PHYSICA/

ELSEVIER

Physica A 238 (1997) 129-148

Time-dependent behavior of granular material

in a vibrating box
Jysoo Lee
Benjamin Levich Institute and Department of Physics, City College of the City University of" New York,
New York, NY 10031, USA

Received 17 May 1996

Abstract

Using numerical and analytic methods, we study the time-dependent behavior of granular
material in a vibrating box. We find, by molecular dynamics simulation, that the temporal fluctuations of the pressure and the height expansion scale in A f , where A ( f ) is the amplitude
(frequency) of the vibration. On the other hand, the fluctuations of the velocity and the granular
temperature do not scale in any simple combination of A and f . Using the kinetic theory of
Haft, we study the temporal behaviors of the hydrodynamic quantities by perturbing about their
time-averaged values in the quasi-incompressible limit. The results of the kinetic theory disagree
with the numerical simulations. The kinetic theory predicts that the whole material oscillates
roughly as a single block. However, the numerical simulations show that the region of active
particle movement is localized and moves with time, behavior very similar to the propagation
of a sound wave.
Keywords: Granular media; Vibration; Wave; Scaling; Kinetic theory

1. Introduction

Systems of granular particles (e.g. sand) exhibit many interesting phenomena, such
as segregation under vibration or shear, density waves in the outflow through a hopper
and a tube, and the formation o f heaps and convection cells under vibration [ 1-5].
These phenomena are consequences of the unusual dynamical response o f the systems,
and are for the most part still poorly understood.
We focus on the vertical vibration of a box containing granular particles. There
are many interesting phenomena associated with this system, such as convection cells
[ 6 - 1 1 ] , heap formation [ 1 2 - 1 7 ] , sub-harmonic instability [18], surface waves
[19,20] and even turbulent flows [21]. The basis for understanding these diverse
0378-4371/97/$17.00 Copyright ~ 1997 Elsevier Science B.V. All rights reserved
PH S0378-4371 (96)00281-6

130

J. Lee/Physica A 238 (1997) 129-148

phenomena is the state of granular media under vibration. The state is characterized
by the hydrodynamic fields of the system, such as the density, velocity and granular
temperature fields.
There have been several studies on the state of granular particles under vibration.
Thomas et al. studied the system in three dimensions, mainly focusing on the behavior
of shallow beds [22]. C16ment and Rajchenbach experimentally measured the density,
velocity and temperature fields of a two-dimensional vertical packing of beads [23].
They found that the temperature increases monotonically with the distance from the
bottom plate. The same system was studied by molecular dynamics (MD) simulation
with similar results [24]. In a series of simulations and experiments, Luding et al.
studied the behavior of the one- and two-dimensional systems [25-27]. They found
that the height expansion, which is the rise of the center of mass due to the vibration,
scales in the variable x = A f . Here, A and f are the amplitude and the frequency
of the vibration. Warr et al. experimentally confirmed the scaling, and they also gave
an argument for its origin [28]. In recent MD simulations of the three-dimensional
system, Lan and Rosato measured the density and temperature fields [29]. They compared the results with the theoretical predictions by Richman and Martin [30], and
found good agreement. Also, an approximate theory was developed for the system in
one dimension, which agrees with simulations in the weak and the strong dissipative
regimes [31 ].
In a previous paper, we studied the time-averaged behavior of the two-dimensional
system using numerical and analytic methods [32]. Using MD simulation, we found
that the time-averaged value of not only the expansion but also the density and the
granular temperature fields scale in x. We also used the kinetic theory of Haft [33]
to determine the time-averaged hydrodynamic fields in the quasi-incompressible limit.
The results are, in general, consistent with the numerical data, and in particular show
scaling behavior in the variable x. We found that the origin of the scaling can be
understood within the framework of the theory.
In the present paper, we extend our study to the time-dependent behavior of the
system, whose understanding is not only essential in studying various time-dependent
phenomena, but also necessary to understand the mechanisms of certain steady-state
phenomena. For example, many of the arguments for the mechanism of the convection involve the variations of certain hydrodynamic quantities over a vibration cycle
[7,8,10,11,16].
We find, by MD simulation, that the temporal fluctuations of the pressure and the
height expansion scale in x, while the velocity and the granular temperature do not
scale in any simple combination of A and f . We also study the system using the kinetic
theory of Haft, where the temporal behavior is studied by perturbing about the time
average in the quasi-incompressible limit. The results of the kinetic theory disagree
with the numerical data. The kinetic theory predicts that the whole system of particles
moves as one effective "block". The numerical simulations, on the other hand, show
that the region of active particle movement is localized, and moves with time. The
presence of the wave is partly responsible for the discrepancy.

J. LeelPhysica A 238 (1997) 129-148

131

The paper is organized as follows. In Section 2, we specify the interaction of the

particles used in the MD simulations. We then present the temporal fluctuations of the
expansion and the fields obtained by the simulations. Analytic results will be discussed
in Section 3. The continuum equations for granular material will be given, and perturbation equations are derived. We present the solution of the equations, and compare
with the numerical data. In Section 4, we check the assumptions used in obtaining
the solution. We also discuss various properties of the waves. Conclusions are given
in Section 5.

