Eur. Phys. J.

Special Topics 146, 331340 (2007)

c EDP Sciences, Springer-Verlag 2007
DOI: 10.1140/epjst/e2007-00191-x
THE EUROPEAN
PHYSICAL JOURNAL
SPECIAL TOPICS
Convection in a vibrated granular layer
A. Garcimartn
a
, J.M. Pastor, R. Arevalo, and D. Maza
Departamento de Fsica y Matem atica Aplicada, Facultad de Ciencias, Universidad de Navarra,
31080 Pamplona, Spain
Abstract. When a granular layer is submitted to an oscillating acceleration with
a peak value larger than gravity, a large scale motion develops. This movement
is in some ways similar to the one displayed by a liquid heated from below, and
it is called granular convection. Dierent conditions beside the parameters of the
forcing can aect it, such as the presence of an interstitial gas or the roughness of
the walls. We have carried out an experiment to study the convective movement
of a granular layer with a temporal resolution high enough to describe the motion
of individual grains within one oscillating period. We also present experimental
results concerning the friction that the lateral walls exert on the grains and its
relevance on granular convection.
1 Introduction
An isolated system in equilibrium will always evolve so as to increase its entropy or at least
to keep it constant. Typically this means that the system is fully disordered and that it lacks
any spatial structure. But this does not hold for a dissipative system, where a continuous supply
of energy can lead it towards an ordered dynamic state out of equilibrium. Benard convection
is a paradigmatic example of such a behavior: a liquid layer heated from below dissipates the
injected energy by adopting an orderly motion: a pattern of convective cells. The features of
these cells depend on geometrical parameters, such as the ratio between the liquid depth and
the lateral dimension of the container, and on the characteristics of the uid. Such kind of
behavior is therefore expected for granular media, which are inherently dissipative. This is so
because although the material from which the grains are made can have a high restitution
coecient so that it can be considered elastic the rate of collisions among grains is so large
that it usually causes a rapid dissipation of energy. It is therefore not surprising that ordered
states out of equilibrium appear easily when energy is supplied to a granular media [1, 2].
Injecting energy into such a system to drive it into a dynamical state can be accom-
plished in four ways: shearing, shaking, through a uid drag, and under the action of gravity.
Here we will deal with the out of equilibrium dynamical states that appear when a granular
layer is subjected to a vibration in the same direction than gravity. In 1831, M. Faraday [3]
reported that grains deposited on a vibrating plate formed a heap, and investigated the inu-
ence of the air on such an structure. The heap has an internal circulation: grains go up under
de apex, descend down the sides and go towards the center at the base, closing the trajectory.
Faradays work was pioneering in the sense that it is one of the rst reports concerning the
self-organization of an out of equilibrium system. Nevertheless, not much attention was paid to
this phenomenon until a couple of decades ago. In 1988, S.B. Savage [4] described the motion
of beads in a rectangular box with a vibrating exible base. The xed side walls and the fact
that the vibration at the base was spatially non-uniform diers from most experiments car-
ried out since then, but many results remain qualitatively unchanged. He observed symmetric
a
e-mail: angel@fisica.unav.es
332 The European Physical Journal Special Topics
closed vortices in which the velocity was upwards at the vertical centreline and downwards along
the vertical sidewalls, resembling the convective cells in heated liquid layers. Maybe due to this
likeness this motion is now known as granular convection. Shortly after this work, Evesque
and Rajchenbach [5] demonstrated experimentally that the pertinent control parameter is
the peak acceleration of the container. (Usually, the excitation is a harmonic oscillation, and
the peak acceleration is normalized to gravity to produce an adimensional quantity: =
Af
2
4
2
g
,
where A is the amplitude and f is the frequency of the driving).
Despite the apparent simplicity of this phenomenon, the origin of the coherent movements
observed in the experiments is not well understood. In Benard-Marangoni convection, the typi-
cal honeycomb pattern appears when the inuence of the lateral boundaries can be neglected [6],
