN
from a twofertilities process and show that it can be used to count the number of atoms
belonging
to clusters of small size for alarge
range of concentrations.Acknowledgments.
- We aregrateful
to ProfessorM. Cadilhac and Professor J. Hervé for useful discus- sions and criticism.
References [1] GAYDA, J. P., BLANCHARD, C., J. Physique 30 (1969) 827.
[2] GAYDA, J. P., J. Physique 32 (1971) 793.
[3] GAYDA, J. P., DEVILLE, A., LENDWAY, E., J. Physique 33 (1972) 935.
[4] SHANTE, V. K. S., KIRKPATRICK, S., Adv. Phys. 20 (1971) 325.
[5] HOLCOMB, D. F., IWASAWA, M., ROBERTS, F. D. K., Biometrika 59 (1972) 207.
[6] DOMB, C., DALTON, N. N., Proc. Phys. Soc. 89 (1966) 859.
[7] SYKES, M. F., ESSAM, J. W., Phys. Rev. 133A (1964) 310.
[8] HOLCOMB, D. F., REHR, J. J., Phys. Rev. 183 (1969) 773.
[9]
HARRIS, T. E., The theory of branching process (Springer- Verlag, Berlin), 1963.[10] DOMB, C., Biometrika 59 (1972) 209.
[11] FISHER, M. E., ESSAM, J. W., J. Math. Phys. 2 (1961) 609.
[12] OTTAVI, H., GAYDA, J. P., J. Physique 34 (1973) 341.