2. Numerical simulation
We start by describing the interactions used in the MD simulations. The simulations
are done in two dimensions with disk-shaped particles. The interaction between the
particles is that of Cundall and Strack [34], which allows the particles to rotate as well
as translate. Particles interact only by contact, and the force between two such particles
i and j is the following. Let the coordinate of the center of particle i ( j ) be Ri (Rj),
and r = R i - R j . We use a new coordinate system defined by two vectors h (normal)
and g (shear). Here, h = r / I r L, and g is obtained by rotating h clockwise by ~/2. The
normal component Fjni of the force acting on particle i from particle j is
Fjn~i = kn(ai 4- aj -

Irl) -

7nme( v" h) ,

(1)

where ai ( a j ) is the radius of particle i (j), and v = d r / d t . The first term is the linear

elastic force, where kn is the elastic constant of the material. The constant Vn of the
second term is the friction coefficient of the velocity-dependent damping force, and
me is the effective mass, mimj/(m i Jr mj). The shear component FjSi is given by
FjSi = -sign (5s)min(kslfSl, #lFj~il).

(2)

The term represents static friction, which requires a finite amount of force (#Fjni)
to break a contact. Here, p is the friction coefficient, 6s the total shear displacement
during a contact, and ks the elastic constant of a virtual tangential spring.
The shear force also affects the rotation of the particles. The torque acting on
particle i due to particle j is
~ . ~ i = ,-~ x

~FjS~,

(3)

where rc is the vector from the center of particle i to the point where particles i and j
overlap. Since the particles used in the simulations are very stiff (large kn), the area
of the overlap is very small. It is thus a good approximation to use - a i h as rc.
A particle can also interact with a wall. The force and torque on particle i, in
contact with a wall, are given by ( 1 ) - ( 3 ) with aj -- 0 and me = mi. A wall is
assumed to be rigid, i.e., it is not moved by collisions with particles. Also, the system
is under a gravitational field 9. A more detailed explanation of the interaction is given
elsewhere [35].

132

J. LeeIPhysica A 238 (1997) 129-148

The movements of the particles are calculated using a fifth-order predictor-corrector

method. We use two Verlet tables. One is a usual table with finite skin thickness.
The other table is a list of pairs of actually interacting particles, which is needed to
calculate the shear force. The interaction parameters used in this study are fixed as
follows, unless otherwise specified: kn = 105, k s = 105, 7n = 2 x 102 and # = 0 . 2 . The
timestep is taken to be 5 x 10 -6. This small timestep is necessary for the large elastic
constant used in the simulations. For too small values of the elastic constant, the
system loses the character of a system of distinct particles, and behaves like a viscous
material. In order to avoid artifacts of a monodisperse system (e.g., hexagonal packing),
we choose the radius of the particles from a Gaussian distribution with the mean 0.1
and the width 0.02. The density of the particles is 0.1. Throughout this paper, CGS
units are implied.
We put the particles in a two-dimensional rectangular box. The box consists of
two horizontal (top and bottom) plates which oscillate sinusoidally along the vertical
direction with given amplitude A and frequency f . The separation between the two
plates H is chosen to be much larger (105 times) than the average radius of the
particles, so the particles do not interact with the top plate for all cases studied here.
We apply a periodic boundary condition in the horizontal direction. The width of the
box is W = 1. We also try different values of W, and find no essential difference in
the following results.
We start the simulation by inserting the particles at random positions in the box.
We let them fall by gravity and wait while they lose energy by collisions. We wait for
105 iterations for the particles to relax, and during this period we keep the plates
fixed. The typical velocity at the end of the relaxation is of order 10 -2 . After the
relaxation, we vibrate the plates for 50 cycles before taking measurements in order to eliminate any transient effect. Measurements are made during the next 200
cycles.
We measure hydrodynamic quantities - density, velocity and granular temperature
which characterize the state of the system. The most detailed information is contained in the time series of their fields (e.g., density field), which will be discussed
later. Here, we want to start with something simple and representative of the
system.
The center of mass of the particles can be loosely related to the density. Let y(t)
be the vertical coordinate of the center of mass at time t. The mean density is related
to the mean interparticle distance, which is also related to y(t). Since y(t) is a scalar
which also can be easily measured, we study it as a representative of the density.
In the same spirit, we study the space-averaged vertical velocity Vy(t) and granular
temperature r(t) instead of the complete local fields. We define the spatial average
of A(x, y,t) as
-

(A(t)) = f f A(x, y,t)p(x, y,t)dxdy

f f p(x, y, t)dx dy

(4)

,L Lee/Physiea A 238 (1997) 129-148

133

where p(x, y, t) is the density field at position (x, y). It thus follows that Vy(t)= (Vy(t))
and z ( t ) = (T(t)), where Vy(X,y,t) and T(x,y,t) are the vertical velocity and the
granular temperature field, respectively. Another quantity of interest is the pressure
field p(x, y, t). For information on this quantity, we will study the total pressure at the
bottom,

po(t) = f p(x, y = O, t)p(x, y = O, t)dx

f p(x,y=O,t)dx

(5)