and in the case of small aspect ratio systems boundary conditions completely dene the result-
ing pattern [7]. Unfortunately, a simple analogy with the granular convection can not be intro-
duced. For instance, the convective threshold in a heated liquid layer is well dened through the
dimensionless Rayleigh number. This number depends on the geometrical parameters such
as the liquid depth and on the departure from the equilibrium state, which is quantied by
the temperature dierence between the top and the bottom surfaces. The instability threshold
can be expressed as the ratio between the time scales of the energy input and the dissipative
processes. In granular media, on the other hand, it is dicult to identify such temporal and
geometrical scales. Moreover, the depth of the material can determine the kind of instability
that will appear. For a shallow granular layer (it is generally understood that this means that
the depth is smaller than about ten grain diameters) there is not a large scale circulation:
instead, waves of dierent kinds appear [8]. If, on the other hand, the height of the conned
layer is large enough, collective large scale circulation (spanning over the whole layer) can be
observed.
Laroche [9] and Clement [10] proceeded to describe convection and raised for the rst time
the question of the inuence of the lateral walls and the presence of interstitial air in this circu-
lation. Another subject raised in those papers was the uidization of the granular medium,
meaning that the vibrated grains ow like a liquid. Further work uncovered the relevance of
three possible mechanisms for the granular convection. Pak and collaborators [11] studied the
inuence of the interstitial gas and found that convection is greatly impaired in an evacuated
contained. Arguments still go on about the signicance of this nding, although many believe
that air acts mainly as a lubricant. Convection therefore would stop in the absence of air
because of internal friction among grains, but air if this is true would not directly cause
convection. Knight et al. [12] reported measurements of the velocity of a tracer particle inside a
tapped granular layer, obtained by magnetic resonance imaging. They concluded that the fric-
tion with walls was related to the streaming velocity. Finally, numerical simulations [13] showed
that a thermal diusion from the vibrating plate could cause similar motions in diluted granular
gases. For a dilute gas (where dissipation is not so fast as in a dense granular medium) hydro-
dynamic theories can be used to describe the problem by introducing a temperature obtained
from the mean kinetic energy. The diusion of the kinetic energy can drive convection, but this
mechanism seems to be restricted to uidized grains with low densities.
In this article, we will rst describe our experiment, and provide an insight into the mea-
surements that are accessible with our equipment. We will rst present results concerning the
motion of the whole granular layer, which shows a period doubling bifurcation in the ight
time as the control parameter is increased. In this section, the inuence of interstitial air will
be analyzed. We will then turn to the motion of individual grains near the walls in order to
study the cause for the relative motion within the granular layer (the granular convection) and
describe the patterns that appear. In the last section, we will oer some conclusions.
2 Experimental setup
In order to describe the motion of the granular layer, we will rely mainly on two measurements:
the instantaneous acceleration of the excitation and the velocity of the grains. The rst one
can be easily obtained by attaching an accelerometer to the container. Its signal is picked up
New Trends, Dynamics and Scales in Pattern Formation 333
C
S
O
A
FG
PC
Fig. 1. Sketch of the experimental setup. C: High-speed camera; S: shaker; FG: function generator;
A: accelerometer; O: oscilloscope; PC: computer.
by an oscilloscope and data are registered by a computer (see gure 1). The velocity of the
grains has been obtained by particle tracking. A high-speed camera (Photron FASTCAM X-
1024 PCI) is used to record a movie at a suitable frame rate, depending on the case. Typically,
1000 frames per second is enough to resolve the fast motion of the grains within one vibrating
cycle. The movie is then downloaded to a computer, where individual grains are tracked with
an image processing software. Standard particle tracking techniques provide a resolution much
better than the particle diameter. Spatio-temporal diagrams can also be used to obtain the
