HAL Id: jpa-00246849
https://hal.archives-ouvertes.fr/jpa-00246849
Submitted on 1 Jan 1993
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.
dynamics study
Jysoo Lee
To cite this version:
Jysoo Lee. Avalanches in (1 + 1)-dimensional piles: a molecular dynamics study. Journal de Physique
I, EDP Sciences, 1993, 3 (10), pp.2017-2027. �10.1051/jp1:1993229�. �jpa-00246849�
Classification Physics Abstracts
05.40 46.10 62.20 64.60
Avalanches in (I + I)-dimensional piles:
amolecular dynamics
study
Jysoo
LeeHLRZ-KFA Jiilich, Postfach1913, W-52425 Jiilich, Germany
(Received
I September 1992, revised 13 June 1993(*),
accepted 16June1993)
Abstract. We numerically study the piles generated by continuously dropping a particle
on the top of a stable pile in
(I
+I)
dimension. We use a code which has an implementationof static friction in the Molecular Dynamics
(MD)
simulations of granular material. We studyseveral properties of the pile-the time evolution of mass and slope, and the avalanche size
(and
duration) distribution. We also study the effect of gravity on the properties of the pile. We find that the angle of repose @R decreases as gravity is increased. On the other hand, gravity seemsto have little effect on the angle of marginal stability @Ms. We also find that the angles (@R and
@Ms) are different between the piles of monodisperse and polydisperse particles. We suggest a possible explanation for these dependencies of the angles.
1 Introduction.
Systems
ofgranular particles (e.g. sand)
exhibit manyinteresting phenomena [1-4].
One of the distinctproperties
ofgranular
systems is thatthey
can behave like both a solid and a fluid. Onecan pour
(like
afluid)
sandgrains
on atable,
andthey
form a stablepile
with finiteslope (like
a
solid). Using
a cellular automation model ofsandpile, Bak, Tang
and Wiesenfeld [5](BTW) showed,
in(2 +1)
and(3 +1)
dimensions, that when onecontinuously drops grains
on a stablepile,
the system evolves into a state with no intrinsiclength
and time scales(critical state).
Experimental
studies on realsandpiles, however,
are notfully
consistent with the results of BTW.Experiments by Jaeger
et al. [6] andEvesque iii
do not show anysign
ofcriticality.
On the other
hand,
experimentsby
Held et al. [8] showed that thepiles
are indeed critical for small sizes, but cease to be critical forlarger
sizes.The behavior of
(I
+I)-d sandpiles
can be very different from their(2
+1)-d
counterparts.For
example,
the(I
+I)-d
version of the BTW model does notdisplay
any critical behavior.On the other
hand,
there exist(I
+I)-d
models which are critical [9]. Whether a(I
+I)-d
realsandpile
shows critical behavior or not iscertainly
an openquestion.
In this paper, westudy
(*) The editor regrets the delay due to lost mail.
the behavior of realistic
(I
+I)-d piles using
MolecularDynamics (MD).
The very reasonwhy
sand can form a
pile
is static friction. In otherwords,
one needs afinite
threshold to breaka contact between
grains (and
between agrain
and awall). Therefore,
one has to include static friction in the MD simulations ofgranular particles
to obtain a stablepile. Here,
theimplementation
of static friction is doneby using
the scheme of Cundall and Strack[10].
Westudy essentially
a(I
+I)-d
version of theexperiments by Jaeger
et al. and Held et al.. In otherwords,
wedrop
aparticle
on apile,
and wait until thepile
becomesstable,
then add anotherparticle,
and so on. We find that the time evolution of the mass andslope
of thepiles
showcomplex
structures. We alsostudy
the avalanche size(and duration)
distribution to determinewhether the system is critical. The avalanche size distribution is consistent with a
power-law
form
(critical). However,
we cannot make a definite statement, since we have studiedonly
very small systems.There are two
important angles
for apile.
One is theangle just
after an avalanchw-theangle
of repose9R,
and the other is the maximumangle
of a stablepilw-the angle
ofmarginal stability 9Ms.
These twoangles
differby
a fewdegrees ill].
