Beruflich Dokumente
Kultur Dokumente
Editorial Board:
E.Guyon Lui Lam Dominique Langevin H.E. Stanley
Ecole Nonnale San Jose State University Laboratoire de Physique ENS Boston University
Advisory Board:
J. Charvolin W. Helfrich Patrick A. Lee
Institut Laue-Langevin Institut fUr Theorie der Massachusetts Institute
Kondensierten Materie of Technology
Springer
Jacques Duran Translated by
Directeur de Recherche Axel Reisinger
CNRS Lockheed Martin
University of Paris VI Lexington, MA
France USA
Editorial Board:
E. Guyon Lui Lam Dominique Langevin H.E. Stanley
Ecole Normale Department of Physics Laboratoire de Physique ENS Center for Polymer
45 RueD'Ulm San Jose State University 24 Rue Lhomond Studies
F-75005 Paris One Washington Square F-75231 Paris, Cedex 05 Physics Department
France San Jose, CA 95192 France Boston University
USA Boston. MA 02215
USA
Translated from the French Introduction it la physique des materiaux granulaires, © 1997
by Editions Eyrolles, Paris, France.
987 6 5 4 3 2 I
ISBN 978-1-4612-6790-4
Foreword
a funnel, a few hoses, and a glass container. As it turns out, a thousand technical
details can upset an experiment and lead to nonsensical results. The physics of
granulars may not require costly hardware, but it demands enormous care. Nor is
it exempt of real dangers: Industrial reactors can generate violent explosions, even
when they are ostensibly empty.
I am thrilled to see French physics actively engaged in the field of grains and
powders, where almost everything is yet to be discovered. Within the last decade,
a number of highly creative teams have popped up in places like Paris, Lyon, and
Rennes. Jacques Duran leads one of those teams at the Pierre and Marie Curie
University in Paris. This book draws on his first-rate expertise. I personally had
the opportunity to read various drafts as it was taking shape. I learned a great deal
from them in the process. There was a time when we had to steer anyone interested
in sand to the classic text by Bagnold. From now on, Duran will be high on the
required reading list. This book will fill a very real need. I sincerely wish it great
success.
Pierre-Gilles de Gennes
Paris, France
Preface
up concepts and theories about a medium whose behavior is most often surprising
and counterintuitive. My goal has been to gather, in as self-contained a manner
as possible and in a unified language, the background necessary to understand the
latest developments, as well as to present the rudiments of granular physics.
Given the current status of this young discipline, any progress is likely to come
from experiments. Henri Poincare once said: "It has been quite some time since
anyone has thought of getting ahead of experimentation or to construct the world
on the basis of a few hasty hypotheses." The remark is particularly relevant to the
physics of granular media. This book is rooted in that very principle. In this spirit, it
contains many descriptions of experimental devices and accounts of experimental
results obtained under conditions that are as controlled as possible. These provide
the backdrop for concepts, models, and ideas that are pushed to levels that some
might find too speculative. The adopted strategy, whose limitations must be borne
in mind, goes with the very nature of this discipline, which is still very much in its
infancy. In this context, this book discusses concepts and results as they are known
to us at the present moment. As such, it is only a snapshot, as our understanding
of this topic is bound to evolve, perhaps even undergo major revisions.
Lest it become a dry compilation of facts, writing a book of this kind entails
somewhat arbitrary choices that remain the sole responsibility of the author. On
more than one occasion, I felt compelled to leave out this particular result or that,
even though I may come to regret it in the future. I did so simply because it did
not readily fit in with the flow of the presentation or because it would have taken
me too far afield. I beg for those whose work is mentioned too cursorily or not at
all to forgive me. This by no means implies a negative judgment on my part. My
choice was guided exclusively by pedagogical concerns and my own concept of
the logical flow of the material. My decisions were often agonizingly difficult.
My thanks go to my colleagues and friends in my own laboratory (Eric Clement,
Jean Rajchenbach, and Touria Mazozi). I am also grateful to the members of the
Physics of Heterogeneous and Complex Matter Research Group affiliated with
the National Center of Scientific Research (CNRS) and with the European HCM
(Human Capital and Mobility) Network. I am enormously indebted to them not
only for much of the technical material in the book but also for infusing in it
their irrepressible enthusiasm which I hope will brighten the experience of the
reader. I acknowledge the careful reading of the french manuscript and comments
by Ko Okumura and VJ.-P. Faroux. Axel Reiseinger successfully carried out the
translation from the French edition. He kept within the spirit of the text while
suggesting numerous improvements and clarifications.
Finally, I owe a large debt of gratitude to Pierre-Gilles de Gennes and Etienne
Guyon, both of whom played an instrumental role in the early stages of the research
discussed here. I was fortunate to have their unrelenting support and encouragement
to write and publish the material in this book. It borrows a great deal from their
own work.
Jacques Duran
Paris, France
Contents
Foreword v
Preface vii
1. Introduction 1
1.1 Some Orders of Magnitude Defining the Problem 1
1.2 Economic Implications and Industrial Problems 3
1.2.1 Industrial Processing of Granulars . 4
Construction Materials . . . . . . . . . . 4
Processing Industries . . . . . . . . . . . 6
An Example: Casting by Sacrificial Polystyrene 7
The Agriculture Industry 10
1.2.2 Flow Problems . 10
1.2.3 Problems of Segregation . . 12
1.3 Granular Materials and Geophysics 14
1.4 A Brief Historical Review . . . . . 16
1.5 Prerequisites and Selected Bibliography . 18
Bibliography 209
Index 213
1
Introduction
lOne such case is the effect of a uniform vertical air flow, which keeps particles in suspension and
imparts to them random movements mimicking a classical Brownian motion [3]. It is also possible to
model the system by way of a thermal agitation of sorts when the particles undergo repeated mutual
collisions, as in a rapid flow [4].
1.2 Economic Implications and Industrial Problems 3
The physical laws governing the behavior of granular media actually apply to
objects whose dimensions cover several orders of magnitude. From grains, a few
hundred microns each, to ice floes drifting across the polar seas (over distances of
1000 km),2 not forgetting Saturn's rings (made of icy particles about 1 cm wide
distributed in a band roughly 1 km thick), the science of granulars covers at least
twelve decades of sizes.
The fact that aggregates seem to obey universal laws applicable over such a wide
range of dimensions and characteristics is a strong incentive to pursue fundamen-
tal studies in that area. For instance, phenomena of segregation and intermittent
blockages are pervasive in numerous industrial processes involving granular ma-
terials. Accordingly, the remainder of this chapter will be devoted to a brief and
selective review of a few techniques and processes that in one way or another are
subject to phenomena such as convection, segregation, blockages by arching, all
of which are routinely encountered in industry. These effects will be examined in
greater depth in subsequent chapters.
2 Studies of the movements of icebergs, notably in the vicinity of port facilities, has been the object of
several research contracts with the Canadian Navy. It turns out to be closely related to the physics of
granular materials.
3 A recent report by
a US government agency stresses the obsolescence of the techniques used to process
granular materials. Significantly, it is entitled "Granular Materials: A Legacy of Neglect."
4 1. Introduction
and pharmaceutical industries, specialized chemistry, and the food industry, which
demand increasingly sophisticated processing technologies.
4The numbers quoted here come from data published by Lafarge-Coppee, a world-wide leader in the
manufacture of all types of construction materials.
5Interestingly, the official list of parameters characterizing industrial aggregates does not include even
one of the micromechanical parameters considered of fundamental importance by physicists. In-
dustrial engineers do not care about things like the coefficient of elastic restitution of the coefficient
of friction. These omissions are ample evidence that practical problems in the real world often de-
mand a more complex, albeit more empirical, set of parameters than seem necessary in the idealized
atmosphere of a laboratory.
6This touches on a fundamental aspect of the scientific method, which can at times seem too reductive.
Physicists have to choose objects of study that are simplified enough to be amenable to analysis, yet not
so simplified as to lose any relevance to the real process they are intended to simulate, which is invari-
ably far more complex. We will confront several such choices during the course of this book. Making
the COlTect decision is a difficult task which, for the most part, rests on no more than sound intuition.
1.2 Economic Implications and Industrial Problems 5
FIGURE 1. Around 200 B.C., Appolonius of Perga invented the arrangement shown here
to tile a surface as fully as possible with circles of various sizes. This object enjoys the
property of self-similarity.
7Very much to the point, there exists an Arab saying that can be loosely translated as: "You think your
basket is full when you have stuffed it with oranges. In truth, it is full of void, for you can still add
nuts, and even chickpeas after thal."
8The quality of a high-strength concrete is often expressed as the height of a column with a one-
squaremeter base that withstands collapse. This maximum height has increased by more than one
order of magnitude in the last few years.
6 1. Introduction
concretes and aggregates have proven remarkably adept at building on their rich
experience and mastering the difficult art of devising fluid binders in which each
grain can find its ideal spot. No one can deny that systematic experimentation has
yielded the right answers, even though manufacturing costs are so high that appli-
cations are for the time being restricted to the construction of highly specialized
buildings. It is reasonable to expect that the possible contributions of physics to a
better understanding of the properties of construction materials may substantially
lower the costs of producing, conveying, and mixing them. These processes are
intimately related to problems of flow and segregation, which are to be taken up
in Sections 1.2.2 and 1.2.3.
We will refrain from treating here the principles of soil mechanics, which fall
largely in the domain of geophysics and geology, or the kinds of problems civil
engineers have long been accustomed to dealing with. There exist in this area
numerous models painstakingly put together, and with a respectable track record.
For instance, most soil-mechanics engineers tend to view granular materials as
continuous objects that obey the usual laws of classical mechanics (including
friction, stress-strain relations, plastic deformation, and the like), such as they
apply to homogeneous solids. As we will see, there are cases when this approach
raises serious questions. From a practical point of view, however, it does offer
the advantage of permitting simple calculations to predict the behavior of various
structures made of granular materials, such as roadways, with generally adequate
safety margins.
There is, however, one related phenomenon that deserves to be singled out,
because it shows up routinely in everyday life. We are referring to segregation
by shearing, an effect that is characteristic of most granular materials. Figure 2
describes the phenomenon in question, which civil engineers are quite famil-
iar with. They have to contend with it as they plan and manage construction
projects.
Processing Industries
Generally speaking, modern manufacturing methods of high-technology materials
rely heavily on the processing of granular materials. We are not about to review
each of the many applications, transformations, and various processes involving
granulars at some stage of the technology. One book would not be enough to
cover them all. We will simply give a few examples selected for the purpose of
highlighting the many-faceted problems faced by industry, and their connections
with the fundamental approach advocated in this book.
We start by making a clear distinction between two separate branches of the
chemical industry. On the one hand are industries that typically deal with large
quantities of materials in granular or powder form. Their problems are generally
similar to the ones we have already mentioned, such as transport through pipes,
storage, or granular segregation. On the other hand are the chemical or pharma-
ceutical industries where materials are prepared in relatively small quantities, but
with high-value-added materials. In this case, the demands in terms of purity and
reproducibility are obviously far more stringent. Accordingly, their manufacturing
1.2 Economic Implications and Industrial Problems 7
FIGURE 2. Grading the embankment on the side of a road with a bulldozer causes segre-
gation by shearing. The concentration of large rocks is much greater on the suti'ace of the
embankment than in the interior.
Polystyrene model
Vibration
In principle, the technique ought to produce nearly perfect results. One of its
advantages is that it eliminates the burrs formed at the joints between separate sec-
tions of conventional molds. Unfortunately, during the shaking process, granular
convection gives rise to a powerful central flux of sand which has a disturbing
tendency to deform-and sometimes to destroy outright-the polystyrene model.
This is a very serious problem that affects about 90% of the products, causing a
flood of rejects.
This problem is quite typical in the physics of granular materials. The solution
is relatively straightforward: Convection can be largely eliminated by using nearly
frictionless containers and granulars. But this ignores the fact that convection also
plays a useful role, as it promotes the filling of every nook and cranny around
the model. The trick is then to preserve the useful part of the convection while
minimizing or inhibiting the undesirable more violent component. Solutions do
exist, but they involve processes and materials so costly as to preclude their use
for mass production. A more pragmatic approach consists in finding the best
compromise between the dictates of physics and manufacturing costs.
This example is a classic illustration of the delicate balance that must be struck
between fundamental advances in physics and industrial requirements. It is often
not enough to propose clever solutions. We also have to contend with cost con-
straints. In the case of low-value-added granulars, technological advances must
be quite substantial before industry will commit itself to investing in an expensive
new process.
Problems are particularly varied in the light chemical synthesis and pharmaceu-
tical industries. We will mention a few typical examples.
1.2 Economic Implications and Industrial Problems 9
(b) D
(e) f\
FIGURE 4. Diagrams (a) and (b) show flow blockages due to the formation of arches near
the orifice of an hourglass or a hopper. Diagram (c) depicts an arch structure with maximum
stability (inverted chain).
in the vicinity of the discharge orifice. Depending on the circumstances, the flow
can either become sporadic or be completely blocked by remarkably stable arches.
For the same reason, it is extremely difficult to reproducibly mix two granular
materials by merging two separate flows into a common discharge channel. That
has proven to be a major problem in the food industry and in chemical plants
producing polymers, where raw materials are typically dry granulars. Blockages
can form spontaneously as a result of some kind of seizing mechanism. 9 The
phenomenon is even known to have caused dangerous explosions by allowing
fermentation gases to accumulate in grain silos plugged up by such arches.
Given the many uncertainties surrounding the static properties of a granular
system, it would seem that describing the mechanism of arch formation would be
quite a daunting problem. Nevertheless, it is possible to state a number of gen-
eral principles which will later prove useful. It is important to realize that arches,
which are a direct consequence of gravitational forces, can form because of purely
geometrical and steric reasons. This can be easily understood by considering a
stack of identical frictionless spheres resting on either side against inclined walls,
as depicted in Figure 4(b). In this example, each sphere is kept in equilibrium at
two contact points whose connecting line lies below the center of gravity. While
such arrangements can obviously form spontaneously in two-dimensional systems,
they would seem rather more improbable in real-life three-dimensional situations,
because they require the cooperation of a much larger number of spheres, each sta-
bilizing its neighbors. Nevertheless, they are frequently encountered in industrial
or laboratory conditions. Figure 5 illustrates some examples actually observed in
simple experiments.
Not surprisingly, the phenomenon of dry friction greatly increases the likelihood
of forming arches anchored onto the side walls. In this case, it is not even necessary
9The phenomenon of seizing is frequently observed in the industry, although it is not clearly understood.
It manifests itself in a gradual hardening of arches forming in hoppers. It is not uncommon for workers
in the agriculture industry to have to use pick axes to remove solid plugs in silos that have remained
idle for some time. These plugs can be the result of high humidity (see Section 2.1) or of poorly
known physical and chemical processes causing grains to stick together during storage.
12 1. Introduction
FIGURE 5. Stable arches form readily when a 1 em diameter test tube filled with fine sand
(100 !Lm) is turned upside down. The photo on the left provides a side view, while the one
on the right is taken from below to reveal the arch pinned against the wall.
for the side walls to be slanted. Indeed, we will run into several examples of arches
pinned against vertical walls (e.g., Section 3.2.4), particularly in silos. Effects of
this type can have dangerous consequences in industrial facilities. Engineers have
come up with a number of more or less effective solutions, such as those illustrated
in Figure 6.
FIGURE 6. Three methods used in industry to prevent or remedy blockages by arch effect.
They include (a) an Archimedes screw, and (b) a conveyor belt with a corrugated surface.
In (c), a plant worker pounds an obstructed hopper with a sledge hammer to get the flow
started again. The latter is the method of choice in industries producing low-value-added
granular materials.
1.2 Economic Implications and Industrial Problems 13
Mixture
FIGURE 7. Two techniques used in industry to try to avoid problems of granular segrega-
tion. The principle is the same in both cases. The idea is to force the materials through a
section of hardware that promotes mixing.
(a) 1
B
(b) 2 <~~
B
(e) 3 A
B
pieces are placed on top of each other (Figure S(c», resulting in an AlB I AI B
structure. The process is repeated until the thickness of each individual layer is
roughly equal to that of the largest of the two particles A or B. At the conclusion
of N operations, the number of layers is obviously 2N . Not only is the process
extremely labor-intensive, but the final product is only a superposition of different
layers, rather than a true three-dimensional mixture.
The process just described has been adapted to mass-production with so-called
Kenics mixers, illustrated in Figure 9, which accomplish, automatically and con-
tinuously, the discrete sequence of steps of the pastry-maker technique. 10
The pastes or fluids to be mixed are injected at points A and B shown in the figure.
The ultimate speed of the process is limited by the onset of turbulence phenomena
which may appear within the reactor and can potentially upset the quality and
smoothness of the mixing. Such phenomena depend primarily on the dimensions
of the apparatus and the properties of the products to be mixed.
It is clear that implementing this technology on the factory floor requires a
large capital investment and has to contend with difficult processing techniques
(the preparation of pastes can itself be a challenge). As such, it is largely incom-
patible with the processing of industrial quantities of low-value-added materials,
whose cost has to be kept as low as possible. A better understanding of the mech-
anism of segregation in granular materials is of great importance not only from a
fundamental physics point of view but for economic reasons as well.
lOMixers working on this same principle manufacture the complex powder mixtures packed in rocket
boosters used to launch satellites into orbit.
11 Thestudy of deserts and desertification is a discipline with considerable impact on human affairs. It
is significant that Bagnold, one of the pioneers in the physics of granular materials, has devoted an
entire book to the topic of desert dunes [9].
1.3 Granular Materials and Geophysics 15
B
FIGURE 9. The first two elements of a Kenics mixer (after [115]).
12Guy Berthault has made an excellent documentary movie about this phenomenon, based on a study
conducted at the University of Colorado [10].
16 1. Introduction
Another discipline that, in many respects, has much in common with the physics
of dry granular materials is seismology. How forces of friction are stored and sud-
denly released is just as much of interest to geophysicists as it is to fundamental
theorists trying to understand the basic properties of granulars. The collective
behavior of particles in quasi-permanent contact is the central focus of the geo-
physical sciences. Granted, geophysicists typically deal with interactions that are
far more complex than those to which we will deliberately restrict our attention.
They have to worry about phenomena such as cohesion effects, contact fatigue,
erosion, strain hardening, and many others that we shall barely touch upon. Yet,
the phenomenon of fracturing, which is at the basis of geophysics, shows up just
as faithfully, albeit in an elementary form, in the most grossly simplified versions
of the physics of granulars, as we will see in Chapter 3.
Avalanches and free flows are treated in Chapter 4, in which we draw liberally
from the work of geophysicists. We do so for at least two reasons. First, an
accurate description of what is called the stick-slip phenomenon requires a clear
understanding of the interdependence between friction and velocity, which also
happens to be at the heart of the physics of granular materials. Second, the simu-
lation of so-called self-organized critical (SOC) systems has been called upon to
explain both granular avalanches and the scaling laws of the distribution of earth-
quakes (typified by the Gutenberg-Richter law). The validity of this approach
remains a topic of lively debate in both cases.
13The poppy seed has become popular again of late as the material of choice to study the properties
of granulars. Since it contains water, it can be imaged by nuclear magnetic resonance (NMR) to
generate three-dimensional pictures of the collective behavior of a collection of grains, as we will
explain in Chapter 3.
1.4 A Brief Historical Review 17
140ur great ancestors in science deserve much admiration for their inquisitiveness. Their curiosity was
rarely restricted to a single discipline.
15This problem remains quite current. Given the various uncertainties concerning the forces at play in
a granular medium-which we will study in some detail-the correct way to model the distribution
of forces on the base of a silo remains in dispute.
18 1. Introduction
that such idealized objects might be fairly accurately approximated in the labora-
tory by placing solid particles in a vacuum chamber. In practice though, things are
rarely that simple because of a major intrinsic problem. As it turns out, the energy
dissipated through mutual collisions or friction is inevitably accompanied by the
appearance of surface electrical charges on both the particles themselves and the
walls of the container. Such surface charges are particularly difficult to eliminate
in vacuum; the resulting medium-range electrostatic interactions can greatly com-
plicate the analysis of experimental results. Adding a surrounding fluid makes it
much easier to drain away those pesky surface charges. For that reason, such a
fluid is not only desirable, but indeed essential. Fortunately, provided a few pre-
cautions are observed, a fluid matrix does not necessarily preclude treating many
laboratory, or even industrial, situations involving granulars as though the mate-
rials were truly dry. The purpose of this section is to define the range of validity
of the methodologies which we will develop throughout this book. For strictly
pedagogical reasons, we start out by discussing the interactions of a single particle
with its fluid environment.
Laminar Drag
Consider a spherical particle of radius R, mass m, and volumetric density Pb,
moving at a velocity v in a fluid of viscosity rJ. The surrounding fluid is contained
within walls close to the central sphere, as shown in Figure lO(a). We wish to
compare the kinetic energy of the particle with the energy loss due to the work
done by the viscous forces existing within the fluid-forces which tend to slow
down any movement.
To that end, we define a parameter R1 (the subscript l stands for laminar), which
is simply a measure of the ratio of the kinetic energy of the particle and the work
done by the viscous force over a characteristic length, which we take to be the
radius R of the sphere. This parameter will enable us to estimate the importance
of viscous dissipation in relation to the kinetic energy. Based on what every student
Laminar Turbulent
FIGURE 10. A sphere moving in a fluid in a laminar or turbulent regime.
2.1 A Single Particle and Its Environment 21
1Some recent experiments, whose significance will become more transparent after reading Chapter 3,
have clarified some of the concepts introduced here [31].
22 2. Interactions in Granular Media
Before leaving this subject, the reader will appreciate that the situation is likely
to be problematic in a fluid such as water, which has a coefficient of viscosity a
hundred times larger than air. Even with particles as small as I mm in diameter,
viscous interactions cannot be neglected in this case.
Turbulent Drag
Hydrodynamics taught us that, as soon as the relative speed of a particle in motion
in a fluid matrix becomes sufficiently high, instabilities appear in the boundary
layer between the fluid and the object in motion (Figure lOb). There is at that
moment a transition from a laminar flow to one that is turbulent. We also know
that the transition between the two regimes occurs for relative speeds of the order
of a few cmfs in dry air. As it happens, most granular flows involve velocities that
are precisely in that range. We can thus anticipate that, depending on the details of
the circumstances, energy dissipation will be either laminar (directly linked to the
viscosity of the fluid) or turbulent (related to the difference in dynamic pressure
ahead of and behind the moving object).
Here again, we define a number R t (where the subscript t now stands for tur-
bulent) as the ratio of the kinetic energy to the work done by the opposing force
of turbulence F t over the applicable characteristic length, which, as before, is
typically of the order of the radius R of the particle.
This time around, though, the force Ft is proportional to the frontal cross-
sectional area S of the particle, to the square of the relative speed v, and to a
coefficient kt whose value is typically 0.24 for a sphere. Specifically, we write
2
Pov
Ft =kt - -S,
2
where Po is the density of the fluid. We then find that the number R t is approxi-
mately equal to
1 Pb
Rt ;::::::--.
kt Po
This equation tells us that the number R t does not depend on the velocity (assuming
that the regime is indeed turbulent). It also shows that the concept of dry granular,
i.e., of a system of particles whose interactions with the surrounding fluid are
negligible, is valid for sufficiently heavy particles such as grains of silicate sand
moving through a gas such as air. 2 In this case the number R t is greater than
2The cmcial importance of the density of the medium in the physics of granulars was vividly illustrated
by a simple observation due to Chladni in the eighteenth century. He had noticed that small pieces of
horse hair torn loose from a bow mbbing against the strings of a violin tended to collect on the wooden
body of the instmment according to specific patterns. He proceeded to carry out a series of simple
experiments by depositing on a violin either fine sand or lycopodium powder. The sand piled up at
the nodes of vibration, while the lycopodium powder gathered at the vibration peaks, giving rise to
complementary figures (known as Chladni's figures). The explanation is that the very light lycopodium
powder responds to the vibrations of the surrounding air, while the sand is sensitive primarily to the
vibrations of the instmment itself. This, incidentally, constitutes a magnificent example of a physicist's
sense of observation, as reported by Michael Faraday in 1830.
2.1 A Single Particle and Its Environment 23
Granular Dendrites
We proceed next to discuss a rather spectacular and poorly understood phe-
nomenon, which is due to an intriguing type of interaction between air and granular
matter [26], [32]. Observed in granular dendrite experiments, it involves effects
that have well-known counterparts in hydrodynamics when one fluid is injected
into another more viscous one. Even the experimental configuration tradition-
ally used with liquids (Hele-Shaw radial cell) is essentially duplicated to study
granulars.
The actual experiment goes as follows: A layer of fairly loose fine sand is placed
between two glass plates. The upper plate has a hole in its center through which a
hose is inserted to blow pressurized air into the granular material. When both the
pressure and the flow are low, nothing happens; the sand behaves as an ordinary
porous medium. But as the flow is increased, the grains of sands start moving
around and a dendritic pattern develops with a fractal structure reminiscent of
what is observed with non-Newtonian fluids. An example is shown in Figure 11.
The outcome of this experiment remains something of a mystery. Many ques-
tions come to mind. How can the sand grains move at all when they are confined
in such a tight space between the glass plates, particularly since they must first
expand before they can deform, as we will learn shortly? In other words, is the
stack compact in the sense of the principle of dilatancy (see Section 3.1.3)? What
determines the size of the dendrites? Can the granular medium be considered a
viscous fluid?
FIGURE 11. Growth of a dendritic pattern when air is injected into a confined granular
medium. The photograph on the left shows the beginning of the process. The one on the
right shows a well-developed pattern roughly 10 cm in diameter (after [26]).
24 2. Interactions in Granular Media
Fc = nYzvrz (1 + ~~),
where YZv is the surface tension at the air-liquid interface. We assume that, when
the weight of the lower spherical particle is just equal to the capillary force, the
radius of the liquid is some fraction a of the radius of the particles, or r2 ~ aR,
and the ratio r2f rl is of the order of 5. This series of approximations leads to a
result, correct within an order of magnitude, given by
R~ j4agPbYtv .
When applied to glass beads (Pb = 2.2g/cm3 ) and water as the liquid (Ytv
73 x 10- 3 N/m), the above expression suggests that the beads will remain stuck
for diameters up to 1 mm when a = 1. With a more modest amount of water
(a = 0.01), the diameter of the particles should not exceed 100 {tm. This exercise
highlights the need to control humidity levels when doing experiments with gran-
ular materials assumed to be dry. We may also conclude that industrial processes
dealing with powders (particles smaller than 100 {tm) in open air are very likely
not to conform to the type of physics presented in this book.
It is less straightforward to incorporate electrostatic interactions in a simple
model, primarily because of the difficulty in quantifying even approximately the
electric charges that develop on the surface of granulars in relative motion. It is
somewhat easier to estimate them empirically on the basis of simple experiments,
such as the following. Drop a few steel balls, 1.5 mm in diameter, into a small
plastic tube (Figure 13). Shake the tube and its contents vigorously and place it
back vertically on a table. You will then notice that a certain number of balls remain
stuck to the walls of the tube. The position of each of these balls in precarious
equilibrium obviously depends on the position of its neighbors, since they all
interact electrostatically.
Knowing the weight of each steel sphere, and assuming that it is exactly offset
by electrostatic repulsion, it is easy to calculate that each one carries an electric
charge of the order of 3 x 10-9 C (Coulomb), which corresponds approximately
to 300 {t C/kg, or a surface charge density of 4 x 10- 8 CJcm2 . These numbers
seem to agree with those usually observed in industry. Experience shows that
organic materials are much less sensitive than minerals (by about a factor of 100),
and that the amount of accumulated charge depends on the physical and chemical
nature of the surfaces. Furthermore, as mentioned earlier, humidity minimizes this
type of perturbation, although the actual mechanism is not entirely clear. Finally,
the existence of these surface charges constitutes a danger in industrial facilities,
which tend to store large quantities of dry grains (such as corn). The reason is that
they can trigger explosions of gases generated in silos by the decomposition of
organic substances. A standard way to mitigate the risk in laboratories is to coat
sensitive surfaces with antistatic sprays. This solution is unfortunately completely
impractical on an industrial scale.
Classification of Granular Materials and Definitions
It is useful to familiarize oneself with the terminology traditionally used in the field
granular materials. The following definitions are taken from Brown and Richards,
26 2. Interactions in Granular Media
FIGURE 13. Small steel spheres suspended by electrostatic interactions on the walls of a
plastic tube.
In light of these definitions and of our previous remarks, it should be clear that
the physics of dry granular materials is limited in practice to granular solids and
aggregates. Excluded are particulate systems such as powders, suspensions, or
fluidized beds, in which interactions with the surrounding fluid are predominant.
2.2 Interactions between Two Particles 27
P3
~
...
:
FIGURE 14. Leonardo da Vinci's traction experiments. The force of traction T necessary
to set the disks PI, P2, and P3 is the same in both configurations.