velocities of the grains. Additional information about this technique can be found in [14].
The container is attached to the shaft of an electromechanical shaker (TiraVib 52100)
capable of delivering a sinusoidal acceleration with a peak value up to 10 g. The shaker is
feeded from a function generator (Stanford Research DS 345) which is in turn commanded
from a computer. In this way, the experiment can be automated and the parameter can be
easily controlled. In this experiment the frequency is xed at f = 110 Hz and is varied by
changing the vibration amplitude.
Almost any granular material can be used to observe convection. As a large number of beads
is desirable, the grains must be small. On the other hand, when grains are too small, cohesive
forces due to humidity and electrostatic forces can exceed the weight of a grain, and undesirable
eects appear. Glass beads with a diameter d = 0.43 0.1 mm oer a good compromise and
have been used in this experiment. The density of the glass is
b
= 2.5 g/cm
3
.
Dierent containers have been used to study the eect of the lateral walls on granular con-
vection. They will be described later with the corresponding results. All of them are cylindrical
boxes with an inner diameter of 52 mm, so that the diameter of the container is much larger
than the diameter of one bead. The height of the layer is measured in units of particle diame-
ters, i.e., N = h/d, where h is the height of the layer. Granular convection is usually observed
if N 10, a threshold for which an explanation has been oered [2]. In our experiment N is
always greater than 10.
3 Bifurcations
The dynamics of the granular layer as a whole is determined by the amount of energy injected,
which is quantied by the control parameter , as suggested in [5]. As the frequency is xed
at f = 110 Hz (the choice of this particular value is not crucial), is varied by changing the
amplitude of the oscillations. As the control parameter is increased, the sequence of events
for the whole layer is the following. For < 1 the granular layer will remain in contact with
the container during all the cycle. But when the peak acceleration of the forcing overcomes
the gravity, the layer will loose contact with the vibrating plate and perform a free ight.
Therefore the primary instability is expected to take place at = 1, because the grains will
then be allowed to move while performing their ight.
Just above the threshold, the motion of a the granular layer consists of cyclic jumps at
the same frequency than the forcing. The layer begins to y at the moment in the cycle where
the instantaneous value of the acceleration is equal to g. The value of the velocity and the phase
at this precise instant provide the initial conditions for the ight. Gravity and whatever forces
334 The European Physical Journal Special Topics
act on the granular layer cause a downward acceleration. The layer falls and it lands when it
bumps into the base. As it is composed of a large number of particles, it can be deemed as
an almost perfectly inelastic body, so the energy is completely lost at the collision. Therefore
the layer gets stuck to the vibrating plate. It begins to y again when the instantaneous accel-
eration overcomes the gravity. Let us call T the period of the forcing (as f is xed at 110 Hz,
T = 0.00909 s); then the granular layer spends a time ying and a time T stuck to the
base. The ight time can be obtained from the accelerometer signal (see for example [8],
gure 3), because the impact of the granular layer against the vibrating plate leaves an imprint
on the acceleration.
As is increased, so does the ight time. But if ight times are calculated by assuming that
the granular layer is a perfectly inelastic body upon which only gravity is acting, then the ex-
perimental results are not recovered. The measured ight times are shorter than predicted [15].
Kroll [16] pointed out that the eect of interstitial air was likely to be the cause. Gutman [17]
proposed an improved model that included the compressibility of air. In brief, the following
simplifying assumptions are made [18]: (1) The granular layer can be likened to a porous pis-
ton (2) The collision between the piston and the vibrating plate is almost instantaneous when
compared to the vibrating period (3) Friction against walls is negligible (4) Permeation of air
through the grains can be described by Darcys law (5) The porosity of the granular layer is
constant.
From these hypothesis, the following can be deduced [18]. Darcys law states that pressure
drop in the granular layer must be proportional to the velocity u with which air passes through
it, so
P
z
=