We studied the time-averaged behavior of the system in our previous paper [32].
Here, we study the temporal behavior, especially focusing on the variations of the fields
within a vibration cycle. We first measure the temporal fluctuation, which is defined as
the standard deviation of a temporal sequence. Let (A)t be the time average of A(t);
then, we use

fiexp = (Yexp)t,

#exp = v/(y2xp),- (Yexp)t2 ,

~0 = (p0),,

(6)

ap0 = V/(P20), - (p0)~,

where the expansion Yexp(t) is defined as the difference between y(t) during and before the vibration. We measure these quantities at every 1/100 of a period for 200
cycles.
In our previous paper, it was shown that )Sexp scales in A f , in agreement with earlier simulations and experiments [25-28]. We find that the rotation of the particles
included in the present simulation does not change this scaling, but does significantly
decrease the value of )Sexp. The decrease is probably due to the fact that the average
translational energy becomes smaller, since some of the energy is transferred to the
rotation. We then consider the fluctuation of the expansion 6Yexp. In Fig. l(a), we show
6Yexp for several values of A and f . The quality of the scaling is not very good, but
the data are still consistent with A f scaling, especially without the persistent deviation
at the low A part of the f = 20 data. This deviation is also present in the scaling

of fexp"
The behavior of z(t) is very similar. It was shown that ~ scales in A f [32]. Again,
we find that the rotation does not change the scaling of ~, but does change its value.
The situation becomes a little different when we consider the fluctuation 6z as shown
in Fig. l(b). It is clear from the figure that be does not scale in A f . In fact, it does
not scale in any of the simple combinations of A and f we have tried.
In the previous paper, the behaviors of Vy(t) and po(t) were not discussed in detail,
since their time-averaged values are trivial. Since the system is in a steady state, ~, is
zero, which is also confirmed by the simulations. Since the pressure po(t) is caused
by the weight of the particles, one might guess /3o is simply the total weight of the

J. Lee/Physica A 238 (1997) 129-148

134
0.4

/!

0.35
/"

f=20 ~--f=50 +~f=lO0

0.3

/,'

0.25

,/

./

.Z

0.2
0.15
0.1
0.05
I

0
0

(a)

10

20

15

25

Af

25

f=20
f = 50 .....
f = 100 ..D-

20

/,//

15

/////

10

.-""

.E2

10

(b)

15

20

25

Af

Fig. 1. Scaling behaviors of the hydrodynamic quantities. Each datum is averaged over at least three samples,
where 20 000 measurements are made in a sample. (a) The fluctuation of the expansion 6Yexp seems to scale
in Af , where (b) the fluctuation of the temperature fiz does not scale. (c) The fluctuation of the velocity
6Vy does not scale, where (d) the fluctuation of the pressure 6po seems to scale in Af.

particles divided by the area o f the bottom ( W = 1). W e find that /30 is indeed a
constant independent o f A and f , w h o s e value is consistent with the total weight. O n
the other hand, the b e h a v i o r o f their fluctuations is far from trivial. The fluctuation o f
the v e l o c i t y 6Vy for several values o f A and f

is s h o w n in Fig. l ( c ) . It is apparent

that 6Vy does not scale in A f . Also, 6Vy does not scale in any simple c o m b i n a t i o n o f
A and f . The data for the fluctuation o f the pressure 6po are consistent with scaling

J. Lee/Physica A 238 (1997) 129-148

135

20
18
f= 20

16
14
12
10
8
..... 4 / / / / / / / "

6
+.+~.~_4.+++ . . . . . . . . . . ~ . . G
,"
.C~'"
'
.~3..B-'~"

(c)

2 -t~ 1 2 E ] ~
0

10

15

20

25

Af
1000
/////"

900
800
700
D.Z3 / / /
.. .r /
~" / /

600

o-.?5 .....

500

~" '

....
f=100 a
f=50

!J"

400
300
o

200

(d)

10

15

20

25

Af
Fig. 1. Continued.

in Af as shown in Fig. l(d). The quality of collapse is again not very good, especially
for the low A part of the f----20 data.
The behavior of the time-averaged quantities is either trivial (l?y and /30), or scales
in Af (35exp and ~). The reason for the trivial behavior has been discussed, and the
scaling in Af can be understood from a kinetic theory of granular particles [32]. The
behavior of the fluctuations is, however, not easy to understand. For example, one might
naively expect that 6Vy behaves as Af, the velocity fluctuation of the bottom plate;
but the numerical data suggests this is not so. Also, one might guess 6po behaves
as Af 2, the variation of the effective gravity; but the numerical simulations suggest
scaling in Af. Thus, the behaviors of the temporal fluctuations seem to be inconsistent

J. LeeIPhysica A 238 (1997) 129 148

136

with an intuitive picture. In the next section, we discuss an attempt to understand this
behavior.