A Microscopic Explanation
Micrographs of the surface of ordinary solids such as metals reveal a ragged topog-
raphy of the type schematically indicated in Figure 15. It is crucial to understand
that Coulomb's second law, which states that the force oftraction required to set a
solid object in motion is independent of the surface area in contact with the base,
is justification enough for studying the physics of friction on the scale of a single
protrusion (since the result does not depend on the number of such features).
These ragged features, whose typical size is of the order of 1 t.tm, are so small
that, even under weak stress conditions, the deformations they experience sub-
stantially exceed the elastic regime in which Hooke's law remains valid (see some
2.2 Interactions between Two Particles 29
E::i..
1:
Ol
'Qj
I
o
o 10
Distance (flm)
FIGURE 15. Schematic magnified view of the surface of a polished metal sample.
orders of magnitude in Section 2.2.2). 3 This implies that the protrusions in contact
deform at constant pressure p until they exactly offset the normal load force N
(normal in the sense of being perpendicular to the average plane of contact, as
shown in Figure 16).
If A is the surface area in contact, then the relation N = pA holds when equi-
librium has been reached. The effect of a sufficiently large tangential force T is
to separate the surfaces in contact. There is a constant s which characterizes the
ability of the material to resist this shearing action; it is given T = sA. In this
picture, the static friction coefficient {Ls defined previously can depend only on the
physical constants of the materials in contact. Its value is
T s
{Ls = - = -
L p
As it happens, the constants sand p vary roughly in the same proportion for many
different surfaces. 4 In particular, metals are characterized by
s
0.6:::: - :::: 1.2.
p
This model accounts for the tightly clustered range of values of {Ls. It also ex-
plains the experimental observation that the contact regions between two solids
deform slowly under the weight of the upper object, which is fully consistent with
the behavior expected in the plastic regime. We shall present in Section 6.4 a
more accurate description of these laws of friction in the context of numerical
simulations.
3 Thisbecomes even more obvious if we appreciate that the real contact surface area between two solids
can be hundred to thousand times smaller than the "apparent" area. When distributed among a few
tiny spots, the stress due to weight can easily be much larger than the elastic limit.
4This statement is not true of materials whose mechanical properties are strongly anisotropic. For
instance, materials with a lamellar structure such as graphite exhibit a strong resistance to deformations
in a direction perpendicular to the stacked layers, but a weak one to shearing in the plane of these
layers. That is precisely the property that is being exploited when mechanical parts are lubricated
with graphite.
30 2. Interactions in Granular Media
where Rand R f are frames of reference attached to solids (S) and (Sf), respectively.
As written, the above equation describes the motion of point M in the frame of
reference of solid (Sf). For the purpose of simplification, we assume the solid (Sf)
to be fixed. At any given instant, a generalized rotation (pivoting or rolling) is
given as a sum of two orthogonal components
W = W n + WI.
The vector WI corresponds to a rotation in the plane of the figure, while Wit describes
spinning about a vertical axis. The dynamic equation written above includes three
5The notion of a rigid solid is purely theoretical. It clearly conflicts with the picture we have painted
of solid friction, which is based on contacts at a number of points undergoing plastic defonnation.
This is only one of several difficult issues raised in making the transition from classical mechanics to
the mechanics of real solids, particularly in granular fonn.
2.2 Interactions between Two Particles 31
(8')
Frustrated rotation
Stick-Slip Motion
We will deal on several occasions (see in particular Sections 3.1.1 and 4.2.2) with a
type of motion frequently observed in situations involving dry friction. The motion
in question is known as stick-slip. It typically results from interactions between
particles that are subject to the Coulomb-Euler laws of friction and also exhibits
an elastic behavior. Given that the granular materials we are specifically interested
in often conform to both requirements simultaneously, it should come as no surprise
that the stick-slip mechanism is commonly observed in granular matter. Although
we will have an opportunity to give a more general treatment of the phenomenon
later (see Section 4.2.2), we nevertheless introduce here a simplistic but instructive
description [35]. It is based on the model illustrated in Figure 19.
An object of mass m is placed on a conveyor belt moving at a speed v. The object
is also tied to a stationary post by a string of stiffness k. The friction between the
object and the belt is characterized by coefficients f-ts (static) and f-td (dynamic),
2.2 Interactions between Two Particles 33
such that j.1,d ::'S j.1,s, in accordance with the Coulomb-Euler laws. For the sake of
simplicity, we assume that j.1,d is negligibly small.
We start at time t = 0 with the spring in a state of rest, at which point its length
is x = xo. Because of dry friction, the object starts moving at a constant speed v.
The horizontal component of the reaction associated with the friction opposes the
traction of the spring, which we write as
The condition is met for a duration t ::'S t[ such that t[ = mgj.1,slkv. As soon as
t > t[, the force of friction gives way (suddenly becomes equal to 0), and the
motion of the mass m obeys the usual differential equation
d 2x
m- = -k(x - xo).
dt 2
Its has a well-known solution, which can be written in terms of a parameter Wo =
~k/m as
c
o
:E
'"
&
Time
FIGURE 20. Dynamics of a periodic stick-slip motion. Straight-line segments (stick phase)
are connected to portions of sinusoids (slip phase).
This leads to t2 = t] + 2[(n - a)/wo]. The length of the spring at that moment is
6The physics of collisions between solid particles is treated in detail in reference [36].
2.2 Interactions between Two Particles 35
FIGURE 21. Two elastic spheres exchange momenta during a head-on collision.
collinear. This is, of course, a highly improbable event with real granular solids,
which most often behave as inelastic objects, subject to friction, and experiencing
collisions at various angles. Nevertheless, the ideal case is useful to look at be-
cause it involves two fundamental tenets of classical mechanics-conservation of
momentum and of kinetic energy.
Referring to the notation indicated in Figure 21, the speed after impact is given by
m2 2m2
+
mj -
Uj = Vj v2.
mj +m2 mj +m2
In actuality, collisions between real granulates always entail some loss of momen-
tum and kinetic energy. Some fraction of the incident momentum is given up to the
colliding particles by exciting sound waves propagating in their interior. In the pro-
cess, a portion of the elastic energy stored in both particles is dissipated in the form
of acoustic waves or phonons that relax by heating up the mass of both particles. A
loss of kinetic energy can also result from a permanent deformation incurred during
a collision. In any event, we observe experimentally that a ball hitting a vertical
wall of infinite mass with a velocity v rebounds with a smaller velocity -cp v (with
cp :s 1). To a first approximation, and in the case of frontal collision between two
identical spheres, it is convenient to describe the velocity shortfall in the frame of
rT :n[~:l
reference of the center of gravity of the system by a matrix equation [37]
(2-3)
with some caution. On the other hand, it is often sufficient to describe experimental
observations, as we ourselves will realize when we discuss numerical simulations
in Chapter 6. As long as we are prepared to adopt this approach, it is worthwhile to
emphasize once again that E is truly a measure of the momentum loss of colliding
particles. If P and pi designate the momentum of the total system immediately
before and immediately after impact, we may write in the frame of reference of
the center of mass
p = m12(vl - V2),
pi = ml2(ul - U2),
where m 12 = m 1m2/ (m 1 + m2) is the reduced mass of the system of two colliding
objects. It is straightforward to verify that (2-2) and (2-3) are indeed consistent
with the definition of E in the form
pi Ul - U2
E = -- = ----
P VI - V2
where the minus signs in front of the ratios account for the fact that the velocities
reverse direction after the collision. The change in kinetic energy llEkin after
impact is easily shown to be given by
x
x
Y~------_-"'l<I>
If, on the other hand, we introduce a nonzero friction between the ball and the wall
in the form of a single coefficient fL, we must distinguish two separate cases.
1. The Gliding Velocity Remains Positive for the Entire Duration of the Collision
In this case, the system of equations describing the exchange of momenta (linear
and angular) reads
m(u x - vx ) = X,
m(u y - vy) = -fLX,
where m is the mass of the ball (assumed solidly filled), and a is its radius. It
is immediately apparent that this system of four equations with four unknowns
(X, u X , u y , and wI> can be solved if all relevant quantities before the collision are
given and if the coefficients of friction fL and elastic restitution 8 are known. We
leave this as an exercise, and focus our attention on the phenomenology of the
collision, i.e., on the gliding and rotation motions of the ball during the collision.
We note first that the normal momentum X = -mvx(l + 8) is indeed positive,
as it should, since V x is negative. The final gliding velocity must also be positive,
38 2. Interactions in Granular Media
7 V y - awo
2: fh (1 + E) < -'---
-v x
2. The Gliding Velocity Drops to Zero at any Time t1 During the Collision
In that case, the general relations read
m(u x - v x ) = X,
m(u y - vy) = Y,
~ma2(w1 - wo) = -aY,
u y - aWl = O.
This time around, we have a system of five equations in five unknowns, which can
easily be solved. We still have to verify that the linear momentum makes an angle
less than tan -1 (fh) and, in addition, that
The last two conditions turn out to be mutually exclusive. It follows that the
behavior will be of one type or the other depending on the value of the friction
coefficient fh. 7
We may now ask ourselves whether these mutually exclusively situations are the
only two possible outcomes. Is it at all possible for the gliding velocity to change
sign during the interval [to, td of the collision? If the answer is yes, the condition
IY I < fhX could still be verified without us being able to make any prediction
whatsoever about the final value of (u y - aWl). The following reasoning gives the
answer.
Consider an instant t E [to, td, and let I] and l; designate the components of the
reaction force during impact. The equations of motion at that particular instant are
dvx
m-- =1],
dt
dv y
md"t =l;,
2 2 dw
sma dt = -l;a.
7We note here that this mechanistic model assumes that the forces of friction come into play instan-
taneously during the collision. From the point of view of a physicist, this condition is unlikely to
be realized in all situations. Indeed, phenomena of friction (Section 2.2.1) and collisions between
solids (Section 2.2.2) involve plastic deformation, a process that is notorious for being not at all
instantaneous.
2.2 Interactions between Two Particles 39
(a) (b)
FIGURE 23. Two spheres (a) before and (b) after a nonfrontal collision. The parameters
are defined in the text.
d
= :is.
7
m dt (V y - aw)
As it turns out, the quantity t;, which is the tangential component of the reaction
force during contact, has a sign opposite to that of (v y - aw). It follows that it can
only decrease in magnitude. Should it reach zero at any time during the collision,
it will remain zero up to the end. Stated in plain language, should the gliding
speed happen to drop to zero at any time during a collision, the ball would then
go on gliding for the remainder of the collision. This simple argument shows that
the two conditions considered previously are indeed mutually exclusive and that
they are the only two possibilities. One or the other must prevail depending on the
values of the friction coefficient JL and of the components of the incident velocity.
The same distinction between two regimes will come up again when we study
nonfrontal collisions between two spherical particles.
n=
8We have already been introduced in Section 2.2.1 to an elementary mechanical treatment of such
problems of gliding without rolling or rolling without gliding.
40 2. Interactions in Granular Media
The relative velocity of the two particles at the point of contact is given by
where Vi and Wi are the translation and rotation velocities, respectively, of particle
i before the collision. Note that the magnitude of the relative speed IVe I increases
when the individual velocities point in opposite directions and when the rotation
vectors of these particles point in the same direction. The velocity Ve has a normal
component V~,) = (v e . n)n, while its tangential component is v~r) = Ve - V~').
The vector v~r) defines a unit tangential vector such that t = v~)/ Iv~t) I. The angle
of impact y is defined as the angle between the normal n and the relative velocity
Ve , and is such that y E [n/2, n]. Figure 23(a) shows the situation of the two
colliding particles in the particular case when WI = W2 = O.
We now consider what happens to the kinetic momentum during the collision,
using the same convention as before, namely that Oi designates the speedfollowing
the collision. We have
(2-4)
The normal component of ~p has no effect on the angular rotation speed, but the
tangential component does. By creating a torque on the arm defined by the vector
-(dJf2)n, the change in momentum ~p causes a change in angular momentum
such that
21 ,
-nx ~p= dew -w), (2-5)
where 1 is the moment of inertia with respect to the center of the particle and Wi
is the unknown angular rotation speed after the collision. It should be appreciated
that (2-5) indicates that the change in angular momentum is the same for both
particles. Figure 25(b) depicts the situation after impact.
If ~p is known, it is possible to calculate all relevant velocities following the
collision with the help of (2-4) and (2-5). The results are
~p
OJ = V j + - ,
mj
~p
02 = V2 - - ,
m2
dj
Wi = Wj - -n x ~p
1 2It '
I d?
W = W2 - -- n
x ~P.
2 2h
The reader will recall the definition of e, given by
0(11) = -ev(lI) (2-6)
e e '
2.2 Interactions between Two Particles 41
which relates the translation velocities before and after the collision. Equa-
tion (2-6) applied to the normal component of the sum ~P/ml + ~P/m2 allows
us to calculate the normal component of the change in linear momentum, giving
the result
where, as before, m 12 is the reduced mass of the system composed of m 1 and m2.
Coulomb's law establishes a relation between the normal and tangential compo-
nents of ~P, namely 1~p(t) 1 = ~ I ~p(n) I. Since collisions are always dissipative,
the vector ~p(t) must be aligned along -to We can thus write
where we have v~n) = -Vc cos(y), since cos(y) is always negative. We also have
the relation t = v~t) /[v c sin(y)]. Combining (2-7) and (2-8) yields an expression
for the change in momentum ~p
(2-9)
(2-10)
The second term in (2-10) has the same form as the first, except for a factor of ~
which comes about through the bias of the moment of inertia of a solidly filled
sphere. 9 In light of our previous remarks, the coefficient of tangential restitution fJ
9This factor is obviously different for other shapes of colliding particles. For instance, it is equal to ~
for disks, and 2 for thin rings. -
42 2. Interactions in Granular Media
must be equal to the smallest of two values fJo or fJl corresponding to one of two
different regimes:
• Large values of the angle y (y :::: Yo) correspond to a gliding contact, in which
the contacts have been broken. In this case, the appropriate choice for fJ is
fJ = fJo, withfJo E [-1, +1] .
• Small values of y (y ::::: Yo) correspond to a collision for which the interaction
can be described in terms of dry friction and elastic restitution. In that case,
the correct choice is fJ = fJl, with fJl = -1 - ~ fL(l + E') cot(y).
The angle Yo marks the dividing line between the two regimes, which occurs when
fJo = fJl. The value of Yo is given by the equation
7 1 + E'
-tan(yo) = -fL--.
2 1 + fJo
Reference [41] gives additional details on some practical aspects of the two modes
of collision by considering the ratio of the normal and tangential collision veloci-
ties.
This all-or-nothing simplification makes it possible to pursue numerical calcu-
lations of situations involving nonfrontal collisions and rotations. We will discuss
such models using rigid spheres to simulate particles undergoing multiple colli-
sions (see Sections 3.2.4 and 6.1.2).
Before doing so, however, it is useful to first explore in more depth a number
of problems associated with the mechanics of collisions. To that end, we proceed
to calculate some orders of magnitude of a few phenomena occurring during the
frontal collision of two spheres that can penetrate each other, as is almost always
the case in the real world.
E e -- lkh 5/ 2
2 '
k = 4~ _E_v'R. (2-11b)
15 1 - (J2
The quantities E and (J entering in (2-11b) are Young's modulus and Poisson's
coefficient, respectively. Upon impact, the initial kinetic energy is converted partly
into a reduced kinetic energy and partly into stored elastic energy. Thus, we write
dh)2
Mv
2
= kh 5/ 2 + M ( dt (2-12)
The velocity drops to zero when the two spheres have penetrated each other by a
distance h o. At that moment, dh/dt = 0, and we have
2/5
ho - M V4/ 5 (2-13)
- ( k ) .
M 2)1/5
r = 2.94 ( k 2 v
E 6 x 1011 dynes/cm 2 ,
(J 8:::i 0.3,
k 7 X 10 10 ( cgs units).
very much larger than the number calculated above. The beads need only
have a velocity two or three times larger to reach the plastic regime. Most
importantly, the deformation is of the order of the height of the surface rough-
ness. In accordance with what was said in Section 2.2.1, the deformation of
the surface protrusions during a collision is necessarily plastic. The impli-
cation is-and experiments confirm it-that the degree of polishing of the
interacting objects plays a crucial role in the process of restitution during
impact [34J. We will see shortly an approach to describe the penetration of
inhomogeneous spheres with a surface layer softer than the core.
• Part of the energy may be dissipated by exciting sound waves inside the solid
beads. As an example, we may calculate the time Tph it takes a phonon to
complete a round trip across a sphere, knowing the propagation velocity Vph'
The result is Tph = 2R/v ph = 0.15 cm/6 x 105 cm/s = 2.5 x 10-7 s. In other
words, there is plenty of time for several round trips during a typical collision.
This effect constitutes a potentially important perturbation that in some cases
may invalidate the model embodied in (2-12).
(a) (b)
FIGURE 25. Schematic ofthe zone affected during the penetration of two (a) homogeneous
and (b) inhomogeneous spheres.
2.2 Interactions between Two Particles 45
homogenous spheres. In order to understand the reason for the change, it might
be worthwhile to first outline an argument-originally advanced by de Gennes-
explaining the usual ~ exponent.
Consistent with the notation of Figure 25, h is the depth of penetration during a
collision, and R is the radius of each sphere. From elementary geometry, we can
easily derive an expression for the radius a of the contact circle. Assuming that
h is very much smaller than R, which is justified if we insist on remaining within
the elastic regime, the result is a 2 ~ Rh.
It is not unreasonable to assume that the deformation of colliding spheres affects
a depth roughly equal to the radius a of the contact circle. Under these conditions,
and in the presence of a stress P, the classical definition of Young's modulus gives
a relation of the type
h
P ~E-.
a
This justifies the ~ power law. If we now consider a sphere covered with a soft cmst
of thickness e much smaller than the radius R of the sphere, but sufficiently large
compared to the penetration depth h, we can assume that the deformation will be lo-
calized within the outer layer whose Young's modulus is E e « E. In that case, the
depth affected will be of the order of e, rather than a as in the previous case. In other
words, a stress P acting on the sphere now produces a penetration depth given by
The relation between penetration depth and force has changed from h ex p2/3,
valid in the case of Hertz's homogeneous spheres, to h ex pI /2. The deformation
increases more slowly with applied force. This makes sense if we note that, in
the presence of a soft crust, the deformation is more localized to regions near the
contact point and does not spread as deeply into the core.
Such a dependence can explain, at least in part, certain experimental observations
that deviate notably from what would be expected on the basis of Hertz's model
[44]. This is not the only possible explanation. Others have been proposed, invoking
notions of disorder and enrichment in the vicinity of contact points under the
influence of stress [45]. We shall discuss a particular example in Section 3.1.2.
46 2. Interactions in Granular Media
FIGURE 26. A sphere rolling down a rough surface looses kinetic energy by friction and
successive collisions. Its trajectory is a mix of ballistic arcs and sections of curves following
the rough surface.
A uniform motion results whenever r = 0, i.e., when the friction force exactly
offsets the effect of gravity. It then becomes evident that the problem boils down
to working out a plausible expression for the force of friction F, as we will see in
several examples later in this book (for instance, in Section 4.2.2). The idealized
2.3 A Single Particle on Top of a Granular Medium 47
case described above helps us to understand that the force of friction corresponds
to a loss of energy as the particle proceeds down the slope. This allows us to write
F = aEjax, where x corresponds to the distance covered. The energy of the
particle of mass m has two components:
(1) its kinetic energy E b which is dissipated over a characteristic distance A as
a result of successive collisions and friction with the support; and
(2) its potential energy E p' which also decreases as the particle drops down
into a succession of wells of typical depth D.
Thus we may write
E k = 'i1 mv 2
and
Ep = mgh cos(e),
where h is the altitude of the particle.
If we consider that the loss of kinetic energy occurs over a characteristic length
A (of the order of a few diameters D), we obtain the first component of the fric-
tion force Fk = m v 2j2A. In order to derive a suitable expression for the second
component F p , corresponding to the loss of potential energy, we use the following
argument:
• At high velocity, the freely moving particle executes ballistic flights whose
typical range is of the order of D. This range can be estimated independently
(!
by considering ballistic flights of height f"...h ~ g T 2 ), where T is the typical
time of free flight of a particle moving at speed v. In this approximation, the
quantity vT corresponds to a typical distance of the order of D. Under these
conditions, we have T ~ Djv, and the force of friction equivalent to the loss
of potential energy can be written as
mgf"...h 1 2D
F p = ~ ~ 'img v 2 ·
The equation just derived suggests that the corresponding dissipative force
becomes less and less effective as the speed of the particle increases, which
certainly conforms to our intuition.
• At low velocity, the path of the particle follows the profile of the surface of
the granular support. The resulting loss of potential energy should no longer
depend on speed. It corresponds only to the dry friction experienced by the
particle as it moves down the surface. In this case we have
F p = mg.
• At intermediate velocity, we simply interpolate between the two extremes
with an equation of the form
1
F ~mg--------:,------
p 1 + {3(v 2 jgD)'
48 2. Interactions in Granular Media
F = Fp + Fc = mg [ t )
l+b v 2 jgD
+ c (~)],
gD
(2-16)
LLI~ 3 a=b=1
c
o
t5 2 c= 0.25
it
""0
.§
co
E
o
z 0 +---+-----.-----.--~
o 2 3
Normalized Velocity ~
FIGURE 27. Effective friction force F as a function of the particle's velocity. The curve
does not go through the origin and exhibits a minimum.
• The shearing force required to set the upper half in lateral motion is strictly
proportional to the load P and is independent of the surface area under shear
F=~sP.
This result is entirely consistent with the first law of solid friction reviewed
in Section 2.2.1. 11
IOSoil-mechanics scientists routinely use boxes of this type, roughly a meter long, to measure granular
friction coefficients out in the field.
II It appears that Dawes did not try to verify that the force F was independent of the surface area of
the boxes he was using [51. It also appears that Coulomb implicitly accepted this independence by
writing that granular materials obey the laws of friction as stated by d' Amontons-La Hire [II]. This
was indeed confirmed in recent experiments.
50 2. Interactions in Granular Media
FIGURE 28. Typical experimental setup used to measure the coefficient of the static friction
fhs of a granular material.
• The coefficient fts is of the order of 0.7, which is quite comparable to values
typically encountered in the case of friction between solids. The angle e,
given by the formula e = tan-lefts), is approximately equal to 35°. It is
useful to keep this value in mind, as it will resurface in Chapter 4.
We note in passing that these results hold equally well for cohesive and nonco-
hesive granular materials (cohesive materials are made of particles kept together
by cohesive forces). The only difference is that in the cohesive case the above
expression must be modified by the addition of a constant term C; the expression
then takes on the form F = fts P + C.
These results have actually been long accepted by people working in the field of
soil mechanics, who routinely make use of the concept of stress. 12 Nevertheless,
these properties constitute a rather stunning discovery. Noone could have predicted
them, given the complexity of granular piles and the many unknowns concerning
their equilibrium (see Section 3.3.1). In particular, it seems quite unlikely that a
model based on plastic deformations, such as we invoked in our discussion of the
laws of friction between solids, could be extrapolated to the present situation. The
best we can do at this point is to accept that a medium as complex and heteroge-
neous as a granular can lead to exceedingly simple macroscopic laws. As an aside,
the same can be said of two sheets of cardboard in contact [48].
12Experts in soil mechanics use Coulomb's law by assuming that the shear stress T is related to the
normal stress (}Il by T = fhs(}n. In doing so, they implicitly admit that a stress tensor can be defined
in a granular material (which is far from evident) and that, consequently, the force required to set
the object in motion does not depend on its surface area. The assumption tums out to be justified in
most situations.
2.4 Interactions Between Several Particles 51
FIGURE 29. The flow of a granular material medium normally involves only a few layers
near the free surface. The mean velocity of the particles decreases rapidly with depth.
- - ~ - - y2
Fe -_(aEe)~mD2 .
ax 2A e
We also note that the quantity 17Y, where 17 is the viscosity of the interstitial
fluid, represents the friction force F v due to viscous interaction with the fluid,
divided by the cross-sectional area of the solid in motion, which is of the order
of D 2
Bagnold introduced a number B defined as the ratio of the forces due to friction
52 2. Interactions in Granular Media
and collisions between solids to the forces exerted by the viscous fluid. 13 In other
words
Fc my
B=-~--
Fv 2A e 17
Two distinct regimes can be defined depending On the relative importance of Fc
and F v . In particular:
• B < 40 corresponds to a regime known as macro-viscous, in which energy
is dissipated primarily by way of viscous interaction with the ambient fluid.
Physically, the fluid does most of the work to set the particles in motion. Only
rarely do the particles come in contact with each other. Actual examples
of macro-viscous systems include muds, moist pastes, and solid particles in
suspension.
• B > 450 corresponds to an essentially granular regime, in which most of the
energy is dissipated by collisions and friction between the solid particles.
The intermediate case requires some caution.
13 Actually,
Bagnold started out by introducing the notion of dilation 8 of a granular mixture, defined as
8 = D/A e . Bagnold's number is then equal to 8 1/ 2 Ps D 2 Y/1J. This quantity merges with the number
B defined in the text in the limit of concentrated solutions, for which (Dy)I/2 is indeed of the order
of D.
3
Fluidization, Decompaction,
and Fragmentation
Common sense tells us that, in the absence of any external perturbation, a pile
of granular material is at rest. It can be inclined, gently shaken, and submitted to
various mild stimuli without anything drastic happening, in other words without
triggering a flow or collapse, and without affecting the relative positions of the
granules. We also know from experience that a sufficiently high inclination can set
off an avalanche, possibly even a continuous or intermittent flow; that vibrations
cause a granular material to behave somewhat like a liquid; and that sand deforms
readily when crushed. All these properties are markedly different from those of
the normal solids, liquids, and gases we are accustomed to. The main difference
appears to be that granular materials must be "sufficiently" perturbed in order to
change. This observation strongly suggests that we will be dealing with issues
such as thresholds, nonlinearities, and hysteresis effects.
The aim of this chapter is to clarify some principles and highlight some of the
difficulties encountered when trying to understand why and how a granular pile
changes configuration. We are particularly interested in phenomena of decom-
paction,jiuidization, and fragmentation. As the chapter progresses, we will come
to realize that these phenomena can overlap, sometimes even act in concert, and
that they constitute different facets of a more complex reality. For instance, we
will show that progressive decompaction and convection, which show up in long-
term observations, can actually result from a subtle superposition of discontinuous
processes and fragmentations that typically occur over short periods of time. What
this means is that, as is sometimes the case in other branches of physics-most
notably in hydrodynamics-the evidence for a particular phenomenon depends on
the duration of the observation.
However, before tackling the problem of how granular stacks deform, we will
take the time to analyze briefly the static properties of a sand pile. Paradoxically,
...
·. . .
.. ·... ..
..
... .· ....' ....
_M"~_·_· ._. .. . _,,_,_
_~
FIGURE 30. The classic "cannon ball" stack. Even though the structure appears to have a
high degree of order, many parameters are actually undetermined.
the static properties pose even more delicate challenges than their dynamic coun-
terparts, which we will examine later.
1It is
interesting to note that the static properties of this structure continue to be the object of a great many
theoretical articles, whose results, incidentally, do not always agree with the (difficnlt) measurements
that can be made.
3.1 The Static Properties of a Granular Pile 55
FIGURE 31. Collision of three particles. Classical dynamics requires a precise knowledge
of the order in which the collisions occur, as well as of the nature of the surfaces.
56 3. Fluidization, Decompaction, and Fragmentation
related to the plastic deformations of the microcontacts, the nature and magnitude
of this reaction force are perfectly known. The tangential reaction, on the other
hand, is quite another matter. The corresponding force, called "resistance to fric-
tion" because it opposes lateral motion, is inherently indeterminate, which shatters
any hope of describing precisely and exactly the static properties of objects in fric-
tion. While it is clear that the force R opposed to gliding cannot exceed a critical
value F/-Ls, where F is the force that keeps the objects together, it is no less clear
that, as long as the system is at rest, R has an unknown value that can be anywhere
between 0 and F/-Ls. We can think of the problem by imagining that the micro-
scopic protrusions get deformed and oppose subsequent displacements, although
we do not know precisely the details of these deformations. To complicate matters
further, the deformations in question may well take place in the plastic regime,
which implies slow changes. In other words, strictly speaking, we ought to treat
the force of resistance to gliding as a time-dependent problem. It is also apparent
that a complete description of all the forces involved between two solids depend on
the prior history of their contact, specifically on how that contact was established.
The situation is illustrated in Figure 32. Part (b) of this figure shows that the
force acting on a brick placed on a support featuring two perpendicular slanted
surfaces depends on the way the brick was placed. It might have first been set
free on the left surface and subsequently pushed against the perpendicular wall, or
the other way around. The same ambiguity exists with regard to the equilibrium
of spheres resting on each other, as shown in Figure 32(c). In all these cases, the
solution to the problem depends on the prior history of the system.
