HAL Id: jpa-00246926
https://hal.archives-ouvertes.fr/jpa-00246926
Submitted on 1 Jan 1994
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.
Density waves in granular flow: a kinetic wave approach
Jysoo Lee, Michael Leibig
To cite this version:
Jysoo Lee, Michael Leibig. Density waves in granular flow: a kinetic wave approach. Journal de
Physique I, EDP Sciences, 1994, 4 (4), pp.507-514. �10.1051/jp1:1994156�. �jpa-00246926�
Classification Physics Abstracts
05.40 46.10 62.20
Density
wavesin granular flow:
akinetic
waveapproach
Jysoo Lee(*)
and MichaelLeibig(**)
HLRZ-KFA Jiilich, Postfach 1913, W-5170 Jiilich, Germany
(Received
15 November 1993, received in final form 16 December 1993, accepted 3 January 1994)
Abstract. Ii was recently observed that Sand flowmg down a vertical tube sometimes forms
a traveling density pattern in which
a number of regions with high density are separated from each other by regions of low density. In this work, we consider this behavior from the point of view of kinetic
wave theory. Similar density patterns are found in molecular dynamic simulations of the system, and a well defined relationship is observed between local flux and local density
a strong mdicator of the presence of kinetic waves. The equations of motion for this system are
also presented, and they allow kinetic wave solutions. Finally, the pattem formation process is investigated using a simple model of interacting kinetic waves.
Systems
ofgranular partiales (e.g. sand)
exhibit manyinteresting phenomena,
such as segre-gation
under vibration orshear, density
waves in trie outfiowthrough
tubes andhoppers,
andprobably
moststrikingly,
trie formation ofheap
and convection cell under vibrationil,
2, 3, 4].In
granular
flowsthrough
a narrow verticaltube,
Pôschel found [Si that triepartiales
do trot flowuniformly,
but formhigh density regions
which travel as coherent structures with a ve-locity
dilferent from trie center of massvelocity.
He alsoreproduced
thesedensity
wavesusing
molecular
dynamics (MD)
simulations [Si.However,
trie motion of thesehigh density regions
and trie mechamsm which is
responsible
for their formation are nonfully
understood.In this paper, we present numencal and theoretical evidence that these
density
waves are of a kinetic nature [6].Using
MDsimulations,
we measure triedependence
of triepartiale
flux on triedensity.
We find a well-definedflux-density
relation an indication that a kinetic wavetheory
descnbes trie behavior. A direct measurement of trievelocity
of thesehigh density regions
showsa
dependence
on trie meandensity
which is ingood
agreement with triepredictions
from kineticwave
theory.
On trie theoreticalside,
we consider one dimensionalequations
of motion for triedensity
and trievelocity
fields in trie tube. Theseequations, together
withBagnold's
law forfriction [7], allow kinetic
density
wave solutions.(*) Present address: Levich Institute, Stemman Hall, 140~~ St and convent av., New York, NY 10040, U-S-A-
(**)Present
address: Institute of Theoretical Physics, University of Califomia, Santa Barbara, CA 93106, U-S-A-508 JOURNAL DE PHYSIQUE I N°4
In order to understand the formation of these
high density regions,
we consider triegeneral problem
ofinteracting
kinetic waves. We first shownumerically
that a system with aninitially
randomdensity
field evolves to aconfiguration
in whichneighboring regions
bave ahigh density
contrast. At trie
early
stage ofdevelopment,
we can showanalytically
that triedensity
contrast betweennearby regions
increaseslinearly
with time.We first discuss trie MD simulations of trie system, and
begin
with a briefdescription
of trieinterparticle
force laws that were used in our calculations. Triepartiales
mteract with each other(or
with awall) only
ifthey
are in contact. Trie force that acts on partide1 due topartide j
cari be divided into two components. Triefirst, F)~~,
isparallel
to trie vector r e R~R~,
where
R~
andR~
are trie coordinates of trie centers ofparticles
andj respectively.
We refer to this as trie normal component. Trie second component,orthogonal
to r, is trie shear comportent(~~~.
Trie normal component isgiven by
~Î-
"~n(~i
+ ~J lTll~~~ 7nme)) ~, (la)
where a~(a~ is trie
radius,
and m~ (m~ trie mass ofpartiale1 (j). Also,
m~ is trie effective mass m~m~/(m~
+m~),
and v +dr/dt.