3. Kinetic theory
In this section, we study the time-dependent behavior of the system using a kinetic
theory of granular material. We use the formalism by Haft [33], which was successfully
applied to the time-averaged behavior of the system [32]. For more details on the
formalism and other kinetic theories of granular material, see Ref. [32] and references
therein.
HaWs formulation consists of equations of motion for mass, momentum and energy
conservation. The mass conservation equation is
~ p + I7. ( p v ) = 0 ,

(7)

where p and v are the density and the velocity fields, respectively. Next is the ith
component of the momentum conservation equation,

vi+p(v. 17)v,=

3 [ \ax,
(~Vj+ axjJJ
~Vi~1+pot,

(81

where summation over index j is implied. The coefficients 2 and r/ are viscosities
which will be determined later. Also, p is the internal pressure, and g~ is the ith component of the gravitational field. Although (8) resembles the Navier-Stokes equation,
the coefficients as well as the internal pressure are now functions of the fields instead
of being constant. The last of the equations of motion is energy conservation,

at

+ G

, \oxj +

x,j

+ pvigi + ~xi
Here, T is the granular temperature field, K is the "thermal conductivity", I is the
rate of the dissipation due to inelastic collisions, and summations over indices i and j
are implied. Although the form of (9) is somewhat different from that of the NavierStokes equations, the equation can still be easily understood. The left-hand side of (9) is
simply the material derivative of the total kinetic energy, where the total kinetic energy
is divided into the convective part (involving v) and the fluctuating part (involving T).
On the right-hand side of the equation, the first three lines are simply the rate of work
done by the internal pressure, viscosity and gravity, respectively. The term involving

J. Lee/Physica A 238 (1997) 129-148

137

K is the rate of energy transported by "thermal conduction". The dissipation term I,

which is a consequence of the inelasticity of the particles, is responsible for many of
the unique properties of granular material.
We now discuss the coefficients which are yet to be determined. Derivation of the
relations of these coefficients to the fields is based on intuitive arguments [33]. Also,
the derivation assumes that the density is not significantly smaller than the closepacked density, i.e., the system is almost incompressible. The relation for the internal
pressure is
T
p = tdp-- ,

(10)

where t is an undetermined constant, and d is the average diameter of the particles. The
variable s, which is roughly the gap between the particles, is related to the density by
m

p~

(d+s)3,

(11)

where m is the average mass of the particles. Then, the viscosity r/ is given as

rl=qd2p v@ ,

(12)

where q is an undetermined constant. In a similar way, the thermal conductivity is

found to be

(13)

K=rd 2~
S

Here again, r is an undetermined constant. Finally, the rate of dissipation is

T3/2
1 =7P--,

(14)

where 7 is an undetermined constant. The viscosity 2 is left undetermined, due to the

fact that, in the range where these relations are valid, the term containing 2 is negligible
and is dropped from the calculation.
We impose two constraints in order to make the equations analytically tractable. The
first is the horizontal periodic boundary condition. Due to the boundary condition, there
are no significant variations of the fields along the horizontal direction. Thus, we only
have to deal with a one-dimensional equation instead of a two or three-dimensional
one. The other constraint is incompressibility, which is a little tricky. Incompressibility
implies, strictly speaking, that the density p is constant. Due to the relation between
p and s (11 ), s also has to be constant. Here, we are interested in the situation where
s is much smaller than d, but still non-zero. In such a case, the variation of the density
can be ignored, but not the variation of a variable that depends directly on s. We call
this condition quasi-incompressibility.

138

J. Lee/Physica A 238 (1997) 129-148

Under these conditions, we solved the equations for the time-averaged fields,
T()(y) ~_ B2v 2
Y
exp(-2y/(),
Wy 0 - y

s()(Y) "~

tZdZPOBZv2
yo
g
w (Yo - y)2 e x p ( - 2 y / ( ) ,

(15)

v{)(y) = 0 ,
p(O)(y) = Pog(Yo - Y)

Here, Vw = 2 n A f the maximum velocity of the bottom, y0 the height of the free surface,
and = v / ~ d the dissipation length. Also, P0 is the density of the maximum packing,
and
B2 _-

(1 + ew) 2
1 - e2w - ( 2 r d / a ( ) ( 1 - (/2y0) '

(16)

where ew is the coefficient of restitution of collisions between a particle and a wall [32].
We study the time-dependent behavior of a quantity by perturbing it from its timeaveraged value. We assume that the perturbation term oscillates with f - the frequency
of the vibration. In general, the assumption is not valid, since one has to consider all
the modes with different frequencies. However, when the amplitude of the vibration is
small enough, in many cases, the mode with frequency f dominates the time-dependent
behavior. We expect there is a range of A in which the assumption is valid, which will
be determined later by the numerical simulations. We thus use
T ( y , t) = T()(y) + T ( I ) ( y ) exp(io~t),
s(y, t) =s()(y) + s ( l ) ( y ) exp(itnt),

(17)
v y ( y , t ) = V v(0)(Y) + v~ )(y) exp(/~ot)
p ( y , t) = p(O)(y) + p ( l ) ( y ) exp(i~ot).