It would then seem that any attempt to describe the forces of contact in a granular
edifice is doomed from the outset. It turns out that this is not always the case, as
calculations are often possible as long as enough information is available about the
way equilibrium was reached. This is in fact quite a general property of granular
materials. We will illustrate the point with a particularly simple example which
we will revisit from a different angle in Chapter 4.
F
(b)
(c)
FIGURE 32. The forces of friction between solids can make it impossible to determine
unambiguously the magnitude of all the forces at equilibrium.
3.1 The Static Properties of a Granular Pile 57
subjected to are perfectly known. They include its weight P, the normal reaction
S of the surface, the reaction R = -kx of the spring (where x is the deviation
of its length from equilibrium and k is its stiffness, or spring constant), and the
force of friction with the surface, i.e., the tangential reaction F. The angle e can
vary between 0 and n /2. We know the coefficient of static friction fhs, and for
simplicity we assume that the coefficient of dynamic friction fhd is zero. Our goal
is to determine the compression x of the spring as a function of the angle e.
Finding the exact solution to this problem requires a bit of caution. We start
with two extreme cases. First, when e = n /2, the solution is trivial since the force
pressing the brick against the surface vanishes. Therefore, the friction force F is
itself zero, in which case we have x (n /2) = P / k. The case of a horizontal sur-
face (e = 0) is already more complicated. The solution is actually undetermined
because the spring may have been left initially in a compressed state defined by x,
with x allowed to have any value as long as the reaction R is less than the tangential
reaction due to friction. In other words, x < Pfhs/ k. Here is a case where the
position XQ of the brick cannot be determined without some additional knowledge
about the way the configuration of the system was arrived at.
Suppose now that we start with the spring at rest (xQ = 0) and we gradually
incline the surface from 0 to n /2. We denote by e+ angles that are reached by
increasing the inclination, and e- in the opposite case. The tangential component
ofthe load due to the weight increases as P sin(e+). Since the spring is initially at
rest, this load is the only force that is offset by an equal and opposite friction force,
until the moment the brick starts moving when tan(e+) = fhs. The brick then
drops down without friction (since we assumed fhd = 0) along the surface until
the spring stops in a new position Xe+, 2
which will happen when kxe2+ = P sin(e2+).
The brick being once again at rest, the friction force is reactivated, and the surface
must be inclined to a new larger angle ei to trigger the next motion. The angle
ei is such that kx e: = P sin(ei) = P sin(ei) - fhs cos(ei). We can thus see
that the descent of the brick (analogous to the stick-slip mechanism discussed in
Section 2.2.1) is marked by successive pauses determined by
(3-2)
since the friction force now opposes upward, rather than downward, displacements
of the brick. In other words, the angles at which the brick stops are different on the
return trip, even though the starting and ending points are the same. The behavior
is illustrated in Figure 33.
Some simple experiments confirm the overall validity of the model just de-
scribed, although a particular behavior is often observed in the vicinity of e = n /2.
58 3. Fluidization, Decompaction, and Fragmentation
~
c
o
~
OJ
C
o
iIi
o 10 20 30 40 50 60 70 80 90
Tilt angle 80
FIGURE 33. Mechanical hysteresis effect. The successive elongations of the spring shown
in Figure 32(a) is plotted as a function of the tilt angle. The calculation assumes a static
friction coefficient f-ts of 0.3.
where Nand T are unit vectors, respectively, parallel and perpendicular to the
displacement of the object and E = ± I depending on the direction of the displace-
ment. We know that f-ts = tanCes ), where es is known as Coulomb's angle. It is
the angle of inclination required for the first slip of the brick when the tilt of the
support is increased from horizontal.
It is easy to appreciate the subtle and ambiguous character of the forces involved
in a system at equilibrium by considering the following situation. Imagine that we
tilt the support by a given angle ei greater than tan -1 Clls) and start with the spring
in a quiescent state. What will the final position of the brick be when we place it
on its support? In light of the previous discussion, it should be obvious that the
result will depend on precisely how we conduct the experiment. In particular, do
we first place the brick on its support and then let the spring do what it will, or do
we first find the equilibrium position of the spring and then let the brick down?
3.1 The Static Properties of a Granular Pile 59
• The contact is made with the spring first and the support second. In this case,
the spring is first compressed until kx l is slightly larger than P sinCe). The
brick is then placed in contact with its support. The spring cannot stretch
back because of the force of friction directed downward. The equilibrium
position is then given by kx l = P sinCe) +ftsP cos(e).
• The contact is made with the support first. The brick is then left to slide
down, compressing the spring. In this case, the friction forces are devel-
oped first. They oppose motion toward the bottom, and are therefore di-
rected upward, opposite to the weight. The spring will contract until kX2 =
P sinCe) - ftsP cos(e).
The two equilibrium positions are obviously not the same. It is impossible to
predict the final position and the applicable forces short of knowing the details of
the sequence of steps.
There is an analogy between the behavior of this system and that observed in
magnetic compounds, in which the hysteresis is due at least in part to the irre-
versibility of the formation and disappearance of domains. A tally of the forces at
play in a granular material, even when at rest, suffers from a number of indeter-
minacies, and we can expect serious difficulties in trying to model the equilibrium
of such a structure.
2For instance, gravel ballast used to stabilize railways has very nonlinear elastic properties. Because
it becomes stiffer when compressed, it can offer a range of malleability and strength for a wide range
of loads.
60 3. Fluidization, Decompaction, and Fragmentation
FIGURE 34. Stress pattern observed in a two-dimensional granular material under com-
pression (after Dantu [51]).
i~
F.
J1~dl
a p gdl
on any of the contact points between contiguous spheres. Some shearing force is
actually tolerable, as long as we remain in the dry friction regime, which requires
particular conditions that are left as an exercise. For simplicity, we simply seek
here the conditions that make these forces, as well as the torque, null. We do this
by writing that the resultant of all the forces involved must be tangent to the curve
at points A and B. Assuming that the segment dl is rigid, we may write
FA +FB + pgdl = O.
e
If is the angle the segment dl makes with the horizontal, the following relations
hold in a Cartesian coordinate system: dxjdl = cos(e), dyjdl = sinCe), and
dx 2 + d y 2 = d1 2 . The equilibrium equation projected on the horizontal axis yields
dx
F - =Fh (3-3)
dl '
(FdYdl ) A
_ (F dYdl ) B
_ pg dl = 0,
Y
-d ( Fd- ) +pg=O.
dl dl
d(d-dxY) +pgFh- -- 0 .
-
dl
(3-4)
62 3. Fluidization, Decompaction, and Fragmentation
We thus arrive at a differential equation describing the curve we are looking for.
It reads
pg dt
---
Fh dx
whose well-known solution is
y = - ;; [COSh (- ;~ x ) - I} (3-5)
As we might have guessed, this last result is similar to the equation describing a
flexible rope hanging loose between its two suspended ends. The requirement of
"flexibility" means in the present context that the particles the arch is made of may
roll on each other but are not allowed to slip laterally. In other words, (3-5) correctly
describes a granular arch if-and only if-the constituent particles are subject to
rolling without gliding, which implies some restrictions specified in Section 2.2.2.
Figure 36 shows a magnificent example of architecture that owes its stability to the
principles we have just discussed. Essentially the same approach can be followed to
derive the equation corresponding to an arch supporting a load whose distribution
is known. We will treat the simple case of an evenly distributed weight.
Arch Supporting an Evenly Distributed Load
We now assume that the arch has a negligible linear density, but that it supports
an evenly distributed linear mass mo. In this case, we have mag dx = pg dt, and
(3-4) must be modified to
whose solution is obviously an inverted parabola. The highest point of the parabola
has a horizontal tangent, from which we get
1 mag 2
y = -z--x
Fh
The force of tension at a point of abscissa x is straightforward to calculate. The
result is
FIGURE 36. Photograph ofthe aqueduct ofMaintenon, built during the reign of Louis XlV.
It is a magnificent example of arches that have remained stable through several centuries.
64 3. Fluidization, Decompaction, and Fragmentation
3A more detailed discussion of this problem can be found in an article by S. Roux [24].
3.1 The Static Properties of a Granular Pile 65
c-
o
'§
E
.2
OJ
8
v
(External Force)
F ex: (D - D o)3.S.
Rubber pouch
containing sand
+ colored liquid
FIGURE 39. Parallelogram used to model the unit cell of a two-dimensional uniform
granular medium.
The lines connecting the centers of the four disks form a parallelogram that
changes shape when forces are applied in the manner shown in Figure 39. We
focus our attention on the closed geometrical figure composed of the four disks
themselves (assumed to be rigid) and the void between them. More particularly,
we are interested in how the surface area SI of this figure varies during the defor-
mation. Let h v and hi be the lengths of the two diagonals of the parallelogram.
We can show that the surface area of the four disk sectors situated within the paral-
lelogram is constant and equal to rr R 2 . It follows that SI has the following simple
expression
(3-6)
The diagonals hi and h v are restricted to values such that ht + h~ = l6R 2 . The
variation /::"SI of the surface area of interest can then be written in terms of one of
the two diagonals only, say hi
where hi can take on values only between 2R (when the left and right disks
touch each other) and 4R cos(rr /6) (when the top and bottom disks come in con-
tact).
Figure 40 is a graph of /::"SI (normalized to 4R 2) plotted against the length hi
of the horizontal diagonal (normalized to 2R). The graph differs markedly from
what would be expected on the basis of the mechanical properties of homogeneous
and isotropic solids. As the horizontal deformation begins in response to a vertical
stress, the curve /::"St(h l ) starts out by increasing, until it reaches a maximum.
In this regime, the volume of the material goes up. This behavior agrees with
Reynolds's principle, and runs counter to what is observed with the usual homoge-
neous solids. We should point out, however, that the material does not necessarily
comply with the requirement that it be "strongly compacted," except to the extent
68 3. Fluidization, Decompaction, and Fragmentation
OJ
a: 1.0
""
~ ""
ell
"" Solid
"
~ regime
Reynolds's " "I
ell 1
CD 0.9
()
{g regime
:::J I" "
CfJ
I ""
I
""
0.8
1.0 1.2 1.4 1.6 1.8
Distance between particles (h e/2R)
FIGURE 40. Total surface area of the object defined in the text (normalized to 4R 2 ) plotted
against the horizontal distance between spheres (normalized to 2R). The inclined straight
line describes the behavior of a traditional homogeneous two-dimensional solid, whose
surface area always decreases when compressed.
that adjacent particles are already in contact. To the right of the maximum, on
the other hand, the volume does decrease, in accordance with the classical laws of
elasticity. The conclusion is that Reynolds's principle can, indeed must, depend
in a subtle way on the mode of stacking of the particles. We begin to appreciate
how ambiguous the statement "strongly compacted" really is. Finally, we have
ignored what might happen toward the left side of Figure 40. We can at least con-
ceive that a "loose" granular medium should see its volume initially shrink as it is
being compressed. Figure 41 illustrates that some arrangements obey Reynolds's
principle, while others do not.
We may attempt to calculate effective deformation parameters by resorting to
the standard methodology used in the mechanics of solids. Since we have no idea
what the elastic stiffness of the object considered might be, except that it must be
(a) (b)
16x16 16x16
Planar square lattice Triangular lattice
FIGURE 41. Two possible lattice configurations in two dimensions. Stack (a) has minimum
compactness; its volume decreases upon application of any external stress. By contrast,
stack (b) exhibits maximum compactness; it does conform to Reynolds's dilatancy principle.
3.1 The Static Properties of a Granular Pile 69
anisotropic, it is out of the question to try to define effective elastic constants. What
we can do, however, is to define the equivalent of Poisson's coefficient (]". In the
case of a homogeneous and isotropic material subjected to a vertical load, Poisson's
coefficient is defined as the ratio of the horizontal dilatation strain Uz = dhz/ h z to
the vertical contraction strain Uv = dh v / h v
Uz dhz h v
(]"=-- = - - - - .
Uv dh v hi
In the case of the parallelogram considered here, the restriction hi + h~ = 16R 2
implies that hi dhz + h v dh v = 0, which leads to an effective Poisson coefficient
given by
When the two disks on the horizontal axis are in contact, we deal with a compact
triangular stack in two dimensions (which is equivalent to a compact hexagonal
stack in three dimensions). In this case, the effective Poisson coefficient is equal
to 3. This value is abnormally large when compared to the usual solids for which
thermodynamic stability considerations impose the restriction (]" :s 0.5. 4 The co-
efficient (]" decreases as the two disks lined up vertically get closer to each other.
When the parallelogram becomes a square, which conesponds to the maximum of
the curve in Figure 40, we have (]" = 1. This can be demonstrated by differentiating
(3-6), which gives
It is clear from this last equation that the variation in total surface area d S! changes
sign when (]" = 1. In other words, Reynolds's dilatancy principle ceases to be valid
as soon as the effective Poisson coefficient becomes smaller than 1.
4This does not imply that the principle of thermodynamic stability does not apply here. The large value
of the coefficient (Y simply results from the local anisotropy of the material considered.
5 As an exercise, the reader can work out the stability conditions of the structure by introducing a
coefficient of static friction between adjacent disks. It turns out that the structure will collapse when
the force applied on the topmost disk exceeds a critical value determined by Coulomb's angle of
friction.
70 3. Fluidization, Decompaction, and Fragmentation
must, of course, be deformable. They effectively play the role of Young's mod-
ulus by resisting lateral movements. In the interests of simplicity, and without
undermining the argument, we assume that the vertical walls undergo a uniform
horizontal displacement Uz = (5z E, where (5z is the stress exerted by the disks on
the walls and E is Young's modulus of the material the walls are made of.
Stability having been ensured and dry friction neglected (the disks are assumed
perfectly rigid and smooth), we are ready to focus on the problem we want to solve:
What is the stress exerted on the lateral walls of the container when a vertical stress
(5 v is applied to the system?6 For the time being, we consider a system made of only
three layers of a compact triangular stack of the type illustrated in Figure 41 (b). The
reason we choose to do so is that such a stack is made of a series of parallelograms
similar to the ones we have dealt with in our previous example. We will see later
how to solve the problem for a large number oflayers (see Section 3.1.4).
By itself, an individual parallelogram exerts no resistance to the stress applied to
the disks lined up vertically. 7 It merely redirects the displacements (or strain in the
language of continuous mechanics) toward the lateral walls. Since the system is
nondissipative, the principle of conservation of work applies and the stress exerted
on the lateral walls is immediately seen to be such that
(3-7)
6There should be no confusion between Poisson's coefficient, which is a scalar, and stress, which is a
tensor.
7 Strictly
speaking, we ought to speak in terms of forces rather than stresses. But we have deliberately
opted to resort somewhat loosely to the traditional terminology in the mechanics of continuous media.
8We will often use the word "redirection" in this book. It means that part of the vertical stress applied
to a granular material creates a horizontal component that presses against the vertical walls. The word
"reorientation" is also sometimes used.
3.1 The Static Properties of a Granular Pile 71
had a marked proclivity to redirect vertically applied stresses toward the sides, as
described above in the context of simple models. 9 The model was elaborated on a
few years later by Lord Rayleigh [16]. It rests on two principles that are important
to understand clearly:
• Since the material is made of discrete constituents with a finite size, the
relevant space variables by definition cannot approach zero, and writing dif-
ferential equations obviously raises serious questions. However, this type of
approximation has been used successfully in other branches of physics and
has led to solutions that for the most part seem to agree well with experiments.
A case in point is the transition from the mechanics of atomic structures to
that of continuous media. It should be kept in mind, however, that the ap-
proach works only as long as we deal with large numbers of particles. It
breaks down when we try to explore properties on a local level, such as when
the number of particles involved is small. We will see an example of this
limitation when we study the dynamic behavior of granulars, where a model
based on continuous media fails to account for the phenomenon of fragmen-
tation in granular media that are vibrated or in forced flow. The message
here is that the approach applies only to structures made of a large number of
partic1es. lO
• Problems related to the lack of cohesion of a dry granular medium are even
more fundamental. By definition, a granular material is heterogeneous and
contains empty spaces. Given the constraints we have specified early on in
this book, namely that the particles not interact with their gas environment, the
intergranular voids inevitably impose some restrictions on how deformations
9This property is not the exclusivity of granular materials. Soft media, such as lUbber, exhibit it
too. Unlike lUbber, however, the granulars we will deal with have no cohesion. We will see some
consequences of this fact a little later in this chapter.
10 Apractical method useful in simulation work will be presented in Section 6.1.3. It permits a
smooth transition from discrete variables (position, individual velocity, mass) to continuous variables
consistent with thermodynamic quantities (density, mean speed, temperature, etc.).
72 3. Fluidization, Decompaction, and Fragmentation
and stresses can propagate within the medium. A differential quantity such as
a variation in pressure becomes meaningless the moment the particles are not
in contact. Likewise, the concept of individual rotations presents difficulties.
Classical mechanics, as it applies to continuous and homogeneous solids, can
handle shear forces routinely but is powerless to deal with local rotations. We
may anticipate that a description based on the theory of continuous media
requires that the network of forces be transmitted throughout the medium
without interruption. A variable such as pressure can be defined in a granular
material only inasmuch as the contact chain is unbroken and rotations of
individual particles do not occur. We will make use on these remarks in what
follows, notably in our analysis of the fragmentation of granulars in forced
flow (see Section 3.2.4).
• Problems related to the finite size and discontinuities of these materials can be
particularly bothersome when it comes to their dynamic properties, as we will
appreciate later on. Given the wide range of relaxation times in such systems,
a continuous theory may be useful for sufficiently long-term observations,
giving the system plenty of time to relax, but completely inadequate for
short-term observations. Indeed, it is not all that unusual in the world of
physics for phenomena and their description to be quite different depending
on the duration of the observations. We will see examples of this in this
chapter.
• Since pressure increases with depth (we adopt the convention that h = 0 at
the top of the cylinder, and h > 0 toward the bottom), the slice of interest
experiences a force directed toward the top and equal to A dpv'
• The weight of a slice of thickness dh constitutes a force directed toward
the bottom and equal to pg dAh, where p is the volumetric density of the
material, assumed constant throughout the slice.
Surface Area A
Perimeter P
FIGURE 42. Cylindrical configuration defining the parameters used in Janssen's model.
3.1 The Static Properties of a Granular Pile 73
• The forces of friction with the walls, resulting from an infinitesimal move-
ment of the slice toward the bottom, are directed upward. This is not an
arbitrary choice, as it amounts to assuming that the material slowly settles
under the action of gravity, a trend opposed by friction. 11 As we will see later,
there are some experimental situations that would justify making the opposite
choice for the direction of the forces of friction. In any case, the relevant force
exerts itself all around the wall, over an area P dh. Its value is thus equal
to fLs Ph P dh. Taking into account (3-8), which relates the horizontal and
vertical components of the stress, the forces of friction become KfLs Pv P dh.
We are now in a position to write the equilibrium condition for the particular slice
considered. It reads
(3-9)
It can be rewritten as
(3-10)
llThis is a crucial point to understand. Janssen and Rayleigh implicitly considered that the forces
of friction were always right on the edge of giving way at the walls and throughout the structure.
This picture completely ignores the indeterminacies affecting the forces of friction which we have
discussed in Section 3.1.1. In other words, both authors replaced the inequality T :: /LsN by
the equality T = /LsN-a rather bold assumption. Another way to put it is that the calculation
corresponds to a fully relaxed pile. As such, the model applies to long-term observations. We shall
see later that reality can be considerably more complex, particular for short-term observations.
74 3. Fluidization, Decompaction, and Fragmentation
Vertical Pressure
o ...-- Hydrostatic Psat
regime
1:
0>
'iii
I
Saturated
regime
h
When h becomes larger than about A/ P Kf.-L" the vertical pressure Pv saturates,
as it approaches asymptotically a limit given by Pv ---+ pg(A/ P Kf.-Ls).
Denoting by X the argument appearing in the decreasing exponential in (3-11),
we have X = (Ph/ A)Kf.-Ls, where Ph represents the outer vertical surface area of
the container, while A is its cross-sectional area. Because it will reappear later on,
we define the ratio of these two areas as a parameter S, which we call the aspect
ratio (S = Ph/A). In terms of this new parameter S, the argument, which we
denote X, appearing in the exponential is simply X = S Kf.-Ls. For reasons that will
become clear later on, we call this argument X the decompaction parameter. It
is a dimensionless number that completely characterizes the distribution of forces
in a cylindrical pile. To show that this is true, consider the mass m of granular
material contained within a depth h counted from the top; it is given by m = pAh.
The vertical force F v exerted on a layer at that particular depth h is equal to
_ _ mg -x
Fv - Pv A - -(1 - e ). (3-12)
X
Through the combined effects of dovetailing and friction with the walls, the ap-
parent weight of the cylindrical column is reduced by a factor that depends solely
on the dimensionless decompaction parameter X.
Specific Applications
We briefly consider some specific geometries shown in Figure 44.
Cylindrical Container ofDiameter D
Equation (3-11) then becomes
Two-Dimensional Container
A single layer of a granular is confined between two large flat frontal plates. The
granular medium has no friction with those plates, although it does with the small
end walls. We will make repeated use of such a two-dimensional stack in the
remainder of this book. Let e be the thickness of the material, L the length of the
cell, and h its height (see Figure 44). The aspect ratio S of such a stack is given
by
Ph 2he 2h
S------ (3-13)
- A - Le - L'
The numerator includes the active part of the perimeter only, ignoring that which
does not contribute to friction. The decompaction parameter reads X = SK!J,s =
2K!J,shjL, and (3-11) becomes
(a) (b)
N
spheres
FIGURE 45. Spheres excited by a vibrating plate. Diagram (a) is the classic problem of a
"bouncing ball." Diagram (b) corresponds to several spheres stacked vertically.
12The importance of making a distinction between these two terms will become apparent as we go
further in our discussion. Fluidization of a dry granular gives it dynamic properties reminiscent of
those of a nonviscous liquid or gas. Decompaction is a phenomenon that gives it the ability to execute
internal movements of reorganization, for instance by convection.
3.2 Dynamic Properties of a Granular Pile 77
where ZiQ and Via are measured immediately following the previous collision. The
collisions are described by linear equations of the form
(3-15)
where Ui and Vi are the velocities of bead i immediately before and immediately
after a collision measured in the frame of reference attached to the center of mass
of the two colliding objects. By convention, Ua and Va refer to the vibrating plate.
The beads are numbered sequentially from bottom to top.
How to implement such a calculation constitutes an important physics problem.
It involves making some decisions on the sequence of events-in this case ballis-
tic flights and collisions between the N particles and the vibrating plate. Without
going into the details of the computer simulation, which will be covered in Chap-
ter 6, we will simply mention the simplest and most natural algorithm. It belongs in
the class of so-called "event-driven" algorithms, which analyze events sequentially
78 3. Fluidization, Decompaction, and Fragmentation
10 .-.:::,....----.----y---,...---,-----,
N
o -",
.........
..........
FIGURE 46. Dynamics of a collection often spheres, calculated with the help of (3-14) and
(3-15), using 8 p = 0.6 and 8 = 1. The upper diagram shows the positions of the spheres
plotted against time. The lower diagram gives the time intervals between successive colli-
sions (after [59]).
in the order in which they occur. 13 For instance, starting at some time t, we can
construct a matrix T which contains a sequence of times ti at which future events
are predicted to take place. In particular, the smallest element tim of that matrix is
the time of the very next event. When it has occurred and its consequences have
been calculated, a new matrix T is constructed, and the cycle is started all over
again. There is absolutely nothing wrong with this procedure as long as we are
not faced with two or more simultaneous events (in the sense of physics or of the
computer's ability to distinguish them), in which case some inherently arbitrary
choices have to be made. Erroneous decisions can, of course, seriously taint the
subsequent chain of events. More pragmatically, the reader may recall a calcula-
tion we did in Section 2.2.2, which convinced us that collisions take some finite
amount of time (of the order of a microsecond in the example treated), and that
this time further depends on the velocity of the colliding objects. Including this
effect leads to additional uncertainties.
Summarizing, the calculation technique based on Newton's equation for bal-
listic flights (3-14) and event-driven collision matrices (3-15) can run into major
problems when the sequence of events becomes difficult to keep track of. The point
is illustrated in Figure 46, which tells the story in the case of ten beads. The figure
shows the trajectories Zi (t) over the duration of one excitation period, as well as the
time intervals separating sequential events. We notice immediately-and that is a
very general result-that the time interval between consecutive collisions becomes
exceedingly small (less than 10-7 s in the present case) when the ten beads get so
tightly clustered as to be practically in contact. From a mathematical point of view,
we may even claim that contact will be achieved when the interval between events
13 "Event-driven" algorithms are to be distinguished from "time-driven" algorithms and other sequen-
tial techniques. They are also sometimes referred to under the heading of "collision method."
3.2 Dynamic Properties of a Granular Pile 79
goes to zero, although that may not be too realistic because even the most simplistic
models indicate that two colliding objects remain in contact for durations as long
as several tens of time intervals in this limiting regime. Hence a serious problem.
It is clear that the difficulty has to do with the existence of two separate character-
istic times of the system-the duration fl Tc of a collision and the interval flt m ,m-l
between two successive events. The validity of the model is assured only to the
degree that fl Tc « fltm ,m-l. Some might further object that in this troublesome
regime the distance between particles becomes comparable to the deformations or
microasperities of the surfaces, which in itself may invalidate (3-15). This compli-
cation, which is a constant source of vexation in numerical simulations, has come
to be known in the literature as inelastic collapse or inelastic catastrophe [57]-
[59].14
It is possible to circumvent these difficulties by introducing the notion of block. IS
To that end, we introduce a threshold velocity V c (see Section 6.2.1) which marks
the separation between beads considered separate from those that are effectively
stuck together. When the velocities of two particles that have just collided differs
by less than vc , we treat them as if they had merged into a single block. In practice,
the velocity V c is chosen as small as possible (for instance, 10-7 m/s) while still
compatible with the precision of the computer. We may also verify that changing
the value of V c does not substantially affect the results of the calculated dynamic
behavior of the system. The key is to find solutions that remain stable with respect
to changes in V c ' It is, of course, imperative to conserve the center of mass of
a block during its subsequent movements. Since masses do not enter explicitly
into the type of calculations discussed here, they have to be handled through the
bias of the time-dependent position z(t). Having developed a criterion for block
formation, it is also important to consider the possibility that once they are formed,
blocks of several particles may disintegrate in part or in whole. We will defer until
Section 6.2.2 a more detailed discussion of this topic. Suffice it to say here that
different approaches produce results that are in substantial agreement. Figure 47
is offered as an example of an event in which particles clustered in two separate
blocks rearrange themselves in a different pattern following a collision between
the blocks treated in accordance with the so-called Largest-Relative-Velocity-or
LRV-criterion. 16
Results: Fluidized Phase and Condensed Phase
Numerical solutions of Newton's equations, obtained with the precautions dis-
cussed above, reveal several specific and generic behaviors that seem extendible
to two- and three-dimensional configurations. We proceed next to describe these
behaviors and to emphasize their differences, although the dividing line between
them is not always clear-cut.
14Needless to say, it is a catastrophe only from the point of view of developing a model. As we might
guess, the term was coined by people doing computer simulations.
15This is a common technique designed to get around problems of accumulation in numerical simula-
tions. We will encounter it again in Chapter 6.
16The LRV criterion will be discussed in detail in Section 6.2.2.
80 3. Fluidization, Decompaction, and Fragmentation
FIGURE 47. Behavior of five spheres initially grouped in two blocks of 3 and 2, respec-
tively. The horizontal axis represents time. The spheres rearrange themselves after the
collision. The simulation was done by applying the LRV criterion (see text) (after [59]).
17 Acollection of clustered particles presents a small global coefficient of elastic restitution even if each
individual member is nearly perfectly elastic. That is why a packed bag of marbles does not bounce
back when falling on a hard floor, even though a single marble does so spectacularly.
3.2 Dynamic Properties of a Granular Pile 81
(a)
(b)
FIGURE 48. Calculated trajectories of ten spheres with 8 p = 1 and 8 = 0.9. The system
is vibrated at 10 Hertz with two different accelerations ['. In (a), r = 8.0, and the system
is "fluidized." In (b), [' = 1.7, and the system is "condensed."
The following simple calculation is informative. Let t* be the time when the
normalized acceleration r becomes equal to 1. It is given by
At that moment, the object is launched up with an initial velocity v* equal to that
of the vibrating plate. All other parameters of the movement are straightforward
to calculate. We might note that, even if the coefficient of elastic restitution is not
strictly zero, the locking on the second harmonic (see Figure 49(b)) is particularly
stable. That is because the motions of the object and of the vibrating plate then
take place in the same direction, minimizing the velocity differences and reducing
(a)
(b)
FIGURE 49. Locking of the trajectory of a column of particles on the vibration of the plate.
Locking can occur on a number of frequencies, including (a) the fundamental, and (b) the
second harmonic.
82 3. Fluidization, Decompaction, and Fragmentation
the likelihood of rebound. 18 The same situation can exist, of course, at other
frequencies besides the second harmonic, including the fundamental.