Trie first term inequation (la)
is trie Hertzian elasticforce,
where kn is a material
dependent
elastic constant. Trie second term is avelocity dependent
friction term, where ~in is a normal
damping
coefficient. Trie shear component isgiven
aslÇ~i
"~'lsLle)jj~, (16)
where s is defined
by rotating
r clockwiseby ~/2.
Trie shearforce, equation (16),
issimply
avelocity dependent
friction term similar to trie second term in trie normal component.Finally,
we must
specify
trie interaction between apartiale
and a wall. Trie force onpartiale1,
in contact with awall,
isgiven by equations (1)
with a~= cc and m~ = cc. Trie choice of trie interactions defined
by equations (1)
is rathertypica1in
trie MD simulations ofgranular
material [8]. A detailedexplanation
of trie interaction isgiven
elsewhere [9].For
simplicity,
westudy granular
fiows in 2 dimensions and use a fifth orderpredictor-
corrector scheme to mtegrate trie
equations
of motion,calculating
both triepositions
andvelocity
of eachparticle
ai all limes. Trie tube is modeledby
two vertical sidewalls oflength
L with a
separation W,
and weapply
aperiodic boundary
condition m trie vertical direction.Between trie
sidewalls, partiales
of radii o-1 aremit1ally
filled with a uniformdensity
of p~(throughout
ibis paper, numerica1values aregiven
in CGSunit).
Trieparticles begin
to move under trie influence ofgravity,
and soon reach asteady
state, where triegravitational
force isbalanced
by
trie frictional force from trie interactions with trie sidewalls.in
figure
1, we show trie time evolution of triedensity
and thevelocity
fields for L= 15 and
W = 1, measured at every 5 ms. At a
given
time, we divide trie tube into 15 verticalregions
ofequal length,
and measure triedensity
and trie averagevelocity
in eachregion.
These fieldsare
displayed
as a verticalstrip
of squareboxes,
where each boxcorresponds
to aregion
in trie tube. Triegrayscale
of trie box isproportional
to trie value of trie field in thatregion.
Trieparameters we used in this simulation are kn
= I.o x 10~,~in = ~is = S-o x 10~, with trie time step S-o x 10~5. Trie initial
density
p~ is 25particles
per unit area. In triefigure,
we find(1)
aregion
oflarge density
fluctuations is formed out of trieinit1ally
uniform system,(2)
trie fluctuations seem to travel witl~ almost constantvelocity (dilferent
from tl~e center of massvelocity),
and(3)
tl~ere seems to be strong correlation between tl~edensity
and tl~evelocity
fields. These
findings
remam true for tl~e simulations we baveperformed
with dilferent values of ~i, kn and p~, except when p~ is verysmall,
where asteady
state is non reached. Thesetraveling density
pattems were first observed in trie simulationsby
Pôschel [Si.Time
Space
(a)
(b)
Fig.
1. Time evolution of(a)
density and(b)
velocity fields. These simulations were doue with a tube,of width W= and length L
= 15. Fields ai a given lime
are shown
as a vertical line of small boxes. The grayscale of each box is proportional to the value of the density or velocity m that region of trie tube. Regions of high density are formed, and travel with almost constant velocity.
In order to
quantitatively study
trie correlation between triedensity
and trievelocity fields,
we measure trie local
partiale
flux as a function of tl~e localdensity
in tl~efollowing
manner.Once tl~e system bas reached a
steady
state, we measure trie meanvelocity
u~ and triedensity
p~ in
region
1. Trie fluxj(p)
for agiven density
p is then taken to be p(u(p)),
where () is a time average over ailregions
which had aparticular density
p. Trieflux-density
curve,obtained
by averaging
over 10,000iterations,
are shown infigure
2.Here,
trie parameters are trie same as those offigure
1. Trie fact that a well-definedflux-density
curve exists suggest that triedensity
waves(traveling density fluctuations)
are kinetic m nature.Furtl~ermore,
tl~eflux-density
curve for tl~egranular
fiow resembles tl~at of a trailicfiow,
wl~icl~ is considered asa prime
example
of trie systems which shows kinetic waves [GI.One additional
piece
of evidence that triedensity
waves are of a kinetic nature is theirdependence
on trie initialdensity
p~. Trietheory
of kinetic wavespredicts
[6] that smalldensity
fluctuations in a uniformdensity background
p~ travel with avelocity
U(Po)
=
fl
lp=p~, 12)which is trie
slope
of trieflux-density
curve at tl~e meandensity.