In order to understand the mechanismsresponsible
for theseangles,
Evesque et al. studied thedependencies
of theseangles
on thedensity
ofpacking
andgravity
[12].They
found that theangles
arestrongly dependent
on thedensity,
but lesssensitively
ongravity. However, they
did notstudy
the effect ofgravity
for a fixeddensity. Rather, they
studied the time evolution of the densitiesduring
severalavalanches,
whilegravity
isbeing changed. Here,
we fix thedensity,
andstudy
thedependence
of bothangles
ongravity.
We measure 9R as follows. We putparticles
atrandomly
chosenpositions
in a box. We wait until theparticles dissipate
their energy, and fill the box ivithout anysignificant
motion. We then remove theright wall,
and wait until thepile
becomes stable(motionless).
We define theangle
ofresulting pile
to be9R. Now,
weslowly
rotate clockwise the box
(with
thepile inside).
Thepile
is stable until we rotate more than thetilting angle
ST Theangle
ofmarginal stability
9Ms is estimated from 9R andST.
We find that9R systematically
decreases asgravity
is increased. On the otherhand,
9Ms seems to show little or nodependence
ongravity.
We also find that bothangles
are different betweenpiles
ofmonodisperse
andpolydisperse particles.
Thisdifference,
as well as the nearindependency
of9Ms,
indicate that the geometry of thepacking plays
a dominant role.2. Interaction of the
particles.
Each
particle
isrepresented by spheres
which interactonly
ifthey
are in contact with each other. Consider twoparticles
I andj
in contact, the force between them is thefollowing.
Let the coordinate of the center ofparticle
Ij)
be B~(Rj),
and r= B~
Rj.
In twodimension,
weuse a new coordinate system defined
by
two vectors n(normal)
and s(shear). Here,
n =r/ (r(,
and s are obtained
by rotating
clockwise nby 7r/2.
The normal componentFp_,
of the forceacting
onparticle
Iby j
islift,
=kn(a,
+ aj(r()~/~ ~fnme(v n), (la)
where a;
(aj)
is the radius ofparticle
I(j),
and v=
dr/dt.
The first term is the Hertzian elasticforce,
where km is the elastic constant of the mater1al. The constant ~fn of the secondterm is the friction coefficient of a
velocity dependent damping
term, m~ is the effective mass, m,m~/(m;
+mj).
The shear componentF]_;
isgiven by
Fj_,
=-~f~me(v s) sign(ds) min(ka(ds(, p(F),(), (16)
where the first term is a
velocity dependent damping
term similar to that ofequation (la).
The second term is to simulate staticfriction,
whichrequires
afinite
amount of force(pF/_;)
to break a contact[10]. Here,
p is the frictioncoefficient,
ds the total sheardisplacement during
a contact, and k~ the elastic constant of a virtual
spring.
There are several studies ongranular
systemsusing
the above interactions[10, 13-15]. However,
all ofthem,
except references[10,
14] and[15],
do not include static friction. Aparticle
can also interact with a wall. The forceon
particle
I, in contact with awall,
isgiven by equations (I)
with aj = oo and m~= m;. A
wall is assumed to be
rigid, I-e-,
theposition
is fixed.Also,
the system is under agravitational
field g. We do not include the rotation of the
particles
in the present simulation. A detailedexplanation
of the interaction isgiven
in reference [15].3.
"Sandpile"
with infinitesimal flux.We
study
the behavior of apile
with infinitesimal continuous addition ofparticles.
The simu- lational setup isessentially
a(I
+I)
dimensional version of thesandpile
experiments doneby
a)
Fig. I. a) The "sandpiles" are built on a L shaped wall of width W and height H with a gravitational
field g pointing downward. Here, W
= 4, H
= 4 and g
= 980. We show in
b)
the number of particles,and in c) the angle of the piles like the
one shown in a). Both quantities are measured whenever the
piles become stable.
140.00
120.oo
@100.00
flw
80.00
%-
° g
~
n oo
o.00 ~
0.00
number of inserted b)
30.00
25 oo 1J
#
C>w 20.oo 1J
/$
JZ 15.00o
i~
1J ~~'°°7$
5.oo
noo
0.00 (1.40 0.80 1.20 1.60 2.00x10
number of inserted balls c)
Fig. I.
(continued)
Jaeger
et al. [6] and Held et al. [8]. Consider an Lshaped
wall of width W andheight H,
anda
gravitational
field of g= 980 is
acting
downward(Fig. la).