Substituting (17) into the mass conservation condition (7) and using the quasiincompressibility condition, we obtain dv~O(y)/dy = 0. Here and in the rest of the
calculation, we consider terms up to the first order in the expansion. Since Vy ( y = 0 , t)
should be the velocity of the bottom plate,
V(yl)(y)=iVw ,

(18)

which is a consequence of quasi-incompressibility and the one-dimensional nature of

the system. Also, momentum conservation equation (8), combined with (17), becomes
pO)(y)=poA~oZ(y o - y),

(19)

which can be easily understood. The pressure at the bottom is proportional to the total
weight of the particles. Thus, the fluctuation of po(t) is the total mass PoYo times the

J. Lee/Physica A 238 (1997) 129-148

139

fluctuation of the effective gravity//o92 . Finally, we consider the energy conservation

equation (9). Combined with (17), it becomes

io9poU () U (1) + ipov(y1) U () d u(O)

dy

rd a (p(O) d_u(1) +
U(0))
t dy
P~l) dy
d--dy
~d(U{1}p {) + U{)p{~)),

(20)

where we introduce the variable U(y, t ) = ~ .

One has to solve (20) for T(y, t).
We cannot, unfortunately, get an analytic solution of the resulting non-linear
differential equation. Since s(y, t) has to be calculated from the relation (10) between
T(y,t), s(y,t) and p(y,t), we also cannot get an expression for s(l)(y).
We compare the results from the kinetic theory with the numerical simulations
(Fig. 1). First, we consider Vy(y, t). Since the temporal fluctuation of vy(y, t) is proportional to Vy(l)(y), the kinetic theory predicts 6Vy ,.~ Af. The data from the simulations,
however, are inconsistent with this scaling. The situation is similar for p(y,t). The
kinetic theory predicts bpo ~ A f 2, while the numerical data scale in A f . We do
not have analytic expressions for T(y, t) and s(y, t) to compare with the simulation
data.
The failure of the kinetic theory, when applied to the time-dependent behavior, is
in sharp contrast with its success in studying the time-averaged behavior. The kinetic
theory correctly predicts the scaling behavior of all the time-averaged hydrodynamic
quantities. We suspect the failure is associated with the breakdown of a key assumption(s) used in the theory. Three key assumptions, besides the ones employed in the
formulation of the kinetic theory, are made to obtain a simple analytic solution for
the time-dependent behavior. The first is quasi-incompressibility, which is shown to
be valid for A f ~ l . 5 from the study of the time-averaged behavior [32]. However,
the scaling behavior of the time-averaged quantities remains unchanged even in the
compressible regime. The second assumption is that the time dependence of a quantity is a sinusoidal oscillation with frequency f , which we expect to be valid for
small A. The last assumption is that the time-dependent term in (17) is much smaller
than its time-averaged value in order for the perturbation to be valid. In the next
section, we check the validity of these assumptions by comparing with the numerical
simulations.

4. Validity of assumptions
First, we check the assumption that the mode with the driving frequency dominates
the time-dependent behavior of the hydrodynamic quantities. We obtain a time series
of yexp(t) by measuring it at every 1/100 of a period for 200 cycles. We then calculate

140

Z Lee/Physica A 238 (1997) 129-148

0.06

0.05
f= 100, A=0.01 - 0.04
0.03
0.02
0.01

(a)

0.002

0.004

0.006
f (x 104 Hz)

0.008

0.01

0.25

0.2
f = 100, A = 0.03 - 0.15

0.05

^~-^

0
(b)

0.002

0.004

_.

0.006
f (x 104 Hz)

, I

0.008

0.01

Fig. 2. Power spectrum of yeo(t) with f = 100 and (a) A 0.01, (b) A =0.03. The expansion is measured
at every 10-4s for 200 cycles. The mode with f = 100 is dominant at A =0.01, but loses its dominance at
A = 0.03.

the p o w e r spectrum o f y e x p ( t ) - )~exp using a F F T routine in the N A G library (c06gbf).

The results with f = 100 are s h o w n in Fig. 2. The m o d e with f = 100 is d o m i n a n t
for A = 0.01 (Fig. 2(a)). W h e n A is increased further to 0.03, h o w e v e r , the f
m o d e b e c o m e s d o m i n a n t (Fig. 2(b)). The results with f = 2 0

_~ 20

are entirely similar. The

f = 20 m o d e loses its d o m i n a n c e w h e n A is increased to about 0.5. In both cases, the

m o d e with frequency f is d o m i n a n t until F ~ 10, where F=Ao92/g.

J. Lee/Physica A 238 (1997) 129-148

2.5
2

L .

//

"',~

i .....

,~-f'--"~ "%
" ,'~
~::~,,

1.5

141

A = 0.005
A =0.01 ......

0.5
>

0
-0.5
-1
-1.5
-2
-2.5

(a)

"',, "'----J

10

20

30

40
50
60
ohase (x 2~ / 100)

70

[ " "

80

.f

90

100

450
400
350

\',,,

300
250

_-o0o5

h = 0.01 . . . . . .