We tl~us expect alarge negative velocity
for small p~, a decrease to zerovelocity
at pa m15,
witl~ anincreasingly large positive velocity
as p~ is increased furtl~er. To cl~ecktl~is,
we measure trie wave velocities for several values of p~(keeping
all otl~er parameters fixed asabove). Writing
tl~e meandensity
p~ and tl~e measuredvelocity U(p~)
as(p~, U(p~)),
we find(10.o,
-41+2), (15.0,
5+9), (18.7,12 +11)
and
(22.5,113.o
+4),
wl~icl~ are ail consistent witl~ tl~e aboveprediction.
We now consider trie theoretical aspect of trie
density
waves. Consider trieequations
of510 JOURNAL DE PHYSIQUE I N°4
1400
1000 _."
,,
~~~~
_.,
..
800
~~~
j
>,600 ,
~
400
~,,
~
l'~'j"~
j'
0 5 10
lj
2fl 25 30 35nUm ero palicles
Fig.
2. Local flux as a function of local particle density. This curve was found for a tube with width W= 1 and length L
= 15, obtained by time averagmg. The parabolic shaped curve resembles the flux-density relation in traflic flows.
motion wl~icl~ govern tl~e time evolution of tl~e
density p(z, t)
and tl~evelocity vii, t)
fields fora
granular
flow in a vertical tube. Tl~e firstequation
is tl~at of mass conservation(P
+)lPu)
= 0>
13a)
and tl~e second is a momentum conservation
equation
p~u
+pu)u
=F(x, t), (3b)
wl~ere
F(x,t)dx
is the total amount of forceacting
on theparticles
in aregion [x,x
+dx].
Tl~e force
F(x, t)
bas tl~ree contributionsgravity,
internai pressure, and friction from tl~e sidewalls. Tl~e exact form of tl~e internai pressure and tl~e friction is trot known.Here,
we useBagnold's law,
wl~icl~ is believed to be valid in tl~egrain
inertiaregime
[7].Therefore,
trie forceF(x, t)
isFlx, t)
=
-PgW SignlU)PBf~ylP)U~ Dl
lPBf~~lP)U~l
14)Here,
g isgravitational acceleration,
pB triedensity
of trie material which forms trieparticles,
ptrie
packing
fraction(p
=pBp)
and D is trie diameter of triepartiales.
We assume tl~e thickness of trie shearlayer
to be of order of D.Also, f~~
andf~y
aregeometry dependent functions,
which contain trie information about trie
density dependence
of trie forces.Trie uniform
density
solution ofequation (3
with trie forcegiven by
equation(4)
isP(x,t)
= pBp~
u(x,t)
=-~ (5)
If we add small
density
fluctuation p= p~ +
dp
in trie uniformdensity fiow,
trie fluctuationtravels with a
velocity
UiPo)
~-l~
~~~~~~°~j~iill~~~°~/~~,
16)which is
given by equation (2). Equations (5)
and(6)
areexactly
what one expects if trie kinetic wavetheory
is toapply
uniform fiow is a solution to trieequations
of motion, anddensity
fluctuations travel with adensity dependent velocity.
Thus,
it is clear then that trie motion of triedensity
pattern can be understoodby applying
trie ideas of kinetic wave
theory. However,
this basic formalismonly
describes trie motion ofa
pre-existing density
pattern. It does notexplain
trie observation thatregions
withlarge density
contrasts arebeing
formed out of trie uniformbackground.
Our simulations show that trielarge
scaledensity
patternbegins
as a collection of small fluctuations in triedensity.
These small fluctuations grow in time and a pattern emerges in whichlarge density
contrasts exist betweenneighboring regions.
Trie evolution to such a state can be understoodby considering
trie system as set of
interacting
kinetic waves. A detailed treatment of triegeneral problem
ofinteracting
kinetic waves can be found elsewhere [11] m this paper, we presentonly
trie results from asimple
model for trie pattern evolution process in sandfiowing
down a tube.Consider trie
early
stages of trie fiow in which triedensity
of sand isnearly
uniform at p m p~.Because of trie
roughness
of trie grains, trieroughness
of trie walls or from trie stochastic nature of trie inelasticcollisions,
smalldensity
fluctuations appear in trie system. In trieinteracting density
waveapproach,
we treat trie fluctuations as a set of distinctdensity regions
with interfaces whose velocities are determinedby
a discrete form ofequation (2).
In this case, trie interfaceseparating
aregion
ofdensity
pi from aregion
withdensity
p2 moves with avelocity, U(1, 2), given by
U(1, 2)
=
~~~~~
~~~~~,
(7)
Pl P2
which is trie kinetic wave
theory
result for interfacial velocitiesinvolving
finitedensity
dilfer-ences [GI. In trie case that trie dilference between pi and p2 is very
small,
trieequation
becomesequation (2).