u (1)
Here P is the pressure, z is the vertical coordinate, is the air viscosity and is the porosity
of the granular layer, that can be approximated by the CarmanKozeny expression [19]
=

3
180(1 )
2
d
2
(2)
In this equation = 1 is the compaction fraction. Under Krolls assumption (5) it can
be shown that the pressure P inside the granular medium obeys the following diusion
equation [20]
P
t
=
P
0

(1 )

2
P

2
z
(3)
where P
0
is the atmospheric pressure. In order to solve equation (3) the following boundary
conditions are used. At the top of the layer, P
z=0
= P
0
. And at the bottom of the layer (z = h),
the continuity equation holds that
d
g
dt
=
g,z=0
, u =
g,z=0

P
z
(4)
where
g,z=0
is the gas density at P
0
and s is the height of the air gap between the vibrating
plate and the granular layer. If the pressure change is small compared to P
0
, then
g
does never
dier a lot from
g,z=0
, so s(d
g
/dt)
g,z=0
(ds/dt) and equation 4 now reads
ds
dt
=

P
z

z=h
(5)
In the laboratory reference frame, the equation of motion for the granular layer is
d
2
(s + x)
dt
2
= g +
P
z=h
P
0

b
h
(6)
where x(t) is the driving vibration and
b
is the density of the beads. By integrating equa-
tions (3)(6) in one dimension, it is possible to obtain the pressure dierence in a cycle, as well
as the trajectory of the mass center of the granular layer and its ight time.
New Trends, Dynamics and Scales in Pattern Formation 335
0 0.5 1 1.5 2 2.5 3 3.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Fig. 2. Flight times normalized to the driving period T as a function of . Results obtained from the
Gutman model (solid line) are shown along with experimental data (open circles). The results where
obtained for N = 20, = 1 = 0.59 and = 1.8 10
5
kg m
2
s
1
.
0 2 3 4
0
0.2
0.4
0.6
0.8
1
7 6 5 1
Fig. 3. The ight time for a granular layer of N = 33 as a function of . Flight times are normalized
to the driving period T. The horizontal solid line marks = T.
The calculated ight times are represented in gure 2 along with the experimental measure-
ments. It is clear that the Gutman model correctly reproduces the ight times of the granular
layer as a whole up to 2.3. The disagreement between the model and the experiment
for greater driving accelerations is not due to the presence of air; the nite duration of the
collision (assumption (2)) is likely to be involved [15].
336 The European Physical Journal Special Topics
d
i
s
t
a
n
c
e
(
m
m
)
time(ms)
0 50 100 150 200
1.6
1.7
1.8
1.9
2
2.1
2.2
Fig. 4. Position of a single bead tracked near the lateral wall ( = 2.6 and N = 100). The dashed line
is the slow drift that makes up the convective motion (the line itself is not a t but just a guide for
the eye). Vertical distances are positive downward; their origin is arbitrary.
If is increased further, the ight time reaches the value of the driving period T. A period
doubling bifurcation takes place, because the granular layer performs long ights (longer than
T) alternated with short ights (shorter than T). This is displayed in gure 3. Thus, every 2T,
the grains in the layer will perform one short ight and one long ight (apart from a certain
time that they will spent stuck to the vibrating plate).
4 Patterns
The dynamics of the whole granular layer, as described in the previous section, can be under-
stood at least for moderate values of the acceleration in terms of a porous piston which
collides inelastically with the vibrating plate at each cycle. Nevertheless, the movement of
the granular layer as a whole does not provide the complete picture. In fact, inside the layer
the grains move dierently from one place to another, and this dierential motion establishes
the convective patterns. Near the onset (for 1) the velocity eld takes the form of a toroidal
roll, as depicted in [15], gure 5. This dierential motion can only be understood if there are
forces acting only at certain zones of the granular layer, aecting exclusively some of the grains.
It is therefore reasonable to look at the grains near the wall, because lateral walls and their
possible eects are not taken into account in the model used in the previous section, and any
eect they may produce will be local, so it might cause the granular convection.
In order to investigate the inuence of the walls, we have tracked single grains near the lateral
wall. One of the obtained trajectories is depicted in gure 4. At each cycle, the individual grain
performs a ight. It is quite easy to spot the landings, because the velocity changes abruptly,
but the take o cannot be pinpointed. The initial velocity is determined by the instantaneous
speed of the plate at the moment of the take o, and the downwards acceleration must be
caused by gravity and whatever other forces (friction, etc.) are acting on it. These periodic
jumps have the same frequency than the forcing, at least near the onset (where 1). This