We use CGS units in this paper,although
we do notexplicitly give
units for each quantities. In thebeginning
of asimulation,
we insert a few
(5
r~
10) particles
on the bottomwall,
and calculate thetrajectories
of theparticles,
until the maximum of both(x
andy) velocity
components of theparticles
is less thata cutoff value u~ut. If the maximum is smaller than u~ut, we insert a new
particle
on the left side and twoparticle
radius above thepile.
Weagain
calculate thetrajectories
of theparticles,
until the maximum becomes smaller than u~ut, insert a new
particle,
and iterate thisprocedure again. Also,
if aparticle
falls out of the box, we remove theparticle
from the system.We want to simulate the situation that the
pile
relaxes to a staticconfiguration
after eachaddition of one
particle.
If apile
isstatic,
thevelocity
components of all theparticles
should be zero, which isonly
true if u~ut" 0. We
study
the system with u~ut"
0.1,
0.01 and0.001,
andfind no
systematic
differences. We use u~ut" 0.I from now on.
Also, changing
theheight
of anewly
insertedparticle
to 4 and 12particle
radii above thepile produces
no systematicchange.
The parameters for the interaction between the
particles
are km= 1.0 x
10~,
k~ = 1.0 x10~,
~fn = 5.0 x 10~,~fs = 5 and p = 0.2. For the interaction between the
particle
and thewall,
we use km = 2.0 x
10~,
and the other parameters are the same as above. We use a fifth orderpredictor-corrector
method with time step 5 x 10~~ In order to avoid thehexagonal packing
formed
by particles
of the samesize,
we choose the radius from a Gaussian distribution with average 0.I and width 0.02. Thedensity
of theparticle
is chosen to be 0.I.Having
defined the setup for thesimulations,
we nowproceed
tostudy
someproperties
of thepile.
Whenever we insert a newparticle,
we calculate the total number ofparticles
m"mass")
and theangle
9 of thepile.
The mass will later be used to extract some information about the avalanches of theparticles.
Theangle
of thepile
is calculated as follows. We divide the box into vertical strips of width0.2,
which is the average diameter of theparticles.
Instrip
I, we define the maximum value of the centers ofparticles
to beh(I). Then,
thepairs (0.2 I, h(I))
is defined as the "surface" of thepile.
We calculate 9 from the least square fit ofa
straight
line to the surface. All theangles
are measured indegrees.
In
figures
16 and lc, we show m and 9 measured whenever a newparticle
is inserted. After a certain transientperiod
ofincrease,
bothquantities
seem to fluctuate in acomplicated
manner.One of the
important
quantities forcharacterizing
the behavior of apile
is theangle
fluctuation d9 and mass fluctuation dm of thepile.
We define d9=
(92) (9)2
and dm=
(m2 (m)2,
where the averages are taken over time. If d9 remains finite in the infinite size
(W
-oo) limit,
theangle
9 does not remain a constant. Onepossible
scenario for thisnon-steady
state is that the pile oscillates between theangle
of repose 9R and theangle
ofmarginal
stability 9Ms, which differby
a fewdegrees (See,
e-g-, Ref. [6]). The mass fluctuation dm can be related to d9by
dm=
)
d9 W2. Infigures
2a and2b,
we show(m)
and dm as a function of W. We average over 10000samples
for W= 2,3,4, 5, and 1500
samples
for W= 6 and 500
samples
for W= 7. The average mass in
figure
2a increases asW2,
asexpected.
On the other hand, it is notpossible
to find asystematic
trend for dm infigure
2b due to too much noise. Sincethe CPU time needed to get one
sample (typically,
10~r~ 10~
iterations)
of W = 7 is about 4 hours on one CRAY-YMP processor, it is very hard to get agood
estimate ofdm, especially
forlarge
systems.We now
study
more details of the avalanche, the distribution of their sizes and durations.Consider two successive steps of
inserting
aparticle,
I and I + I. Let the mass at step I be m,.If m,+i = m, + I, the particle inserted at step I just stays in the
pile. Therefore,
noparticle
flows out of the system.Otherwise,
an avalanche of size s= m; + I m,+i has occurred. We define the avalanche size distribution
D(s)
as theprobability
that an avalanche of size s occurs(including
s=
0),
when we add oneparticle
to apile.