Trie evolution of trie system is determinedby
trie motion of trie interfacesand,
as shall be shown, trie nature of their interactions leads to a final state m which
large density
contrasts occur.
In trie
computational
andanalytic
results thatfollow,
we choose aspecific
form for triedensity
fluctuations m trie system. In ourmodel,
trie initialpositions
of trie interfaces are taken as a set ofNo points placed randomly
on trie interval [o,L],
withregions
between successive interfacesbeing assigned
adensity randomly
in trie range [p~W,
p~ +W].
Trieprincipal
virtues of this model are itssimplicity
and trie fact that there are no correlations in trie initial state whichmight
influence trie final structure. A more realistic model for trie fluctuations of trie system would require amicroscopic understanding
of eachspecific
source of noise.It is also necessary to choose a form for trie flux curve
j(p).
We bave taken trieparabolic
formj(p)
= J~ ~(1-
~,
(8)
where R is trie
density
at which no flow occurs, and J~ is one quarter of trie maximum flux of trie system. This curve was chosen for several reason. Trie first is that itssimplicity
eases some of triehardships
ofanalytical
calculations. Trie second reason is that fordensity
fluctuationsover a
sufficiently
small range, trie true flux response can beapproximated by
this form(with
R and J~being fitting parameters).
Andfinally,
it is a firstapproximation
to trie form observed for triej(p)
observed infigure
2.512 JOURNAL DE PHYSIQUE I N°4
(ai
(b)
Fig.
3. Evolution of interacting kinetic waves. Both strips use a linear grayscale with whitemdicating po = 0 and black pa = R, the jamming density.
(a)
shows the initial configuration of 400 interfaces with densities in the range[0.3R, 0.8R]. (b)
shows the configuration when only 33 interfacesremain, and illustrates the tendency for
alternating
high and low density regions.Numerical simulation of this system is a very
straight-forward
exercise. Trie values of trie densities in twoadjacent regions
determine trievelocity
of trie associated interface. Consider three successivedensity
regionsA,
B and C.During
trie course of triesimulation,
trie interface A-B may encounter trie interface B-C. This indicates that ail of trie mass that was insideregion
B bas been
completely
"swallowed up"by
trieregions
A and C. In this case, trie interfaces A-B and B-C arereplaced by
asingle
A-C interface. Trievelocity
of this interface cari be calculated from trie densities inregions
A and C.Thus,
it is a matter oftracking
all of trie interfaces,checking
forcollisions,
and whenthey
occur,replacing
trie two old interface with asingle
newone.
Therefore,
thistechnique
does trot allow for anydensity
values other than thoseinitially
present, and trie number of interfaces isalways decreasing.
For convenience, trie simulationswere done
using periodic boundary
condition.Trie first set of results shown below are from a simulation m which there are 400 interfaces
initially placed randomly
in trie interval [o,ii
(1.e. L =1).
We also choose trie values J~= 1
and R
= 1. Trie densities are chosen at random from trie interval
[o.3,o.8]
(1.e. p~ = o.55R, W =o.25R). Figure
3a shows trie initialdensity configuration,
whilefigure
3b shows trie system alter a time t = o.486(where
time is measured in trie units ofRL/J~),
and there areonly
33 interfaces which remainalong
trie interval. Trie system bas evolved to a state in which triedensity
contrast is veryhigh
betweenneighboring regions,
and this behavior was observed for all values of p~ and W.This increase in trie
density
contrast can be characterizedquantitatively
in triefollowing
way. Let trie
density
of eachregion
be p~, with1indexing
trie dilferentregions,
andN(t)
be trie number ofregions
at time t. Define triequantity
N(t)
M(t)
+q ~
(PiPi+il,
19)where pN(t)+i e pi Trie
larger
trie value ofMit)
trielarger
triedensity
contrast betweenneighboring regions. Figure
4 shows triequantity M(t) M(o) averaged
over 10 simulationswith
No
" 10,ooo interfaces. Atearly times,
there is a linear increase inM(t)
with a crossoverto a
nearly
constant value at late times.At
early limes,
it ispossible
to calculateMit) analytically
and trie results are shown as trie dotted line infigure
4. In thisregime,
triechanges
mMit)
are dominatedby
trie interaction of interfaces whose movements are determinedby
trie initialconfiguration
of trie system. Trie1-où
'~
É
CJ
~Î
1
~# -
Ù
fl ~
~f
-5.oo
-
,'-6.oo
-8.oo logjo(t)
Fig.