Returning to the case of a column of beads, the considerations developed above
provide a plausible picture as to why a large-amplitude excitation can lead to
fluidization, as depicted in Figure 48(a). Even if we start with a situation in which
all the beads are clustered and form effectively a single block, enough energy can be
transmitted along the entire string to cause even the topmost bead to separate from
the block and initiate its own individual ballistic trajectory. 19 What can happen to
the topmost bead can a fortiori happen to any other bead in the stack as well.
Summarizing, a column of beads or other elastic granular material vibrated
vertically exhibits two distinct regimes depending on the acceleration imparted by
the bottom plate. For small accelerations, the system is in a "condensed" state,
with the beads practically in contact with each other and moving in unison. For
large accelerations, the system is in a "fluidized" state where the beads move
about individually much like particles in a gas or a fluid. In this latter state, the
acceleration Aw 2 is no longer the pertinent variable. Instead, that role is taken
over by the kinetic energy, proportional to A 2w 2 , or, equivalently, by the square
of the mean vibration velocity. The transition between the two regimes involves
portions of the system that are fluidized and others that are still condensed. All
of this has been satisfactorily simulated numerically and verified experimentally
[56], [60].
The existence of these two regimes and the transition from one to the other also
depend on the value of the coefficient of elastic restitution as well as on the number
of particles involved. This point is examined further in the next development.
Fluidization and Condensation as Functions ofHeight
As noted above in our discussion of a column of a sufficiently large number
of beads, the collision energy may be attenuated as it propagates up the string
of particles when the interactions are not purely elastic. If so, it may become
impossible to fluidize the upper portion of the system, which will then continue to
behave as a single compact block.
Computer simulations and theoretical calculations agree that the parameter con-
trolling this effect is a reduced variable defined by X = N (l - E:). The theory indi-
cates that there exists a critical value Xc for X such that Xc = If [37]. Simulations
suggest a somewhat less well-defined critical value Xc "'::: 3 [56].
The following calculation clarifies the order of magnitude concerning the re-
duced variable X. The coefficient E: is equal to 0.6 for aluminum and 0.92 for
18Tennis aficionados might appreciate that this very principle is exploited in the dreaded "drop shot,"
whose secret consists in deadening the rebound of the ball, with potentially devastating effects on
the opponent.
19That is the principle behind a toy sold in some specialty stores under the name "Newton's Cradle." It
consists of a linear array of a few metal balls individually suspended by threads. All the balls are in
contact when at rest and form a "block" in the sense defined in the text. When the rightmost ball is
pulled apart from the block and let go, it comes crashing into the remaining block, at which point the
leftmost ball is knocked away in the opposite direction. All the while, the inner balls remain still.
3.2 Dynamic Properties of a Granular Pile 83
hardened steel. We then find that the critical number of balls is 8 for aluminum
and 39 for steel. With very few exceptions, the granular materials typically han-
dled by the food or pharmaceutical industries have much a lower coefficient E;.
Therefore, in the vast majority of practical situations, the variable X is such that
X» 3, and the materials are most often in a condensed regime.
ob .J
3
:-<
t>
t>
-<
.....
..... 2
o t>
o
g~
Eo-<
t> t>
5
r 10
FIGURE 50. Bifurcation diagram of the time of occurrence of collisions. The product f x
Teall is plotted against the normalized acceleration ['. In the present case, N = 10, f = 30
Hertz, and sp = s = 0.6. Circles are experimental points, while triangles show the results
of numerical simulations. The diagram clearly shows locking on the fundamental, as well
as on the second and third harmonics (after [56]).
84 3. Fluidization, Decompaction, and Fragmentation
• The effective coefficient of elastic restitution is too low. This happens when
the particles are clustered and most of the collisional energy is dissipated in
a large number of impacts.
• The collisional energy is dissipated in the column because the collision pro-
cesses are inherently dissipative or because the column is tall. Either way,
the excitation cannot make its way to the top of the column.
• The ballistic flight of the column locks onto one of the harmonics of the
excitation period, in such a way that the velocities of the column and of the
vibrating plate are nearly synchronized. In this case, collisions are virtually
eliminated and no elastic energy is transmitted.
Z°It is the reason why water vapor condenses into separate droplets on a cold windshield, rather than
in a uniform layer. The resulting pattern is referred to as breath figures.
3.2 Dynamic Properties of a Granular Pile 85
Pz
15
z+
I
I
I_~~
X
FIGURE 51. Statistics of the trajectories of elastic and frictionless beads at the surface
of a two-dimensional pile. The photograph reveals that the lower part of the pile remains
compacted. The histograms show the distribution of velocities along the two principal axes
(after [61]).
Faraday instabilities in liquids and are currently the focus of active research. We
will briefly discuss them later in this chapter (see Section 3.2.5).
The experimental apparatus described above allows us to study the nature of
the fluidized top few layers of pellets which, as shown in Figure 51, seem to
follow complex ballistic trajectories above the compacted reservoir. This can be
done by visually recording the pellets with a CCD camera over a duration of
many thousands of excitation periods. A particularly useful technique is to use a
stroboscopic lighting system flashing pulses of light at regularly spaced instants
synchronized with multiples of the excitation period. The technique lends itself
to direct measurements of the instantaneous velocities of the fluidized particles.
Subsequent data manipulation generates the statistics of the velocities at different
times during one period. The results can then be plotted in the form of histograms.
Figure 51 includes such histograms, showing the probability distribution functions
of the horizontal and vertical components of the velocities centered about their
mean values (v x ) and (v z ). The two histograms have approximately the same
width, indicating that the medium behaves isotropically, in conformance with the
symmetry of the problem. As it turns out, this is one of the characteristics of a
fluid or gas in thermal equilibrium.
Generic Model
Once again we resort to the same notation we used in our discussion of Janssen's
model, particularly when we derived (3-9), in which A represents the surface area
of the cylinder's base and P its perimeter. However, for reasons of convenience,
we adopt a different convention for the heights h. From here on, they will be
referenced to the base (h = 0) and counted positively toward the top. The geometry
is illustrated in Figure 52.
88 3. Fluidization, Decompaction, and Fragmentation
A ho
Decompacted
dm
Compacted
p
Ih Ih'
tr
FIGURE 52. Gradual decompaction model of a cylindrical stack of a granular material.
Equation (3-16) enables us to define a threshold height h t below which the mate-
rial is not allowed to separate from the wall and must therefore remain compacted,
while it will undergo decompaction above hi. The equation defining hi can be
recast in the form
where h o is the total height of the stack. Combining the last two equations yields
an expression giving hi
(3-17)
hi In(2 - r) In(2 - r)
a=-=l+ =1+ . (3-18)
ho SoKfLs X
The rate of decompaction a turns out to depend only on the dimensionless pa-
rameter X (aside from the normalized acceleration r). This is the reason why we
chose earlier to give X the name decompaction parameter.
(3-19)
This equation predicts that the normalized acceleration required to induce levitation
of an infinitely high stack is precisely equal to 2.
1.0
i 0.8
~
0-
c:
o
0f$
<1\ 3
0-
E
o
o
21 Thetechnique makes use of medical imaging technology that has become relatively commonplace in
recent years. One drawback is that it requires particles lich in resonant radicals, most typically water.
Mustard or poppy seeds are the most popular choices of materials to work with. Unfortunately, their
mechanical properties are not very well defined.
3.2 Dynamic Properties of a Granular Pile 91
220ne technique is based on Hough's criterion. It relies on numerical processing to extract the positions
of particles from very noisy signals [64].
23This is a trickier problem than appears at first glance. For instance, it is often necessary to light up
the scene from the side, in which case a correction algorithm must be used to infer the actual position
of the centers of the particles.
92 3. Fluidization, Decompaction, and Fragmentation
Image I]
o I1
IImageK-
h 1 tHE
1 1
This table tells us that the resulting display would show not only the points that have
not moved between the two shots, but also the traces of those that did move. That
is how the images of convection reproduced in a later paragraph were produced.
With an "exclusive or" (or) operation, the image would show only the traces of
the particles that had moved, which may also be of interest.
The sequence of operations just described must obviously be repeated over
a sufficiently long time for something interesting to show up. In other words,
the technique is suitable to examine longer-term phenomena (compared to the
excitation period). This technique, known as "Computer Posed Photograph," or
CPP for short, was used to produce the images shown in Figure 54.
We now describe the experiment that was used to generate Figure 54. Roughly
50 x 50 aluminum beads are stacked in the form of a regular triangular lattice inside
a two-dimensional cell characterized by a shape factor So = 2h o/ L. The outer
surface of the beads is roughened by means of an appropriate surface treatment.
This is accomplished by shaking the beads for some time in air. The work hardening
resulting from the many collisions increases the coefficient offriction between balls
from an initial value of 0.2 to about 0.6. We recall that the coefficient of elastic
restitution s of aluminum balls is also about 0.6. We showed in Section 3.2.2 that
the reduced variable describing the dissipation of the collisional energy in such a
configuration is given by X z = Nz(l-s). Under the conditions of this experiment,
we are therefore definitely in a condensed regime since a column six balls high is
FIGURE 54. Computer posed photographs of a two-dimensional vibrated stacle The photo
on the right is a magnification of the region marked at the left by a circle. Convection rolls
generated by shearing at the wall are clearly evident (after [67]).
3.2 Dynamic Properties of a Granular Pile 93
• A stack is prepared with an initially fiat upper surface, and the vibration is
started with an acceleration [' > 1.
• Convection rolls are spontaneously initiated in the upper comers of the stack,
which transport granules residing in the top portion of the structure, giving
rise to two mounds which are quite evident in Figure 55(a).
(a)
(b)
r >1
Vibrations t
FIGURE 55. (a) Photograph of a "Chinese hat" in the process of developing. (b) A simple
experiment demonstrating the crucial role played by lateral walls in pile formation.
3.2 Dynamic Properties of a Granular Pile 95
A simple experiment can demonstrate the decisive role played by the lateral walls-
more specifically, the friction between particles and walls-in the mechanism of
pile formation. It is depicted in Figure 55(b), and goes as follows:
1. The experiment uses a special container made of two concentric glass cylin-
ders, the radii of which differ by slightly more than the diameter of oxidized
aluminum pellets that make up the granular medium. Filling the space be-
tween the two cylinders with those pellets creates a two-dimensional stack
that can move freely and lacks any lateral boundary.
2. When such a structure is subjected to a vertical vibration of amplitude such
that r > I, we observe that the upper free surface remains horizontal. There
is no evidence whatsoever of pile formation.
3. With the vibration uninterrupted, a cylindrical rod of thickness comparable
to the diameter of the granules can be inserted in the space between the two
concentric cylinders, where it acts as lateral boundaries for the stack. This
immediately causes the appearance of pile formation, similar to what was
observed in the cell shown in Figure 55(a). What is more, the moment the
stick is removed, the surface returns to horizontal.
4. If the same experiment is repeated with frictionless granules, for instance,
granules with a specular polish, no pile forms whether the rod is in or out.
(0)
FIGURE 56. Three-dimensional convection experiments. In configuration (a), the left wall
is polished smooth. Only the right wall, which has been roughened to intensify friction,
induces convection movements. In configuration (b), convection takes place in a direction
opposite to that which is observed in a cylindrical container (after [20]).
24The definition of threshold needs to be clarified. It is possible to observe slight movements of the
particles on the surface before convection is truly initiated. The ambiguous nature of this definition,
as well as other factors (such as degree of disorder, whether the stack had initially settled or not,
etc.), probably explain why these particular experiments gave a value of the acceleration r of about
1.2 at threshold, while subsequent work with two-dimensional structures gave a value of 1, within a
measurement uncertainty of ±O.05.
3.2 Dynamic Properties of a Granular Pile 97
..
Bead diameter
0 ... 0.2mm
.........
Q) 0 0 0.4 mm
"'0 0
~
::l 0 0 1.0 mm
c.. C\a
E
« Co
'--"
O'l
0
-1 '"
oOe
\
0
D
°8
,0
0
-2
1.0 1.2 1.4 1.6 1.8 2.0 2.2
log(m/2n)
FIGURE 57. Log-log plot of the vibration's amplitude threshold for three-dimensional pile
formation against frequency, for several bead sizes. The diagram shows that the acceleration
is indeed the determining factor, since the straight line has a slope of -2 (after [66]).
- x) = aexp (-cx) .
dxdt = Cexp (B-C-
25We will later on encounter another phenomenon with a logCt) dependence, for which we will propose
a more detailed explanation (see Section 4.2.1).
98 3. Fluidization, Decompaction, and Fragmentation
en
"0 25
ro
Q)
.0
.....0
r = 1.39
20
0
5 15
en
~
ro
Q)
0- 10
.....0
c:
0 5
:;:::;
'iii
0
a.. 0
1 2 3 4 5
10 10 10 10 10
Time (seconds)
FIGURE 58. Separation between each of the two mounds and the nearest wall plotted
against time for two accelerations. The experiment covers about 30 hours. Distances are
expressed in numbers of particles (after [67]).
The variable x is a measure of the distance from the summit of each of the two
mounds to the wall. Given the geometry of the system, it is also proportional to the
depth of the Chinese hat. In the above equation, the coefficient a, which describes
the efficiency of the pile formation process, turns out to depend on the acceleration
r through the quantity r - 1, and the coupling via friction between beads and
walls. 26 The coefficient C which figures in the exponential defines a characteristic
length in the experiment; it can be assimilated with the typical dimension of a
convection roll, involving the coupling between beads. In the absence of any solid
theoretical basis, it would be hazardous to venture any further interpretation of
what are after all purely phenomenological parameters. We simply conclude by
listing the experimental values of the parameters a and C, in units of number of
beads, for two values of the acceleration r.
r 1.15 1.39
a 3.9 16.8
C 1.0 2.7
26No pile formation occurs with frictionless balls or walls, or a fortiori if there are no walls at all,
as was the case in our earlier two-dimensional experiment. The same will prove true of gradnal
decompaction, to be discussed later on.
3.2 Dynamic Properties of a Granular Pile 99
1.0 '.
~::~I
t$ 0.8 " (e) O•
hh' :•..
.
0 0 .6 Decompacted t
a= hl/h o
~ 004
I.L 0.2 Compacted
I 10 So =0.67
0.0 -I-----_-----"~--o- __1
1.0 1.0 1.2 104 1.6
0.0 0.2 004 0.6 0.6 1.0 1.2
So
0.6 1.0
(b) (d)
EOA 0.8 '"
~
Decompacted
t$
0.6 '.
S
L..
0 So= 1.26
.c 0.2 t5 004
<1 ~.. 'Kf=011 CIl
I.L 0.2
Compacted
10
0.0
a 4 0.0 I
6
1.0 1.2 104 1.6
h (cm) Ie
FIGURE 59. Experimental results confirming the gradual decompaction model (after [67]).
See text.
• Figures 59(c) and 59(d) show the results of experiments intended to test
the relevance of the phase diagrams (a, r, X) discussed previously. The
experiments consisted in filling a two-dimensional cell with pellets of oxidized
aluminum, and placing the system on a vibrating plate whose acceleration
r = au} / g was adjustable. As predicted by the model, the experimental data
points line up with the theoretical curves corresponding to a value of the
product Kf (where f designates the friction coefficient) equal to 0.29. At
least that is the case for the two particular stack heights investigated. We
may recall that the triangular stack model described in Section 3.1.3 gave a
value K = 0.58. Since the friction coefficient fpw between pellet and wall
can vary from one cell to another between 0.2 and 0.5, we should indeed
expect the product Kf to range from 0.1 to 0.3 depending on the quality of
the surfaces of both the walls and the aluminum pellets. This turns out to be
quite consistent with experimental results. We should also note that, as the
pellets and the walls wear out during an experiment, the coefficient of friction
may actually change with time. This warrants some care in the interpretation
of the results .
• Figure 59(a) shows the lift-off acceleration rIo (equal to 2 - e- X ) as a function
of the height of a stack in a particular cell. This experiment is relatively easy
to do. It consists in measuring the acceleration required for the lower row
of the stack to lift off, which can be monitored with a CCD video camera
with a fairly high image magnification. In agreement with the model, the
entire lower row turns out to lift off in unison, providing an a posteriori
justification for simulating the stack as a superposition of sheets. In fairness,
100 3. Fluidization, Decompaction, and Fragmentation
FIGURE 60. Synopsis of results consistent with the gradual decompaction model. Three
configurations correspond to different combinations of the parameters K and /-L (after [97]).
we will see later that this behavior holds only in cells of relatively small
lateral dimensions. Section 3.2.5 will describe what happens in larger cells. As
predicted by the model, the acceleration needed for collective lift-off of a stack
is indeed a monotonically increasing function of its height. The data points fall
on a theoretical curve with a single adjustable parameter Kf = 0.11. These
results were obtained with a different cell than the one used in the previous
experiment.
• Figure 59(b) refers to a completely different experiment, although, this time,
it is done in the same cell as above. Stimulated by the vibrations, the pellets
can move relative to the cell. Provided that the particles leave a record of their
movements on the front and back windows of the cell, we can then calculate
the amplitudes of the displacements /')"h(h) as a function of height h. Here
again, there is excellent agreement between theory and experiment.
FIGURE 61. Stacking in a triangular lattice and its dual counterpart, rotated by 90°. The
configuration on the left is the only one that redirects stresses toward the walls.
the range of 1 to 1.5, which are typical of the experiments conducted previously
with conventional triangular lattices. Unlike the conventional case, though, and
in full agreement with the model, there is here neither gradual decompaction, nor
convection, nor pile formation. In other words, the meshed character of such
structures is indeed one of their fundamental properties explaining not only their
dilatancy properties but their resistance to decompaction under vibrations.
Short- Term Decompaction: Fragmentation
Although the model described in the previous paragraphs is based on the mechanics
of continuous media, it accounts fairly satisfactorily for the gradual decompaction
properties of granular stacks subjected to vibrations. Before proceeding any fur-
ther, it is essential to appreciate the deeper implications of this theory. The model
rests fundamentally on Janssen's hypotheses, supplemented by the dynamical law
expressed in (3-16), which was arrived at empirically. Our purpose is not to reex-
amine here the derivation of Janssen's equation (see Section 3.2.3), but we do need
to take time out to reflect on the nature of the approximations underlying (3-16),
which is rewritten here for convenience:
fg dm - g dm ::: dFtric!.
We will focus on two key points: (1) the assumption that the system can be de-
composed in uniform horizontal sheets; and (2) the simplified description of dry
friction .
• To begin with, we recall that the model assumes that the material is made
of horizontal sheets of mass dm, rubbing against the walls where a fric-
tional force d Ffrie! is developed. This force can be calculated with the help
of Janssen's hypothesis. Furthermore, (3-16) implies that the acceleration
fg imparted by the vibrating plate is homogeneous throughout any hori-
zontal slice of material. Assuming that acceleration, mass-and, therefore,
friction-are all uniform in a given sheet is manifestly in flagrant contradic-
tion with our earlier description of the network of contact points within a
granular material (see Section 1.2.2). The reader will recall that we invoked
102 3. Fluidization, Decompaction, and Fragmentation
arches anchored onto the lateral walls to describe the equilibrium forces in
such materials. The notion of arch is an essential and unavoidable component
of the static and dynamic properties of granulars. From this vantage point
alone, we can expect substantial deviations between theory and reality.
• The extremely simplified way the friction forces with the walls are treated
may have even more serious consequences. Without going into details that
will be discussed in greater depth in Section 6.4, we merely point out here that
(3-16) expresses nothing more than a simple rupture of the forces of contact at
the boundaries when a sheet experiences a sufficient force. This completely
neglects the indeterminacy of forces which was discussed in Section 3.1.1,
and assumes that these forces are fully mobilized and directed straight toward
the top, in accordance with Janssen's hypotheses at equilibrium. It is highly
likely that the random nature of the contact chains, for one thing, and the
more or less elastic coupling between particles involved in arch formation,
for another, will in fact lead to a rather large range of contact forces at the
walls. Invoking elementary uniform horizontal sheets to model a medium that
is inherently inhomogeneous and partly undetermined may seem fraught with
danger. Furthermore, (3-16) also ignores the coefficient of dynamic friction
and, perhaps more importantly, the velocity dependence of the friction force.
All these considerations may induce the skeptics to think that the agreement be-
tween theory and experiment is purely fortuitous. That conclusion would be
wrong, though, at least as long as the model is used with proper caution and within
its own limitations. It is appropriate to point out, for instance, that the simula-
tion of gradual decompaction merely establishes a phase diagram that predicts
the average height at which a stack undergoes decompaction, without specify-
ing anything further about what might be going on in the decompacted phase.
Except for the calculation of particle trajectories, which requires an additional
assumption to be discussed shortly, the model does not go beyond working out
the height of the compacted phase. The procedure usually takes into account the
irregular character of Coulomb's friction law, as we will discuss in more detail in
Section 6.4.
The experiments we have described thus far essentially deal with long-duration
observations. In actuality, the decompaction processes observed experimentally
proceed by way of successive relaxation states in which a system remains long
enough (typically several seconds) to be recorded by conventional techniques, such
as CPP photography. This approach misses the details of any fast event (lasting
no more than a fraction of the excitation period) involving short-duration and re-
versible distortions. While that much is clear from an experimental point of view,
what consequences does it entail in terms of the simplified model we have devel-
oped? In what fundamental way is it-in spite or because of its imperfections-
more suitable to account for long-duration phenomena? There is no clear answer
to this question. All we can do at this point is to simply acknowledge that the model
does seem to describe fairly accurately systems in their reorganized states, after
they have relaxed back to equilibrium. It does not address the largely indeterminate
3.2 Dynamic Properties of a Granular Pile 103
Fracture
Lines
FIGURE 62. Short-term observations, using a strobe light synchronized with the vibration
period, reveal that decompaction of a vibrated two-dimensional pile actually results from
a series of fragmentations propagating from top to bottom during ballistic flights. Such
fragmentations occur during each cycle of the excitation (after [69]).
solutions, subject to random fluctuations, that would result from a more complete
analysis of the problem.
Given the fundamentally dual nature-short-duration and long-duration-of
the relevant phenomena, it is naturally tempting to observe the time evolution of a
granular stack by imposing a slight shift in the frequency of the stroboscopic light
flashes relative to that of the excitation. Figure 62 shows the results of just such
an experiment. 27
The image contains a lot of information about the short-term behavior of a
granular structure, which deserves a number of comments. 28 First of all-and
that cannot be conveyed by a frozen snapshot-there are fractures, such as the
one indicated in the figure, that have dynamic properties of their own. A video
would show that these fractures are initiated near the top and propagate toward the
bottom. This mechanism of successive fragmentation occurs during each cycle
of the excitation and persists during the ballistic phase of the stack, which can be
quite short compared to the excitation period (typically in the ratio of 10 ms to
0.2 s). These fractures have a characteristic V-shaped form (in the present case
pointing toward the bottom) and are more or less dislocated as they progress within
27 This discussion does not respect the actual chronological sequence of events that led to this exper-
iment. In reality, it was while trying to measure directly with a video camera the movements of
particles in a granular stack that researchers came to realize that gradual decompaction resulted from
a series of more or less irreversible fragmentations.
28Note that CPP images and traces left on frontal windows can only provide information averaged over
large numbers of periods.
104 3. Fluidization, Decompaction, and Fragmentation
the structure. We shall see in the next section how we can extract more detailed
information about this phenomenon, which appears to be a very general mode of
decompaction of a guided stack in ballistic flight.
Two-Dimensional Experiment
Particles are stacked in a regular triangular arrangement in a vertical cell of the
type described previously. The bottom of the cell is blocked by a spring-loaded
metal plate that can be dropped rapidly (with an initial acceleration of 3g). A
CCD camera records the fall of the stack, which lasts approximately one-tenth of
a second. Figure 63 shows a time-lapse sequence obtained during the experiment.
As expected in light of the results presented in the preceding paragraph, the stack
breaks up into separate V-shaped blocks whose apex now points upward, always
in a direction opposite to gravity. 3D Also as noted before, successive fractures
propagate toward the top of the pile during the fall. All these observations are fully
consistent with those described in connection with a vibrated stack, provided that
the sign of the acceleration r imparted to the structure in ballistic flight be reversed.
While it was directed toward the top in the case of a vibrated cell (upward ballistic
flight prior to recompaction), it now points down. Given our previous discussion
of the formation arches (see Section 3.1.1), the V-shaped structures observed in the
present experiment can be understood as contact chains with enhanced strength
supporting the granular mass above them.
29Drake reported experiments involving inclined falls in which he was studying structural modifica-
tions by direct observation, although he did not draw the conclusions derived here concerning the
fragmentation process [68].
30 Similar observations apply to three-dimensional structures. If a small tube about I cm in diameter is
half filled with compacted sand and the bottom is suddenly opened, the content escapes by fragmenting
from the bottom up into a series of blocks separated by arches analogous to those shown in Figure 5.
3.2 Dynamic Properties of a Granular Pile 105
Theoretical Modeling
An interesting question is whether our previous theoretical model, which is a dy-
namic extension ofJanssen's model, can account for the process offragmentation. 31
We will later show how a properly adapted numerical simulation can indeed re-
produce the appearance and propagation of fractures within granular piles [69].
L [ 1 - exp ( --L-(h
PvCh) = pg 2KfLs 2KfLs - h o )] ,
- h o))] = pg~ [ 1 - exp (h-~-
31Savage studied the inclined or vertical fall of a noncohesive granular system [70]. Based as it
is on equations describing the mechanics of continuous media, Savage's model is essentially a
hydrodynamic description, while ours, which relies on the concept of fractures, is more akin to
models favored by geophysicists.
106 3. Fluidization, Decompaction, and Fragmentation
where, for convenience, the characteristic coefficients are lumped into a single
parameter C; = Lj(2Kfls).
A straightforward solution can be obtained, considering the stack as a whole
whose apparent weight has been given by equation 3.12. In this case and supposing
that the pressure and friction forces remain constant during the downfall process
of a pile whose height is h o, we get the reduced acceleration of the stack as
where X = ®. !;
An alternative approach would first consist in supposing that the pressure equi-
librium has been established before opening the spring-loaded metal plate. Second,
and because of the particular discontinuous nature of the material, we may con-
sider a thin slab of granulate and suppose that the previously mobilized friction
forces are still acting at the wall thereby reducing the downwards acceleration of
this particular piece of material. We know that the friction force developed at the
walls at this particular thin slab of granulate can be written as
- - = -Pv = g [ 1 -
dFfrict -- -
exp (h hO)]
dm pC; C;
h - ho )
f(h) = exp ( -C;- where h E [0, hoJ.
Strictly speaking, and in the context of the hypotheses discussed earlier, this equa-
tion is valid only at the precise instant when the downward fall is initiated. In the
interest of simplification, and for a semiquantitative analysis, we shall assume that
it holds for the entire duration of the fall. We will see later on that experiments
justify this approximation.
We begin by noting that r(h) is an increasing function of the height h. This
implies that a stack subject to its own weight and to the applicable forces of friction
will tend to remain compacted as it falls. In other words, the stack experiences no
spontaneous tendency to fracture, unless external causes come into play.
To identify external causes likely to initiate fractures that can propagate and
lead to fragmentation, we need to modify the surface properties of the lateral
walls. When we do so, we observe the following:
3.2 Dynamic Properties of a Granular Pile 107
I
I
Fractures
400 One particle
E
.s..9! 300
ho
ho=20 mm
"6. ho=44 mm
Q)
:S
'0 200 ho==' 93mm
Cl.
B
Q) h o= 130mm
:S
'0 100
1: ho==' 195mm
0>
"iii
J:
0
FIGURE 64. Altitude at the top of the pile as a function of fall time. The filled symbols are
experimental data. The solid curves correspond to parabolas calculated with KfLs = 0.12.
The arrows indicate the appearance of at least one fracture in the pile (after [69]).
• Surfaces prepared with a specular polish, that is to say, with a roughness of the
order of a fraction of a micron, have a coefficient of friction fLs that is virtually
the same as for regular surfaces after they have gone through normal wear-
out. With polished lateral surfaces, stacks can fall without fragmentation,
although with a reduced acceleration compared to that due to gravity alone, in
accordance with (3-20). These conclusions emerge from experiments whose
results are shown in Figure 64. They were obtained with identical oxidized
aluminum pellets placed in the same cell, but stacked to different heights.
The data indicate that within experimental errors, and as long as no fracture
appears, the stack falls with a normalized acceleration that is governed by the
initial height and is adequately described by our equation for r(h). The solid
curves in Figure 64 were calculated using a single adjustable parameter KfLs
of 0.12, which is somewhat smaller than the value determined previously. The
discrepancy is probably due to the fact that what matters here is the dynamic,
rather than static, coefficient of friction.
• Lateral surfaces with a roughness of a few microns or more trigger the ap-
parently random appearance of a series of ascending fractures, as noted pre-
viously.
appears at height h f measured, as usual, from the base of the pile. The fracture
causes an interruption in the contact chain and a reorganization (assumed to be
instantaneous) of the equilibrium forces in what is now two daughter piles-pile
A on top and pile B at the bottom. Each daughter pile is from now on subjected
to its own acceleration r A and r B .