The distributionD(s)
shouldsatisfy
thefollowing
conditionsf Djs)
= 1,
j2a)
s=o
f
s
Djs)
= 1.(2b)
smo
Equation (2a)
is the normalization ofD(s).
Since the system is in asteady
state, the average number ofparticles dropping
outby adding
aparticle
should be unity, which isequation (2b)
300.00
250.00
~
uo 200.00E
~~j150.00
~
i
loo.oo ol50.oo
o-w
2.oo 3.oo 4.oo 5.0n 6.00 7.00
width a)
i i.oo
lo-w
9.oo
~j
j~
~~#
7.00l~$
uo 6.00
~i
£
5.004.oo
3.oo
2.00 3.oo 4.00 5,oo 6.00 7.00
width
b)
Fig. 2. The average "mass"
a)
and its fluctuationb)
of the piles for several system sizes.[16].
Infigure
3, the distributionD(s),
obtained for several values ofW,
is shown in a semi-log plot. Also,
we show thelog-log plot
of the same curves in the inset. Thedistribution, especially
forlarger W,
seems to bendupwards
in thesemi-log plot, suggesting
that itdecays
slower than
exponential.
On the otherhand,
it is also notpossible
tounambiguously identify
a
power-law regime
in the inset. The curves shown infigure
3 are consistent withpower-law decay,
butthey
aresimply
insufficient tosingle
out the behavior. One shouldstudy
sufficientnumber
(r~ 10000)
ofsamples
forlarger
systems, which is notpossible
asexplained
before. We alsostudy
the avalanche durationdistribution,
which is toonoisy
to find out anysystematic
behavior.
10o
10~
i' ~ ii milj .1~ i'r~
~ ~
j~ ' 10'~ fi W
" 3
+
10 ~
i
w= 4
j
lo ~ Q
~ j w
# 5
~"~~
~~
~° ~ '
/~
.5Cl ..:.-.l',im
~°
i°°
1°~jo h_ I '
"",' ,' ', fl"?h" ',j '
~.~ ', '-'i I'
[ ",
'O'~'(
'',i .-.--.---z~' ',
4 / "_ ' ,
10 1,I h " ''
~,' h,
~ s
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00
s
Fig. 3. The avalanche size distributions
D(s),
measured for several system sizes, are plotted in asemi-log plot. The log-log plot of the same curves are shown in the inset.
4. Effect of
gravity.
We now
study
the effect ofgravity
on behaviors of apile
with infinitesimal flux.Especially,
we are interested in the
gravity dependence
of twoquantities,
the average mass of thepile
and its fluctuation. We start from the same setup describedabove,
and wechange gravity. Here,
we define lG
= 980, the value of g used in the above simulations. In
figure
4, we show the average mass measured for several different values of g with W= 4 and p
= 0.2. Simulations for
larger
values ofgravity
takelonger
time, since theparticles
havelarger potential
energy todissipate.
It is clear from thefigure
that the average mass of thepile
decreases forlarger
g.In order to understand the
dependence,
we first consider thefollowing possibility.
Since we fix thedropping height
of aparticle,
the momentum of theparticle
when it collides with thepile
is greater forlarger gravity.
Thelarger
momentum could make thepile
lessstable,
and results in smaller average mass. We use differentdropping heights
for different values of g tomake the momentum of the
particle
about the same. We indeed find that9R systematically
decreases as the
dropping height
is increased.However,
the effect due to the momentum ofdropping
oneparticle
is too small(typically,
less than onedegree)
to account for the observedgravity dependence.
We now consider another
possibility.
Thedynamics
of thepile
can be divided into two stages. One is to build apile
until it becomes unstable(angle
ofmarginal stability, 9Ms),
and the other is to make thepile
stableagain by
an avalanche(angle
of repose,9R).
In order tostudy
thegravity dependence
of eachangle,
we measured 9R and ST as described in section I.We
study
thegravity dependence
of bothangles,
and combine the results to find the behavior of9Ms.