4. Contrast,M(t),
as function of time, t. The sohd line shows trie results from averaging
over 10 simulations with
N(0)
= 10,000, and densities chosen in the range
[0.3,0.8].
Ai early limes, the increase in contrast is linear and at long limes ii becomes a constant. The dashed line shows theresults from an analytical calculation of trie short lime behavior.
calculation averages over all
possible configurations
of trie initial random densities and inter-faces,
determines trie time ai which each interface collision occurs and how much that collisionchanges
trie value ofMit).
In thisregime,
trie agreement with trie simulation isgood.
It is alsopossible
to showexactly
thatM(o)
=2W/3.
At latertimes,
after there bave been manycollisions between
interfaces,
trie structure of trie systemdepends
on tl~e nature of tl~e earlier evolution.Tl~us,
finislong
lime bel~avior is much more difficult to calculate. Trie results from trie calculation described above break down in finisregime
because trie distribution ofdensity regions
is nolonger
that of trie initial random distribution.At
long
limesMit)
m 2W.Thus,
triedensity
contrast atlong
timesis,
on average, aslarge
as trielargest density
contrasts present in trie initialconfiguration.
lt turns out that trieinteracting
kinetic waves do non createlarge
contrasts.Rallier,
trie interfaces from trie initial distribution which survive are those that bave a verylarge density
contrast [11]. Thus, while trie noise in tl~e system mayprovide
avariety
of sucl~ contrasts, tl~einteracting
kinetic waves willdestroy
all but tl~e verylargest.
This paper outlines a kinetic wave
approacl~
tounderstanding
triedensity
patterns observed in sand flowalong
a vertical tube(many
of trie details omitted here con be found in references [10,11]). However,
these ideascertainly
do not constitute acomplete theory
for trie pattems observed in trieexperimental
system. Trie role that trie flow of airplays
in this process[12],
as well as trie sources of noise in trie system, are
certainly
not well understood. Furtherexperimental investigation
of tl~ese issues would be mostenligl~tening.
From a tl~eoretical point ofview,
it is not dear wl~etl~er tl~e frictiona1force at tl~e watt and tl~e intemal pressureobey Bagnold's
law. While this form bas been observed in trie sheer cell geometry [7,13],
tl~ere bas been no direct measurement of tl~e frictiona1force forgravity
driven flow.Finally,
it is known that trie interface between two
regions
ofdiflering
densities may not be a stable514 JOURNAL DE PHYSIQUE I N°4
structure [6], and that diflusive eflects may
strongly
influence trielong
time behavior of a system ofinteracting
kinetic waves.Acknowledgments.
Trie authors would like to thank trie members of trie HLRZ
Many Body Group
forstimulating
discussions
throughout
this work.References
iii
Savage S. B., Adv. Appl. Mecll. 24(1984)
289;Savage S. B., Disorder and Granular Media, D. Bideau Ed.
(North-Holland,
Amsterdam,1992).
[2] Campbell C. S., Annu. 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;Mehta A., Granular Materials
(Springer
Verlag, New York,1993).
[5] Pôschel T., J. Phys. I France 4
(1994).
[6] Lighthill M. J. and Whitham G. B., Froc. Roy Soc. A 229
(1955)
281 and 317.[7] Bagnold R, A., Froc. R. Soc. London A 225
(1954)
49.[8] For example, Cundall P. A. and Strack O. D. L., Géotechnique 29
(1979)
47;Half P. K. and Werner B. T., Powder Technol. 48
(1986)
239;Bashir Y. M. and Goddard J. D., J. Rheol. 35
(1991)
849;Thompson P. A. and Grest G. S., Phys. Rev. Lett. 67
(1991)
1751;Ristow G., J. Phys. I France 2
(1992)
649;Taguchi Y-h., Pllys. Rev. Lett. 69
(1992)
1371;Gallas J. A. C., Herrmann H. J. and Sokolowski S., Phys. Rev. Lett. 69
(1992)
1375;Hong D. C. and McLennan J. A., Physica A187
(1992)
159[9] Lee J. and Herrmann H. J., J. Phys. A 26
(1993)
373.[loi
Lee J., Pllys. Rev. E 49(1994)
281.[Il]
Leibig M., Pllys. Rev. E 49(1994)
184.[12] Bideau D., private communication;
Hansen A., private communication.
[13] For example, Savage S. and Sayed M., J. FJuid Mecll. 142
(1984)
391;Campbell C., J. Fluid Mecll. 203