Since r is a monotonic function of height, it is immediately apparent that a
fracture will remain stable or can even get amplified as long as r A ~ r B, which
implies the condition h f .:::; h al2. Stated in words, the fracture must be initiated
in the lower half of the stack if it is not to close up on itself during the fall.
We next examine how an ab initio simulation can reproduce many of the results
we have just discussed.
32The introduction of an artificial thennal agitation to simulate, through the bias of a dynamic numerical
technique, the static properties of a stack hides a profound reality. As we shall see in Section 6.4, the
dynamic properties of a stack are less subject to indeterminacies (in the sense of Section 3.1.1) than
are its equilibrium characteristics. Forces are mobilized and well defined at all times during multiple
collisions, whereas unknowns persist at rest. Furthennore, thennal agitation implies a fictitious
source of fluctuations which has no basis in reality. This touches on a fundamental problem in
matters of simulation and understanding of the behavior of granular materials.
3.2 Dynamic Properties of a Granular Pile 109
FIGURE 65. Simulation of the fracturing of a stack in guided fall under conditions similar
to the experiments. Note that thermal agitation, required by the simulation, causes a kind of
artificial boiling effect at the top of the pile. The time t is recorded in seconds. The aspect
ratio and micromechanical coefficients match the experimental values (after [69]).
Having recognized that simulations must take into account the rotation of particles
if they are to produce results in agreement with reality, we still have to understand
why that is. The best strategy to that end is to rely on experiments. That is precisely
the course of action we will follow. We will begin by considering the distribution
33 A cantilevered fracture occurs, for instance, when we attempt to cut a solid object with a knife. A
fracture often opens up at the point of contact of the knife and propagates according to a dynamics
familiar in geophysics.
110 3. Fluidization, Decompaction, and Fragmentation
0.25
t=0.02 s -+-
t=0.04 s .......
t
~ t=O.06 S .". •..
~ 0.20 Ijl
t=0.08 S ..·H....·
c
:::J
~
~ 0.15
:e
~
[1! 0.10
:::J
<Jl
<Jl
[1!
Q.. 0.05
0.00
-0.05 0.00 0.05 0.10
z(m)
FIGURE 66. Numerical simulation of the pressure against the walls as a function of height.
The integration time is 10 ms, and each point is obtained by averaging over a height
equivalent to six rows (after [69]).
of stresses in a stack in guided fall, and go on from there to examine in some detail
the modes of self-organization of the rotation of a collection of particles in the
immediate vicinity of a fracture.
Distribution of Pressure in a Stack: Arch Effects
Numerical simulations allow us to keep track of the velocities of the particles and
the rate of collisions between themselves as well as with the walls at any point
in space and time. As we have emphasized earlier, we are dealing here with a
dynamic model. As such, we are not in a position to calculate a static pressure.
Rather, we consider that the pressure exerted by the particles results from repeated
collisions and transfer of momenta to the lateral walls such as described by (2-10),
in which the normal component is conserved. This approach is entirely consistent
with the traditional kinetic theory of gases. The time integration of the momentum
I~"t P dt is carried out over a reasonable duration of the order of one-hundredth of
a second. The results of such calculations, obtained for a series of different times,
are shown in Figure 66.
The appearance offractures within the stack (at 0.04 and 0.06 s) corresponds to
a significant increase-by one order of magnitude-of the local pressure on the
walls. This is very much in keeping with our intuitive notion of an arch whose
primary function, as we have seen in Section 3.1.1, is precisely to transfer the
pressure exerted by the upper portions of the stack laterally to the walls. Thus,
through the bias of a numerical simulation, we arrive at some objective and quan-
tifiable information about the mechanism of fragmentation. It appears to result
from the successive formation and disappearance of triangularly shaped arches
pointed upward, as we may have anticipated.
It is likewise possible to study in detail what causes fractures to form in the stack.
The model suggests that a fracture opens up with a single particle whose translation
and rotation velocities (see Section 2.2.2) suddenly have to adjust to the relative
3.2 Dynamic Properties of a Granular Pile III
(b)
I""-~
I
I
(c)
Impossible Configuration
FIGURE 67. Diagram (a) shows a simulation of rotations in a stack in guided fall. The
rotations organize themselves around fractures indicated by arrows. Diagram (b) shows an
organization compatible with V-shaped contact chains of the type observed experimentally.
The arrangement depicted in diagram (c) is incompatible with the arch model and is in fact
never observed in simulations.
speed of the wall. Stated differently, unlike its cousins in the interior of the stack, a
particle interacting with a wall goes into a regime of rolling without gliding, in the
sense defined in Section 2.2.2, which triggers the onset of a fracture. Furthermore,
the model also favors the propagation of a fracture toward the interior along easy
dislocation lines of the lattice. 34 This propagation results in a dramatic increase
in the number of collisions between particles along and around the contact chain.
It would then appear that the mechanism responsible for initiating a fracture is
a kind of organization of the rotations of particles in relation to the wall. Given
the concomitant increase in the number of collisions in the immediate vicinity
of the event, it is logical to inquire whether such a local organization can in fact
propagate along a contact chain. That is our next topic.
Self-Organization of Rotations
Figure 67 depicts a map of the "spins" of particles in a stack right after the mo-
ment two fractures, indicated by arrows, have opened up at the wall. 35 White circles
indicate a counterclockwise rotation with an axis perpendicular to the plane of the
figure, while black circles correspond to a rotation in the opposite direction.
34We use here the standard terminology in crystallography. Given the triangular symmetry of a compact
two-dimensional stack, the dislocation lines run horizontally or at a 60° angle with the horizontal.
35 In the present context, the word "spin" designates the rotation of a particle with an angular momentum
perpendicular to the plane of the two-dimensional stack.
112 3. Fluidization, Decompaction, and Fragmentation
Figure 67(a), obtained by numerical simulation, shows clearly that the spins
tend to organize themselves in an alternating pattern in the immediate vicinity of
and just above the fractures. Figure 67(b) and (c) may help us understand that the
angle of these patterns is compatible with the notion of arch developed earlier in our
discussion. A contact chain inclined at a 60° angle relative to the horizontal implies
that the rotations of its strongly bound particles alternate. It is intuitively obvious
that the particles forming such a contact chain, being in intimate contact with each
other, will experience frustrated rotation relative to the chains immediately above
and below (see Section 2.2.1). It follows that the weakest points of the structure
run between superposed contact chains-parallel to the direction of the arch-and
that fractures can develop only along those lines. Here again, we see that the
self-organization of the rotations of particles conforms to the notion of arches as
contact chains which we developed earlier. 36
36In this picture, the formation of arches and fractures is a consequence of a process of self-organization
of rotations that gives rise to lines of easy fracture (particles in frustrated rotation) on the one hand,
and lines of energetic contacts (particles with alternating rotations). Extrapolating this picture, we
might even envision that an initially amorphous granular material may "crystallize" in sheets that are
easy to peel apart but are quite strong in their own plane. Given our present state of knowledge, this
possibility remains purely speculative.
3.2 Dynamic Properties of a Granular Pile 113
(a) (b)
A
~ Vibrations
Vibrations ~
FIGURE 68. Schematic diagrams of experiments to observe instabilities in thin granular
layers under vibration. Patterns are monitored from the top in three-dimensional configu-
rations, and sideways in two dimensions.
Ph 4h
S == - = - Rj 0.01.
A nd
Remarkably, in this three-dimensional situation, large changes in the coefficient
of restitution (from 0.5 to 0.95), of the density (from 2.3 to 11.4), of the aspect
ratio, and of the number of stacked layers, seem to have virtually no influence on
the phenomena we are about to describe. This is not true of a two-dimensional
configuration, however. In that case, we will see that the number of layers has a
pronounced effect.
Figure 68(b) illustrates a typical two-dimensional experiment. Once again, the
cell is made of two glass plates separated by a distance slightly greater than the
diameter of the particles, which in this case are 1.5-mm aluminum beads. The lat-
eral dimension of the cell is 300 mm, and it contains a number Nh of layers
varying from 4 to 27, which corresponds to aspect ratios of a few percent. Given
the much larger size of the beads, it is not essential this time around to work in
vacuum, and electrostatic interactions are all but negligible. We may even use
114 3. Fluidization, Decompaction, and Fragmentation
37There is a deeper significance attached to this observation. Our earlier decompaction model es-
tablished that in a vibrated granular material with a large aspect ratio, the behavior of the stack
was determined by the friction between particles and walls. Even though the present experiment is
fundamentally different, we nevertheless find again that friction between particles does not seem to
influence the overall behavior of thin stacks under vibrations. We might add that no one understand
clearly why that is.
38Faraday was studying the behavior of viscous liquids by subjecting their container to periodic vertical
vibrations. He observed the development of surface waves rather similar to the ones described here.
He could not, of course, have seen bifurcation phenomena, which are characteristic of inelastic stacks
(see Section 3.2.1).
3.2 Dynamic Properties of a Granular Pile ll5
3
Disorder
Hexagons
.....
..c
~ 2
'5
Ql
E
i=
Flat
o
2 3 4 5 6 7 8 9
Acceleration r
FIGURE 70. Bifurcation diagram of inelastic particles and organization patterns of the
surface of an extended three-dimensional granular medium (after [74]).
The formation of these geometrical patterns results from the combined effects
of two phenomena, at least one of which is specific to granular matter:
39Considering how thin the layers are in this experiment, we might legitimately be concerned that the
condition N" (1- 8) :c: 3, required to ensure that the stack will not be fully fluidized, may be violated.
The reader may convince himself that eight balls with a coefficient of restitution of 0.6 are sufficient
for the condition to be verified. As it turns out, compliance with the above condition is not crucial
for this particular experiment.
116 3. Fluidization, Decompaction, and Fragmentation
30
o
.000 .002 .004 .006
2
1/f (sec 2 )
FIGURE 71. Wavelength of organization patterns plotted against the inverse of the square of
the excitation frequency. Both straight lines have a slope of 1, indicating a linear dependence
for spheres of diameter 0.4 mm (square data points) or 0.2 mm (triangular data points). Only
the ordinate at the origin depends on the size of the spheres. In this case, r = 3.5 (after
[74]).
and 100 Hertz, these experiments confirm that it is indeed the parameter r that
governs the appearance of self-organized surface patterns, as well as their contrast,
validating the model of an inelastic ball.
where Amin depends only on the diameter d of the particles and is approximately
equal to lId. A similar equation appears in the book Fluid Mechanics by Landau
and Lifshitz, in a paragraph dealing with gravitational surface waves. 40 In this
light, the parameter geff can be construed as an effective acceleration which must
be a fraction of gravity's acceleration. That is indeed the case here, since we find
2
geff ~ 3.1 m/s .
40See page 36 of L. D. Landau and E. M. Lifshitz, Fluid Mechanics (Oxford: Pergamon Press, 1987).
3.2 Dynamic Properties of a Granular Pile 117
(a)
At the present time, the reason for the existence of two distinct regimes remains
unknown, as does the -JNh dependence at low frequencies.
Summary
The self-organization of surface patterns in extended two- or three-dimensional
granular media result from the superposition of two well-known phenomena. One
118 3. Fluidization, Decompaction, and Fragmentation
is related to the bifurcation diagram associated with an inelastic ball, and the
other is Faraday's instabilities in liquids. The latter connection is intriguing and
suggests the possibility of making a parallel between the viscosity of a liquid and
the degree of looseness of a granular sheet. Experiments indicate that a loose
granular medium indeed has a lower apparent viscosity (in the sense defined for
liquids) than a compacted one (see, for instance, Section 2.4.2). Intriguing as it
may be, this type of analogy remains largely in the realm of speculation.
4
Granular Media in a State of Flow
FIGURE 73. The figure on the left shows the embankment angle of a two-dimensional pile.
The one on the right depicts one of many ways to create a cone-shaped sand pile. Grains
are dropped one at a time on a roughened horizontal base.
of angles in the plural, for it turns out that there is more than one, as we shall see
shortly. The phenomenon is illustrated in Figure 73.
The first individual to make quantitative observations on the angle of repose was
Charles de Coulomb, who back in the eighteenth century was a military engineer
responsible for building fortifications [11]. Being well acquainted with friction
between solids, as mentioned in Section 2.2.1, he proposed a simple explanation
that is still considered the authoritative word on the topic. His model is rooted in
the idea that two contiguous sheets of a dry granular material cannot slide relative
to each other unless their inclination e is at least equal to tan-j(ILs)' By analogy
with what we know about friction between solids, ILs is to be interpreted as a
coefficient characteristic of the friction forces involved.
This analogy suggests that Coulomb's friction law applies to granular media as
well, which is indeed supported by experimental observations, as pointed out in
Section 2.4.1. Yet, for all its simplicity, this line of thinking raises a number of
questions on closer examination. For starters, how do we define the "weight" of a
granular sheet? We have already emphasized repeatedly that the distribution of
forces exerted on the surface of such a sheet is anything but uniform and a far
cry from the simple description applicable to a massive solid placed on a support.
Based on the arguments we invoked earlier concerning the microscopic mecha-
nisms of solid friction, we can easily appreciate that the problem is even thornier in
the case of a granular sheet. The many indeterminacies and hysteresis phenomena
alluded to earlier all contribute to complicating the task of defining unambiguously
what we mean by angle of repose. Indeed, both experiments and theory show that
there are not one but several angles of repose depending, among other things, on
how a pile was prepared.
A wealth of technical details on this topic can be found in the book by Brown
and Richards [5]. We will simply summarize them here by stating that the relative
indeterminacy of the angle of repose stems from two main factors:
4.1 A Sand Pile in Equilibrium: The Angle of Repose 121
Concave Convex
FIGURE 74. A convex pile generally has a lower embankment angle than a concave file,
such as exists near the opening of an hourglass .
• The first factor is geometrical, as it involves the shape or, more precisely, the
curvature of a pile. The point is illustrated in Figure 74, which may help us
realize intuitively that particles in the vicinity of the free surface tend to be
more densely surrounded in a concave pile (in the form of a crater) than in a
convex one (shaped like a mound).
Any difference is expected to vanish when the radius of curvature of the
surface becomes much larger than the mean diameter of the granules. These
predictions are essentially confirmed by experimental results such as those
reproduced in the following table:
FIGURE 75. Definitions of the angle of movement em and the angle of repose er . The
difference.5 = em - er is called the relaxation angle. Its value is typically about 2°.
FIGURE 76. In a pile made of a small number of particles, the relaxation angle 8 goes to
zero.
concerns the transition from a regime of intermittent flow to one that is continuous.
We go on to examine these two problems.
r-3~3'
given by
Nmin = ~ (~
We have pointed out earlier that (, is typically equal to 2° for real piles involving
a large number of particles. N min is then approximately equal to 8000 granules.
These considerations show that in matters related to the embankment angle, it is
important to keep track of the number of particles constituting the pile.
(a) (b)
FIGURE 77. Profile of the free surface of a pile in a slowly rotating drum. (a) The surface
is essentially straight when the rotation speed is just beyond the threshold for continuous
flow. (b) At higher speeds, we witness the appearance of a surge wave, as the influence of
the boundaries at the top and bottom of the flowing layer become noticeable.
Intermittent Continuous
flux " flux
Q Q+
- - - - - + - 1- - - - + - 1~----> Q
FIGURE 78. The hysteresis effect between intermittent and continuous flow. The arrows
indicate whether the rotation speed of the drum is being increased or decreased.
4.1 A Sand Pile in Equilibrium: The Angle of Repose 125
same in the intermittent and in the continuous regimes, for which the fall times are
designated f) and f2, respectively. The transition from intermittent to continuous
flow occurs when the fall time of a granule becomes synchronized with the interval
T separating two successive avalanches. As we will see shortly, T is actually a
random variable. Nevertheless, its fluctuations are small enough for the argument
to hold. In particular, the regimes switch over when Q+ = 8/f) and Q_ = 8/f2,
with f) < f2.
While the angle of movement 8m is clearly a fundamental property of sand and
evidently reflects a critical phenomenon, things are not so clear when it comes
to the angle of repose 8r after the system has relaxed. Even for large-sized piles,
the angle of repose is generally affected by the fact that the flowing sheet has to
come to a stop at the bottom of the drum. Such finite-size effects are less of a
nuisance than in the case of small-sized piles, but they are real nonetheless. The
influence of the wall must be taken into account if we are to develop an accurate
description of a phenomenon that turns out to be far more complex than it appears
at first.
Since we are dealing with a critical phenomenon, it is logical to want to examine
how the flux of material varies as a function of the inclination angle 8. ) In particular,
the question is whether this particular phenomenon might fit in with other critical
transitions by obeying a law of the form
(4-1)
(4-2)
where L is the length of the drum, and R its radius. Note that J is expressed as a
volume per unit time.
The results shown in Figure 79 correspond to a 19-cm-Iong drum half-filled
with particles of diameter 0.3 mm. Practical considerations impose that the rota-
tion speed Q be restricted to a range between 0.5 and 12 rpm. The lower limit
corresponds to the onset of continuous flow, while at the upper end a substantial
fraction of the flowing sheet is sent flying off due to the centrifugal force. The
experiment consists simply in measuring the inclination angle of the flowing sheet
as a function of rotation speed of the drum. Since we are looking for a power law,
the results are plotted in a log-log format. The graph shows that in the range of
variables investigated there is indeed a power law of the type J ex (8 - 8e n , with r
m = 0.5 ± 0.1. It is interesting to look for the physical significance of the expo-
nent m. Some useful insight can be obtained by reviewing the case, well known in
hydrodynamics, of an ordinary flowing liquid (Brownian flow).
lThe angle e, called dynamic angle, defines one of the characteristics of avalanches. It obviously
depends on the speed of rotation. Under certain conditions, it also depends on the size of the granules.
We will see an application of this poorly understood property in Section 5.4.
126 4. Granular Media in a State of Flow
Sc-S (deg)
20 0
0
0
10
f*
5
Y
4
3
f
2
t
2 3 4 5 10
.n (rpm)
FIGURE 79. Log-log plot of the angular deviation from the critical value versus transport
flux during avalanches in the continuous regime (after [75]).
pgh 3
J = - - sinCe),
31]
where p is the density of the fluid.
In other words, for an ordinary liquid, which happens to have a critical angle
ec = 0, we find an exponent m = 1. The phenomenological explanation for this
result is that the Brownian particles making up the fluid have an instantaneous
velocity far exceeding the net mean speed of the flow. If so, the viscosity 1] reflects
the loss of momentum resulting from multiple collisions. In this context, what can
be said of a granular flow?
(1) The rate of collisions between particles is proportional to the velocity gradi-
ent V v. What this means is that the greater the difference in velocity between
adjacent sheets, the more frequent the collisions between particles.
(2) The loss of momentum upon each collision is also proportional to Vv.
In other words, the greater the relative speed between two particles, the
more dissipation upon impact. We might note that this is hardly consistent
with the elementary picture of frontal collisions described in Section 2.2.2.
4.2 Avalanche Models 127
In any event, according to this model, the friction force resulting from a velocity
gradient V v is proportional to (V v)2. We can thus write the equilibrium between
the friction force and the force of motion as
av)2
-ex ( az + pgz[sin(B) - JLcos(B)] = 0, (4-3)
2
v(z) = '3
This explains the J ex (B -Be) 1(2 dependence suggested by de Gennes and observed
experimentally. Based on this elementary calculation, Bagnold's law in (VV)2 and
a simple law of dry friction are all that is needed to account for the exponent !
characterizing the behavior of a granular flow near the critical point.
2The phrase "system on the verge of avalanching" refers to a situation in which the free surface of a
granular makes an angle e between em and er with respect to the horizontal.
128 4. Granular Media in a State of Flow
ill
FIGURE 80. Principle of the one-dimensional cellular automaton model (CAM) (after
[77]).
model that establishes a relation between the slope of a pile and the flux of parti-
cles in motion. Finally, we will examine a model based on coupled variables that
accounts satisfactorily for a number of characteristics of avalanches.
The Principle
Squares are stacked on top of each other to form contiguous columns according
to a set of extremely simple rules:
(l) The height difference between two adjacent columns cannot be greater than
two units. This effectively simulates the angle of repose, which cannot
exceed a critical value without collapse.
(2) When a column rearranges itself because of excessive height relative to its
neighbors, it involves the movement of a set of two unit cells. This is akin
to the domino effect in an avalanche.
3In this chapter, the term "self-organized criticality" (SOC) applies to a system evolving spontaneously
toward a critical state with no memory of the initial conditions. In other words, it is a system for which
the critical state is an aUraetor as far as its dynamic properties are concerned. The interested reader
is referred to the abundant literature on this topic. There has been much debate about whether this
model is at all pertinent to avalanches. Suffice to say that caution is strongly advised, if for no other
reason than because of the existence of two different angles em and er , rather than a single one.
4.2 Avalanche Models 129
same rules. Since the bottom surface is assumed to be of finite size, squares can-
not accumulate there but, instead, drop off the base. The objective of this thought
experiment is to count the number of squares that drop off after each new square
is added to the pile. The release of one square can trigger mini "avalanches" of 0,
2, 4, and so on, squares escaping off the side. As it turns out, small avalanches are
far more frequent than larger ones.
This elementary CAM process, described here in one dimension, can be gen-
eralized to the case of 2, 3, or more, dimensions by developing suitable computer
algorithms. As artificial as such generalizations may seem in terms of describ-
ing real avalanches, they are of considerable interest because theories can often
predict the value of the exponent in the power laws applicable to any dimension,
even higher than 3. Exercising such algorithms often requires the use of parallel
computing so as to cut down on the computation time and minimize round-off
errors which are the bane of conventional sequential machines.
As an example, we can go through the algorithm describing the sequence of steps
in the one-dimensional case discussed above. The difference in height between
adjacent columns is defined as Zn = hen) - hen + 1). When a single square is
added to the nth column, it entails the following changes:
Zn ---J> Zn +1
and
Whenever the height difference becomes greater than a critical value Zc, the system
relaxes back in such a way that
and
Zn±l ---J> Zn±l - 1
The pile is bounded on the left (corresponding to the index 0), and open on the right
(index N), where squares drop off the pile. Therefore, events at the boundaries are
written as
Zo = 0,
ZN ---J> ZN -1,
ZN-l ---J> ZN-l +1 for Zn > Zc.
It is easy to count the number of stable configurations of a pile made of N columns
by writing the stability condition Zn < Zc, where n = 1,2,3, ... , N. This
gives a total number Z;: of stable configurations. They are not all equally stable,
however. A simple technique to arrive at the least stable of these configurations is to
artificially construct a system in which all columns are unstable (meaning that Zn >
Zc for all values of n), and let it relax spontaneously to some equilibrium which
becomes the fresh starting point from which new squares are then added. Such a
minimum-stability configuration is a critical state in the sense that any subsequent
130 4. Granular Media in a State of Flow
disturbance can propagate through the entire pile. This situation is, therefore,
analogous to the well-known problem of percolation in one dimension. The issue
of stability is considerably more complicated in structures with dimensions greater
than 1. It is decidedly beyond the scope of this book, but we refer the interested
reader to the many articles on this topic published in the wake of the original BTW
paper. Extrapolating the algorithm itself, on the other hand, to more than one
dimension is straightforward. We will do it here for the two-dimensional case. For
a square pile described by Z (x, y), the equivalent of the chain of events considered
previously becomes
Z(x - 1, y) ~ Z(x - 1, y) - 1,
Z(x, y - 1) ~ Z(x, y - 1) - 1,
Z(x, y) ~ Z(x, y) + 2.
If the height difference exceeds a critical value Zc, we have
Z(x, y) ~ Z(x, y) - 4,
Z(x, y ± 1) ~ Z(x, y ± 1) + 1,
Z(x ± 1, y) ~ Z(x ± 1, y) + 1 for Z(x, y) > Zc.
where the exponent T turns out to be approximately equal to 1.0 in two dimensions,
and 1.37 in three dimensions.
It is also of interest to examine the probability distribution function of the
durations (or lifetimes) of these avalanches. In other words, given a system in a
critical state on which an additional square is dropped randomly, how long will the
ensuing slide last? Intuitively, we expect some degree of correlation between the
size of a slide and its lifetime. An event triggering a large number of cells to fall
down the side ought to take longer than another one triggering just a few. Although
it takes a little more work to prove it than in the case of the size, "experiments"
4.2 Avalanche Models 131
show that lifetimes do in fact obey similar power laws, expressed in the form
where ex is found to be 0.43 and 0.92 in two and three dimensions, respectively.
(4-5)
4A good-quality hi-fi amplifier, for instance, transmits all frequencies between 20 Hz and 20 KHz
without distortions. We might wonder what the frequency spectrnm of the output noise of such an
amplifier might be. Under ideal conditions (with all other noise sources eliminated), the noise turns
out to have a II! spectrum. This remarkable property is due to the shot noise associated with the
input impedance of the amplifier.
iiiii
132 4. Granular Media in a State of Flow
p'EJwer
Spectrum
-----------
11 fO
Low-correlation shot noise ~
rS229 ~
I~~~~~~~'''~rn Time I log f
FIGURE 81. Three types of commonly encountered noise, and their power spectral densi-
ties (f0, r 1, and r 2) (after [78]).
of attention on the part of researchers. This type of noise is one of the characteristics
of systems that enjoy self-similarity, also known as fractal objects. The reason is
that the noise power in a small frequency window df is given by S(f) df; when
S(f) = Ilf, this noise power becomes df/f. In other words, the noise power
remains invariant provided the bandwidth df is scaled to match the mean frequency
f [79]. That is one of the fundamental properties of so-called self-similar systems,
which exhibit the same properties on all scales.
As we noted in connection with (4-5), the cellular automaton model (CAM)
leads to a frequency probability distribution function very close to a 1If law. This
immediately suggests the possibility that the behavior of such automatons may be
assimilated with that of a self-similar system. As we are about to see, this turns
out to be a complicated issue whose answer is not black and white.
-I r -I r f7
Scale Condenser Condenser Microphone
FIGURE 82. Four techniques used to study the statistics of avalanches (after [20]).
(1) Many avalanches, particularly small ones, never make it to the bottom of the
slope and, as such, never get recorded. The reason is that matter accumulates
at the bottom of the structure, which tends to diminish the angle the free
134 4. Granular Media in a State of Flow
~;~J 0 1 2
t (ks)
2.------.----r----,.-------,
{j) 0
C;
..Q
-2
(b)
-4
-3 -2 -1 0
lag(r)
FIGURE 83. Statistics of avalanches observed in a rotating drum. Diagram (a) shows the
number of particles transported as a function of time as the drum rotates at a constant speed
of 1.3 degree/min. Diagram (b) shows the power spectral density. The dotted line indicates
a 1/f dependence. The granular material is made of particles 0.5 mm in diameter (after
[80]).
surface makes with the horizontal, thereby dropping it below the critical
value, with drastic repercussions for the physics of the process. This is a
classic example of a finite-size effect. It acts somewhat like a high-pass
filter, as it were, favoring larger events.
(2) As noted above, the splitting of critical angles into two values-angle of
movement and angle of repose-implies that a system that has just experi-
enced an avalanche is no longer in a critical state. It is necessary to incline
it by an additional angle em - er in order to reach a new critical state. It is
then not obvious that real avalanches involving many particles truly con-
stitute critical systems, even if they can still be considered self-organized.
In light of some of our previous observations, however, it seem likely that
small avalanches may not suffer from these restrictions. To some extent, the
experiments we are about to describe support this view.
10 100
Avalanche size (number of particles)
FIGURE 84. Results of experiments on limited-size piles made of different particles. The
fractional rate of avalanche occurrence (i.e., the number of avalanches of size s normalized
to the total number) is plotted vertically against the number of particles s involved in the
corresponding event. In (a), the triangles correspond to steel balls, and the circles to glass
beads. In (b) the squares refer to polystyrene beads and the circles to glass beads (after
[78]).
pile supported in the pan of a balance. The balance keeps track of the weight of
material falling off the pan during each avalanche event. We may expect that the
shape of the pile should playa fairly important role, since a convex pile is inherently
two-dimensional, in accordance with the model developed in Section 4.2.1.
In a configuration of this type, the sizes of successive avalanches do exhibit a
considerable dispersion [78], [81]. Small-size avalanches are plentiful, while large
ones are relatively rare. Figure 84 also reveals that, regardless of the nature of
the materials (steel, glass, or polystyrene), the statistics of avalanche sizes indeed
conforms to a power law, as predicted by the cellular automaton model. It may be
objected, however, that the agreement has been demonstrated over not much more
than one decade, which some may deem too small a dynamic range to truly test
the validity of the law.
Although several experiments conducted under different conditions have all
confirmed that the behavior of a granular pile differs according to whether it is
made of a large or a small number of particles, the reasons for the crossover
from one regime to another are still not fully elucidated. It is possible that a
more sophisticated cellular automaton model may be able to account for such
finite-size effects. Yet, the existence of the two angles em and en as well as the
relaxation oscillations between these two extreme states, constitute a fundamental
objection difficult to overcome when modeling a dry granular material. A number
136 4. Granular Media in a State of Flow
.