The interaction parameters we use in this simulation arekept
to be the same as above.In
figure
5a, we show 9R for different values of g. Theangle
9R is measuredby
the method15.oo
1J
# lo.oo
C>w
los.oo i~~
° loo oo
( g
95no
1J bo 9fl on i~
~j i~ fl(1
31
80 on
n.0n 0.50 n0 50 2 nn 2~l) ~ 0n 3.5n
gravity [G]
Fig. 4. The average mass of the piles for different values of gravity. Here, W
= 4 and p
= 0.2.
described before. Each
point
is obtainedby averaging
over 10samples.
One can see that 9Rclearly
decreases forlarger gravity.
We interpret the result as follows. When an avalanche isstarted,
theparticles,
which arefalling down, probably
have alarger
momentum forlarger
gravity.
Since thepile
is less stable in a collision withparticles
withbigger
momentumj 9R becomes smaller forlarger
g [17]. This argument is also consistent with the abovestudy
of themomentum of the
dropping
ball.Next,
we consider thetilting angle,
ST We show ST vs. g infigure 5b,
where the averagesare taken over 10
samples.
Theangle
seems to getlarger
forbigger
values of g,although
the curve is too
noisy.
We now estimate 9Ms from the twoangles,
9R and ST- We proposethe
following approximation.
Consider apile
ofangle 9R,
rotatedby
ST(Fig. 6). Along
thebottom
@,
we define the component of the force normal
(tangential)
to the surface to be Fn(F~).
Since thepile
ismarginally
stable at thetilting angle,
F~ =pfn,
where p is the frictioncoefficient between a
particle
and a wall. We now fill thetriangle
OAB with the same materialas the
pile,
and remove the bottom@.
The forces(F[, F() along
the new bottom OB can beobtained,
in the firstapproximation, by rotating
the(Fn, F~) by
9T. That isF[
= cos ST Fn + sin ST F~F(
= sin ST Fn + cos ST F~.(3)
The
stability
of the newpile
can be checkedby calculating
the ratioF] IFS. Using
equation(3),
the ratio becomes
~
,~
/~ ~~~~T
F[
I + p tan ST j~~The ratio becomes less than p for small
positive
ST, therefore a pile withangle
9R + 9T is stable. When theangle
is increasedadditionally by 9',
thepile
becomesmarginally
stable. If ST « I, it isplausible
that 9' = A9T up toO(9[),
where A is aproportionality
constant.Furthermore, we can show [18], from an
approximate
stress distribution inside apile,
that 9' is indeedproportional
to9i.
Therefore, 9Ms'- 9R +
(1+ A)9T.
The constant A is,however,
19.oo
ii.oo
0J
I
17.00
fl
flw%- 1600
O
~15.00
0J~
14 00
13.00
0.00 2.0(1 400 600 8.00 In,n(1
gravity [G]
a)
7 oo
6 50
6.00
~
0J 5.50~
~ 500£
~ 450~ 400
3.50
3,no
0 00 1.25 2 50 375 5.00 6.25 7.50 8.75 10 00
gravity [G]
b)
Fig. 5. a) The angle of repose @R and b) the tilting angle @T for different values of gravity, obtained
by removing a wall and rotating the box (see text).
too much
dependent
on the details of theapproximation.
If Ar~ 2, there is no
systematic dependence
of9Ms
on g. Weconclude,
from thetilting simulations,
that 9Rclearly
decreasesas g is increased. On the other
hand,
9Ms seems to show very weak or nodependence
on g.How can we understand the near constancy of 9Ms over
gravity?
Consider contacts between theparticles land
between aparticle
and awall).
Thestability
of a contact is determiRedby
the ratio
f~ / fn,
wherefn( f~)
is the normal(shear)
force between the contact. If the networkof contacts remains
unchanged,
both forcesfn's
andf~'s depend linearly
on g. Since all theratios
fn/f~
are constant for different g, the ratioFn/F~
is alsoindependent
of g, so is the4
F~ °R
F~
D
F( F]
H,r oFig. 6. A pile of angle @R is tilted by @T. The forces along M is related to the force along $ by
rotation.
stability
of thepile.
This couldexplain
the near constancy of HMS overgravity. However,
one should note that ifgravity
islarge enough
tochange
the contacts of thenetwork,
then the above argument does not hold.Finally,
we repeat thetilting experiment
with theparticles
of the same radius(0.I).