,," /

A = 0.03

',i"'

/" ..t/

200
150
100
50
0
0
(b)

10

20

30

40
50
60
phase (x 2 ~ / 100)

70

80

90

100

Fig. 3. Time evolution of (a) Vy(t), (b) po(t) in one cycle, where f = 100 and the data are averaged over
200 cycles, Deviation from a sinusoidal is apparent for A =0.03.

The m e a s u r e m e n t s o f Vy(t) and po(t) also support the a b o v e observations. In

Fig. 3(a), we s h o w the variation o f Vy(t) in one cycle, w h e r e f = 100 and the data
are a v e r a g e d o v e r 200 periods. The curve with A = 0.01 is nearly sinusoidal, w h i l e
clear deviation is seen for A = 0.03. A n important point to note is the b e h a v i o r o f the
m a x i m u m value o f Vy(t). The kinetic theory predicts, in (18), the m a x i m u m value to
be proportional to A, w h i c h is clearly not consistent with the data. The m e a s u r e d value
is quite smaller than what is predicted. For example, for A = 0 . 0 3 , the predicted v a l u e is

J. Lee/Physica A 238 (1997) 129 148

142

Vy(

10
5
0

100
90

-5
-10

(X 2 ~ / 100)
(a)

T(y,t)

01 'q
0.6 ~
0:4
0.5
(b)

0100

~
y

.
~
1~ - - - - - . . . . . ~
1.5

40

phase (x 2 rc / 100)

~f2030
10

Fig. 4. Time evolution of (a) Vy(y,t), (b) T(y,t) and (c) p(y,t) fields, where f = 100, A=0.01 and the
data are averaged over 200 cycles. The density field is normalized to be the volume fraction.

6n, while the measured value is about 2.4. In Fig. 3(b), we show po(t) in one period,
where again f = 100 and the data are averaged over 200 cycles. The curve seems to
deviate from the sinusoidal even for small A, and it is difficult to determine whether
the mode with f = 100 dominates. Again, the predicted behavior of the maximum
value of po(t) is not consistent with the measurement. The maximum value of po(t)
is predicted to increase linearly with A, as in (19), which is clearly not consistent with

143

J. LeelPhysica A 238 (1997) 129-148

p(:

0.8
0.6
0.4

100
)0

(x 2 rt / 100)
(c)
Fig. 4. Continued.

the data. The observed maximum value (~400) is quite smaller than the predicted
value (~4000).
The mode with the driving frequency dominates the time-dependent behavior for
small values of A, where a rough criterion for the dominance is F,~ 10. Also, it was
shown that quasi-incompressibility is valid for A f < 1.5 [32]. The validity of the
linear perturbation approximation (17) is a little tricky. We require that the perturbation terms are smaller than the time-averaged terms, which is valid for the expansion
and the temperature. The two terms are, however, comparable for the pressure even
at a small value of A = 0.05. The consequence of the large pressure fluctuation on
the validity of the perturbation is unclear. For large A, all of the above assumptions
are not valid, which complicates the analysis of the system. For example, in order to
study the time-dependent behavior, one has to consider additional modes with different
frequencies.
The surprise is that the predictions for the maximum values of the vertical velocity
and the pressure are not correct even when all the assumptions seem to be valid (e.g.,
A = 0.01 and f = 100). We inspect again the predictions of the kinetic theory. As given
in (l 8), the velocity field is uniform, and its value is Vw the velocity of the bottom.
Therefore, the solution of the perturbation expansion suggests that the whole system of
particles is moving as a single "block" attached to the bottom. The spatial and temporal
variations of the other fields do not change the single block picture, but rather describe
the structure of the block. A consequence of the picture is that the pressure at the
bottom is proportional to the effective gravity which reaches its maximum (F + 1 )9 at
phase 3zr/2 of the vibration, which is exactly (19).
However, the measurements of Vy(t) and po(t) are not consistent with the picture.
The measured maximum value of Vy(t) is much smaller than what is predicted, which