............................. 24
21
.... I/A ..... .o\l,
~..
....
~ ..
'2 "........
o _, 18
.... .....
1
~
g; 15
~ 12
ff
Vibrations -3 -2 .. , 0
.9u =26.1°
.. 8..=22.2°
loglOCf(sec- 1)J
.9..=19.5°
.9..=16.4°
FIGURE 85. Results of experiments done with a rotating drum subjected to vertical vi-
brations. The angle ess, which corresponds to the steady-state angle obtained for a rotation
speed of 1.3 degree/min, is a measure of the excitation intensity. The dotted line in (b)
indicates the 1/f behavior predicted theoretically. In (c), Q = 0, and the structure relaxes
according to a log(t) behavior, whereas the CAM model predicts a relaxation proportional
to t (after [80]).
4.2 Avalanche Models 137
in the static case. The question we seek to answer is whether vertical vibrations
can put a pile in a supercritical state characterized by a single critical angle, such
that the power law predicted by the cellular automaton model is observed. Based
on the curves shown in Figure 85(b), the answer is decidedly no. Even when the
avalanche angle, which started out at 39°, is lowered by more than 2°-enough
to cancel the difference 8-the power spectrum of the avalanches continues to
exhibit a peak, which is inconsistent with the SOC model. As the critical angle is
lowered by even greater amounts, the power spectrum begins to look more like a
power law, but with an exponent that deviates substantially from the 1If behavior
expected theoretically. The experiments suggest a dependence closer to II fo. 8 .
Nevertheless, this is not enough to conclude that the poor agreement between the
CAM model and the real systems is solely due to the existence of the two angles
em and er •5
Other measurements done with the same apparatus provide a wealth of infor-
mation. The pertinent results are shown in Figure 85(c). Here the drum no longer
rotates, but it is stopped in a position such that the free surface of the granular
material is at an angle e just below the value er • When the vertical vibration is
then turned on, the angle of repose turns out to decrease (or relax) with time.
Furthermore, the rate of relaxation depends on the amplitude of the current in the
loudspeaker. The results shown in Figure 85(c) constitute fairly compelling evi-
dence that the embankment angle e evolves as log(t), while the CAM-SOC model
suggests that it should vary proportionally to t. This rather remarkable result has
been interpreted in terms of a simple model involving a relaxation process induced
by an agitation (caused by the loudspeaker) that may be considered the equivalent
of a "thermal" agitation.
In the context of this model, the intensity of the vibration plays the role of an
"effective" temperature Teff [21], [80]. There is a perhaps rather bold analogy to be
made with the phenomenon of electrical conductivity, where a flux j of electrons
results from an applied electrical field E through the bias of the conductivity a,
the latter being of a magnitude governed by the degree to which electrons are
released from randomly distributed traps. Here, the electric field E-the driving
force-is assimilated with the angle e, and the current density j with the rate of
change deldt. By analogy with the electrical case, a granule can be thought of
as trapped, much like electrons in a conductor, by neighboring granules. Rather
than reviewing the many assumptions involved in this problem, we will simply
calculate the average height of an effective barrier V as a function of the angle e.
Expanding V to first order about the starting point of the experiment, we may write
V ~ V o + VI (e - er ). Knowing that a flow is triggered when e = em, we impose
the condition V (em) = 0, which leads to 8 == e - er = Vol VI. Pursuing our
electrical analogy, we write that the rate of change del dt, like the current density,
5 As we will show in Section 4.2.2, the dragging effect of the rotating disk, even as slow as it is in
this case, can introduce an artificial periodicity in the sequence of avalanches. This point is still
being actively debated. Nevertheless, very recent experiments have for the most part confirmed the
conclusions presented in this paragraph.
138 4. Granular Media in a State of Flow
de
- = -Ae exp[tJ(e - er )],
dt
where both A == A o exp( -Vol kTeff) and 1'3 = VI! kTeff are independent of e.
The solution of the differential equation can be expressed by means of the
exponential integral function E 1 (tJe). Approximations valid when the argument
M» I lead to
I
e ~ er - -loglQ(tJAert + 1),
1'3
which is consistent with the loglQCt) behavior observed in Figure 85(c) for times
greater than to = 111'3 Aer . The agreement remains reasonably good even for shorter
times. This derivation thus accounts for the loglQCt) behavior of the embankment
angle of a vibrated granular pile.
Interpreting avalanches in terms of thermal detrapping is fundamentally different
from the critical angle model proposed by BTW. The experiments described above
clearly show that the simplistic view of the angle of repose er as a critical angle
does not stand the test of a detailed analysis of the relaxation mechanism.
6The paTticles in flow are sometimes referred to as a "moving species." The idea of describing the
problem by means of these two coupled variables has recently been exploited from a different angle.
It has led to several interesting developments which we will discuss later on.
4.2 Avalanche Models 139
(a) (b)
FIGURE 86. Illustration of the correspondence between (a) avalanche processes and (b) the
stick-slip mechanism. Both phenomena can be described by the same set of differential
equations (after [27]).
cylinder depicted in Figure 86(a), it turns out that the two systems are governed
by the same set of equations, with the following correspondence
w +------+ V.
In particular, when the pad is in motion, the elongation of the spring is given by
the equation
dx
if -=0 and
dt
Figure 87 illustrates graphically the behavior of any stick-slip mechanism. Upon
stretching the spring at a constant rate, which is equivalent to moving toward the
right on the diagram, the pad is made to execute a periodic motion characteristic
of a relaxation oscillation. On the one hand, the system returns to its equilibrium
140 4. Granular Media in a State of Flow
0=0
·x
o
o
til
OJ
:0 (jJ/y
co
~
When a flow occurs, the preceding analogy allows us to use the equations estab-
lished in Sections 2.2.1 and 3.1.1
de
-=w-yD,
dt
dD
dt = P[sin(e) - Itd(D) cos(e)],
if
where P = gh, h is the thickness of the moving sheet, and y depends on the
geometry of the flow; to a first approximation, y can be considered a constant. The
first of these equations expresses the conservation of the mass of material driven
4.2 Avalanche Models 141
to the bottom and replenished at the top by the rotation of the cylinder. The second
equation is obtained by analogy with the results found in Section 3.1.1.
The notation can be simplified by introducing an angle cD d defined by
or
We should bear in mind that fJ--d depends on the flux D, which is itself given by
D = _w_-_d_e/_d_t .
y
Eliminating D between the first two of our set of equations yields a second-order
differential equation in e
where p is defined as p = p / cos (cD d). This last equation describes oscillations
of the angle e with a period (l/y p )1/2. The steady-state solution, corresponding
to a constant flow of the rolling species, is given by
and
As expected, the stability of this solution and, more generally, the overall behavior
of the system, is critically dependent on how the coefficient of dynamic friction
fJ--d depends on the flux D.
Expanding cDd(D) to first order around the steady-state value DO, we get
cDd - d)
cD °+ (acD (D - D ) - d) --,
° = cD °- (acD 1 de
aD °y dt
~
aD 0
which can be substituted back into the second-order differential equation to give
(4-9)
In plain language, at very low rotation speeds, the relative velocity imparted by the
cylinder becomes negligible, and the angular windows ,6.(} and ,6. <P both converge
toward the difference em - er , which is what we called the relaxation angle in
Section 4.1 and is defined unambiguously only when the rotation speed is slow.
Next, we determine the effect of the rotation of the disk on the duration r of an
avalanche. That quantity can be derived from (4-9) and is given by
r = ffi
2 { rr - tan -1[ffi
~(<Ps - <Pd) ]} .
o
~
a
1:5
CD
.~
I-
o cD d cDs
Angle e
FIGURE 88. Numerical calculations of the trajectories D(e) with a very low rotation
speed (w = 10- 3 rad/s). The coefficient of dynamic friction decreases with speed, which
introduces an asymmetry in the curves, while the angular difference between trajectories
gradually increases (after [27]).
increasing function of the flux D. For instance, in the limit of a small and constant
variation a = - ~ (B<Pd/B D)o, (4-9) can be solved in closed from, giving the result
More general cases can only be solved numerically. Figure 88 shows an example
of such a calculation. Experiments done with noninvasive observation techniques
reveal remarkably good agreement with the present model, in spite of its great
simplicity.7 Further details on the comparison between theory and experiment can
be found in the original article on this topic [27]. By way of example, Figures 89 and
90 reproduce the results of two experiments conducted in the same drum rotating
at two different speeds. The similarities with Figures 87 and 88 are striking.
As a general comment, we might point out that avalanches are rather chaotic
for very slow rotations and tend to become more regularly periodic as the rotation
speeds up. This is suggested by the model and certainly appears confirmed ex-
perimentally. An important consequence of this observation is that it is generally
difficult to carry out reliable and controlled measurements in a rotating cylinder.
The same comment actually holds true for any other configuration. Gradually in-
creasing the inclination angle so as to drive the system out of equilibrium must
always be done as slowly as possible.
As pointed out previously, this particular stick-slip model for avalanches leads
to a second-order differential equation, in which the dissipative term (in de / dt)
involves a nonlinear dependence of the friction coefficient f.-id on the flux D. The
7The term "noninvasive" technique means that the measurements do not disturb the object being
measured. These techniques are most often optical in nature. They include the imaging methods
described in Chapter 3.
144 4. Granular Media in a State of Flow
3
'CD
I 2
o
-1
FIGURE 89. Optical measurements of the slope of piles and its time derivative for a series
of avalanches. The relation between the amplitude of avalanches and the angle at which
they begin is evident. The trajectories exhibit the asymmetry predicted by calculations. The
rotation speed used here is a rather slow 0.023 rpm (after [27]).
• CD
I
-2
0.5 1.0 1.5 2.0
8 (Arbitrary units)
FIGURE 90. Same as Figure 89, but with a faster rotation speed of 0.52 rpm. The trajectories
become more symmetrical and nearly periodic (after [27]).
4.2 Avalanche Models 145
dependence in question can take on varied fonns, which determine to a great extent
the precise behavior of the flow.
Put another way, the key to any analysis of stick-slip mechanisms is to learn
something about the F (v) dependence, F being the force of friction and v the
velocity. The functional shape of F (v) has in fact been the focus of a great many
investigations, notably among geophysicists and geologists, who are interested
primarily in earthquakes. A comprehensive review of these largely heuristic models
would be beyond the scope of this book. It nonetheless serves a useful purpose to
review a few general principles, which we now proceed to do.
Different Friction Models
Burridge-Knopoff (BK) Pads
The Burridge-Knopoff pad model is in many ways the paradigm among a slew
of approaches aimed at deriving functional laws for F (v). The model does suffer
from some weaknesses, which have inspired numerous variations attempting to
remedy them. We will present only a brief description.
The principle behind the BK model is illustrated in Figure 91 [82]. A series of
pads, each of mass m, rests on a horizontal base. The pads are connected to a rope
by meanS of identical springs of stiffness k. The pads are also interconnected by
a different set of springs with stiffness K. As it moves horizontally at a constant
velocity v, the rope drags along with it all the pads and causes them to slide against
the base (it is important that the pads not bounce off the base during their motion so
as to maintain friction at all times). We sense intuitively that such a system can be
the seat of complex oscillations. Our objective is to find the applicable fonn of the
force of friction, which we know must depend on both k and K. The differential
equation governing the motion of the j th pad, flanked On either side by pads j - 1
and j + 1, reads
2
d_X,
_ J = K(X+ 1 - 2X, + X J'-I) - k(X - vt) - F (dX
_J
,)
dt2 J J J dt '
which leads to an elementary solution of the type
1
Xj = vt - kF(v).
Constant velocity
Velocity
FIGURE 92. Dependence of the dynamical friction force on velocity in the BK model.
A detailed analysis of these equations shows that the F (v) has a functional depen-
dence such as is shown in Figure 92. Except very near the origin, the curve F(v)
features a negative slope which, as described earlier as well as in Section 2.3, is
likely to give rise to amplification and possibly chaotic oscillations. This partic-
ular function, and others of similar shape, translate a very common phenomenon
known in geophysics as the velocity weakening friction law. In plain language, it
simply means that the friction forces decrease as the velocity of the relative motion
increases. 8
Other Friction Laws F(v)
Figure 93 illustrates several functional dependencies of F (v) proposed by a number
of authors on the basis of different arguments [48], [83]. Various phenomena such
as contact wear-out and fatigue, frictional heating, and others, can come into play
and give rise to exotic shapes for the F (v) curve. There exists a fairly abundant
literature on this topic, which the reader may want to consult.
• The discontinuity of the force of friction when the flux of particles is zero, in
accordance with Coulomb's law of dry friction, which specifies that there are
two different friction coefficients, depending on whether there is movement
or not.
• The existence of a "negative resistance" coupling affecting the function F (v).
8A deadly manifestation of this trend is the phenomenon of aquaplaning of vehicle tries on wet roads.
4.2 Avalanche Models 147
A sphere on an
inclined sheet. V-shaped curve
"Half- W" curve
C C
IC719~)
~
Cal
2:Q O'u
c al
.g1l5 t;:E
'C; ~
LLo LL
(.J U
Velocity Velocity
vv
Contact fatigue:
"Inverted-V" curve N-shaped curve
(Carlson and Langer, 1989) (Barenblatt et al. 1981)
C C
~
c al cal
0'(3 0'(3
"u ~~
"i::
UtE
"C ~
LL LL
U U
Velocity Velocity
FIGURE 93. Various proposed functional dependencies of the friction force on velocity.
• A spatial coupling that accounts for the propagation of the rolling species after
the movement has been initiated. It also explains the duration of avalanches.
Obviously, the model has limitations. For one thing, it does not enable us to
calculate the profile of the granular flux. There exist other, more sophisticated,
models that can do that. They too start from a system of coupled differential
equations, but they involve a different set of variables (in this case, the flux of
rolling species and the height of the pile) and are based on a rather different
principle. We will take the time to briefly discuss the broad outlines of these
models, which are still in the process of being developed. We deliberately limit
ourselves to a phenomenological description.
3R(x, t)
--- = -3x (vR) + 3x (D3 x R) + feR, h), (4-10)
3t
where v represents the velocity of the sheet toward the bottom of the pile, and D is
a diffusion constant that can drive particles in both directions. The first two terms
on the right-hand side of (4-10) are quite familiar; they correspond to the usual
convection and diffusion mechanisms. The last term feR, h), on the other hand,
is the crux of the matter, because it must be able to account for the properties of
avalanches. As such, what form to chose for it bears some discussion. The function
f (R, h) must act as a mathematical operator capable of stopping a particle in mo-
tion or, conversely, setting one that is standing still in motion. Taking our cue from
the phenomenology discussed earlier in this chapter, we may begin to have a feeling
for what this function might look like by imposing a certain number of conditions:
Given these constraints, all based on what we already know about avalanches, we
are now in a position to look for a plausible form for f. The simplest possible way
for f to comply with our four requirements is to be of the form
(4-11)
where y and K are two positive constants. We might point out that this expression
depends on hand R to first order, which is a distinct advantage from the standpoint
of analytical calculations, as we shall see shortly. We also note that the operator f
is now proportional to R. That is a major difference from the equations considered
earlier. Previously, the flux of rolling species experienced amplification not as a
result of a domino effect, as is the case here, but solely because of the negative
4.2 Avalanche Models 149
curvature of the function a<t>lax in the vicinity of its equilibrium value [see (4-9)].
In the previous stick-slip model, the driving force determining the flux was an an-
gular deviation. Here, the flux is determined both by the deviation from the critical
slope and by the quantity of material already in motion.
In order to account for material at rest, all we need to do is write h = - r in the
previous equation. Thus, the height of the pile obeys the equation
(4-12)
so that the total quantity of matter h + R is conserved locally. We further note that
(4-11) does reproduce the metastability of a pile on the verge of avalanching which,
as we know, is a defining characteristic of granular piles. We can indeed convince
ourselves that, in the absence of any flow, the equation does not spontaneously
generate any avalanche. Thus the surface appears as if frozen in a static equilibrium.
If, on the other hand, we start in such a situation and introduce a perturbation on
the surface of the pile, at least a few grains will start rolling down the slope. The
slide will take place in a finite amount of time and lead to a stable state. That too
can be recognized as one of the fundamental characteristics of avalanches.
It is not essential here to examine further the details of this formalism. Suffice
it to say that it can be made use of either analytically (in the simplest cases) or
numerically (in the tougher ones). The interested reader is encouraged to consult the
basic literature about the theory [84]-[87]. We will restrict ourselves to discussing
two applications of this model.
FIGURE 94. The model predicts that a perturbation created anywhere on the slope works
its way back up. In the process, it gets attenuated and widened because of diffusion effects.
150 4. Granular Media in a State of Flow
efficiency of this process as measured by ah / at, and hence its propagation velocity,
depends linearly on y and R o, as is evident in (4-12). Things are far less intuitively
obvious, on the other hand, in the case of a mound. We would be more inclined
to think that the lower part of the mound would collapse. Instead, because it is
based on a set of coupled equations, the model suggests that the material located
above the mound can somehow "sense" the perturbation below and undergo local
minislides that rearrange material in such a way as to generate a new mound that
appears to propagate toward the top. Perhaps this unexpected property is a generic
signature of this type of model.
Simulation ofAvalanches
The ultimate test of the model is, of course, whether it can correctly simulate an
avalanche. To resolve that question, we consider a pile in a metastable state such
that axh(x, 0) = -SoC <0). That can be easily accomplished by increasing-at least
mentally-the inclination of the pile, initially assumed to be at the angle of repose,
until it approaches the angle of movement. Barring any additional perturbation, the
new state is metastable, and nothing happens (as long as the angle of movement
is not exceeded). In this situation, the upper part of the pile is at an angle greater
than the angle of repose.
What happens now if we add a few grains near the bottom of the pile? Those who
have correctly grasped the essence of the correlations associated with the coupled
equations will readily guess that the result is going to depend quite critically on
the value chosen for So.
To clarify this point, we need to consider in some detail how the process un-
folds. Assume that at time t = 0 a localized perturbation is created at point x in
the form of a few moving grains, which may be written as R (x', 0) = flo (x' - x).
Two opposite effects combine to produce the final outcome. First, as we have just
seen, the small perturbation propagates up the slope toward the top of the pile.
That is directly implied by (4-10) in which we neglect the coupling term r. The
ascending perturbation decreases with time as R(x, t) ex exp( -v 2 t /4D), as is
characteristic of all diffusion processes. On the other hand-and that is the second
competing phenomenon-the grains rolling down at abscissa x will dislodge other
grains initially at rest. Equation (4-10) shows that, when v = D = 0 (velocity and
diffusion are both zero), the density of the rolling species grows exponentially as
R(x, t) ex exp(y Sot). Stated in words, the initial perturbation sets off two compet-
ing phenomena that both grow exponentially. The situation is critical in the sense
that either one can win out. We find that if So > Sd ~ v 2 / y D, the number of grains
that are dislodged grows faster than the number of grains that make it back up the
slope. A runaway event-runaway because governed by an exponential-is trig-
gered in the form of an avalanche. 9 In the other case (corresponding to So < Sd),
the diffusive term dominates; it attenuates the effect of the perturbation and the
system remains at rest.
9We have chosen to preserve the notation used by the authors of this model. This should help those
inspired to read the original articles. The correspondence with the angles em and e,., used earlier in
this chapter, as well as their difference 8 = em - e,., is obvious. Here Sc' (== e,.') is set equal to 0, and
Sd(== em) becomes equivalent to what we called the angle of movement.
4.2 Avalanche Models 151
Starting from a simple consideration of an angle of repose Sc', the present model
leads quite naturally to the definition of a critical angle Sd which corresponds to our
familiar angle of movement em. It is the angle required to trigger an avalanche. 10
We might add that numerical solutions of the applicable equations confirm these
interpretations.
The model can, of course, be applied to a variety of situations, from filling
a silo to far more complex scenarios. We must keep in mind, however, that the
formalism accords a great deal of importance to what we have referred to as a
diffusive process, which in actuality boils down to a mechanism of attenuation and
upward propagation competing with the rate at which grains are knocked loose.
De Gennes has pointed out that such a diffusive term was unsatisfactory on at least
two grounds [86]:
• The first has to do with the range of this diffusion. For any variation of the
density of a rolling species R of size L, the diffusive term corresponds to a
perturbation of the order of D / lv, which is itself of the order of d/ L (where d
is the mean diameter of the grains) relative to the convection term. In practice,
this means that the diffusive term must be rather small for small-size slides.
• The second objection is that the physical interpretation of the diffusive term
is a sort of Brownian motion that enables some of the particles to make their
way back up the slope. The problem is that such an interpretation is plausible
on a very small spatial scale only, of the order of D / v ex d, in which case a
continuous description looses its meaning.
This opens to door to the possibility of modifying the model, while still preserving
the spirit of the initial equations.
lOWe also note, incidentally, that the present model leads to a process analogous to a first-order
phase transition, consistent with the conclusions arrived at in Section 4.2.1. This is, of course, in
contradiction with the predictions of the SOC model.
152 4. Granular Media in a State of Flow
the form
We retain the same functional form here, except that R i (x, t) now represents a
source of ambient noise due to the mechanical agitation of the system, and ee is
an angle corresponding to equilibrium. This can be justified by picturing particles
trapped in a potential well U that depends on the angle e through a relation of the
type U = mg dj(e), where j(e) is adecreasingfunctionofe which must obviously
vanish when e = Jr 12. The mechanical noise is described in terms of an effective
temperature Teff. The source of rolling species is then of the formd exp( -U 1kTeff),
which varies as d exp[a(e - em)], where em is defined as the angle at which the
detrapping potential equals the effective "thermal" energy associated with the
ambient noise. Under these conditions, we have U ~ kTeff and = mgdl kTeff,
and it can be verified that nucleation is suppressed as long as e < em. It only
begins when e = em' We should point out that in this model the angle em depends
critically on noise.
In short, the essence of this model is to give a preeminent role to the ambient
noise. This noise is responsible for detrapping the rolling species and is instru-
mental in determining the angle of movement.
De Gennes has shown that this approach accounts for both types of movement
identified in Section 4.I-a continuous flow (class A) at high velocities, and an
intermittent one (class B) at low velocities. In other words, there are always some
grains in motion in a class-A movement and, in particular, there exists a steady-
state solution for which the profile h(x, t) is constant, which is perfectly consistent
with experimental observations.
Intermittent or continuous avalanches in a rotating drum can be described by
the following system of equations
ah ah
-=-yR-+wx,
at ax
aR aR ah
-=v-+yR-,
at ax ax
where the flow is assumed to take place from left to right. Note the presence of the
wx term, describing the effect of the rotation. How to solve these equations is left
as an exercise. In case of difficulties, the reader is referred to the original article
[86]. We conclude this chapter by seeking the steady-state solution pertaining to a
class-A regime in a rotating drum of diameter 2L. The geometry is illustrated in
Figure 95.
The relevant boundary conditions are R = 0 at x = ±L. By further imposing
the restriction that the time-derivatives vanish throughout, we find steady-state
solutions of the form
W 2 2
Rss(x) = -(L - x )
2v
4.2 Avalanche Models 153
(a) (b)
-L+-_------:
-L o +L
FIGURE 95. Geometries used in conjunction with de Gennes's amended model based on
a system of coupled differential equations (see text).
and
These solutions predict a slope with a parabolic correction, illustrated in Figure 95,
although the deviation from a straight line remains small.
It is useful to remember that solutions of type A (continuous flow) and type B
(intermittent slides) crop up systematically in granular flow problems. They reflect
the metastable nature of piles whose slope evolves between the angles em and e,..
As a further example, we may want to determine the steady-state solutions when
a silo of diameter 2L is being filled from the top by a flux Q of particles falling
through a hopper. We assume that the flow is continuous and free of arch effects.
The orifice of the hopper is at x = 0, and the lateral walls of the silo are at x = ±L.
Using the same conditions as in the previous example, we find that the steady-state
solutions are
Q ( 1-
Rss(x) = 2v LIX I)
and
ess - e = (ah)
y
ax ss
-----
v
y L
I
-Ixl
Here again, the deviation of the steady-state angle from the angle of repose is
small, being only of the order of d / L.
This modified model allows a variety of situations to be analyzed, although,
at the present time, it does have limitations related to the fact that it treats a
granular as a continuous medium. One thing it cannot do, for instance, is account
for phenomena of fracturing and fragmentation (see Section 3.2.4) which, as we
now know, are likely to occur in piles subjected to any sudden change, such as
when the door of a filled silo is opened abruptly.
5
Mixing and Segregation
5.1. Introduction
Liquids have a well-known predisposition for miscibility.l Dry granular materials,
by contrast, are notorious for being difficult to mix homogeneously.2 Two granular
materials differing by their density, shape, size, or even by their micromechanical
properties (such as coefficient of elastic restitution and friction), exhibit a dis-
tinct propensity for segregation. This phenomenon is a fundamental property of
the granular state and an unending source of frustration in industry. Whenever a
mixture undergoes a flow, a vibration, or a shearing action, the components tend
to separate partially or completely, depending on the circumstances. By analogy
with chemical reactions, we may say that under the influence of various stimuli
a granular mixture inexorably tends to self-organize so as to locally reconstruct
clusters of identical particles.
As we have pointed out in the Introduction to this book, the segregation of dry
granular materials can be routinely observed even in the most primitive table-top
1Since liquids are made of particles subject to Brownian motion, thermal agitation alone produces
mixing. As pointed out in Chapter I, the Brownian motion of granular particles is entirely negligible.
Another source of energy is then required to achieve mixing. Vibration is a logical candidate, even
though the final result is often the exact opposite of what is being sought!
2The concept of "homogeneous mixture" needs to be clarified in a granular material. A mixture
composed of a fraction u: of granules A and f3 of granules B (with u: + f3 = I) is said to be homo-
geneous on a scale;" if a volume ;,,3 contains the two ingredients in the right proportions. We can
appreciate that optimum homogeneity will be achieved when the smallest scale ;", for which the
mixture can still be considered homogeneous in the sense just defined, is of the order of the size of
the largest of the two types of particle.
:-------.:
Wavelength A
FIGURE 96. Oyama's horizontal drum rotates around its axis. An initially homogenous
granular mixture segregates in vertical bands with a spatial periodicity A.
experiments by shaking a mixture of different grains (wheat, corn, rice, salt, are
all good candidates) in a test tube. 3 Likewise, farmers know very well that tilling
fields causes large rocks, in seemingly endless supply, to work their way up to
the surface. Peasants in India take advantage of the segregation properties of dry
granulars by shaking their harvest of chickpeas in baskets in order to separate
them from other materials. When Brazil nuts, mixed with other smaller nuts, are
transported in pick-up trucks over the rough back roads of South America, they
invariably end up on top of the load. 4 In short, the phenomenon of segregation in
dry granulars is universally recognized, even though it may not be well understood.
The first recorded observation of segregation in a three-dimensional medium
was described by Oyama in 1939.
3The importance of using dry materials cannot be overemphasized. It is essential that any interaction
with the ambient fluid be negligible. If not, the problems are of a radically different nature. Well-
designed cement trucks, for instance, are perfectly capable of adequately mixing gravel of various
sizes with cement and water.
4The example of Brazil nuts has become a paragon in matters of granular segregation following the
publication in Physical Review Letters of a paper entitled "Why do Brazil nuts are on top?". To speak
of "Brazil nut" phenomenon has become synonymous with the problem of segregation by size.
156 5. Mixing and Segregation
knows why. Spectacular as it may be, this experiment remains poorly understood. It
continues to be the object of numerous studies, some of which rely on sophisticated
techniques such as nuclear magnetic resonance (see Section 3.2.3) [90].
Be that as it may, the experiment underscores the amazing efficacy of segregation
by rotation or, to be more precise, by shearing. As mentioned earlier, segregation is
a very general property of mixtures in which particles are in motion relative to one
other. A careful observation of the mechanisms potentially involved has led some
authors to distinguish at least two different modes of granular segregation. They are:
M and m, and radii Rand r (with M > m and R > r), stacked on top of each
other. If the large sphere is on top, the potential energy E p of the stack can be
written as
E p ocr 4 + R 4 + 2rR 3 .
Interestingly, this expression is not symmetrical in rand R, which means that the
energy depends on whether the large or the small sphere is on top. Not surprisingly,
the configuration with the small sphere on top is energetically favorable. We may
find a degree of solace in this result. Yet, our intuition clearly fails us when it comes
to the phenomenon of segregation by size which, as already indicated, invariably
drives the largest-hence, heaviest-particles toward the top.
FIGURE 97. Two types of particles A and B of different sizes are stacked in superposed
layers. In the absence of any structural defect, AlB and BIA configurations are equivalent
from an energy point of view.