Theangle
HR (HT) is measured to be 24.97 +1.14(6.90
+0.83), compared
to 18.0 + 0.6(4.8 +1-1)
obtained before. These differences can be
explained
from the "networkpicture"
as follows. Ina
pile,
we draw a line betweenparticles
which are in contact with each other. The stress dis- tribution inside thepile depends
on the structure of the network. Since we expect the networkto be different between the
monodisperse (triangular lattice)
andpolydisperse (disordered
lat-tice) particles,
it isquite possible
that the stress distribution(and
HR, HMS) alsodepends
on the"dispersity".
Furthermore, the network picture can alsoexplain
thedependence
of HMS on thedensity
ofpacking,
which is foundexperimentally by Evesque
et al. [12]. If one fixes thedensity
of thepacking,
one alsospecifies
a certain structure of the network. In otherwords,
networks of different densities should have different structures, which can result to different HMS. Thesearguments, of course, should be checked
by
direct calculations of the stress distribution.5 Conclusion.
We
study
properties of(I
+I)-d sandpiles by
moleculardynamics
simulations. The avalanche size distribution ofpiles,
builtby continuously adding particles,
is consistent with a power lawbehavior.
However,
since the size of the systems westudy
is toosmall,
we cannot make a definite statement about theasymptotic
behavior. The CPU time needed forobtaining
the distribution issignificant,
even for a small system size(150 particles).
The reason is thatwe have to wait a
long
time(typically,
10~ r~ 10~iterations)
for thepile
torelax,
for each addition of aparticle
on thepile.
We alsostudy
thedependences
of HR and HMS ongravity.
we find that HR is decreased, as
gravity
is increased. On the other hand, HMS shows no or littledependence
ongravity.
We try toexplain qualitatively
the dependenceusing
the "network"~f f~rces It would be interesting if one can
quantitatively
relate the stress distribution(and
o~s)
toproperties
of the network.Acknowledgements.
thank H. J. Herrmann for many useful discussions.
References
ill
Savage S. B., Adv. Appl. Mech. 24(1984)
289; Disorder and Granular Media, D. Bideau Ed.(North-Holland,
Amsterdam,1992).
[2] Campbell C. S., Ann. Rev. Fluid Mech. 22
(1990)
57.[3] Jaeger H. M. and Nagel S. R., Science 255
(1992)
1523.[4] Mehta A., Physica A 186
(1992)
121.[5] Bak P., Tang C. and Wiesenfeld K., Phys. Rev. Lett. 59
(1987)
381.[6] Jaeger H. M., Liu C.-h. and Nagel S., Phys. Rev. Lett. 62
(1988)
40.[7] Evesque P., Phys. Rev. A 43
(1991)
2720.[8] Held G. A., Solina, II D. H., Keane D. T., Haag W. J., Horn P. M. and Grinstein G., Phys. Rev.
Lett. 65
(1990)
l120.[9] Kadanoff L. P., Nagel S. R., Wu L. and Zhou S., Phys. Rev. A 39
(1989)
6524.[10] Cundall P. A. and Strack O. D. L., G£otechnique 29
(1979)
47.[iii
See, e-g-, Briscoe B. J., Pope L. and Adams M. J., Powder Technol. 37(1984)
169.[12] Evesque P., Fargeix D., Habib P., Luong M. P. and Porion P., J. Phys. I France 2
(1992)
1271.[13] For example, Haff P. K. and Werner B. T., Powder Technol. 48
(1986)
239;Thompson P. A. and Grest G. S., Phys. Rev. Lett. 67
(1991)
1751;Ristow G., J. Pliys. I £Yonce 2 (1992) 649;
Taguchi Y-h., Phys. Rev. Lett. 69
(1992)1371;
Gallas J. A. C., Herrmann H. J. and Sokolowski S., Pliys. Rev. Lett. 69
(1992)
1375.[14] Bashir Y. M. and Goddard J. D., J. Rheol. 35
(1991)
849.[15] Lee J. and Herrmann H. J., J. Phys. A 26
(1993)
373.[16] Tang C. and Bak P., Phys. Rev. Lett. 60
(1988)
2347.[17] A similar argument, applied in a different context, is given by Prado C. P. and Olami Z., Pliys.
Rev. A 45
(1992)
665.[18] This calculation is done by using the approximate stress distribution obtained in reference [15].