144

.L Lee/Physica A 238 (1997) 129 148

suggests that only a small fraction of the particles move together at a given time. Also,
the fact that the measured maximum value of po(t) is smaller than the prediction also
supports this observation. Furthermore, the phase at which po(t) reaches the maximum
is about 0, in contrast to 3n/2 suggested by the single block picture. The discrepancy
can be understood as follows. Since the maximum acceleration of the bottom is larger
than that of gravity, particles initially lying on the bottom will be "launched" at a
certain phase of the vibration. The maximum pressure at the bottom will occur when
most of the launched particles come back and collide with the bottom, which occurs
around qS=0.
The direct evidence against the single block picture is the time evolution of the
whole velocity field shown in Fig. 4(a). Here, we use f = 100, A--0.01, at which the
assumptions used to derive the predictions of the kinetic theory seem to be valid. It
is clear from the figure that the region of significant motion is localized, and travels
like a wave. The propagation of the disturbance seems to be very similar to that of
sound waves in a gas. Also, the maximum velocity is about 6, close to the prediction 2n. The localization of the particle motion can also be seen in the time evolution of the granular temperature field shown in Fig. 4(b). Again, the region of high
temperature is localized, and travels upwards [36]. Furthermore, the location of the
high-temperature region coincides with that of the large velocity region. The density
field, on the other hand, does not vary significantly as shown in Fig. 4(c), which
agrees with the previous experiment [23]. It is clear that the presence of the "waves"
changes the behaviors of the fields, and possibly their scaling properties. The absence
of the waves in the single block solution is, at least, partly responsible for the failure of the kinetic theory. The absence is due to quasi-incompressibility. In fact, it is
easy to derive the single block picture only from the quasi-incompressibility condition and one-dimensional nature of the equation. It is thus necessary to consider the
general case of the kinetic theory of a compressible gas, which unfortunately is quite
complicated.
We want to finish this section by discussing some properties of the waves. We first
consider the motion of the maximum disturbance. In Fig. 5(a), we show the phase
q~max(Y) at which t~(y,t) reaches a maximum with f = 100 and several values of
A. In other words, we plot the position of the maximum velocity in the y-q~ plane.
The velocity of the wave, inversely proportional to the slope of the q~max(Y) curve,
is a bit small. The time needed for the wave to propagate from the bottom to the
top of the pile is of the order of the period of the vibration. The simulations with
./=50 show that, for the same values of A, the velocity of the wave does not change
significantly, suggesting that there is a fixed time scale for the wave propagation. Also,
it can be seen from the figure that the velocity decreases when either y or A increases
with the other parameters fixed. The decrease probably results from the decrease of
the collision frequency between particles, due to the decrease of the density. Also,
the location of the maximum granular temperature in the y-q5 plane is shown in
Fig. 5(b), where one can see the close correlation with Fig. 5(a). In fact, the maximum
temperature always occurs just above the maximum velocity at a given phase, which

Z Lee/Physica A 238 (1997) 129-148

90

145

80
A=001
A = 0.05 ....

70

,~

,':,".',()"A
,(,.'/ ~
=::9 /

.. . . . . . .

60
50
40
30
20
10
0
(a)

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Y
90

80
A = 0.01
A = 0.05

70

......

. ........

.i:i'~"<
7 /

60
50
-e-

40
30
20
10
0

(b)

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Fig. 5. The position of (a) the maximum velocity and (b) the maximum temperature in the y - plane with
f = 100 and several values of A. The data are averaged over 200 cycles. Note that the two sets of the
curves are almost identical.

probably is the point o f the largest v e l o c i t y gradient. W e n o w discuss the values o f the
m a x i m u m disturbances. In Fig. 6(a), w e s h o w the m a x i m u m value o f Vy(y,t) for g i v e n
y - Vy(y, ~bmax(Y)). The m a x i m u m value decreases with y r o u g h l y as an exponential,
and the rate o f the decrease is larger for larger A with fixed f . The decrease is due to
two causes: (1) s o m e o f the vertical v e l o c i t y c o m p o n e n t is transferred to the horizontal
one by interparticle collisions; (2) the kinetic e n e r g y is lost by inelastic collisions. The
results with f = 20 are essentially the same. Also, the b e h a v i o r o f the value o f the
m a x i m u m temperature T(y,C~max(Y)), as s h o w n in Fig. 6(b), is v e r y similar to that

146

Lee/Physica A 238 (1997) 129-148

J.

........"'---.iiiii' . . . .

A = 0.05
A=0.1

....
.....

10
e:

0.2

0.4

0.6

0.8

(a)

1.2

1.4

1.6

1.8

10

""""

..........
,,

A = O . O I

- -

0.1

0.01

(b)

0.5

1.5

2
Y

2.5

3.5

Fig. 6. The maximum value of (a) the vertical velocity Vy(y,qbmax(Y)) and (b) the temperature
at height y with f 200. The data are averaged over 200 cycles.

of

Vy(Y,~max(t)).

T(y, ~max(y))

It d e c r e a s e s r o u g h l y as a n e x p o n e n t i a l , a n d it d e c a y s faster w i t h

larger A w i t h fixed f ,

w h e r e the o r i g i n o f the d e c r e a s e is the s a m e as the v e l o c i t y

field.

5. Conclusion
W e h a v e studied the t e m p o r a l b e h a v i o r o f g r a n u l a r m a t e r i a l in a v i b r a t i n g box. W e
find t h a t the t e m p o r a l fluctuations o f Yexp(t) a n d

po(t)

scale in

Af,

while no scaling

J. Lee/Physica A 238 (1997) 129-148

147

is found for the fluctuations of z(t) and Vy(t). We study the behavior using the kinetic
theory of Haft, where we perturb the hydrodynamic quantities from their time-averaged
values. The results of the kinetic theory are not consistent with the numerical data. We
argue that the failure of the theory is, at least, partly due to the waves found in the
simulations.
We discuss some possible ways to study the time-dependent behavior of the system.
Since the main problem of the present theory, as discussed above, is the condition
of quasi-incompressibility, it sounds reasonable to study the system of equations in a
fully compressible regime. However, the system of equations becomes too complicated.
The three conservation equations ( 7 ) - ( 9 ) are written in general form, and do not
need any modification. The relations of p, rl, K , I to the fields (10)-(14) as well as
the boundary conditions have to be modified. It is not the modifications themselves,
but the complexity of the resulting equations, that makes an analytic solution too
difficult to obtain. However, we can still gain some information about the system by a
perturbative or an approximate method, as well as the numerical solution of the kinetic
equations.