158 5. Mixing and Segregation
(a) (b)
16x16 16x16
Planar square lattice Triangular lattice
FIGURE 98. Two stacking modes for two-dimensional structures. The triangular config-
uration (b), being as compact as possible, has a lower potential energy than configuration
(a). It is therefore energetically favorable and more stable.
the upper part of the container, the potential energy of the entire pile is given
by
This time, which type of marble is on top has no bearing from an energy point of
view, and there is no reason to anticipate any segregation, at least as long as inter-
face effects are absent. This result is obvious, inasmuch as two layers of identical
density are always in equilibrium.
Without the benefit of a more detailed analysis, we may begin to suspect that
structural defects in a lattice made of particles of different sizes may have some-
thing to do with the tendency of large particles to work their way back up. To
explore this possibility, we return to two configurations already discussed in Sec-
tion 3.1.3. The relevant structures are reproduced in Figure 98.
An elementary analysis of the two stacks shows that the triangular lattice is-
based on simple energy arguments-more probable and stable than its square
counterpart. As we have seen, the compact triangular lattice is the only one to
exhibit characteristics consistent with Reynolds's dilatancy principle. The reader
will recall that a compact stack subjected to any distortion can only respond by
expanding, which increases its potential energy. If we realize that distorting a
compact triangular lattice necessarily entails the creation of defects in the stack,
we may legitimately inquire whether such defects may tend to concentrate in the
lower or the upper portion of the pile. In particular, what can energy arguments
tell us about this question?
Faults
Compacted
Zone
FIGURE 99. Defects are created when a large disk is inserted into a two-dimensional stack.
The photograph was obtained by back-illuminating a real stack. The lower part of the stack
remains compacted, while the triangular symmetry of the upper part is greatly disturbed by
the introduction of the large disk (after [93]).
energy is denoted E pu if the defects are in the upper region, and E pi otherwise. If
the defects involve an increase in volume dv, the calculation indicates that
E pl - E pu ex vdv > 0,
which implies that defects near the top are indeed energetically preferred. Another
way to express the same result is that, in a cylindrical container, the potential
energy is minimum when the less dense material is on top.
An equivalent two-dimensional experimental configuration, similar to those we
have discussed in Chapter 3, is easy to implement with suitable provisions for
agitation. s All we need to do is to introduce into the container a single larger disk
of the same volumetric density as the rest of the pile, and we can readily witness
the phenomenon just described. As demonstrated in Figure 99, the intruder causes
local distortions in the lattice by creating defects which tend to migrate to the upper
part of the structure. 6
When viewed in this context, the process of segregation by size emerges as one
of the consequences of the dilatancy principle. The introduction of an intruder
necessarily causes a local distortion of the lattice manifesting itself in a local
expansion. The expanded and, therefore, less dense portion of the pile tends to
5The meaning of the phrase "suitable agitation" deserves to be thought out carefully. It may be said
that shaking (or vibrating) a complex granular edifice enables us to explore perhaps not all but at least
many of the possible configurations of a pile. Several simulation approaches (notably the Monte Carlo
technique, to be discussed in Chapter 6) capitalize on this observation by minimizing the energy after
each perturbational event through various relaxation processes.
6We will often use the term "intruder" to refer to a particle whose size or other properties differ from
those of the "normal" sea of particles which, for the sake of simplicity, we will consider uniform.
160 5. Mixing and Segregation
move toward the top, dragging the intruder along with it. This would suggest that
the shape of the intruder may playa crucial role in the process of segregation,
depending on how readily it may fit in the surrounding lattice.
There are too many potential objections to accept such a crude explanation at
face value. In an attempt to get a better grasp of the phenomenon, we will have to
refine our understanding of the successive relaxation states of a pile undergoing
segregation. Until then, the one idea that should stand out and be remembered in
what follows is that size segregation implies defects created by the intruder in its
environment.
(a) (b)
(e) (d)
FIGURE 100. Different configurations of a two-dimensional inhomogeneous stack. Dia-
grams (a) and (b) are computer-generated structures, while (c) and (d) are photographs of
actual real-life stacks. Note that a large intruder can be in a stable position without having
to be in contact with particles immediately below (arch effect) (after [93]).
rest on lattice lines associated with the matrix. Instead, they can be propped up
above such lines by lateral particles, somewhat like the arch of a cathedral rests on
stones that transmit its weight to side columns. Pursuing this metaphor, we refer
to this phenomenon as the arch effect or vault effect.
When trying to model the dynamic properties of such a system, we need to
inventory all possible stable positions of the intruder. Stability occurs under one
of two conditions:
Modeling this situation involves solving a topological problem, which goes some-
thing as follows: The intruder is raised a very small step at a time, as depicted
in Figure 102, and the system is left to reorganize itself by relaxing around the
intruder. The new arrangement is examined to determine if it is stable or unsta-
ble. If it is unstable, the intruder is raised some more by a tiny amount, and the
stability is examined again. The process makes it possible to find eventually all
stable configurations.
162 5. Mixing and Segregation
FIGURE 101. Simulation of the equilibrium of an intruder by the arch effect. On the left
is a photograph of an actual stack of metal spheres containing a foreign disk (after [93]).
Two-Dimensional Model
The relevant geometry is illustrated in Figure 103(a). Let cD = R/ r be the ratio
(> 1) of the radii of the particles. As indicated earlier, the effective part of the pile
is confined within walls Bl (T) and B 2 (T), which in the present case intersect at an
angle of 60°. Our purpose is to determine all stable positions of the intruder as it is
being raised in a step-by-step fashion up the pile. The first thing we notice is that,
because of the geometry of the structure, we do not have to explore a height greater
than one period of the structure, which is given by 8 = 2r~, or 3.46 times the
radius of the dominant particles. Let h be the altitude of the intruder's center, as
shown in Figure 103(a). We keep track of each particle by its row (index i) and
column (index j). Simple geometry considerations show that the stable positions
FIGURE l02. Method for seeking the equilibrium positions in a stack by exploring different
possible configurations.
5.2 Segregation by Vibration 163
(a) (b)
FIGURE 103. Diagrams of two- and three-dimensional stacks used to develop a model of
segregation by the arch effect.
of the intruder, i.e., when it rests on a lattice line, are obtained when
where the superscript S indicates that the position is stable. The index k runs
from [(_l)i+l + 1]/2 to Int((i + 1) /2), the operatorInt(m) designating the nearest
truncated integer of the argument m.
Next we look for stable positions of the intruder via vault effects, that is to say,
when it rests on two particles located below its center of gravity. We start from
a situation where the intruder is in contact with the two lateral walls B 1(T) and
B2(T), which occurs when h vI = 2r. When the intruder is raised gradually, it will
find a new stable position when two small particles can just squeeze below its
center of gravity. This happens when h v2 = (R + 8),J3 ;:;" r,J3(cD + 2), where 8
is the space between the intruder and the lateral walls.? The fraction S of stable
vault configurations over a period e is given by
h v2 - h v1 2- ,J3
S=l- =~cD;:;"O.077cD.
e 2y3
A smooth rise through a continuum of vault configurations is obtained when S = 1,
which happens when cD~D ;:;" 12.9. As such, the quantity cD~D can be construed as
a critical ratio ofdiameters marking the boundary between two types of behavior
[92], [93]. On one side of the dividing line, the intruder rises continuously, while
on the other it goes through a series of discrete steps determined by the size of the
smaller particles. This line of reasoning makes it possible to calculate analytically,
7In this approximation, we assume 8 '" 2r. The exact solution would require lining the boundary with
small particles. The walls would then no longer be straight but made of a succession of connected
half-circles. The approximation is justified a posteriori by numerical simulations.
164 5. Mixing and Segregation
--
<I>
~ 20
'"
OJ
_. 13
-- -
.0 11
'0 15
d
E. -
I--
~ 10 1-----
--
---- -
9
7
5
o - - 3
~o t--- 1.2
c. 5
OJ
:0
.!9 (
lih )
(f) 0
o 1 2 3 4
Intruder's displacements (no. of beads)
Three-Dimensional Model
It is relatively straightforward to extend the preceding model to the case of three
dimensions. The triangle (T) becomes a tetrahedron which obeys the same sym-
metry. Using the notation indicated in Figure 103(b), we have
I R-r R+r
h-h = - - = - -
sin \II tan \II I '
where h = 3R, hi = 3r, \II = cos-1(1), and \II' = tan- 1eJ2/2).
The critical diameter for continuous rise via vault effects in three dimensions is
given by
(a) (b)
2-D
1
FIGURE 105. Diagram of experimental methods used to study the segregation by shaking
of a large particle in either (a) two-dimensions or (b) three-dimensions. In (a), a large
marked intruder is immersed in a population that includes a few tracers (black particles),
whose progression is tracked by image processing techniques. In (b), the movement of
tracer particles in their environment is monitored by direct visual observation.
Marked intruder
Vibration source
Camera monitoring
the vibration
amplitude
FIGURE 106. Photograph of an experimental setup designed to track the ascending move-
ment of an intruder in a two-dimensional configuration.
166 5. Mixing and Segregation
(a) (b)
:'I%i\\§~H~~1ii~~!~~
FIGURE 107. Experimental observations of the movement of an intruder. Diagram (a)
corresponds to a continuous rise (<I:> = 16 and [' = 1.2), while (b) reveals successive
plateaus (<I:> = 2 and [' = 1.4). The doted white line in (a) is an artifact due to the image
processing technique. It is quite real, on the other hand, in (b), which was generated by
directly displaying the height of the moving object as a function of time. The horizontal
scale corresponds to about 1 hour (after [93]).
plate, measures accurately the amplitude of the vibration. The second is mounted
on a translation stage and connected to the image processing electronics. With
thresholding and the image addition techniques described in Section 3.2.3, it is
possible to track one or more particles during the course of an experiment. 8 Alter-
natively, by moving the camera horizontally at constant speed, the vertical position
h(t) of the intruder can be displayed directly on the monitor's screen as a function
of time. We will see several examples of this technique below.
S"Thresholding" consists in defining a particular level of brightness. Anything brighter than that level
is considered white (or I), and anything darker becomes black (or 0).
5.2 Segregation by Vibration 167
Long-lived
fractu res
FIGURE 108. Photograph showing the relatively random appearance of fractures around
the intruder (after [93]).
Referring to the ascent diagram of Figure 104, we can make the following obser-
vations:
• A small intruder (one that is characterized by <P < 12.9) requires a large rel-
ative displacement 8 between itself and its environment in order to overcome
each step in the ascent diagram. Accordingly, we may anticipate a sizable vi-
bration threshold to initiate this kind of motion. Experiments indeed confirm
this .
• A large intruder (one with <P> 12.9) proceeds much more readily while re-
maining continuously in an arch configuration. We can predict-and it is
experimentally verified-that the amplitude threshold is much lower than in
the previous case. In other words, a larger intruder moves up far more easily
than a small one.
As we will see, this is further confirmed by diagrams of upward speed as a function
of the ratio <P = R / r, at least on condition that we avoid the convection regime.
In this respect, experiments that make it possible to observe simultaneously the
upward drift of the intruder and the relative motions of the surrounding particles
are highly revealing.
Finally, we might point out that the phenomena of continuous and discontinu-
ous ascent just discussed have never been observed up to now in three dimensions,
even though simulations predict them. That does not necessarily mean that they
do not exist.
Convection or Arch Effect?
Experiments conducted in two or three dimensions show that, when the vibration
is sufficiently intense, particles experience a phenomenon of convection similar
to what we have already encountered in Chapter 3. We proceed to provide ad-
ditional details pertinent to these two situations. We start with three-dimensional
configurations.
168 5. Mixing and Segregation
t
lo} (b) (e)
9Some degree of caution is warranted in making this statement. It has not actually been proven that
the intruder itself does not somehow induce convection. The dynamic maps presented in Figure 111
would suggest that such is not the case and that the intruder does not promote convection, at least not
below its own height. This point is still being debated [96].
5.2 Segregation by Vibration 169
E 0
~
Q)
l.l
~ -5
::J
til
.8 -10
Q)
l.l
c
rn
1il -15
0 0
FIGURE 110. Successive positions of different intruders plotted against the number of
impulsions applied to the container. The ratio <t> is equal to 9.5 (crosses), 6 (circles), and
1 (squares), respectively. Note that large intruders remain trapped at the surface after com-
pleting their climb, while small ones (square data points) are dragged back down to the
bottom by convection (after [95]).
two-dimensional pile vibrated vertically with varying accelerations. Figure III (a)
reveals a process of convection identical to the one just discussed in three dimen-
sions. By contrast, Figure 111 (b) is indicative of an arch effect squeezing the in-
truder. As emphasized above, this phenomenon depends on the intruder's size. The
velocity of the upward movement is presumably a function of the diameter ratio <P.
This was entirely confirmed by a set of experiments recording the altitude h(t) of
intruders of various sizes as a function of time for a given acceleration and container
configuration. The results of these experiments are reproduced in Figure 112.
convection
,.... >,;'", ", ~onvection
I /<I>=12~9
<1>=16.3
10 .",. /<1>-10.5
," <1>=9
i) I
~
<1>=5.3
.- 8 . I
E I I
()
'-"
Q) 6 / ..
"'C
.....
::l
4
t1r.... .---lntermittent
«
;;:::; Events
2 <1>=3
0
0 10 20 30 40 50 60
Time (minutes)
FIGURE 112. Positions h(t) of intruders of various sizes immersed in a bath of particles
1.5 mm in diameter. The larger the size of the intruder, the faster its ascent (after [93]).
Here the acceleration was held constant at r = 1.25. The various curves h(t)
have been displaced along the horizontal axis to make it easier to tell them apart.
The results clearly demonstrate a process of size segregation consistent with the
arch effect model discussed earlier. Small-size intruders (<l> < 12.9) experience a
discontinuous ascending motion marked by a series of steps and plateaus. The
greater the size of the intruder, the less discontinuous its upward movement, in
agreement with the model. All of this is consistent with the ascent diagram shown
in Figure 104. Furthermore, with small enough intruders (characterized by <l> < 3),
no ascending movement occurs at all, at least not for this particular acceleration
and over the duration of the experiment (about 1 hour). The results, summarized
in Figure 113, prove that there is indeed a threshold diameter below which any
14
c
'E
......
12
E 10
.s 8
~
'u
0 6
~ 4
C
OJ
u 2
«
Ul
0
0 5 10 15 20
Diameter Ratio <l>
FIGURE 113. Ascending velocity measured from Figure 112 as a function of the diameter
ratio <P (after [93]).
5.3 Segregation by Shearing 171
ascending motion is inhibited. The existence of the threshold diameter makes sense
on the basis of the ascent diagram in Figure 104. We noted at the time that, for
weak accelerations, the fluctuations 8h of the intruder's positions can be smaller
than what is required to bridge the quantum jumps between successive steps. This
is consistent with the nonlinear behavior observed in Figure 113.
Finally, the size dependence of the ascending velocity of intruders may find use-
ful industrial applications, since it provides a means to separate particles immersed
in a granular medium. We can envision the possibility of "filtering out" components
of different sizes by a proper choice of the acceleration imparted to the cell.
lOPigure 114 is a drawing, not a photograph taken during an actual experiment. The distinction is
significant, inasmuch as it is not obvious that the configuration depicted here can truly occur in a
real rotating drum. This particular structure was arranged so as to ensure the local equilibrium of
every disk during the stacking process. There is a fundamental difference between that and the global
stability ofthe entire edifice. An avalanche boils down to disrupting the local stability, which in turn
affects the global stability of a granular pile.
172 5. Mixing and Segregation
Rotation
Witness
particle
Roughened
inner wall
portion of the pile remains effectively bound to the cylinder. As the tracer particle
is being dragged along by an avalanche to the bottom of the pile, it is reinserted
and buried into the solid phase. The rotation of the drum then causes it to rise back
up toward the free surface.
It is important to have a clear picture of the sequence of events. The evolution
of the intruder is illustrated in Figure 115, which shows the outcome of a real
experiment. The experiment reveals that the process of segregation toward the
center or edges of the drum occurs in the flowing region at the surface of the
pile. Based on what we know about avalanches, particularly about their statistical
FIGURE 115. Velocity diagram of a tracer particle dragged in the liquid phase and rein-
serted in the solid phase. In the present case, R < r. The intruder is seen to converge toward
the center (after [90]).
5.3 Segregation by Shearing 173
IIThis is an important observation. It means that segregation by shearing does not result merely from
reinsertion of an intruder during successive avalanches. A more plausible picture is for the intruder
to undergo segregation by size during an avalanche and be transported across a distance that is
determined by its relative size.
174 5. Mixing and Segregation
ri+l~
•• "~;.'..
ri+1J if£.'· :/;:I~':.:;~J
0.,,..,,0
~ rj ri
The results show that, depending on the size of the intruder relative to the
majority particles, it tends to converge toward the center, to explore all the available
space, or to take refuge near the periphery. A uniformly gray area indicates near
perfect mixing.
We note first that the correlation diagrams ri+ 1 = f (ri) are all symmetrical with
respect to the principal diagonal. This observation is not insignificant. It means that
a true steady state is reached as early as the first iteration. If that were not the case,
we would see over the course of measurements a flight of the intruder either toward
large radii or toward small ones. That would translate into an accumulation of data
points either above or below the principal diagonal in the correlation diagrams. It
also means that the steady state can be described in terms of a relation of the type
IT (rj+rI rj)
IT(r;jri+1)
where P (r) is the probability of finding the intruder in region S, and IT (rj +1/ rj)
is the conditional probability of finding the particle at rj+1 when it is at rj during
the preceding frame. These results can be normalized by writing
l IT (r /r')P(r') dr
R2
per) =
Rr
and
l
R2
per) dr = 1.
Rl
5.3 Segregation by Shearing 175
5.0 - , - - - - - - - - - - - - - - - - ,
4.0
a 0.04 0'08~
0.00
-0.04
---~...... .
30
-0.08
.....- \ 0.8 1.0 1.2 1.5
\~ CD
I::" .........
Q 2.0 ''''''
1.0
~ .....c .....I:l..~:~~~~<·P-··.q· ...o/:::: ....c ...
~ '";1. ~
0.····P
The experimental results have been averaged over a distance /:).r equal to three
particle diameters.
Figure 116 shows convincingly that the region of space covered by the intruder
depends on whether its diameter is larger or smaller than that of the majority
particles. The same information is conveyed by a graph displaying the probability
P (r), as shown in Figure 117. To a first approximation, the probability P (r) can
be described by a function of the type per) ex exp(ar), where a represents the
inverse of a length characteristic of the segregation process. Evidently, a changes
sign when the ratio <1> goes through unity. A positive sign indicates that the intruder
flees toward the periphery, a negative sign that it tends to converge toward the center.
Without pursuing the analysis of experimental results any further, we may note
that our avalanche model discussed in Chapter 4 is not completely accurate. In
particular, we had claimed that an avalanche, being of random size, should reinsert
a particle anywhere along a flowing sheet. Actually, as revealed by the maps of
first iteration or by tracking an individual marked particle of congruent size, both
the center and the periphery of the cylinder are attractors for the dynamics of
the system. If a particle is introduced near the periphery of the cylinder, it will
tend to stay there. Likewise, a particle introduced near the center will tend not to
wander off vary far. This implies a degree of correlation between trapping events
taking place within avalanches, somewhat consistent with the model developed in
Section 4.2.2. We might conclude that the dynamics of size segregation is governed
by two attractors, one at the center, and the other on the periphery of the cylinder.
According to this hypothesis, segregation in a rotating cylinder could be construed
176 5. Mixing and Segregation
as a mechanism favoring one attractor over the other. Such situations often involve
the phenomenon of bistability. 12
12 Abistable system has two equilibrium states. It can switch from one to the other under the effect of
some external perturbation.
13 A mass is said to be "connected" when its particles actually touch each other.
5.3 Segregation by Shearing 177
after an infinite time-in reality, this time is quite short, as we will see-is de-
noted Soo.
We proceed to highlight a few general ideas relevant to this situation. A more
detailed analysis can be found in [90]. In order to characterize the state of the
mixture, it is useful to introduce a parameter describing its degree of order. Let
S Ct) be the surface area of type- A disks absorbed in the reference mass at time t. It
is clear that S Ct) has to be smaller than Soo. The degree of segregation is quantified
by a parameter aCt) defined as
Set)
aCt) ==-.
Soo
At this point, it is natural to introduce an ordering parameter poet) that can vary
between 0 (for a completely random and homogenous mixture) and 1 (for a fully
developed reference mass). This parameter is defined in terms of aCt) by
1.0
-
~
C....L 0.8
"*E~ 0.6
O;~o
I1l o
a..
....Q) 0.4 o 0 00
rn
~ 0
"E 0
o .Q
Time constant 'r
0.2
Time
0.0
o 200 400 600
Time t (seconds)
FIGURE 119. Order parameter PaCt) plotted against time for a mixture of disks with
diameters of 6 and 10 mm. Thirty percent of the surface is occupied by the smaller disks.
The inset shows the dependence of [1 - PaCt)] on time Cafter ref. [90]).
178 5. Mixing and Segregation
• The ordering parameter grows exponentially. The inset in Figure 119 shows
that it evolves as Po(t) ex 1 - exp( -t Ir), which describes a kinetic process
of first order. This remarkably simple behavior remains totally unexplained.
When studying mixtures of different concentrations of particles A and B, the
conclusion is that the time constant T is practically independent of the composition.
As of yet, this too is unexplained.
This is entirely consistent with the results reported in Section 4.1, namely, that
the flow switches from discontinuous to continuous at around 0.3 rpm. Second,
the speed of 1.3 rpm is well below the 12 rpm needed for the onset of inertial
effects.
Both the previous experiments and the present one were done with 600 small
disks 6 mm in diameter, and 720 large disks 10 mm in diameter, giving area ratios
of 25% and 75%, respectively. Two surprising results emerged. First, the time
constant remained unchanged at 25 s as the rotation velocity varied from 1.3 to
8 rpm. Second, at velocities higher than 8 rpm, segregation disappeared altogether
and the mixture remained substantially homogeneous, even over periods of several
hours.
These results are completely unexpected and, for the time being, without ex-
planation. We would normally anticipate that the process of segregation should
become more efficient as the small disks pass more frequently through the portion
of granular material flowing down the slope. In fact, nothing of the sort happens.
Quite on the contrary, segregation proceeds at the same pace even though the
number of crossings through the liquid phase varies by more than a factor of 6.
Furthermore, since segregation was shown to behave as a kinetic process of first
order, we would expect it to depend monotonically on rotation speed, rather than
to suddenly drop to zero at 8 rpm.
Perhaps a plausible interpretation can be proposed if we go back to the principles
we invoked to explain the role of arch effects in the phenomenon of segregation
by vibration. We argued that a granular system must have enough time to relax
between excitations in order to adapt to the intruder's geometry. Only then can the
intruder migrate efficiently. In other words, segregation is sensitive to the size and
geometry of the objects involved and requires a finite amount of time to manifest
itself. When viewed in that light, the results described above may not be so puzzling
after all. As the rotation velocity increases, the granular surface flow becomes too
fast and chaotic, leaving too little time to adjust to the geometry of the particles.
We may even push the argument a step fmther and envision that the flowing
sheet, which is liquid-like at low speed, gradually turns into a gas as the number
of collisions between particles increases with speed. This hypothetical "phase
change" between liquid and gas may occur rather abruptly and could very well
explain why segregation suddenly disappears above a certain rotation velocity (in
the present case, at 8 rpm). Between 1.3 and 8 rpm-the regime where segregation
proceeds efficiently-the experimental results show that T increases linearly with
rotation velocity (provided that T be expressed in number of revolutions).
(a)
Boundary line
(b)
FIGURE 120. Diagram (a) shows particle clustering after the drum, filled as indicated in the
text, has rotated for 300 s. Diagram (b) shows a portion of the boundary of the accumulated
cluster.
a
0.0 0.4 0.8 1.2 1.6
log(r)
FIGURE 121. Normalized length M plotted against the radius r of the measurement circle.
here have yet to be confirmed by more careful experimental work and further
theoretical analysis. 14
14De Gennes has recently developed a model for segregation via avalanches [43]. The model is based
on a set of coupled variables discussed in Section 4.2.3. Although it deals with a situation that is
different from a rotating drum, it leads to power laws with fractional exponents which may well have
a bearing on the experiments described here.
182 5. Mixing and Segregation
~
Wavelength A
axis at speeds ranging from 15 to 65 rpm. After 10 or 20 min, the mixture has
separated in bands characterized by a wavelength A, as depicted in Figure 122.
The most salient features of the experimental results can be summarized thus:
Between 15 and 65 rpm, there is little or no dependence of the wavelength on the
angular velocity. When the rotation speed drops below 15 rpm, segregation ceases
altogether, and the smallest bands tend to be the first ones to vanish.
Several authors have reported that the steady state, reached after long periods,
consists of three bands.
15 Curiously,
the dependence of the kinetic angle on particle size has apparently never been observed
in two-dimensional geometries, perhaps because the total number of particles is then too limited.
Expanding on an idea advanced in Section 4.1 in connection with avalanches of varions sizes, we
may surmise that the aspect ratio manifests itself only when the number of particles is large enough,
regardless of dimensionality.
5.4 Segregation in Oyama's Three-Dimensional Drum 183
This we write
with
Focusing our attention on type-B particles, their flux will result from two com-
peting effects. On the one hand, there is a flux <PBx (,6,,8) due to the difference
in kinetic angles, which tends to drive particles along the direction of the x-axis
(see Figure 122). On the other, there is an opposing flux <P BD due to diffusion and
described by Fick's law involving a diffusion coefficient D. The total flux <P Bx of
B-type particles along the x-axis is then given by
aCB aCB
= -,6,,8-- - D--.
<PBx
ax ax
To summarize, the horizontal flux is created by differences in the kinetic angle, but
it is opposed by a diffusive component that tends to equalize the concentrations.
The formation of bands would result from competition between the two effects.
6
Numerical Simulations
6.1. Introduction
Numerical simulations aimed at modeling various aspects of the physics of gran-
ular materials, which we have touched upon throughout the earlier chapters, have
a twofold objective. l On the one hand, there are pressing incentives to solve a
number of practical problems related to the treatment of granular matter in indus-
try. Whether the issue is pesky segregation, blockages of flows by arch effects,
or disruptive internal convection phenomena (see Chapter 1), the requirements
of the industrial sector are many and, needless to say, almost always immediate.
The urgency of industry's needs and the increasingly rapid developments of cre-
ative numerical simulation techniques have prompted many researchers to devote
a great deal of effort to devising algorithms suitable for describing the behavior of
granular materials. 2
On the other hand, numerical simulations are of considerable interest from a
more fundamental point view as well. They offer the possibility to explore the
effect of many parameters which are simply not accessible to experimentation. 3
In that sense, numerical simulation has truly become an integral part of basic
1 An excellent introduction to the topic of numerical simulations of granular materials can be found in
[59].
2 At
the present time, the number of researchers engaged in numerical simulations in this particular
domain of physics substantially exceeds those pursuing experimental work.
3That is the case, notably, for the coefficients of elastic restitution E and the coefficients of friction
fL, which can be valied at will on a computer, whereas nature offers the experimenter a very limited
range of choices.
• The first option consists in sampling the system at regular intervals, using
a time step small enough to avoid "missing" an event, which could com-
pletely change the subsequent chain of events. However, as we have already
pointed out, the duration of contact between hard spheres is infinitesimally
short. Several collisions may occur in rapid succession and possibly set up
oscillations at a rate exceeding the sampling frequency, as suggested in Sec-
tion 3.2.1 and illustrated in Figure 123. Therefore, a sequential algorithm is
not a wise choice in the hard-sphere approximation. It would be more suitable
for a molecular dynamics model, where collisions have a finite duration.
• Researchers have devised an algorithm in which the timing of the sampling is
governed not by a fixed external clock, but by events themselves. This type of
event-driven approach guarantees that no event will be missed. On the down-
side, there is a risk of getting trapped in situations where a particular event,
such as an oscillation triggered by a collision, lasts for a very long time. It
then becomes necessary to rely on some test criterion to get around the trap.
We have already mentioned such a criterion in Section 3.2.1, known as LRV
(Largest Relative Velocity).
6.1 Introduction 187
Loss of
Sequential Event
Method II I I I I I I
?
Events 11111
Time
Jammed
Event-Driven Point
Method 1111
Events 11111
Time
There are clearly several possible answers to this question, some more realistic
than others. The technique we describe next has the merit of being fairly intuitive.
It is rooted in the notion of "cloud," which effectively spreads out the mass of each
particle over a region larger than its actual volume [57]. In this picture, the clouds
associated with two particles can overlap, which ensures a continuous passage
from one to the other. The cloud function her) must satisfy several conditions.
They are
1 00
h(r)2:rrr dr = 1, (6-1)
(r
her) = -1-2 exp - - ) ,
2:rr 0' 20'2
2
where 0' is larger than the diameter d of a particle (for instance, 0' = 6d) and deter-
mines the extent of the clould. This enables us to define a density p, a macroscopic
4We have already defined in Section 4.2.1 a granular temperature as the driving force of a thermal
agitation by vibration which causes the detrapping necessary to set off an avalanche. It is not in the
least proven that this definition is identical to the one we are about to put forth in this section.