is what can be called the fast motion.
But there is a conspicuous feature of the motion of a bead near the lateral wall: it descends
along the wall. As can be noticed in gure 4, each landing position is lower than the previous
New Trends, Dynamics and Scales in Pattern Formation 337
t(s)
z
(
m
m
)
0 3 4
-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
8 7 6 5 2 1
Fig. 5. The positions of the grains near the lateral wall, as obtained from particle tracking (dots). All
the data have been repositioned to an arbitrary initial position, keeping the phase information. The
solid line is the position of the vibrating plate. The results where obtained for = 1.9 and N = 20.
Also shown is the trajectory for mass center as obtained from the Gutman model (dashed line).
one. This means that another force, apart from gravity, must be acting upon the bead so as
to produce a net drift downwards. This drift is shown as a dashed line in the plot. We can
therefore distinguish between a fast motion, which are the jumps at the forcing frequency, and
a slow motion, that is the downward drift. This slow downward drift of the grains, that takes
place near the walls, is at the origin of the convective motion. Grains descend along the lateral
wall, travel radially towards the axis at the bottom plate, rise near the center of the layer and
move outwards at the upper surface, closing the ow.
In order to quantify the eect of the lateral wall, the motion of individual grains adjacent
to the container wall is to be compared to the motion of the granular layer as a whole. This
shown in gure 5 for = 1.9, where the positions of large number of grains, taken to a
common reference, are plotted with small dots. Along with it, the trajectory of the mass center
is displayed as obtained from the porous piston model, which describes it quite correctly for
= 1.9, as can be seen in gure 2.
The most obvious peculiarity of these measurements is the dierence between the trajectories
of the grains near the wall and the motion of the mass center of the granular layer. This relative
motion causes the convection. In fact, if we take the dierence between the landing point of
the mass center and the landing point of the grains near the wall (h in gure 5) and divide
it by T, we get roughly the convection velocity (albeit with a large uncertainty; we never used
this method to obtain the velocities). This means that another force, apart from gravity and
the eect due to the interstitial air, is acting on the grains near the wall so as to produce a
net downward motion at each cycle. The quantication of this force by means of an eective
acceleration will be presented elsewhere [15].
We have also changed the properties of the lateral wall in order to investigate their inuence
on the convection. This was accomplished, for example, by gluing sandpaper (with a grain of
150 microns and a grain of 30 microns) to the wall, or by using a cell made of steel into which
338 The European Physical Journal Special Topics
Table 1. Reference velocities for dierent wall materials, for = 2.
Material Velocity (mm/s)
glass 1.2
smooth steel 1.1
grooved steel 0.4
sandpaper, 30 microns 1.2
sandpaper, 100 microns 0.4
2 4 5 6
0.7
0.8
0.9
1
1.1
1.2
9 8 7 3
Fig. 6. Flight times for the granular layer as a whole near the bifurcation point for lateral walls having
dierent properties (each symbol corresponds to one of them). The ight times are normalized to the
forcing period T.
grooves were carved. For each one we measured the velocity of the beads at the same point on
the upper surface. This value is meaningful if we use it just to compare the dierent cases. The
values obtained are shown in table 1. The value reported is the mean velocity (i.e. averaged
over many cycles) at a point on the upper surface of the granular layer 1 cm apart from the
lateral wall (therefore these velocities have a radial direction).
From these results it stems that smooth walls are associated with higher convection veloci-
ties. It is probable that in changing the lateral wall we are mainly modifying the static friction
coecient, and the eect of this would be to reduce the ight time of the grains near the wall,
which in turn would switch o the driving for their relative movement by keeping the grains
stuck to the wall. Note that if grains take o at a later phase then the initial velocity for the
ight would be smaller, and the ight shorter. In this regard, taps and a sinusoidal vibration
may well reveal to produce dierent results [12], although this aspect should be studied in
more depth.
It is important to note that the velocity proles have the same shape (the prole along
the vertical wall is shown in [14] for glass), and they only change by a scale factor. Moreover,
the ight times measured from the accelerometer signal (corresponding to the mass center) is
always the same before the bifurcation point (see gure 6). This means that the wall properties