Acknowledgements
We thank Joel Koplik for many useful discussions and comments on the manuscript,
and Michael Tanksley for a critical reading of the manuscript. This work is supported
in part by the Department of Energy under grant DE-FG02-93-ER14327.

References
[1] S.B. Savage, Adv. Appl. Mech. 24 (1984) 289; S.B. Savage, Disorder and Granular Media, ed.
D. Bideau (North-Holland, Amsterdam, 1992).
[2] C.S. Campbell, Annu. Rev. Fluid Mech. 22 (1990) 57.
[3] H.M. Jaeger and S.R. Nagel, Science 255 (1992) 1523.
[4] A. Mehta, Granular Matter: an Interdisciplinary Approach, ed. A. Mehta (Springer, New York, 1993).
[5] G. Ristow, Annu. Rev. Comp. Phys. I, ed. D. Stauffer (World Scientific, Singapore, 1994).
[6] O. Zik and J. Stavans, Europhys. Lett. 16 (1991) 255.
[7] Y.-H. Taguchi, Phys. Rev. Lett. 69 (1992) 1371 .
[8] J.A.C. Gallas, H.J. Herrmann and S. Sokolowski, Phys. Rev. Lett. 69 (1992) 1375.
[9] S. Luding, E. C16ment, A. Blumen, J. Rajchenbach and J. Duran, Phys. Rev. E 50 (1994) R1762.
[10] E.E. Ehrichs, H.M. Jaeger, G.S. Karczmar, J.B. Knight, V.Yu. Kupperman and S.R. Nagel, Science 267
(1995) 1632.
[11] H. Hayakawa, S. Yue and D. Hong, Phys. Rev. Lett. 75 (1995) 2328.
[12] M. Faraday, Phil. Trans. Roy. Soc. London 52 (1831) 299.
[13] P. Evesque and J. Rajchenbach, Phys. Rev. Lett. 62 (1989) 44.
[14] C. Laroche, S. Douady and S. Fauve, J. de Physique 50 (1989) 699.
[15] E. Clement, J. Duran and J. Rajchenbach, Phys. Rev. Lett. 69 (1992) 1189.
[16] J. Lee, J. Phys. 27 (1994) L257.
[17] H. Pak, E. Van Doom and R. Behringer, Phys. Rev. Lett. 74 (1995) 4643.
[18] S. Douady, S. Fauve and C. Laroche, Europhys. Lett. 8 (1989) 621.
[19] H.K. Pak and R.P. Behringer, Phys. Rev. Lett. 71 (1993) 1832.
[20] F. Melo, P. Umbanhower and H.L. Swinney, Phys. Rev. Lett. 72 (1994) 172.

148
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]

J. Lee/Physica A 238 (1997) 129-148

Y.-H. Taguchi, J. Phys. 1I (France) 2 (1992) 2103.
B. Thomas, M.O. Mason, Y.A. Liu and A.M. Squires, Powder Technol. 57 (1989) 267.
E. Cl6ment and J. Rajchenbach, Europhys. Lett. 16 (1991) 133.
J.A.C. Gallas, H.J. Herrmann and S. Sokotowski, Physica A 189 (1992) 437.
S. Luding, E. Cl6ment, A. Blumen, J. Rajchenbach and J. Duran, Phys. Rev. E 49 (1994) 1634.
S. Luding, H.J. Henmann and A. Blumen, Phys. Rev. E 50 (1994) 3100.
S. Luding, Ph.D. Thesis, University of Freiburg (1994).
S. Warr, J. Huntley and G. Jacques, Phys. Rev. E 52 (1995) 5583.
Y. Lan and A.D. Rosato, Phys. Fluid A 7 (1995) 1818.
M.W. Richman and R.E. Martin, Proc. 9th Conf. Eng. Mech., ed. L.D. Lutes and J.N. Niezwecki
(New York, 1992).
R. Mazighi, B. Bernu and F. Delyon, Phys. Rev. E 50 (1994) 4551.
J. Lee, Physica A 219 (1995) 305.
P.K. Haft, J. Fluid Mech. 134 (1983) 401.
P.A. Cundall and O.D.L. Strack, G6otechnique 29 (1979) 47.
J. Lee and H.J. Herrmann, J. Phys. A 26 (1993) 373. Note that particles were not allowed to rotate.
The wave was first seen in the video made from the MD simulations of granular particles by G. Ristow
and H. Herrmann (1992).