6.2 Simulations of Collisions 189
p(x)T(x) =m L N
;=1
U2
--'-h(lxi -
2
xl) - p(X)--,
V 2 (X)
2
where N is the number of particles used in the simulation, and m is the mass of
each. These macroscopic quantities are continuous in space and time. It is pos-
sible to calculate their gradients, and as such they can be viewed as the usual
thermodynamic variables.
5 Researchers doing numerical simulations frequently resort to such predictive techniques.which consist
in bypassing accumulation points leading to endless loops by deciding ahead of time the state a system
will find itself in. This type of trick, as it were, speeds up the computation time. It only works, of
course, to the extent that the predictions are indeed correct.
190 6. Numerical Simulations
FIGURE 124. Evolution of a group of five spheres, initially grouped in two blocks, for
E = 0.8 (after [59]).
Consider five spheres originally arranged in two blocks of two and three mem-
bers, respectively, about to collide. What is the trajectory of the five spheres after
the collision? Do the spheres coalesce into a single block? If not, do they separate
and perhaps rearrange themselves in a different pattern? Before solving this prob-
lem, it is useful to recall the definition given earlier of a "block" in the context of a
numerical simulation. Two colliding particles form a block if their relative velocity
is smaller than a predetermined value V c chosen according to the characteristics of
the computer used. From a practical standpoint, the velocity differences between
all pairs of adjacent particles forming a block have to be computed at every instant
identified by the ED algorithm. Let 1'..vi = Vi-l - Vi designate those differences
for all adjacent pairs of a block. Pairs for which 1'.. Vi < 0 do not collide, while those
for which 1'.. Vi > 0 are likely to. The LRV procedure works as follows:
(1) We pick the adjacent pair with the largest value of 1'.. Vi , or 1'..Vj = max(1'..vi),
at the moment of impact, and we let the particles (j, j - 1) collide.
(2) Collision matrices of the type described in Section 3.2.1 are used to calculate
an updated set of differences 1'.. Vi.
(3) The previous two steps are repeated until all the differences 1'.. Vi become
either smaller than V c (in which case the corresponding particles form a
block) or negative (in which case they fly apart).
It has been shown that this type of predictive approach does indeed lead to the
same result as the conventional ED method.
of soft spheres and on a sequential calculation. The primary difference with event-
driven methods is that in the present case the duration tc of a collision is not zero.
The principle behind MD methods is to solve in regular incremental steps the
equations governing the changes in linear and angular momenta of the colliding
particles. The objective is to solve the following vector equations
and
where f is the moment of inertia of the solid around its axis of rotation, r x F is
the torque exerted by the force F, and Fern is the component of the force acting on
the center of mass. Once again, we emphasize that this strategy is quite different
from the one followed in ED models, which starts from the equations governing
momentum exchanges, in the manner described by (2-2) and (2-3). In the present
case, solving (6-4) requires a knowledge of the forces F and Fern, of how they vary
in time, and of the duration tc of the collision. As a prerequisite to any molecular
dynamics simulation, it is essential to model as exactly as possible the forces
of elastic restitution and friction involved during collisions between particles. We
have already stressed on numerous occasions how fundamentally difficult this task
can be (see, in particular, Section 3.1.1), due to the inherently indeterminate nature
of the equilibrium forces in a granular stack, as they depend on its prior history, or
to our limited understanding of contact interactions between solids. This explains
the many forms of equations proposed by various researchers working on this
problem. The situation is not unlike that discussed in Section 4.2.2, in which we
reviewed a variety of functional dependences of the friction forces on velocity.
The goal of the next few paragraphs is to discuss the different types of behavior
we might encounter in modeling the contact forces F and Fern, which feed directly
into (6-4).
contact forces are involved, at least under the assumption that angular momenta
can be neglected. In simplified vector notation, these forces are:
• A force of elastic restitution, related to the elastic energy stored during the
penetration of the two particles. This force is given by
(6-5)
where llij is the unit vector along the line connecting the centers of the parti-
cles i and j. This is simply the usual relation for the deformation of a spring
with stiffness K. It is obviously linear and, as such, incompatible with Hertz's
penetration model (Section 2.2.2), which predicts a power law with an expo-
nent of ~ to describe how the force depends on the penetration distance. To
allow for this nonlinearity, (6-5) is modified to a slightly more general form
(6-6)
where f3 = ~ in the Hertz model, and f3 = - ~ in the case of a soft crust (see
Section 2.2.2).
• A friction force which opposes the rupture of contacts. It plays a dissipative
role similar to that of the Euler-Coulomb dynamic friction. For generality,
two components are distinguished. The normal component is
(6-7)
where mij is the reduced mass of the system of two colliding particles i
and j, V;j is the difference between their velocities, and D n is a dissipation
coefficient characterizing the separation of contact along the direction of llij.
Likewise, the tangential component of the friction force is
(6-8)
where tij is a vector tangent to the contact, that is to say, perpendicular to llij,
along the slip direction, and D t describes the corresponding dissipation.
Here again the linear approximation contained in (6-7) and (6-8) is sometimes too
limiting. The equations are often generalized in the form
D
Oashpot
Mechanical Analogies
The preceding equations were introduced purely phenomenologically. They may,
however, be interpreted in terms of more concrete models that give a physical
meaning to the parameters figuring in the equations. The simplest analogy is de-
picted in Figure 125. It features a spring (simulating elastic restitution) coupled to
a linear dampener. 6
Such a simple system obviously cannot account for the subtleties of contact
interactions, such as the plastic deformations that typically occur when two collid-
ing spheres penetrate each other, as pointed out in Section 2.2.2. More elaborate
variants have been proposed to simulate these more complicated effects [38]. An
example is depicted in Figure 126. With enough creative imagination, other ar-
rangements can undoubtedly be contrived, but we should keep in mind that such
mechanical analogies have limitations and remain crude pictures of reality. Figure
126(b) describes the behavior of the system shown in Figure 126(a). The spring of
stiffness K 1 is compressed, simulating the two particles colliding and penetrating
each other. On Figure 126(b), the operating point moves up along the straight line
of slope K 1 until its abscissa is equal to et. At that point, the ratchet mechanism
jumps down one notch, which causes the stiffness to suddenly increase to K2.
If the system is allowed to relax in its new configuration, the operating point
moves back down along a different line of slope K 2 , until it reaches the point of
abscissa eto. Since the force has now returned to zero, the system is clearly left in a
different state relative to what it was, which is consistent with the phenomenon of
6 Such a dampener is sometimes referred to as LSD, for linear spring dashpot. Those fond of acronyms
will shortly be treated to an example of PLS, for partially latching spring.
194 6. Numerical Simulations
(a) (b)
~o
u.
Ratchet "'iii
Mechanism E
o
z
Normal Elongation
FIGURE 126. Mechanical model of the phenomenon of plasticity. It uses a set of coupled
springs, one of which activates a ratchet mechanism.
plastic deformation. As more and more notches get engaged, the operating point
describes increasingly skinny triangular sectors, as shown in Figure 126(b). The
plastic limit corresponds to the first notch on the ratchet. As long as that condition
is not exceeded, the regime remains linear with a stiffness K j. Beyond that point,
two offset springs act in parallel, with a net stiffness K j - K 2 . As intriguing as
this device may be, it still cannot reproduce Hertz's nonlinear penetration regime.
mi mj
where fei) = fe~) + f//), since only the normal force comes into play in a head-
on collision. For simplicity, the vector notation has been dropped. The previous
equation applies only when x = ~ (d; + d j ) - rij > O. Under these conditions, we
have
d 2x dx 2
- 2 +~-+wox=O, (6-9)
dt dt
7 Soas to limit ourselves to these regimes, we will refrain from discussing the simulations done by
Taguchi, who added to the equations a viscous dissipative term [105], [106].
6.3 Molecular Dynamics (MD) Simulations 195
Contact ends when x(tc) becomes negative. Note that in the present model, tc is
independent of the relative velocity of the particles. We may define the equivalent
of the coefficient of restitution E: introduced in Section 2.2.2 by writing
[dx / dt]t=t,
E == - ,
[dx/dt]t=o
which leads to
This last relation clearly demonstrates the link between the loss of momentum
during collision and the dissipative term D n (or ft). The coefficient of restitution
also turns out to be independent of the relative velocity.
We are now in a position to calculate the maximum penetration depth X max along
the same line we followed for Hertz's model (Section 2.2.2). Maximum penetration
is obtained when the penetration velocity dx/dt vanishes at time t = tmax ' From
(6-10) and (6-11), the result is
If the system is only slightly dissipative (for instance, when E: ::: 0.9), then wo» ft,
and tmax approaches the value 2tc , as in the case of Hertz's model. Under these
circumstances, X max reduces to
Vo
X max =-.
Wo
In other words, the penetration depth is then proportional to the relative velocity of
the colliding particles. This result differs significantly from Hertz's model, which
196 6. Numerical Simulations
a
predicts a much weaker dependence (as v I/5 ). We thus arrive at the conclusion
that a linear elastic model deviates substantially from the physical behavior of
real collisions. 8 It seems necessary to devise a more realistic model incorporating
the nonlinear nature of the contact interactions. That is the objective of the next
paragraph.
d x
2 (X)Y dx (X)fJ
m-+1]d - -+Ed - x=O, (6-12)
dt 2 d dt f3
where E depends on Young's modulus and Poisson's coefficient of the material,
and 1] depends on the compression as well as the viscosity with respect to shearing.
We note in passing that the dissipative term in this last equation corresponds to a
purely viscoelastic interaction. As such, the equation does not account for plastic
deformations, permanent distortions, or dissipation of vibrational excitations via
phonons, all of which were mentioned during our discussion of Hertz's model in
Section 2.2.2.
It is informative to consider a few particular cases in terms of values of the
exponents f3 and y:
(1) f3 = 0 and y = 0 corresponds to the linear interaction described by (6-9).
(2) f3 = ~ and y = 0 corresponds to the situation described by Hertz's equation.
This can be verified as an exercise.
(3) f3 = ~ and y = ~ corresponds to a generalized situation (Kuwabara and
Kono model) in which a viscoelastic compression is added to the normal
elastic interaction [107]. In this model, the nonlinearity stems from purely
geometrical properties of the penetration.
It should be fairly evident by now that modeling collisions between particles is not
easy. The physics of contact interactions is inherently complex and remains poorly
elucidated. Furthermore, good numerical algorithms are tricky to develop, because
they have to scrupulously take into account all the time constants involved (such
as the duration of collisions, the relative velocities, the time of free flights, and
others). Carelessness is likely to lead to unphysical results. To illustrate the point,
we proceed to discuss a completely artificial effect that comes up in models based
8We might come to the elToneous conclusion that the present simple model, based on coupled spring
and dampener, is useless. In fact, it can be shown that, as long as the contact duration tc is chosen
judiciously, in other words, realistically from the standpoint of the physics of the materials involved,
MD simulations yield results that turn out to be fairly satisfactory.
6.3 Molecular Dynamics (MD) Simulations 197
on soft spheres [60]. It has come to be known as the "detachment effect," because
it causes an unphysical separation of particles undergoing multiple collisions. It
can best be understood by pursuing the simple model used in Section 3.2.1 to
describe the behavior of a one-dimensional stack of spheres subjected to a vertical
sinusoidal excitation.
ceff=&,
where Eo and E f are the initial and final kinetic energies (before and after the
collision). It is important to choose a suitable variable to analyze this problem.
Numerical simulations suggest that one such variable is the ratio cr = so/(votc ),
where So is the initial distance separating the colliding particles. Figure 127 shows
how ceff depends on cr. The trend indicated in the figure seems to be "universal" in
0.5
ED-LRV
0.0
-6 -4 -2 o 2
log(a)
FIGURE 127. Effective restitution coefficient 8 plotted against the ratio SO/vote (see text).
The horizontal line at 8eff "'" 0.34 corresponds to the result of the ED-LRV procedure
described in Section 6.2.2. The present results were obtained for a column of ten spheres
and with fixed walls. The parameters used for the calculation were: d = 1 mm, 8 = 0.9
(the true coefficient of restitution), tc = 0.0022 s, and va = 0.03 m/s (after [59]).
198 6. Numerical Simulations
the sense that the results of MD simulations obtained for an extremely wide range
of values of tc (over three decades) and Va (varying by a factor of 400) all line up
along the same curve.
The graph reveals a sudden change in the behavior of a column of spheres when
CJ = 1, that is to say, when the separation between particles becomes comparable to
the distance traveled during the duration of a collision. As the separation becomes
smaller than this critical value, the effective coefficient of restitution approaches
unity. Such a result is in flagrant contradiction not only with experiments, but
also with theoretical predictions that 8 eff should be a decreasing function of the
number of particles. Here we find, instead, that it becomes equal to or larger than
the coefficient of a single sphere, for which 8 = 0.9. Put another way, the column
appears far too "elastic," which from a practical point of view leads to an artificial
separation of the colliding particles. No such problem exists when the initial sep-
aration is sufficiently large (CJ > 1), in which case the molecular dynamics model
agrees quite well with the results of the ED method. The latter technique, coupled
with an LRV procedure, correctly predicts that 8 eff does not depend on CJ. The arti-
ficial decompaction just discussed is at the origin of the designation "detachment
effect," whose meaning is further illustrated in Figure 128.
Here the effect is clearly demonstrated when the particles are initially in contact.
If we were to repeat the same calculations with particles initially separated by about
0.01 mm, the two techniques would produce virtually identical results.
There is, incidentally, another related phenomenon, known as the brake failure
effect, when particles collide tangentially [108]. It comes about for very much
the same reason. Here again, particles are slowed down considerably less in MD
simulations than in other, more realistic, mechanics-based models.
We conclude this brief review of molecular-dynamics models with a more
general remark, which in fact applies to all other simulation techniques as well.
0.2
ED
E
oS 0.1
~
0
t5Q)
.~
I-
0.0
FIGURE 128. Trajectories of the centers of ten spherical particles. The MD model was
carried out using exponents f3 = ~ and y = 0 (Hertz's model). Other parameters were:
6
E: = 0.86, and t c = 6 x 10- s for a binary collision, va = -0.2 mis, and Sa = O. The ED
model used the same values (after [59]).
6.4 Simulation of the Dynamics of Contacts 199
The results generally converge as long as the colliding particles spend most of
their time sufficiently far apart, in which case the dynamic behavior of the entire
system can be accurately modeled by a series of binary collisions. As soon as
more than two particles come in contact at the same time, several questions come
up. Are the collisions binary or ternary, or worse? Are we dealing with blocks?
The answers are never simple, even from a straight physics point of view. All
simulation techniques pay a price for this fundamental indeterminacy, although
the symptoms may differ in each case. These difficulties manifest themselves in
the form of inelastic collapse in ED models, or the detachment effect in their MD
counterparts. As we pointed out in Section 3.1.1, short of knowing the details
of interactions on a microscopic scale, we find ourselves rather helpless when it
comes to predicting the dynamical behavior of a simple stack of as few as three
particles when they are almost in contact.
T N
/-!N
o o
v D
o
v D
FIGURE 129. The diagrams on the left correspond to Coulomb's law of dry friction. T and
N are tangential and normal forces. The diagrams on the right show Signorini's conditions.
D is the distance between contact points. Both upper diagrams are discontinuous. The dis-
continuities have been partly mitigated in the lower figures. The gentler form of Coulomb's
law implies a viscous interaction in the vicinity of the contact, while that corresponding to
Signorini's condition assumes an elastic reaction when the solids get close to each other
(after [Ill]).
A similar analysis can be done on the basis of Signorini's conditions, which deal
with the normal, rather than tangential, forces. They apply to hard objects, consid-
ered impenetrable in the sense defined earlier:
From this point of view, Signorini's conditions exhibit very much the same type
of discontinuity as Coulomb's law.
6.4 Simulation of the Dynamics of Contacts 20 I
Static
T Dynamic T
!-I,N
!-IN !-Id N
o 0
v V
-!-IdN
-!-I,N
and
where <t>t and <t>n are the normal and tangential components of the reaction force
due to friction. These components depend on the mode of contact between the
two particles, but not on the external forces, since we deliberately treat the two
separately. If we work in the frame of reference attached to the contact point
between the two particles, the fundamental equations are represented in the above
diagrams by straight lines with a positive slope. These straight lines would intersect
the discontinuous curves at a single point, as shown in Figure 130, which implies
a unique solution. The problem is somewhat more complicated in the case of
dry friction. Whether the solution is unique or not depends on the experimental
conditions and the way Coulomb's friction is modeled. Figure 130 reveals the
following:
• If dry friction is modeled with a single coefficient fh = fhs = fhd, the solution
is always unique for a dynamic interaction.
• For a static interaction, the straight line describing the fundamental equation
becomes vertical, and the solution becomes undetermined (with an infinity of
solutions).
9 An excellent analysis of these indeterminacies and how to handle them mathematically can be found
in [111].
lOHere we neglect any possible rotation of the particles. It could easily be added to the equations, but
it would not materially change the argument.
202 6. Numerical Simulations
• If dry friction is modeled with two different friction coefficients /-is and fLd,
with fLs > /-id, the straight line can intersect Coulomb's graph at two distinct
points, and the solution is obviously not unique. Which solution the system
chooses depends on its prior history, which opens the door to the kinds of
hysteresis effects discussed in Sections 2.3 and 3.1.1. That is a commonly
recognized characteristic studied in structural analysis [112].
These considerations may well elicit growing skepticism that it will ever be possible
to accurately model any granular system that is subject to such intrinsic indetermi-
nacies. As we now know, these indeterminacies all come from the discontinuous
character of the static resistance force. One way to get around this problem is to
resort to well-behaved functional dependences of the type depicted in the lower
part of Figure 129. Another way is to consider the static situation simply as a
limiting case of the dynamic problem (when v -+ 0). Such arguments may indeed
be viewed as a posteriori justifications of the MD and ED simulations techniques
which, being inherently dynamical approaches, avoid these problems entirely. It
is also essential to bear in mind that we have considered only hard objects (in the
sense of the hard spheres in ED simulations). The creation and rupture of micro-
contacts is unlikely to be as discontinuous as implied by the standard constitutive
laws. It is in fact quite plausible that smoother functions might describe real phe-
nomena more realistically. What we can say with some confidence is that various
models based on the arguments presented above generally lead to results in good
agreement with experiments [Ill]. This includes ED and MD simulations, as well
as others to be discussed in the latter part of this chapter. There is no compelling
reason to promote anyone technique over another. In all likelihood, a particular
approach, based on specific simplifying assumptions, can be perfectly adequate in
certain circumstances, and completely break down in others. The best strategy is
to be flexible and keep an open mind.
We proceed next to discuss two more simulation techniques, based on pro-
cedures for synthesizing piles. These methods may appear somewhat primitive
when compared to the ones reviewed thus far. Yet, they too turn out to produce
very satisfactory results, at least when the geometry of a pile is an important factor.
might be worthwhile to reread the portion of Section 3.2.1 dealing with the excitation period T in
11 It
relation to the relaxation time T of the system. We can also appreciate that the Me method should
be applied preferably to materials with a low coefficient of elastic restitution 8, simply because the
relaxation following excitation is typically a fairly rapid phenomenon.
204 6. Numerical Simulations
by following fairly closely the traditional way of using Monte Carlo calculations.
We will subsequently discuss the specifics of applying the technique to granular
materials.
Although we could, without unduly complicating the problem, treat the case of
a three-dimensional pile of dissimilar granules, we consider, instead, a collection
of identical disks of diameter d. These disks, assumed to be impenetrable, are
initially arranged randomly in a hypothetical vertical two-dimensional container
without walls. 12 In practice, this is approximated by using a ring-shaped container.
The initial configuration of such a system of N disks is described by a generalized
vector encompassing the coordinates of all centers
where m is the mass of an individual disk and Z j is the altitude of its center.
The Monte Carlo method is based on analyzing the probability P of different
configurations r, each of which has an energy E g (1). Thermodynamics tells us that
(6-l4a)
and
(6-l4b)
where ~x and ~z are independent random variables equally distributed in the in-
terval [-1, + 1], and 8 > O. So as to ensure that the disks do not penetrate each
other during the successive trials, we require that the interaction between adjacent
particles be governed by a potential energy U (s) of a pair such that
12This is a crucial restriction. We have seen in Chapters 3 and 5 that walls induce convection effects
in granular media. By getting rid of them, convection is conveniently eliminated. All that is then left
are geometrical phenomena, such as "arch effects" of the type described in Section 5.2.1.
6.5 Monte Carlo (MC) Simulations 205
and
The trials conducted according to (6-13) through (6-15) must be evaluated for
plausibility against the following criteria:
• If the quantity
the new configuration has a lower energy than the one we started with. It is
therefore retained for subsequent calculations.
• If t"E > 0, the solution I' is not necessarily rejected, as it may well be
accessible via simple thermal agitation. It is therefore assigned a probability
given by
P t"E - P[E(I")] e x
(t"E)
( ) - P[E(i')] P -kT'
- (6-17)
The procedure calls for jiggling every single particle of the system, until each
member of the population has had its tum, which completes one iteration. The
new configuration is then used as a fresh starting point for the next iteration, in
which all particles are moved about all over again and allowed to relax, and so
forth.
Some comments on the temperature of the system are in order. The method
we have just described is essentially what is used traditionally for Monte Carlo
simulations of Brownian systems. When it is applied to macroscopic objects like
granular materials, the significance of equations involving the thermal energy kT
raises some legitimate questions.
As pointed out in Chapter 1, the Brownian motion of typical systems of inter-
est here is entirely negligible, the ratio mg t"z/ kT being of the order of 10 12 at
ordinary temperatures. If so, (6-17), and the criterion associated with it, gives a
probability that is always practically zero. In other words, the only really relevant
equation for a granular system is (6-16), which means that the potential energy
can only decrease at each step of the iteration. This is equivalent to assuming that
the temperature of the system is at absolute zero. The clear implication is that the
system traps particles in potential wells from which they cannot escape without
collisions on a microscopic level. In accordance with our earlier discussions, it
is clear that this simulation strategy amounts to neglecting the short-range inter-
actions normally associated with multiple collisions-which are equivalent to a
local temperature of the granular. Rather, it deals fundamentally with systems in
their relaxed states.
206 6. Numerical Simulations
Despite these restrictions, this type ofsimulation technique has proven extremely
useful to model a number of situations, such as a pile of dissimilar particles. Byway
of summary, we emphasize again that, by its very nature, this particular technique
is not a good choice to describe nonrelaxed configurations, where particles spend
only a fraction of their time in actual contact. One important example is that of
fluidized beds, which are more suitably treated by ED or MD simulations.
5
Stop
the centers of the two underlying particles. We note, incidentally, that this
assumes the absence of any rebound when the particle raches that favorable
spot, in accordance with earlier remarks on these stacking methods.
• Once a particle stops, it becomes permanently embedded in the pile.
Agitation can be simulated, for instance, by perturbing the entire system upward
and leaving it to relax on its own. This can be accomplished in the following
manner:
(1) A pile is first generated, by randomly depositing particles one at a time
and allowing them to relax after each addition, using the algorithm just
described.
(2) The stacked particles are numbered in ascending order starting from the
bottom.
(3) The entire pile is raised (fictitiously), and each particle is left to fall down
individually, again using the above algorithm. The process starts with the
lowest-numbered particles and gradually works its way up. To some extent,
it preserves a memory of the pile's prior configuration. 13
(4) Steps (2) and (3) are repeated many times, thereby simulating a vertical
vibration.
This type of simulation is relatively frugal in terms of computation time. It is the
technique of choice to treat cases involving large numbers of particles in three
dimensions. However, the limitations discussed in the context of the Monte Carlo
method apply here as well. Both techniques are good choices to treat a series of
relaxed states, to the specific exclusion of rapid interactions and multiple collisions
that may occur in real systems. The method of steepest descent is particularly well
suited to dealing with problems in which geometry is of paramount importance.
It has produced results in relatively good agreement with experiments. Perhaps its
greatest claim to fame is to have predicted the existence of critical diameters in the
Brazil nut problem, similar to the ones we found analytically in Section 5.2.1. 14
13We encourage the reader to refresh his or her memory by going back to the part of Chapter 3 that
describes the various modes of decompaction of a pile nnder vertical excitation, particularly in one
and two dimensions. This will provide further opportunities to reflect on the degree of realism of the
present algorithm.
14The algorithm described here was originally developed by Jullien et al. [91]. Interestingly, early
versions did not include noise, that is to say, random fluctuations of the particles's positions during
the stacking of the pile, making it entirely deterministic. One consequence was that segregation was
precluded for <P < <Pc, whereas the analytical model in Section 5.2.1 predicts merely a change in
behavior as the critical value <Pc is crossed. A noise source was subsequently added to the model,
and a more realistic behavior did indeed result from this improvement.
Bibliography
[22] J. Duran, "La Physique du tas de sable," (The physics of a sand pile), Revue du
Palais de la Decouverte 21 (1994),23-224.
[23] R. M. Nedderman, Statics and Kinematics ofGranular Materials (Cambridge, UK:
Cambridge University Press, 1992).
[24] D. Bideau and J. Dodds, eds., Physics of Granular Media (Commack, NY: Nova
Science, 1991).
[25] H. J. Heimann and S. Roux, eds., Statistical Modelsfor the Fracture ofDisordered
Media (New York: Elsevier, 1990).
[26] K. K. Bardhan, B. K. Chakrabarti, and A. Hansen, eds., Non-Linearity and Break-
down in Soft Condensed Matter (Berlin: Springer-Verlag, 1994).
[27] E. Guazzelli and L. Oger, eds., Mobile Particulate Systems (Dordrecht: Kluwer
Academic, 1995).
[28] S. B. Savage and D. I Jeffrey, J. Fluid Mech. 110 (1981), 255.
[29] M. W. Richman, Acta Mech. 75 (1988), 227.
[30] M. W. Richman, J. Rheo/. 33 (1989), 1293.
[31] H. K. Pak, E. Van Doorn, and R. P. Behringer, Phys. Rev. Lett. 74 (1995), 4643.
[32] D. D. Joseph, European J. Mech. B (Fluids) 9 (1990), 565.
[33] K. Hotta, K. Takeda, and K, Iinoya, Powder Technol. 10 (1974), 231.
[34] F. P. Bowden and D. Tabor, The Friction and Lubrication of Solids (Oxford:
Clarendon Press, 1950).
[35] E. Rabinowicz, Friction and Wear of Materials (New York: Wiley, 1965).
[36] W. Goldsmith, Impact: The Theory of Physical Behaviour of Colliding Solids
(London: Arnold, 1960).
[37] B. Bernu and R. Mazhigi, J. Phys. A 23 (1990), 5745.
[38] O. R. Walton and R. L. Braun, J. Rheol. 30 (1986), 949.
[39] S. Luding, Phys. Rev. E 52 (1995), 4442.
[40] L. D. Landau and E. M. Lifshitz, Mechanics (Oxford: Pergamon Press, 1976).
[41] S. F. Foerster, M. Y. Louge, H. Chang, and K. Allia, Phys. Fluids 6 (1994), 1108.
[42] J. Duffy and R. D. Mindlin, J. Appl. Mech. (ASME) 24 (1957), 585.
[43] P.-G. de Gennes, private communication.
[44] S. N. Domenico, Geophysics 42 (1977), 1339.
[45] J. D. Goddard, Proc. Roy. Soc. (London) 430 (1990), 105.
[46] F. X. Riguidel, R. Jullien, G. Ristow, A. Hansen, and D. Bideau, J. Physique I4
(1994),4973.
[47] G. G. Batrouni, S. Dippel, and L. Samson, Phys. Rev. E 53 (1996), 6496.
[48] F. Heslot, T. Baumberger, B. Perrin, B. Caroli, and C. Caroli, Phys. Rev. E (1994),
4973.
[49] J. Grindlay, Amer. J. Phys. 61 (1993), 469.
[50] E. Guyon, in Physics of Granular Media, edited by D. Bideau and J. Dodds
(New York: Nova Science, 1991).
[51] P. Dantu, in Proc. of the Fourth International Con! on Soil Mechanics and Foun-
dation Eng., vol. 1 (London: Butterworths Scientific, 1957), p. 144.
[52] S. F. Edwards and C. C. Mounfield, Phys. A 210 (1994), 290.
[53] D. Stauffer, H. J. Hermann, and S. Roux, J. Physique 48 (1987), 437.
[54] T. Travers, M. Amrni, D. Bideau, A. Gervois, J.-C. Messager, and I-P. Troadec,
Europhys. Lett. 4 (1987),329.
Bibliography 211
CAM,128 Fluctuations, 1
Cannon ball, 54, 64, 101 Fluidization, 53, 76, 82
Chaos, 80, 85 Fluidized bed, 15
Cohesion, 5, 16 Fractal, 23, 132, 176, 179
Collision, 34, 55, 77,193 Fracture, 16, 107, 166
Compaction, 7 Fragmentation, 53, 72, 101,
Condensation, 82 104
213
214 Index