do not aect the ight time and the dynamics of the granular layer as a whole, but they do
aect the convection velocity.
New Trends, Dynamics and Scales in Pattern Formation 339
Fig. 7. The vibrated granular layer under lateral lighting shows dierent regions (two in the one at the
left, three in the one at the right) that are performing short and long ights out of step with respect
to the adjacent one. The regions are visible because a small heap is formed in each of them. Note that
the frontiers between them nish perpendicularly to the wall.
After the bifurcation point at = T new patterns can appear. The grains must perform
one short ight and one long ight in a time span of 2T, but the granular layer can do this in
two ways: a short ight followed by a long ight, or a long ight followed by a short ight. The
possibility that one zone of the granular layer performs one kind of ight while other zone is
performing the other is at the origin of the patterns sometimes seen in convecting granular layers
above the value of corresponding to the period doubling bifurcation for the ight times. Two
examples of such patterns are shown in gure 7. The photographs show the upper surface of
the granular layer under lateral lighting. Distinct regions can be seen because the surface
is not at: small heaps form over each zone. An interesting feature that remains unexplained is
that the borders between these regions always try to be perpendicular to the lateral wall, like
in thermal convection (see also [14], gure 2).
5 Conclusions
The experiments reported here allowed us to investigate the factors that inuence the motion
of a vibrated granular layer. For the granular layer as a whole, a free ight under the action of
gravity alone that ends with an inelastic collision will not reproduce the experimental results.
Nevertheless, taking into account the eect of interstitial air the experimental results are faith-
fully reproduced for moderate values of the dimensionless acceleration ( 2.3). For beyond
that value, the model fails to reproduce the experimental results. The disagreement is not
caused by the presence of air, and more work is needed to understand the phenomenon.
It has been shown that near the lateral wall the grains suer a downward drift which is
caused by a local force acting on them. This drift is at the origin of granular convection.
By changing the properties of the wall we have shown that the dynamics of the layer as a
whole is not changed (the ight times are the same and the bifurcation point does not change
signicantly), but the convection velocity is greatly inuenced. This means that the extra
force responsible for the dierential movement that the grains carry out is a local eect of
the walls.
This work has been funded by Spanish Government project FIS 2005-03881 and by the Universidad de
Navarra (PIUNA). J.M.P. and R.A. thank the Asociaci on de Amigos de la Universidad de Navarra for
a fellowship. We are also grateful to I. Zuriguel for his comments and suggestions.
We are indebted to Carlos Perez Garca in many ways; we would like just to mention his support
when some of us started the line of research concerning granular media in our laboratory.
340 The European Physical Journal Special Topics
References
1. G.H. Ristow, Pattern Formation in Granular Materials (Springer, Berlin, 2000)
2. J. Duran, Sables, poudres et grains (Eyrolles, Paris, 1997)
3. M. Faraday, Philos. Trans. R. Soc. Lond. 52, 299 (1831)
4. S.B. Savage, J. Fluid Mech. 194, 457 (1988)
5. P. Evesque, J. Rajchenbach, Phys. Rev. Lett. 62, 44 (1989)
6. E.L. Koschmieder, Benard Cells and Taylor Vortices (Cambridge University Press, Cambridge,
1993)
7. T. Ondar cuhu, J. Mill an-Rodrguez, H.L. Mancini, A. Garcimartn, C. Perez-Garca, Phys. Rev. E
48, 1051 (1993)
8. F. Melo, P.B. Umbanhowar, H.L. Swinney, Phys. Rev. Lett. 75, 3838 (1995)
9. C. Laroche, S. Douady, S. Fauve, J. Phys. (France) 50, 699 (1989)
10. E. Clement, J. Duran, J. Rajchenbach, Phys. Rev. Lett. 69, 1189 (1992)
11. H.K. Pak, E. Van Doorn, R.P. Behringer, Phys. Rev. Lett. 74, 4643 (1995)
12. J.B. Knight, E.E. Ehrichs, V.Y. Kuperman, J.K. Flint, H.M. Jaeger, S.R. Nagel, Phys. Rev. E 54,
5726 (1996)
13. R. Ramrez, D. Risso, P. Cordero, Phys. Rev. Lett. 85, 1230 (2000)
14. A. Garcimartn, D. Maza, J.L. Ilquimiche, I. Zuriguel, Phys. Rev. E 65, 031303-1-5 (2002)
15. J.M. Pastor, D. Maza, I. Zuriguel, J.-F. Boudet, A. Garcimartn, Physica D (submitted)
16. W. Kroll, Forschung auf der Gebiete des Ingenieurwesen 20, 2 (1954)
17. R.G. Gutman, Trans. Instn. Chem. Engrs. 54, 174 (1976)
18. M.E. M obius, X. Cheng, P. Eshuis, G.S. Karczmar, S.R. Nagel, H.M. Jaeger, Phys. Rev. E 72,
011304-1-13 (2005)
19. R.M. Nedderman, Statics and Kinematics of Granular Materials (Cambridge University Press,
Cambridge, 1992)
20. P.C. Carman, Flow of Gases Through Porous Media (Butterworth Scientic, London, 1956)